1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.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/sched/isolation.h>
58 #include <linux/mem_encrypt.h>
60 #include <trace/events/kvm.h>
62 #include <asm/debugreg.h>
66 #include <linux/kernel_stat.h>
67 #include <asm/fpu/internal.h> /* Ugh! */
68 #include <asm/pvclock.h>
69 #include <asm/div64.h>
70 #include <asm/irq_remapping.h>
71 #include <asm/mshyperv.h>
72 #include <asm/hypervisor.h>
73 #include <asm/intel_pt.h>
74 #include <asm/emulate_prefix.h>
75 #include <clocksource/hyperv_timer.h>
77 #define CREATE_TRACE_POINTS
80 #define MAX_IO_MSRS 256
81 #define KVM_MAX_MCE_BANKS 32
82 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
83 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
85 #define emul_to_vcpu(ctxt) \
86 ((struct kvm_vcpu *)(ctxt)->vcpu)
89 * - enable syscall per default because its emulated by KVM
90 * - enable LME and LMA per default on 64 bit KVM
94 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
96 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
99 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
101 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
102 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
104 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
105 static void process_nmi(struct kvm_vcpu *vcpu);
106 static void enter_smm(struct kvm_vcpu *vcpu);
107 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
108 static void store_regs(struct kvm_vcpu *vcpu);
109 static int sync_regs(struct kvm_vcpu *vcpu);
111 struct kvm_x86_ops kvm_x86_ops __read_mostly;
112 EXPORT_SYMBOL_GPL(kvm_x86_ops);
114 static bool __read_mostly ignore_msrs = 0;
115 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
117 static bool __read_mostly report_ignored_msrs = true;
118 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
120 unsigned int min_timer_period_us = 200;
121 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
123 static bool __read_mostly kvmclock_periodic_sync = true;
124 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
126 bool __read_mostly kvm_has_tsc_control;
127 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
128 u32 __read_mostly kvm_max_guest_tsc_khz;
129 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
130 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
131 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
132 u64 __read_mostly kvm_max_tsc_scaling_ratio;
133 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
134 u64 __read_mostly kvm_default_tsc_scaling_ratio;
135 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
137 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
138 static u32 __read_mostly tsc_tolerance_ppm = 250;
139 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
142 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
143 * adaptive tuning starting from default advancment of 1000ns. '0' disables
144 * advancement entirely. Any other value is used as-is and disables adaptive
145 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
147 static int __read_mostly lapic_timer_advance_ns = -1;
148 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
150 static bool __read_mostly vector_hashing = true;
151 module_param(vector_hashing, bool, S_IRUGO);
153 bool __read_mostly enable_vmware_backdoor = false;
154 module_param(enable_vmware_backdoor, bool, S_IRUGO);
155 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
157 static bool __read_mostly force_emulation_prefix = false;
158 module_param(force_emulation_prefix, bool, S_IRUGO);
160 int __read_mostly pi_inject_timer = -1;
161 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
163 #define KVM_NR_SHARED_MSRS 16
165 struct kvm_shared_msrs_global {
167 u32 msrs[KVM_NR_SHARED_MSRS];
170 struct kvm_shared_msrs {
171 struct user_return_notifier urn;
173 struct kvm_shared_msr_values {
176 } values[KVM_NR_SHARED_MSRS];
179 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
180 static struct kvm_shared_msrs __percpu *shared_msrs;
182 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
183 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
184 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
185 | XFEATURE_MASK_PKRU)
187 u64 __read_mostly host_efer;
188 EXPORT_SYMBOL_GPL(host_efer);
190 static u64 __read_mostly host_xss;
191 u64 __read_mostly supported_xss;
192 EXPORT_SYMBOL_GPL(supported_xss);
194 struct kvm_stats_debugfs_item debugfs_entries[] = {
195 VCPU_STAT("pf_fixed", pf_fixed),
196 VCPU_STAT("pf_guest", pf_guest),
197 VCPU_STAT("tlb_flush", tlb_flush),
198 VCPU_STAT("invlpg", invlpg),
199 VCPU_STAT("exits", exits),
200 VCPU_STAT("io_exits", io_exits),
201 VCPU_STAT("mmio_exits", mmio_exits),
202 VCPU_STAT("signal_exits", signal_exits),
203 VCPU_STAT("irq_window", irq_window_exits),
204 VCPU_STAT("nmi_window", nmi_window_exits),
205 VCPU_STAT("halt_exits", halt_exits),
206 VCPU_STAT("halt_successful_poll", halt_successful_poll),
207 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
208 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
209 VCPU_STAT("halt_wakeup", halt_wakeup),
210 VCPU_STAT("hypercalls", hypercalls),
211 VCPU_STAT("request_irq", request_irq_exits),
212 VCPU_STAT("irq_exits", irq_exits),
213 VCPU_STAT("host_state_reload", host_state_reload),
214 VCPU_STAT("fpu_reload", fpu_reload),
215 VCPU_STAT("insn_emulation", insn_emulation),
216 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
217 VCPU_STAT("irq_injections", irq_injections),
218 VCPU_STAT("nmi_injections", nmi_injections),
219 VCPU_STAT("req_event", req_event),
220 VCPU_STAT("l1d_flush", l1d_flush),
221 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
222 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
223 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
224 VM_STAT("mmu_pte_write", mmu_pte_write),
225 VM_STAT("mmu_pte_updated", mmu_pte_updated),
226 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
227 VM_STAT("mmu_flooded", mmu_flooded),
228 VM_STAT("mmu_recycled", mmu_recycled),
229 VM_STAT("mmu_cache_miss", mmu_cache_miss),
230 VM_STAT("mmu_unsync", mmu_unsync),
231 VM_STAT("remote_tlb_flush", remote_tlb_flush),
232 VM_STAT("largepages", lpages, .mode = 0444),
233 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
234 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
238 u64 __read_mostly host_xcr0;
239 u64 __read_mostly supported_xcr0;
240 EXPORT_SYMBOL_GPL(supported_xcr0);
242 static struct kmem_cache *x86_fpu_cache;
244 static struct kmem_cache *x86_emulator_cache;
246 static struct kmem_cache *kvm_alloc_emulator_cache(void)
248 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
249 unsigned int size = sizeof(struct x86_emulate_ctxt);
251 return kmem_cache_create_usercopy("x86_emulator", size,
252 __alignof__(struct x86_emulate_ctxt),
253 SLAB_ACCOUNT, useroffset,
254 size - useroffset, NULL);
257 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
259 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
262 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
263 vcpu->arch.apf.gfns[i] = ~0;
266 static void kvm_on_user_return(struct user_return_notifier *urn)
269 struct kvm_shared_msrs *locals
270 = container_of(urn, struct kvm_shared_msrs, urn);
271 struct kvm_shared_msr_values *values;
275 * Disabling irqs at this point since the following code could be
276 * interrupted and executed through kvm_arch_hardware_disable()
278 local_irq_save(flags);
279 if (locals->registered) {
280 locals->registered = false;
281 user_return_notifier_unregister(urn);
283 local_irq_restore(flags);
284 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
285 values = &locals->values[slot];
286 if (values->host != values->curr) {
287 wrmsrl(shared_msrs_global.msrs[slot], values->host);
288 values->curr = values->host;
293 void kvm_define_shared_msr(unsigned slot, u32 msr)
295 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
296 shared_msrs_global.msrs[slot] = msr;
297 if (slot >= shared_msrs_global.nr)
298 shared_msrs_global.nr = slot + 1;
300 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
302 static void kvm_shared_msr_cpu_online(void)
304 unsigned int cpu = smp_processor_id();
305 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
309 for (i = 0; i < shared_msrs_global.nr; ++i) {
310 rdmsrl_safe(shared_msrs_global.msrs[i], &value);
311 smsr->values[i].host = value;
312 smsr->values[i].curr = value;
316 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
318 unsigned int cpu = smp_processor_id();
319 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
322 value = (value & mask) | (smsr->values[slot].host & ~mask);
323 if (value == smsr->values[slot].curr)
325 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
329 smsr->values[slot].curr = value;
330 if (!smsr->registered) {
331 smsr->urn.on_user_return = kvm_on_user_return;
332 user_return_notifier_register(&smsr->urn);
333 smsr->registered = true;
337 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
339 static void drop_user_return_notifiers(void)
341 unsigned int cpu = smp_processor_id();
342 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
344 if (smsr->registered)
345 kvm_on_user_return(&smsr->urn);
348 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
350 return vcpu->arch.apic_base;
352 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
354 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
356 return kvm_apic_mode(kvm_get_apic_base(vcpu));
358 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
360 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
362 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
363 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
364 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
365 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
367 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
369 if (!msr_info->host_initiated) {
370 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
372 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
376 kvm_lapic_set_base(vcpu, msr_info->data);
377 kvm_recalculate_apic_map(vcpu->kvm);
380 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
382 asmlinkage __visible void kvm_spurious_fault(void)
384 /* Fault while not rebooting. We want the trace. */
385 BUG_ON(!kvm_rebooting);
387 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
389 #define EXCPT_BENIGN 0
390 #define EXCPT_CONTRIBUTORY 1
393 static int exception_class(int vector)
403 return EXCPT_CONTRIBUTORY;
410 #define EXCPT_FAULT 0
412 #define EXCPT_ABORT 2
413 #define EXCPT_INTERRUPT 3
415 static int exception_type(int vector)
419 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
420 return EXCPT_INTERRUPT;
424 /* #DB is trap, as instruction watchpoints are handled elsewhere */
425 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
428 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
431 /* Reserved exceptions will result in fault */
435 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
437 unsigned nr = vcpu->arch.exception.nr;
438 bool has_payload = vcpu->arch.exception.has_payload;
439 unsigned long payload = vcpu->arch.exception.payload;
447 * "Certain debug exceptions may clear bit 0-3. The
448 * remaining contents of the DR6 register are never
449 * cleared by the processor".
451 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
453 * DR6.RTM is set by all #DB exceptions that don't clear it.
455 vcpu->arch.dr6 |= DR6_RTM;
456 vcpu->arch.dr6 |= payload;
458 * Bit 16 should be set in the payload whenever the #DB
459 * exception should clear DR6.RTM. This makes the payload
460 * compatible with the pending debug exceptions under VMX.
461 * Though not currently documented in the SDM, this also
462 * makes the payload compatible with the exit qualification
463 * for #DB exceptions under VMX.
465 vcpu->arch.dr6 ^= payload & DR6_RTM;
468 * The #DB payload is defined as compatible with the 'pending
469 * debug exceptions' field under VMX, not DR6. While bit 12 is
470 * defined in the 'pending debug exceptions' field (enabled
471 * breakpoint), it is reserved and must be zero in DR6.
473 vcpu->arch.dr6 &= ~BIT(12);
476 vcpu->arch.cr2 = payload;
480 vcpu->arch.exception.has_payload = false;
481 vcpu->arch.exception.payload = 0;
483 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
485 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
486 unsigned nr, bool has_error, u32 error_code,
487 bool has_payload, unsigned long payload, bool reinject)
492 kvm_make_request(KVM_REQ_EVENT, vcpu);
494 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
496 if (has_error && !is_protmode(vcpu))
500 * On vmentry, vcpu->arch.exception.pending is only
501 * true if an event injection was blocked by
502 * nested_run_pending. In that case, however,
503 * vcpu_enter_guest requests an immediate exit,
504 * and the guest shouldn't proceed far enough to
507 WARN_ON_ONCE(vcpu->arch.exception.pending);
508 vcpu->arch.exception.injected = true;
509 if (WARN_ON_ONCE(has_payload)) {
511 * A reinjected event has already
512 * delivered its payload.
518 vcpu->arch.exception.pending = true;
519 vcpu->arch.exception.injected = false;
521 vcpu->arch.exception.has_error_code = has_error;
522 vcpu->arch.exception.nr = nr;
523 vcpu->arch.exception.error_code = error_code;
524 vcpu->arch.exception.has_payload = has_payload;
525 vcpu->arch.exception.payload = payload;
526 if (!is_guest_mode(vcpu))
527 kvm_deliver_exception_payload(vcpu);
531 /* to check exception */
532 prev_nr = vcpu->arch.exception.nr;
533 if (prev_nr == DF_VECTOR) {
534 /* triple fault -> shutdown */
535 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
538 class1 = exception_class(prev_nr);
539 class2 = exception_class(nr);
540 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
541 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
543 * Generate double fault per SDM Table 5-5. Set
544 * exception.pending = true so that the double fault
545 * can trigger a nested vmexit.
547 vcpu->arch.exception.pending = true;
548 vcpu->arch.exception.injected = false;
549 vcpu->arch.exception.has_error_code = true;
550 vcpu->arch.exception.nr = DF_VECTOR;
551 vcpu->arch.exception.error_code = 0;
552 vcpu->arch.exception.has_payload = false;
553 vcpu->arch.exception.payload = 0;
555 /* replace previous exception with a new one in a hope
556 that instruction re-execution will regenerate lost
561 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
563 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
565 EXPORT_SYMBOL_GPL(kvm_queue_exception);
567 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
569 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
571 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
573 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
574 unsigned long payload)
576 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
578 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
580 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
581 u32 error_code, unsigned long payload)
583 kvm_multiple_exception(vcpu, nr, true, error_code,
584 true, payload, false);
587 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
590 kvm_inject_gp(vcpu, 0);
592 return kvm_skip_emulated_instruction(vcpu);
596 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
598 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
600 ++vcpu->stat.pf_guest;
601 vcpu->arch.exception.nested_apf =
602 is_guest_mode(vcpu) && fault->async_page_fault;
603 if (vcpu->arch.exception.nested_apf) {
604 vcpu->arch.apf.nested_apf_token = fault->address;
605 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
607 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
611 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
613 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
614 struct x86_exception *fault)
616 struct kvm_mmu *fault_mmu;
617 WARN_ON_ONCE(fault->vector != PF_VECTOR);
619 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
623 * Invalidate the TLB entry for the faulting address, if it exists,
624 * else the access will fault indefinitely (and to emulate hardware).
626 if ((fault->error_code & PFERR_PRESENT_MASK) &&
627 !(fault->error_code & PFERR_RSVD_MASK))
628 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
629 fault_mmu->root_hpa);
631 fault_mmu->inject_page_fault(vcpu, fault);
632 return fault->nested_page_fault;
634 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
636 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
638 atomic_inc(&vcpu->arch.nmi_queued);
639 kvm_make_request(KVM_REQ_NMI, vcpu);
641 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
643 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
645 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
647 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
649 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
651 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
653 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
656 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
657 * a #GP and return false.
659 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
661 if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
663 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
666 EXPORT_SYMBOL_GPL(kvm_require_cpl);
668 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
670 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
673 kvm_queue_exception(vcpu, UD_VECTOR);
676 EXPORT_SYMBOL_GPL(kvm_require_dr);
679 * This function will be used to read from the physical memory of the currently
680 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
681 * can read from guest physical or from the guest's guest physical memory.
683 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
684 gfn_t ngfn, void *data, int offset, int len,
687 struct x86_exception exception;
691 ngpa = gfn_to_gpa(ngfn);
692 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
693 if (real_gfn == UNMAPPED_GVA)
696 real_gfn = gpa_to_gfn(real_gfn);
698 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
700 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
702 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
703 void *data, int offset, int len, u32 access)
705 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
706 data, offset, len, access);
709 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
711 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
716 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
718 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
720 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
721 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
724 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
726 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
727 offset * sizeof(u64), sizeof(pdpte),
728 PFERR_USER_MASK|PFERR_WRITE_MASK);
733 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
734 if ((pdpte[i] & PT_PRESENT_MASK) &&
735 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
742 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
743 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
749 EXPORT_SYMBOL_GPL(load_pdptrs);
751 bool pdptrs_changed(struct kvm_vcpu *vcpu)
753 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
758 if (!is_pae_paging(vcpu))
761 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
764 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
765 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
766 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
767 PFERR_USER_MASK | PFERR_WRITE_MASK);
771 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
773 EXPORT_SYMBOL_GPL(pdptrs_changed);
775 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
777 unsigned long old_cr0 = kvm_read_cr0(vcpu);
778 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
783 if (cr0 & 0xffffffff00000000UL)
787 cr0 &= ~CR0_RESERVED_BITS;
789 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
792 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
795 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
797 if ((vcpu->arch.efer & EFER_LME)) {
802 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
807 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
812 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
815 kvm_x86_ops.set_cr0(vcpu, cr0);
817 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
818 kvm_clear_async_pf_completion_queue(vcpu);
819 kvm_async_pf_hash_reset(vcpu);
822 if ((cr0 ^ old_cr0) & update_bits)
823 kvm_mmu_reset_context(vcpu);
825 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
826 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
827 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
828 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
832 EXPORT_SYMBOL_GPL(kvm_set_cr0);
834 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
836 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
838 EXPORT_SYMBOL_GPL(kvm_lmsw);
840 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
842 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
844 if (vcpu->arch.xcr0 != host_xcr0)
845 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
847 if (vcpu->arch.xsaves_enabled &&
848 vcpu->arch.ia32_xss != host_xss)
849 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
852 if (static_cpu_has(X86_FEATURE_PKU) &&
853 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
854 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
855 vcpu->arch.pkru != vcpu->arch.host_pkru)
856 __write_pkru(vcpu->arch.pkru);
858 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
860 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
862 if (static_cpu_has(X86_FEATURE_PKU) &&
863 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
864 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
865 vcpu->arch.pkru = rdpkru();
866 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
867 __write_pkru(vcpu->arch.host_pkru);
870 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
872 if (vcpu->arch.xcr0 != host_xcr0)
873 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
875 if (vcpu->arch.xsaves_enabled &&
876 vcpu->arch.ia32_xss != host_xss)
877 wrmsrl(MSR_IA32_XSS, host_xss);
881 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
883 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
886 u64 old_xcr0 = vcpu->arch.xcr0;
889 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
890 if (index != XCR_XFEATURE_ENABLED_MASK)
892 if (!(xcr0 & XFEATURE_MASK_FP))
894 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
898 * Do not allow the guest to set bits that we do not support
899 * saving. However, xcr0 bit 0 is always set, even if the
900 * emulated CPU does not support XSAVE (see fx_init).
902 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
903 if (xcr0 & ~valid_bits)
906 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
907 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
910 if (xcr0 & XFEATURE_MASK_AVX512) {
911 if (!(xcr0 & XFEATURE_MASK_YMM))
913 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
916 vcpu->arch.xcr0 = xcr0;
918 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
919 kvm_update_cpuid(vcpu);
923 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
925 if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
926 __kvm_set_xcr(vcpu, index, xcr)) {
927 kvm_inject_gp(vcpu, 0);
932 EXPORT_SYMBOL_GPL(kvm_set_xcr);
934 #define __cr4_reserved_bits(__cpu_has, __c) \
936 u64 __reserved_bits = CR4_RESERVED_BITS; \
938 if (!__cpu_has(__c, X86_FEATURE_XSAVE)) \
939 __reserved_bits |= X86_CR4_OSXSAVE; \
940 if (!__cpu_has(__c, X86_FEATURE_SMEP)) \
941 __reserved_bits |= X86_CR4_SMEP; \
942 if (!__cpu_has(__c, X86_FEATURE_SMAP)) \
943 __reserved_bits |= X86_CR4_SMAP; \
944 if (!__cpu_has(__c, X86_FEATURE_FSGSBASE)) \
945 __reserved_bits |= X86_CR4_FSGSBASE; \
946 if (!__cpu_has(__c, X86_FEATURE_PKU)) \
947 __reserved_bits |= X86_CR4_PKE; \
948 if (!__cpu_has(__c, X86_FEATURE_LA57)) \
949 __reserved_bits |= X86_CR4_LA57; \
950 if (!__cpu_has(__c, X86_FEATURE_UMIP)) \
951 __reserved_bits |= X86_CR4_UMIP; \
955 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
957 if (cr4 & cr4_reserved_bits)
960 if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu))
966 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
968 unsigned long old_cr4 = kvm_read_cr4(vcpu);
969 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
970 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
972 if (kvm_valid_cr4(vcpu, cr4))
975 if (is_long_mode(vcpu)) {
976 if (!(cr4 & X86_CR4_PAE))
978 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
980 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
981 && ((cr4 ^ old_cr4) & pdptr_bits)
982 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
986 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
987 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
990 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
991 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
995 if (kvm_x86_ops.set_cr4(vcpu, cr4))
998 if (((cr4 ^ old_cr4) & pdptr_bits) ||
999 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1000 kvm_mmu_reset_context(vcpu);
1002 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1003 kvm_update_cpuid(vcpu);
1007 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1009 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1011 bool skip_tlb_flush = false;
1012 #ifdef CONFIG_X86_64
1013 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1016 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1017 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1021 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1022 if (!skip_tlb_flush) {
1023 kvm_mmu_sync_roots(vcpu);
1024 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1029 if (is_long_mode(vcpu) &&
1030 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
1032 else if (is_pae_paging(vcpu) &&
1033 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1036 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1037 vcpu->arch.cr3 = cr3;
1038 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1042 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1044 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1046 if (cr8 & CR8_RESERVED_BITS)
1048 if (lapic_in_kernel(vcpu))
1049 kvm_lapic_set_tpr(vcpu, cr8);
1051 vcpu->arch.cr8 = cr8;
1054 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1056 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1058 if (lapic_in_kernel(vcpu))
1059 return kvm_lapic_get_cr8(vcpu);
1061 return vcpu->arch.cr8;
1063 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1065 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1069 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1070 for (i = 0; i < KVM_NR_DB_REGS; i++)
1071 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1072 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1076 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1080 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1081 dr7 = vcpu->arch.guest_debug_dr7;
1083 dr7 = vcpu->arch.dr7;
1084 kvm_x86_ops.set_dr7(vcpu, dr7);
1085 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1086 if (dr7 & DR7_BP_EN_MASK)
1087 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1089 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1091 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1093 u64 fixed = DR6_FIXED_1;
1095 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1100 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1102 size_t size = ARRAY_SIZE(vcpu->arch.db);
1106 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1107 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1108 vcpu->arch.eff_db[dr] = val;
1113 if (val & 0xffffffff00000000ULL)
1114 return -1; /* #GP */
1115 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1120 if (!kvm_dr7_valid(val))
1121 return -1; /* #GP */
1122 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1123 kvm_update_dr7(vcpu);
1130 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1132 if (__kvm_set_dr(vcpu, dr, val)) {
1133 kvm_inject_gp(vcpu, 0);
1138 EXPORT_SYMBOL_GPL(kvm_set_dr);
1140 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1142 size_t size = ARRAY_SIZE(vcpu->arch.db);
1146 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1151 *val = vcpu->arch.dr6;
1156 *val = vcpu->arch.dr7;
1161 EXPORT_SYMBOL_GPL(kvm_get_dr);
1163 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1165 u32 ecx = kvm_rcx_read(vcpu);
1169 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1172 kvm_rax_write(vcpu, (u32)data);
1173 kvm_rdx_write(vcpu, data >> 32);
1176 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1179 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1180 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1182 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1183 * extract the supported MSRs from the related const lists.
1184 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1185 * capabilities of the host cpu. This capabilities test skips MSRs that are
1186 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1187 * may depend on host virtualization features rather than host cpu features.
1190 static const u32 msrs_to_save_all[] = {
1191 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1193 #ifdef CONFIG_X86_64
1194 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1196 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1197 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1199 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1200 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1201 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1202 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1203 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1204 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1205 MSR_IA32_UMWAIT_CONTROL,
1207 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1208 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1209 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1210 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1211 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1212 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1213 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1214 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1215 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1216 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1217 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1218 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1219 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1220 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1221 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1222 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1223 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1224 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1225 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1226 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1227 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1228 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1231 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1232 static unsigned num_msrs_to_save;
1234 static const u32 emulated_msrs_all[] = {
1235 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1236 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1237 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1238 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1239 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1240 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1241 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1243 HV_X64_MSR_VP_INDEX,
1244 HV_X64_MSR_VP_RUNTIME,
1245 HV_X64_MSR_SCONTROL,
1246 HV_X64_MSR_STIMER0_CONFIG,
1247 HV_X64_MSR_VP_ASSIST_PAGE,
1248 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1249 HV_X64_MSR_TSC_EMULATION_STATUS,
1250 HV_X64_MSR_SYNDBG_OPTIONS,
1251 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1252 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1253 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1255 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1256 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1258 MSR_IA32_TSC_ADJUST,
1259 MSR_IA32_TSCDEADLINE,
1260 MSR_IA32_ARCH_CAPABILITIES,
1261 MSR_IA32_PERF_CAPABILITIES,
1262 MSR_IA32_MISC_ENABLE,
1263 MSR_IA32_MCG_STATUS,
1265 MSR_IA32_MCG_EXT_CTL,
1269 MSR_MISC_FEATURES_ENABLES,
1270 MSR_AMD64_VIRT_SPEC_CTRL,
1275 * The following list leaves out MSRs whose values are determined
1276 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1277 * We always support the "true" VMX control MSRs, even if the host
1278 * processor does not, so I am putting these registers here rather
1279 * than in msrs_to_save_all.
1282 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1283 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1284 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1285 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1287 MSR_IA32_VMX_CR0_FIXED0,
1288 MSR_IA32_VMX_CR4_FIXED0,
1289 MSR_IA32_VMX_VMCS_ENUM,
1290 MSR_IA32_VMX_PROCBASED_CTLS2,
1291 MSR_IA32_VMX_EPT_VPID_CAP,
1292 MSR_IA32_VMX_VMFUNC,
1295 MSR_KVM_POLL_CONTROL,
1298 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1299 static unsigned num_emulated_msrs;
1302 * List of msr numbers which are used to expose MSR-based features that
1303 * can be used by a hypervisor to validate requested CPU features.
1305 static const u32 msr_based_features_all[] = {
1307 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1308 MSR_IA32_VMX_PINBASED_CTLS,
1309 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1310 MSR_IA32_VMX_PROCBASED_CTLS,
1311 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1312 MSR_IA32_VMX_EXIT_CTLS,
1313 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1314 MSR_IA32_VMX_ENTRY_CTLS,
1316 MSR_IA32_VMX_CR0_FIXED0,
1317 MSR_IA32_VMX_CR0_FIXED1,
1318 MSR_IA32_VMX_CR4_FIXED0,
1319 MSR_IA32_VMX_CR4_FIXED1,
1320 MSR_IA32_VMX_VMCS_ENUM,
1321 MSR_IA32_VMX_PROCBASED_CTLS2,
1322 MSR_IA32_VMX_EPT_VPID_CAP,
1323 MSR_IA32_VMX_VMFUNC,
1327 MSR_IA32_ARCH_CAPABILITIES,
1328 MSR_IA32_PERF_CAPABILITIES,
1331 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1332 static unsigned int num_msr_based_features;
1334 static u64 kvm_get_arch_capabilities(void)
1338 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1339 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1342 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1343 * the nested hypervisor runs with NX huge pages. If it is not,
1344 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1345 * L1 guests, so it need not worry about its own (L2) guests.
1347 data |= ARCH_CAP_PSCHANGE_MC_NO;
1350 * If we're doing cache flushes (either "always" or "cond")
1351 * we will do one whenever the guest does a vmlaunch/vmresume.
1352 * If an outer hypervisor is doing the cache flush for us
1353 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1354 * capability to the guest too, and if EPT is disabled we're not
1355 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1356 * require a nested hypervisor to do a flush of its own.
1358 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1359 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1361 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1362 data |= ARCH_CAP_RDCL_NO;
1363 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1364 data |= ARCH_CAP_SSB_NO;
1365 if (!boot_cpu_has_bug(X86_BUG_MDS))
1366 data |= ARCH_CAP_MDS_NO;
1369 * On TAA affected systems:
1370 * - nothing to do if TSX is disabled on the host.
