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 struct kmem_cache *x86_fpu_cache;
243 EXPORT_SYMBOL_GPL(x86_fpu_cache);
245 static struct kmem_cache *x86_emulator_cache;
247 static struct kmem_cache *kvm_alloc_emulator_cache(void)
249 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
250 unsigned int size = sizeof(struct x86_emulate_ctxt);
252 return kmem_cache_create_usercopy("x86_emulator", size,
253 __alignof__(struct x86_emulate_ctxt),
254 SLAB_ACCOUNT, useroffset,
255 size - useroffset, NULL);
258 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
260 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
263 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
264 vcpu->arch.apf.gfns[i] = ~0;
267 static void kvm_on_user_return(struct user_return_notifier *urn)
270 struct kvm_shared_msrs *locals
271 = container_of(urn, struct kvm_shared_msrs, urn);
272 struct kvm_shared_msr_values *values;
276 * Disabling irqs at this point since the following code could be
277 * interrupted and executed through kvm_arch_hardware_disable()
279 local_irq_save(flags);
280 if (locals->registered) {
281 locals->registered = false;
282 user_return_notifier_unregister(urn);
284 local_irq_restore(flags);
285 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
286 values = &locals->values[slot];
287 if (values->host != values->curr) {
288 wrmsrl(shared_msrs_global.msrs[slot], values->host);
289 values->curr = values->host;
294 void kvm_define_shared_msr(unsigned slot, u32 msr)
296 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
297 shared_msrs_global.msrs[slot] = msr;
298 if (slot >= shared_msrs_global.nr)
299 shared_msrs_global.nr = slot + 1;
301 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
303 static void kvm_shared_msr_cpu_online(void)
305 unsigned int cpu = smp_processor_id();
306 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
310 for (i = 0; i < shared_msrs_global.nr; ++i) {
311 rdmsrl_safe(shared_msrs_global.msrs[i], &value);
312 smsr->values[i].host = value;
313 smsr->values[i].curr = value;
317 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
319 unsigned int cpu = smp_processor_id();
320 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
323 value = (value & mask) | (smsr->values[slot].host & ~mask);
324 if (value == smsr->values[slot].curr)
326 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
330 smsr->values[slot].curr = value;
331 if (!smsr->registered) {
332 smsr->urn.on_user_return = kvm_on_user_return;
333 user_return_notifier_register(&smsr->urn);
334 smsr->registered = true;
338 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
340 static void drop_user_return_notifiers(void)
342 unsigned int cpu = smp_processor_id();
343 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
345 if (smsr->registered)
346 kvm_on_user_return(&smsr->urn);
349 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
351 return vcpu->arch.apic_base;
353 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
355 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
357 return kvm_apic_mode(kvm_get_apic_base(vcpu));
359 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
361 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
363 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
364 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
365 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
366 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
368 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
370 if (!msr_info->host_initiated) {
371 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
373 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
377 kvm_lapic_set_base(vcpu, msr_info->data);
378 kvm_recalculate_apic_map(vcpu->kvm);
381 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
383 asmlinkage __visible void kvm_spurious_fault(void)
385 /* Fault while not rebooting. We want the trace. */
386 BUG_ON(!kvm_rebooting);
388 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
390 #define EXCPT_BENIGN 0
391 #define EXCPT_CONTRIBUTORY 1
394 static int exception_class(int vector)
404 return EXCPT_CONTRIBUTORY;
411 #define EXCPT_FAULT 0
413 #define EXCPT_ABORT 2
414 #define EXCPT_INTERRUPT 3
416 static int exception_type(int vector)
420 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
421 return EXCPT_INTERRUPT;
425 /* #DB is trap, as instruction watchpoints are handled elsewhere */
426 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
429 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
432 /* Reserved exceptions will result in fault */
436 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
438 unsigned nr = vcpu->arch.exception.nr;
439 bool has_payload = vcpu->arch.exception.has_payload;
440 unsigned long payload = vcpu->arch.exception.payload;
448 * "Certain debug exceptions may clear bit 0-3. The
449 * remaining contents of the DR6 register are never
450 * cleared by the processor".
452 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
454 * DR6.RTM is set by all #DB exceptions that don't clear it.
456 vcpu->arch.dr6 |= DR6_RTM;
457 vcpu->arch.dr6 |= payload;
459 * Bit 16 should be set in the payload whenever the #DB
460 * exception should clear DR6.RTM. This makes the payload
461 * compatible with the pending debug exceptions under VMX.
462 * Though not currently documented in the SDM, this also
463 * makes the payload compatible with the exit qualification
464 * for #DB exceptions under VMX.
466 vcpu->arch.dr6 ^= payload & DR6_RTM;
469 * The #DB payload is defined as compatible with the 'pending
470 * debug exceptions' field under VMX, not DR6. While bit 12 is
471 * defined in the 'pending debug exceptions' field (enabled
472 * breakpoint), it is reserved and must be zero in DR6.
474 vcpu->arch.dr6 &= ~BIT(12);
477 vcpu->arch.cr2 = payload;
481 vcpu->arch.exception.has_payload = false;
482 vcpu->arch.exception.payload = 0;
484 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
486 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
487 unsigned nr, bool has_error, u32 error_code,
488 bool has_payload, unsigned long payload, bool reinject)
493 kvm_make_request(KVM_REQ_EVENT, vcpu);
495 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
497 if (has_error && !is_protmode(vcpu))
501 * On vmentry, vcpu->arch.exception.pending is only
502 * true if an event injection was blocked by
503 * nested_run_pending. In that case, however,
504 * vcpu_enter_guest requests an immediate exit,
505 * and the guest shouldn't proceed far enough to
508 WARN_ON_ONCE(vcpu->arch.exception.pending);
509 vcpu->arch.exception.injected = true;
510 if (WARN_ON_ONCE(has_payload)) {
512 * A reinjected event has already
513 * delivered its payload.
519 vcpu->arch.exception.pending = true;
520 vcpu->arch.exception.injected = false;
522 vcpu->arch.exception.has_error_code = has_error;
523 vcpu->arch.exception.nr = nr;
524 vcpu->arch.exception.error_code = error_code;
525 vcpu->arch.exception.has_payload = has_payload;
526 vcpu->arch.exception.payload = payload;
527 if (!is_guest_mode(vcpu))
528 kvm_deliver_exception_payload(vcpu);
532 /* to check exception */
533 prev_nr = vcpu->arch.exception.nr;
534 if (prev_nr == DF_VECTOR) {
535 /* triple fault -> shutdown */
536 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
539 class1 = exception_class(prev_nr);
540 class2 = exception_class(nr);
541 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
542 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
544 * Generate double fault per SDM Table 5-5. Set
545 * exception.pending = true so that the double fault
546 * can trigger a nested vmexit.
548 vcpu->arch.exception.pending = true;
549 vcpu->arch.exception.injected = false;
550 vcpu->arch.exception.has_error_code = true;
551 vcpu->arch.exception.nr = DF_VECTOR;
552 vcpu->arch.exception.error_code = 0;
553 vcpu->arch.exception.has_payload = false;
554 vcpu->arch.exception.payload = 0;
556 /* replace previous exception with a new one in a hope
557 that instruction re-execution will regenerate lost
562 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
564 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
566 EXPORT_SYMBOL_GPL(kvm_queue_exception);
568 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
570 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
572 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
574 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
575 unsigned long payload)
577 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
579 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
581 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
582 u32 error_code, unsigned long payload)
584 kvm_multiple_exception(vcpu, nr, true, error_code,
585 true, payload, false);
588 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
591 kvm_inject_gp(vcpu, 0);
593 return kvm_skip_emulated_instruction(vcpu);
597 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
599 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
601 ++vcpu->stat.pf_guest;
602 vcpu->arch.exception.nested_apf =
603 is_guest_mode(vcpu) && fault->async_page_fault;
604 if (vcpu->arch.exception.nested_apf) {
605 vcpu->arch.apf.nested_apf_token = fault->address;
606 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
608 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
612 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
614 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
615 struct x86_exception *fault)
617 struct kvm_mmu *fault_mmu;
618 WARN_ON_ONCE(fault->vector != PF_VECTOR);
620 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
624 * Invalidate the TLB entry for the faulting address, if it exists,
625 * else the access will fault indefinitely (and to emulate hardware).
627 if ((fault->error_code & PFERR_PRESENT_MASK) &&
628 !(fault->error_code & PFERR_RSVD_MASK))
629 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
630 fault_mmu->root_hpa);
632 fault_mmu->inject_page_fault(vcpu, fault);
633 return fault->nested_page_fault;
635 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
637 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
639 atomic_inc(&vcpu->arch.nmi_queued);
640 kvm_make_request(KVM_REQ_NMI, vcpu);
642 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
644 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
646 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
648 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
650 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
652 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
654 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
657 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
658 * a #GP and return false.
660 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
662 if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
664 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
667 EXPORT_SYMBOL_GPL(kvm_require_cpl);
669 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
671 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
674 kvm_queue_exception(vcpu, UD_VECTOR);
677 EXPORT_SYMBOL_GPL(kvm_require_dr);
680 * This function will be used to read from the physical memory of the currently
681 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
682 * can read from guest physical or from the guest's guest physical memory.
684 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
685 gfn_t ngfn, void *data, int offset, int len,
688 struct x86_exception exception;
692 ngpa = gfn_to_gpa(ngfn);
693 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
694 if (real_gfn == UNMAPPED_GVA)
697 real_gfn = gpa_to_gfn(real_gfn);
699 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
701 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
703 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
704 void *data, int offset, int len, u32 access)
706 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
707 data, offset, len, access);
710 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
712 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
717 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
719 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
721 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
722 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
725 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
727 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
728 offset * sizeof(u64), sizeof(pdpte),
729 PFERR_USER_MASK|PFERR_WRITE_MASK);
734 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
735 if ((pdpte[i] & PT_PRESENT_MASK) &&
736 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
743 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
744 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
750 EXPORT_SYMBOL_GPL(load_pdptrs);
752 bool pdptrs_changed(struct kvm_vcpu *vcpu)
754 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
759 if (!is_pae_paging(vcpu))
762 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
765 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
766 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
767 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
768 PFERR_USER_MASK | PFERR_WRITE_MASK);
772 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
774 EXPORT_SYMBOL_GPL(pdptrs_changed);
776 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
778 unsigned long old_cr0 = kvm_read_cr0(vcpu);
779 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
784 if (cr0 & 0xffffffff00000000UL)
788 cr0 &= ~CR0_RESERVED_BITS;
790 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
793 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
796 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
798 if ((vcpu->arch.efer & EFER_LME)) {
803 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
808 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
813 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
816 kvm_x86_ops.set_cr0(vcpu, cr0);
818 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
819 kvm_clear_async_pf_completion_queue(vcpu);
820 kvm_async_pf_hash_reset(vcpu);
823 if ((cr0 ^ old_cr0) & update_bits)
824 kvm_mmu_reset_context(vcpu);
826 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
827 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
828 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
829 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
833 EXPORT_SYMBOL_GPL(kvm_set_cr0);
835 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
837 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
839 EXPORT_SYMBOL_GPL(kvm_lmsw);
841 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
843 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
845 if (vcpu->arch.xcr0 != host_xcr0)
846 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
848 if (vcpu->arch.xsaves_enabled &&
849 vcpu->arch.ia32_xss != host_xss)
850 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
853 if (static_cpu_has(X86_FEATURE_PKU) &&
854 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
855 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
856 vcpu->arch.pkru != vcpu->arch.host_pkru)
857 __write_pkru(vcpu->arch.pkru);
859 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
861 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
863 if (static_cpu_has(X86_FEATURE_PKU) &&
864 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
865 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
866 vcpu->arch.pkru = rdpkru();
867 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
868 __write_pkru(vcpu->arch.host_pkru);
871 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
873 if (vcpu->arch.xcr0 != host_xcr0)
874 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
876 if (vcpu->arch.xsaves_enabled &&
877 vcpu->arch.ia32_xss != host_xss)
878 wrmsrl(MSR_IA32_XSS, host_xss);
882 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
884 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
887 u64 old_xcr0 = vcpu->arch.xcr0;
890 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
891 if (index != XCR_XFEATURE_ENABLED_MASK)
893 if (!(xcr0 & XFEATURE_MASK_FP))
895 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
899 * Do not allow the guest to set bits that we do not support
900 * saving. However, xcr0 bit 0 is always set, even if the
901 * emulated CPU does not support XSAVE (see fx_init).
903 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
904 if (xcr0 & ~valid_bits)
907 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
908 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
911 if (xcr0 & XFEATURE_MASK_AVX512) {
912 if (!(xcr0 & XFEATURE_MASK_YMM))
914 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
917 vcpu->arch.xcr0 = xcr0;
919 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
920 kvm_update_cpuid(vcpu);
924 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
926 if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
927 __kvm_set_xcr(vcpu, index, xcr)) {
928 kvm_inject_gp(vcpu, 0);
933 EXPORT_SYMBOL_GPL(kvm_set_xcr);
935 #define __cr4_reserved_bits(__cpu_has, __c) \
937 u64 __reserved_bits = CR4_RESERVED_BITS; \
939 if (!__cpu_has(__c, X86_FEATURE_XSAVE)) \
940 __reserved_bits |= X86_CR4_OSXSAVE; \
941 if (!__cpu_has(__c, X86_FEATURE_SMEP)) \
942 __reserved_bits |= X86_CR4_SMEP; \
943 if (!__cpu_has(__c, X86_FEATURE_SMAP)) \
944 __reserved_bits |= X86_CR4_SMAP; \
945 if (!__cpu_has(__c, X86_FEATURE_FSGSBASE)) \
946 __reserved_bits |= X86_CR4_FSGSBASE; \
947 if (!__cpu_has(__c, X86_FEATURE_PKU)) \
948 __reserved_bits |= X86_CR4_PKE; \
949 if (!__cpu_has(__c, X86_FEATURE_LA57)) \
950 __reserved_bits |= X86_CR4_LA57; \
951 if (!__cpu_has(__c, X86_FEATURE_UMIP)) \
952 __reserved_bits |= X86_CR4_UMIP; \
956 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
958 if (cr4 & cr4_reserved_bits)
961 if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu))
967 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
969 unsigned long old_cr4 = kvm_read_cr4(vcpu);
970 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
971 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
973 if (kvm_valid_cr4(vcpu, cr4))
976 if (is_long_mode(vcpu)) {
977 if (!(cr4 & X86_CR4_PAE))
979 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
980 && ((cr4 ^ old_cr4) & pdptr_bits)
981 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
985 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
986 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
989 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
990 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
994 if (kvm_x86_ops.set_cr4(vcpu, cr4))
997 if (((cr4 ^ old_cr4) & pdptr_bits) ||
998 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
999 kvm_mmu_reset_context(vcpu);
1001 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1002 kvm_update_cpuid(vcpu);
1006 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1008 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1010 bool skip_tlb_flush = false;
1011 #ifdef CONFIG_X86_64
1012 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1015 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1016 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1020 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1021 if (!skip_tlb_flush) {
1022 kvm_mmu_sync_roots(vcpu);
1023 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1028 if (is_long_mode(vcpu) &&
1029 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
1031 else if (is_pae_paging(vcpu) &&
1032 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1035 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1036 vcpu->arch.cr3 = cr3;
1037 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1041 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1043 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1045 if (cr8 & CR8_RESERVED_BITS)
1047 if (lapic_in_kernel(vcpu))
1048 kvm_lapic_set_tpr(vcpu, cr8);
1050 vcpu->arch.cr8 = cr8;
1053 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1055 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1057 if (lapic_in_kernel(vcpu))
1058 return kvm_lapic_get_cr8(vcpu);
1060 return vcpu->arch.cr8;
1062 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1064 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1068 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1069 for (i = 0; i < KVM_NR_DB_REGS; i++)
1070 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1071 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1075 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1079 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1080 dr7 = vcpu->arch.guest_debug_dr7;
1082 dr7 = vcpu->arch.dr7;
1083 kvm_x86_ops.set_dr7(vcpu, dr7);
1084 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1085 if (dr7 & DR7_BP_EN_MASK)
1086 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1088 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1090 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1092 u64 fixed = DR6_FIXED_1;
1094 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1099 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1101 size_t size = ARRAY_SIZE(vcpu->arch.db);
1105 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1106 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1107 vcpu->arch.eff_db[dr] = val;
1112 if (val & 0xffffffff00000000ULL)
1113 return -1; /* #GP */
1114 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1119 if (!kvm_dr7_valid(val))
1120 return -1; /* #GP */
1121 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1122 kvm_update_dr7(vcpu);
1129 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1131 if (__kvm_set_dr(vcpu, dr, val)) {
1132 kvm_inject_gp(vcpu, 0);
1137 EXPORT_SYMBOL_GPL(kvm_set_dr);
1139 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1141 size_t size = ARRAY_SIZE(vcpu->arch.db);
1145 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1150 *val = vcpu->arch.dr6;
1155 *val = vcpu->arch.dr7;
1160 EXPORT_SYMBOL_GPL(kvm_get_dr);
1162 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1164 u32 ecx = kvm_rcx_read(vcpu);
1168 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1171 kvm_rax_write(vcpu, (u32)data);
1172 kvm_rdx_write(vcpu, data >> 32);
1175 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1178 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1179 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1181 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1182 * extract the supported MSRs from the related const lists.
1183 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1184 * capabilities of the host cpu. This capabilities test skips MSRs that are
1185 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1186 * may depend on host virtualization features rather than host cpu features.
1189 static const u32 msrs_to_save_all[] = {
1190 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1192 #ifdef CONFIG_X86_64
1193 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1195 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1196 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1198 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1199 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1200 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1201 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1202 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1203 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1204 MSR_IA32_UMWAIT_CONTROL,
1206 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1207 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1208 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1209 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1210 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1211 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1212 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1213 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1214 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1215 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1216 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1217 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1218 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1219 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1220 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1221 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1222 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1223 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1224 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1225 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1226 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1227 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1230 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1231 static unsigned num_msrs_to_save;
1233 static const u32 emulated_msrs_all[] = {
1234 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1235 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1236 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1237 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1238 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1239 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1240 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1242 HV_X64_MSR_VP_INDEX,
1243 HV_X64_MSR_VP_RUNTIME,
1244 HV_X64_MSR_SCONTROL,
1245 HV_X64_MSR_STIMER0_CONFIG,
1246 HV_X64_MSR_VP_ASSIST_PAGE,
1247 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1248 HV_X64_MSR_TSC_EMULATION_STATUS,
1250 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1253 MSR_IA32_TSC_ADJUST,
1254 MSR_IA32_TSCDEADLINE,
1255 MSR_IA32_ARCH_CAPABILITIES,
1256 MSR_IA32_MISC_ENABLE,
1257 MSR_IA32_MCG_STATUS,
1259 MSR_IA32_MCG_EXT_CTL,
1263 MSR_MISC_FEATURES_ENABLES,
1264 MSR_AMD64_VIRT_SPEC_CTRL,
1269 * The following list leaves out MSRs whose values are determined
1270 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1271 * We always support the "true" VMX control MSRs, even if the host
1272 * processor does not, so I am putting these registers here rather
1273 * than in msrs_to_save_all.
1276 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1277 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1278 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1279 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1281 MSR_IA32_VMX_CR0_FIXED0,
1282 MSR_IA32_VMX_CR4_FIXED0,
1283 MSR_IA32_VMX_VMCS_ENUM,
1284 MSR_IA32_VMX_PROCBASED_CTLS2,
1285 MSR_IA32_VMX_EPT_VPID_CAP,
1286 MSR_IA32_VMX_VMFUNC,
1289 MSR_KVM_POLL_CONTROL,
1292 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1293 static unsigned num_emulated_msrs;
1296 * List of msr numbers which are used to expose MSR-based features that
1297 * can be used by a hypervisor to validate requested CPU features.
1299 static const u32 msr_based_features_all[] = {
1301 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1302 MSR_IA32_VMX_PINBASED_CTLS,
1303 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1304 MSR_IA32_VMX_PROCBASED_CTLS,
1305 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1306 MSR_IA32_VMX_EXIT_CTLS,
1307 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1308 MSR_IA32_VMX_ENTRY_CTLS,
1310 MSR_IA32_VMX_CR0_FIXED0,
1311 MSR_IA32_VMX_CR0_FIXED1,
1312 MSR_IA32_VMX_CR4_FIXED0,
1313 MSR_IA32_VMX_CR4_FIXED1,
1314 MSR_IA32_VMX_VMCS_ENUM,
1315 MSR_IA32_VMX_PROCBASED_CTLS2,
1316 MSR_IA32_VMX_EPT_VPID_CAP,
1317 MSR_IA32_VMX_VMFUNC,
1321 MSR_IA32_ARCH_CAPABILITIES,
1324 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1325 static unsigned int num_msr_based_features;
1327 static u64 kvm_get_arch_capabilities(void)
1331 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1332 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1335 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1336 * the nested hypervisor runs with NX huge pages. If it is not,
1337 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1338 * L1 guests, so it need not worry about its own (L2) guests.
1340 data |= ARCH_CAP_PSCHANGE_MC_NO;
1343 * If we're doing cache flushes (either "always" or "cond")
1344 * we will do one whenever the guest does a vmlaunch/vmresume.
1345 * If an outer hypervisor is doing the cache flush for us
1346 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1347 * capability to the guest too, and if EPT is disabled we're not
1348 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1349 * require a nested hypervisor to do a flush of its own.
1351 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1352 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1354 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1355 data |= ARCH_CAP_RDCL_NO;
1356 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1357 data |= ARCH_CAP_SSB_NO;
1358 if (!boot_cpu_has_bug(X86_BUG_MDS))
1359 data |= ARCH_CAP_MDS_NO;
1362 * On TAA affected systems:
1363 * - nothing to do if TSX is disabled on the host.
1364 * - we emulate TSX_CTRL if present on the host.
1365 * This lets the guest use VERW to clear CPU buffers.
1367 if (!boot_cpu_has(X86_FEATURE_RTM))
1368 data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1369 else if (!boot_cpu_has_bug(X86_BUG_TAA))
1370 data |= ARCH_CAP_TAA_NO;
1375 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1377 switch (msr->index) {
1378 case MSR_IA32_ARCH_CAPABILITIES:
1379 msr->data = kvm_get_arch_capabilities();
1381 case MSR_IA32_UCODE_REV:
1382 rdmsrl_safe(msr->index, &msr->data);
1385 if (kvm_x86_ops.get_msr_feature(msr))
1391 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1393 struct kvm_msr_entry msr;
1397 r = kvm_get_msr_feature(&msr);
1406 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1408 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1411 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1414 if (efer & (EFER_LME | EFER_LMA) &&
1415 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1418 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1424 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1426 if (efer & efer_reserved_bits)
1429 return __kvm_valid_efer(vcpu, efer);
1431 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1433 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1435 u64 old_efer = vcpu->arch.efer;
1436 u64 efer = msr_info->data;
1438 if (efer & efer_reserved_bits)
1441 if (!msr_info->host_initiated) {
1442 if (!__kvm_valid_efer(vcpu, efer))
1445 if (is_paging(vcpu) &&
1446 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1451 efer |= vcpu->arch.efer & EFER_LMA;
1453 kvm_x86_ops.set_efer(vcpu, efer);
1455 /* Update reserved bits */
1456 if ((efer ^ old_efer) & EFER_NX)
1457 kvm_mmu_reset_context(vcpu);
1462 void kvm_enable_efer_bits(u64 mask)
1464 efer_reserved_bits &= ~mask;
1466 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1469 * Write @data into the MSR specified by @index. Select MSR specific fault
1470 * checks are bypassed if @host_initiated is %true.
1471 * Returns 0 on success, non-0 otherwise.
1472 * Assumes vcpu_load() was already called.
1474 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1475 bool host_initiated)
1477 struct msr_data msr;
1482 case MSR_KERNEL_GS_BASE:
1485 if (is_noncanonical_address(data, vcpu))
1488 case MSR_IA32_SYSENTER_EIP:
1489 case MSR_IA32_SYSENTER_ESP:
1491 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1492 * non-canonical address is written on Intel but not on
1493 * AMD (which ignores the top 32-bits, because it does
1494 * not implement 64-bit SYSENTER).
1496 * 64-bit code should hence be able to write a non-canonical
1497 * value on AMD. Making the address canonical ensures that
1498 * vmentry does not fail on Intel after writing a non-canonical
1499 * value, and that something deterministic happens if the guest
1500 * invokes 64-bit SYSENTER.
1502 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1507 msr.host_initiated = host_initiated;
1509 return kvm_x86_ops.set_msr(vcpu, &msr);
1513 * Read the MSR specified by @index into @data. Select MSR specific fault
1514 * checks are bypassed if @host_initiated is %true.
1515 * Returns 0 on success, non-0 otherwise.
1516 * Assumes vcpu_load() was already called.
1518 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1519 bool host_initiated)
1521 struct msr_data msr;
1525 msr.host_initiated = host_initiated;
1527 ret = kvm_x86_ops.get_msr(vcpu, &msr);
1533 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1535 return __kvm_get_msr(vcpu, index, data, false);
1537 EXPORT_SYMBOL_GPL(kvm_get_msr);
1539 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1541 return __kvm_set_msr(vcpu, index, data, false);
1543 EXPORT_SYMBOL_GPL(kvm_set_msr);
1545 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1547 u32 ecx = kvm_rcx_read(vcpu);
1550 if (kvm_get_msr(vcpu, ecx, &data)) {
1551 trace_kvm_msr_read_ex(ecx);
1552 kvm_inject_gp(vcpu, 0);
1556 trace_kvm_msr_read(ecx, data);
1558 kvm_rax_write(vcpu, data & -1u);
1559 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1560 return kvm_skip_emulated_instruction(vcpu);
1562 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1564 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1566 u32 ecx = kvm_rcx_read(vcpu);
1567 u64 data = kvm_read_edx_eax(vcpu);
1569 if (kvm_set_msr(vcpu, ecx, data)) {
1570 trace_kvm_msr_write_ex(ecx, data);
1571 kvm_inject_gp(vcpu, 0);
1575 trace_kvm_msr_write(ecx, data);
1576 return kvm_skip_emulated_instruction(vcpu);
1578 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1580 bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1582 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1583 need_resched() || signal_pending(current);
1585 EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request);
1588 * The fast path for frequent and performance sensitive wrmsr emulation,
1589 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1590 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1591 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1592 * other cases which must be called after interrupts are enabled on the host.
1594 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1596 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1599 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1600 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1601 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1602 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1605 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1606 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1607 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1608 trace_kvm_apic_write(APIC_ICR, (u32)data);
1615 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1617 if (!kvm_can_use_hv_timer(vcpu))
1620 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1624 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1626 u32 msr = kvm_rcx_read(vcpu);
1628 fastpath_t ret = EXIT_FASTPATH_NONE;
1631 case APIC_BASE_MSR + (APIC_ICR >> 4):
1632 data = kvm_read_edx_eax(vcpu);
1633 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1634 kvm_skip_emulated_instruction(vcpu);
1635 ret = EXIT_FASTPATH_EXIT_HANDLED;
1638 case MSR_IA32_TSCDEADLINE:
1639 data = kvm_read_edx_eax(vcpu);
1640 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1641 kvm_skip_emulated_instruction(vcpu);
1642 ret = EXIT_FASTPATH_REENTER_GUEST;
1649 if (ret != EXIT_FASTPATH_NONE)
1650 trace_kvm_msr_write(msr, data);
1654 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1657 * Adapt set_msr() to msr_io()'s calling convention
1659 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1661 return __kvm_get_msr(vcpu, index, data, true);
1664 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1666 return __kvm_set_msr(vcpu, index, *data, true);
1669 #ifdef CONFIG_X86_64
1670 struct pvclock_clock {
1680 struct pvclock_gtod_data {
1683 struct pvclock_clock clock; /* extract of a clocksource struct */
1684 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1690 static struct pvclock_gtod_data pvclock_gtod_data;
1692 static void update_pvclock_gtod(struct timekeeper *tk)
1694 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1696 write_seqcount_begin(&vdata->seq);
1698 /* copy pvclock gtod data */
1699 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1700 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1701 vdata->clock.mask = tk->tkr_mono.mask;
1702 vdata->clock.mult = tk->tkr_mono.mult;
1703 vdata->clock.shift = tk->tkr_mono.shift;
1704 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1705 vdata->clock.offset = tk->tkr_mono.base;
1707 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1708 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1709 vdata->raw_clock.mask = tk->tkr_raw.mask;
1710 vdata->raw_clock.mult = tk->tkr_raw.mult;
1711 vdata->raw_clock.shift = tk->tkr_raw.shift;
1712 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1713 vdata->raw_clock.offset = tk->tkr_raw.base;
1715 vdata->wall_time_sec = tk->xtime_sec;
1717 vdata->offs_boot = tk->offs_boot;
1719 write_seqcount_end(&vdata->seq);
1722 static s64 get_kvmclock_base_ns(void)
1724 /* Count up from boot time, but with the frequency of the raw clock. */
1725 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1728 static s64 get_kvmclock_base_ns(void)
1730 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1731 return ktime_get_boottime_ns();
1735 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1737 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1738 kvm_vcpu_kick(vcpu);
1741 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1745 struct pvclock_wall_clock wc;
1751 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1756 ++version; /* first time write, random junk */
1760 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1764 * The guest calculates current wall clock time by adding
1765 * system time (updated by kvm_guest_time_update below) to the
1766 * wall clock specified here. We do the reverse here.
1768 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1770 wc.nsec = do_div(wall_nsec, 1000000000);
1771 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1772 wc.version = version;
1774 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1777 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1780 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1782 do_shl32_div32(dividend, divisor);
1786 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1787 s8 *pshift, u32 *pmultiplier)
1795 scaled64 = scaled_hz;
1796 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1801 tps32 = (uint32_t)tps64;
1802 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1803 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1811 *pmultiplier = div_frac(scaled64, tps32);
1814 #ifdef CONFIG_X86_64
1815 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1818 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1819 static unsigned long max_tsc_khz;
1821 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1823 u64 v = (u64)khz * (1000000 + ppm);
1828 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1832 /* Guest TSC same frequency as host TSC? */
1834 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1838 /* TSC scaling supported? */
1839 if (!kvm_has_tsc_control) {
1840 if (user_tsc_khz > tsc_khz) {
1841 vcpu->arch.tsc_catchup = 1;
1842 vcpu->arch.tsc_always_catchup = 1;
1845 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1850 /* TSC scaling required - calculate ratio */
1851 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1852 user_tsc_khz, tsc_khz);
1854 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1855 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1860 vcpu->arch.tsc_scaling_ratio = ratio;
1864 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1866 u32 thresh_lo, thresh_hi;
1867 int use_scaling = 0;
1869 /* tsc_khz can be zero if TSC calibration fails */
1870 if (user_tsc_khz == 0) {
1871 /* set tsc_scaling_ratio to a safe value */
1872 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1876 /* Compute a scale to convert nanoseconds in TSC cycles */
1877 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1878 &vcpu->arch.virtual_tsc_shift,
1879 &vcpu->arch.virtual_tsc_mult);
1880 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1883 * Compute the variation in TSC rate which is acceptable
1884 * within the range of tolerance and decide if the
1885 * rate being applied is within that bounds of the hardware
1886 * rate. If so, no scaling or compensation need be done.
1888 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1889 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1890 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1891 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1894 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1897 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1899 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1900 vcpu->arch.virtual_tsc_mult,
1901 vcpu->arch.virtual_tsc_shift);
1902 tsc += vcpu->arch.this_tsc_write;
1906 static inline int gtod_is_based_on_tsc(int mode)
1908 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
1911 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1913 #ifdef CONFIG_X86_64
1915 struct kvm_arch *ka = &vcpu->kvm->arch;
1916 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1918 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1919 atomic_read(&vcpu->kvm->online_vcpus));
1922 * Once the masterclock is enabled, always perform request in
1923 * order to update it.
1925 * In order to enable masterclock, the host clocksource must be TSC
1926 * and the vcpus need to have matched TSCs. When that happens,
1927 * perform request to enable masterclock.
1929 if (ka->use_master_clock ||
1930 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1931 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1933 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1934 atomic_read(&vcpu->kvm->online_vcpus),
1935 ka->use_master_clock, gtod->clock.vclock_mode);
1939 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1941 u64 curr_offset = vcpu->arch.l1_tsc_offset;
1942 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1946 * Multiply tsc by a fixed point number represented by ratio.
1948 * The most significant 64-N bits (mult) of ratio represent the
1949 * integral part of the fixed point number; the remaining N bits
1950 * (frac) represent the fractional part, ie. ratio represents a fixed
1951 * point number (mult + frac * 2^(-N)).
1953 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1955 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1957 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1960 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1963 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1965 if (ratio != kvm_default_tsc_scaling_ratio)
1966 _tsc = __scale_tsc(ratio, tsc);
1970 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1972 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1976 tsc = kvm_scale_tsc(vcpu, rdtsc());
1978 return target_tsc - tsc;
1981 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1983 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1985 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1987 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1989 vcpu->arch.l1_tsc_offset = offset;
1990 vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
1993 static inline bool kvm_check_tsc_unstable(void)
1995 #ifdef CONFIG_X86_64
1997 * TSC is marked unstable when we're running on Hyper-V,
1998 * 'TSC page' clocksource is good.
2000 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2003 return check_tsc_unstable();
2006 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
2008 struct kvm *kvm = vcpu->kvm;
2009 u64 offset, ns, elapsed;
2010 unsigned long flags;
2012 bool already_matched;
2013 u64 data = msr->data;
2014 bool synchronizing = false;
2016 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2017 offset = kvm_compute_tsc_offset(vcpu, data);
2018 ns = get_kvmclock_base_ns();
2019 elapsed = ns - kvm->arch.last_tsc_nsec;
2021 if (vcpu->arch.virtual_tsc_khz) {
2022 if (data == 0 && msr->host_initiated) {
2024 * detection of vcpu initialization -- need to sync
2025 * with other vCPUs. This particularly helps to keep
2026 * kvm_clock stable after CPU hotplug
2028 synchronizing = true;
2030 u64 tsc_exp = kvm->arch.last_tsc_write +
2031 nsec_to_cycles(vcpu, elapsed);
2032 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2034 * Special case: TSC write with a small delta (1 second)
2035 * of virtual cycle time against real time is
2036 * interpreted as an attempt to synchronize the CPU.
2038 synchronizing = data < tsc_exp + tsc_hz &&
2039 data + tsc_hz > tsc_exp;
2044 * For a reliable TSC, we can match TSC offsets, and for an unstable
2045 * TSC, we add elapsed time in this computation. We could let the
2046 * compensation code attempt to catch up if we fall behind, but
2047 * it's better to try to match offsets from the beginning.
2049 if (synchronizing &&
2050 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2051 if (!kvm_check_tsc_unstable()) {
2052 offset = kvm->arch.cur_tsc_offset;
2054 u64 delta = nsec_to_cycles(vcpu, elapsed);
2056 offset = kvm_compute_tsc_offset(vcpu, data);
2059 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2062 * We split periods of matched TSC writes into generations.
2063 * For each generation, we track the original measured
2064 * nanosecond time, offset, and write, so if TSCs are in
2065 * sync, we can match exact offset, and if not, we can match
2066 * exact software computation in compute_guest_tsc()
2068 * These values are tracked in kvm->arch.cur_xxx variables.
2070 kvm->arch.cur_tsc_generation++;
2071 kvm->arch.cur_tsc_nsec = ns;
2072 kvm->arch.cur_tsc_write = data;
2073 kvm->arch.cur_tsc_offset = offset;
2078 * We also track th most recent recorded KHZ, write and time to
2079 * allow the matching interval to be extended at each write.
2081 kvm->arch.last_tsc_nsec = ns;
2082 kvm->arch.last_tsc_write = data;
2083 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2085 vcpu->arch.last_guest_tsc = data;
2087 /* Keep track of which generation this VCPU has synchronized to */
2088 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2089 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2090 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2092 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
2093 update_ia32_tsc_adjust_msr(vcpu, offset);
2095 kvm_vcpu_write_tsc_offset(vcpu, offset);
2096 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2098 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2100 kvm->arch.nr_vcpus_matched_tsc = 0;
2101 } else if (!already_matched) {
2102 kvm->arch.nr_vcpus_matched_tsc++;
2105 kvm_track_tsc_matching(vcpu);
2106 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2109 EXPORT_SYMBOL_GPL(kvm_write_tsc);
2111 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2114 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2115 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2118 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2120 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2121 WARN_ON(adjustment < 0);
2122 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2123 adjust_tsc_offset_guest(vcpu, adjustment);
2126 #ifdef CONFIG_X86_64
2128 static u64 read_tsc(void)
2130 u64 ret = (u64)rdtsc_ordered();
2131 u64 last = pvclock_gtod_data.clock.cycle_last;
2133 if (likely(ret >= last))
2137 * GCC likes to generate cmov here, but this branch is extremely
2138 * predictable (it's just a function of time and the likely is
2139 * very likely) and there's a data dependence, so force GCC
2140 * to generate a branch instead. I don't barrier() because
2141 * we don't actually need a barrier, and if this function
2142 * ever gets inlined it will generate worse code.
2148 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2154 switch (clock->vclock_mode) {
2155 case VDSO_CLOCKMODE_HVCLOCK:
2156 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2158 if (tsc_pg_val != U64_MAX) {
2159 /* TSC page valid */
2160 *mode = VDSO_CLOCKMODE_HVCLOCK;
2161 v = (tsc_pg_val - clock->cycle_last) &
2164 /* TSC page invalid */
2165 *mode = VDSO_CLOCKMODE_NONE;
2168 case VDSO_CLOCKMODE_TSC:
2169 *mode = VDSO_CLOCKMODE_TSC;
2170 *tsc_timestamp = read_tsc();
2171 v = (*tsc_timestamp - clock->cycle_last) &
2175 *mode = VDSO_CLOCKMODE_NONE;
2178 if (*mode == VDSO_CLOCKMODE_NONE)
2179 *tsc_timestamp = v = 0;
2181 return v * clock->mult;
2184 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2186 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2192 seq = read_seqcount_begin(>od->seq);
2193 ns = gtod->raw_clock.base_cycles;
2194 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2195 ns >>= gtod->raw_clock.shift;
2196 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2197 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2203 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2205 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2211 seq = read_seqcount_begin(>od->seq);
2212 ts->tv_sec = gtod->wall_time_sec;
2213 ns = gtod->clock.base_cycles;
2214 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2215 ns >>= gtod->clock.shift;
2216 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2218 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2224 /* returns true if host is using TSC based clocksource */
2225 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2227 /* checked again under seqlock below */
2228 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2231 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2235 /* returns true if host is using TSC based clocksource */
2236 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2239 /* checked again under seqlock below */
2240 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2243 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2249 * Assuming a stable TSC across physical CPUS, and a stable TSC
2250 * across virtual CPUs, the following condition is possible.
2251 * Each numbered line represents an event visible to both
2252 * CPUs at the next numbered event.
2254 * "timespecX" represents host monotonic time. "tscX" represents
2257 * VCPU0 on CPU0 | VCPU1 on CPU1
2259 * 1. read timespec0,tsc0
2260 * 2. | timespec1 = timespec0 + N
2262 * 3. transition to guest | transition to guest
2263 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2264 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2265 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2267 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2270 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2272 * - 0 < N - M => M < N
2274 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2275 * always the case (the difference between two distinct xtime instances
2276 * might be smaller then the difference between corresponding TSC reads,
2277 * when updating guest vcpus pvclock areas).
2279 * To avoid that problem, do not allow visibility of distinct
2280 * system_timestamp/tsc_timestamp values simultaneously: use a master
2281 * copy of host monotonic time values. Update that master copy
2284 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2288 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2290 #ifdef CONFIG_X86_64
2291 struct kvm_arch *ka = &kvm->arch;
2293 bool host_tsc_clocksource, vcpus_matched;
2295 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2296 atomic_read(&kvm->online_vcpus));
2299 * If the host uses TSC clock, then passthrough TSC as stable
2302 host_tsc_clocksource = kvm_get_time_and_clockread(
2303 &ka->master_kernel_ns,
2304 &ka->master_cycle_now);
2306 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2307 && !ka->backwards_tsc_observed
2308 && !ka->boot_vcpu_runs_old_kvmclock;
2310 if (ka->use_master_clock)
2311 atomic_set(&kvm_guest_has_master_clock, 1);
2313 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2314 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2319 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2321 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2324 static void kvm_gen_update_masterclock(struct kvm *kvm)
2326 #ifdef CONFIG_X86_64
2328 struct kvm_vcpu *vcpu;
2329 struct kvm_arch *ka = &kvm->arch;
2331 spin_lock(&ka->pvclock_gtod_sync_lock);
2332 kvm_make_mclock_inprogress_request(kvm);
2333 /* no guest entries from this point */
2334 pvclock_update_vm_gtod_copy(kvm);
2336 kvm_for_each_vcpu(i, vcpu, kvm)
2337 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2339 /* guest entries allowed */
2340 kvm_for_each_vcpu(i, vcpu, kvm)
2341 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2343 spin_unlock(&ka->pvclock_gtod_sync_lock);
2347 u64 get_kvmclock_ns(struct kvm *kvm)
2349 struct kvm_arch *ka = &kvm->arch;
2350 struct pvclock_vcpu_time_info hv_clock;
2353 spin_lock(&ka->pvclock_gtod_sync_lock);
2354 if (!ka->use_master_clock) {
2355 spin_unlock(&ka->pvclock_gtod_sync_lock);
2356 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2359 hv_clock.tsc_timestamp = ka->master_cycle_now;
2360 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2361 spin_unlock(&ka->pvclock_gtod_sync_lock);
2363 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2366 if (__this_cpu_read(cpu_tsc_khz)) {
2367 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2368 &hv_clock.tsc_shift,
2369 &hv_clock.tsc_to_system_mul);
2370 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2372 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2379 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2381 struct kvm_vcpu_arch *vcpu = &v->arch;
2382 struct pvclock_vcpu_time_info guest_hv_clock;
2384 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2385 &guest_hv_clock, sizeof(guest_hv_clock))))
2388 /* This VCPU is paused, but it's legal for a guest to read another
2389 * VCPU's kvmclock, so we really have to follow the specification where
2390 * it says that version is odd if data is being modified, and even after
2393 * Version field updates must be kept separate. This is because
2394 * kvm_write_guest_cached might use a "rep movs" instruction, and
2395 * writes within a string instruction are weakly ordered. So there
2396 * are three writes overall.
2398 * As a small optimization, only write the version field in the first
2399 * and third write. The vcpu->pv_time cache is still valid, because the
2400 * version field is the first in the struct.
2402 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2404 if (guest_hv_clock.version & 1)
2405 ++guest_hv_clock.version; /* first time write, random junk */
2407 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2408 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2410 sizeof(vcpu->hv_clock.version));
2414 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2415 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2417 if (vcpu->pvclock_set_guest_stopped_request) {
2418 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2419 vcpu->pvclock_set_guest_stopped_request = false;
2422 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2424 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2426 sizeof(vcpu->hv_clock));
2430 vcpu->hv_clock.version++;
2431 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2433 sizeof(vcpu->hv_clock.version));
2436 static int kvm_guest_time_update(struct kvm_vcpu *v)
2438 unsigned long flags, tgt_tsc_khz;
2439 struct kvm_vcpu_arch *vcpu = &v->arch;
2440 struct kvm_arch *ka = &v->kvm->arch;
2442 u64 tsc_timestamp, host_tsc;
2444 bool use_master_clock;
2450 * If the host uses TSC clock, then passthrough TSC as stable
2453 spin_lock(&ka->pvclock_gtod_sync_lock);
2454 use_master_clock = ka->use_master_clock;
2455 if (use_master_clock) {
2456 host_tsc = ka->master_cycle_now;
2457 kernel_ns = ka->master_kernel_ns;
2459 spin_unlock(&ka->pvclock_gtod_sync_lock);
2461 /* Keep irq disabled to prevent changes to the clock */
2462 local_irq_save(flags);
2463 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2464 if (unlikely(tgt_tsc_khz == 0)) {
2465 local_irq_restore(flags);
2466 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2469 if (!use_master_clock) {
2471 kernel_ns = get_kvmclock_base_ns();
2474 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2477 * We may have to catch up the TSC to match elapsed wall clock
2478 * time for two reasons, even if kvmclock is used.
2479 * 1) CPU could have been running below the maximum TSC rate
2480 * 2) Broken TSC compensation resets the base at each VCPU
2481 * entry to avoid unknown leaps of TSC even when running
2482 * again on the same CPU. This may cause apparent elapsed
2483 * time to disappear, and the guest to stand still or run
2486 if (vcpu->tsc_catchup) {
2487 u64 tsc = compute_guest_tsc(v, kernel_ns);
2488 if (tsc > tsc_timestamp) {
2489 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2490 tsc_timestamp = tsc;
2494 local_irq_restore(flags);
2496 /* With all the info we got, fill in the values */
2498 if (kvm_has_tsc_control)
2499 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2501 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2502 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2503 &vcpu->hv_clock.tsc_shift,
2504 &vcpu->hv_clock.tsc_to_system_mul);
2505 vcpu->hw_tsc_khz = tgt_tsc_khz;
2508 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2509 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2510 vcpu->last_guest_tsc = tsc_timestamp;
2512 /* If the host uses TSC clocksource, then it is stable */
2514 if (use_master_clock)
2515 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2517 vcpu->hv_clock.flags = pvclock_flags;
2519 if (vcpu->pv_time_enabled)
2520 kvm_setup_pvclock_page(v);
2521 if (v == kvm_get_vcpu(v->kvm, 0))
2522 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2527 * kvmclock updates which are isolated to a given vcpu, such as
2528 * vcpu->cpu migration, should not allow system_timestamp from
2529 * the rest of the vcpus to remain static. Otherwise ntp frequency
2530 * correction applies to one vcpu's system_timestamp but not
2533 * So in those cases, request a kvmclock update for all vcpus.
2534 * We need to rate-limit these requests though, as they can
2535 * considerably slow guests that have a large number of vcpus.
2536 * The time for a remote vcpu to update its kvmclock is bound
2537 * by the delay we use to rate-limit the updates.
2540 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2542 static void kvmclock_update_fn(struct work_struct *work)
2545 struct delayed_work *dwork = to_delayed_work(work);
2546 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2547 kvmclock_update_work);
2548 struct kvm *kvm = container_of(ka, struct kvm, arch);
2549 struct kvm_vcpu *vcpu;
2551 kvm_for_each_vcpu(i, vcpu, kvm) {
2552 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2553 kvm_vcpu_kick(vcpu);
2557 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2559 struct kvm *kvm = v->kvm;
2561 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2562 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2563 KVMCLOCK_UPDATE_DELAY);
2566 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2568 static void kvmclock_sync_fn(struct work_struct *work)
2570 struct delayed_work *dwork = to_delayed_work(work);
2571 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2572 kvmclock_sync_work);
2573 struct kvm *kvm = container_of(ka, struct kvm, arch);
2575 if (!kvmclock_periodic_sync)
2578 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2579 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2580 KVMCLOCK_SYNC_PERIOD);
2584 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2586 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2588 /* McStatusWrEn enabled? */
2589 if (guest_cpuid_is_amd_or_hygon(vcpu))
2590 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2595 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2597 u64 mcg_cap = vcpu->arch.mcg_cap;
2598 unsigned bank_num = mcg_cap & 0xff;
2599 u32 msr = msr_info->index;
2600 u64 data = msr_info->data;
2603 case MSR_IA32_MCG_STATUS:
2604 vcpu->arch.mcg_status = data;
2606 case MSR_IA32_MCG_CTL:
2607 if (!(mcg_cap & MCG_CTL_P) &&
2608 (data || !msr_info->host_initiated))
2610 if (data != 0 && data != ~(u64)0)
2612 vcpu->arch.mcg_ctl = data;
2615 if (msr >= MSR_IA32_MC0_CTL &&
2616 msr < MSR_IA32_MCx_CTL(bank_num)) {
2617 u32 offset = array_index_nospec(
2618 msr - MSR_IA32_MC0_CTL,
2619 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2621 /* only 0 or all 1s can be written to IA32_MCi_CTL
2622 * some Linux kernels though clear bit 10 in bank 4 to
2623 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2624 * this to avoid an uncatched #GP in the guest
2626 if ((offset & 0x3) == 0 &&
2627 data != 0 && (data | (1 << 10)) != ~(u64)0)
2631 if (!msr_info->host_initiated &&
2632 (offset & 0x3) == 1 && data != 0) {
2633 if (!can_set_mci_status(vcpu))
2637 vcpu->arch.mce_banks[offset] = data;
2645 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2647 struct kvm *kvm = vcpu->kvm;
2648 int lm = is_long_mode(vcpu);
2649 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2650 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2651 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2652 : kvm->arch.xen_hvm_config.blob_size_32;
2653 u32 page_num = data & ~PAGE_MASK;
2654 u64 page_addr = data & PAGE_MASK;
2659 if (page_num >= blob_size)
2662 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2667 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2676 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2678 gpa_t gpa = data & ~0x3f;
2680 /* Bits 3:5 are reserved, Should be zero */
2684 vcpu->arch.apf.msr_val = data;
2686 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2687 kvm_clear_async_pf_completion_queue(vcpu);
2688 kvm_async_pf_hash_reset(vcpu);
2692 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2696 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2697 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2698 kvm_async_pf_wakeup_all(vcpu);
2702 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2704 vcpu->arch.pv_time_enabled = false;
2705 vcpu->arch.time = 0;
2708 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2710 ++vcpu->stat.tlb_flush;
2711 kvm_x86_ops.tlb_flush_all(vcpu);
2714 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2716 ++vcpu->stat.tlb_flush;
2717 kvm_x86_ops.tlb_flush_guest(vcpu);
2720 static void record_steal_time(struct kvm_vcpu *vcpu)
2722 struct kvm_host_map map;
2723 struct kvm_steal_time *st;
2725 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2728 /* -EAGAIN is returned in atomic context so we can just return. */
2729 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2730 &map, &vcpu->arch.st.cache, false))
2734 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2737 * Doing a TLB flush here, on the guest's behalf, can avoid
2740 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2741 st->preempted & KVM_VCPU_FLUSH_TLB);
2742 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2743 kvm_vcpu_flush_tlb_guest(vcpu);
2745 vcpu->arch.st.preempted = 0;
2747 if (st->version & 1)
2748 st->version += 1; /* first time write, random junk */
2754 st->steal += current->sched_info.run_delay -
2755 vcpu->arch.st.last_steal;
2756 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2762 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2765 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2768 u32 msr = msr_info->index;
2769 u64 data = msr_info->data;
2772 case MSR_AMD64_NB_CFG:
2773 case MSR_IA32_UCODE_WRITE:
2774 case MSR_VM_HSAVE_PA:
2775 case MSR_AMD64_PATCH_LOADER:
2776 case MSR_AMD64_BU_CFG2:
2777 case MSR_AMD64_DC_CFG:
2778 case MSR_F15H_EX_CFG:
2781 case MSR_IA32_UCODE_REV:
2782 if (msr_info->host_initiated)
2783 vcpu->arch.microcode_version = data;
2785 case MSR_IA32_ARCH_CAPABILITIES:
2786 if (!msr_info->host_initiated)
2788 vcpu->arch.arch_capabilities = data;
2791 return set_efer(vcpu, msr_info);
2793 data &= ~(u64)0x40; /* ignore flush filter disable */
2794 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2795 data &= ~(u64)0x8; /* ignore TLB cache disable */
2797 /* Handle McStatusWrEn */
2798 if (data == BIT_ULL(18)) {
2799 vcpu->arch.msr_hwcr = data;
2800 } else if (data != 0) {
2801 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2806 case MSR_FAM10H_MMIO_CONF_BASE:
2808 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2813 case MSR_IA32_DEBUGCTLMSR:
2815 /* We support the non-activated case already */
2817 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2818 /* Values other than LBR and BTF are vendor-specific,
2819 thus reserved and should throw a #GP */
2822 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2825 case 0x200 ... 0x2ff:
2826 return kvm_mtrr_set_msr(vcpu, msr, data);
2827 case MSR_IA32_APICBASE:
2828 return kvm_set_apic_base(vcpu, msr_info);
2829 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2830 return kvm_x2apic_msr_write(vcpu, msr, data);
2831 case MSR_IA32_TSCDEADLINE:
2832 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2834 case MSR_IA32_TSC_ADJUST:
2835 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2836 if (!msr_info->host_initiated) {
2837 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2838 adjust_tsc_offset_guest(vcpu, adj);
2840 vcpu->arch.ia32_tsc_adjust_msr = data;
2843 case MSR_IA32_MISC_ENABLE:
2844 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2845 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2846 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2848 vcpu->arch.ia32_misc_enable_msr = data;
2849 kvm_update_cpuid(vcpu);
2851 vcpu->arch.ia32_misc_enable_msr = data;
2854 case MSR_IA32_SMBASE:
2855 if (!msr_info->host_initiated)
2857 vcpu->arch.smbase = data;
2859 case MSR_IA32_POWER_CTL:
2860 vcpu->arch.msr_ia32_power_ctl = data;
2863 kvm_write_tsc(vcpu, msr_info);
2866 if (!msr_info->host_initiated &&
2867 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2870 * KVM supports exposing PT to the guest, but does not support
2871 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
2872 * XSAVES/XRSTORS to save/restore PT MSRs.