1371 * - we emulate TSX_CTRL if present on the host.
1372 * This lets the guest use VERW to clear CPU buffers.
1374 if (!boot_cpu_has(X86_FEATURE_RTM))
1375 data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1376 else if (!boot_cpu_has_bug(X86_BUG_TAA))
1377 data |= ARCH_CAP_TAA_NO;
1382 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1384 switch (msr->index) {
1385 case MSR_IA32_ARCH_CAPABILITIES:
1386 msr->data = kvm_get_arch_capabilities();
1388 case MSR_IA32_UCODE_REV:
1389 rdmsrl_safe(msr->index, &msr->data);
1392 if (kvm_x86_ops.get_msr_feature(msr))
1398 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1400 struct kvm_msr_entry msr;
1404 r = kvm_get_msr_feature(&msr);
1413 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1415 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1418 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1421 if (efer & (EFER_LME | EFER_LMA) &&
1422 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1425 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1431 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1433 if (efer & efer_reserved_bits)
1436 return __kvm_valid_efer(vcpu, efer);
1438 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1440 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1442 u64 old_efer = vcpu->arch.efer;
1443 u64 efer = msr_info->data;
1445 if (efer & efer_reserved_bits)
1448 if (!msr_info->host_initiated) {
1449 if (!__kvm_valid_efer(vcpu, efer))
1452 if (is_paging(vcpu) &&
1453 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1458 efer |= vcpu->arch.efer & EFER_LMA;
1460 kvm_x86_ops.set_efer(vcpu, efer);
1462 /* Update reserved bits */
1463 if ((efer ^ old_efer) & EFER_NX)
1464 kvm_mmu_reset_context(vcpu);
1469 void kvm_enable_efer_bits(u64 mask)
1471 efer_reserved_bits &= ~mask;
1473 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1476 * Write @data into the MSR specified by @index. Select MSR specific fault
1477 * checks are bypassed if @host_initiated is %true.
1478 * Returns 0 on success, non-0 otherwise.
1479 * Assumes vcpu_load() was already called.
1481 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1482 bool host_initiated)
1484 struct msr_data msr;
1489 case MSR_KERNEL_GS_BASE:
1492 if (is_noncanonical_address(data, vcpu))
1495 case MSR_IA32_SYSENTER_EIP:
1496 case MSR_IA32_SYSENTER_ESP:
1498 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1499 * non-canonical address is written on Intel but not on
1500 * AMD (which ignores the top 32-bits, because it does
1501 * not implement 64-bit SYSENTER).
1503 * 64-bit code should hence be able to write a non-canonical
1504 * value on AMD. Making the address canonical ensures that
1505 * vmentry does not fail on Intel after writing a non-canonical
1506 * value, and that something deterministic happens if the guest
1507 * invokes 64-bit SYSENTER.
1509 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1514 msr.host_initiated = host_initiated;
1516 return kvm_x86_ops.set_msr(vcpu, &msr);
1520 * Read the MSR specified by @index into @data. Select MSR specific fault
1521 * checks are bypassed if @host_initiated is %true.
1522 * Returns 0 on success, non-0 otherwise.
1523 * Assumes vcpu_load() was already called.
1525 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1526 bool host_initiated)
1528 struct msr_data msr;
1532 msr.host_initiated = host_initiated;
1534 ret = kvm_x86_ops.get_msr(vcpu, &msr);
1540 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1542 return __kvm_get_msr(vcpu, index, data, false);
1544 EXPORT_SYMBOL_GPL(kvm_get_msr);
1546 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1548 return __kvm_set_msr(vcpu, index, data, false);
1550 EXPORT_SYMBOL_GPL(kvm_set_msr);
1552 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1554 u32 ecx = kvm_rcx_read(vcpu);
1557 if (kvm_get_msr(vcpu, ecx, &data)) {
1558 trace_kvm_msr_read_ex(ecx);
1559 kvm_inject_gp(vcpu, 0);
1563 trace_kvm_msr_read(ecx, data);
1565 kvm_rax_write(vcpu, data & -1u);
1566 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1567 return kvm_skip_emulated_instruction(vcpu);
1569 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1571 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1573 u32 ecx = kvm_rcx_read(vcpu);
1574 u64 data = kvm_read_edx_eax(vcpu);
1576 if (kvm_set_msr(vcpu, ecx, data)) {
1577 trace_kvm_msr_write_ex(ecx, data);
1578 kvm_inject_gp(vcpu, 0);
1582 trace_kvm_msr_write(ecx, data);
1583 return kvm_skip_emulated_instruction(vcpu);
1585 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1587 bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1589 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1590 need_resched() || signal_pending(current);
1592 EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request);
1595 * The fast path for frequent and performance sensitive wrmsr emulation,
1596 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1597 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1598 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1599 * other cases which must be called after interrupts are enabled on the host.
1601 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1603 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1606 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1607 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1608 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1609 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1612 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1613 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1614 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1615 trace_kvm_apic_write(APIC_ICR, (u32)data);
1622 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1624 if (!kvm_can_use_hv_timer(vcpu))
1627 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1631 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1633 u32 msr = kvm_rcx_read(vcpu);
1635 fastpath_t ret = EXIT_FASTPATH_NONE;
1638 case APIC_BASE_MSR + (APIC_ICR >> 4):
1639 data = kvm_read_edx_eax(vcpu);
1640 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1641 kvm_skip_emulated_instruction(vcpu);
1642 ret = EXIT_FASTPATH_EXIT_HANDLED;
1645 case MSR_IA32_TSCDEADLINE:
1646 data = kvm_read_edx_eax(vcpu);
1647 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1648 kvm_skip_emulated_instruction(vcpu);
1649 ret = EXIT_FASTPATH_REENTER_GUEST;
1656 if (ret != EXIT_FASTPATH_NONE)
1657 trace_kvm_msr_write(msr, data);
1661 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1664 * Adapt set_msr() to msr_io()'s calling convention
1666 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1668 return __kvm_get_msr(vcpu, index, data, true);
1671 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1673 return __kvm_set_msr(vcpu, index, *data, true);
1676 #ifdef CONFIG_X86_64
1677 struct pvclock_clock {
1687 struct pvclock_gtod_data {
1690 struct pvclock_clock clock; /* extract of a clocksource struct */
1691 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1697 static struct pvclock_gtod_data pvclock_gtod_data;
1699 static void update_pvclock_gtod(struct timekeeper *tk)
1701 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1703 write_seqcount_begin(&vdata->seq);
1705 /* copy pvclock gtod data */
1706 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1707 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1708 vdata->clock.mask = tk->tkr_mono.mask;
1709 vdata->clock.mult = tk->tkr_mono.mult;
1710 vdata->clock.shift = tk->tkr_mono.shift;
1711 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1712 vdata->clock.offset = tk->tkr_mono.base;
1714 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1715 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1716 vdata->raw_clock.mask = tk->tkr_raw.mask;
1717 vdata->raw_clock.mult = tk->tkr_raw.mult;
1718 vdata->raw_clock.shift = tk->tkr_raw.shift;
1719 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1720 vdata->raw_clock.offset = tk->tkr_raw.base;
1722 vdata->wall_time_sec = tk->xtime_sec;
1724 vdata->offs_boot = tk->offs_boot;
1726 write_seqcount_end(&vdata->seq);
1729 static s64 get_kvmclock_base_ns(void)
1731 /* Count up from boot time, but with the frequency of the raw clock. */
1732 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1735 static s64 get_kvmclock_base_ns(void)
1737 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1738 return ktime_get_boottime_ns();
1742 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1744 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1745 kvm_vcpu_kick(vcpu);
1748 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1752 struct pvclock_wall_clock wc;
1758 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1763 ++version; /* first time write, random junk */
1767 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1771 * The guest calculates current wall clock time by adding
1772 * system time (updated by kvm_guest_time_update below) to the
1773 * wall clock specified here. We do the reverse here.
1775 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1777 wc.nsec = do_div(wall_nsec, 1000000000);
1778 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1779 wc.version = version;
1781 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1784 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1787 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1789 do_shl32_div32(dividend, divisor);
1793 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1794 s8 *pshift, u32 *pmultiplier)
1802 scaled64 = scaled_hz;
1803 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1808 tps32 = (uint32_t)tps64;
1809 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1810 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1818 *pmultiplier = div_frac(scaled64, tps32);
1821 #ifdef CONFIG_X86_64
1822 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1825 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1826 static unsigned long max_tsc_khz;
1828 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1830 u64 v = (u64)khz * (1000000 + ppm);
1835 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1839 /* Guest TSC same frequency as host TSC? */
1841 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1845 /* TSC scaling supported? */
1846 if (!kvm_has_tsc_control) {
1847 if (user_tsc_khz > tsc_khz) {
1848 vcpu->arch.tsc_catchup = 1;
1849 vcpu->arch.tsc_always_catchup = 1;
1852 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1857 /* TSC scaling required - calculate ratio */
1858 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1859 user_tsc_khz, tsc_khz);
1861 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1862 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1867 vcpu->arch.tsc_scaling_ratio = ratio;
1871 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1873 u32 thresh_lo, thresh_hi;
1874 int use_scaling = 0;
1876 /* tsc_khz can be zero if TSC calibration fails */
1877 if (user_tsc_khz == 0) {
1878 /* set tsc_scaling_ratio to a safe value */
1879 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1883 /* Compute a scale to convert nanoseconds in TSC cycles */
1884 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1885 &vcpu->arch.virtual_tsc_shift,
1886 &vcpu->arch.virtual_tsc_mult);
1887 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1890 * Compute the variation in TSC rate which is acceptable
1891 * within the range of tolerance and decide if the
1892 * rate being applied is within that bounds of the hardware
1893 * rate. If so, no scaling or compensation need be done.
1895 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1896 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1897 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1898 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1901 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1904 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1906 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1907 vcpu->arch.virtual_tsc_mult,
1908 vcpu->arch.virtual_tsc_shift);
1909 tsc += vcpu->arch.this_tsc_write;
1913 static inline int gtod_is_based_on_tsc(int mode)
1915 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
1918 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1920 #ifdef CONFIG_X86_64
1922 struct kvm_arch *ka = &vcpu->kvm->arch;
1923 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1925 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1926 atomic_read(&vcpu->kvm->online_vcpus));
1929 * Once the masterclock is enabled, always perform request in
1930 * order to update it.
1932 * In order to enable masterclock, the host clocksource must be TSC
1933 * and the vcpus need to have matched TSCs. When that happens,
1934 * perform request to enable masterclock.
1936 if (ka->use_master_clock ||
1937 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1938 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1940 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1941 atomic_read(&vcpu->kvm->online_vcpus),
1942 ka->use_master_clock, gtod->clock.vclock_mode);
1946 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1948 u64 curr_offset = vcpu->arch.l1_tsc_offset;
1949 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1953 * Multiply tsc by a fixed point number represented by ratio.
1955 * The most significant 64-N bits (mult) of ratio represent the
1956 * integral part of the fixed point number; the remaining N bits
1957 * (frac) represent the fractional part, ie. ratio represents a fixed
1958 * point number (mult + frac * 2^(-N)).
1960 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1962 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1964 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1967 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1970 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1972 if (ratio != kvm_default_tsc_scaling_ratio)
1973 _tsc = __scale_tsc(ratio, tsc);
1977 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1979 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1983 tsc = kvm_scale_tsc(vcpu, rdtsc());
1985 return target_tsc - tsc;
1988 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1990 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1992 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1994 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1996 vcpu->arch.l1_tsc_offset = offset;
1997 vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
2000 static inline bool kvm_check_tsc_unstable(void)
2002 #ifdef CONFIG_X86_64
2004 * TSC is marked unstable when we're running on Hyper-V,
2005 * 'TSC page' clocksource is good.
2007 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2010 return check_tsc_unstable();
2013 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
2015 struct kvm *kvm = vcpu->kvm;
2016 u64 offset, ns, elapsed;
2017 unsigned long flags;
2019 bool already_matched;
2020 u64 data = msr->data;
2021 bool synchronizing = false;
2023 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2024 offset = kvm_compute_tsc_offset(vcpu, data);
2025 ns = get_kvmclock_base_ns();
2026 elapsed = ns - kvm->arch.last_tsc_nsec;
2028 if (vcpu->arch.virtual_tsc_khz) {
2029 if (data == 0 && msr->host_initiated) {
2031 * detection of vcpu initialization -- need to sync
2032 * with other vCPUs. This particularly helps to keep
2033 * kvm_clock stable after CPU hotplug
2035 synchronizing = true;
2037 u64 tsc_exp = kvm->arch.last_tsc_write +
2038 nsec_to_cycles(vcpu, elapsed);
2039 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2041 * Special case: TSC write with a small delta (1 second)
2042 * of virtual cycle time against real time is
2043 * interpreted as an attempt to synchronize the CPU.
2045 synchronizing = data < tsc_exp + tsc_hz &&
2046 data + tsc_hz > tsc_exp;
2051 * For a reliable TSC, we can match TSC offsets, and for an unstable
2052 * TSC, we add elapsed time in this computation. We could let the
2053 * compensation code attempt to catch up if we fall behind, but
2054 * it's better to try to match offsets from the beginning.
2056 if (synchronizing &&
2057 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2058 if (!kvm_check_tsc_unstable()) {
2059 offset = kvm->arch.cur_tsc_offset;
2061 u64 delta = nsec_to_cycles(vcpu, elapsed);
2063 offset = kvm_compute_tsc_offset(vcpu, data);
2066 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2069 * We split periods of matched TSC writes into generations.
2070 * For each generation, we track the original measured
2071 * nanosecond time, offset, and write, so if TSCs are in
2072 * sync, we can match exact offset, and if not, we can match
2073 * exact software computation in compute_guest_tsc()
2075 * These values are tracked in kvm->arch.cur_xxx variables.
2077 kvm->arch.cur_tsc_generation++;
2078 kvm->arch.cur_tsc_nsec = ns;
2079 kvm->arch.cur_tsc_write = data;
2080 kvm->arch.cur_tsc_offset = offset;
2085 * We also track th most recent recorded KHZ, write and time to
2086 * allow the matching interval to be extended at each write.
2088 kvm->arch.last_tsc_nsec = ns;
2089 kvm->arch.last_tsc_write = data;
2090 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2092 vcpu->arch.last_guest_tsc = data;
2094 /* Keep track of which generation this VCPU has synchronized to */
2095 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2096 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2097 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2099 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
2100 update_ia32_tsc_adjust_msr(vcpu, offset);
2102 kvm_vcpu_write_tsc_offset(vcpu, offset);
2103 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2105 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2107 kvm->arch.nr_vcpus_matched_tsc = 0;
2108 } else if (!already_matched) {
2109 kvm->arch.nr_vcpus_matched_tsc++;
2112 kvm_track_tsc_matching(vcpu);
2113 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2116 EXPORT_SYMBOL_GPL(kvm_write_tsc);
2118 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2121 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2122 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2125 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2127 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2128 WARN_ON(adjustment < 0);
2129 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2130 adjust_tsc_offset_guest(vcpu, adjustment);
2133 #ifdef CONFIG_X86_64
2135 static u64 read_tsc(void)
2137 u64 ret = (u64)rdtsc_ordered();
2138 u64 last = pvclock_gtod_data.clock.cycle_last;
2140 if (likely(ret >= last))
2144 * GCC likes to generate cmov here, but this branch is extremely
2145 * predictable (it's just a function of time and the likely is
2146 * very likely) and there's a data dependence, so force GCC
2147 * to generate a branch instead. I don't barrier() because
2148 * we don't actually need a barrier, and if this function
2149 * ever gets inlined it will generate worse code.
2155 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2161 switch (clock->vclock_mode) {
2162 case VDSO_CLOCKMODE_HVCLOCK:
2163 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2165 if (tsc_pg_val != U64_MAX) {
2166 /* TSC page valid */
2167 *mode = VDSO_CLOCKMODE_HVCLOCK;
2168 v = (tsc_pg_val - clock->cycle_last) &
2171 /* TSC page invalid */
2172 *mode = VDSO_CLOCKMODE_NONE;
2175 case VDSO_CLOCKMODE_TSC:
2176 *mode = VDSO_CLOCKMODE_TSC;
2177 *tsc_timestamp = read_tsc();
2178 v = (*tsc_timestamp - clock->cycle_last) &
2182 *mode = VDSO_CLOCKMODE_NONE;
2185 if (*mode == VDSO_CLOCKMODE_NONE)
2186 *tsc_timestamp = v = 0;
2188 return v * clock->mult;
2191 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2193 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2199 seq = read_seqcount_begin(>od->seq);
2200 ns = gtod->raw_clock.base_cycles;
2201 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2202 ns >>= gtod->raw_clock.shift;
2203 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2204 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2210 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2212 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2218 seq = read_seqcount_begin(>od->seq);
2219 ts->tv_sec = gtod->wall_time_sec;
2220 ns = gtod->clock.base_cycles;
2221 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2222 ns >>= gtod->clock.shift;
2223 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2225 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2231 /* returns true if host is using TSC based clocksource */
2232 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2234 /* checked again under seqlock below */
2235 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2238 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2242 /* returns true if host is using TSC based clocksource */
2243 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2246 /* checked again under seqlock below */
2247 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2250 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2256 * Assuming a stable TSC across physical CPUS, and a stable TSC
2257 * across virtual CPUs, the following condition is possible.
2258 * Each numbered line represents an event visible to both
2259 * CPUs at the next numbered event.
2261 * "timespecX" represents host monotonic time. "tscX" represents
2264 * VCPU0 on CPU0 | VCPU1 on CPU1
2266 * 1. read timespec0,tsc0
2267 * 2. | timespec1 = timespec0 + N
2269 * 3. transition to guest | transition to guest
2270 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2271 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2272 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2274 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2277 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2279 * - 0 < N - M => M < N
2281 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2282 * always the case (the difference between two distinct xtime instances
2283 * might be smaller then the difference between corresponding TSC reads,
2284 * when updating guest vcpus pvclock areas).
2286 * To avoid that problem, do not allow visibility of distinct
2287 * system_timestamp/tsc_timestamp values simultaneously: use a master
2288 * copy of host monotonic time values. Update that master copy
2291 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2295 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2297 #ifdef CONFIG_X86_64
2298 struct kvm_arch *ka = &kvm->arch;
2300 bool host_tsc_clocksource, vcpus_matched;
2302 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2303 atomic_read(&kvm->online_vcpus));
2306 * If the host uses TSC clock, then passthrough TSC as stable
2309 host_tsc_clocksource = kvm_get_time_and_clockread(
2310 &ka->master_kernel_ns,
2311 &ka->master_cycle_now);
2313 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2314 && !ka->backwards_tsc_observed
2315 && !ka->boot_vcpu_runs_old_kvmclock;
2317 if (ka->use_master_clock)
2318 atomic_set(&kvm_guest_has_master_clock, 1);
2320 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2321 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2326 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2328 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2331 static void kvm_gen_update_masterclock(struct kvm *kvm)
2333 #ifdef CONFIG_X86_64
2335 struct kvm_vcpu *vcpu;
2336 struct kvm_arch *ka = &kvm->arch;
2338 spin_lock(&ka->pvclock_gtod_sync_lock);
2339 kvm_make_mclock_inprogress_request(kvm);
2340 /* no guest entries from this point */
2341 pvclock_update_vm_gtod_copy(kvm);
2343 kvm_for_each_vcpu(i, vcpu, kvm)
2344 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2346 /* guest entries allowed */
2347 kvm_for_each_vcpu(i, vcpu, kvm)
2348 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2350 spin_unlock(&ka->pvclock_gtod_sync_lock);
2354 u64 get_kvmclock_ns(struct kvm *kvm)
2356 struct kvm_arch *ka = &kvm->arch;
2357 struct pvclock_vcpu_time_info hv_clock;
2360 spin_lock(&ka->pvclock_gtod_sync_lock);
2361 if (!ka->use_master_clock) {
2362 spin_unlock(&ka->pvclock_gtod_sync_lock);
2363 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2366 hv_clock.tsc_timestamp = ka->master_cycle_now;
2367 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2368 spin_unlock(&ka->pvclock_gtod_sync_lock);
2370 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2373 if (__this_cpu_read(cpu_tsc_khz)) {
2374 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2375 &hv_clock.tsc_shift,
2376 &hv_clock.tsc_to_system_mul);
2377 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2379 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2386 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2388 struct kvm_vcpu_arch *vcpu = &v->arch;
2389 struct pvclock_vcpu_time_info guest_hv_clock;
2391 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2392 &guest_hv_clock, sizeof(guest_hv_clock))))
2395 /* This VCPU is paused, but it's legal for a guest to read another
2396 * VCPU's kvmclock, so we really have to follow the specification where
2397 * it says that version is odd if data is being modified, and even after
2400 * Version field updates must be kept separate. This is because
2401 * kvm_write_guest_cached might use a "rep movs" instruction, and
2402 * writes within a string instruction are weakly ordered. So there
2403 * are three writes overall.
2405 * As a small optimization, only write the version field in the first
2406 * and third write. The vcpu->pv_time cache is still valid, because the
2407 * version field is the first in the struct.
2409 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2411 if (guest_hv_clock.version & 1)
2412 ++guest_hv_clock.version; /* first time write, random junk */
2414 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2415 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2417 sizeof(vcpu->hv_clock.version));
2421 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2422 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2424 if (vcpu->pvclock_set_guest_stopped_request) {
2425 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2426 vcpu->pvclock_set_guest_stopped_request = false;
2429 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2431 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2433 sizeof(vcpu->hv_clock));
2437 vcpu->hv_clock.version++;
2438 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2440 sizeof(vcpu->hv_clock.version));
2443 static int kvm_guest_time_update(struct kvm_vcpu *v)
2445 unsigned long flags, tgt_tsc_khz;
2446 struct kvm_vcpu_arch *vcpu = &v->arch;
2447 struct kvm_arch *ka = &v->kvm->arch;
2449 u64 tsc_timestamp, host_tsc;
2451 bool use_master_clock;
2457 * If the host uses TSC clock, then passthrough TSC as stable
2460 spin_lock(&ka->pvclock_gtod_sync_lock);
2461 use_master_clock = ka->use_master_clock;
2462 if (use_master_clock) {
2463 host_tsc = ka->master_cycle_now;
2464 kernel_ns = ka->master_kernel_ns;
2466 spin_unlock(&ka->pvclock_gtod_sync_lock);
2468 /* Keep irq disabled to prevent changes to the clock */
2469 local_irq_save(flags);
2470 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2471 if (unlikely(tgt_tsc_khz == 0)) {
2472 local_irq_restore(flags);
2473 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2476 if (!use_master_clock) {
2478 kernel_ns = get_kvmclock_base_ns();
2481 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2484 * We may have to catch up the TSC to match elapsed wall clock
2485 * time for two reasons, even if kvmclock is used.
2486 * 1) CPU could have been running below the maximum TSC rate
2487 * 2) Broken TSC compensation resets the base at each VCPU
2488 * entry to avoid unknown leaps of TSC even when running
2489 * again on the same CPU. This may cause apparent elapsed
2490 * time to disappear, and the guest to stand still or run
2493 if (vcpu->tsc_catchup) {
2494 u64 tsc = compute_guest_tsc(v, kernel_ns);
2495 if (tsc > tsc_timestamp) {
2496 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2497 tsc_timestamp = tsc;
2501 local_irq_restore(flags);
2503 /* With all the info we got, fill in the values */
2505 if (kvm_has_tsc_control)
2506 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2508 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2509 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2510 &vcpu->hv_clock.tsc_shift,
2511 &vcpu->hv_clock.tsc_to_system_mul);
2512 vcpu->hw_tsc_khz = tgt_tsc_khz;
2515 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2516 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2517 vcpu->last_guest_tsc = tsc_timestamp;
2519 /* If the host uses TSC clocksource, then it is stable */
2521 if (use_master_clock)
2522 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2524 vcpu->hv_clock.flags = pvclock_flags;
2526 if (vcpu->pv_time_enabled)
2527 kvm_setup_pvclock_page(v);
2528 if (v == kvm_get_vcpu(v->kvm, 0))
2529 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2534 * kvmclock updates which are isolated to a given vcpu, such as
2535 * vcpu->cpu migration, should not allow system_timestamp from
2536 * the rest of the vcpus to remain static. Otherwise ntp frequency
2537 * correction applies to one vcpu's system_timestamp but not
2540 * So in those cases, request a kvmclock update for all vcpus.
2541 * We need to rate-limit these requests though, as they can
2542 * considerably slow guests that have a large number of vcpus.
2543 * The time for a remote vcpu to update its kvmclock is bound
2544 * by the delay we use to rate-limit the updates.