2874 if (data & ~supported_xss)
2876 vcpu->arch.ia32_xss = data;
2879 if (!msr_info->host_initiated)
2881 vcpu->arch.smi_count = data;
2883 case MSR_KVM_WALL_CLOCK_NEW:
2884 case MSR_KVM_WALL_CLOCK:
2885 vcpu->kvm->arch.wall_clock = data;
2886 kvm_write_wall_clock(vcpu->kvm, data);
2888 case MSR_KVM_SYSTEM_TIME_NEW:
2889 case MSR_KVM_SYSTEM_TIME: {
2890 struct kvm_arch *ka = &vcpu->kvm->arch;
2892 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2893 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2895 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2896 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2898 ka->boot_vcpu_runs_old_kvmclock = tmp;
2901 vcpu->arch.time = data;
2902 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2904 /* we verify if the enable bit is set... */
2905 vcpu->arch.pv_time_enabled = false;
2909 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2910 &vcpu->arch.pv_time, data & ~1ULL,
2911 sizeof(struct pvclock_vcpu_time_info)))
2912 vcpu->arch.pv_time_enabled = true;
2916 case MSR_KVM_ASYNC_PF_EN:
2917 if (kvm_pv_enable_async_pf(vcpu, data))
2920 case MSR_KVM_STEAL_TIME:
2922 if (unlikely(!sched_info_on()))
2925 if (data & KVM_STEAL_RESERVED_MASK)
2928 vcpu->arch.st.msr_val = data;
2930 if (!(data & KVM_MSR_ENABLED))
2933 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2936 case MSR_KVM_PV_EOI_EN:
2937 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2941 case MSR_KVM_POLL_CONTROL:
2942 /* only enable bit supported */
2943 if (data & (-1ULL << 1))
2946 vcpu->arch.msr_kvm_poll_control = data;
2949 case MSR_IA32_MCG_CTL:
2950 case MSR_IA32_MCG_STATUS:
2951 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2952 return set_msr_mce(vcpu, msr_info);
2954 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2955 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2956 pr = true; /* fall through */
2957 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2958 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2959 if (kvm_pmu_is_valid_msr(vcpu, msr))
2960 return kvm_pmu_set_msr(vcpu, msr_info);
2962 if (pr || data != 0)
2963 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2964 "0x%x data 0x%llx\n", msr, data);
2966 case MSR_K7_CLK_CTL:
2968 * Ignore all writes to this no longer documented MSR.
2969 * Writes are only relevant for old K7 processors,
2970 * all pre-dating SVM, but a recommended workaround from
2971 * AMD for these chips. It is possible to specify the
2972 * affected processor models on the command line, hence
2973 * the need to ignore the workaround.
2976 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2977 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2978 case HV_X64_MSR_CRASH_CTL:
2979 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2980 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2981 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2982 case HV_X64_MSR_TSC_EMULATION_STATUS:
2983 return kvm_hv_set_msr_common(vcpu, msr, data,
2984 msr_info->host_initiated);
2985 case MSR_IA32_BBL_CR_CTL3:
2986 /* Drop writes to this legacy MSR -- see rdmsr
2987 * counterpart for further detail.
2989 if (report_ignored_msrs)
2990 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2993 case MSR_AMD64_OSVW_ID_LENGTH:
2994 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2996 vcpu->arch.osvw.length = data;
2998 case MSR_AMD64_OSVW_STATUS:
2999 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3001 vcpu->arch.osvw.status = data;
3003 case MSR_PLATFORM_INFO:
3004 if (!msr_info->host_initiated ||
3005 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3006 cpuid_fault_enabled(vcpu)))
3008 vcpu->arch.msr_platform_info = data;
3010 case MSR_MISC_FEATURES_ENABLES:
3011 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3012 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3013 !supports_cpuid_fault(vcpu)))
3015 vcpu->arch.msr_misc_features_enables = data;
3018 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
3019 return xen_hvm_config(vcpu, data);
3020 if (kvm_pmu_is_valid_msr(vcpu, msr))
3021 return kvm_pmu_set_msr(vcpu, msr_info);
3023 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
3027 if (report_ignored_msrs)
3029 "ignored wrmsr: 0x%x data 0x%llx\n",
3036 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3038 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3041 u64 mcg_cap = vcpu->arch.mcg_cap;
3042 unsigned bank_num = mcg_cap & 0xff;
3045 case MSR_IA32_P5_MC_ADDR:
3046 case MSR_IA32_P5_MC_TYPE:
3049 case MSR_IA32_MCG_CAP:
3050 data = vcpu->arch.mcg_cap;
3052 case MSR_IA32_MCG_CTL:
3053 if (!(mcg_cap & MCG_CTL_P) && !host)
3055 data = vcpu->arch.mcg_ctl;
3057 case MSR_IA32_MCG_STATUS:
3058 data = vcpu->arch.mcg_status;
3061 if (msr >= MSR_IA32_MC0_CTL &&
3062 msr < MSR_IA32_MCx_CTL(bank_num)) {
3063 u32 offset = array_index_nospec(
3064 msr - MSR_IA32_MC0_CTL,
3065 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3067 data = vcpu->arch.mce_banks[offset];
3076 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3078 switch (msr_info->index) {
3079 case MSR_IA32_PLATFORM_ID:
3080 case MSR_IA32_EBL_CR_POWERON:
3081 case MSR_IA32_DEBUGCTLMSR:
3082 case MSR_IA32_LASTBRANCHFROMIP:
3083 case MSR_IA32_LASTBRANCHTOIP:
3084 case MSR_IA32_LASTINTFROMIP:
3085 case MSR_IA32_LASTINTTOIP:
3087 case MSR_K8_TSEG_ADDR:
3088 case MSR_K8_TSEG_MASK:
3089 case MSR_VM_HSAVE_PA:
3090 case MSR_K8_INT_PENDING_MSG:
3091 case MSR_AMD64_NB_CFG:
3092 case MSR_FAM10H_MMIO_CONF_BASE:
3093 case MSR_AMD64_BU_CFG2:
3094 case MSR_IA32_PERF_CTL:
3095 case MSR_AMD64_DC_CFG:
3096 case MSR_F15H_EX_CFG:
3098 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3099 * limit) MSRs. Just return 0, as we do not want to expose the host
3100 * data here. Do not conditionalize this on CPUID, as KVM does not do
3101 * so for existing CPU-specific MSRs.
3103 case MSR_RAPL_POWER_UNIT:
3104 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3105 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3106 case MSR_PKG_ENERGY_STATUS: /* Total package */
3107 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3110 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3111 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3112 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3113 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3114 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3115 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3116 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3119 case MSR_IA32_UCODE_REV:
3120 msr_info->data = vcpu->arch.microcode_version;
3122 case MSR_IA32_ARCH_CAPABILITIES:
3123 if (!msr_info->host_initiated &&
3124 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3126 msr_info->data = vcpu->arch.arch_capabilities;
3128 case MSR_IA32_POWER_CTL:
3129 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3132 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
3135 case 0x200 ... 0x2ff:
3136 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3137 case 0xcd: /* fsb frequency */
3141 * MSR_EBC_FREQUENCY_ID
3142 * Conservative value valid for even the basic CPU models.
3143 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3144 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3145 * and 266MHz for model 3, or 4. Set Core Clock
3146 * Frequency to System Bus Frequency Ratio to 1 (bits
3147 * 31:24) even though these are only valid for CPU
3148 * models > 2, however guests may end up dividing or
3149 * multiplying by zero otherwise.
3151 case MSR_EBC_FREQUENCY_ID:
3152 msr_info->data = 1 << 24;
3154 case MSR_IA32_APICBASE:
3155 msr_info->data = kvm_get_apic_base(vcpu);
3157 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
3158 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3159 case MSR_IA32_TSCDEADLINE:
3160 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3162 case MSR_IA32_TSC_ADJUST:
3163 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3165 case MSR_IA32_MISC_ENABLE:
3166 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3168 case MSR_IA32_SMBASE:
3169 if (!msr_info->host_initiated)
3171 msr_info->data = vcpu->arch.smbase;
3174 msr_info->data = vcpu->arch.smi_count;
3176 case MSR_IA32_PERF_STATUS:
3177 /* TSC increment by tick */
3178 msr_info->data = 1000ULL;
3179 /* CPU multiplier */
3180 msr_info->data |= (((uint64_t)4ULL) << 40);
3183 msr_info->data = vcpu->arch.efer;
3185 case MSR_KVM_WALL_CLOCK:
3186 case MSR_KVM_WALL_CLOCK_NEW:
3187 msr_info->data = vcpu->kvm->arch.wall_clock;
3189 case MSR_KVM_SYSTEM_TIME:
3190 case MSR_KVM_SYSTEM_TIME_NEW:
3191 msr_info->data = vcpu->arch.time;
3193 case MSR_KVM_ASYNC_PF_EN:
3194 msr_info->data = vcpu->arch.apf.msr_val;
3196 case MSR_KVM_STEAL_TIME:
3197 msr_info->data = vcpu->arch.st.msr_val;
3199 case MSR_KVM_PV_EOI_EN:
3200 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3202 case MSR_KVM_POLL_CONTROL:
3203 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3205 case MSR_IA32_P5_MC_ADDR:
3206 case MSR_IA32_P5_MC_TYPE:
3207 case MSR_IA32_MCG_CAP:
3208 case MSR_IA32_MCG_CTL:
3209 case MSR_IA32_MCG_STATUS:
3210 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3211 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3212 msr_info->host_initiated);
3214 if (!msr_info->host_initiated &&
3215 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3217 msr_info->data = vcpu->arch.ia32_xss;
3219 case MSR_K7_CLK_CTL:
3221 * Provide expected ramp-up count for K7. All other
3222 * are set to zero, indicating minimum divisors for
3225 * This prevents guest kernels on AMD host with CPU
3226 * type 6, model 8 and higher from exploding due to
3227 * the rdmsr failing.
3229 msr_info->data = 0x20000000;
3231 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3232 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3233 case HV_X64_MSR_CRASH_CTL:
3234 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3235 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3236 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3237 case HV_X64_MSR_TSC_EMULATION_STATUS:
3238 return kvm_hv_get_msr_common(vcpu,
3239 msr_info->index, &msr_info->data,
3240 msr_info->host_initiated);
3241 case MSR_IA32_BBL_CR_CTL3:
3242 /* This legacy MSR exists but isn't fully documented in current
3243 * silicon. It is however accessed by winxp in very narrow
3244 * scenarios where it sets bit #19, itself documented as
3245 * a "reserved" bit. Best effort attempt to source coherent
3246 * read data here should the balance of the register be
3247 * interpreted by the guest:
3249 * L2 cache control register 3: 64GB range, 256KB size,
3250 * enabled, latency 0x1, configured
3252 msr_info->data = 0xbe702111;
3254 case MSR_AMD64_OSVW_ID_LENGTH:
3255 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3257 msr_info->data = vcpu->arch.osvw.length;
3259 case MSR_AMD64_OSVW_STATUS:
3260 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3262 msr_info->data = vcpu->arch.osvw.status;
3264 case MSR_PLATFORM_INFO:
3265 if (!msr_info->host_initiated &&
3266 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3268 msr_info->data = vcpu->arch.msr_platform_info;
3270 case MSR_MISC_FEATURES_ENABLES:
3271 msr_info->data = vcpu->arch.msr_misc_features_enables;
3274 msr_info->data = vcpu->arch.msr_hwcr;
3277 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3278 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3280 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3284 if (report_ignored_msrs)
3285 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3293 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3296 * Read or write a bunch of msrs. All parameters are kernel addresses.
3298 * @return number of msrs set successfully.
3300 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3301 struct kvm_msr_entry *entries,
3302 int (*do_msr)(struct kvm_vcpu *vcpu,
3303 unsigned index, u64 *data))
3307 for (i = 0; i < msrs->nmsrs; ++i)
3308 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3315 * Read or write a bunch of msrs. Parameters are user addresses.
3317 * @return number of msrs set successfully.
3319 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3320 int (*do_msr)(struct kvm_vcpu *vcpu,
3321 unsigned index, u64 *data),
3324 struct kvm_msrs msrs;
3325 struct kvm_msr_entry *entries;
3330 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3334 if (msrs.nmsrs >= MAX_IO_MSRS)
3337 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3338 entries = memdup_user(user_msrs->entries, size);
3339 if (IS_ERR(entries)) {
3340 r = PTR_ERR(entries);
3344 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3349 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3360 static inline bool kvm_can_mwait_in_guest(void)
3362 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3363 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3364 boot_cpu_has(X86_FEATURE_ARAT);
3367 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3372 case KVM_CAP_IRQCHIP:
3374 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3375 case KVM_CAP_SET_TSS_ADDR:
3376 case KVM_CAP_EXT_CPUID:
3377 case KVM_CAP_EXT_EMUL_CPUID:
3378 case KVM_CAP_CLOCKSOURCE:
3380 case KVM_CAP_NOP_IO_DELAY:
3381 case KVM_CAP_MP_STATE:
3382 case KVM_CAP_SYNC_MMU:
3383 case KVM_CAP_USER_NMI:
3384 case KVM_CAP_REINJECT_CONTROL:
3385 case KVM_CAP_IRQ_INJECT_STATUS:
3386 case KVM_CAP_IOEVENTFD:
3387 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3389 case KVM_CAP_PIT_STATE2:
3390 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3391 case KVM_CAP_XEN_HVM:
3392 case KVM_CAP_VCPU_EVENTS:
3393 case KVM_CAP_HYPERV:
3394 case KVM_CAP_HYPERV_VAPIC:
3395 case KVM_CAP_HYPERV_SPIN:
3396 case KVM_CAP_HYPERV_SYNIC:
3397 case KVM_CAP_HYPERV_SYNIC2:
3398 case KVM_CAP_HYPERV_VP_INDEX:
3399 case KVM_CAP_HYPERV_EVENTFD:
3400 case KVM_CAP_HYPERV_TLBFLUSH:
3401 case KVM_CAP_HYPERV_SEND_IPI:
3402 case KVM_CAP_HYPERV_CPUID:
3403 case KVM_CAP_PCI_SEGMENT:
3404 case KVM_CAP_DEBUGREGS:
3405 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3407 case KVM_CAP_ASYNC_PF:
3408 case KVM_CAP_GET_TSC_KHZ:
3409 case KVM_CAP_KVMCLOCK_CTRL:
3410 case KVM_CAP_READONLY_MEM:
3411 case KVM_CAP_HYPERV_TIME:
3412 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3413 case KVM_CAP_TSC_DEADLINE_TIMER:
3414 case KVM_CAP_DISABLE_QUIRKS:
3415 case KVM_CAP_SET_BOOT_CPU_ID:
3416 case KVM_CAP_SPLIT_IRQCHIP:
3417 case KVM_CAP_IMMEDIATE_EXIT:
3418 case KVM_CAP_PMU_EVENT_FILTER:
3419 case KVM_CAP_GET_MSR_FEATURES:
3420 case KVM_CAP_MSR_PLATFORM_INFO:
3421 case KVM_CAP_EXCEPTION_PAYLOAD:
3422 case KVM_CAP_SET_GUEST_DEBUG:
3425 case KVM_CAP_SYNC_REGS:
3426 r = KVM_SYNC_X86_VALID_FIELDS;
3428 case KVM_CAP_ADJUST_CLOCK:
3429 r = KVM_CLOCK_TSC_STABLE;
3431 case KVM_CAP_X86_DISABLE_EXITS:
3432 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3433 KVM_X86_DISABLE_EXITS_CSTATE;
3434 if(kvm_can_mwait_in_guest())
3435 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3437 case KVM_CAP_X86_SMM:
3438 /* SMBASE is usually relocated above 1M on modern chipsets,
3439 * and SMM handlers might indeed rely on 4G segment limits,
3440 * so do not report SMM to be available if real mode is
3441 * emulated via vm86 mode. Still, do not go to great lengths
3442 * to avoid userspace's usage of the feature, because it is a
3443 * fringe case that is not enabled except via specific settings
3444 * of the module parameters.
3446 r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3449 r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3451 case KVM_CAP_NR_VCPUS:
3452 r = KVM_SOFT_MAX_VCPUS;
3454 case KVM_CAP_MAX_VCPUS:
3457 case KVM_CAP_MAX_VCPU_ID:
3458 r = KVM_MAX_VCPU_ID;
3460 case KVM_CAP_PV_MMU: /* obsolete */
3464 r = KVM_MAX_MCE_BANKS;
3467 r = boot_cpu_has(X86_FEATURE_XSAVE);
3469 case KVM_CAP_TSC_CONTROL:
3470 r = kvm_has_tsc_control;
3472 case KVM_CAP_X2APIC_API:
3473 r = KVM_X2APIC_API_VALID_FLAGS;
3475 case KVM_CAP_NESTED_STATE:
3476 r = kvm_x86_ops.nested_ops->get_state ?
3477 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3479 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3480 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3482 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3483 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3492 long kvm_arch_dev_ioctl(struct file *filp,
3493 unsigned int ioctl, unsigned long arg)
3495 void __user *argp = (void __user *)arg;
3499 case KVM_GET_MSR_INDEX_LIST: {
3500 struct kvm_msr_list __user *user_msr_list = argp;
3501 struct kvm_msr_list msr_list;
3505 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3508 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3509 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3512 if (n < msr_list.nmsrs)
3515 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3516 num_msrs_to_save * sizeof(u32)))
3518 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3520 num_emulated_msrs * sizeof(u32)))
3525 case KVM_GET_SUPPORTED_CPUID:
3526 case KVM_GET_EMULATED_CPUID: {
3527 struct kvm_cpuid2 __user *cpuid_arg = argp;
3528 struct kvm_cpuid2 cpuid;
3531 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3534 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3540 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3545 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3547 if (copy_to_user(argp, &kvm_mce_cap_supported,
3548 sizeof(kvm_mce_cap_supported)))
3552 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3553 struct kvm_msr_list __user *user_msr_list = argp;
3554 struct kvm_msr_list msr_list;
3558 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3561 msr_list.nmsrs = num_msr_based_features;
3562 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3565 if (n < msr_list.nmsrs)
3568 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3569 num_msr_based_features * sizeof(u32)))
3575 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3585 static void wbinvd_ipi(void *garbage)
3590 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3592 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3595 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3597 /* Address WBINVD may be executed by guest */
3598 if (need_emulate_wbinvd(vcpu)) {
3599 if (kvm_x86_ops.has_wbinvd_exit())
3600 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3601 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3602 smp_call_function_single(vcpu->cpu,
3603 wbinvd_ipi, NULL, 1);
3606 kvm_x86_ops.vcpu_load(vcpu, cpu);
3608 /* Save host pkru register if supported */
3609 vcpu->arch.host_pkru = read_pkru();
3611 /* Apply any externally detected TSC adjustments (due to suspend) */
3612 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3613 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3614 vcpu->arch.tsc_offset_adjustment = 0;
3615 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3618 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3619 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3620 rdtsc() - vcpu->arch.last_host_tsc;
3622 mark_tsc_unstable("KVM discovered backwards TSC");
3624 if (kvm_check_tsc_unstable()) {
3625 u64 offset = kvm_compute_tsc_offset(vcpu,
3626 vcpu->arch.last_guest_tsc);
3627 kvm_vcpu_write_tsc_offset(vcpu, offset);
3628 vcpu->arch.tsc_catchup = 1;
3631 if (kvm_lapic_hv_timer_in_use(vcpu))
3632 kvm_lapic_restart_hv_timer(vcpu);
3635 * On a host with synchronized TSC, there is no need to update
3636 * kvmclock on vcpu->cpu migration
3638 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3639 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3640 if (vcpu->cpu != cpu)
3641 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3645 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3648 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3650 struct kvm_host_map map;
3651 struct kvm_steal_time *st;
3653 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3656 if (vcpu->arch.st.preempted)
3659 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3660 &vcpu->arch.st.cache, true))
3664 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3666 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3668 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3671 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3675 if (vcpu->preempted)
3676 vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
3679 * Disable page faults because we're in atomic context here.
3680 * kvm_write_guest_offset_cached() would call might_fault()
3681 * that relies on pagefault_disable() to tell if there's a
3682 * bug. NOTE: the write to guest memory may not go through if
3683 * during postcopy live migration or if there's heavy guest
3686 pagefault_disable();
3688 * kvm_memslots() will be called by
3689 * kvm_write_guest_offset_cached() so take the srcu lock.
3691 idx = srcu_read_lock(&vcpu->kvm->srcu);
3692 kvm_steal_time_set_preempted(vcpu);
3693 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3695 kvm_x86_ops.vcpu_put(vcpu);
3696 vcpu->arch.last_host_tsc = rdtsc();
3698 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3699 * on every vmexit, but if not, we might have a stale dr6 from the
3700 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3705 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3706 struct kvm_lapic_state *s)
3708 if (vcpu->arch.apicv_active)
3709 kvm_x86_ops.sync_pir_to_irr(vcpu);
3711 return kvm_apic_get_state(vcpu, s);
3714 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3715 struct kvm_lapic_state *s)
3719 r = kvm_apic_set_state(vcpu, s);
3722 update_cr8_intercept(vcpu);
3727 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3729 return (!lapic_in_kernel(vcpu) ||
3730 kvm_apic_accept_pic_intr(vcpu));
3734 * if userspace requested an interrupt window, check that the
3735 * interrupt window is open.
3737 * No need to exit to userspace if we already have an interrupt queued.
3739 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3741 return kvm_arch_interrupt_allowed(vcpu) &&
3742 !kvm_cpu_has_interrupt(vcpu) &&
3743 !kvm_event_needs_reinjection(vcpu) &&
3744 kvm_cpu_accept_dm_intr(vcpu);
3747 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3748 struct kvm_interrupt *irq)
3750 if (irq->irq >= KVM_NR_INTERRUPTS)
3753 if (!irqchip_in_kernel(vcpu->kvm)) {
3754 kvm_queue_interrupt(vcpu, irq->irq, false);
3755 kvm_make_request(KVM_REQ_EVENT, vcpu);
3760 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3761 * fail for in-kernel 8259.
3763 if (pic_in_kernel(vcpu->kvm))
3766 if (vcpu->arch.pending_external_vector != -1)
3769 vcpu->arch.pending_external_vector = irq->irq;
3770 kvm_make_request(KVM_REQ_EVENT, vcpu);
3774 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3776 kvm_inject_nmi(vcpu);
3781 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3783 kvm_make_request(KVM_REQ_SMI, vcpu);
3788 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3789 struct kvm_tpr_access_ctl *tac)
3793 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3797 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3801 unsigned bank_num = mcg_cap & 0xff, bank;
3804 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
3806 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3809 vcpu->arch.mcg_cap = mcg_cap;
3810 /* Init IA32_MCG_CTL to all 1s */
3811 if (mcg_cap & MCG_CTL_P)
3812 vcpu->arch.mcg_ctl = ~(u64)0;
3813 /* Init IA32_MCi_CTL to all 1s */
3814 for (bank = 0; bank < bank_num; bank++)
3815 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3817 kvm_x86_ops.setup_mce(vcpu);
3822 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3823 struct kvm_x86_mce *mce)
3825 u64 mcg_cap = vcpu->arch.mcg_cap;
3826 unsigned bank_num = mcg_cap & 0xff;
3827 u64 *banks = vcpu->arch.mce_banks;
3829 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3832 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3833 * reporting is disabled
3835 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3836 vcpu->arch.mcg_ctl != ~(u64)0)
3838 banks += 4 * mce->bank;
3840 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3841 * reporting is disabled for the bank
3843 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3845 if (mce->status & MCI_STATUS_UC) {
3846 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3847 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3848 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3851 if (banks[1] & MCI_STATUS_VAL)
3852 mce->status |= MCI_STATUS_OVER;
3853 banks[2] = mce->addr;
3854 banks[3] = mce->misc;
3855 vcpu->arch.mcg_status = mce->mcg_status;
3856 banks[1] = mce->status;
3857 kvm_queue_exception(vcpu, MC_VECTOR);
3858 } else if (!(banks[1] & MCI_STATUS_VAL)
3859 || !(banks[1] & MCI_STATUS_UC)) {
3860 if (banks[1] & MCI_STATUS_VAL)
3861 mce->status |= MCI_STATUS_OVER;
3862 banks[2] = mce->addr;
3863 banks[3] = mce->misc;
3864 banks[1] = mce->status;
3866 banks[1] |= MCI_STATUS_OVER;
3870 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3871 struct kvm_vcpu_events *events)
3876 * In guest mode, payload delivery should be deferred,
3877 * so that the L1 hypervisor can intercept #PF before
3878 * CR2 is modified (or intercept #DB before DR6 is
3879 * modified under nVMX). Unless the per-VM capability,
3880 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
3881 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
3882 * opportunistically defer the exception payload, deliver it if the
3883 * capability hasn't been requested before processing a
3884 * KVM_GET_VCPU_EVENTS.
3886 if (!vcpu->kvm->arch.exception_payload_enabled &&
3887 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
3888 kvm_deliver_exception_payload(vcpu);
3891 * The API doesn't provide the instruction length for software
3892 * exceptions, so don't report them. As long as the guest RIP
3893 * isn't advanced, we should expect to encounter the exception
3896 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3897 events->exception.injected = 0;
3898 events->exception.pending = 0;
3900 events->exception.injected = vcpu->arch.exception.injected;
3901 events->exception.pending = vcpu->arch.exception.pending;
3903 * For ABI compatibility, deliberately conflate
3904 * pending and injected exceptions when
3905 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3907 if (!vcpu->kvm->arch.exception_payload_enabled)
3908 events->exception.injected |=
3909 vcpu->arch.exception.pending;
3911 events->exception.nr = vcpu->arch.exception.nr;
3912 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3913 events->exception.error_code = vcpu->arch.exception.error_code;
3914 events->exception_has_payload = vcpu->arch.exception.has_payload;
3915 events->exception_payload = vcpu->arch.exception.payload;
3917 events->interrupt.injected =
3918 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3919 events->interrupt.nr = vcpu->arch.interrupt.nr;
3920 events->interrupt.soft = 0;
3921 events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
3923 events->nmi.injected = vcpu->arch.nmi_injected;
3924 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3925 events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
3926 events->nmi.pad = 0;
3928 events->sipi_vector = 0; /* never valid when reporting to user space */
3930 events->smi.smm = is_smm(vcpu);
3931 events->smi.pending = vcpu->arch.smi_pending;
3932 events->smi.smm_inside_nmi =
3933 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3934 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3936 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3937 | KVM_VCPUEVENT_VALID_SHADOW
3938 | KVM_VCPUEVENT_VALID_SMM);
3939 if (vcpu->kvm->arch.exception_payload_enabled)
3940 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3942 memset(&events->reserved, 0, sizeof(events->reserved));
3945 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3947 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3948 struct kvm_vcpu_events *events)
3950 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3951 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3952 | KVM_VCPUEVENT_VALID_SHADOW
3953 | KVM_VCPUEVENT_VALID_SMM
3954 | KVM_VCPUEVENT_VALID_PAYLOAD))
3957 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3958 if (!vcpu->kvm->arch.exception_payload_enabled)
3960 if (events->exception.pending)
3961 events->exception.injected = 0;
3963 events->exception_has_payload = 0;
3965 events->exception.pending = 0;
3966 events->exception_has_payload = 0;
3969 if ((events->exception.injected || events->exception.pending) &&
3970 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3973 /* INITs are latched while in SMM */
3974 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3975 (events->smi.smm || events->smi.pending) &&
3976 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3980 vcpu->arch.exception.injected = events->exception.injected;
3981 vcpu->arch.exception.pending = events->exception.pending;
3982 vcpu->arch.exception.nr = events->exception.nr;
3983 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3984 vcpu->arch.exception.error_code = events->exception.error_code;
3985 vcpu->arch.exception.has_payload = events->exception_has_payload;
3986 vcpu->arch.exception.payload = events->exception_payload;
3988 vcpu->arch.interrupt.injected = events->interrupt.injected;
3989 vcpu->arch.interrupt.nr = events->interrupt.nr;
3990 vcpu->arch.interrupt.soft = events->interrupt.soft;
3991 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3992 kvm_x86_ops.set_interrupt_shadow(vcpu,
3993 events->interrupt.shadow);
3995 vcpu->arch.nmi_injected = events->nmi.injected;
3996 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3997 vcpu->arch.nmi_pending = events->nmi.pending;
3998 kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
4000 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4001 lapic_in_kernel(vcpu))
4002 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4004 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4005 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4006 if (events->smi.smm)
4007 vcpu->arch.hflags |= HF_SMM_MASK;
4009 vcpu->arch.hflags &= ~HF_SMM_MASK;
4010 kvm_smm_changed(vcpu);
4013 vcpu->arch.smi_pending = events->smi.pending;
4015 if (events->smi.smm) {
4016 if (events->smi.smm_inside_nmi)
4017 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4019 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4022 if (lapic_in_kernel(vcpu)) {
4023 if (events->smi.latched_init)
4024 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4026 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4030 kvm_make_request(KVM_REQ_EVENT, vcpu);
4035 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4036 struct kvm_debugregs *dbgregs)
4040 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4041 kvm_get_dr(vcpu, 6, &val);
4043 dbgregs->dr7 = vcpu->arch.dr7;
4045 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4048 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4049 struct kvm_debugregs *dbgregs)
4054 if (dbgregs->dr6 & ~0xffffffffull)
4056 if (dbgregs->dr7 & ~0xffffffffull)
4059 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4060 kvm_update_dr0123(vcpu);
4061 vcpu->arch.dr6 = dbgregs->dr6;
4062 vcpu->arch.dr7 = dbgregs->dr7;
4063 kvm_update_dr7(vcpu);
4068 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4070 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4072 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4073 u64 xstate_bv = xsave->header.xfeatures;
4077 * Copy legacy XSAVE area, to avoid complications with CPUID
4078 * leaves 0 and 1 in the loop below.
4080 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4083 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4084 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4087 * Copy each region from the possibly compacted offset to the
4088 * non-compacted offset.
4090 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4092 u64 xfeature_mask = valid & -valid;
4093 int xfeature_nr = fls64(xfeature_mask) - 1;
4094 void *src = get_xsave_addr(xsave, xfeature_nr);
4097 u32 size, offset, ecx, edx;
4098 cpuid_count(XSTATE_CPUID, xfeature_nr,
4099 &size, &offset, &ecx, &edx);
4100 if (xfeature_nr == XFEATURE_PKRU)
4101 memcpy(dest + offset, &vcpu->arch.pkru,
4102 sizeof(vcpu->arch.pkru));
4104 memcpy(dest + offset, src, size);
4108 valid -= xfeature_mask;
4112 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4114 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4115 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4119 * Copy legacy XSAVE area, to avoid complications with CPUID
4120 * leaves 0 and 1 in the loop below.
4122 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4124 /* Set XSTATE_BV and possibly XCOMP_BV. */
4125 xsave->header.xfeatures = xstate_bv;
4126 if (boot_cpu_has(X86_FEATURE_XSAVES))
4127 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4130 * Copy each region from the non-compacted offset to the
4131 * possibly compacted offset.
4133 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4135 u64 xfeature_mask = valid & -valid;
4136 int xfeature_nr = fls64(xfeature_mask) - 1;
4137 void *dest = get_xsave_addr(xsave, xfeature_nr);
4140 u32 size, offset, ecx, edx;
4141 cpuid_count(XSTATE_CPUID, xfeature_nr,
4142 &size, &offset, &ecx, &edx);
4143 if (xfeature_nr == XFEATURE_PKRU)
4144 memcpy(&vcpu->arch.pkru, src + offset,
4145 sizeof(vcpu->arch.pkru));
4147 memcpy(dest, src + offset, size);
4150 valid -= xfeature_mask;
4154 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4155 struct kvm_xsave *guest_xsave)
4157 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4158 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4159 fill_xsave((u8 *) guest_xsave->region, vcpu);
4161 memcpy(guest_xsave->region,
4162 &vcpu->arch.guest_fpu->state.fxsave,
4163 sizeof(struct fxregs_state));
4164 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4165 XFEATURE_MASK_FPSSE;
4169 #define XSAVE_MXCSR_OFFSET 24
4171 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4172 struct kvm_xsave *guest_xsave)
4175 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4176 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4178 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4180 * Here we allow setting states that are not present in
4181 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4182 * with old userspace.
4184 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4186 load_xsave(vcpu, (u8 *)guest_xsave->region);
4188 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4189 mxcsr & ~mxcsr_feature_mask)
4191 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4192 guest_xsave->region, sizeof(struct fxregs_state));
4197 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4198 struct kvm_xcrs *guest_xcrs)
4200 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4201 guest_xcrs->nr_xcrs = 0;
4205 guest_xcrs->nr_xcrs = 1;
4206 guest_xcrs->flags = 0;
4207 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4208 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4211 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4212 struct kvm_xcrs *guest_xcrs)
4216 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4219 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4222 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4223 /* Only support XCR0 currently */
4224 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4225 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4226 guest_xcrs->xcrs[i].value);
4235 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4236 * stopped by the hypervisor. This function will be called from the host only.
4237 * EINVAL is returned when the host attempts to set the flag for a guest that
4238 * does not support pv clocks.
4240 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4242 if (!vcpu->arch.pv_time_enabled)
4244 vcpu->arch.pvclock_set_guest_stopped_request = true;
4245 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4249 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4250 struct kvm_enable_cap *cap)
4253 uint16_t vmcs_version;
4254 void __user *user_ptr;
4260 case KVM_CAP_HYPERV_SYNIC2:
4265 case KVM_CAP_HYPERV_SYNIC:
4266 if (!irqchip_in_kernel(vcpu->kvm))
4268 return kvm_hv_activate_synic(vcpu, cap->cap ==
4269 KVM_CAP_HYPERV_SYNIC2);
4270 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4271 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4273 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4275 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4276 if (copy_to_user(user_ptr, &vmcs_version,
4277 sizeof(vmcs_version)))
4281 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4282 if (!kvm_x86_ops.enable_direct_tlbflush)
4285 return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4292 long kvm_arch_vcpu_ioctl(struct file *filp,
4293 unsigned int ioctl, unsigned long arg)
4295 struct kvm_vcpu *vcpu = filp->private_data;
4296 void __user *argp = (void __user *)arg;
4299 struct kvm_lapic_state *lapic;
4300 struct kvm_xsave *xsave;
4301 struct kvm_xcrs *xcrs;
4309 case KVM_GET_LAPIC: {
4311 if (!lapic_in_kernel(vcpu))
4313 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4314 GFP_KERNEL_ACCOUNT);
4319 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4323 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4328 case KVM_SET_LAPIC: {
4330 if (!lapic_in_kernel(vcpu))
4332 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4333 if (IS_ERR(u.lapic)) {
4334 r = PTR_ERR(u.lapic);
4338 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4341 case KVM_INTERRUPT: {
4342 struct kvm_interrupt irq;
4345 if (copy_from_user(&irq, argp, sizeof(irq)))
4347 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4351 r = kvm_vcpu_ioctl_nmi(vcpu);
4355 r = kvm_vcpu_ioctl_smi(vcpu);
4358 case KVM_SET_CPUID: {
4359 struct kvm_cpuid __user *cpuid_arg = argp;
4360 struct kvm_cpuid cpuid;
4363 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4365 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4368 case KVM_SET_CPUID2: {
4369 struct kvm_cpuid2 __user *cpuid_arg = argp;
4370 struct kvm_cpuid2 cpuid;
4373 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4375 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4376 cpuid_arg->entries);
4379 case KVM_GET_CPUID2: {
4380 struct kvm_cpuid2 __user *cpuid_arg = argp;
4381 struct kvm_cpuid2 cpuid;
4384 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4386 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4387 cpuid_arg->entries);
4391 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4396 case KVM_GET_MSRS: {
4397 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4398 r = msr_io(vcpu, argp, do_get_msr, 1);
4399 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4402 case KVM_SET_MSRS: {
4403 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4404 r = msr_io(vcpu, argp, do_set_msr, 0);
4405 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4408 case KVM_TPR_ACCESS_REPORTING: {
4409 struct kvm_tpr_access_ctl tac;
4412 if (copy_from_user(&tac, argp, sizeof(tac)))
4414 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4418 if (copy_to_user(argp, &tac, sizeof(tac)))
4423 case KVM_SET_VAPIC_ADDR: {
4424 struct kvm_vapic_addr va;
4428 if (!lapic_in_kernel(vcpu))
4431 if (copy_from_user(&va, argp, sizeof(va)))
4433 idx = srcu_read_lock(&vcpu->kvm->srcu);
4434 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4435 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4438 case KVM_X86_SETUP_MCE: {
4442 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4444 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4447 case KVM_X86_SET_MCE: {
4448 struct kvm_x86_mce mce;
4451 if (copy_from_user(&mce, argp, sizeof(mce)))
4453 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4456 case KVM_GET_VCPU_EVENTS: {
4457 struct kvm_vcpu_events events;
4459 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4462 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4467 case KVM_SET_VCPU_EVENTS: {
4468 struct kvm_vcpu_events events;
4471 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4474 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4477 case KVM_GET_DEBUGREGS: {
4478 struct kvm_debugregs dbgregs;
4480 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4483 if (copy_to_user(argp, &dbgregs,
4484 sizeof(struct kvm_debugregs)))
4489 case KVM_SET_DEBUGREGS: {
4490 struct kvm_debugregs dbgregs;
4493 if (copy_from_user(&dbgregs, argp,
4494 sizeof(struct kvm_debugregs)))
4497 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4500 case KVM_GET_XSAVE: {
4501 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4506 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4509 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4514 case KVM_SET_XSAVE: {
4515 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4516 if (IS_ERR(u.xsave)) {
4517 r = PTR_ERR(u.xsave);
4521 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4524 case KVM_GET_XCRS: {
4525 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4530 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4533 if (copy_to_user(argp, u.xcrs,
4534 sizeof(struct kvm_xcrs)))
4539 case KVM_SET_XCRS: {
4540 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4541 if (IS_ERR(u.xcrs)) {
4542 r = PTR_ERR(u.xcrs);
4546 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4549 case KVM_SET_TSC_KHZ: {
4553 user_tsc_khz = (u32)arg;
4555 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4558 if (user_tsc_khz == 0)
4559 user_tsc_khz = tsc_khz;
4561 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4566 case KVM_GET_TSC_KHZ: {
4567 r = vcpu->arch.virtual_tsc_khz;
4570 case KVM_KVMCLOCK_CTRL: {
4571 r = kvm_set_guest_paused(vcpu);
4574 case KVM_ENABLE_CAP: {
4575 struct kvm_enable_cap cap;
4578 if (copy_from_user(&cap, argp, sizeof(cap)))
4580 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4583 case KVM_GET_NESTED_STATE: {
4584 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4588 if (!kvm_x86_ops.nested_ops->get_state)
4591 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4593 if (get_user(user_data_size, &user_kvm_nested_state->size))
4596 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4601 if (r > user_data_size) {
4602 if (put_user(r, &user_kvm_nested_state->size))
4612 case KVM_SET_NESTED_STATE: {
4613 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4614 struct kvm_nested_state kvm_state;
4618 if (!kvm_x86_ops.nested_ops->set_state)
4622 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4626 if (kvm_state.size < sizeof(kvm_state))
4629 if (kvm_state.flags &
4630 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4631 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING))
4634 /* nested_run_pending implies guest_mode. */
4635 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4636 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4639 idx = srcu_read_lock(&vcpu->kvm->srcu);
4640 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
4641 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4644 case KVM_GET_SUPPORTED_HV_CPUID: {
4645 struct kvm_cpuid2 __user *cpuid_arg = argp;
4646 struct kvm_cpuid2 cpuid;
4649 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4652 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4653 cpuid_arg->entries);
4658 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4673 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4675 return VM_FAULT_SIGBUS;
4678 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4682 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4684 ret = kvm_x86_ops.set_tss_addr(kvm, addr);
4688 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4691 return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
4694 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4695 unsigned long kvm_nr_mmu_pages)
4697 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4700 mutex_lock(&kvm->slots_lock);
4702 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4703 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4705 mutex_unlock(&kvm->slots_lock);
4709 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4711 return kvm->arch.n_max_mmu_pages;
4714 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4716 struct kvm_pic *pic = kvm->arch.vpic;
4720 switch (chip->chip_id) {
4721 case KVM_IRQCHIP_PIC_MASTER:
4722 memcpy(&chip->chip.pic, &pic->pics[0],
4723 sizeof(struct kvm_pic_state));
4725 case KVM_IRQCHIP_PIC_SLAVE:
4726 memcpy(&chip->chip.pic, &pic->pics[1],
4727 sizeof(struct kvm_pic_state));
4729 case KVM_IRQCHIP_IOAPIC:
4730 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4739 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4741 struct kvm_pic *pic = kvm->arch.vpic;
4745 switch (chip->chip_id) {
4746 case KVM_IRQCHIP_PIC_MASTER:
4747 spin_lock(&pic->lock);
4748 memcpy(&pic->pics[0], &chip->chip.pic,
4749 sizeof(struct kvm_pic_state));
4750 spin_unlock(&pic->lock);
4752 case KVM_IRQCHIP_PIC_SLAVE:
4753 spin_lock(&pic->lock);
4754 memcpy(&pic->pics[1], &chip->chip.pic,
4755 sizeof(struct kvm_pic_state));
4756 spin_unlock(&pic->lock);
4758 case KVM_IRQCHIP_IOAPIC:
4759 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4765 kvm_pic_update_irq(pic);
4769 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4771 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4773 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4775 mutex_lock(&kps->lock);
4776 memcpy(ps, &kps->channels, sizeof(*ps));
4777 mutex_unlock(&kps->lock);
4781 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4784 struct kvm_pit *pit = kvm->arch.vpit;
4786 mutex_lock(&pit->pit_state.lock);
4787 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4788 for (i = 0; i < 3; i++)
4789 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4790 mutex_unlock(&pit->pit_state.lock);
4794 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4796 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4797 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4798 sizeof(ps->channels));
4799 ps->flags = kvm->arch.vpit->pit_state.flags;
4800 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4801 memset(&ps->reserved, 0, sizeof(ps->reserved));
4805 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4809 u32 prev_legacy, cur_legacy;
4810 struct kvm_pit *pit = kvm->arch.vpit;
4812 mutex_lock(&pit->pit_state.lock);
4813 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4814 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4815 if (!prev_legacy && cur_legacy)
4817 memcpy(&pit->pit_state.channels, &ps->channels,
4818 sizeof(pit->pit_state.channels));
4819 pit->pit_state.flags = ps->flags;
4820 for (i = 0; i < 3; i++)
4821 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4823 mutex_unlock(&pit->pit_state.lock);
4827 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4828 struct kvm_reinject_control *control)
4830 struct kvm_pit *pit = kvm->arch.vpit;
4832 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4833 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4834 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4836 mutex_lock(&pit->pit_state.lock);
4837 kvm_pit_set_reinject(pit, control->pit_reinject);
4838 mutex_unlock(&pit->pit_state.lock);
4843 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
4846 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4848 if (kvm_x86_ops.flush_log_dirty)
4849 kvm_x86_ops.flush_log_dirty(kvm);
4852 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4855 if (!irqchip_in_kernel(kvm))
4858 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4859 irq_event->irq, irq_event->level,
4864 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4865 struct kvm_enable_cap *cap)
4873 case KVM_CAP_DISABLE_QUIRKS:
4874 kvm->arch.disabled_quirks = cap->args[0];
4877 case KVM_CAP_SPLIT_IRQCHIP: {
4878 mutex_lock(&kvm->lock);
4880 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4881 goto split_irqchip_unlock;
4883 if (irqchip_in_kernel(kvm))
4884 goto split_irqchip_unlock;
4885 if (kvm->created_vcpus)
4886 goto split_irqchip_unlock;
4887 r = kvm_setup_empty_irq_routing(kvm);
4889 goto split_irqchip_unlock;
4890 /* Pairs with irqchip_in_kernel. */
4892 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4893 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4895 split_irqchip_unlock:
4896 mutex_unlock(&kvm->lock);
4899 case KVM_CAP_X2APIC_API:
4901 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4904 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4905 kvm->arch.x2apic_format = true;
4906 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4907 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4911 case KVM_CAP_X86_DISABLE_EXITS:
4913 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4916 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4917 kvm_can_mwait_in_guest())
4918 kvm->arch.mwait_in_guest = true;
4919 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4920 kvm->arch.hlt_in_guest = true;
4921 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4922 kvm->arch.pause_in_guest = true;
4923 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4924 kvm->arch.cstate_in_guest = true;
4927 case KVM_CAP_MSR_PLATFORM_INFO:
4928 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4931 case KVM_CAP_EXCEPTION_PAYLOAD:
4932 kvm->arch.exception_payload_enabled = cap->args[0];
4942 long kvm_arch_vm_ioctl(struct file *filp,
4943 unsigned int ioctl, unsigned long arg)
4945 struct kvm *kvm = filp->private_data;
4946 void __user *argp = (void __user *)arg;
4949 * This union makes it completely explicit to gcc-3.x
4950 * that these two variables' stack usage should be
4951 * combined, not added together.
4954 struct kvm_pit_state ps;
4955 struct kvm_pit_state2 ps2;
4956 struct kvm_pit_config pit_config;
4960 case KVM_SET_TSS_ADDR:
4961 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4963 case KVM_SET_IDENTITY_MAP_ADDR: {
4966 mutex_lock(&kvm->lock);
4968 if (kvm->created_vcpus)
4969 goto set_identity_unlock;
4971 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4972 goto set_identity_unlock;
4973 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4974 set_identity_unlock:
4975 mutex_unlock(&kvm->lock);
4978 case KVM_SET_NR_MMU_PAGES:
4979 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4981 case KVM_GET_NR_MMU_PAGES:
4982 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4984 case KVM_CREATE_IRQCHIP: {
4985 mutex_lock(&kvm->lock);
4988 if (irqchip_in_kernel(kvm))
4989 goto create_irqchip_unlock;
4992 if (kvm->created_vcpus)
4993 goto create_irqchip_unlock;
4995 r = kvm_pic_init(kvm);
4997 goto create_irqchip_unlock;
4999 r = kvm_ioapic_init(kvm);
5001 kvm_pic_destroy(kvm);
5002 goto create_irqchip_unlock;
5005 r = kvm_setup_default_irq_routing(kvm);
5007 kvm_ioapic_destroy(kvm);
5008 kvm_pic_destroy(kvm);
5009 goto create_irqchip_unlock;
5011 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5013 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5014 create_irqchip_unlock:
5015 mutex_unlock(&kvm->lock);
5018 case KVM_CREATE_PIT:
5019 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5021 case KVM_CREATE_PIT2:
5023 if (copy_from_user(&u.pit_config, argp,
5024 sizeof(struct kvm_pit_config)))
5027 mutex_lock(&kvm->lock);
5030 goto create_pit_unlock;
5032 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5036 mutex_unlock(&kvm->lock);
5038 case KVM_GET_IRQCHIP: {
5039 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5040 struct kvm_irqchip *chip;
5042 chip = memdup_user(argp, sizeof(*chip));
5049 if (!irqchip_kernel(kvm))
5050 goto get_irqchip_out;
5051 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5053 goto get_irqchip_out;
5055 if (copy_to_user(argp, chip, sizeof(*chip)))
5056 goto get_irqchip_out;
5062 case KVM_SET_IRQCHIP: {
5063 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5064 struct kvm_irqchip *chip;
5066 chip = memdup_user(argp, sizeof(*chip));
5073 if (!irqchip_kernel(kvm))
5074 goto set_irqchip_out;
5075 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5082 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5085 if (!kvm->arch.vpit)
5087 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5091 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5098 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5100 mutex_lock(&kvm->lock);
5102 if (!kvm->arch.vpit)
5104 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5106 mutex_unlock(&kvm->lock);
5109 case KVM_GET_PIT2: {
5111 if (!kvm->arch.vpit)
5113 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5117 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5122 case KVM_SET_PIT2: {
5124 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5126 mutex_lock(&kvm->lock);
5128 if (!kvm->arch.vpit)
5130 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5132 mutex_unlock(&kvm->lock);
5135 case KVM_REINJECT_CONTROL: {
5136 struct kvm_reinject_control control;
5138 if (copy_from_user(&control, argp, sizeof(control)))
5141 if (!kvm->arch.vpit)
5143 r = kvm_vm_ioctl_reinject(kvm, &control);
5146 case KVM_SET_BOOT_CPU_ID:
5148 mutex_lock(&kvm->lock);
5149 if (kvm->created_vcpus)
5152 kvm->arch.bsp_vcpu_id = arg;
5153 mutex_unlock(&kvm->lock);
5155 case KVM_XEN_HVM_CONFIG: {
5156 struct kvm_xen_hvm_config xhc;
5158 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5163 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5167 case KVM_SET_CLOCK: {
5168 struct kvm_clock_data user_ns;
5172 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5181 * TODO: userspace has to take care of races with VCPU_RUN, so
5182 * kvm_gen_update_masterclock() can be cut down to locked
5183 * pvclock_update_vm_gtod_copy().