2547 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2549 static void kvmclock_update_fn(struct work_struct *work)
2552 struct delayed_work *dwork = to_delayed_work(work);
2553 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2554 kvmclock_update_work);
2555 struct kvm *kvm = container_of(ka, struct kvm, arch);
2556 struct kvm_vcpu *vcpu;
2558 kvm_for_each_vcpu(i, vcpu, kvm) {
2559 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2560 kvm_vcpu_kick(vcpu);
2564 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2566 struct kvm *kvm = v->kvm;
2568 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2569 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2570 KVMCLOCK_UPDATE_DELAY);
2573 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2575 static void kvmclock_sync_fn(struct work_struct *work)
2577 struct delayed_work *dwork = to_delayed_work(work);
2578 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2579 kvmclock_sync_work);
2580 struct kvm *kvm = container_of(ka, struct kvm, arch);
2582 if (!kvmclock_periodic_sync)
2585 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2586 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2587 KVMCLOCK_SYNC_PERIOD);
2591 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2593 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2595 /* McStatusWrEn enabled? */
2596 if (guest_cpuid_is_amd_or_hygon(vcpu))
2597 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2602 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2604 u64 mcg_cap = vcpu->arch.mcg_cap;
2605 unsigned bank_num = mcg_cap & 0xff;
2606 u32 msr = msr_info->index;
2607 u64 data = msr_info->data;
2610 case MSR_IA32_MCG_STATUS:
2611 vcpu->arch.mcg_status = data;
2613 case MSR_IA32_MCG_CTL:
2614 if (!(mcg_cap & MCG_CTL_P) &&
2615 (data || !msr_info->host_initiated))
2617 if (data != 0 && data != ~(u64)0)
2619 vcpu->arch.mcg_ctl = data;
2622 if (msr >= MSR_IA32_MC0_CTL &&
2623 msr < MSR_IA32_MCx_CTL(bank_num)) {
2624 u32 offset = array_index_nospec(
2625 msr - MSR_IA32_MC0_CTL,
2626 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2628 /* only 0 or all 1s can be written to IA32_MCi_CTL
2629 * some Linux kernels though clear bit 10 in bank 4 to
2630 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2631 * this to avoid an uncatched #GP in the guest
2633 if ((offset & 0x3) == 0 &&
2634 data != 0 && (data | (1 << 10)) != ~(u64)0)
2638 if (!msr_info->host_initiated &&
2639 (offset & 0x3) == 1 && data != 0) {
2640 if (!can_set_mci_status(vcpu))
2644 vcpu->arch.mce_banks[offset] = data;
2652 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2654 struct kvm *kvm = vcpu->kvm;
2655 int lm = is_long_mode(vcpu);
2656 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2657 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2658 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2659 : kvm->arch.xen_hvm_config.blob_size_32;
2660 u32 page_num = data & ~PAGE_MASK;
2661 u64 page_addr = data & PAGE_MASK;
2666 if (page_num >= blob_size)
2669 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2674 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2683 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2685 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2687 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2690 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2692 gpa_t gpa = data & ~0x3f;
2694 /* Bits 4:5 are reserved, Should be zero */
2698 if (!lapic_in_kernel(vcpu))
2701 vcpu->arch.apf.msr_en_val = data;
2703 if (!kvm_pv_async_pf_enabled(vcpu)) {
2704 kvm_clear_async_pf_completion_queue(vcpu);
2705 kvm_async_pf_hash_reset(vcpu);
2709 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2713 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2714 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2716 kvm_async_pf_wakeup_all(vcpu);
2721 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2723 /* Bits 8-63 are reserved */
2727 if (!lapic_in_kernel(vcpu))
2730 vcpu->arch.apf.msr_int_val = data;
2732 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2737 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2739 vcpu->arch.pv_time_enabled = false;
2740 vcpu->arch.time = 0;
2743 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2745 ++vcpu->stat.tlb_flush;
2746 kvm_x86_ops.tlb_flush_all(vcpu);
2749 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2751 ++vcpu->stat.tlb_flush;
2752 kvm_x86_ops.tlb_flush_guest(vcpu);
2755 static void record_steal_time(struct kvm_vcpu *vcpu)
2757 struct kvm_host_map map;
2758 struct kvm_steal_time *st;
2760 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2763 /* -EAGAIN is returned in atomic context so we can just return. */
2764 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2765 &map, &vcpu->arch.st.cache, false))
2769 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2772 * Doing a TLB flush here, on the guest's behalf, can avoid
2775 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2776 st->preempted & KVM_VCPU_FLUSH_TLB);
2777 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2778 kvm_vcpu_flush_tlb_guest(vcpu);
2780 vcpu->arch.st.preempted = 0;
2782 if (st->version & 1)
2783 st->version += 1; /* first time write, random junk */
2789 st->steal += current->sched_info.run_delay -
2790 vcpu->arch.st.last_steal;
2791 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2797 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2800 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2803 u32 msr = msr_info->index;
2804 u64 data = msr_info->data;
2807 case MSR_AMD64_NB_CFG:
2808 case MSR_IA32_UCODE_WRITE:
2809 case MSR_VM_HSAVE_PA:
2810 case MSR_AMD64_PATCH_LOADER:
2811 case MSR_AMD64_BU_CFG2:
2812 case MSR_AMD64_DC_CFG:
2813 case MSR_F15H_EX_CFG:
2816 case MSR_IA32_UCODE_REV:
2817 if (msr_info->host_initiated)
2818 vcpu->arch.microcode_version = data;
2820 case MSR_IA32_ARCH_CAPABILITIES:
2821 if (!msr_info->host_initiated)
2823 vcpu->arch.arch_capabilities = data;
2826 return set_efer(vcpu, msr_info);
2828 data &= ~(u64)0x40; /* ignore flush filter disable */
2829 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2830 data &= ~(u64)0x8; /* ignore TLB cache disable */
2832 /* Handle McStatusWrEn */
2833 if (data == BIT_ULL(18)) {
2834 vcpu->arch.msr_hwcr = data;
2835 } else if (data != 0) {
2836 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2841 case MSR_FAM10H_MMIO_CONF_BASE:
2843 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2848 case MSR_IA32_DEBUGCTLMSR:
2850 /* We support the non-activated case already */
2852 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2853 /* Values other than LBR and BTF are vendor-specific,
2854 thus reserved and should throw a #GP */
2857 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2860 case 0x200 ... 0x2ff:
2861 return kvm_mtrr_set_msr(vcpu, msr, data);
2862 case MSR_IA32_APICBASE:
2863 return kvm_set_apic_base(vcpu, msr_info);
2864 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
2865 return kvm_x2apic_msr_write(vcpu, msr, data);
2866 case MSR_IA32_TSCDEADLINE:
2867 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2869 case MSR_IA32_TSC_ADJUST:
2870 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2871 if (!msr_info->host_initiated) {
2872 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2873 adjust_tsc_offset_guest(vcpu, adj);
2875 vcpu->arch.ia32_tsc_adjust_msr = data;
2878 case MSR_IA32_MISC_ENABLE:
2879 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2880 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2881 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2883 vcpu->arch.ia32_misc_enable_msr = data;
2884 kvm_update_cpuid(vcpu);
2886 vcpu->arch.ia32_misc_enable_msr = data;
2889 case MSR_IA32_SMBASE:
2890 if (!msr_info->host_initiated)
2892 vcpu->arch.smbase = data;
2894 case MSR_IA32_POWER_CTL:
2895 vcpu->arch.msr_ia32_power_ctl = data;
2898 kvm_write_tsc(vcpu, msr_info);
2901 if (!msr_info->host_initiated &&
2902 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2905 * KVM supports exposing PT to the guest, but does not support
2906 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
2907 * XSAVES/XRSTORS to save/restore PT MSRs.
2909 if (data & ~supported_xss)
2911 vcpu->arch.ia32_xss = data;
2914 if (!msr_info->host_initiated)
2916 vcpu->arch.smi_count = data;
2918 case MSR_KVM_WALL_CLOCK_NEW:
2919 case MSR_KVM_WALL_CLOCK:
2920 vcpu->kvm->arch.wall_clock = data;
2921 kvm_write_wall_clock(vcpu->kvm, data);
2923 case MSR_KVM_SYSTEM_TIME_NEW:
2924 case MSR_KVM_SYSTEM_TIME: {
2925 struct kvm_arch *ka = &vcpu->kvm->arch;
2927 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2928 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2930 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2931 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2933 ka->boot_vcpu_runs_old_kvmclock = tmp;
2936 vcpu->arch.time = data;
2937 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2939 /* we verify if the enable bit is set... */
2940 vcpu->arch.pv_time_enabled = false;
2944 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2945 &vcpu->arch.pv_time, data & ~1ULL,
2946 sizeof(struct pvclock_vcpu_time_info)))
2947 vcpu->arch.pv_time_enabled = true;
2951 case MSR_KVM_ASYNC_PF_EN:
2952 if (kvm_pv_enable_async_pf(vcpu, data))
2955 case MSR_KVM_ASYNC_PF_INT:
2956 if (kvm_pv_enable_async_pf_int(vcpu, data))
2959 case MSR_KVM_ASYNC_PF_ACK:
2961 vcpu->arch.apf.pageready_pending = false;
2962 kvm_check_async_pf_completion(vcpu);
2965 case MSR_KVM_STEAL_TIME:
2967 if (unlikely(!sched_info_on()))
2970 if (data & KVM_STEAL_RESERVED_MASK)
2973 vcpu->arch.st.msr_val = data;
2975 if (!(data & KVM_MSR_ENABLED))
2978 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2981 case MSR_KVM_PV_EOI_EN:
2982 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2986 case MSR_KVM_POLL_CONTROL:
2987 /* only enable bit supported */
2988 if (data & (-1ULL << 1))
2991 vcpu->arch.msr_kvm_poll_control = data;
2994 case MSR_IA32_MCG_CTL:
2995 case MSR_IA32_MCG_STATUS:
2996 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2997 return set_msr_mce(vcpu, msr_info);
2999 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3000 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3001 pr = true; /* fall through */
3002 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3003 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3004 if (kvm_pmu_is_valid_msr(vcpu, msr))
3005 return kvm_pmu_set_msr(vcpu, msr_info);
3007 if (pr || data != 0)
3008 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3009 "0x%x data 0x%llx\n", msr, data);
3011 case MSR_K7_CLK_CTL:
3013 * Ignore all writes to this no longer documented MSR.
3014 * Writes are only relevant for old K7 processors,
3015 * all pre-dating SVM, but a recommended workaround from
3016 * AMD for these chips. It is possible to specify the
3017 * affected processor models on the command line, hence
3018 * the need to ignore the workaround.
3021 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3022 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3023 case HV_X64_MSR_SYNDBG_OPTIONS:
3024 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3025 case HV_X64_MSR_CRASH_CTL:
3026 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3027 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3028 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3029 case HV_X64_MSR_TSC_EMULATION_STATUS:
3030 return kvm_hv_set_msr_common(vcpu, msr, data,
3031 msr_info->host_initiated);
3032 case MSR_IA32_BBL_CR_CTL3:
3033 /* Drop writes to this legacy MSR -- see rdmsr
3034 * counterpart for further detail.
3036 if (report_ignored_msrs)
3037 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3040 case MSR_AMD64_OSVW_ID_LENGTH:
3041 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3043 vcpu->arch.osvw.length = data;
3045 case MSR_AMD64_OSVW_STATUS:
3046 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3048 vcpu->arch.osvw.status = data;
3050 case MSR_PLATFORM_INFO:
3051 if (!msr_info->host_initiated ||
3052 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3053 cpuid_fault_enabled(vcpu)))
3055 vcpu->arch.msr_platform_info = data;
3057 case MSR_MISC_FEATURES_ENABLES:
3058 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3059 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3060 !supports_cpuid_fault(vcpu)))
3062 vcpu->arch.msr_misc_features_enables = data;
3065 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
3066 return xen_hvm_config(vcpu, data);
3067 if (kvm_pmu_is_valid_msr(vcpu, msr))
3068 return kvm_pmu_set_msr(vcpu, msr_info);
3070 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
3074 if (report_ignored_msrs)
3076 "ignored wrmsr: 0x%x data 0x%llx\n",
3083 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3085 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3088 u64 mcg_cap = vcpu->arch.mcg_cap;
3089 unsigned bank_num = mcg_cap & 0xff;
3092 case MSR_IA32_P5_MC_ADDR:
3093 case MSR_IA32_P5_MC_TYPE:
3096 case MSR_IA32_MCG_CAP:
3097 data = vcpu->arch.mcg_cap;
3099 case MSR_IA32_MCG_CTL:
3100 if (!(mcg_cap & MCG_CTL_P) && !host)
3102 data = vcpu->arch.mcg_ctl;
3104 case MSR_IA32_MCG_STATUS:
3105 data = vcpu->arch.mcg_status;
3108 if (msr >= MSR_IA32_MC0_CTL &&
3109 msr < MSR_IA32_MCx_CTL(bank_num)) {
3110 u32 offset = array_index_nospec(
3111 msr - MSR_IA32_MC0_CTL,
3112 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3114 data = vcpu->arch.mce_banks[offset];
3123 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3125 switch (msr_info->index) {
3126 case MSR_IA32_PLATFORM_ID:
3127 case MSR_IA32_EBL_CR_POWERON:
3128 case MSR_IA32_DEBUGCTLMSR:
3129 case MSR_IA32_LASTBRANCHFROMIP:
3130 case MSR_IA32_LASTBRANCHTOIP:
3131 case MSR_IA32_LASTINTFROMIP:
3132 case MSR_IA32_LASTINTTOIP:
3134 case MSR_K8_TSEG_ADDR:
3135 case MSR_K8_TSEG_MASK:
3136 case MSR_VM_HSAVE_PA:
3137 case MSR_K8_INT_PENDING_MSG:
3138 case MSR_AMD64_NB_CFG:
3139 case MSR_FAM10H_MMIO_CONF_BASE:
3140 case MSR_AMD64_BU_CFG2:
3141 case MSR_IA32_PERF_CTL:
3142 case MSR_AMD64_DC_CFG:
3143 case MSR_F15H_EX_CFG:
3145 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3146 * limit) MSRs. Just return 0, as we do not want to expose the host
3147 * data here. Do not conditionalize this on CPUID, as KVM does not do
3148 * so for existing CPU-specific MSRs.
3150 case MSR_RAPL_POWER_UNIT:
3151 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3152 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3153 case MSR_PKG_ENERGY_STATUS: /* Total package */
3154 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3157 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3158 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3159 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3160 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3161 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3162 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3163 return kvm_pmu_get_msr(vcpu, msr_info);
3166 case MSR_IA32_UCODE_REV:
3167 msr_info->data = vcpu->arch.microcode_version;
3169 case MSR_IA32_ARCH_CAPABILITIES:
3170 if (!msr_info->host_initiated &&
3171 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3173 msr_info->data = vcpu->arch.arch_capabilities;
3175 case MSR_IA32_POWER_CTL:
3176 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3179 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
3182 case 0x200 ... 0x2ff:
3183 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3184 case 0xcd: /* fsb frequency */
3188 * MSR_EBC_FREQUENCY_ID
3189 * Conservative value valid for even the basic CPU models.
3190 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3191 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3192 * and 266MHz for model 3, or 4. Set Core Clock
3193 * Frequency to System Bus Frequency Ratio to 1 (bits
3194 * 31:24) even though these are only valid for CPU
3195 * models > 2, however guests may end up dividing or
3196 * multiplying by zero otherwise.
3198 case MSR_EBC_FREQUENCY_ID:
3199 msr_info->data = 1 << 24;
3201 case MSR_IA32_APICBASE:
3202 msr_info->data = kvm_get_apic_base(vcpu);
3204 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3205 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3206 case MSR_IA32_TSCDEADLINE:
3207 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3209 case MSR_IA32_TSC_ADJUST:
3210 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3212 case MSR_IA32_MISC_ENABLE:
3213 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3215 case MSR_IA32_SMBASE:
3216 if (!msr_info->host_initiated)
3218 msr_info->data = vcpu->arch.smbase;
3221 msr_info->data = vcpu->arch.smi_count;
3223 case MSR_IA32_PERF_STATUS:
3224 /* TSC increment by tick */
3225 msr_info->data = 1000ULL;
3226 /* CPU multiplier */
3227 msr_info->data |= (((uint64_t)4ULL) << 40);
3230 msr_info->data = vcpu->arch.efer;
3232 case MSR_KVM_WALL_CLOCK:
3233 case MSR_KVM_WALL_CLOCK_NEW:
3234 msr_info->data = vcpu->kvm->arch.wall_clock;
3236 case MSR_KVM_SYSTEM_TIME:
3237 case MSR_KVM_SYSTEM_TIME_NEW:
3238 msr_info->data = vcpu->arch.time;
3240 case MSR_KVM_ASYNC_PF_EN:
3241 msr_info->data = vcpu->arch.apf.msr_en_val;
3243 case MSR_KVM_ASYNC_PF_INT:
3244 msr_info->data = vcpu->arch.apf.msr_int_val;
3246 case MSR_KVM_ASYNC_PF_ACK:
3249 case MSR_KVM_STEAL_TIME:
3250 msr_info->data = vcpu->arch.st.msr_val;
3252 case MSR_KVM_PV_EOI_EN:
3253 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3255 case MSR_KVM_POLL_CONTROL:
3256 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3258 case MSR_IA32_P5_MC_ADDR:
3259 case MSR_IA32_P5_MC_TYPE:
3260 case MSR_IA32_MCG_CAP:
3261 case MSR_IA32_MCG_CTL:
3262 case MSR_IA32_MCG_STATUS:
3263 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3264 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3265 msr_info->host_initiated);
3267 if (!msr_info->host_initiated &&
3268 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3270 msr_info->data = vcpu->arch.ia32_xss;
3272 case MSR_K7_CLK_CTL:
3274 * Provide expected ramp-up count for K7. All other
3275 * are set to zero, indicating minimum divisors for
3278 * This prevents guest kernels on AMD host with CPU
3279 * type 6, model 8 and higher from exploding due to
3280 * the rdmsr failing.
3282 msr_info->data = 0x20000000;
3284 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3285 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3286 case HV_X64_MSR_SYNDBG_OPTIONS:
3287 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3288 case HV_X64_MSR_CRASH_CTL:
3289 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3290 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3291 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3292 case HV_X64_MSR_TSC_EMULATION_STATUS:
3293 return kvm_hv_get_msr_common(vcpu,
3294 msr_info->index, &msr_info->data,
3295 msr_info->host_initiated);
3296 case MSR_IA32_BBL_CR_CTL3:
3297 /* This legacy MSR exists but isn't fully documented in current
3298 * silicon. It is however accessed by winxp in very narrow
3299 * scenarios where it sets bit #19, itself documented as
3300 * a "reserved" bit. Best effort attempt to source coherent
3301 * read data here should the balance of the register be
3302 * interpreted by the guest:
3304 * L2 cache control register 3: 64GB range, 256KB size,
3305 * enabled, latency 0x1, configured
3307 msr_info->data = 0xbe702111;
3309 case MSR_AMD64_OSVW_ID_LENGTH:
3310 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3312 msr_info->data = vcpu->arch.osvw.length;
3314 case MSR_AMD64_OSVW_STATUS:
3315 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3317 msr_info->data = vcpu->arch.osvw.status;
3319 case MSR_PLATFORM_INFO:
3320 if (!msr_info->host_initiated &&
3321 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3323 msr_info->data = vcpu->arch.msr_platform_info;
3325 case MSR_MISC_FEATURES_ENABLES:
3326 msr_info->data = vcpu->arch.msr_misc_features_enables;
3329 msr_info->data = vcpu->arch.msr_hwcr;
3332 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3333 return kvm_pmu_get_msr(vcpu, msr_info);
3335 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3339 if (report_ignored_msrs)
3340 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3348 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3351 * Read or write a bunch of msrs. All parameters are kernel addresses.
3353 * @return number of msrs set successfully.
3355 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3356 struct kvm_msr_entry *entries,
3357 int (*do_msr)(struct kvm_vcpu *vcpu,
3358 unsigned index, u64 *data))
3362 for (i = 0; i < msrs->nmsrs; ++i)
3363 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3370 * Read or write a bunch of msrs. Parameters are user addresses.
3372 * @return number of msrs set successfully.
3374 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3375 int (*do_msr)(struct kvm_vcpu *vcpu,
3376 unsigned index, u64 *data),
3379 struct kvm_msrs msrs;
3380 struct kvm_msr_entry *entries;
3385 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3389 if (msrs.nmsrs >= MAX_IO_MSRS)
3392 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3393 entries = memdup_user(user_msrs->entries, size);
3394 if (IS_ERR(entries)) {
3395 r = PTR_ERR(entries);
3399 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3404 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3415 static inline bool kvm_can_mwait_in_guest(void)
3417 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3418 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3419 boot_cpu_has(X86_FEATURE_ARAT);
3422 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3427 case KVM_CAP_IRQCHIP:
3429 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3430 case KVM_CAP_SET_TSS_ADDR:
3431 case KVM_CAP_EXT_CPUID:
3432 case KVM_CAP_EXT_EMUL_CPUID:
3433 case KVM_CAP_CLOCKSOURCE:
3435 case KVM_CAP_NOP_IO_DELAY:
3436 case KVM_CAP_MP_STATE:
3437 case KVM_CAP_SYNC_MMU:
3438 case KVM_CAP_USER_NMI:
3439 case KVM_CAP_REINJECT_CONTROL:
3440 case KVM_CAP_IRQ_INJECT_STATUS:
3441 case KVM_CAP_IOEVENTFD:
3442 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3444 case KVM_CAP_PIT_STATE2:
3445 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3446 case KVM_CAP_XEN_HVM:
3447 case KVM_CAP_VCPU_EVENTS:
3448 case KVM_CAP_HYPERV:
3449 case KVM_CAP_HYPERV_VAPIC:
3450 case KVM_CAP_HYPERV_SPIN:
3451 case KVM_CAP_HYPERV_SYNIC:
3452 case KVM_CAP_HYPERV_SYNIC2:
3453 case KVM_CAP_HYPERV_VP_INDEX:
3454 case KVM_CAP_HYPERV_EVENTFD:
3455 case KVM_CAP_HYPERV_TLBFLUSH:
3456 case KVM_CAP_HYPERV_SEND_IPI:
3457 case KVM_CAP_HYPERV_CPUID:
3458 case KVM_CAP_PCI_SEGMENT:
3459 case KVM_CAP_DEBUGREGS:
3460 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3462 case KVM_CAP_ASYNC_PF:
3463 case KVM_CAP_ASYNC_PF_INT:
3464 case KVM_CAP_GET_TSC_KHZ:
3465 case KVM_CAP_KVMCLOCK_CTRL:
3466 case KVM_CAP_READONLY_MEM:
3467 case KVM_CAP_HYPERV_TIME:
3468 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3469 case KVM_CAP_TSC_DEADLINE_TIMER:
3470 case KVM_CAP_DISABLE_QUIRKS:
3471 case KVM_CAP_SET_BOOT_CPU_ID:
3472 case KVM_CAP_SPLIT_IRQCHIP:
3473 case KVM_CAP_IMMEDIATE_EXIT:
3474 case KVM_CAP_PMU_EVENT_FILTER:
3475 case KVM_CAP_GET_MSR_FEATURES:
3476 case KVM_CAP_MSR_PLATFORM_INFO:
3477 case KVM_CAP_EXCEPTION_PAYLOAD:
3478 case KVM_CAP_SET_GUEST_DEBUG:
3481 case KVM_CAP_SYNC_REGS:
3482 r = KVM_SYNC_X86_VALID_FIELDS;
3484 case KVM_CAP_ADJUST_CLOCK:
3485 r = KVM_CLOCK_TSC_STABLE;
3487 case KVM_CAP_X86_DISABLE_EXITS:
3488 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3489 KVM_X86_DISABLE_EXITS_CSTATE;
3490 if(kvm_can_mwait_in_guest())
3491 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3493 case KVM_CAP_X86_SMM:
3494 /* SMBASE is usually relocated above 1M on modern chipsets,
3495 * and SMM handlers might indeed rely on 4G segment limits,
3496 * so do not report SMM to be available if real mode is
3497 * emulated via vm86 mode. Still, do not go to great lengths
3498 * to avoid userspace's usage of the feature, because it is a
3499 * fringe case that is not enabled except via specific settings
3500 * of the module parameters.
3502 r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3505 r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3507 case KVM_CAP_NR_VCPUS:
3508 r = KVM_SOFT_MAX_VCPUS;
3510 case KVM_CAP_MAX_VCPUS:
3513 case KVM_CAP_MAX_VCPU_ID:
3514 r = KVM_MAX_VCPU_ID;
3516 case KVM_CAP_PV_MMU: /* obsolete */
3520 r = KVM_MAX_MCE_BANKS;
3523 r = boot_cpu_has(X86_FEATURE_XSAVE);
3525 case KVM_CAP_TSC_CONTROL:
3526 r = kvm_has_tsc_control;
3528 case KVM_CAP_X2APIC_API:
3529 r = KVM_X2APIC_API_VALID_FLAGS;
3531 case KVM_CAP_NESTED_STATE:
3532 r = kvm_x86_ops.nested_ops->get_state ?
3533 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3535 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3536 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3538 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3539 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3548 long kvm_arch_dev_ioctl(struct file *filp,
3549 unsigned int ioctl, unsigned long arg)
3551 void __user *argp = (void __user *)arg;
3555 case KVM_GET_MSR_INDEX_LIST: {
3556 struct kvm_msr_list __user *user_msr_list = argp;
3557 struct kvm_msr_list msr_list;
3561 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3564 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3565 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3568 if (n < msr_list.nmsrs)
3571 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3572 num_msrs_to_save * sizeof(u32)))
3574 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3576 num_emulated_msrs * sizeof(u32)))
3581 case KVM_GET_SUPPORTED_CPUID:
3582 case KVM_GET_EMULATED_CPUID: {
3583 struct kvm_cpuid2 __user *cpuid_arg = argp;
3584 struct kvm_cpuid2 cpuid;
3587 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3590 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3596 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3601 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3603 if (copy_to_user(argp, &kvm_mce_cap_supported,
3604 sizeof(kvm_mce_cap_supported)))
3608 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3609 struct kvm_msr_list __user *user_msr_list = argp;
3610 struct kvm_msr_list msr_list;
3614 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3617 msr_list.nmsrs = num_msr_based_features;
3618 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3621 if (n < msr_list.nmsrs)
3624 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3625 num_msr_based_features * sizeof(u32)))
3631 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3641 static void wbinvd_ipi(void *garbage)
3646 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3648 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3651 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3653 /* Address WBINVD may be executed by guest */
3654 if (need_emulate_wbinvd(vcpu)) {
3655 if (kvm_x86_ops.has_wbinvd_exit())
3656 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3657 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3658 smp_call_function_single(vcpu->cpu,
3659 wbinvd_ipi, NULL, 1);
3662 kvm_x86_ops.vcpu_load(vcpu, cpu);
3664 /* Save host pkru register if supported */
3665 vcpu->arch.host_pkru = read_pkru();
3667 /* Apply any externally detected TSC adjustments (due to suspend) */
3668 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3669 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3670 vcpu->arch.tsc_offset_adjustment = 0;
3671 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3674 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3675 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3676 rdtsc() - vcpu->arch.last_host_tsc;
3678 mark_tsc_unstable("KVM discovered backwards TSC");
3680 if (kvm_check_tsc_unstable()) {
3681 u64 offset = kvm_compute_tsc_offset(vcpu,
3682 vcpu->arch.last_guest_tsc);
3683 kvm_vcpu_write_tsc_offset(vcpu, offset);
3684 vcpu->arch.tsc_catchup = 1;
3687 if (kvm_lapic_hv_timer_in_use(vcpu))
3688 kvm_lapic_restart_hv_timer(vcpu);
3691 * On a host with synchronized TSC, there is no need to update
3692 * kvmclock on vcpu->cpu migration
3694 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3695 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3696 if (vcpu->cpu != cpu)
3697 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3701 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3704 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3706 struct kvm_host_map map;
3707 struct kvm_steal_time *st;
3709 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3712 if (vcpu->arch.st.preempted)
3715 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3716 &vcpu->arch.st.cache, true))
3720 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3722 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3724 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3727 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3731 if (vcpu->preempted)
3732 vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
3735 * Disable page faults because we're in atomic context here.
3736 * kvm_write_guest_offset_cached() would call might_fault()
3737 * that relies on pagefault_disable() to tell if there's a
3738 * bug. NOTE: the write to guest memory may not go through if
3739 * during postcopy live migration or if there's heavy guest
3742 pagefault_disable();
3744 * kvm_memslots() will be called by
3745 * kvm_write_guest_offset_cached() so take the srcu lock.
3747 idx = srcu_read_lock(&vcpu->kvm->srcu);
3748 kvm_steal_time_set_preempted(vcpu);
3749 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3751 kvm_x86_ops.vcpu_put(vcpu);
3752 vcpu->arch.last_host_tsc = rdtsc();
3754 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3755 * on every vmexit, but if not, we might have a stale dr6 from the
3756 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3761 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3762 struct kvm_lapic_state *s)
3764 if (vcpu->arch.apicv_active)
3765 kvm_x86_ops.sync_pir_to_irr(vcpu);
3767 return kvm_apic_get_state(vcpu, s);
3770 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3771 struct kvm_lapic_state *s)
3775 r = kvm_apic_set_state(vcpu, s);
3778 update_cr8_intercept(vcpu);
3783 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3785 return (!lapic_in_kernel(vcpu) ||
3786 kvm_apic_accept_pic_intr(vcpu));
3790 * if userspace requested an interrupt window, check that the
3791 * interrupt window is open.
3793 * No need to exit to userspace if we already have an interrupt queued.
3795 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3797 return kvm_arch_interrupt_allowed(vcpu) &&
3798 !kvm_cpu_has_interrupt(vcpu) &&
3799 !kvm_event_needs_reinjection(vcpu) &&
3800 kvm_cpu_accept_dm_intr(vcpu);
3803 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3804 struct kvm_interrupt *irq)
3806 if (irq->irq >= KVM_NR_INTERRUPTS)
3809 if (!irqchip_in_kernel(vcpu->kvm)) {
3810 kvm_queue_interrupt(vcpu, irq->irq, false);
3811 kvm_make_request(KVM_REQ_EVENT, vcpu);
3816 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3817 * fail for in-kernel 8259.