5185 kvm_gen_update_masterclock(kvm);
5186 now_ns = get_kvmclock_ns(kvm);
5187 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5188 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5191 case KVM_GET_CLOCK: {
5192 struct kvm_clock_data user_ns;
5195 now_ns = get_kvmclock_ns(kvm);
5196 user_ns.clock = now_ns;
5197 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5198 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5201 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5206 case KVM_MEMORY_ENCRYPT_OP: {
5208 if (kvm_x86_ops.mem_enc_op)
5209 r = kvm_x86_ops.mem_enc_op(kvm, argp);
5212 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5213 struct kvm_enc_region region;
5216 if (copy_from_user(®ion, argp, sizeof(region)))
5220 if (kvm_x86_ops.mem_enc_reg_region)
5221 r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion);
5224 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5225 struct kvm_enc_region region;
5228 if (copy_from_user(®ion, argp, sizeof(region)))
5232 if (kvm_x86_ops.mem_enc_unreg_region)
5233 r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion);
5236 case KVM_HYPERV_EVENTFD: {
5237 struct kvm_hyperv_eventfd hvevfd;
5240 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5242 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5245 case KVM_SET_PMU_EVENT_FILTER:
5246 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5255 static void kvm_init_msr_list(void)
5257 struct x86_pmu_capability x86_pmu;
5261 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5262 "Please update the fixed PMCs in msrs_to_saved_all[]");
5264 perf_get_x86_pmu_capability(&x86_pmu);
5266 num_msrs_to_save = 0;
5267 num_emulated_msrs = 0;
5268 num_msr_based_features = 0;
5270 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5271 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5275 * Even MSRs that are valid in the host may not be exposed
5276 * to the guests in some cases.
5278 switch (msrs_to_save_all[i]) {
5279 case MSR_IA32_BNDCFGS:
5280 if (!kvm_mpx_supported())
5284 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5287 case MSR_IA32_UMWAIT_CONTROL:
5288 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5291 case MSR_IA32_RTIT_CTL:
5292 case MSR_IA32_RTIT_STATUS:
5293 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5296 case MSR_IA32_RTIT_CR3_MATCH:
5297 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5298 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5301 case MSR_IA32_RTIT_OUTPUT_BASE:
5302 case MSR_IA32_RTIT_OUTPUT_MASK:
5303 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5304 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5305 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5308 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5309 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5310 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5311 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5314 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5315 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5316 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5319 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5320 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5321 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5328 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5331 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5332 if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5335 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5338 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5339 struct kvm_msr_entry msr;
5341 msr.index = msr_based_features_all[i];
5342 if (kvm_get_msr_feature(&msr))
5345 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5349 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5357 if (!(lapic_in_kernel(vcpu) &&
5358 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5359 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5370 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5377 if (!(lapic_in_kernel(vcpu) &&
5378 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5380 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5382 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5392 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5393 struct kvm_segment *var, int seg)
5395 kvm_x86_ops.set_segment(vcpu, var, seg);
5398 void kvm_get_segment(struct kvm_vcpu *vcpu,
5399 struct kvm_segment *var, int seg)
5401 kvm_x86_ops.get_segment(vcpu, var, seg);
5404 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5405 struct x86_exception *exception)
5409 BUG_ON(!mmu_is_nested(vcpu));
5411 /* NPT walks are always user-walks */
5412 access |= PFERR_USER_MASK;
5413 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5418 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5419 struct x86_exception *exception)
5421 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5422 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5425 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5426 struct x86_exception *exception)
5428 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5429 access |= PFERR_FETCH_MASK;
5430 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5433 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5434 struct x86_exception *exception)
5436 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5437 access |= PFERR_WRITE_MASK;
5438 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5441 /* uses this to access any guest's mapped memory without checking CPL */
5442 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5443 struct x86_exception *exception)
5445 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5448 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5449 struct kvm_vcpu *vcpu, u32 access,
5450 struct x86_exception *exception)
5453 int r = X86EMUL_CONTINUE;
5456 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5458 unsigned offset = addr & (PAGE_SIZE-1);
5459 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5462 if (gpa == UNMAPPED_GVA)
5463 return X86EMUL_PROPAGATE_FAULT;
5464 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5467 r = X86EMUL_IO_NEEDED;
5479 /* used for instruction fetching */
5480 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5481 gva_t addr, void *val, unsigned int bytes,
5482 struct x86_exception *exception)
5484 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5485 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5489 /* Inline kvm_read_guest_virt_helper for speed. */
5490 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5492 if (unlikely(gpa == UNMAPPED_GVA))
5493 return X86EMUL_PROPAGATE_FAULT;
5495 offset = addr & (PAGE_SIZE-1);
5496 if (WARN_ON(offset + bytes > PAGE_SIZE))
5497 bytes = (unsigned)PAGE_SIZE - offset;
5498 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5500 if (unlikely(ret < 0))
5501 return X86EMUL_IO_NEEDED;
5503 return X86EMUL_CONTINUE;
5506 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5507 gva_t addr, void *val, unsigned int bytes,
5508 struct x86_exception *exception)
5510 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5513 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5514 * is returned, but our callers are not ready for that and they blindly
5515 * call kvm_inject_page_fault. Ensure that they at least do not leak
5516 * uninitialized kernel stack memory into cr2 and error code.
5518 memset(exception, 0, sizeof(*exception));
5519 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5522 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5524 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5525 gva_t addr, void *val, unsigned int bytes,
5526 struct x86_exception *exception, bool system)
5528 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5531 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5532 access |= PFERR_USER_MASK;
5534 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5537 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5538 unsigned long addr, void *val, unsigned int bytes)
5540 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5541 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5543 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5546 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5547 struct kvm_vcpu *vcpu, u32 access,
5548 struct x86_exception *exception)
5551 int r = X86EMUL_CONTINUE;
5554 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5557 unsigned offset = addr & (PAGE_SIZE-1);
5558 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5561 if (gpa == UNMAPPED_GVA)
5562 return X86EMUL_PROPAGATE_FAULT;
5563 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5565 r = X86EMUL_IO_NEEDED;
5577 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5578 unsigned int bytes, struct x86_exception *exception,
5581 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5582 u32 access = PFERR_WRITE_MASK;
5584 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5585 access |= PFERR_USER_MASK;
5587 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5591 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5592 unsigned int bytes, struct x86_exception *exception)
5594 /* kvm_write_guest_virt_system can pull in tons of pages. */
5595 vcpu->arch.l1tf_flush_l1d = true;
5598 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5599 * is returned, but our callers are not ready for that and they blindly
5600 * call kvm_inject_page_fault. Ensure that they at least do not leak
5601 * uninitialized kernel stack memory into cr2 and error code.
5603 memset(exception, 0, sizeof(*exception));
5604 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5605 PFERR_WRITE_MASK, exception);
5607 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5609 int handle_ud(struct kvm_vcpu *vcpu)
5611 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5612 int emul_type = EMULTYPE_TRAP_UD;
5613 char sig[5]; /* ud2; .ascii "kvm" */
5614 struct x86_exception e;
5616 if (force_emulation_prefix &&
5617 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5618 sig, sizeof(sig), &e) == 0 &&
5619 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5620 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5621 emul_type = EMULTYPE_TRAP_UD_FORCED;
5624 return kvm_emulate_instruction(vcpu, emul_type);
5626 EXPORT_SYMBOL_GPL(handle_ud);
5628 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5629 gpa_t gpa, bool write)
5631 /* For APIC access vmexit */
5632 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5635 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5636 trace_vcpu_match_mmio(gva, gpa, write, true);
5643 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5644 gpa_t *gpa, struct x86_exception *exception,
5647 u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5648 | (write ? PFERR_WRITE_MASK : 0);
5651 * currently PKRU is only applied to ept enabled guest so
5652 * there is no pkey in EPT page table for L1 guest or EPT
5653 * shadow page table for L2 guest.
5655 if (vcpu_match_mmio_gva(vcpu, gva)
5656 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5657 vcpu->arch.mmio_access, 0, access)) {
5658 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5659 (gva & (PAGE_SIZE - 1));
5660 trace_vcpu_match_mmio(gva, *gpa, write, false);
5664 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5666 if (*gpa == UNMAPPED_GVA)
5669 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5672 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5673 const void *val, int bytes)
5677 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5680 kvm_page_track_write(vcpu, gpa, val, bytes);
5684 struct read_write_emulator_ops {
5685 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5687 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5688 void *val, int bytes);
5689 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5690 int bytes, void *val);
5691 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5692 void *val, int bytes);
5696 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5698 if (vcpu->mmio_read_completed) {
5699 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5700 vcpu->mmio_fragments[0].gpa, val);
5701 vcpu->mmio_read_completed = 0;
5708 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5709 void *val, int bytes)
5711 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5714 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5715 void *val, int bytes)
5717 return emulator_write_phys(vcpu, gpa, val, bytes);
5720 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5722 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5723 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5726 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5727 void *val, int bytes)
5729 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5730 return X86EMUL_IO_NEEDED;
5733 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5734 void *val, int bytes)
5736 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5738 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5739 return X86EMUL_CONTINUE;
5742 static const struct read_write_emulator_ops read_emultor = {
5743 .read_write_prepare = read_prepare,
5744 .read_write_emulate = read_emulate,
5745 .read_write_mmio = vcpu_mmio_read,
5746 .read_write_exit_mmio = read_exit_mmio,
5749 static const struct read_write_emulator_ops write_emultor = {
5750 .read_write_emulate = write_emulate,
5751 .read_write_mmio = write_mmio,
5752 .read_write_exit_mmio = write_exit_mmio,
5756 static int emulator_read_write_onepage(unsigned long addr, void *val,
5758 struct x86_exception *exception,
5759 struct kvm_vcpu *vcpu,
5760 const struct read_write_emulator_ops *ops)
5764 bool write = ops->write;
5765 struct kvm_mmio_fragment *frag;
5766 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
5769 * If the exit was due to a NPF we may already have a GPA.
5770 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5771 * Note, this cannot be used on string operations since string
5772 * operation using rep will only have the initial GPA from the NPF
5775 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
5776 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
5777 gpa = ctxt->gpa_val;
5778 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5780 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5782 return X86EMUL_PROPAGATE_FAULT;
5785 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5786 return X86EMUL_CONTINUE;
5789 * Is this MMIO handled locally?
5791 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5792 if (handled == bytes)
5793 return X86EMUL_CONTINUE;
5799 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5800 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5804 return X86EMUL_CONTINUE;
5807 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5809 void *val, unsigned int bytes,
5810 struct x86_exception *exception,
5811 const struct read_write_emulator_ops *ops)
5813 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5817 if (ops->read_write_prepare &&
5818 ops->read_write_prepare(vcpu, val, bytes))
5819 return X86EMUL_CONTINUE;
5821 vcpu->mmio_nr_fragments = 0;
5823 /* Crossing a page boundary? */
5824 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5827 now = -addr & ~PAGE_MASK;
5828 rc = emulator_read_write_onepage(addr, val, now, exception,
5831 if (rc != X86EMUL_CONTINUE)
5834 if (ctxt->mode != X86EMUL_MODE_PROT64)
5840 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5842 if (rc != X86EMUL_CONTINUE)
5845 if (!vcpu->mmio_nr_fragments)
5848 gpa = vcpu->mmio_fragments[0].gpa;
5850 vcpu->mmio_needed = 1;
5851 vcpu->mmio_cur_fragment = 0;
5853 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5854 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5855 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5856 vcpu->run->mmio.phys_addr = gpa;
5858 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5861 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5865 struct x86_exception *exception)
5867 return emulator_read_write(ctxt, addr, val, bytes,
5868 exception, &read_emultor);
5871 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5875 struct x86_exception *exception)
5877 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5878 exception, &write_emultor);
5881 #define CMPXCHG_TYPE(t, ptr, old, new) \
5882 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5884 #ifdef CONFIG_X86_64
5885 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5887 # define CMPXCHG64(ptr, old, new) \
5888 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5891 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5896 struct x86_exception *exception)
5898 struct kvm_host_map map;
5899 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5905 /* guests cmpxchg8b have to be emulated atomically */
5906 if (bytes > 8 || (bytes & (bytes - 1)))
5909 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5911 if (gpa == UNMAPPED_GVA ||
5912 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5916 * Emulate the atomic as a straight write to avoid #AC if SLD is
5917 * enabled in the host and the access splits a cache line.
5919 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
5920 page_line_mask = ~(cache_line_size() - 1);
5922 page_line_mask = PAGE_MASK;
5924 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
5927 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5930 kaddr = map.hva + offset_in_page(gpa);
5934 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5937 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5940 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5943 exchanged = CMPXCHG64(kaddr, old, new);
5949 kvm_vcpu_unmap(vcpu, &map, true);
5952 return X86EMUL_CMPXCHG_FAILED;
5954 kvm_page_track_write(vcpu, gpa, new, bytes);
5956 return X86EMUL_CONTINUE;
5959 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5961 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5964 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5968 for (i = 0; i < vcpu->arch.pio.count; i++) {
5969 if (vcpu->arch.pio.in)
5970 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5971 vcpu->arch.pio.size, pd);
5973 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5974 vcpu->arch.pio.port, vcpu->arch.pio.size,
5978 pd += vcpu->arch.pio.size;
5983 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5984 unsigned short port, void *val,
5985 unsigned int count, bool in)
5987 vcpu->arch.pio.port = port;
5988 vcpu->arch.pio.in = in;
5989 vcpu->arch.pio.count = count;
5990 vcpu->arch.pio.size = size;
5992 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5993 vcpu->arch.pio.count = 0;
5997 vcpu->run->exit_reason = KVM_EXIT_IO;
5998 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5999 vcpu->run->io.size = size;
6000 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6001 vcpu->run->io.count = count;
6002 vcpu->run->io.port = port;
6007 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6008 unsigned short port, void *val, unsigned int count)
6012 if (vcpu->arch.pio.count)
6015 memset(vcpu->arch.pio_data, 0, size * count);
6017 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6020 memcpy(val, vcpu->arch.pio_data, size * count);
6021 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6022 vcpu->arch.pio.count = 0;
6029 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6030 int size, unsigned short port, void *val,
6033 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6037 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6038 unsigned short port, const void *val,
6041 memcpy(vcpu->arch.pio_data, val, size * count);
6042 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6043 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6046 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6047 int size, unsigned short port,
6048 const void *val, unsigned int count)
6050 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6053 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6055 return kvm_x86_ops.get_segment_base(vcpu, seg);
6058 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6060 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6063 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6065 if (!need_emulate_wbinvd(vcpu))
6066 return X86EMUL_CONTINUE;
6068 if (kvm_x86_ops.has_wbinvd_exit()) {
6069 int cpu = get_cpu();
6071 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6072 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6073 wbinvd_ipi, NULL, 1);
6075 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6078 return X86EMUL_CONTINUE;
6081 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6083 kvm_emulate_wbinvd_noskip(vcpu);
6084 return kvm_skip_emulated_instruction(vcpu);
6086 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6090 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6092 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6095 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6096 unsigned long *dest)
6098 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6101 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6102 unsigned long value)
6105 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6108 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6110 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6113 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6115 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6116 unsigned long value;
6120 value = kvm_read_cr0(vcpu);
6123 value = vcpu->arch.cr2;
6126 value = kvm_read_cr3(vcpu);
6129 value = kvm_read_cr4(vcpu);
6132 value = kvm_get_cr8(vcpu);
6135 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6142 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6144 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6149 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6152 vcpu->arch.cr2 = val;
6155 res = kvm_set_cr3(vcpu, val);
6158 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6161 res = kvm_set_cr8(vcpu, val);
6164 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6171 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6173 return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6176 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6178 kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6181 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6183 kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6186 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6188 kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6191 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6193 kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6196 static unsigned long emulator_get_cached_segment_base(
6197 struct x86_emulate_ctxt *ctxt, int seg)
6199 return get_segment_base(emul_to_vcpu(ctxt), seg);
6202 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6203 struct desc_struct *desc, u32 *base3,
6206 struct kvm_segment var;
6208 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6209 *selector = var.selector;
6212 memset(desc, 0, sizeof(*desc));
6220 set_desc_limit(desc, var.limit);
6221 set_desc_base(desc, (unsigned long)var.base);
6222 #ifdef CONFIG_X86_64
6224 *base3 = var.base >> 32;
6226 desc->type = var.type;
6228 desc->dpl = var.dpl;
6229 desc->p = var.present;
6230 desc->avl = var.avl;
6238 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6239 struct desc_struct *desc, u32 base3,
6242 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6243 struct kvm_segment var;
6245 var.selector = selector;
6246 var.base = get_desc_base(desc);
6247 #ifdef CONFIG_X86_64
6248 var.base |= ((u64)base3) << 32;
6250 var.limit = get_desc_limit(desc);
6252 var.limit = (var.limit << 12) | 0xfff;
6253 var.type = desc->type;
6254 var.dpl = desc->dpl;
6259 var.avl = desc->avl;
6260 var.present = desc->p;
6261 var.unusable = !var.present;
6264 kvm_set_segment(vcpu, &var, seg);
6268 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6269 u32 msr_index, u64 *pdata)
6271 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6274 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6275 u32 msr_index, u64 data)
6277 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6280 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6282 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6284 return vcpu->arch.smbase;
6287 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6289 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6291 vcpu->arch.smbase = smbase;
6294 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6297 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6300 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6301 u32 pmc, u64 *pdata)
6303 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6306 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6308 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6311 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6312 struct x86_instruction_info *info,
6313 enum x86_intercept_stage stage)
6315 return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6319 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6320 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6323 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6326 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6328 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6331 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6333 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6336 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6338 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6341 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6343 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6346 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6348 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6351 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6353 kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6356 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6358 return emul_to_vcpu(ctxt)->arch.hflags;
6361 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6363 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6366 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6367 const char *smstate)
6369 return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6372 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6374 kvm_smm_changed(emul_to_vcpu(ctxt));
6377 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6379 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6382 static const struct x86_emulate_ops emulate_ops = {
6383 .read_gpr = emulator_read_gpr,
6384 .write_gpr = emulator_write_gpr,
6385 .read_std = emulator_read_std,
6386 .write_std = emulator_write_std,
6387 .read_phys = kvm_read_guest_phys_system,
6388 .fetch = kvm_fetch_guest_virt,
6389 .read_emulated = emulator_read_emulated,
6390 .write_emulated = emulator_write_emulated,
6391 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6392 .invlpg = emulator_invlpg,
6393 .pio_in_emulated = emulator_pio_in_emulated,
6394 .pio_out_emulated = emulator_pio_out_emulated,
6395 .get_segment = emulator_get_segment,
6396 .set_segment = emulator_set_segment,
6397 .get_cached_segment_base = emulator_get_cached_segment_base,
6398 .get_gdt = emulator_get_gdt,
6399 .get_idt = emulator_get_idt,
6400 .set_gdt = emulator_set_gdt,
6401 .set_idt = emulator_set_idt,
6402 .get_cr = emulator_get_cr,
6403 .set_cr = emulator_set_cr,
6404 .cpl = emulator_get_cpl,
6405 .get_dr = emulator_get_dr,
6406 .set_dr = emulator_set_dr,
6407 .get_smbase = emulator_get_smbase,
6408 .set_smbase = emulator_set_smbase,
6409 .set_msr = emulator_set_msr,
6410 .get_msr = emulator_get_msr,
6411 .check_pmc = emulator_check_pmc,
6412 .read_pmc = emulator_read_pmc,
6413 .halt = emulator_halt,
6414 .wbinvd = emulator_wbinvd,
6415 .fix_hypercall = emulator_fix_hypercall,
6416 .intercept = emulator_intercept,
6417 .get_cpuid = emulator_get_cpuid,
6418 .guest_has_long_mode = emulator_guest_has_long_mode,
6419 .guest_has_movbe = emulator_guest_has_movbe,
6420 .guest_has_fxsr = emulator_guest_has_fxsr,
6421 .set_nmi_mask = emulator_set_nmi_mask,
6422 .get_hflags = emulator_get_hflags,
6423 .set_hflags = emulator_set_hflags,
6424 .pre_leave_smm = emulator_pre_leave_smm,
6425 .post_leave_smm = emulator_post_leave_smm,
6426 .set_xcr = emulator_set_xcr,
6429 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6431 u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6433 * an sti; sti; sequence only disable interrupts for the first
6434 * instruction. So, if the last instruction, be it emulated or
6435 * not, left the system with the INT_STI flag enabled, it
6436 * means that the last instruction is an sti. We should not
6437 * leave the flag on in this case. The same goes for mov ss
6439 if (int_shadow & mask)
6441 if (unlikely(int_shadow || mask)) {
6442 kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6444 kvm_make_request(KVM_REQ_EVENT, vcpu);
6448 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6450 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6451 if (ctxt->exception.vector == PF_VECTOR)
6452 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
6454 if (ctxt->exception.error_code_valid)
6455 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6456 ctxt->exception.error_code);
6458 kvm_queue_exception(vcpu, ctxt->exception.vector);
6462 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6464 struct x86_emulate_ctxt *ctxt;
6466 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6468 pr_err("kvm: failed to allocate vcpu's emulator\n");
6473 ctxt->ops = &emulate_ops;
6474 vcpu->arch.emulate_ctxt = ctxt;
6479 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6481 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6484 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6486 ctxt->gpa_available = false;
6487 ctxt->eflags = kvm_get_rflags(vcpu);
6488 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6490 ctxt->eip = kvm_rip_read(vcpu);
6491 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6492 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6493 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6494 cs_db ? X86EMUL_MODE_PROT32 :
6495 X86EMUL_MODE_PROT16;
6496 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6497 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6498 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6500 init_decode_cache(ctxt);
6501 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6504 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6506 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6509 init_emulate_ctxt(vcpu);
6513 ctxt->_eip = ctxt->eip + inc_eip;
6514 ret = emulate_int_real(ctxt, irq);
6516 if (ret != X86EMUL_CONTINUE) {
6517 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6519 ctxt->eip = ctxt->_eip;
6520 kvm_rip_write(vcpu, ctxt->eip);
6521 kvm_set_rflags(vcpu, ctxt->eflags);
6524 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6526 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6528 ++vcpu->stat.insn_emulation_fail;
6529 trace_kvm_emulate_insn_failed(vcpu);
6531 if (emulation_type & EMULTYPE_VMWARE_GP) {
6532 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6536 if (emulation_type & EMULTYPE_SKIP) {
6537 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6538 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6539 vcpu->run->internal.ndata = 0;
6543 kvm_queue_exception(vcpu, UD_VECTOR);
6545 if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
6546 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6547 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6548 vcpu->run->internal.ndata = 0;
6555 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6556 bool write_fault_to_shadow_pgtable,
6559 gpa_t gpa = cr2_or_gpa;
6562 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6565 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6566 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6569 if (!vcpu->arch.mmu->direct_map) {
6571 * Write permission should be allowed since only
6572 * write access need to be emulated.
6574 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6577 * If the mapping is invalid in guest, let cpu retry
6578 * it to generate fault.
6580 if (gpa == UNMAPPED_GVA)
6585 * Do not retry the unhandleable instruction if it faults on the
6586 * readonly host memory, otherwise it will goto a infinite loop:
6587 * retry instruction -> write #PF -> emulation fail -> retry
6588 * instruction -> ...
6590 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6593 * If the instruction failed on the error pfn, it can not be fixed,
6594 * report the error to userspace.
6596 if (is_error_noslot_pfn(pfn))
6599 kvm_release_pfn_clean(pfn);
6601 /* The instructions are well-emulated on direct mmu. */
6602 if (vcpu->arch.mmu->direct_map) {
6603 unsigned int indirect_shadow_pages;
6605 spin_lock(&vcpu->kvm->mmu_lock);
6606 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6607 spin_unlock(&vcpu->kvm->mmu_lock);
6609 if (indirect_shadow_pages)
6610 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6616 * if emulation was due to access to shadowed page table
6617 * and it failed try to unshadow page and re-enter the
6618 * guest to let CPU execute the instruction.
6620 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6623 * If the access faults on its page table, it can not
6624 * be fixed by unprotecting shadow page and it should
6625 * be reported to userspace.
6627 return !write_fault_to_shadow_pgtable;
6630 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6631 gpa_t cr2_or_gpa, int emulation_type)
6633 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6634 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
6636 last_retry_eip = vcpu->arch.last_retry_eip;
6637 last_retry_addr = vcpu->arch.last_retry_addr;
6640 * If the emulation is caused by #PF and it is non-page_table
6641 * writing instruction, it means the VM-EXIT is caused by shadow
6642 * page protected, we can zap the shadow page and retry this
6643 * instruction directly.
6645 * Note: if the guest uses a non-page-table modifying instruction
6646 * on the PDE that points to the instruction, then we will unmap
6647 * the instruction and go to an infinite loop. So, we cache the
6648 * last retried eip and the last fault address, if we meet the eip
6649 * and the address again, we can break out of the potential infinite
6652 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6654 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6657 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6658 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6661 if (x86_page_table_writing_insn(ctxt))
6664 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
6667 vcpu->arch.last_retry_eip = ctxt->eip;
6668 vcpu->arch.last_retry_addr = cr2_or_gpa;
6670 if (!vcpu->arch.mmu->direct_map)
6671 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6673 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6678 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6679 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6681 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6683 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6684 /* This is a good place to trace that we are exiting SMM. */
6685 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6687 /* Process a latched INIT or SMI, if any. */
6688 kvm_make_request(KVM_REQ_EVENT, vcpu);
6691 kvm_mmu_reset_context(vcpu);
6694 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6703 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6704 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6709 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6711 struct kvm_run *kvm_run = vcpu->run;
6713 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6714 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6715 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6716 kvm_run->debug.arch.exception = DB_VECTOR;
6717 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6720 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6724 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6726 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6729 r = kvm_x86_ops.skip_emulated_instruction(vcpu);
6734 * rflags is the old, "raw" value of the flags. The new value has
6735 * not been saved yet.