3819 if (pic_in_kernel(vcpu->kvm))
3822 if (vcpu->arch.pending_external_vector != -1)
3825 vcpu->arch.pending_external_vector = irq->irq;
3826 kvm_make_request(KVM_REQ_EVENT, vcpu);
3830 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3832 kvm_inject_nmi(vcpu);
3837 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3839 kvm_make_request(KVM_REQ_SMI, vcpu);
3844 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3845 struct kvm_tpr_access_ctl *tac)
3849 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3853 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3857 unsigned bank_num = mcg_cap & 0xff, bank;
3860 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
3862 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3865 vcpu->arch.mcg_cap = mcg_cap;
3866 /* Init IA32_MCG_CTL to all 1s */
3867 if (mcg_cap & MCG_CTL_P)
3868 vcpu->arch.mcg_ctl = ~(u64)0;
3869 /* Init IA32_MCi_CTL to all 1s */
3870 for (bank = 0; bank < bank_num; bank++)
3871 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3873 kvm_x86_ops.setup_mce(vcpu);
3878 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3879 struct kvm_x86_mce *mce)
3881 u64 mcg_cap = vcpu->arch.mcg_cap;
3882 unsigned bank_num = mcg_cap & 0xff;
3883 u64 *banks = vcpu->arch.mce_banks;
3885 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3888 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3889 * reporting is disabled
3891 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3892 vcpu->arch.mcg_ctl != ~(u64)0)
3894 banks += 4 * mce->bank;
3896 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3897 * reporting is disabled for the bank
3899 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3901 if (mce->status & MCI_STATUS_UC) {
3902 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3903 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3904 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3907 if (banks[1] & MCI_STATUS_VAL)
3908 mce->status |= MCI_STATUS_OVER;
3909 banks[2] = mce->addr;
3910 banks[3] = mce->misc;
3911 vcpu->arch.mcg_status = mce->mcg_status;
3912 banks[1] = mce->status;
3913 kvm_queue_exception(vcpu, MC_VECTOR);
3914 } else if (!(banks[1] & MCI_STATUS_VAL)
3915 || !(banks[1] & MCI_STATUS_UC)) {
3916 if (banks[1] & MCI_STATUS_VAL)
3917 mce->status |= MCI_STATUS_OVER;
3918 banks[2] = mce->addr;
3919 banks[3] = mce->misc;
3920 banks[1] = mce->status;
3922 banks[1] |= MCI_STATUS_OVER;
3926 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3927 struct kvm_vcpu_events *events)
3932 * In guest mode, payload delivery should be deferred,
3933 * so that the L1 hypervisor can intercept #PF before
3934 * CR2 is modified (or intercept #DB before DR6 is
3935 * modified under nVMX). Unless the per-VM capability,
3936 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
3937 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
3938 * opportunistically defer the exception payload, deliver it if the
3939 * capability hasn't been requested before processing a
3940 * KVM_GET_VCPU_EVENTS.
3942 if (!vcpu->kvm->arch.exception_payload_enabled &&
3943 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
3944 kvm_deliver_exception_payload(vcpu);
3947 * The API doesn't provide the instruction length for software
3948 * exceptions, so don't report them. As long as the guest RIP
3949 * isn't advanced, we should expect to encounter the exception
3952 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3953 events->exception.injected = 0;
3954 events->exception.pending = 0;
3956 events->exception.injected = vcpu->arch.exception.injected;
3957 events->exception.pending = vcpu->arch.exception.pending;
3959 * For ABI compatibility, deliberately conflate
3960 * pending and injected exceptions when
3961 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3963 if (!vcpu->kvm->arch.exception_payload_enabled)
3964 events->exception.injected |=
3965 vcpu->arch.exception.pending;
3967 events->exception.nr = vcpu->arch.exception.nr;
3968 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3969 events->exception.error_code = vcpu->arch.exception.error_code;
3970 events->exception_has_payload = vcpu->arch.exception.has_payload;
3971 events->exception_payload = vcpu->arch.exception.payload;
3973 events->interrupt.injected =
3974 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3975 events->interrupt.nr = vcpu->arch.interrupt.nr;
3976 events->interrupt.soft = 0;
3977 events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
3979 events->nmi.injected = vcpu->arch.nmi_injected;
3980 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3981 events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
3982 events->nmi.pad = 0;
3984 events->sipi_vector = 0; /* never valid when reporting to user space */
3986 events->smi.smm = is_smm(vcpu);
3987 events->smi.pending = vcpu->arch.smi_pending;
3988 events->smi.smm_inside_nmi =
3989 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3990 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3992 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3993 | KVM_VCPUEVENT_VALID_SHADOW
3994 | KVM_VCPUEVENT_VALID_SMM);
3995 if (vcpu->kvm->arch.exception_payload_enabled)
3996 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3998 memset(&events->reserved, 0, sizeof(events->reserved));
4001 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4003 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4004 struct kvm_vcpu_events *events)
4006 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4007 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4008 | KVM_VCPUEVENT_VALID_SHADOW
4009 | KVM_VCPUEVENT_VALID_SMM
4010 | KVM_VCPUEVENT_VALID_PAYLOAD))
4013 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4014 if (!vcpu->kvm->arch.exception_payload_enabled)
4016 if (events->exception.pending)
4017 events->exception.injected = 0;
4019 events->exception_has_payload = 0;
4021 events->exception.pending = 0;
4022 events->exception_has_payload = 0;
4025 if ((events->exception.injected || events->exception.pending) &&
4026 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4029 /* INITs are latched while in SMM */
4030 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4031 (events->smi.smm || events->smi.pending) &&
4032 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4036 vcpu->arch.exception.injected = events->exception.injected;
4037 vcpu->arch.exception.pending = events->exception.pending;
4038 vcpu->arch.exception.nr = events->exception.nr;
4039 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4040 vcpu->arch.exception.error_code = events->exception.error_code;
4041 vcpu->arch.exception.has_payload = events->exception_has_payload;
4042 vcpu->arch.exception.payload = events->exception_payload;
4044 vcpu->arch.interrupt.injected = events->interrupt.injected;
4045 vcpu->arch.interrupt.nr = events->interrupt.nr;
4046 vcpu->arch.interrupt.soft = events->interrupt.soft;
4047 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4048 kvm_x86_ops.set_interrupt_shadow(vcpu,
4049 events->interrupt.shadow);
4051 vcpu->arch.nmi_injected = events->nmi.injected;
4052 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4053 vcpu->arch.nmi_pending = events->nmi.pending;
4054 kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
4056 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4057 lapic_in_kernel(vcpu))
4058 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4060 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4061 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4062 if (events->smi.smm)
4063 vcpu->arch.hflags |= HF_SMM_MASK;
4065 vcpu->arch.hflags &= ~HF_SMM_MASK;
4066 kvm_smm_changed(vcpu);
4069 vcpu->arch.smi_pending = events->smi.pending;
4071 if (events->smi.smm) {
4072 if (events->smi.smm_inside_nmi)
4073 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4075 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4078 if (lapic_in_kernel(vcpu)) {
4079 if (events->smi.latched_init)
4080 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4082 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4086 kvm_make_request(KVM_REQ_EVENT, vcpu);
4091 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4092 struct kvm_debugregs *dbgregs)
4096 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4097 kvm_get_dr(vcpu, 6, &val);
4099 dbgregs->dr7 = vcpu->arch.dr7;
4101 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4104 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4105 struct kvm_debugregs *dbgregs)
4110 if (dbgregs->dr6 & ~0xffffffffull)
4112 if (dbgregs->dr7 & ~0xffffffffull)
4115 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4116 kvm_update_dr0123(vcpu);
4117 vcpu->arch.dr6 = dbgregs->dr6;
4118 vcpu->arch.dr7 = dbgregs->dr7;
4119 kvm_update_dr7(vcpu);
4124 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4126 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4128 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4129 u64 xstate_bv = xsave->header.xfeatures;
4133 * Copy legacy XSAVE area, to avoid complications with CPUID
4134 * leaves 0 and 1 in the loop below.
4136 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4139 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4140 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4143 * Copy each region from the possibly compacted offset to the
4144 * non-compacted offset.
4146 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4148 u64 xfeature_mask = valid & -valid;
4149 int xfeature_nr = fls64(xfeature_mask) - 1;
4150 void *src = get_xsave_addr(xsave, xfeature_nr);
4153 u32 size, offset, ecx, edx;
4154 cpuid_count(XSTATE_CPUID, xfeature_nr,
4155 &size, &offset, &ecx, &edx);
4156 if (xfeature_nr == XFEATURE_PKRU)
4157 memcpy(dest + offset, &vcpu->arch.pkru,
4158 sizeof(vcpu->arch.pkru));
4160 memcpy(dest + offset, src, size);
4164 valid -= xfeature_mask;
4168 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4170 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4171 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4175 * Copy legacy XSAVE area, to avoid complications with CPUID
4176 * leaves 0 and 1 in the loop below.
4178 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4180 /* Set XSTATE_BV and possibly XCOMP_BV. */
4181 xsave->header.xfeatures = xstate_bv;
4182 if (boot_cpu_has(X86_FEATURE_XSAVES))
4183 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4186 * Copy each region from the non-compacted offset to the
4187 * possibly compacted offset.
4189 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4191 u64 xfeature_mask = valid & -valid;
4192 int xfeature_nr = fls64(xfeature_mask) - 1;
4193 void *dest = get_xsave_addr(xsave, xfeature_nr);
4196 u32 size, offset, ecx, edx;
4197 cpuid_count(XSTATE_CPUID, xfeature_nr,
4198 &size, &offset, &ecx, &edx);
4199 if (xfeature_nr == XFEATURE_PKRU)
4200 memcpy(&vcpu->arch.pkru, src + offset,
4201 sizeof(vcpu->arch.pkru));
4203 memcpy(dest, src + offset, size);
4206 valid -= xfeature_mask;
4210 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4211 struct kvm_xsave *guest_xsave)
4213 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4214 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4215 fill_xsave((u8 *) guest_xsave->region, vcpu);
4217 memcpy(guest_xsave->region,
4218 &vcpu->arch.guest_fpu->state.fxsave,
4219 sizeof(struct fxregs_state));
4220 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4221 XFEATURE_MASK_FPSSE;
4225 #define XSAVE_MXCSR_OFFSET 24
4227 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4228 struct kvm_xsave *guest_xsave)
4231 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4232 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4234 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4236 * Here we allow setting states that are not present in
4237 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4238 * with old userspace.
4240 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4242 load_xsave(vcpu, (u8 *)guest_xsave->region);
4244 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4245 mxcsr & ~mxcsr_feature_mask)
4247 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4248 guest_xsave->region, sizeof(struct fxregs_state));
4253 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4254 struct kvm_xcrs *guest_xcrs)
4256 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4257 guest_xcrs->nr_xcrs = 0;
4261 guest_xcrs->nr_xcrs = 1;
4262 guest_xcrs->flags = 0;
4263 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4264 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4267 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4268 struct kvm_xcrs *guest_xcrs)
4272 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4275 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4278 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4279 /* Only support XCR0 currently */
4280 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4281 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4282 guest_xcrs->xcrs[i].value);
4291 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4292 * stopped by the hypervisor. This function will be called from the host only.
4293 * EINVAL is returned when the host attempts to set the flag for a guest that
4294 * does not support pv clocks.
4296 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4298 if (!vcpu->arch.pv_time_enabled)
4300 vcpu->arch.pvclock_set_guest_stopped_request = true;
4301 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4305 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4306 struct kvm_enable_cap *cap)
4309 uint16_t vmcs_version;
4310 void __user *user_ptr;
4316 case KVM_CAP_HYPERV_SYNIC2:
4321 case KVM_CAP_HYPERV_SYNIC:
4322 if (!irqchip_in_kernel(vcpu->kvm))
4324 return kvm_hv_activate_synic(vcpu, cap->cap ==
4325 KVM_CAP_HYPERV_SYNIC2);
4326 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4327 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4329 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4331 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4332 if (copy_to_user(user_ptr, &vmcs_version,
4333 sizeof(vmcs_version)))
4337 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4338 if (!kvm_x86_ops.enable_direct_tlbflush)
4341 return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4348 long kvm_arch_vcpu_ioctl(struct file *filp,
4349 unsigned int ioctl, unsigned long arg)
4351 struct kvm_vcpu *vcpu = filp->private_data;
4352 void __user *argp = (void __user *)arg;
4355 struct kvm_lapic_state *lapic;
4356 struct kvm_xsave *xsave;
4357 struct kvm_xcrs *xcrs;
4365 case KVM_GET_LAPIC: {
4367 if (!lapic_in_kernel(vcpu))
4369 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4370 GFP_KERNEL_ACCOUNT);
4375 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4379 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4384 case KVM_SET_LAPIC: {
4386 if (!lapic_in_kernel(vcpu))
4388 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4389 if (IS_ERR(u.lapic)) {
4390 r = PTR_ERR(u.lapic);
4394 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4397 case KVM_INTERRUPT: {
4398 struct kvm_interrupt irq;
4401 if (copy_from_user(&irq, argp, sizeof(irq)))
4403 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4407 r = kvm_vcpu_ioctl_nmi(vcpu);
4411 r = kvm_vcpu_ioctl_smi(vcpu);
4414 case KVM_SET_CPUID: {
4415 struct kvm_cpuid __user *cpuid_arg = argp;
4416 struct kvm_cpuid cpuid;
4419 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4421 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4424 case KVM_SET_CPUID2: {
4425 struct kvm_cpuid2 __user *cpuid_arg = argp;
4426 struct kvm_cpuid2 cpuid;
4429 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4431 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4432 cpuid_arg->entries);
4435 case KVM_GET_CPUID2: {
4436 struct kvm_cpuid2 __user *cpuid_arg = argp;
4437 struct kvm_cpuid2 cpuid;
4440 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4442 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4443 cpuid_arg->entries);
4447 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4452 case KVM_GET_MSRS: {
4453 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4454 r = msr_io(vcpu, argp, do_get_msr, 1);
4455 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4458 case KVM_SET_MSRS: {
4459 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4460 r = msr_io(vcpu, argp, do_set_msr, 0);
4461 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4464 case KVM_TPR_ACCESS_REPORTING: {
4465 struct kvm_tpr_access_ctl tac;
4468 if (copy_from_user(&tac, argp, sizeof(tac)))
4470 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4474 if (copy_to_user(argp, &tac, sizeof(tac)))
4479 case KVM_SET_VAPIC_ADDR: {
4480 struct kvm_vapic_addr va;
4484 if (!lapic_in_kernel(vcpu))
4487 if (copy_from_user(&va, argp, sizeof(va)))
4489 idx = srcu_read_lock(&vcpu->kvm->srcu);
4490 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4491 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4494 case KVM_X86_SETUP_MCE: {
4498 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4500 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4503 case KVM_X86_SET_MCE: {
4504 struct kvm_x86_mce mce;
4507 if (copy_from_user(&mce, argp, sizeof(mce)))
4509 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4512 case KVM_GET_VCPU_EVENTS: {
4513 struct kvm_vcpu_events events;
4515 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4518 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4523 case KVM_SET_VCPU_EVENTS: {
4524 struct kvm_vcpu_events events;
4527 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4530 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4533 case KVM_GET_DEBUGREGS: {
4534 struct kvm_debugregs dbgregs;
4536 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4539 if (copy_to_user(argp, &dbgregs,
4540 sizeof(struct kvm_debugregs)))
4545 case KVM_SET_DEBUGREGS: {
4546 struct kvm_debugregs dbgregs;
4549 if (copy_from_user(&dbgregs, argp,
4550 sizeof(struct kvm_debugregs)))
4553 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4556 case KVM_GET_XSAVE: {
4557 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4562 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4565 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4570 case KVM_SET_XSAVE: {
4571 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4572 if (IS_ERR(u.xsave)) {
4573 r = PTR_ERR(u.xsave);
4577 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4580 case KVM_GET_XCRS: {
4581 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4586 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4589 if (copy_to_user(argp, u.xcrs,
4590 sizeof(struct kvm_xcrs)))
4595 case KVM_SET_XCRS: {
4596 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4597 if (IS_ERR(u.xcrs)) {
4598 r = PTR_ERR(u.xcrs);
4602 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4605 case KVM_SET_TSC_KHZ: {
4609 user_tsc_khz = (u32)arg;
4611 if (kvm_has_tsc_control &&
4612 user_tsc_khz >= kvm_max_guest_tsc_khz)
4615 if (user_tsc_khz == 0)
4616 user_tsc_khz = tsc_khz;
4618 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4623 case KVM_GET_TSC_KHZ: {
4624 r = vcpu->arch.virtual_tsc_khz;
4627 case KVM_KVMCLOCK_CTRL: {
4628 r = kvm_set_guest_paused(vcpu);
4631 case KVM_ENABLE_CAP: {
4632 struct kvm_enable_cap cap;
4635 if (copy_from_user(&cap, argp, sizeof(cap)))
4637 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4640 case KVM_GET_NESTED_STATE: {
4641 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4645 if (!kvm_x86_ops.nested_ops->get_state)
4648 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4650 if (get_user(user_data_size, &user_kvm_nested_state->size))
4653 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4658 if (r > user_data_size) {
4659 if (put_user(r, &user_kvm_nested_state->size))
4669 case KVM_SET_NESTED_STATE: {
4670 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4671 struct kvm_nested_state kvm_state;
4675 if (!kvm_x86_ops.nested_ops->set_state)
4679 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4683 if (kvm_state.size < sizeof(kvm_state))
4686 if (kvm_state.flags &
4687 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4688 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
4689 | KVM_STATE_NESTED_GIF_SET))
4692 /* nested_run_pending implies guest_mode. */
4693 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4694 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4697 idx = srcu_read_lock(&vcpu->kvm->srcu);
4698 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
4699 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4702 case KVM_GET_SUPPORTED_HV_CPUID: {
4703 struct kvm_cpuid2 __user *cpuid_arg = argp;
4704 struct kvm_cpuid2 cpuid;
4707 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4710 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4711 cpuid_arg->entries);
4716 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4731 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4733 return VM_FAULT_SIGBUS;
4736 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4740 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4742 ret = kvm_x86_ops.set_tss_addr(kvm, addr);
4746 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4749 return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
4752 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4753 unsigned long kvm_nr_mmu_pages)
4755 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4758 mutex_lock(&kvm->slots_lock);
4760 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4761 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4763 mutex_unlock(&kvm->slots_lock);
4767 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4769 return kvm->arch.n_max_mmu_pages;
4772 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4774 struct kvm_pic *pic = kvm->arch.vpic;
4778 switch (chip->chip_id) {
4779 case KVM_IRQCHIP_PIC_MASTER:
4780 memcpy(&chip->chip.pic, &pic->pics[0],
4781 sizeof(struct kvm_pic_state));
4783 case KVM_IRQCHIP_PIC_SLAVE:
4784 memcpy(&chip->chip.pic, &pic->pics[1],
4785 sizeof(struct kvm_pic_state));
4787 case KVM_IRQCHIP_IOAPIC:
4788 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4797 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4799 struct kvm_pic *pic = kvm->arch.vpic;
4803 switch (chip->chip_id) {
4804 case KVM_IRQCHIP_PIC_MASTER:
4805 spin_lock(&pic->lock);
4806 memcpy(&pic->pics[0], &chip->chip.pic,
4807 sizeof(struct kvm_pic_state));
4808 spin_unlock(&pic->lock);
4810 case KVM_IRQCHIP_PIC_SLAVE:
4811 spin_lock(&pic->lock);
4812 memcpy(&pic->pics[1], &chip->chip.pic,
4813 sizeof(struct kvm_pic_state));
4814 spin_unlock(&pic->lock);
4816 case KVM_IRQCHIP_IOAPIC:
4817 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4823 kvm_pic_update_irq(pic);
4827 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4829 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4831 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4833 mutex_lock(&kps->lock);
4834 memcpy(ps, &kps->channels, sizeof(*ps));
4835 mutex_unlock(&kps->lock);
4839 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4842 struct kvm_pit *pit = kvm->arch.vpit;
4844 mutex_lock(&pit->pit_state.lock);
4845 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4846 for (i = 0; i < 3; i++)
4847 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4848 mutex_unlock(&pit->pit_state.lock);
4852 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4854 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4855 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4856 sizeof(ps->channels));
4857 ps->flags = kvm->arch.vpit->pit_state.flags;
4858 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4859 memset(&ps->reserved, 0, sizeof(ps->reserved));
4863 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4867 u32 prev_legacy, cur_legacy;
4868 struct kvm_pit *pit = kvm->arch.vpit;
4870 mutex_lock(&pit->pit_state.lock);
4871 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4872 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4873 if (!prev_legacy && cur_legacy)
4875 memcpy(&pit->pit_state.channels, &ps->channels,
4876 sizeof(pit->pit_state.channels));
4877 pit->pit_state.flags = ps->flags;
4878 for (i = 0; i < 3; i++)
4879 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4881 mutex_unlock(&pit->pit_state.lock);
4885 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4886 struct kvm_reinject_control *control)
4888 struct kvm_pit *pit = kvm->arch.vpit;
4890 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4891 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4892 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4894 mutex_lock(&pit->pit_state.lock);
4895 kvm_pit_set_reinject(pit, control->pit_reinject);
4896 mutex_unlock(&pit->pit_state.lock);
4901 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
4904 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4906 if (kvm_x86_ops.flush_log_dirty)
4907 kvm_x86_ops.flush_log_dirty(kvm);
4910 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4913 if (!irqchip_in_kernel(kvm))
4916 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4917 irq_event->irq, irq_event->level,
4922 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4923 struct kvm_enable_cap *cap)
4931 case KVM_CAP_DISABLE_QUIRKS:
4932 kvm->arch.disabled_quirks = cap->args[0];
4935 case KVM_CAP_SPLIT_IRQCHIP: {
4936 mutex_lock(&kvm->lock);
4938 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4939 goto split_irqchip_unlock;
4941 if (irqchip_in_kernel(kvm))
4942 goto split_irqchip_unlock;
4943 if (kvm->created_vcpus)
4944 goto split_irqchip_unlock;
4945 r = kvm_setup_empty_irq_routing(kvm);
4947 goto split_irqchip_unlock;
4948 /* Pairs with irqchip_in_kernel. */
4950 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4951 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4953 split_irqchip_unlock:
4954 mutex_unlock(&kvm->lock);
4957 case KVM_CAP_X2APIC_API:
4959 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4962 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4963 kvm->arch.x2apic_format = true;
4964 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4965 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4969 case KVM_CAP_X86_DISABLE_EXITS:
4971 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4974 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4975 kvm_can_mwait_in_guest())
4976 kvm->arch.mwait_in_guest = true;
4977 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4978 kvm->arch.hlt_in_guest = true;
4979 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4980 kvm->arch.pause_in_guest = true;
4981 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4982 kvm->arch.cstate_in_guest = true;
4985 case KVM_CAP_MSR_PLATFORM_INFO:
4986 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4989 case KVM_CAP_EXCEPTION_PAYLOAD:
4990 kvm->arch.exception_payload_enabled = cap->args[0];
5000 long kvm_arch_vm_ioctl(struct file *filp,
5001 unsigned int ioctl, unsigned long arg)
5003 struct kvm *kvm = filp->private_data;
5004 void __user *argp = (void __user *)arg;
5007 * This union makes it completely explicit to gcc-3.x
5008 * that these two variables' stack usage should be
5009 * combined, not added together.
5012 struct kvm_pit_state ps;
5013 struct kvm_pit_state2 ps2;
5014 struct kvm_pit_config pit_config;
5018 case KVM_SET_TSS_ADDR:
5019 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5021 case KVM_SET_IDENTITY_MAP_ADDR: {
5024 mutex_lock(&kvm->lock);
5026 if (kvm->created_vcpus)
5027 goto set_identity_unlock;
5029 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5030 goto set_identity_unlock;
5031 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5032 set_identity_unlock:
5033 mutex_unlock(&kvm->lock);
5036 case KVM_SET_NR_MMU_PAGES:
5037 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5039 case KVM_GET_NR_MMU_PAGES:
5040 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5042 case KVM_CREATE_IRQCHIP: {
5043 mutex_lock(&kvm->lock);
5046 if (irqchip_in_kernel(kvm))
5047 goto create_irqchip_unlock;
5050 if (kvm->created_vcpus)
5051 goto create_irqchip_unlock;
5053 r = kvm_pic_init(kvm);
5055 goto create_irqchip_unlock;
5057 r = kvm_ioapic_init(kvm);
5059 kvm_pic_destroy(kvm);
5060 goto create_irqchip_unlock;
5063 r = kvm_setup_default_irq_routing(kvm);
5065 kvm_ioapic_destroy(kvm);
5066 kvm_pic_destroy(kvm);
5067 goto create_irqchip_unlock;
5069 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5071 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5072 create_irqchip_unlock:
5073 mutex_unlock(&kvm->lock);
5076 case KVM_CREATE_PIT:
5077 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5079 case KVM_CREATE_PIT2:
5081 if (copy_from_user(&u.pit_config, argp,
5082 sizeof(struct kvm_pit_config)))
5085 mutex_lock(&kvm->lock);
5088 goto create_pit_unlock;
5090 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5094 mutex_unlock(&kvm->lock);
5096 case KVM_GET_IRQCHIP: {
5097 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5098 struct kvm_irqchip *chip;
5100 chip = memdup_user(argp, sizeof(*chip));
5107 if (!irqchip_kernel(kvm))
5108 goto get_irqchip_out;
5109 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5111 goto get_irqchip_out;
5113 if (copy_to_user(argp, chip, sizeof(*chip)))
5114 goto get_irqchip_out;
5120 case KVM_SET_IRQCHIP: {
5121 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5122 struct kvm_irqchip *chip;
5124 chip = memdup_user(argp, sizeof(*chip));
5131 if (!irqchip_kernel(kvm))
5132 goto set_irqchip_out;
5133 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5140 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5143 if (!kvm->arch.vpit)
5145 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5149 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5156 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5158 mutex_lock(&kvm->lock);
5160 if (!kvm->arch.vpit)
5162 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5164 mutex_unlock(&kvm->lock);
5167 case KVM_GET_PIT2: {
5169 if (!kvm->arch.vpit)
5171 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5175 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5180 case KVM_SET_PIT2: {
5182 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5184 mutex_lock(&kvm->lock);
5186 if (!kvm->arch.vpit)
5188 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5190 mutex_unlock(&kvm->lock);
5193 case KVM_REINJECT_CONTROL: {
5194 struct kvm_reinject_control control;
5196 if (copy_from_user(&control, argp, sizeof(control)))
5199 if (!kvm->arch.vpit)
5201 r = kvm_vm_ioctl_reinject(kvm, &control);
5204 case KVM_SET_BOOT_CPU_ID:
5206 mutex_lock(&kvm->lock);
5207 if (kvm->created_vcpus)
5210 kvm->arch.bsp_vcpu_id = arg;
5211 mutex_unlock(&kvm->lock);
5213 case KVM_XEN_HVM_CONFIG: {
5214 struct kvm_xen_hvm_config xhc;
5216 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5221 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5225 case KVM_SET_CLOCK: {
5226 struct kvm_clock_data user_ns;
5230 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5239 * TODO: userspace has to take care of races with VCPU_RUN, so
5240 * kvm_gen_update_masterclock() can be cut down to locked
5241 * pvclock_update_vm_gtod_copy().