6737 * This is correct even for TF set by the guest, because "the
6738 * processor will not generate this exception after the instruction
6739 * that sets the TF flag".
6741 if (unlikely(rflags & X86_EFLAGS_TF))
6742 r = kvm_vcpu_do_singlestep(vcpu);
6745 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6747 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6749 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6750 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6751 struct kvm_run *kvm_run = vcpu->run;
6752 unsigned long eip = kvm_get_linear_rip(vcpu);
6753 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6754 vcpu->arch.guest_debug_dr7,
6758 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6759 kvm_run->debug.arch.pc = eip;
6760 kvm_run->debug.arch.exception = DB_VECTOR;
6761 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6767 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6768 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6769 unsigned long eip = kvm_get_linear_rip(vcpu);
6770 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6775 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
6784 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6786 switch (ctxt->opcode_len) {
6793 case 0xe6: /* OUT */
6797 case 0x6c: /* INS */
6799 case 0x6e: /* OUTS */
6806 case 0x33: /* RDPMC */
6815 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6816 int emulation_type, void *insn, int insn_len)
6819 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6820 bool writeback = true;
6821 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6823 vcpu->arch.l1tf_flush_l1d = true;
6826 * Clear write_fault_to_shadow_pgtable here to ensure it is
6829 vcpu->arch.write_fault_to_shadow_pgtable = false;
6830 kvm_clear_exception_queue(vcpu);
6832 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6833 init_emulate_ctxt(vcpu);
6836 * We will reenter on the same instruction since
6837 * we do not set complete_userspace_io. This does not
6838 * handle watchpoints yet, those would be handled in
6841 if (!(emulation_type & EMULTYPE_SKIP) &&
6842 kvm_vcpu_check_breakpoint(vcpu, &r))
6845 ctxt->interruptibility = 0;
6846 ctxt->have_exception = false;
6847 ctxt->exception.vector = -1;
6848 ctxt->perm_ok = false;
6850 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6852 r = x86_decode_insn(ctxt, insn, insn_len);
6854 trace_kvm_emulate_insn_start(vcpu);
6855 ++vcpu->stat.insn_emulation;
6856 if (r != EMULATION_OK) {
6857 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6858 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6859 kvm_queue_exception(vcpu, UD_VECTOR);
6862 if (reexecute_instruction(vcpu, cr2_or_gpa,
6866 if (ctxt->have_exception) {
6868 * #UD should result in just EMULATION_FAILED, and trap-like
6869 * exception should not be encountered during decode.
6871 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6872 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6873 inject_emulated_exception(vcpu);
6876 return handle_emulation_failure(vcpu, emulation_type);
6880 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6881 !is_vmware_backdoor_opcode(ctxt)) {
6882 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6887 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6888 * for kvm_skip_emulated_instruction(). The caller is responsible for
6889 * updating interruptibility state and injecting single-step #DBs.
6891 if (emulation_type & EMULTYPE_SKIP) {
6892 kvm_rip_write(vcpu, ctxt->_eip);
6893 if (ctxt->eflags & X86_EFLAGS_RF)
6894 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6898 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
6901 /* this is needed for vmware backdoor interface to work since it
6902 changes registers values during IO operation */
6903 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6904 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6905 emulator_invalidate_register_cache(ctxt);
6909 if (emulation_type & EMULTYPE_PF) {
6910 /* Save the faulting GPA (cr2) in the address field */
6911 ctxt->exception.address = cr2_or_gpa;
6913 /* With shadow page tables, cr2 contains a GVA or nGPA. */
6914 if (vcpu->arch.mmu->direct_map) {
6915 ctxt->gpa_available = true;
6916 ctxt->gpa_val = cr2_or_gpa;
6919 /* Sanitize the address out of an abundance of paranoia. */
6920 ctxt->exception.address = 0;
6923 r = x86_emulate_insn(ctxt);
6925 if (r == EMULATION_INTERCEPTED)
6928 if (r == EMULATION_FAILED) {
6929 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
6933 return handle_emulation_failure(vcpu, emulation_type);
6936 if (ctxt->have_exception) {
6938 if (inject_emulated_exception(vcpu))
6940 } else if (vcpu->arch.pio.count) {
6941 if (!vcpu->arch.pio.in) {
6942 /* FIXME: return into emulator if single-stepping. */
6943 vcpu->arch.pio.count = 0;
6946 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6949 } else if (vcpu->mmio_needed) {
6950 ++vcpu->stat.mmio_exits;
6952 if (!vcpu->mmio_is_write)
6955 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6956 } else if (r == EMULATION_RESTART)
6962 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6963 toggle_interruptibility(vcpu, ctxt->interruptibility);
6964 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6965 if (!ctxt->have_exception ||
6966 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6967 kvm_rip_write(vcpu, ctxt->eip);
6969 r = kvm_vcpu_do_singlestep(vcpu);
6970 if (kvm_x86_ops.update_emulated_instruction)
6971 kvm_x86_ops.update_emulated_instruction(vcpu);
6972 __kvm_set_rflags(vcpu, ctxt->eflags);
6976 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6977 * do nothing, and it will be requested again as soon as
6978 * the shadow expires. But we still need to check here,
6979 * because POPF has no interrupt shadow.
6981 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6982 kvm_make_request(KVM_REQ_EVENT, vcpu);
6984 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6989 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6991 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6993 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6995 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6996 void *insn, int insn_len)
6998 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7000 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7002 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7004 vcpu->arch.pio.count = 0;
7008 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7010 vcpu->arch.pio.count = 0;
7012 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7015 return kvm_skip_emulated_instruction(vcpu);
7018 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7019 unsigned short port)
7021 unsigned long val = kvm_rax_read(vcpu);
7022 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7028 * Workaround userspace that relies on old KVM behavior of %rip being
7029 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7032 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7033 vcpu->arch.complete_userspace_io =
7034 complete_fast_pio_out_port_0x7e;
7035 kvm_skip_emulated_instruction(vcpu);
7037 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7038 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7043 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7047 /* We should only ever be called with arch.pio.count equal to 1 */
7048 BUG_ON(vcpu->arch.pio.count != 1);
7050 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7051 vcpu->arch.pio.count = 0;
7055 /* For size less than 4 we merge, else we zero extend */
7056 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7059 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7060 * the copy and tracing
7062 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7063 kvm_rax_write(vcpu, val);
7065 return kvm_skip_emulated_instruction(vcpu);
7068 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7069 unsigned short port)
7074 /* For size less than 4 we merge, else we zero extend */
7075 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7077 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7079 kvm_rax_write(vcpu, val);
7083 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7084 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7089 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7094 ret = kvm_fast_pio_in(vcpu, size, port);
7096 ret = kvm_fast_pio_out(vcpu, size, port);
7097 return ret && kvm_skip_emulated_instruction(vcpu);
7099 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7101 static int kvmclock_cpu_down_prep(unsigned int cpu)
7103 __this_cpu_write(cpu_tsc_khz, 0);
7107 static void tsc_khz_changed(void *data)
7109 struct cpufreq_freqs *freq = data;
7110 unsigned long khz = 0;
7114 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7115 khz = cpufreq_quick_get(raw_smp_processor_id());
7118 __this_cpu_write(cpu_tsc_khz, khz);
7121 #ifdef CONFIG_X86_64
7122 static void kvm_hyperv_tsc_notifier(void)
7125 struct kvm_vcpu *vcpu;
7128 mutex_lock(&kvm_lock);
7129 list_for_each_entry(kvm, &vm_list, vm_list)
7130 kvm_make_mclock_inprogress_request(kvm);
7132 hyperv_stop_tsc_emulation();
7134 /* TSC frequency always matches when on Hyper-V */
7135 for_each_present_cpu(cpu)
7136 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7137 kvm_max_guest_tsc_khz = tsc_khz;
7139 list_for_each_entry(kvm, &vm_list, vm_list) {
7140 struct kvm_arch *ka = &kvm->arch;
7142 spin_lock(&ka->pvclock_gtod_sync_lock);
7144 pvclock_update_vm_gtod_copy(kvm);
7146 kvm_for_each_vcpu(cpu, vcpu, kvm)
7147 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7149 kvm_for_each_vcpu(cpu, vcpu, kvm)
7150 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7152 spin_unlock(&ka->pvclock_gtod_sync_lock);
7154 mutex_unlock(&kvm_lock);
7158 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7161 struct kvm_vcpu *vcpu;
7162 int i, send_ipi = 0;
7165 * We allow guests to temporarily run on slowing clocks,
7166 * provided we notify them after, or to run on accelerating
7167 * clocks, provided we notify them before. Thus time never
7170 * However, we have a problem. We can't atomically update
7171 * the frequency of a given CPU from this function; it is
7172 * merely a notifier, which can be called from any CPU.
7173 * Changing the TSC frequency at arbitrary points in time
7174 * requires a recomputation of local variables related to
7175 * the TSC for each VCPU. We must flag these local variables
7176 * to be updated and be sure the update takes place with the
7177 * new frequency before any guests proceed.
7179 * Unfortunately, the combination of hotplug CPU and frequency
7180 * change creates an intractable locking scenario; the order
7181 * of when these callouts happen is undefined with respect to
7182 * CPU hotplug, and they can race with each other. As such,
7183 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7184 * undefined; you can actually have a CPU frequency change take
7185 * place in between the computation of X and the setting of the
7186 * variable. To protect against this problem, all updates of
7187 * the per_cpu tsc_khz variable are done in an interrupt
7188 * protected IPI, and all callers wishing to update the value
7189 * must wait for a synchronous IPI to complete (which is trivial
7190 * if the caller is on the CPU already). This establishes the
7191 * necessary total order on variable updates.
7193 * Note that because a guest time update may take place
7194 * anytime after the setting of the VCPU's request bit, the
7195 * correct TSC value must be set before the request. However,
7196 * to ensure the update actually makes it to any guest which
7197 * starts running in hardware virtualization between the set
7198 * and the acquisition of the spinlock, we must also ping the
7199 * CPU after setting the request bit.
7203 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7205 mutex_lock(&kvm_lock);
7206 list_for_each_entry(kvm, &vm_list, vm_list) {
7207 kvm_for_each_vcpu(i, vcpu, kvm) {
7208 if (vcpu->cpu != cpu)
7210 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7211 if (vcpu->cpu != raw_smp_processor_id())
7215 mutex_unlock(&kvm_lock);
7217 if (freq->old < freq->new && send_ipi) {
7219 * We upscale the frequency. Must make the guest
7220 * doesn't see old kvmclock values while running with
7221 * the new frequency, otherwise we risk the guest sees
7222 * time go backwards.
7224 * In case we update the frequency for another cpu
7225 * (which might be in guest context) send an interrupt
7226 * to kick the cpu out of guest context. Next time
7227 * guest context is entered kvmclock will be updated,
7228 * so the guest will not see stale values.
7230 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7234 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7237 struct cpufreq_freqs *freq = data;
7240 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7242 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7245 for_each_cpu(cpu, freq->policy->cpus)
7246 __kvmclock_cpufreq_notifier(freq, cpu);
7251 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7252 .notifier_call = kvmclock_cpufreq_notifier
7255 static int kvmclock_cpu_online(unsigned int cpu)
7257 tsc_khz_changed(NULL);
7261 static void kvm_timer_init(void)
7263 max_tsc_khz = tsc_khz;
7265 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7266 #ifdef CONFIG_CPU_FREQ
7267 struct cpufreq_policy *policy;
7271 policy = cpufreq_cpu_get(cpu);
7273 if (policy->cpuinfo.max_freq)
7274 max_tsc_khz = policy->cpuinfo.max_freq;
7275 cpufreq_cpu_put(policy);
7279 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7280 CPUFREQ_TRANSITION_NOTIFIER);
7283 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7284 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7287 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7288 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7290 int kvm_is_in_guest(void)
7292 return __this_cpu_read(current_vcpu) != NULL;
7295 static int kvm_is_user_mode(void)
7299 if (__this_cpu_read(current_vcpu))
7300 user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7302 return user_mode != 0;
7305 static unsigned long kvm_get_guest_ip(void)
7307 unsigned long ip = 0;
7309 if (__this_cpu_read(current_vcpu))
7310 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7315 static void kvm_handle_intel_pt_intr(void)
7317 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7319 kvm_make_request(KVM_REQ_PMI, vcpu);
7320 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7321 (unsigned long *)&vcpu->arch.pmu.global_status);
7324 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7325 .is_in_guest = kvm_is_in_guest,
7326 .is_user_mode = kvm_is_user_mode,
7327 .get_guest_ip = kvm_get_guest_ip,
7328 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7331 #ifdef CONFIG_X86_64
7332 static void pvclock_gtod_update_fn(struct work_struct *work)
7336 struct kvm_vcpu *vcpu;
7339 mutex_lock(&kvm_lock);
7340 list_for_each_entry(kvm, &vm_list, vm_list)
7341 kvm_for_each_vcpu(i, vcpu, kvm)
7342 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7343 atomic_set(&kvm_guest_has_master_clock, 0);
7344 mutex_unlock(&kvm_lock);
7347 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7350 * Notification about pvclock gtod data update.
7352 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7355 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7356 struct timekeeper *tk = priv;
7358 update_pvclock_gtod(tk);
7360 /* disable master clock if host does not trust, or does not
7361 * use, TSC based clocksource.
7363 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7364 atomic_read(&kvm_guest_has_master_clock) != 0)
7365 queue_work(system_long_wq, &pvclock_gtod_work);
7370 static struct notifier_block pvclock_gtod_notifier = {
7371 .notifier_call = pvclock_gtod_notify,
7375 int kvm_arch_init(void *opaque)
7377 struct kvm_x86_init_ops *ops = opaque;
7380 if (kvm_x86_ops.hardware_enable) {
7381 printk(KERN_ERR "kvm: already loaded the other module\n");
7386 if (!ops->cpu_has_kvm_support()) {
7387 pr_err_ratelimited("kvm: no hardware support\n");
7391 if (ops->disabled_by_bios()) {
7392 pr_err_ratelimited("kvm: disabled by bios\n");
7398 * KVM explicitly assumes that the guest has an FPU and
7399 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7400 * vCPU's FPU state as a fxregs_state struct.
7402 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7403 printk(KERN_ERR "kvm: inadequate fpu\n");
7409 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7410 __alignof__(struct fpu), SLAB_ACCOUNT,
7412 if (!x86_fpu_cache) {
7413 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7417 x86_emulator_cache = kvm_alloc_emulator_cache();
7418 if (!x86_emulator_cache) {
7419 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7420 goto out_free_x86_fpu_cache;
7423 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7425 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7426 goto out_free_x86_emulator_cache;
7429 r = kvm_mmu_module_init();
7431 goto out_free_percpu;
7433 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7434 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7435 PT_PRESENT_MASK, 0, sme_me_mask);
7438 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7440 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7441 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7442 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7446 if (pi_inject_timer == -1)
7447 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7448 #ifdef CONFIG_X86_64
7449 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7451 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7452 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7458 free_percpu(shared_msrs);
7459 out_free_x86_emulator_cache:
7460 kmem_cache_destroy(x86_emulator_cache);
7461 out_free_x86_fpu_cache:
7462 kmem_cache_destroy(x86_fpu_cache);
7467 void kvm_arch_exit(void)
7469 #ifdef CONFIG_X86_64
7470 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7471 clear_hv_tscchange_cb();
7474 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7476 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7477 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7478 CPUFREQ_TRANSITION_NOTIFIER);
7479 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7480 #ifdef CONFIG_X86_64
7481 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7483 kvm_x86_ops.hardware_enable = NULL;
7484 kvm_mmu_module_exit();
7485 free_percpu(shared_msrs);
7486 kmem_cache_destroy(x86_fpu_cache);
7489 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7491 ++vcpu->stat.halt_exits;
7492 if (lapic_in_kernel(vcpu)) {
7493 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7496 vcpu->run->exit_reason = KVM_EXIT_HLT;
7500 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7502 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7504 int ret = kvm_skip_emulated_instruction(vcpu);
7506 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7507 * KVM_EXIT_DEBUG here.
7509 return kvm_vcpu_halt(vcpu) && ret;
7511 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7513 #ifdef CONFIG_X86_64
7514 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7515 unsigned long clock_type)
7517 struct kvm_clock_pairing clock_pairing;
7518 struct timespec64 ts;
7522 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7523 return -KVM_EOPNOTSUPP;
7525 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7526 return -KVM_EOPNOTSUPP;
7528 clock_pairing.sec = ts.tv_sec;
7529 clock_pairing.nsec = ts.tv_nsec;
7530 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7531 clock_pairing.flags = 0;
7532 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7535 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7536 sizeof(struct kvm_clock_pairing)))
7544 * kvm_pv_kick_cpu_op: Kick a vcpu.
7546 * @apicid - apicid of vcpu to be kicked.
7548 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7550 struct kvm_lapic_irq lapic_irq;
7552 lapic_irq.shorthand = APIC_DEST_NOSHORT;
7553 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7554 lapic_irq.level = 0;
7555 lapic_irq.dest_id = apicid;
7556 lapic_irq.msi_redir_hint = false;
7558 lapic_irq.delivery_mode = APIC_DM_REMRD;
7559 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7562 bool kvm_apicv_activated(struct kvm *kvm)
7564 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
7566 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
7568 void kvm_apicv_init(struct kvm *kvm, bool enable)
7571 clear_bit(APICV_INHIBIT_REASON_DISABLE,
7572 &kvm->arch.apicv_inhibit_reasons);
7574 set_bit(APICV_INHIBIT_REASON_DISABLE,
7575 &kvm->arch.apicv_inhibit_reasons);
7577 EXPORT_SYMBOL_GPL(kvm_apicv_init);
7579 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7581 struct kvm_vcpu *target = NULL;
7582 struct kvm_apic_map *map;
7585 map = rcu_dereference(kvm->arch.apic_map);
7587 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7588 target = map->phys_map[dest_id]->vcpu;
7592 if (target && READ_ONCE(target->ready))
7593 kvm_vcpu_yield_to(target);
7596 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7598 unsigned long nr, a0, a1, a2, a3, ret;
7601 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7602 return kvm_hv_hypercall(vcpu);
7604 nr = kvm_rax_read(vcpu);
7605 a0 = kvm_rbx_read(vcpu);
7606 a1 = kvm_rcx_read(vcpu);
7607 a2 = kvm_rdx_read(vcpu);
7608 a3 = kvm_rsi_read(vcpu);
7610 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7612 op_64_bit = is_64_bit_mode(vcpu);
7621 if (kvm_x86_ops.get_cpl(vcpu) != 0) {
7627 case KVM_HC_VAPIC_POLL_IRQ:
7630 case KVM_HC_KICK_CPU:
7631 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7632 kvm_sched_yield(vcpu->kvm, a1);
7635 #ifdef CONFIG_X86_64
7636 case KVM_HC_CLOCK_PAIRING:
7637 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7640 case KVM_HC_SEND_IPI:
7641 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7643 case KVM_HC_SCHED_YIELD:
7644 kvm_sched_yield(vcpu->kvm, a0);
7654 kvm_rax_write(vcpu, ret);
7656 ++vcpu->stat.hypercalls;
7657 return kvm_skip_emulated_instruction(vcpu);
7659 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7661 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7663 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7664 char instruction[3];
7665 unsigned long rip = kvm_rip_read(vcpu);
7667 kvm_x86_ops.patch_hypercall(vcpu, instruction);
7669 return emulator_write_emulated(ctxt, rip, instruction, 3,
7673 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7675 return vcpu->run->request_interrupt_window &&
7676 likely(!pic_in_kernel(vcpu->kvm));
7679 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7681 struct kvm_run *kvm_run = vcpu->run;
7683 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7684 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7685 kvm_run->cr8 = kvm_get_cr8(vcpu);
7686 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7687 kvm_run->ready_for_interrupt_injection =
7688 pic_in_kernel(vcpu->kvm) ||
7689 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7692 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7696 if (!kvm_x86_ops.update_cr8_intercept)
7699 if (!lapic_in_kernel(vcpu))
7702 if (vcpu->arch.apicv_active)
7705 if (!vcpu->arch.apic->vapic_addr)
7706 max_irr = kvm_lapic_find_highest_irr(vcpu);
7713 tpr = kvm_lapic_get_cr8(vcpu);
7715 kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
7718 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
7721 bool can_inject = true;
7723 /* try to reinject previous events if any */
7725 if (vcpu->arch.exception.injected) {
7726 kvm_x86_ops.queue_exception(vcpu);
7730 * Do not inject an NMI or interrupt if there is a pending
7731 * exception. Exceptions and interrupts are recognized at
7732 * instruction boundaries, i.e. the start of an instruction.
7733 * Trap-like exceptions, e.g. #DB, have higher priority than
7734 * NMIs and interrupts, i.e. traps are recognized before an
7735 * NMI/interrupt that's pending on the same instruction.
7736 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7737 * priority, but are only generated (pended) during instruction
7738 * execution, i.e. a pending fault-like exception means the
7739 * fault occurred on the *previous* instruction and must be
7740 * serviced prior to recognizing any new events in order to
7741 * fully complete the previous instruction.
7743 else if (!vcpu->arch.exception.pending) {
7744 if (vcpu->arch.nmi_injected) {
7745 kvm_x86_ops.set_nmi(vcpu);
7747 } else if (vcpu->arch.interrupt.injected) {
7748 kvm_x86_ops.set_irq(vcpu);
7753 WARN_ON_ONCE(vcpu->arch.exception.injected &&
7754 vcpu->arch.exception.pending);
7757 * Call check_nested_events() even if we reinjected a previous event
7758 * in order for caller to determine if it should require immediate-exit
7759 * from L2 to L1 due to pending L1 events which require exit
7762 if (is_guest_mode(vcpu)) {
7763 r = kvm_x86_ops.nested_ops->check_events(vcpu);
7768 /* try to inject new event if pending */
7769 if (vcpu->arch.exception.pending) {
7770 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7771 vcpu->arch.exception.has_error_code,
7772 vcpu->arch.exception.error_code);
7774 vcpu->arch.exception.pending = false;
7775 vcpu->arch.exception.injected = true;
7777 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7778 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7781 if (vcpu->arch.exception.nr == DB_VECTOR) {
7782 kvm_deliver_exception_payload(vcpu);
7783 if (vcpu->arch.dr7 & DR7_GD) {
7784 vcpu->arch.dr7 &= ~DR7_GD;
7785 kvm_update_dr7(vcpu);
7789 kvm_x86_ops.queue_exception(vcpu);
7794 * Finally, inject interrupt events. If an event cannot be injected
7795 * due to architectural conditions (e.g. IF=0) a window-open exit
7796 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
7797 * and can architecturally be injected, but we cannot do it right now:
7798 * an interrupt could have arrived just now and we have to inject it
7799 * as a vmexit, or there could already an event in the queue, which is
7800 * indicated by can_inject. In that case we request an immediate exit
7801 * in order to make progress and get back here for another iteration.
7802 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
7804 if (vcpu->arch.smi_pending) {
7805 r = can_inject ? kvm_x86_ops.smi_allowed(vcpu, true) : -EBUSY;
7809 vcpu->arch.smi_pending = false;
7810 ++vcpu->arch.smi_count;
7814 kvm_x86_ops.enable_smi_window(vcpu);
7817 if (vcpu->arch.nmi_pending) {
7818 r = can_inject ? kvm_x86_ops.nmi_allowed(vcpu, true) : -EBUSY;
7822 --vcpu->arch.nmi_pending;
7823 vcpu->arch.nmi_injected = true;
7824 kvm_x86_ops.set_nmi(vcpu);
7826 WARN_ON(kvm_x86_ops.nmi_allowed(vcpu, true) < 0);
7828 if (vcpu->arch.nmi_pending)
7829 kvm_x86_ops.enable_nmi_window(vcpu);
7832 if (kvm_cpu_has_injectable_intr(vcpu)) {
7833 r = can_inject ? kvm_x86_ops.interrupt_allowed(vcpu, true) : -EBUSY;
7837 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
7838 kvm_x86_ops.set_irq(vcpu);
7839 WARN_ON(kvm_x86_ops.interrupt_allowed(vcpu, true) < 0);
7841 if (kvm_cpu_has_injectable_intr(vcpu))
7842 kvm_x86_ops.enable_irq_window(vcpu);
7845 if (is_guest_mode(vcpu) &&
7846 kvm_x86_ops.nested_ops->hv_timer_pending &&
7847 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
7848 *req_immediate_exit = true;
7850 WARN_ON(vcpu->arch.exception.pending);
7854 *req_immediate_exit = true;
7858 static void process_nmi(struct kvm_vcpu *vcpu)
7863 * x86 is limited to one NMI running, and one NMI pending after it.
7864 * If an NMI is already in progress, limit further NMIs to just one.
7865 * Otherwise, allow two (and we'll inject the first one immediately).