5243 kvm_gen_update_masterclock(kvm);
5244 now_ns = get_kvmclock_ns(kvm);
5245 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5246 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5249 case KVM_GET_CLOCK: {
5250 struct kvm_clock_data user_ns;
5253 now_ns = get_kvmclock_ns(kvm);
5254 user_ns.clock = now_ns;
5255 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5256 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5259 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5264 case KVM_MEMORY_ENCRYPT_OP: {
5266 if (kvm_x86_ops.mem_enc_op)
5267 r = kvm_x86_ops.mem_enc_op(kvm, argp);
5270 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5271 struct kvm_enc_region region;
5274 if (copy_from_user(®ion, argp, sizeof(region)))
5278 if (kvm_x86_ops.mem_enc_reg_region)
5279 r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion);
5282 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5283 struct kvm_enc_region region;
5286 if (copy_from_user(®ion, argp, sizeof(region)))
5290 if (kvm_x86_ops.mem_enc_unreg_region)
5291 r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion);
5294 case KVM_HYPERV_EVENTFD: {
5295 struct kvm_hyperv_eventfd hvevfd;
5298 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5300 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5303 case KVM_SET_PMU_EVENT_FILTER:
5304 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5313 static void kvm_init_msr_list(void)
5315 struct x86_pmu_capability x86_pmu;
5319 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5320 "Please update the fixed PMCs in msrs_to_saved_all[]");
5322 perf_get_x86_pmu_capability(&x86_pmu);
5324 num_msrs_to_save = 0;
5325 num_emulated_msrs = 0;
5326 num_msr_based_features = 0;
5328 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5329 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5333 * Even MSRs that are valid in the host may not be exposed
5334 * to the guests in some cases.
5336 switch (msrs_to_save_all[i]) {
5337 case MSR_IA32_BNDCFGS:
5338 if (!kvm_mpx_supported())
5342 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5345 case MSR_IA32_UMWAIT_CONTROL:
5346 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5349 case MSR_IA32_RTIT_CTL:
5350 case MSR_IA32_RTIT_STATUS:
5351 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5354 case MSR_IA32_RTIT_CR3_MATCH:
5355 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5356 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5359 case MSR_IA32_RTIT_OUTPUT_BASE:
5360 case MSR_IA32_RTIT_OUTPUT_MASK:
5361 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5362 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5363 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5366 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5367 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5368 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5369 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5372 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5373 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5374 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5377 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5378 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5379 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5386 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5389 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5390 if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5393 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5396 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5397 struct kvm_msr_entry msr;
5399 msr.index = msr_based_features_all[i];
5400 if (kvm_get_msr_feature(&msr))
5403 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5407 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5415 if (!(lapic_in_kernel(vcpu) &&
5416 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5417 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5428 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5435 if (!(lapic_in_kernel(vcpu) &&
5436 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5438 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5440 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5450 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5451 struct kvm_segment *var, int seg)
5453 kvm_x86_ops.set_segment(vcpu, var, seg);
5456 void kvm_get_segment(struct kvm_vcpu *vcpu,
5457 struct kvm_segment *var, int seg)
5459 kvm_x86_ops.get_segment(vcpu, var, seg);
5462 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5463 struct x86_exception *exception)
5467 BUG_ON(!mmu_is_nested(vcpu));
5469 /* NPT walks are always user-walks */
5470 access |= PFERR_USER_MASK;
5471 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5476 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5477 struct x86_exception *exception)
5479 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5480 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5483 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5484 struct x86_exception *exception)
5486 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5487 access |= PFERR_FETCH_MASK;
5488 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5491 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5492 struct x86_exception *exception)
5494 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5495 access |= PFERR_WRITE_MASK;
5496 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5499 /* uses this to access any guest's mapped memory without checking CPL */
5500 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5501 struct x86_exception *exception)
5503 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5506 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5507 struct kvm_vcpu *vcpu, u32 access,
5508 struct x86_exception *exception)
5511 int r = X86EMUL_CONTINUE;
5514 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5516 unsigned offset = addr & (PAGE_SIZE-1);
5517 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5520 if (gpa == UNMAPPED_GVA)
5521 return X86EMUL_PROPAGATE_FAULT;
5522 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5525 r = X86EMUL_IO_NEEDED;
5537 /* used for instruction fetching */
5538 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5539 gva_t addr, void *val, unsigned int bytes,
5540 struct x86_exception *exception)
5542 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5543 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5547 /* Inline kvm_read_guest_virt_helper for speed. */
5548 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5550 if (unlikely(gpa == UNMAPPED_GVA))
5551 return X86EMUL_PROPAGATE_FAULT;
5553 offset = addr & (PAGE_SIZE-1);
5554 if (WARN_ON(offset + bytes > PAGE_SIZE))
5555 bytes = (unsigned)PAGE_SIZE - offset;
5556 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5558 if (unlikely(ret < 0))
5559 return X86EMUL_IO_NEEDED;
5561 return X86EMUL_CONTINUE;
5564 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5565 gva_t addr, void *val, unsigned int bytes,
5566 struct x86_exception *exception)
5568 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5571 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5572 * is returned, but our callers are not ready for that and they blindly
5573 * call kvm_inject_page_fault. Ensure that they at least do not leak
5574 * uninitialized kernel stack memory into cr2 and error code.
5576 memset(exception, 0, sizeof(*exception));
5577 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5580 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5582 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5583 gva_t addr, void *val, unsigned int bytes,
5584 struct x86_exception *exception, bool system)
5586 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5589 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5590 access |= PFERR_USER_MASK;
5592 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5595 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5596 unsigned long addr, void *val, unsigned int bytes)
5598 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5599 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5601 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5604 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5605 struct kvm_vcpu *vcpu, u32 access,
5606 struct x86_exception *exception)
5609 int r = X86EMUL_CONTINUE;
5612 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5615 unsigned offset = addr & (PAGE_SIZE-1);
5616 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5619 if (gpa == UNMAPPED_GVA)
5620 return X86EMUL_PROPAGATE_FAULT;
5621 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5623 r = X86EMUL_IO_NEEDED;
5635 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5636 unsigned int bytes, struct x86_exception *exception,
5639 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5640 u32 access = PFERR_WRITE_MASK;
5642 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5643 access |= PFERR_USER_MASK;
5645 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5649 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5650 unsigned int bytes, struct x86_exception *exception)
5652 /* kvm_write_guest_virt_system can pull in tons of pages. */
5653 vcpu->arch.l1tf_flush_l1d = true;
5655 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5656 PFERR_WRITE_MASK, exception);
5658 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5660 int handle_ud(struct kvm_vcpu *vcpu)
5662 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5663 int emul_type = EMULTYPE_TRAP_UD;
5664 char sig[5]; /* ud2; .ascii "kvm" */
5665 struct x86_exception e;
5667 if (force_emulation_prefix &&
5668 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5669 sig, sizeof(sig), &e) == 0 &&
5670 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5671 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5672 emul_type = EMULTYPE_TRAP_UD_FORCED;
5675 return kvm_emulate_instruction(vcpu, emul_type);
5677 EXPORT_SYMBOL_GPL(handle_ud);
5679 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5680 gpa_t gpa, bool write)
5682 /* For APIC access vmexit */
5683 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5686 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5687 trace_vcpu_match_mmio(gva, gpa, write, true);
5694 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5695 gpa_t *gpa, struct x86_exception *exception,
5698 u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5699 | (write ? PFERR_WRITE_MASK : 0);
5702 * currently PKRU is only applied to ept enabled guest so
5703 * there is no pkey in EPT page table for L1 guest or EPT
5704 * shadow page table for L2 guest.
5706 if (vcpu_match_mmio_gva(vcpu, gva)
5707 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5708 vcpu->arch.mmio_access, 0, access)) {
5709 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5710 (gva & (PAGE_SIZE - 1));
5711 trace_vcpu_match_mmio(gva, *gpa, write, false);
5715 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5717 if (*gpa == UNMAPPED_GVA)
5720 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5723 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5724 const void *val, int bytes)
5728 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5731 kvm_page_track_write(vcpu, gpa, val, bytes);
5735 struct read_write_emulator_ops {
5736 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5738 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5739 void *val, int bytes);
5740 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5741 int bytes, void *val);
5742 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5743 void *val, int bytes);
5747 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5749 if (vcpu->mmio_read_completed) {
5750 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5751 vcpu->mmio_fragments[0].gpa, val);
5752 vcpu->mmio_read_completed = 0;
5759 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5760 void *val, int bytes)
5762 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5765 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5766 void *val, int bytes)
5768 return emulator_write_phys(vcpu, gpa, val, bytes);
5771 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5773 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5774 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5777 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5778 void *val, int bytes)
5780 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5781 return X86EMUL_IO_NEEDED;
5784 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5785 void *val, int bytes)
5787 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5789 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5790 return X86EMUL_CONTINUE;
5793 static const struct read_write_emulator_ops read_emultor = {
5794 .read_write_prepare = read_prepare,
5795 .read_write_emulate = read_emulate,
5796 .read_write_mmio = vcpu_mmio_read,
5797 .read_write_exit_mmio = read_exit_mmio,
5800 static const struct read_write_emulator_ops write_emultor = {
5801 .read_write_emulate = write_emulate,
5802 .read_write_mmio = write_mmio,
5803 .read_write_exit_mmio = write_exit_mmio,
5807 static int emulator_read_write_onepage(unsigned long addr, void *val,
5809 struct x86_exception *exception,
5810 struct kvm_vcpu *vcpu,
5811 const struct read_write_emulator_ops *ops)
5815 bool write = ops->write;
5816 struct kvm_mmio_fragment *frag;
5817 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
5820 * If the exit was due to a NPF we may already have a GPA.
5821 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5822 * Note, this cannot be used on string operations since string
5823 * operation using rep will only have the initial GPA from the NPF
5826 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
5827 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
5828 gpa = ctxt->gpa_val;
5829 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5831 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5833 return X86EMUL_PROPAGATE_FAULT;
5836 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5837 return X86EMUL_CONTINUE;
5840 * Is this MMIO handled locally?
5842 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5843 if (handled == bytes)
5844 return X86EMUL_CONTINUE;
5850 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5851 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5855 return X86EMUL_CONTINUE;
5858 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5860 void *val, unsigned int bytes,
5861 struct x86_exception *exception,
5862 const struct read_write_emulator_ops *ops)
5864 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5868 if (ops->read_write_prepare &&
5869 ops->read_write_prepare(vcpu, val, bytes))
5870 return X86EMUL_CONTINUE;
5872 vcpu->mmio_nr_fragments = 0;
5874 /* Crossing a page boundary? */
5875 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5878 now = -addr & ~PAGE_MASK;
5879 rc = emulator_read_write_onepage(addr, val, now, exception,
5882 if (rc != X86EMUL_CONTINUE)
5885 if (ctxt->mode != X86EMUL_MODE_PROT64)
5891 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5893 if (rc != X86EMUL_CONTINUE)
5896 if (!vcpu->mmio_nr_fragments)
5899 gpa = vcpu->mmio_fragments[0].gpa;
5901 vcpu->mmio_needed = 1;
5902 vcpu->mmio_cur_fragment = 0;
5904 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5905 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5906 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5907 vcpu->run->mmio.phys_addr = gpa;
5909 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5912 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5916 struct x86_exception *exception)
5918 return emulator_read_write(ctxt, addr, val, bytes,
5919 exception, &read_emultor);
5922 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5926 struct x86_exception *exception)
5928 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5929 exception, &write_emultor);
5932 #define CMPXCHG_TYPE(t, ptr, old, new) \
5933 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5935 #ifdef CONFIG_X86_64
5936 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5938 # define CMPXCHG64(ptr, old, new) \
5939 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5942 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5947 struct x86_exception *exception)
5949 struct kvm_host_map map;
5950 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5956 /* guests cmpxchg8b have to be emulated atomically */
5957 if (bytes > 8 || (bytes & (bytes - 1)))
5960 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5962 if (gpa == UNMAPPED_GVA ||
5963 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5967 * Emulate the atomic as a straight write to avoid #AC if SLD is
5968 * enabled in the host and the access splits a cache line.
5970 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
5971 page_line_mask = ~(cache_line_size() - 1);
5973 page_line_mask = PAGE_MASK;
5975 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
5978 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5981 kaddr = map.hva + offset_in_page(gpa);
5985 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5988 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5991 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5994 exchanged = CMPXCHG64(kaddr, old, new);
6000 kvm_vcpu_unmap(vcpu, &map, true);
6003 return X86EMUL_CMPXCHG_FAILED;
6005 kvm_page_track_write(vcpu, gpa, new, bytes);
6007 return X86EMUL_CONTINUE;
6010 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6012 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6015 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6019 for (i = 0; i < vcpu->arch.pio.count; i++) {
6020 if (vcpu->arch.pio.in)
6021 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6022 vcpu->arch.pio.size, pd);
6024 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6025 vcpu->arch.pio.port, vcpu->arch.pio.size,
6029 pd += vcpu->arch.pio.size;
6034 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6035 unsigned short port, void *val,
6036 unsigned int count, bool in)
6038 vcpu->arch.pio.port = port;
6039 vcpu->arch.pio.in = in;
6040 vcpu->arch.pio.count = count;
6041 vcpu->arch.pio.size = size;
6043 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6044 vcpu->arch.pio.count = 0;
6048 vcpu->run->exit_reason = KVM_EXIT_IO;
6049 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6050 vcpu->run->io.size = size;
6051 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6052 vcpu->run->io.count = count;
6053 vcpu->run->io.port = port;
6058 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6059 unsigned short port, void *val, unsigned int count)
6063 if (vcpu->arch.pio.count)
6066 memset(vcpu->arch.pio_data, 0, size * count);
6068 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6071 memcpy(val, vcpu->arch.pio_data, size * count);
6072 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6073 vcpu->arch.pio.count = 0;
6080 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6081 int size, unsigned short port, void *val,
6084 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6088 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6089 unsigned short port, const void *val,
6092 memcpy(vcpu->arch.pio_data, val, size * count);
6093 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6094 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6097 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6098 int size, unsigned short port,
6099 const void *val, unsigned int count)
6101 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6104 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6106 return kvm_x86_ops.get_segment_base(vcpu, seg);
6109 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6111 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6114 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6116 if (!need_emulate_wbinvd(vcpu))
6117 return X86EMUL_CONTINUE;
6119 if (kvm_x86_ops.has_wbinvd_exit()) {
6120 int cpu = get_cpu();
6122 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6123 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6124 wbinvd_ipi, NULL, 1);
6126 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6129 return X86EMUL_CONTINUE;
6132 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6134 kvm_emulate_wbinvd_noskip(vcpu);
6135 return kvm_skip_emulated_instruction(vcpu);
6137 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6141 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6143 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6146 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6147 unsigned long *dest)
6149 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6152 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6153 unsigned long value)
6156 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6159 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6161 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6164 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6166 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6167 unsigned long value;
6171 value = kvm_read_cr0(vcpu);
6174 value = vcpu->arch.cr2;
6177 value = kvm_read_cr3(vcpu);
6180 value = kvm_read_cr4(vcpu);
6183 value = kvm_get_cr8(vcpu);
6186 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6193 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6195 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6200 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6203 vcpu->arch.cr2 = val;
6206 res = kvm_set_cr3(vcpu, val);
6209 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6212 res = kvm_set_cr8(vcpu, val);
6215 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6222 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6224 return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6227 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6229 kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6232 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6234 kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6237 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6239 kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6242 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6244 kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6247 static unsigned long emulator_get_cached_segment_base(
6248 struct x86_emulate_ctxt *ctxt, int seg)
6250 return get_segment_base(emul_to_vcpu(ctxt), seg);
6253 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6254 struct desc_struct *desc, u32 *base3,
6257 struct kvm_segment var;
6259 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6260 *selector = var.selector;
6263 memset(desc, 0, sizeof(*desc));
6271 set_desc_limit(desc, var.limit);
6272 set_desc_base(desc, (unsigned long)var.base);
6273 #ifdef CONFIG_X86_64
6275 *base3 = var.base >> 32;
6277 desc->type = var.type;
6279 desc->dpl = var.dpl;
6280 desc->p = var.present;
6281 desc->avl = var.avl;
6289 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6290 struct desc_struct *desc, u32 base3,
6293 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6294 struct kvm_segment var;
6296 var.selector = selector;
6297 var.base = get_desc_base(desc);
6298 #ifdef CONFIG_X86_64
6299 var.base |= ((u64)base3) << 32;
6301 var.limit = get_desc_limit(desc);
6303 var.limit = (var.limit << 12) | 0xfff;
6304 var.type = desc->type;
6305 var.dpl = desc->dpl;
6310 var.avl = desc->avl;
6311 var.present = desc->p;
6312 var.unusable = !var.present;
6315 kvm_set_segment(vcpu, &var, seg);
6319 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6320 u32 msr_index, u64 *pdata)
6322 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6325 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6326 u32 msr_index, u64 data)
6328 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6331 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6333 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6335 return vcpu->arch.smbase;
6338 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6340 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6342 vcpu->arch.smbase = smbase;
6345 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6348 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6351 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6352 u32 pmc, u64 *pdata)
6354 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6357 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6359 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6362 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6363 struct x86_instruction_info *info,
6364 enum x86_intercept_stage stage)
6366 return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6370 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6371 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6374 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6377 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6379 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6382 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6384 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6387 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6389 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6392 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6394 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6397 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6399 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6402 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6404 kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6407 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6409 return emul_to_vcpu(ctxt)->arch.hflags;
6412 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6414 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6417 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6418 const char *smstate)
6420 return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6423 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6425 kvm_smm_changed(emul_to_vcpu(ctxt));
6428 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6430 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6433 static const struct x86_emulate_ops emulate_ops = {
6434 .read_gpr = emulator_read_gpr,
6435 .write_gpr = emulator_write_gpr,
6436 .read_std = emulator_read_std,
6437 .write_std = emulator_write_std,
6438 .read_phys = kvm_read_guest_phys_system,
6439 .fetch = kvm_fetch_guest_virt,
6440 .read_emulated = emulator_read_emulated,
6441 .write_emulated = emulator_write_emulated,
6442 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6443 .invlpg = emulator_invlpg,
6444 .pio_in_emulated = emulator_pio_in_emulated,
6445 .pio_out_emulated = emulator_pio_out_emulated,
6446 .get_segment = emulator_get_segment,
6447 .set_segment = emulator_set_segment,
6448 .get_cached_segment_base = emulator_get_cached_segment_base,
6449 .get_gdt = emulator_get_gdt,
6450 .get_idt = emulator_get_idt,
6451 .set_gdt = emulator_set_gdt,
6452 .set_idt = emulator_set_idt,
6453 .get_cr = emulator_get_cr,
6454 .set_cr = emulator_set_cr,
6455 .cpl = emulator_get_cpl,
6456 .get_dr = emulator_get_dr,
6457 .set_dr = emulator_set_dr,
6458 .get_smbase = emulator_get_smbase,
6459 .set_smbase = emulator_set_smbase,
6460 .set_msr = emulator_set_msr,
6461 .get_msr = emulator_get_msr,
6462 .check_pmc = emulator_check_pmc,
6463 .read_pmc = emulator_read_pmc,
6464 .halt = emulator_halt,
6465 .wbinvd = emulator_wbinvd,
6466 .fix_hypercall = emulator_fix_hypercall,
6467 .intercept = emulator_intercept,
6468 .get_cpuid = emulator_get_cpuid,
6469 .guest_has_long_mode = emulator_guest_has_long_mode,
6470 .guest_has_movbe = emulator_guest_has_movbe,
6471 .guest_has_fxsr = emulator_guest_has_fxsr,
6472 .set_nmi_mask = emulator_set_nmi_mask,
6473 .get_hflags = emulator_get_hflags,
6474 .set_hflags = emulator_set_hflags,
6475 .pre_leave_smm = emulator_pre_leave_smm,
6476 .post_leave_smm = emulator_post_leave_smm,
6477 .set_xcr = emulator_set_xcr,
6480 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6482 u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6484 * an sti; sti; sequence only disable interrupts for the first
6485 * instruction. So, if the last instruction, be it emulated or
6486 * not, left the system with the INT_STI flag enabled, it
6487 * means that the last instruction is an sti. We should not
6488 * leave the flag on in this case. The same goes for mov ss
6490 if (int_shadow & mask)
6492 if (unlikely(int_shadow || mask)) {
6493 kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6495 kvm_make_request(KVM_REQ_EVENT, vcpu);
6499 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6501 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6502 if (ctxt->exception.vector == PF_VECTOR)
6503 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
6505 if (ctxt->exception.error_code_valid)
6506 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6507 ctxt->exception.error_code);
6509 kvm_queue_exception(vcpu, ctxt->exception.vector);
6513 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6515 struct x86_emulate_ctxt *ctxt;
6517 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6519 pr_err("kvm: failed to allocate vcpu's emulator\n");
6524 ctxt->ops = &emulate_ops;
6525 vcpu->arch.emulate_ctxt = ctxt;
6530 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6532 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6535 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6537 ctxt->gpa_available = false;
6538 ctxt->eflags = kvm_get_rflags(vcpu);
6539 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6541 ctxt->eip = kvm_rip_read(vcpu);
6542 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6543 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6544 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6545 cs_db ? X86EMUL_MODE_PROT32 :
6546 X86EMUL_MODE_PROT16;
6547 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6548 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6549 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6551 init_decode_cache(ctxt);
6552 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6555 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6557 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6560 init_emulate_ctxt(vcpu);
6564 ctxt->_eip = ctxt->eip + inc_eip;
6565 ret = emulate_int_real(ctxt, irq);
6567 if (ret != X86EMUL_CONTINUE) {
6568 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6570 ctxt->eip = ctxt->_eip;
6571 kvm_rip_write(vcpu, ctxt->eip);
6572 kvm_set_rflags(vcpu, ctxt->eflags);
6575 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6577 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6579 ++vcpu->stat.insn_emulation_fail;
6580 trace_kvm_emulate_insn_failed(vcpu);
6582 if (emulation_type & EMULTYPE_VMWARE_GP) {
6583 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6587 if (emulation_type & EMULTYPE_SKIP) {
6588 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6589 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6590 vcpu->run->internal.ndata = 0;
6594 kvm_queue_exception(vcpu, UD_VECTOR);
6596 if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
6597 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6598 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6599 vcpu->run->internal.ndata = 0;
6606 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6607 bool write_fault_to_shadow_pgtable,
6610 gpa_t gpa = cr2_or_gpa;
6613 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6616 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6617 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6620 if (!vcpu->arch.mmu->direct_map) {
6622 * Write permission should be allowed since only
6623 * write access need to be emulated.
6625 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6628 * If the mapping is invalid in guest, let cpu retry
6629 * it to generate fault.
6631 if (gpa == UNMAPPED_GVA)
6636 * Do not retry the unhandleable instruction if it faults on the
6637 * readonly host memory, otherwise it will goto a infinite loop:
6638 * retry instruction -> write #PF -> emulation fail -> retry
6639 * instruction -> ...
6641 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6644 * If the instruction failed on the error pfn, it can not be fixed,
6645 * report the error to userspace.
6647 if (is_error_noslot_pfn(pfn))
6650 kvm_release_pfn_clean(pfn);
6652 /* The instructions are well-emulated on direct mmu. */
6653 if (vcpu->arch.mmu->direct_map) {
6654 unsigned int indirect_shadow_pages;
6656 spin_lock(&vcpu->kvm->mmu_lock);
6657 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6658 spin_unlock(&vcpu->kvm->mmu_lock);
6660 if (indirect_shadow_pages)
6661 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6667 * if emulation was due to access to shadowed page table
6668 * and it failed try to unshadow page and re-enter the
6669 * guest to let CPU execute the instruction.
6671 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6674 * If the access faults on its page table, it can not
6675 * be fixed by unprotecting shadow page and it should
6676 * be reported to userspace.
6678 return !write_fault_to_shadow_pgtable;
6681 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6682 gpa_t cr2_or_gpa, int emulation_type)
6684 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6685 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
6687 last_retry_eip = vcpu->arch.last_retry_eip;
6688 last_retry_addr = vcpu->arch.last_retry_addr;
6691 * If the emulation is caused by #PF and it is non-page_table
6692 * writing instruction, it means the VM-EXIT is caused by shadow
6693 * page protected, we can zap the shadow page and retry this
6694 * instruction directly.
6696 * Note: if the guest uses a non-page-table modifying instruction
6697 * on the PDE that points to the instruction, then we will unmap
6698 * the instruction and go to an infinite loop. So, we cache the
6699 * last retried eip and the last fault address, if we meet the eip
6700 * and the address again, we can break out of the potential infinite
6703 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6705 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6708 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6709 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6712 if (x86_page_table_writing_insn(ctxt))
6715 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
6718 vcpu->arch.last_retry_eip = ctxt->eip;
6719 vcpu->arch.last_retry_addr = cr2_or_gpa;
6721 if (!vcpu->arch.mmu->direct_map)
6722 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6724 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6729 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6730 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6732 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6734 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6735 /* This is a good place to trace that we are exiting SMM. */
6736 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6738 /* Process a latched INIT or SMI, if any. */
6739 kvm_make_request(KVM_REQ_EVENT, vcpu);
6742 kvm_mmu_reset_context(vcpu);
6745 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6754 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6755 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6760 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6762 struct kvm_run *kvm_run = vcpu->run;
6764 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6765 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6766 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6767 kvm_run->debug.arch.exception = DB_VECTOR;
6768 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6771 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6775 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6777 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6780 r = kvm_x86_ops.skip_emulated_instruction(vcpu);
6785 * rflags is the old, "raw" value of the flags. The new value has
6786 * not been saved yet.
6788 * This is correct even for TF set by the guest, because "the
6789 * processor will not generate this exception after the instruction
6790 * that sets the TF flag".
6792 if (unlikely(rflags & X86_EFLAGS_TF))
6793 r = kvm_vcpu_do_singlestep(vcpu);
6796 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6798 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6800 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6801 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6802 struct kvm_run *kvm_run = vcpu->run;
6803 unsigned long eip = kvm_get_linear_rip(vcpu);
6804 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6805 vcpu->arch.guest_debug_dr7,
6809 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6810 kvm_run->debug.arch.pc = eip;
6811 kvm_run->debug.arch.exception = DB_VECTOR;
6812 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6818 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6819 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6820 unsigned long eip = kvm_get_linear_rip(vcpu);
6821 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6826 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
6835 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6837 switch (ctxt->opcode_len) {
6844 case 0xe6: /* OUT */
6848 case 0x6c: /* INS */
6850 case 0x6e: /* OUTS */
6857 case 0x33: /* RDPMC */
6866 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6867 int emulation_type, void *insn, int insn_len)
6870 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6871 bool writeback = true;
6872 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6874 vcpu->arch.l1tf_flush_l1d = true;
6877 * Clear write_fault_to_shadow_pgtable here to ensure it is
6880 vcpu->arch.write_fault_to_shadow_pgtable = false;
6881 kvm_clear_exception_queue(vcpu);
6883 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6884 init_emulate_ctxt(vcpu);
6887 * We will reenter on the same instruction since
6888 * we do not set complete_userspace_io. This does not
6889 * handle watchpoints yet, those would be handled in
6892 if (!(emulation_type & EMULTYPE_SKIP) &&
6893 kvm_vcpu_check_breakpoint(vcpu, &r))
6896 ctxt->interruptibility = 0;
6897 ctxt->have_exception = false;
6898 ctxt->exception.vector = -1;
6899 ctxt->perm_ok = false;
6901 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6903 r = x86_decode_insn(ctxt, insn, insn_len);
6905 trace_kvm_emulate_insn_start(vcpu);
6906 ++vcpu->stat.insn_emulation;
6907 if (r != EMULATION_OK) {
6908 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6909 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6910 kvm_queue_exception(vcpu, UD_VECTOR);
6913 if (reexecute_instruction(vcpu, cr2_or_gpa,
6917 if (ctxt->have_exception) {
6919 * #UD should result in just EMULATION_FAILED, and trap-like
6920 * exception should not be encountered during decode.
6922 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6923 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6924 inject_emulated_exception(vcpu);
6927 return handle_emulation_failure(vcpu, emulation_type);
6931 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6932 !is_vmware_backdoor_opcode(ctxt)) {
6933 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6938 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6939 * for kvm_skip_emulated_instruction(). The caller is responsible for
6940 * updating interruptibility state and injecting single-step #DBs.