7867 if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7870 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7871 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7872 kvm_make_request(KVM_REQ_EVENT, vcpu);
7875 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7878 flags |= seg->g << 23;
7879 flags |= seg->db << 22;
7880 flags |= seg->l << 21;
7881 flags |= seg->avl << 20;
7882 flags |= seg->present << 15;
7883 flags |= seg->dpl << 13;
7884 flags |= seg->s << 12;
7885 flags |= seg->type << 8;
7889 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7891 struct kvm_segment seg;
7894 kvm_get_segment(vcpu, &seg, n);
7895 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7898 offset = 0x7f84 + n * 12;
7900 offset = 0x7f2c + (n - 3) * 12;
7902 put_smstate(u32, buf, offset + 8, seg.base);
7903 put_smstate(u32, buf, offset + 4, seg.limit);
7904 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7907 #ifdef CONFIG_X86_64
7908 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7910 struct kvm_segment seg;
7914 kvm_get_segment(vcpu, &seg, n);
7915 offset = 0x7e00 + n * 16;
7917 flags = enter_smm_get_segment_flags(&seg) >> 8;
7918 put_smstate(u16, buf, offset, seg.selector);
7919 put_smstate(u16, buf, offset + 2, flags);
7920 put_smstate(u32, buf, offset + 4, seg.limit);
7921 put_smstate(u64, buf, offset + 8, seg.base);
7925 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7928 struct kvm_segment seg;
7932 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7933 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7934 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7935 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7937 for (i = 0; i < 8; i++)
7938 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7940 kvm_get_dr(vcpu, 6, &val);
7941 put_smstate(u32, buf, 0x7fcc, (u32)val);
7942 kvm_get_dr(vcpu, 7, &val);
7943 put_smstate(u32, buf, 0x7fc8, (u32)val);
7945 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7946 put_smstate(u32, buf, 0x7fc4, seg.selector);
7947 put_smstate(u32, buf, 0x7f64, seg.base);
7948 put_smstate(u32, buf, 0x7f60, seg.limit);
7949 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7951 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7952 put_smstate(u32, buf, 0x7fc0, seg.selector);
7953 put_smstate(u32, buf, 0x7f80, seg.base);
7954 put_smstate(u32, buf, 0x7f7c, seg.limit);
7955 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7957 kvm_x86_ops.get_gdt(vcpu, &dt);
7958 put_smstate(u32, buf, 0x7f74, dt.address);
7959 put_smstate(u32, buf, 0x7f70, dt.size);
7961 kvm_x86_ops.get_idt(vcpu, &dt);
7962 put_smstate(u32, buf, 0x7f58, dt.address);
7963 put_smstate(u32, buf, 0x7f54, dt.size);
7965 for (i = 0; i < 6; i++)
7966 enter_smm_save_seg_32(vcpu, buf, i);
7968 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7971 put_smstate(u32, buf, 0x7efc, 0x00020000);
7972 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7975 #ifdef CONFIG_X86_64
7976 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7979 struct kvm_segment seg;
7983 for (i = 0; i < 16; i++)
7984 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7986 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7987 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7989 kvm_get_dr(vcpu, 6, &val);
7990 put_smstate(u64, buf, 0x7f68, val);
7991 kvm_get_dr(vcpu, 7, &val);
7992 put_smstate(u64, buf, 0x7f60, val);
7994 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7995 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7996 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7998 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8001 put_smstate(u32, buf, 0x7efc, 0x00020064);
8003 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8005 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8006 put_smstate(u16, buf, 0x7e90, seg.selector);
8007 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8008 put_smstate(u32, buf, 0x7e94, seg.limit);
8009 put_smstate(u64, buf, 0x7e98, seg.base);
8011 kvm_x86_ops.get_idt(vcpu, &dt);
8012 put_smstate(u32, buf, 0x7e84, dt.size);
8013 put_smstate(u64, buf, 0x7e88, dt.address);
8015 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8016 put_smstate(u16, buf, 0x7e70, seg.selector);
8017 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8018 put_smstate(u32, buf, 0x7e74, seg.limit);
8019 put_smstate(u64, buf, 0x7e78, seg.base);
8021 kvm_x86_ops.get_gdt(vcpu, &dt);
8022 put_smstate(u32, buf, 0x7e64, dt.size);
8023 put_smstate(u64, buf, 0x7e68, dt.address);
8025 for (i = 0; i < 6; i++)
8026 enter_smm_save_seg_64(vcpu, buf, i);
8030 static void enter_smm(struct kvm_vcpu *vcpu)
8032 struct kvm_segment cs, ds;
8037 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8038 memset(buf, 0, 512);
8039 #ifdef CONFIG_X86_64
8040 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8041 enter_smm_save_state_64(vcpu, buf);
8044 enter_smm_save_state_32(vcpu, buf);
8047 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8048 * vCPU state (e.g. leave guest mode) after we've saved the state into
8049 * the SMM state-save area.
8051 kvm_x86_ops.pre_enter_smm(vcpu, buf);
8053 vcpu->arch.hflags |= HF_SMM_MASK;
8054 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8056 if (kvm_x86_ops.get_nmi_mask(vcpu))
8057 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8059 kvm_x86_ops.set_nmi_mask(vcpu, true);
8061 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8062 kvm_rip_write(vcpu, 0x8000);
8064 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8065 kvm_x86_ops.set_cr0(vcpu, cr0);
8066 vcpu->arch.cr0 = cr0;
8068 kvm_x86_ops.set_cr4(vcpu, 0);
8070 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8071 dt.address = dt.size = 0;
8072 kvm_x86_ops.set_idt(vcpu, &dt);
8074 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8076 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8077 cs.base = vcpu->arch.smbase;
8082 cs.limit = ds.limit = 0xffffffff;
8083 cs.type = ds.type = 0x3;
8084 cs.dpl = ds.dpl = 0;
8089 cs.avl = ds.avl = 0;
8090 cs.present = ds.present = 1;
8091 cs.unusable = ds.unusable = 0;
8092 cs.padding = ds.padding = 0;
8094 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8095 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8096 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8097 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8098 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8099 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8101 #ifdef CONFIG_X86_64
8102 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8103 kvm_x86_ops.set_efer(vcpu, 0);
8106 kvm_update_cpuid(vcpu);
8107 kvm_mmu_reset_context(vcpu);
8110 static void process_smi(struct kvm_vcpu *vcpu)
8112 vcpu->arch.smi_pending = true;
8113 kvm_make_request(KVM_REQ_EVENT, vcpu);
8116 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8117 unsigned long *vcpu_bitmap)
8121 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8123 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8124 NULL, vcpu_bitmap, cpus);
8126 free_cpumask_var(cpus);
8129 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8131 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8134 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8136 if (!lapic_in_kernel(vcpu))
8139 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8140 kvm_apic_update_apicv(vcpu);
8141 kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8143 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8146 * NOTE: Do not hold any lock prior to calling this.
8148 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8149 * locked, because it calls __x86_set_memory_region() which does
8150 * synchronize_srcu(&kvm->srcu).
8152 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8154 struct kvm_vcpu *except;
8155 unsigned long old, new, expected;
8157 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8158 !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8161 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8163 expected = new = old;
8165 __clear_bit(bit, &new);
8167 __set_bit(bit, &new);
8170 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8171 } while (old != expected);
8176 trace_kvm_apicv_update_request(activate, bit);
8177 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8178 kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8181 * Sending request to update APICV for all other vcpus,
8182 * while update the calling vcpu immediately instead of
8183 * waiting for another #VMEXIT to handle the request.
8185 except = kvm_get_running_vcpu();
8186 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8189 kvm_vcpu_update_apicv(except);
8191 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8193 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8195 if (!kvm_apic_present(vcpu))
8198 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8200 if (irqchip_split(vcpu->kvm))
8201 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8203 if (vcpu->arch.apicv_active)
8204 kvm_x86_ops.sync_pir_to_irr(vcpu);
8205 if (ioapic_in_kernel(vcpu->kvm))
8206 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8209 if (is_guest_mode(vcpu))
8210 vcpu->arch.load_eoi_exitmap_pending = true;
8212 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8215 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8217 u64 eoi_exit_bitmap[4];
8219 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8222 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8223 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8224 kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8227 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8228 unsigned long start, unsigned long end,
8231 unsigned long apic_address;
8234 * The physical address of apic access page is stored in the VMCS.
8235 * Update it when it becomes invalid.
8237 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8238 if (start <= apic_address && apic_address < end)
8239 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8244 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8246 if (!lapic_in_kernel(vcpu))
8249 if (!kvm_x86_ops.set_apic_access_page_addr)
8252 kvm_x86_ops.set_apic_access_page_addr(vcpu);
8255 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8257 smp_send_reschedule(vcpu->cpu);
8259 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8262 * Returns 1 to let vcpu_run() continue the guest execution loop without
8263 * exiting to the userspace. Otherwise, the value will be returned to the
8266 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8270 dm_request_for_irq_injection(vcpu) &&
8271 kvm_cpu_accept_dm_intr(vcpu);
8272 fastpath_t exit_fastpath;
8274 bool req_immediate_exit = false;
8276 if (kvm_request_pending(vcpu)) {
8277 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8278 if (unlikely(!kvm_x86_ops.nested_ops->get_vmcs12_pages(vcpu))) {
8283 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8284 kvm_mmu_unload(vcpu);
8285 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8286 __kvm_migrate_timers(vcpu);
8287 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8288 kvm_gen_update_masterclock(vcpu->kvm);
8289 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8290 kvm_gen_kvmclock_update(vcpu);
8291 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8292 r = kvm_guest_time_update(vcpu);
8296 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8297 kvm_mmu_sync_roots(vcpu);
8298 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8299 kvm_mmu_load_pgd(vcpu);
8300 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8301 kvm_vcpu_flush_tlb_all(vcpu);
8303 /* Flushing all ASIDs flushes the current ASID... */
8304 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8306 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8307 kvm_vcpu_flush_tlb_current(vcpu);
8308 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8309 kvm_vcpu_flush_tlb_guest(vcpu);
8311 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8312 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8316 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8317 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8318 vcpu->mmio_needed = 0;
8322 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8323 /* Page is swapped out. Do synthetic halt */
8324 vcpu->arch.apf.halted = true;
8328 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8329 record_steal_time(vcpu);
8330 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8332 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8334 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8335 kvm_pmu_handle_event(vcpu);
8336 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8337 kvm_pmu_deliver_pmi(vcpu);
8338 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8339 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8340 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8341 vcpu->arch.ioapic_handled_vectors)) {
8342 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8343 vcpu->run->eoi.vector =
8344 vcpu->arch.pending_ioapic_eoi;
8349 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8350 vcpu_scan_ioapic(vcpu);
8351 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8352 vcpu_load_eoi_exitmap(vcpu);
8353 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8354 kvm_vcpu_reload_apic_access_page(vcpu);
8355 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8356 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8357 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8361 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8362 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8363 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8367 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8368 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8369 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8375 * KVM_REQ_HV_STIMER has to be processed after
8376 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8377 * depend on the guest clock being up-to-date
8379 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8380 kvm_hv_process_stimers(vcpu);
8381 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8382 kvm_vcpu_update_apicv(vcpu);
8385 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8386 ++vcpu->stat.req_event;
8387 kvm_apic_accept_events(vcpu);
8388 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8393 inject_pending_event(vcpu, &req_immediate_exit);
8395 kvm_x86_ops.enable_irq_window(vcpu);
8397 if (kvm_lapic_enabled(vcpu)) {
8398 update_cr8_intercept(vcpu);
8399 kvm_lapic_sync_to_vapic(vcpu);
8403 r = kvm_mmu_reload(vcpu);
8405 goto cancel_injection;
8410 kvm_x86_ops.prepare_guest_switch(vcpu);
8413 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8414 * IPI are then delayed after guest entry, which ensures that they
8415 * result in virtual interrupt delivery.
8417 local_irq_disable();
8418 vcpu->mode = IN_GUEST_MODE;
8420 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8423 * 1) We should set ->mode before checking ->requests. Please see
8424 * the comment in kvm_vcpu_exiting_guest_mode().
8426 * 2) For APICv, we should set ->mode before checking PID.ON. This
8427 * pairs with the memory barrier implicit in pi_test_and_set_on
8428 * (see vmx_deliver_posted_interrupt).
8430 * 3) This also orders the write to mode from any reads to the page
8431 * tables done while the VCPU is running. Please see the comment
8432 * in kvm_flush_remote_tlbs.
8434 smp_mb__after_srcu_read_unlock();
8437 * This handles the case where a posted interrupt was
8438 * notified with kvm_vcpu_kick.
8440 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8441 kvm_x86_ops.sync_pir_to_irr(vcpu);
8443 if (kvm_vcpu_exit_request(vcpu)) {
8444 vcpu->mode = OUTSIDE_GUEST_MODE;
8448 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8450 goto cancel_injection;
8453 if (req_immediate_exit) {
8454 kvm_make_request(KVM_REQ_EVENT, vcpu);
8455 kvm_x86_ops.request_immediate_exit(vcpu);
8458 trace_kvm_entry(vcpu->vcpu_id);
8459 guest_enter_irqoff();
8461 fpregs_assert_state_consistent();
8462 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8463 switch_fpu_return();
8465 if (unlikely(vcpu->arch.switch_db_regs)) {
8467 set_debugreg(vcpu->arch.eff_db[0], 0);
8468 set_debugreg(vcpu->arch.eff_db[1], 1);
8469 set_debugreg(vcpu->arch.eff_db[2], 2);
8470 set_debugreg(vcpu->arch.eff_db[3], 3);
8471 set_debugreg(vcpu->arch.dr6, 6);
8472 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8475 exit_fastpath = kvm_x86_ops.run(vcpu);
8478 * Do this here before restoring debug registers on the host. And
8479 * since we do this before handling the vmexit, a DR access vmexit
8480 * can (a) read the correct value of the debug registers, (b) set
8481 * KVM_DEBUGREG_WONT_EXIT again.
8483 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8484 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8485 kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8486 kvm_update_dr0123(vcpu);
8487 kvm_update_dr7(vcpu);
8488 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8492 * If the guest has used debug registers, at least dr7
8493 * will be disabled while returning to the host.
8494 * If we don't have active breakpoints in the host, we don't
8495 * care about the messed up debug address registers. But if
8496 * we have some of them active, restore the old state.
8498 if (hw_breakpoint_active())
8499 hw_breakpoint_restore();
8501 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8503 vcpu->mode = OUTSIDE_GUEST_MODE;
8506 kvm_x86_ops.handle_exit_irqoff(vcpu);
8509 * Consume any pending interrupts, including the possible source of
8510 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8511 * An instruction is required after local_irq_enable() to fully unblock
8512 * interrupts on processors that implement an interrupt shadow, the
8513 * stat.exits increment will do nicely.
8515 kvm_before_interrupt(vcpu);
8518 local_irq_disable();
8519 kvm_after_interrupt(vcpu);
8521 guest_exit_irqoff();
8522 if (lapic_in_kernel(vcpu)) {
8523 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8524 if (delta != S64_MIN) {
8525 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8526 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8533 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8536 * Profile KVM exit RIPs:
8538 if (unlikely(prof_on == KVM_PROFILING)) {
8539 unsigned long rip = kvm_rip_read(vcpu);
8540 profile_hit(KVM_PROFILING, (void *)rip);
8543 if (unlikely(vcpu->arch.tsc_always_catchup))
8544 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8546 if (vcpu->arch.apic_attention)
8547 kvm_lapic_sync_from_vapic(vcpu);
8549 r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
8553 if (req_immediate_exit)
8554 kvm_make_request(KVM_REQ_EVENT, vcpu);
8555 kvm_x86_ops.cancel_injection(vcpu);
8556 if (unlikely(vcpu->arch.apic_attention))
8557 kvm_lapic_sync_from_vapic(vcpu);
8562 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8564 if (!kvm_arch_vcpu_runnable(vcpu) &&
8565 (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
8566 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8567 kvm_vcpu_block(vcpu);
8568 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8570 if (kvm_x86_ops.post_block)
8571 kvm_x86_ops.post_block(vcpu);
8573 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8577 kvm_apic_accept_events(vcpu);
8578 switch(vcpu->arch.mp_state) {
8579 case KVM_MP_STATE_HALTED:
8580 vcpu->arch.pv.pv_unhalted = false;
8581 vcpu->arch.mp_state =
8582 KVM_MP_STATE_RUNNABLE;
8584 case KVM_MP_STATE_RUNNABLE:
8585 vcpu->arch.apf.halted = false;
8587 case KVM_MP_STATE_INIT_RECEIVED:
8595 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8597 if (is_guest_mode(vcpu))
8598 kvm_x86_ops.nested_ops->check_events(vcpu);
8600 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8601 !vcpu->arch.apf.halted);
8604 static int vcpu_run(struct kvm_vcpu *vcpu)
8607 struct kvm *kvm = vcpu->kvm;
8609 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8610 vcpu->arch.l1tf_flush_l1d = true;
8613 if (kvm_vcpu_running(vcpu)) {
8614 r = vcpu_enter_guest(vcpu);
8616 r = vcpu_block(kvm, vcpu);
8622 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8623 if (kvm_cpu_has_pending_timer(vcpu))
8624 kvm_inject_pending_timer_irqs(vcpu);
8626 if (dm_request_for_irq_injection(vcpu) &&
8627 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8629 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8630 ++vcpu->stat.request_irq_exits;
8634 kvm_check_async_pf_completion(vcpu);
8636 if (signal_pending(current)) {
8638 vcpu->run->exit_reason = KVM_EXIT_INTR;
8639 ++vcpu->stat.signal_exits;
8642 if (need_resched()) {
8643 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8645 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8649 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8654 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8658 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8659 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8660 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8664 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8666 BUG_ON(!vcpu->arch.pio.count);
8668 return complete_emulated_io(vcpu);
8672 * Implements the following, as a state machine:
8676 * for each mmio piece in the fragment
8684 * for each mmio piece in the fragment
8689 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8691 struct kvm_run *run = vcpu->run;
8692 struct kvm_mmio_fragment *frag;
8695 BUG_ON(!vcpu->mmio_needed);
8697 /* Complete previous fragment */
8698 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8699 len = min(8u, frag->len);
8700 if (!vcpu->mmio_is_write)
8701 memcpy(frag->data, run->mmio.data, len);
8703 if (frag->len <= 8) {
8704 /* Switch to the next fragment. */
8706 vcpu->mmio_cur_fragment++;
8708 /* Go forward to the next mmio piece. */
8714 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8715 vcpu->mmio_needed = 0;
8717 /* FIXME: return into emulator if single-stepping. */
8718 if (vcpu->mmio_is_write)
8720 vcpu->mmio_read_completed = 1;
8721 return complete_emulated_io(vcpu);
8724 run->exit_reason = KVM_EXIT_MMIO;
8725 run->mmio.phys_addr = frag->gpa;
8726 if (vcpu->mmio_is_write)
8727 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8728 run->mmio.len = min(8u, frag->len);
8729 run->mmio.is_write = vcpu->mmio_is_write;
8730 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8734 static void kvm_save_current_fpu(struct fpu *fpu)
8737 * If the target FPU state is not resident in the CPU registers, just
8738 * memcpy() from current, else save CPU state directly to the target.
8740 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8741 memcpy(&fpu->state, ¤t->thread.fpu.state,
8742 fpu_kernel_xstate_size);
8744 copy_fpregs_to_fpstate(fpu);
8747 /* Swap (qemu) user FPU context for the guest FPU context. */
8748 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8752 kvm_save_current_fpu(vcpu->arch.user_fpu);
8754 /* PKRU is separately restored in kvm_x86_ops.run. */
8755 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8756 ~XFEATURE_MASK_PKRU);
8758 fpregs_mark_activate();
8764 /* When vcpu_run ends, restore user space FPU context. */
8765 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8769 kvm_save_current_fpu(vcpu->arch.guest_fpu);
8771 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8773 fpregs_mark_activate();
8776 ++vcpu->stat.fpu_reload;
8780 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
8782 struct kvm_run *kvm_run = vcpu->run;
8786 kvm_sigset_activate(vcpu);
8787 kvm_load_guest_fpu(vcpu);
8789 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8790 if (kvm_run->immediate_exit) {
8794 kvm_vcpu_block(vcpu);
8795 kvm_apic_accept_events(vcpu);
8796 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8798 if (signal_pending(current)) {
8800 kvm_run->exit_reason = KVM_EXIT_INTR;
8801 ++vcpu->stat.signal_exits;
8806 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8811 if (kvm_run->kvm_dirty_regs) {
8812 r = sync_regs(vcpu);
8817 /* re-sync apic's tpr */
8818 if (!lapic_in_kernel(vcpu)) {
8819 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8825 if (unlikely(vcpu->arch.complete_userspace_io)) {
8826 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8827 vcpu->arch.complete_userspace_io = NULL;
8832 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8834 if (kvm_run->immediate_exit)
8840 kvm_put_guest_fpu(vcpu);
8841 if (kvm_run->kvm_valid_regs)
8843 post_kvm_run_save(vcpu);
8844 kvm_sigset_deactivate(vcpu);
8850 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8852 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8854 * We are here if userspace calls get_regs() in the middle of
8855 * instruction emulation. Registers state needs to be copied
8856 * back from emulation context to vcpu. Userspace shouldn't do
8857 * that usually, but some bad designed PV devices (vmware
8858 * backdoor interface) need this to work
8860 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
8861 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8863 regs->rax = kvm_rax_read(vcpu);
8864 regs->rbx = kvm_rbx_read(vcpu);
8865 regs->rcx = kvm_rcx_read(vcpu);
8866 regs->rdx = kvm_rdx_read(vcpu);
8867 regs->rsi = kvm_rsi_read(vcpu);
8868 regs->rdi = kvm_rdi_read(vcpu);
8869 regs->rsp = kvm_rsp_read(vcpu);
8870 regs->rbp = kvm_rbp_read(vcpu);
8871 #ifdef CONFIG_X86_64
8872 regs->r8 = kvm_r8_read(vcpu);
8873 regs->r9 = kvm_r9_read(vcpu);
8874 regs->r10 = kvm_r10_read(vcpu);
8875 regs->r11 = kvm_r11_read(vcpu);
8876 regs->r12 = kvm_r12_read(vcpu);
8877 regs->r13 = kvm_r13_read(vcpu);
8878 regs->r14 = kvm_r14_read(vcpu);
8879 regs->r15 = kvm_r15_read(vcpu);
8882 regs->rip = kvm_rip_read(vcpu);
8883 regs->rflags = kvm_get_rflags(vcpu);
8886 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8889 __get_regs(vcpu, regs);
8894 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8896 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8897 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8899 kvm_rax_write(vcpu, regs->rax);
8900 kvm_rbx_write(vcpu, regs->rbx);
8901 kvm_rcx_write(vcpu, regs->rcx);
8902 kvm_rdx_write(vcpu, regs->rdx);
8903 kvm_rsi_write(vcpu, regs->rsi);
8904 kvm_rdi_write(vcpu, regs->rdi);
8905 kvm_rsp_write(vcpu, regs->rsp);
8906 kvm_rbp_write(vcpu, regs->rbp);
8907 #ifdef CONFIG_X86_64
8908 kvm_r8_write(vcpu, regs->r8);
8909 kvm_r9_write(vcpu, regs->r9);
8910 kvm_r10_write(vcpu, regs->r10);
8911 kvm_r11_write(vcpu, regs->r11);
8912 kvm_r12_write(vcpu, regs->r12);
8913 kvm_r13_write(vcpu, regs->r13);
8914 kvm_r14_write(vcpu, regs->r14);
8915 kvm_r15_write(vcpu, regs->r15);
8918 kvm_rip_write(vcpu, regs->rip);
8919 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8921 vcpu->arch.exception.pending = false;
8923 kvm_make_request(KVM_REQ_EVENT, vcpu);
8926 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8929 __set_regs(vcpu, regs);
8934 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8936 struct kvm_segment cs;
8938 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8942 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8944 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8948 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8949 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8950 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8951 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8952 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8953 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8955 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8956 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8958 kvm_x86_ops.get_idt(vcpu, &dt);
8959 sregs->idt.limit = dt.size;
8960 sregs->idt.base = dt.address;
8961 kvm_x86_ops.get_gdt(vcpu, &dt);
8962 sregs->gdt.limit = dt.size;
8963 sregs->gdt.base = dt.address;
8965 sregs->cr0 = kvm_read_cr0(vcpu);
8966 sregs->cr2 = vcpu->arch.cr2;
8967 sregs->cr3 = kvm_read_cr3(vcpu);
8968 sregs->cr4 = kvm_read_cr4(vcpu);
8969 sregs->cr8 = kvm_get_cr8(vcpu);
8970 sregs->efer = vcpu->arch.efer;
8971 sregs->apic_base = kvm_get_apic_base(vcpu);
8973 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8975 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8976 set_bit(vcpu->arch.interrupt.nr,
8977 (unsigned long *)sregs->interrupt_bitmap);
8980 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8981 struct kvm_sregs *sregs)
8984 __get_sregs(vcpu, sregs);
8989 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8990 struct kvm_mp_state *mp_state)
8993 if (kvm_mpx_supported())
8994 kvm_load_guest_fpu(vcpu);
8996 kvm_apic_accept_events(vcpu);
8997 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8998 vcpu->arch.pv.pv_unhalted)
8999 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9001 mp_state->mp_state = vcpu->arch.mp_state;
9003 if (kvm_mpx_supported())
9004 kvm_put_guest_fpu(vcpu);
9009 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9010 struct kvm_mp_state *mp_state)
9016 if (!lapic_in_kernel(vcpu) &&
9017 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9021 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9022 * INIT state; latched init should be reported using
9023 * KVM_SET_VCPU_EVENTS, so reject it here.
9025 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9026 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9027 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9030 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9031 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9032 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9034 vcpu->arch.mp_state = mp_state->mp_state;
9035 kvm_make_request(KVM_REQ_EVENT, vcpu);
9043 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9044 int reason, bool has_error_code, u32 error_code)
9046 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9049 init_emulate_ctxt(vcpu);
9051 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9052 has_error_code, error_code);
9054 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9055 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9056 vcpu->run->internal.ndata = 0;
9060 kvm_rip_write(vcpu, ctxt->eip);
9061 kvm_set_rflags(vcpu, ctxt->eflags);
9064 EXPORT_SYMBOL_GPL(kvm_task_switch);
9066 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9068 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9070 * When EFER.LME and CR0.PG are set, the processor is in
9071 * 64-bit mode (though maybe in a 32-bit code segment).