6942 if (emulation_type & EMULTYPE_SKIP) {
6943 kvm_rip_write(vcpu, ctxt->_eip);
6944 if (ctxt->eflags & X86_EFLAGS_RF)
6945 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6949 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
6952 /* this is needed for vmware backdoor interface to work since it
6953 changes registers values during IO operation */
6954 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6955 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6956 emulator_invalidate_register_cache(ctxt);
6960 if (emulation_type & EMULTYPE_PF) {
6961 /* Save the faulting GPA (cr2) in the address field */
6962 ctxt->exception.address = cr2_or_gpa;
6964 /* With shadow page tables, cr2 contains a GVA or nGPA. */
6965 if (vcpu->arch.mmu->direct_map) {
6966 ctxt->gpa_available = true;
6967 ctxt->gpa_val = cr2_or_gpa;
6970 /* Sanitize the address out of an abundance of paranoia. */
6971 ctxt->exception.address = 0;
6974 r = x86_emulate_insn(ctxt);
6976 if (r == EMULATION_INTERCEPTED)
6979 if (r == EMULATION_FAILED) {
6980 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
6984 return handle_emulation_failure(vcpu, emulation_type);
6987 if (ctxt->have_exception) {
6989 if (inject_emulated_exception(vcpu))
6991 } else if (vcpu->arch.pio.count) {
6992 if (!vcpu->arch.pio.in) {
6993 /* FIXME: return into emulator if single-stepping. */
6994 vcpu->arch.pio.count = 0;
6997 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7000 } else if (vcpu->mmio_needed) {
7001 ++vcpu->stat.mmio_exits;
7003 if (!vcpu->mmio_is_write)
7006 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7007 } else if (r == EMULATION_RESTART)
7013 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7014 toggle_interruptibility(vcpu, ctxt->interruptibility);
7015 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7016 if (!ctxt->have_exception ||
7017 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7018 kvm_rip_write(vcpu, ctxt->eip);
7019 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7020 r = kvm_vcpu_do_singlestep(vcpu);
7021 if (kvm_x86_ops.update_emulated_instruction)
7022 kvm_x86_ops.update_emulated_instruction(vcpu);
7023 __kvm_set_rflags(vcpu, ctxt->eflags);
7027 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7028 * do nothing, and it will be requested again as soon as
7029 * the shadow expires. But we still need to check here,
7030 * because POPF has no interrupt shadow.
7032 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7033 kvm_make_request(KVM_REQ_EVENT, vcpu);
7035 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7040 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7042 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7044 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7046 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7047 void *insn, int insn_len)
7049 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7051 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7053 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7055 vcpu->arch.pio.count = 0;
7059 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7061 vcpu->arch.pio.count = 0;
7063 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7066 return kvm_skip_emulated_instruction(vcpu);
7069 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7070 unsigned short port)
7072 unsigned long val = kvm_rax_read(vcpu);
7073 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7079 * Workaround userspace that relies on old KVM behavior of %rip being
7080 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7083 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7084 vcpu->arch.complete_userspace_io =
7085 complete_fast_pio_out_port_0x7e;
7086 kvm_skip_emulated_instruction(vcpu);
7088 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7089 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7094 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7098 /* We should only ever be called with arch.pio.count equal to 1 */
7099 BUG_ON(vcpu->arch.pio.count != 1);
7101 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7102 vcpu->arch.pio.count = 0;
7106 /* For size less than 4 we merge, else we zero extend */
7107 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7110 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7111 * the copy and tracing
7113 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7114 kvm_rax_write(vcpu, val);
7116 return kvm_skip_emulated_instruction(vcpu);
7119 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7120 unsigned short port)
7125 /* For size less than 4 we merge, else we zero extend */
7126 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7128 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7130 kvm_rax_write(vcpu, val);
7134 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7135 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7140 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7145 ret = kvm_fast_pio_in(vcpu, size, port);
7147 ret = kvm_fast_pio_out(vcpu, size, port);
7148 return ret && kvm_skip_emulated_instruction(vcpu);
7150 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7152 static int kvmclock_cpu_down_prep(unsigned int cpu)
7154 __this_cpu_write(cpu_tsc_khz, 0);
7158 static void tsc_khz_changed(void *data)
7160 struct cpufreq_freqs *freq = data;
7161 unsigned long khz = 0;
7165 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7166 khz = cpufreq_quick_get(raw_smp_processor_id());
7169 __this_cpu_write(cpu_tsc_khz, khz);
7172 #ifdef CONFIG_X86_64
7173 static void kvm_hyperv_tsc_notifier(void)
7176 struct kvm_vcpu *vcpu;
7179 mutex_lock(&kvm_lock);
7180 list_for_each_entry(kvm, &vm_list, vm_list)
7181 kvm_make_mclock_inprogress_request(kvm);
7183 hyperv_stop_tsc_emulation();
7185 /* TSC frequency always matches when on Hyper-V */
7186 for_each_present_cpu(cpu)
7187 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7188 kvm_max_guest_tsc_khz = tsc_khz;
7190 list_for_each_entry(kvm, &vm_list, vm_list) {
7191 struct kvm_arch *ka = &kvm->arch;
7193 spin_lock(&ka->pvclock_gtod_sync_lock);
7195 pvclock_update_vm_gtod_copy(kvm);
7197 kvm_for_each_vcpu(cpu, vcpu, kvm)
7198 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7200 kvm_for_each_vcpu(cpu, vcpu, kvm)
7201 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7203 spin_unlock(&ka->pvclock_gtod_sync_lock);
7205 mutex_unlock(&kvm_lock);
7209 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7212 struct kvm_vcpu *vcpu;
7213 int i, send_ipi = 0;
7216 * We allow guests to temporarily run on slowing clocks,
7217 * provided we notify them after, or to run on accelerating
7218 * clocks, provided we notify them before. Thus time never
7221 * However, we have a problem. We can't atomically update
7222 * the frequency of a given CPU from this function; it is
7223 * merely a notifier, which can be called from any CPU.
7224 * Changing the TSC frequency at arbitrary points in time
7225 * requires a recomputation of local variables related to
7226 * the TSC for each VCPU. We must flag these local variables
7227 * to be updated and be sure the update takes place with the
7228 * new frequency before any guests proceed.
7230 * Unfortunately, the combination of hotplug CPU and frequency
7231 * change creates an intractable locking scenario; the order
7232 * of when these callouts happen is undefined with respect to
7233 * CPU hotplug, and they can race with each other. As such,
7234 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7235 * undefined; you can actually have a CPU frequency change take
7236 * place in between the computation of X and the setting of the
7237 * variable. To protect against this problem, all updates of
7238 * the per_cpu tsc_khz variable are done in an interrupt
7239 * protected IPI, and all callers wishing to update the value
7240 * must wait for a synchronous IPI to complete (which is trivial
7241 * if the caller is on the CPU already). This establishes the
7242 * necessary total order on variable updates.
7244 * Note that because a guest time update may take place
7245 * anytime after the setting of the VCPU's request bit, the
7246 * correct TSC value must be set before the request. However,
7247 * to ensure the update actually makes it to any guest which
7248 * starts running in hardware virtualization between the set
7249 * and the acquisition of the spinlock, we must also ping the
7250 * CPU after setting the request bit.
7254 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7256 mutex_lock(&kvm_lock);
7257 list_for_each_entry(kvm, &vm_list, vm_list) {
7258 kvm_for_each_vcpu(i, vcpu, kvm) {
7259 if (vcpu->cpu != cpu)
7261 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7262 if (vcpu->cpu != raw_smp_processor_id())
7266 mutex_unlock(&kvm_lock);
7268 if (freq->old < freq->new && send_ipi) {
7270 * We upscale the frequency. Must make the guest
7271 * doesn't see old kvmclock values while running with
7272 * the new frequency, otherwise we risk the guest sees
7273 * time go backwards.
7275 * In case we update the frequency for another cpu
7276 * (which might be in guest context) send an interrupt
7277 * to kick the cpu out of guest context. Next time
7278 * guest context is entered kvmclock will be updated,
7279 * so the guest will not see stale values.
7281 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7285 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7288 struct cpufreq_freqs *freq = data;
7291 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7293 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7296 for_each_cpu(cpu, freq->policy->cpus)
7297 __kvmclock_cpufreq_notifier(freq, cpu);
7302 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7303 .notifier_call = kvmclock_cpufreq_notifier
7306 static int kvmclock_cpu_online(unsigned int cpu)
7308 tsc_khz_changed(NULL);
7312 static void kvm_timer_init(void)
7314 max_tsc_khz = tsc_khz;
7316 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7317 #ifdef CONFIG_CPU_FREQ
7318 struct cpufreq_policy *policy;
7322 policy = cpufreq_cpu_get(cpu);
7324 if (policy->cpuinfo.max_freq)
7325 max_tsc_khz = policy->cpuinfo.max_freq;
7326 cpufreq_cpu_put(policy);
7330 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7331 CPUFREQ_TRANSITION_NOTIFIER);
7334 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7335 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7338 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7339 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7341 int kvm_is_in_guest(void)
7343 return __this_cpu_read(current_vcpu) != NULL;
7346 static int kvm_is_user_mode(void)
7350 if (__this_cpu_read(current_vcpu))
7351 user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7353 return user_mode != 0;
7356 static unsigned long kvm_get_guest_ip(void)
7358 unsigned long ip = 0;
7360 if (__this_cpu_read(current_vcpu))
7361 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7366 static void kvm_handle_intel_pt_intr(void)
7368 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7370 kvm_make_request(KVM_REQ_PMI, vcpu);
7371 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7372 (unsigned long *)&vcpu->arch.pmu.global_status);
7375 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7376 .is_in_guest = kvm_is_in_guest,
7377 .is_user_mode = kvm_is_user_mode,
7378 .get_guest_ip = kvm_get_guest_ip,
7379 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7382 #ifdef CONFIG_X86_64
7383 static void pvclock_gtod_update_fn(struct work_struct *work)
7387 struct kvm_vcpu *vcpu;
7390 mutex_lock(&kvm_lock);
7391 list_for_each_entry(kvm, &vm_list, vm_list)
7392 kvm_for_each_vcpu(i, vcpu, kvm)
7393 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7394 atomic_set(&kvm_guest_has_master_clock, 0);
7395 mutex_unlock(&kvm_lock);
7398 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7401 * Notification about pvclock gtod data update.
7403 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7406 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7407 struct timekeeper *tk = priv;
7409 update_pvclock_gtod(tk);
7411 /* disable master clock if host does not trust, or does not
7412 * use, TSC based clocksource.
7414 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7415 atomic_read(&kvm_guest_has_master_clock) != 0)
7416 queue_work(system_long_wq, &pvclock_gtod_work);
7421 static struct notifier_block pvclock_gtod_notifier = {
7422 .notifier_call = pvclock_gtod_notify,
7426 int kvm_arch_init(void *opaque)
7428 struct kvm_x86_init_ops *ops = opaque;
7431 if (kvm_x86_ops.hardware_enable) {
7432 printk(KERN_ERR "kvm: already loaded the other module\n");
7437 if (!ops->cpu_has_kvm_support()) {
7438 pr_err_ratelimited("kvm: no hardware support\n");
7442 if (ops->disabled_by_bios()) {
7443 pr_err_ratelimited("kvm: disabled by bios\n");
7449 * KVM explicitly assumes that the guest has an FPU and
7450 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7451 * vCPU's FPU state as a fxregs_state struct.
7453 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7454 printk(KERN_ERR "kvm: inadequate fpu\n");
7460 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7461 __alignof__(struct fpu), SLAB_ACCOUNT,
7463 if (!x86_fpu_cache) {
7464 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7468 x86_emulator_cache = kvm_alloc_emulator_cache();
7469 if (!x86_emulator_cache) {
7470 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7471 goto out_free_x86_fpu_cache;
7474 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7476 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7477 goto out_free_x86_emulator_cache;
7480 r = kvm_mmu_module_init();
7482 goto out_free_percpu;
7484 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7485 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7486 PT_PRESENT_MASK, 0, sme_me_mask);
7489 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7491 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7492 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7493 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7497 if (pi_inject_timer == -1)
7498 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7499 #ifdef CONFIG_X86_64
7500 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7502 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7503 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7509 free_percpu(shared_msrs);
7510 out_free_x86_emulator_cache:
7511 kmem_cache_destroy(x86_emulator_cache);
7512 out_free_x86_fpu_cache:
7513 kmem_cache_destroy(x86_fpu_cache);
7518 void kvm_arch_exit(void)
7520 #ifdef CONFIG_X86_64
7521 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7522 clear_hv_tscchange_cb();
7525 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7527 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7528 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7529 CPUFREQ_TRANSITION_NOTIFIER);
7530 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7531 #ifdef CONFIG_X86_64
7532 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7534 kvm_x86_ops.hardware_enable = NULL;
7535 kvm_mmu_module_exit();
7536 free_percpu(shared_msrs);
7537 kmem_cache_destroy(x86_fpu_cache);
7540 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7542 ++vcpu->stat.halt_exits;
7543 if (lapic_in_kernel(vcpu)) {
7544 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7547 vcpu->run->exit_reason = KVM_EXIT_HLT;
7551 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7553 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7555 int ret = kvm_skip_emulated_instruction(vcpu);
7557 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7558 * KVM_EXIT_DEBUG here.
7560 return kvm_vcpu_halt(vcpu) && ret;
7562 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7564 #ifdef CONFIG_X86_64
7565 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7566 unsigned long clock_type)
7568 struct kvm_clock_pairing clock_pairing;
7569 struct timespec64 ts;
7573 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7574 return -KVM_EOPNOTSUPP;
7576 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7577 return -KVM_EOPNOTSUPP;
7579 clock_pairing.sec = ts.tv_sec;
7580 clock_pairing.nsec = ts.tv_nsec;
7581 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7582 clock_pairing.flags = 0;
7583 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7586 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7587 sizeof(struct kvm_clock_pairing)))
7595 * kvm_pv_kick_cpu_op: Kick a vcpu.
7597 * @apicid - apicid of vcpu to be kicked.
7599 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7601 struct kvm_lapic_irq lapic_irq;
7603 lapic_irq.shorthand = APIC_DEST_NOSHORT;
7604 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7605 lapic_irq.level = 0;
7606 lapic_irq.dest_id = apicid;
7607 lapic_irq.msi_redir_hint = false;
7609 lapic_irq.delivery_mode = APIC_DM_REMRD;
7610 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7613 bool kvm_apicv_activated(struct kvm *kvm)
7615 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
7617 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
7619 void kvm_apicv_init(struct kvm *kvm, bool enable)
7622 clear_bit(APICV_INHIBIT_REASON_DISABLE,
7623 &kvm->arch.apicv_inhibit_reasons);
7625 set_bit(APICV_INHIBIT_REASON_DISABLE,
7626 &kvm->arch.apicv_inhibit_reasons);
7628 EXPORT_SYMBOL_GPL(kvm_apicv_init);
7630 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7632 struct kvm_vcpu *target = NULL;
7633 struct kvm_apic_map *map;
7636 map = rcu_dereference(kvm->arch.apic_map);
7638 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7639 target = map->phys_map[dest_id]->vcpu;
7643 if (target && READ_ONCE(target->ready))
7644 kvm_vcpu_yield_to(target);
7647 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7649 unsigned long nr, a0, a1, a2, a3, ret;
7652 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7653 return kvm_hv_hypercall(vcpu);
7655 nr = kvm_rax_read(vcpu);
7656 a0 = kvm_rbx_read(vcpu);
7657 a1 = kvm_rcx_read(vcpu);
7658 a2 = kvm_rdx_read(vcpu);
7659 a3 = kvm_rsi_read(vcpu);
7661 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7663 op_64_bit = is_64_bit_mode(vcpu);
7672 if (kvm_x86_ops.get_cpl(vcpu) != 0) {
7678 case KVM_HC_VAPIC_POLL_IRQ:
7681 case KVM_HC_KICK_CPU:
7682 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7683 kvm_sched_yield(vcpu->kvm, a1);
7686 #ifdef CONFIG_X86_64
7687 case KVM_HC_CLOCK_PAIRING:
7688 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7691 case KVM_HC_SEND_IPI:
7692 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7694 case KVM_HC_SCHED_YIELD:
7695 kvm_sched_yield(vcpu->kvm, a0);
7705 kvm_rax_write(vcpu, ret);
7707 ++vcpu->stat.hypercalls;
7708 return kvm_skip_emulated_instruction(vcpu);
7710 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7712 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7714 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7715 char instruction[3];
7716 unsigned long rip = kvm_rip_read(vcpu);
7718 kvm_x86_ops.patch_hypercall(vcpu, instruction);
7720 return emulator_write_emulated(ctxt, rip, instruction, 3,
7724 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7726 return vcpu->run->request_interrupt_window &&
7727 likely(!pic_in_kernel(vcpu->kvm));
7730 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7732 struct kvm_run *kvm_run = vcpu->run;
7734 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7735 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7736 kvm_run->cr8 = kvm_get_cr8(vcpu);
7737 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7738 kvm_run->ready_for_interrupt_injection =
7739 pic_in_kernel(vcpu->kvm) ||
7740 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7743 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7747 if (!kvm_x86_ops.update_cr8_intercept)
7750 if (!lapic_in_kernel(vcpu))
7753 if (vcpu->arch.apicv_active)
7756 if (!vcpu->arch.apic->vapic_addr)
7757 max_irr = kvm_lapic_find_highest_irr(vcpu);
7764 tpr = kvm_lapic_get_cr8(vcpu);
7766 kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
7769 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
7772 bool can_inject = true;
7774 /* try to reinject previous events if any */
7776 if (vcpu->arch.exception.injected) {
7777 kvm_x86_ops.queue_exception(vcpu);
7781 * Do not inject an NMI or interrupt if there is a pending
7782 * exception. Exceptions and interrupts are recognized at
7783 * instruction boundaries, i.e. the start of an instruction.
7784 * Trap-like exceptions, e.g. #DB, have higher priority than
7785 * NMIs and interrupts, i.e. traps are recognized before an
7786 * NMI/interrupt that's pending on the same instruction.
7787 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7788 * priority, but are only generated (pended) during instruction
7789 * execution, i.e. a pending fault-like exception means the
7790 * fault occurred on the *previous* instruction and must be
7791 * serviced prior to recognizing any new events in order to
7792 * fully complete the previous instruction.
7794 else if (!vcpu->arch.exception.pending) {
7795 if (vcpu->arch.nmi_injected) {
7796 kvm_x86_ops.set_nmi(vcpu);
7798 } else if (vcpu->arch.interrupt.injected) {
7799 kvm_x86_ops.set_irq(vcpu);
7804 WARN_ON_ONCE(vcpu->arch.exception.injected &&
7805 vcpu->arch.exception.pending);
7808 * Call check_nested_events() even if we reinjected a previous event
7809 * in order for caller to determine if it should require immediate-exit
7810 * from L2 to L1 due to pending L1 events which require exit
7813 if (is_guest_mode(vcpu)) {
7814 r = kvm_x86_ops.nested_ops->check_events(vcpu);
7819 /* try to inject new event if pending */
7820 if (vcpu->arch.exception.pending) {
7821 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7822 vcpu->arch.exception.has_error_code,
7823 vcpu->arch.exception.error_code);
7825 vcpu->arch.exception.pending = false;
7826 vcpu->arch.exception.injected = true;
7828 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7829 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7832 if (vcpu->arch.exception.nr == DB_VECTOR) {
7833 kvm_deliver_exception_payload(vcpu);
7834 if (vcpu->arch.dr7 & DR7_GD) {
7835 vcpu->arch.dr7 &= ~DR7_GD;
7836 kvm_update_dr7(vcpu);
7840 kvm_x86_ops.queue_exception(vcpu);
7845 * Finally, inject interrupt events. If an event cannot be injected
7846 * due to architectural conditions (e.g. IF=0) a window-open exit
7847 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
7848 * and can architecturally be injected, but we cannot do it right now:
7849 * an interrupt could have arrived just now and we have to inject it
7850 * as a vmexit, or there could already an event in the queue, which is
7851 * indicated by can_inject. In that case we request an immediate exit
7852 * in order to make progress and get back here for another iteration.
7853 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
7855 if (vcpu->arch.smi_pending) {
7856 r = can_inject ? kvm_x86_ops.smi_allowed(vcpu, true) : -EBUSY;
7860 vcpu->arch.smi_pending = false;
7861 ++vcpu->arch.smi_count;
7865 kvm_x86_ops.enable_smi_window(vcpu);
7868 if (vcpu->arch.nmi_pending) {
7869 r = can_inject ? kvm_x86_ops.nmi_allowed(vcpu, true) : -EBUSY;
7873 --vcpu->arch.nmi_pending;
7874 vcpu->arch.nmi_injected = true;
7875 kvm_x86_ops.set_nmi(vcpu);
7877 WARN_ON(kvm_x86_ops.nmi_allowed(vcpu, true) < 0);
7879 if (vcpu->arch.nmi_pending)
7880 kvm_x86_ops.enable_nmi_window(vcpu);
7883 if (kvm_cpu_has_injectable_intr(vcpu)) {
7884 r = can_inject ? kvm_x86_ops.interrupt_allowed(vcpu, true) : -EBUSY;
7888 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
7889 kvm_x86_ops.set_irq(vcpu);
7890 WARN_ON(kvm_x86_ops.interrupt_allowed(vcpu, true) < 0);
7892 if (kvm_cpu_has_injectable_intr(vcpu))
7893 kvm_x86_ops.enable_irq_window(vcpu);
7896 if (is_guest_mode(vcpu) &&
7897 kvm_x86_ops.nested_ops->hv_timer_pending &&
7898 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
7899 *req_immediate_exit = true;
7901 WARN_ON(vcpu->arch.exception.pending);
7905 *req_immediate_exit = true;
7909 static void process_nmi(struct kvm_vcpu *vcpu)
7914 * x86 is limited to one NMI running, and one NMI pending after it.
7915 * If an NMI is already in progress, limit further NMIs to just one.
7916 * Otherwise, allow two (and we'll inject the first one immediately).
7918 if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7921 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7922 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7923 kvm_make_request(KVM_REQ_EVENT, vcpu);
7926 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7929 flags |= seg->g << 23;
7930 flags |= seg->db << 22;
7931 flags |= seg->l << 21;
7932 flags |= seg->avl << 20;
7933 flags |= seg->present << 15;
7934 flags |= seg->dpl << 13;
7935 flags |= seg->s << 12;
7936 flags |= seg->type << 8;
7940 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7942 struct kvm_segment seg;
7945 kvm_get_segment(vcpu, &seg, n);
7946 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7949 offset = 0x7f84 + n * 12;
7951 offset = 0x7f2c + (n - 3) * 12;
7953 put_smstate(u32, buf, offset + 8, seg.base);
7954 put_smstate(u32, buf, offset + 4, seg.limit);
7955 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7958 #ifdef CONFIG_X86_64
7959 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7961 struct kvm_segment seg;
7965 kvm_get_segment(vcpu, &seg, n);
7966 offset = 0x7e00 + n * 16;
7968 flags = enter_smm_get_segment_flags(&seg) >> 8;
7969 put_smstate(u16, buf, offset, seg.selector);
7970 put_smstate(u16, buf, offset + 2, flags);
7971 put_smstate(u32, buf, offset + 4, seg.limit);
7972 put_smstate(u64, buf, offset + 8, seg.base);
7976 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7979 struct kvm_segment seg;
7983 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7984 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7985 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7986 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7988 for (i = 0; i < 8; i++)
7989 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7991 kvm_get_dr(vcpu, 6, &val);
7992 put_smstate(u32, buf, 0x7fcc, (u32)val);
7993 kvm_get_dr(vcpu, 7, &val);
7994 put_smstate(u32, buf, 0x7fc8, (u32)val);
7996 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7997 put_smstate(u32, buf, 0x7fc4, seg.selector);
7998 put_smstate(u32, buf, 0x7f64, seg.base);
7999 put_smstate(u32, buf, 0x7f60, seg.limit);
8000 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8002 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8003 put_smstate(u32, buf, 0x7fc0, seg.selector);
8004 put_smstate(u32, buf, 0x7f80, seg.base);
8005 put_smstate(u32, buf, 0x7f7c, seg.limit);
8006 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8008 kvm_x86_ops.get_gdt(vcpu, &dt);
8009 put_smstate(u32, buf, 0x7f74, dt.address);
8010 put_smstate(u32, buf, 0x7f70, dt.size);
8012 kvm_x86_ops.get_idt(vcpu, &dt);
8013 put_smstate(u32, buf, 0x7f58, dt.address);
8014 put_smstate(u32, buf, 0x7f54, dt.size);
8016 for (i = 0; i < 6; i++)
8017 enter_smm_save_seg_32(vcpu, buf, i);
8019 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8022 put_smstate(u32, buf, 0x7efc, 0x00020000);
8023 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8026 #ifdef CONFIG_X86_64
8027 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8030 struct kvm_segment seg;
8034 for (i = 0; i < 16; i++)
8035 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8037 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8038 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8040 kvm_get_dr(vcpu, 6, &val);
8041 put_smstate(u64, buf, 0x7f68, val);
8042 kvm_get_dr(vcpu, 7, &val);
8043 put_smstate(u64, buf, 0x7f60, val);
8045 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8046 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8047 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8049 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8052 put_smstate(u32, buf, 0x7efc, 0x00020064);
8054 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8056 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8057 put_smstate(u16, buf, 0x7e90, seg.selector);
8058 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8059 put_smstate(u32, buf, 0x7e94, seg.limit);
8060 put_smstate(u64, buf, 0x7e98, seg.base);
8062 kvm_x86_ops.get_idt(vcpu, &dt);
8063 put_smstate(u32, buf, 0x7e84, dt.size);
8064 put_smstate(u64, buf, 0x7e88, dt.address);
8066 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8067 put_smstate(u16, buf, 0x7e70, seg.selector);
8068 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8069 put_smstate(u32, buf, 0x7e74, seg.limit);
8070 put_smstate(u64, buf, 0x7e78, seg.base);
8072 kvm_x86_ops.get_gdt(vcpu, &dt);
8073 put_smstate(u32, buf, 0x7e64, dt.size);
8074 put_smstate(u64, buf, 0x7e68, dt.address);
8076 for (i = 0; i < 6; i++)
8077 enter_smm_save_seg_64(vcpu, buf, i);
8081 static void enter_smm(struct kvm_vcpu *vcpu)
8083 struct kvm_segment cs, ds;
8088 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8089 memset(buf, 0, 512);
8090 #ifdef CONFIG_X86_64
8091 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8092 enter_smm_save_state_64(vcpu, buf);
8095 enter_smm_save_state_32(vcpu, buf);
8098 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8099 * vCPU state (e.g. leave guest mode) after we've saved the state into
8100 * the SMM state-save area.
8102 kvm_x86_ops.pre_enter_smm(vcpu, buf);
8104 vcpu->arch.hflags |= HF_SMM_MASK;
8105 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8107 if (kvm_x86_ops.get_nmi_mask(vcpu))
8108 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8110 kvm_x86_ops.set_nmi_mask(vcpu, true);
8112 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8113 kvm_rip_write(vcpu, 0x8000);
8115 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8116 kvm_x86_ops.set_cr0(vcpu, cr0);
8117 vcpu->arch.cr0 = cr0;
8119 kvm_x86_ops.set_cr4(vcpu, 0);
8121 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8122 dt.address = dt.size = 0;
8123 kvm_x86_ops.set_idt(vcpu, &dt);
8125 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8127 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8128 cs.base = vcpu->arch.smbase;
8133 cs.limit = ds.limit = 0xffffffff;
8134 cs.type = ds.type = 0x3;
8135 cs.dpl = ds.dpl = 0;
8140 cs.avl = ds.avl = 0;
8141 cs.present = ds.present = 1;
8142 cs.unusable = ds.unusable = 0;
8143 cs.padding = ds.padding = 0;
8145 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8146 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8147 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8148 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8149 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8150 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8152 #ifdef CONFIG_X86_64
8153 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8154 kvm_x86_ops.set_efer(vcpu, 0);
8157 kvm_update_cpuid(vcpu);
8158 kvm_mmu_reset_context(vcpu);
8161 static void process_smi(struct kvm_vcpu *vcpu)
8163 vcpu->arch.smi_pending = true;
8164 kvm_make_request(KVM_REQ_EVENT, vcpu);
8167 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8168 unsigned long *vcpu_bitmap)
8172 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8174 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8175 NULL, vcpu_bitmap, cpus);
8177 free_cpumask_var(cpus);
8180 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8182 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8185 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8187 if (!lapic_in_kernel(vcpu))
8190 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8191 kvm_apic_update_apicv(vcpu);
8192 kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8194 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8197 * NOTE: Do not hold any lock prior to calling this.
8199 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8200 * locked, because it calls __x86_set_memory_region() which does
8201 * synchronize_srcu(&kvm->srcu).
8203 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8205 struct kvm_vcpu *except;
8206 unsigned long old, new, expected;
8208 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8209 !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8212 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8214 expected = new = old;
8216 __clear_bit(bit, &new);
8218 __set_bit(bit, &new);
8221 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8222 } while (old != expected);
8227 trace_kvm_apicv_update_request(activate, bit);
8228 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8229 kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8232 * Sending request to update APICV for all other vcpus,
8233 * while update the calling vcpu immediately instead of
8234 * waiting for another #VMEXIT to handle the request.
8236 except = kvm_get_running_vcpu();
8237 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8240 kvm_vcpu_update_apicv(except);
8242 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8244 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8246 if (!kvm_apic_present(vcpu))
8249 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8251 if (irqchip_split(vcpu->kvm))
8252 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8254 if (vcpu->arch.apicv_active)
8255 kvm_x86_ops.sync_pir_to_irr(vcpu);
8256 if (ioapic_in_kernel(vcpu->kvm))
8257 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8260 if (is_guest_mode(vcpu))
8261 vcpu->arch.load_eoi_exitmap_pending = true;
8263 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8266 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8268 u64 eoi_exit_bitmap[4];
8270 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8273 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8274 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8275 kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8278 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8279 unsigned long start, unsigned long end)
8281 unsigned long apic_address;
8284 * The physical address of apic access page is stored in the VMCS.
8285 * Update it when it becomes invalid.
8287 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8288 if (start <= apic_address && apic_address < end)
8289 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8292 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8294 if (!lapic_in_kernel(vcpu))
8297 if (!kvm_x86_ops.set_apic_access_page_addr)
8300 kvm_x86_ops.set_apic_access_page_addr(vcpu);
8303 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8305 smp_send_reschedule(vcpu->cpu);
8307 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8310 * Returns 1 to let vcpu_run() continue the guest execution loop without
8311 * exiting to the userspace. Otherwise, the value will be returned to the
8314 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8318 dm_request_for_irq_injection(vcpu) &&
8319 kvm_cpu_accept_dm_intr(vcpu);
8320 fastpath_t exit_fastpath;
8322 bool req_immediate_exit = false;
8324 if (kvm_request_pending(vcpu)) {
8325 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8326 if (unlikely(!kvm_x86_ops.nested_ops->get_vmcs12_pages(vcpu))) {
8331 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8332 kvm_mmu_unload(vcpu);
8333 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8334 __kvm_migrate_timers(vcpu);
8335 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8336 kvm_gen_update_masterclock(vcpu->kvm);
8337 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8338 kvm_gen_kvmclock_update(vcpu);
8339 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8340 r = kvm_guest_time_update(vcpu);
8344 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8345 kvm_mmu_sync_roots(vcpu);
8346 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8347 kvm_mmu_load_pgd(vcpu);
8348 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8349 kvm_vcpu_flush_tlb_all(vcpu);
8351 /* Flushing all ASIDs flushes the current ASID... */
8352 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8354 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8355 kvm_vcpu_flush_tlb_current(vcpu);
8356 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8357 kvm_vcpu_flush_tlb_guest(vcpu);
8359 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8360 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8364 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8365 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8366 vcpu->mmio_needed = 0;
8370 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8371 /* Page is swapped out. Do synthetic halt */
8372 vcpu->arch.apf.halted = true;
8376 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8377 record_steal_time(vcpu);
8378 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8380 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8382 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8383 kvm_pmu_handle_event(vcpu);
8384 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8385 kvm_pmu_deliver_pmi(vcpu);
8386 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8387 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8388 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8389 vcpu->arch.ioapic_handled_vectors)) {
8390 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8391 vcpu->run->eoi.vector =
8392 vcpu->arch.pending_ioapic_eoi;
8397 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8398 vcpu_scan_ioapic(vcpu);
8399 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8400 vcpu_load_eoi_exitmap(vcpu);
8401 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8402 kvm_vcpu_reload_apic_access_page(vcpu);
8403 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8404 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8405 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8409 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8410 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8411 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8415 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8416 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8417 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8423 * KVM_REQ_HV_STIMER has to be processed after
8424 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8425 * depend on the guest clock being up-to-date
8427 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8428 kvm_hv_process_stimers(vcpu);
8429 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8430 kvm_vcpu_update_apicv(vcpu);
8431 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
8432 kvm_check_async_pf_completion(vcpu);
8435 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8436 ++vcpu->stat.req_event;
8437 kvm_apic_accept_events(vcpu);
8438 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8443 inject_pending_event(vcpu, &req_immediate_exit);
8445 kvm_x86_ops.enable_irq_window(vcpu);
8447 if (kvm_lapic_enabled(vcpu)) {
8448 update_cr8_intercept(vcpu);
8449 kvm_lapic_sync_to_vapic(vcpu);
8453 r = kvm_mmu_reload(vcpu);
8455 goto cancel_injection;
8460 kvm_x86_ops.prepare_guest_switch(vcpu);
8463 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8464 * IPI are then delayed after guest entry, which ensures that they
8465 * result in virtual interrupt delivery.
8467 local_irq_disable();
8468 vcpu->mode = IN_GUEST_MODE;
8470 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8473 * 1) We should set ->mode before checking ->requests. Please see
8474 * the comment in kvm_vcpu_exiting_guest_mode().
8476 * 2) For APICv, we should set ->mode before checking PID.ON. This
8477 * pairs with the memory barrier implicit in pi_test_and_set_on
8478 * (see vmx_deliver_posted_interrupt).
8480 * 3) This also orders the write to mode from any reads to the page
8481 * tables done while the VCPU is running. Please see the comment
8482 * in kvm_flush_remote_tlbs.
8484 smp_mb__after_srcu_read_unlock();
8487 * This handles the case where a posted interrupt was
8488 * notified with kvm_vcpu_kick.
8490 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8491 kvm_x86_ops.sync_pir_to_irr(vcpu);
8493 if (kvm_vcpu_exit_request(vcpu)) {
8494 vcpu->mode = OUTSIDE_GUEST_MODE;
8498 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8500 goto cancel_injection;
8503 if (req_immediate_exit) {
8504 kvm_make_request(KVM_REQ_EVENT, vcpu);
8505 kvm_x86_ops.request_immediate_exit(vcpu);
8508 trace_kvm_entry(vcpu->vcpu_id);
8509 guest_enter_irqoff();
8511 fpregs_assert_state_consistent();
8512 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8513 switch_fpu_return();
8515 if (unlikely(vcpu->arch.switch_db_regs)) {
8517 set_debugreg(vcpu->arch.eff_db[0], 0);
8518 set_debugreg(vcpu->arch.eff_db[1], 1);
8519 set_debugreg(vcpu->arch.eff_db[2], 2);
8520 set_debugreg(vcpu->arch.eff_db[3], 3);
8521 set_debugreg(vcpu->arch.dr6, 6);
8522 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8525 exit_fastpath = kvm_x86_ops.run(vcpu);
8528 * Do this here before restoring debug registers on the host. And
8529 * since we do this before handling the vmexit, a DR access vmexit
8530 * can (a) read the correct value of the debug registers, (b) set
8531 * KVM_DEBUGREG_WONT_EXIT again.
8533 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8534 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8535 kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8536 kvm_update_dr0123(vcpu);
8537 kvm_update_dr7(vcpu);
8538 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8542 * If the guest has used debug registers, at least dr7
8543 * will be disabled while returning to the host.
8544 * If we don't have active breakpoints in the host, we don't
8545 * care about the messed up debug address registers. But if
8546 * we have some of them active, restore the old state.
8548 if (hw_breakpoint_active())
8549 hw_breakpoint_restore();
8551 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8553 vcpu->mode = OUTSIDE_GUEST_MODE;
8556 kvm_x86_ops.handle_exit_irqoff(vcpu);
8559 * Consume any pending interrupts, including the possible source of
8560 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8561 * An instruction is required after local_irq_enable() to fully unblock
8562 * interrupts on processors that implement an interrupt shadow, the
8563 * stat.exits increment will do nicely.
8565 kvm_before_interrupt(vcpu);
8568 local_irq_disable();
8569 kvm_after_interrupt(vcpu);
8571 guest_exit_irqoff();
8572 if (lapic_in_kernel(vcpu)) {
8573 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8574 if (delta != S64_MIN) {
8575 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8576 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8583 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8586 * Profile KVM exit RIPs:
8588 if (unlikely(prof_on == KVM_PROFILING)) {
8589 unsigned long rip = kvm_rip_read(vcpu);
8590 profile_hit(KVM_PROFILING, (void *)rip);
8593 if (unlikely(vcpu->arch.tsc_always_catchup))
8594 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8596 if (vcpu->arch.apic_attention)
8597 kvm_lapic_sync_from_vapic(vcpu);
8599 r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
8603 if (req_immediate_exit)
8604 kvm_make_request(KVM_REQ_EVENT, vcpu);
8605 kvm_x86_ops.cancel_injection(vcpu);
8606 if (unlikely(vcpu->arch.apic_attention))
8607 kvm_lapic_sync_from_vapic(vcpu);
8612 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8614 if (!kvm_arch_vcpu_runnable(vcpu) &&
8615 (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
8616 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8617 kvm_vcpu_block(vcpu);
8618 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8620 if (kvm_x86_ops.post_block)
8621 kvm_x86_ops.post_block(vcpu);
8623 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8627 kvm_apic_accept_events(vcpu);
8628 switch(vcpu->arch.mp_state) {
8629 case KVM_MP_STATE_HALTED:
8630 vcpu->arch.pv.pv_unhalted = false;
8631 vcpu->arch.mp_state =
8632 KVM_MP_STATE_RUNNABLE;
8634 case KVM_MP_STATE_RUNNABLE:
8635 vcpu->arch.apf.halted = false;
8637 case KVM_MP_STATE_INIT_RECEIVED:
8645 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8647 if (is_guest_mode(vcpu))
8648 kvm_x86_ops.nested_ops->check_events(vcpu);
8650 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8651 !vcpu->arch.apf.halted);
8654 static int vcpu_run(struct kvm_vcpu *vcpu)
8657 struct kvm *kvm = vcpu->kvm;
8659 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8660 vcpu->arch.l1tf_flush_l1d = true;
8663 if (kvm_vcpu_running(vcpu)) {
8664 r = vcpu_enter_guest(vcpu);
8666 r = vcpu_block(kvm, vcpu);
8672 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8673 if (kvm_cpu_has_pending_timer(vcpu))
8674 kvm_inject_pending_timer_irqs(vcpu);
8676 if (dm_request_for_irq_injection(vcpu) &&
8677 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8679 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8680 ++vcpu->stat.request_irq_exits;
8684 if (signal_pending(current)) {
8686 vcpu->run->exit_reason = KVM_EXIT_INTR;
8687 ++vcpu->stat.signal_exits;
8690 if (need_resched()) {
8691 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8693 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8697 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8702 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8706 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8707 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8708 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8712 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8714 BUG_ON(!vcpu->arch.pio.count);
8716 return complete_emulated_io(vcpu);
8720 * Implements the following, as a state machine:
8724 * for each mmio piece in the fragment
8732 * for each mmio piece in the fragment
8737 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8739 struct kvm_run *run = vcpu->run;
8740 struct kvm_mmio_fragment *frag;
8743 BUG_ON(!vcpu->mmio_needed);
8745 /* Complete previous fragment */
8746 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8747 len = min(8u, frag->len);
8748 if (!vcpu->mmio_is_write)
8749 memcpy(frag->data, run->mmio.data, len);
8751 if (frag->len <= 8) {
8752 /* Switch to the next fragment. */
8754 vcpu->mmio_cur_fragment++;
8756 /* Go forward to the next mmio piece. */
8762 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8763 vcpu->mmio_needed = 0;
8765 /* FIXME: return into emulator if single-stepping. */
8766 if (vcpu->mmio_is_write)
8768 vcpu->mmio_read_completed = 1;
8769 return complete_emulated_io(vcpu);
8772 run->exit_reason = KVM_EXIT_MMIO;
8773 run->mmio.phys_addr = frag->gpa;
8774 if (vcpu->mmio_is_write)
8775 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8776 run->mmio.len = min(8u, frag->len);
8777 run->mmio.is_write = vcpu->mmio_is_write;
8778 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8782 static void kvm_save_current_fpu(struct fpu *fpu)
8785 * If the target FPU state is not resident in the CPU registers, just
8786 * memcpy() from current, else save CPU state directly to the target.
8788 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8789 memcpy(&fpu->state, ¤t->thread.fpu.state,
8790 fpu_kernel_xstate_size);
8792 copy_fpregs_to_fpstate(fpu);
8795 /* Swap (qemu) user FPU context for the guest FPU context. */
8796 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8800 kvm_save_current_fpu(vcpu->arch.user_fpu);
8802 /* PKRU is separately restored in kvm_x86_ops.run. */
8803 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8804 ~XFEATURE_MASK_PKRU);
8806 fpregs_mark_activate();
8812 /* When vcpu_run ends, restore user space FPU context. */
8813 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8817 kvm_save_current_fpu(vcpu->arch.guest_fpu);
8819 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8821 fpregs_mark_activate();
8824 ++vcpu->stat.fpu_reload;
8828 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
8830 struct kvm_run *kvm_run = vcpu->run;
8834 kvm_sigset_activate(vcpu);
8835 kvm_load_guest_fpu(vcpu);
8837 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8838 if (kvm_run->immediate_exit) {
8842 kvm_vcpu_block(vcpu);
8843 kvm_apic_accept_events(vcpu);
8844 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8846 if (signal_pending(current)) {
8848 kvm_run->exit_reason = KVM_EXIT_INTR;
8849 ++vcpu->stat.signal_exits;
8854 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8859 if (kvm_run->kvm_dirty_regs) {
8860 r = sync_regs(vcpu);
8865 /* re-sync apic's tpr */
8866 if (!lapic_in_kernel(vcpu)) {
8867 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8873 if (unlikely(vcpu->arch.complete_userspace_io)) {
8874 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8875 vcpu->arch.complete_userspace_io = NULL;
8880 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8882 if (kvm_run->immediate_exit)
8888 kvm_put_guest_fpu(vcpu);
8889 if (kvm_run->kvm_valid_regs)
8891 post_kvm_run_save(vcpu);
8892 kvm_sigset_deactivate(vcpu);
8898 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8900 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8902 * We are here if userspace calls get_regs() in the middle of
8903 * instruction emulation. Registers state needs to be copied
8904 * back from emulation context to vcpu. Userspace shouldn't do
8905 * that usually, but some bad designed PV devices (vmware
8906 * backdoor interface) need this to work
8908 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
8909 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8911 regs->rax = kvm_rax_read(vcpu);
8912 regs->rbx = kvm_rbx_read(vcpu);
8913 regs->rcx = kvm_rcx_read(vcpu);
8914 regs->rdx = kvm_rdx_read(vcpu);
8915 regs->rsi = kvm_rsi_read(vcpu);
8916 regs->rdi = kvm_rdi_read(vcpu);
8917 regs->rsp = kvm_rsp_read(vcpu);
8918 regs->rbp = kvm_rbp_read(vcpu);
8919 #ifdef CONFIG_X86_64
8920 regs->r8 = kvm_r8_read(vcpu);
8921 regs->r9 = kvm_r9_read(vcpu);
8922 regs->r10 = kvm_r10_read(vcpu);
8923 regs->r11 = kvm_r11_read(vcpu);
8924 regs->r12 = kvm_r12_read(vcpu);
8925 regs->r13 = kvm_r13_read(vcpu);
8926 regs->r14 = kvm_r14_read(vcpu);
8927 regs->r15 = kvm_r15_read(vcpu);
8930 regs->rip = kvm_rip_read(vcpu);
8931 regs->rflags = kvm_get_rflags(vcpu);
8934 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8937 __get_regs(vcpu, regs);
8942 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8944 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8945 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8947 kvm_rax_write(vcpu, regs->rax);
8948 kvm_rbx_write(vcpu, regs->rbx);
8949 kvm_rcx_write(vcpu, regs->rcx);
8950 kvm_rdx_write(vcpu, regs->rdx);
8951 kvm_rsi_write(vcpu, regs->rsi);
8952 kvm_rdi_write(vcpu, regs->rdi);
8953 kvm_rsp_write(vcpu, regs->rsp);
8954 kvm_rbp_write(vcpu, regs->rbp);
8955 #ifdef CONFIG_X86_64
8956 kvm_r8_write(vcpu, regs->r8);
8957 kvm_r9_write(vcpu, regs->r9);
8958 kvm_r10_write(vcpu, regs->r10);
8959 kvm_r11_write(vcpu, regs->r11);
8960 kvm_r12_write(vcpu, regs->r12);
8961 kvm_r13_write(vcpu, regs->r13);
8962 kvm_r14_write(vcpu, regs->r14);
8963 kvm_r15_write(vcpu, regs->r15);
8966 kvm_rip_write(vcpu, regs->rip);
8967 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8969 vcpu->arch.exception.pending = false;
8971 kvm_make_request(KVM_REQ_EVENT, vcpu);
8974 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8977 __set_regs(vcpu, regs);
8982 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8984 struct kvm_segment cs;
8986 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8990 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8992 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8996 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8997 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8998 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8999 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9000 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9001 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9003 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9004 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9006 kvm_x86_ops.get_idt(vcpu, &dt);
9007 sregs->idt.limit = dt.size;
9008 sregs->idt.base = dt.address;
9009 kvm_x86_ops.get_gdt(vcpu, &dt);
9010 sregs->gdt.limit = dt.size;
9011 sregs->gdt.base = dt.address;
9013 sregs->cr0 = kvm_read_cr0(vcpu);
9014 sregs->cr2 = vcpu->arch.cr2;
9015 sregs->cr3 = kvm_read_cr3(vcpu);
9016 sregs->cr4 = kvm_read_cr4(vcpu);
9017 sregs->cr8 = kvm_get_cr8(vcpu);
9018 sregs->efer = vcpu->arch.efer;
9019 sregs->apic_base = kvm_get_apic_base(vcpu);
9021 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9023 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9024 set_bit(vcpu->arch.interrupt.nr,
9025 (unsigned long *)sregs->interrupt_bitmap);
9028 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9029 struct kvm_sregs *sregs)
9032 __get_sregs(vcpu, sregs);
9037 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9038 struct kvm_mp_state *mp_state)
9041 if (kvm_mpx_supported())
9042 kvm_load_guest_fpu(vcpu);
9044 kvm_apic_accept_events(vcpu);
9045 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
9046 vcpu->arch.pv.pv_unhalted)
9047 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9049 mp_state->mp_state = vcpu->arch.mp_state;
9051 if (kvm_mpx_supported())
9052 kvm_put_guest_fpu(vcpu);
9057 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9058 struct kvm_mp_state *mp_state)
9064 if (!lapic_in_kernel(vcpu) &&
9065 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9069 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9070 * INIT state; latched init should be reported using
9071 * KVM_SET_VCPU_EVENTS, so reject it here.
9073 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9074 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9075 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9078 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9079 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9080 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9082 vcpu->arch.mp_state = mp_state->mp_state;
9083 kvm_make_request(KVM_REQ_EVENT, vcpu);
9091 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9092 int reason, bool has_error_code, u32 error_code)
9094 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9097 init_emulate_ctxt(vcpu);
9099 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9100 has_error_code, error_code);
9102 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9103 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9104 vcpu->run->internal.ndata = 0;
9108 kvm_rip_write(vcpu, ctxt->eip);
9109 kvm_set_rflags(vcpu, ctxt->eflags);
9112 EXPORT_SYMBOL_GPL(kvm_task_switch);
9114 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9116 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9118 * When EFER.LME and CR0.PG are set, the processor is in
9119 * 64-bit mode (though maybe in a 32-bit code segment).
9120 * CR4.PAE and EFER.LMA must be set.
9122 if (!(sregs->cr4 & X86_CR4_PAE)
9123 || !(sregs->efer & EFER_LMA))
9127 * Not in 64-bit mode: EFER.LMA is clear and the code
9128 * segment cannot be 64-bit.
9130 if (sregs->efer & EFER_LMA || sregs->cs.l)
9134 return kvm_valid_cr4(vcpu, sregs->cr4);
9137 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9139 struct msr_data apic_base_msr;
9140 int mmu_reset_needed = 0;
9141 int cpuid_update_needed = 0;
9142 int pending_vec, max_bits, idx;
9146 if (kvm_valid_sregs(vcpu, sregs))
9149 apic_base_msr.data = sregs->apic_base;
9150 apic_base_msr.host_initiated = true;
9151 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9154 dt.size = sregs->idt.limit;
9155 dt.address = sregs->idt.base;
9156 kvm_x86_ops.set_idt(vcpu, &dt);
9157 dt.size = sregs->gdt.limit;
9158 dt.address = sregs->gdt.base;
9159 kvm_x86_ops.set_gdt(vcpu, &dt);
9161 vcpu->arch.cr2 = sregs->cr2;
9162 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9163 vcpu->arch.cr3 = sregs->cr3;
9164 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9166 kvm_set_cr8(vcpu, sregs->cr8);
9168 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9169 kvm_x86_ops.set_efer(vcpu, sregs->efer);
9171 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9172 kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9173 vcpu->arch.cr0 = sregs->cr0;
9175 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9176 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9177 (X86_CR4_OSXSAVE | X86_CR4_PKE));
9178 kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9179 if (cpuid_update_needed)
9180 kvm_update_cpuid(vcpu);
9182 idx = srcu_read_lock(&vcpu->kvm->srcu);
9183 if (is_pae_paging(vcpu)) {
9184 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9185 mmu_reset_needed = 1;
9187 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9189 if (mmu_reset_needed)
9190 kvm_mmu_reset_context(vcpu);
9192 max_bits = KVM_NR_INTERRUPTS;
9193 pending_vec = find_first_bit(
9194 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9195 if (pending_vec < max_bits) {
9196 kvm_queue_interrupt(vcpu, pending_vec, false);
9197 pr_debug("Set back pending irq %d\n", pending_vec);
9200 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9201 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9202 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9203 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9204 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9205 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9207 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9208 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9210 update_cr8_intercept(vcpu);
9212 /* Older userspace won't unhalt the vcpu on reset. */
9213 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9214 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9216 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9218 kvm_make_request(KVM_REQ_EVENT, vcpu);
9225 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9226 struct kvm_sregs *sregs)
9231 ret = __set_sregs(vcpu, sregs);
9236 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9237 struct kvm_guest_debug *dbg)
9239 unsigned long rflags;
9244 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9246 if (vcpu->arch.exception.pending)
9248 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9249 kvm_queue_exception(vcpu, DB_VECTOR);
9251 kvm_queue_exception(vcpu, BP_VECTOR);
9255 * Read rflags as long as potentially injected trace flags are still
9258 rflags = kvm_get_rflags(vcpu);
9260 vcpu->guest_debug = dbg->control;
9261 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9262 vcpu->guest_debug = 0;
9264 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9265 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9266 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9267 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9269 for (i = 0; i < KVM_NR_DB_REGS; i++)
9270 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9272 kvm_update_dr7(vcpu);
9274 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9275 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9276 get_segment_base(vcpu, VCPU_SREG_CS);
9279 * Trigger an rflags update that will inject or remove the trace
9282 kvm_set_rflags(vcpu, rflags);
9284 kvm_x86_ops.update_bp_intercept(vcpu);
9294 * Translate a guest virtual address to a guest physical address.
9296 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9297 struct kvm_translation *tr)
9299 unsigned long vaddr = tr->linear_address;
9305 idx = srcu_read_lock(&vcpu->kvm->srcu);
9306 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9307 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9308 tr->physical_address = gpa;
9309 tr->valid = gpa != UNMAPPED_GVA;
9317 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9319 struct fxregs_state *fxsave;
9323 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9324 memcpy(fpu->fpr, fxsave->st_space, 128);
9325 fpu->fcw = fxsave->cwd;
9326 fpu->fsw = fxsave->swd;
9327 fpu->ftwx = fxsave->twd;
9328 fpu->last_opcode = fxsave->fop;
9329 fpu->last_ip = fxsave->rip;
9330 fpu->last_dp = fxsave->rdp;
9331 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9337 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9339 struct fxregs_state *fxsave;
9343 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9345 memcpy(fxsave->st_space, fpu->fpr, 128);
9346 fxsave->cwd = fpu->fcw;
9347 fxsave->swd = fpu->fsw;
9348 fxsave->twd = fpu->ftwx;
9349 fxsave->fop = fpu->last_opcode;
9350 fxsave->rip = fpu->last_ip;
9351 fxsave->rdp = fpu->last_dp;
9352 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9358 static void store_regs(struct kvm_vcpu *vcpu)
9360 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9362 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9363 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9365 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9366 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9368 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9369 kvm_vcpu_ioctl_x86_get_vcpu_events(
9370 vcpu, &vcpu->run->s.regs.events);
9373 static int sync_regs(struct kvm_vcpu *vcpu)
9375 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9378 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9379 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9380 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9382 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9383 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9385 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9387 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9388 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9389 vcpu, &vcpu->run->s.regs.events))
9391 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9397 static void fx_init(struct kvm_vcpu *vcpu)
9399 fpstate_init(&vcpu->arch.guest_fpu->state);
9400 if (boot_cpu_has(X86_FEATURE_XSAVES))
9401 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9402 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9405 * Ensure guest xcr0 is valid for loading
9407 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9409 vcpu->arch.cr0 |= X86_CR0_ET;
9412 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9414 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9415 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9416 "guest TSC will not be reliable\n");
9421 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9426 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9427 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9429 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9431 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9433 r = kvm_mmu_create(vcpu);
9437 if (irqchip_in_kernel(vcpu->kvm)) {
9438 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9440 goto fail_mmu_destroy;
9441 if (kvm_apicv_activated(vcpu->kvm))
9442 vcpu->arch.apicv_active = true;
9444 static_key_slow_inc(&kvm_no_apic_vcpu);
9448 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9450 goto fail_free_lapic;
9451 vcpu->arch.pio_data = page_address(page);
9453 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9454 GFP_KERNEL_ACCOUNT);
9455 if (!vcpu->arch.mce_banks)
9456 goto fail_free_pio_data;
9457 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9459 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9460 GFP_KERNEL_ACCOUNT))
9461 goto fail_free_mce_banks;
9463 if (!alloc_emulate_ctxt(vcpu))
9464 goto free_wbinvd_dirty_mask;
9466 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9467 GFP_KERNEL_ACCOUNT);
9468 if (!vcpu->arch.user_fpu) {
9469 pr_err("kvm: failed to allocate userspace's fpu\n");
9470 goto free_emulate_ctxt;
9473 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9474 GFP_KERNEL_ACCOUNT);
9475 if (!vcpu->arch.guest_fpu) {
9476 pr_err("kvm: failed to allocate vcpu's fpu\n");
9481 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9482 vcpu->arch.tdp_level = kvm_x86_ops.get_tdp_level(vcpu);
9484 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9486 kvm_async_pf_hash_reset(vcpu);
9489 vcpu->arch.pending_external_vector = -1;
9490 vcpu->arch.preempted_in_kernel = false;
9492 kvm_hv_vcpu_init(vcpu);
9494 r = kvm_x86_ops.vcpu_create(vcpu);
9496 goto free_guest_fpu;
9498 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9499 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9500 kvm_vcpu_mtrr_init(vcpu);
9502 kvm_vcpu_reset(vcpu, false);
9503 kvm_init_mmu(vcpu, false);
9508 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9510 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9512 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9513 free_wbinvd_dirty_mask:
9514 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9515 fail_free_mce_banks:
9516 kfree(vcpu->arch.mce_banks);
9518 free_page((unsigned long)vcpu->arch.pio_data);
9520 kvm_free_lapic(vcpu);
9522 kvm_mmu_destroy(vcpu);
9526 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9528 struct msr_data msr;
9529 struct kvm *kvm = vcpu->kvm;
9531 kvm_hv_vcpu_postcreate(vcpu);
9533 if (mutex_lock_killable(&vcpu->mutex))
9537 msr.index = MSR_IA32_TSC;
9538 msr.host_initiated = true;
9539 kvm_write_tsc(vcpu, &msr);
9542 /* poll control enabled by default */
9543 vcpu->arch.msr_kvm_poll_control = 1;
9545 mutex_unlock(&vcpu->mutex);
9547 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9548 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9549 KVMCLOCK_SYNC_PERIOD);
9552 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9554 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9557 kvm_release_pfn(cache->pfn, cache->dirty, cache);
9559 kvmclock_reset(vcpu);
9561 kvm_x86_ops.vcpu_free(vcpu);
9563 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9564 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9565 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9566 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9568 kvm_hv_vcpu_uninit(vcpu);
9569 kvm_pmu_destroy(vcpu);
9570 kfree(vcpu->arch.mce_banks);
9571 kvm_free_lapic(vcpu);
9572 idx = srcu_read_lock(&vcpu->kvm->srcu);
9573 kvm_mmu_destroy(vcpu);
9574 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9575 free_page((unsigned long)vcpu->arch.pio_data);
9576 if (!lapic_in_kernel(vcpu))
9577 static_key_slow_dec(&kvm_no_apic_vcpu);
9580 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9582 kvm_lapic_reset(vcpu, init_event);
9584 vcpu->arch.hflags = 0;
9586 vcpu->arch.smi_pending = 0;
9587 vcpu->arch.smi_count = 0;
9588 atomic_set(&vcpu->arch.nmi_queued, 0);
9589 vcpu->arch.nmi_pending = 0;
9590 vcpu->arch.nmi_injected = false;
9591 kvm_clear_interrupt_queue(vcpu);
9592 kvm_clear_exception_queue(vcpu);
9594 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9595 kvm_update_dr0123(vcpu);
9596 vcpu->arch.dr6 = DR6_INIT;
9597 vcpu->arch.dr7 = DR7_FIXED_1;
9598 kvm_update_dr7(vcpu);
9602 kvm_make_request(KVM_REQ_EVENT, vcpu);
9603 vcpu->arch.apf.msr_en_val = 0;
9604 vcpu->arch.apf.msr_int_val = 0;
9605 vcpu->arch.st.msr_val = 0;
9607 kvmclock_reset(vcpu);
9609 kvm_clear_async_pf_completion_queue(vcpu);
9610 kvm_async_pf_hash_reset(vcpu);
9611 vcpu->arch.apf.halted = false;
9613 if (kvm_mpx_supported()) {
9614 void *mpx_state_buffer;
9617 * To avoid have the INIT path from kvm_apic_has_events() that be
9618 * called with loaded FPU and does not let userspace fix the state.
9621 kvm_put_guest_fpu(vcpu);
9622 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9624 if (mpx_state_buffer)
9625 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9626 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9628 if (mpx_state_buffer)
9629 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9631 kvm_load_guest_fpu(vcpu);
9635 kvm_pmu_reset(vcpu);
9636 vcpu->arch.smbase = 0x30000;
9638 vcpu->arch.msr_misc_features_enables = 0;
9640 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9643 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9644 vcpu->arch.regs_avail = ~0;
9645 vcpu->arch.regs_dirty = ~0;
9647 vcpu->arch.ia32_xss = 0;
9649 kvm_x86_ops.vcpu_reset(vcpu, init_event);
9652 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9654 struct kvm_segment cs;
9656 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9657 cs.selector = vector << 8;
9658 cs.base = vector << 12;
9659 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9660 kvm_rip_write(vcpu, 0);
9663 int kvm_arch_hardware_enable(void)
9666 struct kvm_vcpu *vcpu;
9671 bool stable, backwards_tsc = false;
9673 kvm_shared_msr_cpu_online();
9674 ret = kvm_x86_ops.hardware_enable();
9678 local_tsc = rdtsc();
9679 stable = !kvm_check_tsc_unstable();
9680 list_for_each_entry(kvm, &vm_list, vm_list) {
9681 kvm_for_each_vcpu(i, vcpu, kvm) {
9682 if (!stable && vcpu->cpu == smp_processor_id())
9683 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9684 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9685 backwards_tsc = true;
9686 if (vcpu->arch.last_host_tsc > max_tsc)
9687 max_tsc = vcpu->arch.last_host_tsc;
9693 * Sometimes, even reliable TSCs go backwards. This happens on
9694 * platforms that reset TSC during suspend or hibernate actions, but
9695 * maintain synchronization. We must compensate. Fortunately, we can
9696 * detect that condition here, which happens early in CPU bringup,
9697 * before any KVM threads can be running. Unfortunately, we can't
9698 * bring the TSCs fully up to date with real time, as we aren't yet far
9699 * enough into CPU bringup that we know how much real time has actually
9700 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9701 * variables that haven't been updated yet.
9703 * So we simply find the maximum observed TSC above, then record the
9704 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9705 * the adjustment will be applied. Note that we accumulate
9706 * adjustments, in case multiple suspend cycles happen before some VCPU
9707 * gets a chance to run again. In the event that no KVM threads get a
9708 * chance to run, we will miss the entire elapsed period, as we'll have
9709 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9710 * loose cycle time. This isn't too big a deal, since the loss will be
9711 * uniform across all VCPUs (not to mention the scenario is extremely
9712 * unlikely). It is possible that a second hibernate recovery happens
9713 * much faster than a first, causing the observed TSC here to be
9714 * smaller; this would require additional padding adjustment, which is
9715 * why we set last_host_tsc to the local tsc observed here.
9717 * N.B. - this code below runs only on platforms with reliable TSC,
9718 * as that is the only way backwards_tsc is set above. Also note
9719 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9720 * have the same delta_cyc adjustment applied if backwards_tsc
9721 * is detected. Note further, this adjustment is only done once,
9722 * as we reset last_host_tsc on all VCPUs to stop this from being
9723 * called multiple times (one for each physical CPU bringup).
9725 * Platforms with unreliable TSCs don't have to deal with this, they
9726 * will be compensated by the logic in vcpu_load, which sets the TSC to
9727 * catchup mode. This will catchup all VCPUs to real time, but cannot
9728 * guarantee that they stay in perfect synchronization.
9730 if (backwards_tsc) {
9731 u64 delta_cyc = max_tsc - local_tsc;
9732 list_for_each_entry(kvm, &vm_list, vm_list) {
9733 kvm->arch.backwards_tsc_observed = true;
9734 kvm_for_each_vcpu(i, vcpu, kvm) {
9735 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9736 vcpu->arch.last_host_tsc = local_tsc;
9737 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9741 * We have to disable TSC offset matching.. if you were
9742 * booting a VM while issuing an S4 host suspend....
9743 * you may have some problem. Solving this issue is
9744 * left as an exercise to the reader.
9746 kvm->arch.last_tsc_nsec = 0;
9747 kvm->arch.last_tsc_write = 0;
9754 void kvm_arch_hardware_disable(void)
9756 kvm_x86_ops.hardware_disable();
9757 drop_user_return_notifiers();
9760 int kvm_arch_hardware_setup(void *opaque)
9762 struct kvm_x86_init_ops *ops = opaque;
9765 rdmsrl_safe(MSR_EFER, &host_efer);
9767 if (boot_cpu_has(X86_FEATURE_XSAVES))
9768 rdmsrl(MSR_IA32_XSS, host_xss);
9770 r = ops->hardware_setup();
9774 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9776 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9779 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
9780 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
9781 #undef __kvm_cpu_cap_has
9783 if (kvm_has_tsc_control) {
9785 * Make sure the user can only configure tsc_khz values that
9786 * fit into a signed integer.
9787 * A min value is not calculated because it will always
9788 * be 1 on all machines.
9790 u64 max = min(0x7fffffffULL,
9791 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9792 kvm_max_guest_tsc_khz = max;
9794 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9797 kvm_init_msr_list();
9801 void kvm_arch_hardware_unsetup(void)
9803 kvm_x86_ops.hardware_unsetup();
9806 int kvm_arch_check_processor_compat(void *opaque)
9808 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
9809 struct kvm_x86_init_ops *ops = opaque;
9811 WARN_ON(!irqs_disabled());
9813 if (__cr4_reserved_bits(cpu_has, c) !=
9814 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
9817 return ops->check_processor_compatibility();
9820 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9822 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9824 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9826 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9828 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9831 struct static_key kvm_no_apic_vcpu __read_mostly;
9832 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9834 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9836 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9838 vcpu->arch.l1tf_flush_l1d = true;
9839 if (pmu->version && unlikely(pmu->event_count)) {
9840 pmu->need_cleanup = true;
9841 kvm_make_request(KVM_REQ_PMU, vcpu);
9843 kvm_x86_ops.sched_in(vcpu, cpu);
9846 void kvm_arch_free_vm(struct kvm *kvm)
9848 kfree(kvm->arch.hyperv.hv_pa_pg);
9853 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9858 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9859 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9860 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9861 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9862 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9863 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9865 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9866 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9867 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9868 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9869 &kvm->arch.irq_sources_bitmap);
9871 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9872 mutex_init(&kvm->arch.apic_map_lock);
9873 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9875 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
9876 pvclock_update_vm_gtod_copy(kvm);
9878 kvm->arch.guest_can_read_msr_platform_info = true;
9880 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9881 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9883 kvm_hv_init_vm(kvm);
9884 kvm_page_track_init(kvm);
9885 kvm_mmu_init_vm(kvm);
9887 return kvm_x86_ops.vm_init(kvm);
9890 int kvm_arch_post_init_vm(struct kvm *kvm)
9892 return kvm_mmu_post_init_vm(kvm);
9895 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9898 kvm_mmu_unload(vcpu);
9902 static void kvm_free_vcpus(struct kvm *kvm)
9905 struct kvm_vcpu *vcpu;
9908 * Unpin any mmu pages first.
9910 kvm_for_each_vcpu(i, vcpu, kvm) {
9911 kvm_clear_async_pf_completion_queue(vcpu);
9912 kvm_unload_vcpu_mmu(vcpu);
9914 kvm_for_each_vcpu(i, vcpu, kvm)
9915 kvm_vcpu_destroy(vcpu);
9917 mutex_lock(&kvm->lock);
9918 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9919 kvm->vcpus[i] = NULL;
9921 atomic_set(&kvm->online_vcpus, 0);
9922 mutex_unlock(&kvm->lock);
9925 void kvm_arch_sync_events(struct kvm *kvm)
9927 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9928 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9932 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9935 unsigned long hva, uninitialized_var(old_npages);
9936 struct kvm_memslots *slots = kvm_memslots(kvm);
9937 struct kvm_memory_slot *slot;
9939 /* Called with kvm->slots_lock held. */
9940 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9943 slot = id_to_memslot(slots, id);
9945 if (slot && slot->npages)
9949 * MAP_SHARED to prevent internal slot pages from being moved
9952 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9953 MAP_SHARED | MAP_ANONYMOUS, 0);
9954 if (IS_ERR((void *)hva))
9955 return PTR_ERR((void *)hva);
9957 if (!slot || !slot->npages)
9960 old_npages = slot->npages;
9964 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9965 struct kvm_userspace_memory_region m;
9967 m.slot = id | (i << 16);
9969 m.guest_phys_addr = gpa;
9970 m.userspace_addr = hva;
9971 m.memory_size = size;
9972 r = __kvm_set_memory_region(kvm, &m);
9978 vm_munmap(hva, old_npages * PAGE_SIZE);
9982 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9984 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9986 kvm_mmu_pre_destroy_vm(kvm);
9989 void kvm_arch_destroy_vm(struct kvm *kvm)
9991 if (current->mm == kvm->mm) {
9993 * Free memory regions allocated on behalf of userspace,
9994 * unless the the memory map has changed due to process exit
9997 mutex_lock(&kvm->slots_lock);
9998 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10000 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10002 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10003 mutex_unlock(&kvm->slots_lock);
10005 if (kvm_x86_ops.vm_destroy)
10006 kvm_x86_ops.vm_destroy(kvm);
10007 kvm_pic_destroy(kvm);
10008 kvm_ioapic_destroy(kvm);
10009 kvm_free_vcpus(kvm);
10010 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10011 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10012 kvm_mmu_uninit_vm(kvm);
10013 kvm_page_track_cleanup(kvm);
10014 kvm_hv_destroy_vm(kvm);
10017 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10021 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10022 kvfree(slot->arch.rmap[i]);
10023 slot->arch.rmap[i] = NULL;
10028 kvfree(slot->arch.lpage_info[i - 1]);
10029 slot->arch.lpage_info[i - 1] = NULL;
10032 kvm_page_track_free_memslot(slot);
10035 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10036 unsigned long npages)
10041 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10042 * old arrays will be freed by __kvm_set_memory_region() if installing
10043 * the new memslot is successful.
10045 memset(&slot->arch, 0, sizeof(slot->arch));
10047 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10048 struct kvm_lpage_info *linfo;
10049 unsigned long ugfn;
10053 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10054 slot->base_gfn, level) + 1;
10056 slot->arch.rmap[i] =
10057 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10058 GFP_KERNEL_ACCOUNT);
10059 if (!slot->arch.rmap[i])
10064 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10068 slot->arch.lpage_info[i - 1] = linfo;
10070 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10071 linfo[0].disallow_lpage = 1;
10072 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10073 linfo[lpages - 1].disallow_lpage = 1;
10074 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10076 * If the gfn and userspace address are not aligned wrt each
10077 * other, disable large page support for this slot.
10079 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10082 for (j = 0; j < lpages; ++j)
10083 linfo[j].disallow_lpage = 1;
10087 if (kvm_page_track_create_memslot(slot, npages))
10093 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10094 kvfree(slot->arch.rmap[i]);
10095 slot->arch.rmap[i] = NULL;
10099 kvfree(slot->arch.lpage_info[i - 1]);
10100 slot->arch.lpage_info[i - 1] = NULL;
10105 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10107 struct kvm_vcpu *vcpu;
10111 * memslots->generation has been incremented.
10112 * mmio generation may have reached its maximum value.
10114 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10116 /* Force re-initialization of steal_time cache */
10117 kvm_for_each_vcpu(i, vcpu, kvm)
10118 kvm_vcpu_kick(vcpu);
10121 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10122 struct kvm_memory_slot *memslot,
10123 const struct kvm_userspace_memory_region *mem,
10124 enum kvm_mr_change change)
10126 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10127 return kvm_alloc_memslot_metadata(memslot,
10128 mem->memory_size >> PAGE_SHIFT);
10132 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10133 struct kvm_memory_slot *old,
10134 struct kvm_memory_slot *new,
10135 enum kvm_mr_change change)
10138 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
10139 * See comments below.
10141 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10145 * Dirty logging tracks sptes in 4k granularity, meaning that large
10146 * sptes have to be split. If live migration is successful, the guest
10147 * in the source machine will be destroyed and large sptes will be
10148 * created in the destination. However, if the guest continues to run
10149 * in the source machine (for example if live migration fails), small
10150 * sptes will remain around and cause bad performance.
10152 * Scan sptes if dirty logging has been stopped, dropping those
10153 * which can be collapsed into a single large-page spte. Later
10154 * page faults will create the large-page sptes.
10156 * There is no need to do this in any of the following cases:
10157 * CREATE: No dirty mappings will already exist.
10158 * MOVE/DELETE: The old mappings will already have been cleaned up by
10159 * kvm_arch_flush_shadow_memslot()
10161 if ((old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10162 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10163 kvm_mmu_zap_collapsible_sptes(kvm, new);
10166 * Enable or disable dirty logging for the slot.
10168 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old
10169 * slot have been zapped so no dirty logging updates are needed for
10171 * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible
10172 * any mappings that might be created in it will consume the
10173 * properties of the new slot and do not need to be updated here.
10175 * When PML is enabled, the kvm_x86_ops dirty logging hooks are
10176 * called to enable/disable dirty logging.
10178 * When disabling dirty logging with PML enabled, the D-bit is set
10179 * for sptes in the slot in order to prevent unnecessary GPA
10180 * logging in the PML buffer (and potential PML buffer full VMEXIT).
10181 * This guarantees leaving PML enabled for the guest's lifetime
10182 * won't have any additional overhead from PML when the guest is
10183 * running with dirty logging disabled.
10185 * When enabling dirty logging, large sptes are write-protected
10186 * so they can be split on first write. New large sptes cannot
10187 * be created for this slot until the end of the logging.
10188 * See the comments in fast_page_fault().
10189 * For small sptes, nothing is done if the dirty log is in the
10190 * initial-all-set state. Otherwise, depending on whether pml
10191 * is enabled the D-bit or the W-bit will be cleared.
10193 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10194 if (kvm_x86_ops.slot_enable_log_dirty) {
10195 kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10198 kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10199 PG_LEVEL_2M : PG_LEVEL_4K;
10202 * If we're with initial-all-set, we don't need
10203 * to write protect any small page because
10204 * they're reported as dirty already. However
10205 * we still need to write-protect huge pages
10206 * so that the page split can happen lazily on
10207 * the first write to the huge page.
10209 kvm_mmu_slot_remove_write_access(kvm, new, level);
10212 if (kvm_x86_ops.slot_disable_log_dirty)
10213 kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10217 void kvm_arch_commit_memory_region(struct kvm *kvm,
10218 const struct kvm_userspace_memory_region *mem,
10219 struct kvm_memory_slot *old,
10220 const struct kvm_memory_slot *new,
10221 enum kvm_mr_change change)
10223 if (!kvm->arch.n_requested_mmu_pages)
10224 kvm_mmu_change_mmu_pages(kvm,
10225 kvm_mmu_calculate_default_mmu_pages(kvm));
10228 * FIXME: const-ify all uses of struct kvm_memory_slot.
10230 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
10232 /* Free the arrays associated with the old memslot. */
10233 if (change == KVM_MR_MOVE)
10234 kvm_arch_free_memslot(kvm, old);
10237 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10239 kvm_mmu_zap_all(kvm);
10242 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10243 struct kvm_memory_slot *slot)
10245 kvm_page_track_flush_slot(kvm, slot);
10248 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10250 return (is_guest_mode(vcpu) &&
10251 kvm_x86_ops.guest_apic_has_interrupt &&
10252 kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10255 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10257 if (!list_empty_careful(&vcpu->async_pf.done))
10260 if (kvm_apic_has_events(vcpu))
10263 if (vcpu->arch.pv.pv_unhalted)
10266 if (vcpu->arch.exception.pending)
10269 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10270 (vcpu->arch.nmi_pending &&
10271 kvm_x86_ops.nmi_allowed(vcpu, false)))
10274 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10275 (vcpu->arch.smi_pending &&
10276 kvm_x86_ops.smi_allowed(vcpu, false)))
10279 if (kvm_arch_interrupt_allowed(vcpu) &&
10280 (kvm_cpu_has_interrupt(vcpu) ||
10281 kvm_guest_apic_has_interrupt(vcpu)))
10284 if (kvm_hv_has_stimer_pending(vcpu))
10287 if (is_guest_mode(vcpu) &&
10288 kvm_x86_ops.nested_ops->hv_timer_pending &&
10289 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10295 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10297 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10300 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10302 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10305 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10306 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10307 kvm_test_request(KVM_REQ_EVENT, vcpu))
10310 if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10316 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10318 return vcpu->arch.preempted_in_kernel;
10321 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10323 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10326 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10328 return kvm_x86_ops.interrupt_allowed(vcpu, false);
10331 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10333 if (is_64_bit_mode(vcpu))
10334 return kvm_rip_read(vcpu);
10335 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10336 kvm_rip_read(vcpu));
10338 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10340 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10342 return kvm_get_linear_rip(vcpu) == linear_rip;
10344 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10346 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10348 unsigned long rflags;
10350 rflags = kvm_x86_ops.get_rflags(vcpu);
10351 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10352 rflags &= ~X86_EFLAGS_TF;
10355 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10357 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10359 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10360 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10361 rflags |= X86_EFLAGS_TF;
10362 kvm_x86_ops.set_rflags(vcpu, rflags);
10365 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10367 __kvm_set_rflags(vcpu, rflags);
10368 kvm_make_request(KVM_REQ_EVENT, vcpu);
10370 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10372 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10376 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10380 r = kvm_mmu_reload(vcpu);
10384 if (!vcpu->arch.mmu->direct_map &&
10385 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10388 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10391 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10393 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
10395 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10398 static inline u32 kvm_async_pf_next_probe(u32 key)
10400 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
10403 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10405 u32 key = kvm_async_pf_hash_fn(gfn);
10407 while (vcpu->arch.apf.gfns[key] != ~0)
10408 key = kvm_async_pf_next_probe(key);
10410 vcpu->arch.apf.gfns[key] = gfn;
10413 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10416 u32 key = kvm_async_pf_hash_fn(gfn);
10418 for (i = 0; i < ASYNC_PF_PER_VCPU &&
10419 (vcpu->arch.apf.gfns[key] != gfn &&
10420 vcpu->arch.apf.gfns[key] != ~0); i++)
10421 key = kvm_async_pf_next_probe(key);
10426 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10428 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10431 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10435 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10437 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
10441 vcpu->arch.apf.gfns[i] = ~0;
10443 j = kvm_async_pf_next_probe(j);
10444 if (vcpu->arch.apf.gfns[j] == ~0)
10446 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10448 * k lies cyclically in ]i,j]
10450 * |....j i.k.| or |.k..j i...|
10452 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10453 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10458 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
10460 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
10462 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
10466 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
10468 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
10470 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
10471 &token, offset, sizeof(token));
10474 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
10476 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
10479 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
10480 &val, offset, sizeof(val)))
10486 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10488 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10491 if (!kvm_pv_async_pf_enabled(vcpu) ||
10492 (vcpu->arch.apf.send_user_only && kvm_x86_ops.get_cpl(vcpu) == 0))
10498 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10500 if (unlikely(!lapic_in_kernel(vcpu) ||
10501 kvm_event_needs_reinjection(vcpu) ||
10502 vcpu->arch.exception.pending))
10505 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10509 * If interrupts are off we cannot even use an artificial
10512 return kvm_arch_interrupt_allowed(vcpu);
10515 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10516 struct kvm_async_pf *work)
10518 struct x86_exception fault;
10520 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10521 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10523 if (kvm_can_deliver_async_pf(vcpu) &&
10524 !apf_put_user_notpresent(vcpu)) {
10525 fault.vector = PF_VECTOR;
10526 fault.error_code_valid = true;
10527 fault.error_code = 0;
10528 fault.nested_page_fault = false;
10529 fault.address = work->arch.token;
10530 fault.async_page_fault = true;
10531 kvm_inject_page_fault(vcpu, &fault);
10535 * It is not possible to deliver a paravirtualized asynchronous
10536 * page fault, but putting the guest in an artificial halt state
10537 * can be beneficial nevertheless: if an interrupt arrives, we
10538 * can deliver it timely and perhaps the guest will schedule
10539 * another process. When the instruction that triggered a page
10540 * fault is retried, hopefully the page will be ready in the host.
10542 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10547 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10548 struct kvm_async_pf *work)
10550 struct kvm_lapic_irq irq = {
10551 .delivery_mode = APIC_DM_FIXED,
10552 .vector = vcpu->arch.apf.vec
10555 if (work->wakeup_all)
10556 work->arch.token = ~0; /* broadcast wakeup */
10558 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10559 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10561 if ((work->wakeup_all || work->notpresent_injected) &&
10562 kvm_pv_async_pf_enabled(vcpu) &&
10563 !apf_put_user_ready(vcpu, work->arch.token)) {
10564 vcpu->arch.apf.pageready_pending = true;
10565 kvm_apic_set_irq(vcpu, &irq, NULL);
10568 vcpu->arch.apf.halted = false;
10569 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10572 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
10574 kvm_make_request(KVM_REQ_APF_READY, vcpu);
10575 if (!vcpu->arch.apf.pageready_pending)
10576 kvm_vcpu_kick(vcpu);
10579 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
10581 if (!kvm_pv_async_pf_enabled(vcpu))
10584 return apf_pageready_slot_free(vcpu);
10587 void kvm_arch_start_assignment(struct kvm *kvm)
10589 atomic_inc(&kvm->arch.assigned_device_count);
10591 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10593 void kvm_arch_end_assignment(struct kvm *kvm)
10595 atomic_dec(&kvm->arch.assigned_device_count);
10597 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10599 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10601 return atomic_read(&kvm->arch.assigned_device_count);
10603 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10605 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10607 atomic_inc(&kvm->arch.noncoherent_dma_count);
10609 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10611 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10613 atomic_dec(&kvm->arch.noncoherent_dma_count);
10615 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10617 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10619 return atomic_read(&kvm->arch.noncoherent_dma_count);
10621 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10623 bool kvm_arch_has_irq_bypass(void)
10628 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10629 struct irq_bypass_producer *prod)
10631 struct kvm_kernel_irqfd *irqfd =
10632 container_of(cons, struct kvm_kernel_irqfd, consumer);
10634 irqfd->producer = prod;
10636 return kvm_x86_ops.update_pi_irte(irqfd->kvm,
10637 prod->irq, irqfd->gsi, 1);
10640 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10641 struct irq_bypass_producer *prod)
10644 struct kvm_kernel_irqfd *irqfd =
10645 container_of(cons, struct kvm_kernel_irqfd, consumer);
10647 WARN_ON(irqfd->producer != prod);
10648 irqfd->producer = NULL;
10651 * When producer of consumer is unregistered, we change back to
10652 * remapped mode, so we can re-use the current implementation
10653 * when the irq is masked/disabled or the consumer side (KVM
10654 * int this case doesn't want to receive the interrupts.
10656 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10658 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10659 " fails: %d\n", irqfd->consumer.token, ret);
10662 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10663 uint32_t guest_irq, bool set)
10665 return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
10668 bool kvm_vector_hashing_enabled(void)
10670 return vector_hashing;
10673 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10675 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10677 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10679 u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu)
10681 uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD;
10683 /* The STIBP bit doesn't fault even if it's not advertised */
10684 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
10685 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
10686 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10687 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
10688 !boot_cpu_has(X86_FEATURE_AMD_IBRS))
10689 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10691 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) &&
10692 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
10693 bits &= ~SPEC_CTRL_SSBD;
10694 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
10695 !boot_cpu_has(X86_FEATURE_AMD_SSBD))
10696 bits &= ~SPEC_CTRL_SSBD;
10700 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits);
10702 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10703 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10704 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10705 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10706 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10707 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10708 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10709 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10710 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10711 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10712 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10713 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10714 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10715 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10716 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10717 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10718 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10719 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10720 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10721 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10722 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
10723 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);