9072 * CR4.PAE and EFER.LMA must be set.
9074 if (!(sregs->cr4 & X86_CR4_PAE)
9075 || !(sregs->efer & EFER_LMA))
9079 * Not in 64-bit mode: EFER.LMA is clear and the code
9080 * segment cannot be 64-bit.
9082 if (sregs->efer & EFER_LMA || sregs->cs.l)
9086 return kvm_valid_cr4(vcpu, sregs->cr4);
9089 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9091 struct msr_data apic_base_msr;
9092 int mmu_reset_needed = 0;
9093 int cpuid_update_needed = 0;
9094 int pending_vec, max_bits, idx;
9098 if (kvm_valid_sregs(vcpu, sregs))
9101 apic_base_msr.data = sregs->apic_base;
9102 apic_base_msr.host_initiated = true;
9103 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9106 dt.size = sregs->idt.limit;
9107 dt.address = sregs->idt.base;
9108 kvm_x86_ops.set_idt(vcpu, &dt);
9109 dt.size = sregs->gdt.limit;
9110 dt.address = sregs->gdt.base;
9111 kvm_x86_ops.set_gdt(vcpu, &dt);
9113 vcpu->arch.cr2 = sregs->cr2;
9114 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9115 vcpu->arch.cr3 = sregs->cr3;
9116 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9118 kvm_set_cr8(vcpu, sregs->cr8);
9120 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9121 kvm_x86_ops.set_efer(vcpu, sregs->efer);
9123 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9124 kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9125 vcpu->arch.cr0 = sregs->cr0;
9127 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9128 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9129 (X86_CR4_OSXSAVE | X86_CR4_PKE));
9130 kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9131 if (cpuid_update_needed)
9132 kvm_update_cpuid(vcpu);
9134 idx = srcu_read_lock(&vcpu->kvm->srcu);
9135 if (is_pae_paging(vcpu)) {
9136 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9137 mmu_reset_needed = 1;
9139 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9141 if (mmu_reset_needed)
9142 kvm_mmu_reset_context(vcpu);
9144 max_bits = KVM_NR_INTERRUPTS;
9145 pending_vec = find_first_bit(
9146 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9147 if (pending_vec < max_bits) {
9148 kvm_queue_interrupt(vcpu, pending_vec, false);
9149 pr_debug("Set back pending irq %d\n", pending_vec);
9152 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9153 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9154 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9155 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9156 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9157 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9159 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9160 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9162 update_cr8_intercept(vcpu);
9164 /* Older userspace won't unhalt the vcpu on reset. */
9165 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9166 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9168 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9170 kvm_make_request(KVM_REQ_EVENT, vcpu);
9177 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9178 struct kvm_sregs *sregs)
9183 ret = __set_sregs(vcpu, sregs);
9188 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9189 struct kvm_guest_debug *dbg)
9191 unsigned long rflags;
9196 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9198 if (vcpu->arch.exception.pending)
9200 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9201 kvm_queue_exception(vcpu, DB_VECTOR);
9203 kvm_queue_exception(vcpu, BP_VECTOR);
9207 * Read rflags as long as potentially injected trace flags are still
9210 rflags = kvm_get_rflags(vcpu);
9212 vcpu->guest_debug = dbg->control;
9213 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9214 vcpu->guest_debug = 0;
9216 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9217 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9218 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9219 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9221 for (i = 0; i < KVM_NR_DB_REGS; i++)
9222 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9224 kvm_update_dr7(vcpu);
9226 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9227 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9228 get_segment_base(vcpu, VCPU_SREG_CS);
9231 * Trigger an rflags update that will inject or remove the trace
9234 kvm_set_rflags(vcpu, rflags);
9236 kvm_x86_ops.update_bp_intercept(vcpu);
9246 * Translate a guest virtual address to a guest physical address.
9248 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9249 struct kvm_translation *tr)
9251 unsigned long vaddr = tr->linear_address;
9257 idx = srcu_read_lock(&vcpu->kvm->srcu);
9258 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9259 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9260 tr->physical_address = gpa;
9261 tr->valid = gpa != UNMAPPED_GVA;
9269 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9271 struct fxregs_state *fxsave;
9275 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9276 memcpy(fpu->fpr, fxsave->st_space, 128);
9277 fpu->fcw = fxsave->cwd;
9278 fpu->fsw = fxsave->swd;
9279 fpu->ftwx = fxsave->twd;
9280 fpu->last_opcode = fxsave->fop;
9281 fpu->last_ip = fxsave->rip;
9282 fpu->last_dp = fxsave->rdp;
9283 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9289 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9291 struct fxregs_state *fxsave;
9295 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9297 memcpy(fxsave->st_space, fpu->fpr, 128);
9298 fxsave->cwd = fpu->fcw;
9299 fxsave->swd = fpu->fsw;
9300 fxsave->twd = fpu->ftwx;
9301 fxsave->fop = fpu->last_opcode;
9302 fxsave->rip = fpu->last_ip;
9303 fxsave->rdp = fpu->last_dp;
9304 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9310 static void store_regs(struct kvm_vcpu *vcpu)
9312 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9314 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9315 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9317 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9318 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9320 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9321 kvm_vcpu_ioctl_x86_get_vcpu_events(
9322 vcpu, &vcpu->run->s.regs.events);
9325 static int sync_regs(struct kvm_vcpu *vcpu)
9327 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9330 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9331 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9332 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9334 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9335 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9337 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9339 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9340 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9341 vcpu, &vcpu->run->s.regs.events))
9343 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9349 static void fx_init(struct kvm_vcpu *vcpu)
9351 fpstate_init(&vcpu->arch.guest_fpu->state);
9352 if (boot_cpu_has(X86_FEATURE_XSAVES))
9353 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9354 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9357 * Ensure guest xcr0 is valid for loading
9359 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9361 vcpu->arch.cr0 |= X86_CR0_ET;
9364 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9366 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9367 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9368 "guest TSC will not be reliable\n");
9373 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9378 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9379 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9381 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9383 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9385 r = kvm_mmu_create(vcpu);
9389 if (irqchip_in_kernel(vcpu->kvm)) {
9390 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9392 goto fail_mmu_destroy;
9393 if (kvm_apicv_activated(vcpu->kvm))
9394 vcpu->arch.apicv_active = true;
9396 static_key_slow_inc(&kvm_no_apic_vcpu);
9400 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9402 goto fail_free_lapic;
9403 vcpu->arch.pio_data = page_address(page);
9405 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9406 GFP_KERNEL_ACCOUNT);
9407 if (!vcpu->arch.mce_banks)
9408 goto fail_free_pio_data;
9409 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9411 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9412 GFP_KERNEL_ACCOUNT))
9413 goto fail_free_mce_banks;
9415 if (!alloc_emulate_ctxt(vcpu))
9416 goto free_wbinvd_dirty_mask;
9418 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9419 GFP_KERNEL_ACCOUNT);
9420 if (!vcpu->arch.user_fpu) {
9421 pr_err("kvm: failed to allocate userspace's fpu\n");
9422 goto free_emulate_ctxt;
9425 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9426 GFP_KERNEL_ACCOUNT);
9427 if (!vcpu->arch.guest_fpu) {
9428 pr_err("kvm: failed to allocate vcpu's fpu\n");
9433 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9434 vcpu->arch.tdp_level = kvm_x86_ops.get_tdp_level(vcpu);
9436 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9438 kvm_async_pf_hash_reset(vcpu);
9441 vcpu->arch.pending_external_vector = -1;
9442 vcpu->arch.preempted_in_kernel = false;
9444 kvm_hv_vcpu_init(vcpu);
9446 r = kvm_x86_ops.vcpu_create(vcpu);
9448 goto free_guest_fpu;
9450 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9451 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9452 kvm_vcpu_mtrr_init(vcpu);
9454 kvm_vcpu_reset(vcpu, false);
9455 kvm_init_mmu(vcpu, false);
9460 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9462 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9464 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9465 free_wbinvd_dirty_mask:
9466 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9467 fail_free_mce_banks:
9468 kfree(vcpu->arch.mce_banks);
9470 free_page((unsigned long)vcpu->arch.pio_data);
9472 kvm_free_lapic(vcpu);
9474 kvm_mmu_destroy(vcpu);
9478 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9480 struct msr_data msr;
9481 struct kvm *kvm = vcpu->kvm;
9483 kvm_hv_vcpu_postcreate(vcpu);
9485 if (mutex_lock_killable(&vcpu->mutex))
9489 msr.index = MSR_IA32_TSC;
9490 msr.host_initiated = true;
9491 kvm_write_tsc(vcpu, &msr);
9494 /* poll control enabled by default */
9495 vcpu->arch.msr_kvm_poll_control = 1;
9497 mutex_unlock(&vcpu->mutex);
9499 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9500 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9501 KVMCLOCK_SYNC_PERIOD);
9504 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9506 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9509 kvm_release_pfn(cache->pfn, cache->dirty, cache);
9511 kvmclock_reset(vcpu);
9513 kvm_x86_ops.vcpu_free(vcpu);
9515 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9516 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9517 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9518 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9520 kvm_hv_vcpu_uninit(vcpu);
9521 kvm_pmu_destroy(vcpu);
9522 kfree(vcpu->arch.mce_banks);
9523 kvm_free_lapic(vcpu);
9524 idx = srcu_read_lock(&vcpu->kvm->srcu);
9525 kvm_mmu_destroy(vcpu);
9526 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9527 free_page((unsigned long)vcpu->arch.pio_data);
9528 if (!lapic_in_kernel(vcpu))
9529 static_key_slow_dec(&kvm_no_apic_vcpu);
9532 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9534 kvm_lapic_reset(vcpu, init_event);
9536 vcpu->arch.hflags = 0;
9538 vcpu->arch.smi_pending = 0;
9539 vcpu->arch.smi_count = 0;
9540 atomic_set(&vcpu->arch.nmi_queued, 0);
9541 vcpu->arch.nmi_pending = 0;
9542 vcpu->arch.nmi_injected = false;
9543 kvm_clear_interrupt_queue(vcpu);
9544 kvm_clear_exception_queue(vcpu);
9546 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9547 kvm_update_dr0123(vcpu);
9548 vcpu->arch.dr6 = DR6_INIT;
9549 vcpu->arch.dr7 = DR7_FIXED_1;
9550 kvm_update_dr7(vcpu);
9554 kvm_make_request(KVM_REQ_EVENT, vcpu);
9555 vcpu->arch.apf.msr_val = 0;
9556 vcpu->arch.st.msr_val = 0;
9558 kvmclock_reset(vcpu);
9560 kvm_clear_async_pf_completion_queue(vcpu);
9561 kvm_async_pf_hash_reset(vcpu);
9562 vcpu->arch.apf.halted = false;
9564 if (kvm_mpx_supported()) {
9565 void *mpx_state_buffer;
9568 * To avoid have the INIT path from kvm_apic_has_events() that be
9569 * called with loaded FPU and does not let userspace fix the state.
9572 kvm_put_guest_fpu(vcpu);
9573 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9575 if (mpx_state_buffer)
9576 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9577 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9579 if (mpx_state_buffer)
9580 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9582 kvm_load_guest_fpu(vcpu);
9586 kvm_pmu_reset(vcpu);
9587 vcpu->arch.smbase = 0x30000;
9589 vcpu->arch.msr_misc_features_enables = 0;
9591 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9594 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9595 vcpu->arch.regs_avail = ~0;
9596 vcpu->arch.regs_dirty = ~0;
9598 vcpu->arch.ia32_xss = 0;
9600 kvm_x86_ops.vcpu_reset(vcpu, init_event);
9603 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9605 struct kvm_segment cs;
9607 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9608 cs.selector = vector << 8;
9609 cs.base = vector << 12;
9610 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9611 kvm_rip_write(vcpu, 0);
9614 int kvm_arch_hardware_enable(void)
9617 struct kvm_vcpu *vcpu;
9622 bool stable, backwards_tsc = false;
9624 kvm_shared_msr_cpu_online();
9625 ret = kvm_x86_ops.hardware_enable();
9629 local_tsc = rdtsc();
9630 stable = !kvm_check_tsc_unstable();
9631 list_for_each_entry(kvm, &vm_list, vm_list) {
9632 kvm_for_each_vcpu(i, vcpu, kvm) {
9633 if (!stable && vcpu->cpu == smp_processor_id())
9634 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9635 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9636 backwards_tsc = true;
9637 if (vcpu->arch.last_host_tsc > max_tsc)
9638 max_tsc = vcpu->arch.last_host_tsc;
9644 * Sometimes, even reliable TSCs go backwards. This happens on
9645 * platforms that reset TSC during suspend or hibernate actions, but
9646 * maintain synchronization. We must compensate. Fortunately, we can
9647 * detect that condition here, which happens early in CPU bringup,
9648 * before any KVM threads can be running. Unfortunately, we can't
9649 * bring the TSCs fully up to date with real time, as we aren't yet far
9650 * enough into CPU bringup that we know how much real time has actually
9651 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9652 * variables that haven't been updated yet.
9654 * So we simply find the maximum observed TSC above, then record the
9655 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9656 * the adjustment will be applied. Note that we accumulate
9657 * adjustments, in case multiple suspend cycles happen before some VCPU
9658 * gets a chance to run again. In the event that no KVM threads get a
9659 * chance to run, we will miss the entire elapsed period, as we'll have
9660 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9661 * loose cycle time. This isn't too big a deal, since the loss will be
9662 * uniform across all VCPUs (not to mention the scenario is extremely
9663 * unlikely). It is possible that a second hibernate recovery happens
9664 * much faster than a first, causing the observed TSC here to be
9665 * smaller; this would require additional padding adjustment, which is
9666 * why we set last_host_tsc to the local tsc observed here.
9668 * N.B. - this code below runs only on platforms with reliable TSC,
9669 * as that is the only way backwards_tsc is set above. Also note
9670 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9671 * have the same delta_cyc adjustment applied if backwards_tsc
9672 * is detected. Note further, this adjustment is only done once,
9673 * as we reset last_host_tsc on all VCPUs to stop this from being
9674 * called multiple times (one for each physical CPU bringup).
9676 * Platforms with unreliable TSCs don't have to deal with this, they
9677 * will be compensated by the logic in vcpu_load, which sets the TSC to
9678 * catchup mode. This will catchup all VCPUs to real time, but cannot
9679 * guarantee that they stay in perfect synchronization.
9681 if (backwards_tsc) {
9682 u64 delta_cyc = max_tsc - local_tsc;
9683 list_for_each_entry(kvm, &vm_list, vm_list) {
9684 kvm->arch.backwards_tsc_observed = true;
9685 kvm_for_each_vcpu(i, vcpu, kvm) {
9686 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9687 vcpu->arch.last_host_tsc = local_tsc;
9688 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9692 * We have to disable TSC offset matching.. if you were
9693 * booting a VM while issuing an S4 host suspend....
9694 * you may have some problem. Solving this issue is
9695 * left as an exercise to the reader.
9697 kvm->arch.last_tsc_nsec = 0;
9698 kvm->arch.last_tsc_write = 0;
9705 void kvm_arch_hardware_disable(void)
9707 kvm_x86_ops.hardware_disable();
9708 drop_user_return_notifiers();
9711 int kvm_arch_hardware_setup(void *opaque)
9713 struct kvm_x86_init_ops *ops = opaque;
9716 rdmsrl_safe(MSR_EFER, &host_efer);
9718 if (boot_cpu_has(X86_FEATURE_XSAVES))
9719 rdmsrl(MSR_IA32_XSS, host_xss);
9721 r = ops->hardware_setup();
9725 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9727 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9730 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
9731 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
9732 #undef __kvm_cpu_cap_has
9734 if (kvm_has_tsc_control) {
9736 * Make sure the user can only configure tsc_khz values that
9737 * fit into a signed integer.
9738 * A min value is not calculated because it will always
9739 * be 1 on all machines.
9741 u64 max = min(0x7fffffffULL,
9742 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9743 kvm_max_guest_tsc_khz = max;
9745 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9748 kvm_init_msr_list();
9752 void kvm_arch_hardware_unsetup(void)
9754 kvm_x86_ops.hardware_unsetup();
9757 int kvm_arch_check_processor_compat(void *opaque)
9759 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
9760 struct kvm_x86_init_ops *ops = opaque;
9762 WARN_ON(!irqs_disabled());
9764 if (__cr4_reserved_bits(cpu_has, c) !=
9765 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
9768 return ops->check_processor_compatibility();
9771 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9773 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9775 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9777 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9779 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9782 struct static_key kvm_no_apic_vcpu __read_mostly;
9783 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9785 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9787 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9789 vcpu->arch.l1tf_flush_l1d = true;
9790 if (pmu->version && unlikely(pmu->event_count)) {
9791 pmu->need_cleanup = true;
9792 kvm_make_request(KVM_REQ_PMU, vcpu);
9794 kvm_x86_ops.sched_in(vcpu, cpu);
9797 void kvm_arch_free_vm(struct kvm *kvm)
9799 kfree(kvm->arch.hyperv.hv_pa_pg);
9804 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9809 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9810 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9811 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9812 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9813 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9814 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9816 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9817 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9818 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9819 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9820 &kvm->arch.irq_sources_bitmap);
9822 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9823 mutex_init(&kvm->arch.apic_map_lock);
9824 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9826 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
9827 pvclock_update_vm_gtod_copy(kvm);
9829 kvm->arch.guest_can_read_msr_platform_info = true;
9831 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9832 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9834 kvm_hv_init_vm(kvm);
9835 kvm_page_track_init(kvm);
9836 kvm_mmu_init_vm(kvm);
9838 return kvm_x86_ops.vm_init(kvm);
9841 int kvm_arch_post_init_vm(struct kvm *kvm)
9843 return kvm_mmu_post_init_vm(kvm);
9846 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9849 kvm_mmu_unload(vcpu);
9853 static void kvm_free_vcpus(struct kvm *kvm)
9856 struct kvm_vcpu *vcpu;
9859 * Unpin any mmu pages first.
9861 kvm_for_each_vcpu(i, vcpu, kvm) {
9862 kvm_clear_async_pf_completion_queue(vcpu);
9863 kvm_unload_vcpu_mmu(vcpu);
9865 kvm_for_each_vcpu(i, vcpu, kvm)
9866 kvm_vcpu_destroy(vcpu);
9868 mutex_lock(&kvm->lock);
9869 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9870 kvm->vcpus[i] = NULL;
9872 atomic_set(&kvm->online_vcpus, 0);
9873 mutex_unlock(&kvm->lock);
9876 void kvm_arch_sync_events(struct kvm *kvm)
9878 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9879 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9883 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9886 unsigned long hva, uninitialized_var(old_npages);
9887 struct kvm_memslots *slots = kvm_memslots(kvm);
9888 struct kvm_memory_slot *slot;
9890 /* Called with kvm->slots_lock held. */
9891 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9894 slot = id_to_memslot(slots, id);
9896 if (slot && slot->npages)
9900 * MAP_SHARED to prevent internal slot pages from being moved
9903 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9904 MAP_SHARED | MAP_ANONYMOUS, 0);
9905 if (IS_ERR((void *)hva))
9906 return PTR_ERR((void *)hva);
9908 if (!slot || !slot->npages)
9912 * Stuff a non-canonical value to catch use-after-delete. This
9913 * ends up being 0 on 32-bit KVM, but there's no better
9916 hva = (unsigned long)(0xdeadull << 48);
9917 old_npages = slot->npages;
9920 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9921 struct kvm_userspace_memory_region m;
9923 m.slot = id | (i << 16);
9925 m.guest_phys_addr = gpa;
9926 m.userspace_addr = hva;
9927 m.memory_size = size;
9928 r = __kvm_set_memory_region(kvm, &m);
9934 vm_munmap(hva, old_npages * PAGE_SIZE);
9938 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9940 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9942 kvm_mmu_pre_destroy_vm(kvm);
9945 void kvm_arch_destroy_vm(struct kvm *kvm)
9947 if (current->mm == kvm->mm) {
9949 * Free memory regions allocated on behalf of userspace,
9950 * unless the the memory map has changed due to process exit
9953 mutex_lock(&kvm->slots_lock);
9954 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
9956 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
9958 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9959 mutex_unlock(&kvm->slots_lock);
9961 if (kvm_x86_ops.vm_destroy)
9962 kvm_x86_ops.vm_destroy(kvm);
9963 kvm_pic_destroy(kvm);
9964 kvm_ioapic_destroy(kvm);
9965 kvm_free_vcpus(kvm);
9966 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9967 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9968 kvm_mmu_uninit_vm(kvm);
9969 kvm_page_track_cleanup(kvm);
9970 kvm_hv_destroy_vm(kvm);
9973 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
9977 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9978 kvfree(slot->arch.rmap[i]);
9979 slot->arch.rmap[i] = NULL;
9984 kvfree(slot->arch.lpage_info[i - 1]);
9985 slot->arch.lpage_info[i - 1] = NULL;
9988 kvm_page_track_free_memslot(slot);
9991 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
9992 unsigned long npages)
9997 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
9998 * old arrays will be freed by __kvm_set_memory_region() if installing
9999 * the new memslot is successful.
10001 memset(&slot->arch, 0, sizeof(slot->arch));
10003 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10004 struct kvm_lpage_info *linfo;
10005 unsigned long ugfn;
10009 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10010 slot->base_gfn, level) + 1;
10012 slot->arch.rmap[i] =
10013 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10014 GFP_KERNEL_ACCOUNT);
10015 if (!slot->arch.rmap[i])
10020 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10024 slot->arch.lpage_info[i - 1] = linfo;
10026 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10027 linfo[0].disallow_lpage = 1;
10028 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10029 linfo[lpages - 1].disallow_lpage = 1;
10030 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10032 * If the gfn and userspace address are not aligned wrt each
10033 * other, disable large page support for this slot.
10035 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10038 for (j = 0; j < lpages; ++j)
10039 linfo[j].disallow_lpage = 1;
10043 if (kvm_page_track_create_memslot(slot, npages))
10049 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10050 kvfree(slot->arch.rmap[i]);
10051 slot->arch.rmap[i] = NULL;
10055 kvfree(slot->arch.lpage_info[i - 1]);
10056 slot->arch.lpage_info[i - 1] = NULL;
10061 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10063 struct kvm_vcpu *vcpu;
10067 * memslots->generation has been incremented.
10068 * mmio generation may have reached its maximum value.
10070 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10072 /* Force re-initialization of steal_time cache */
10073 kvm_for_each_vcpu(i, vcpu, kvm)
10074 kvm_vcpu_kick(vcpu);
10077 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10078 struct kvm_memory_slot *memslot,
10079 const struct kvm_userspace_memory_region *mem,
10080 enum kvm_mr_change change)
10082 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10083 return kvm_alloc_memslot_metadata(memslot,
10084 mem->memory_size >> PAGE_SHIFT);
10088 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10089 struct kvm_memory_slot *new)
10091 /* Still write protect RO slot */
10092 if (new->flags & KVM_MEM_READONLY) {
10093 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
10098 * Call kvm_x86_ops dirty logging hooks when they are valid.
10100 * kvm_x86_ops.slot_disable_log_dirty is called when:
10102 * - KVM_MR_CREATE with dirty logging is disabled
10103 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
10105 * The reason is, in case of PML, we need to set D-bit for any slots
10106 * with dirty logging disabled in order to eliminate unnecessary GPA
10107 * logging in PML buffer (and potential PML buffer full VMEXIT). This
10108 * guarantees leaving PML enabled during guest's lifetime won't have
10109 * any additional overhead from PML when guest is running with dirty
10110 * logging disabled for memory slots.
10112 * kvm_x86_ops.slot_enable_log_dirty is called when switching new slot
10113 * to dirty logging mode.
10115 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
10117 * In case of write protect:
10119 * Write protect all pages for dirty logging.
10121 * All the sptes including the large sptes which point to this
10122 * slot are set to readonly. We can not create any new large
10123 * spte on this slot until the end of the logging.
10125 * See the comments in fast_page_fault().
10127 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10128 if (kvm_x86_ops.slot_enable_log_dirty) {
10129 kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10132 kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10133 PG_LEVEL_2M : PG_LEVEL_4K;
10136 * If we're with initial-all-set, we don't need
10137 * to write protect any small page because
10138 * they're reported as dirty already. However
10139 * we still need to write-protect huge pages
10140 * so that the page split can happen lazily on
10141 * the first write to the huge page.
10143 kvm_mmu_slot_remove_write_access(kvm, new, level);
10146 if (kvm_x86_ops.slot_disable_log_dirty)
10147 kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10151 void kvm_arch_commit_memory_region(struct kvm *kvm,
10152 const struct kvm_userspace_memory_region *mem,
10153 struct kvm_memory_slot *old,
10154 const struct kvm_memory_slot *new,
10155 enum kvm_mr_change change)
10157 if (!kvm->arch.n_requested_mmu_pages)
10158 kvm_mmu_change_mmu_pages(kvm,
10159 kvm_mmu_calculate_default_mmu_pages(kvm));
10162 * Dirty logging tracks sptes in 4k granularity, meaning that large
10163 * sptes have to be split. If live migration is successful, the guest
10164 * in the source machine will be destroyed and large sptes will be
10165 * created in the destination. However, if the guest continues to run
10166 * in the source machine (for example if live migration fails), small
10167 * sptes will remain around and cause bad performance.
10169 * Scan sptes if dirty logging has been stopped, dropping those
10170 * which can be collapsed into a single large-page spte. Later
10171 * page faults will create the large-page sptes.
10173 * There is no need to do this in any of the following cases:
10174 * CREATE: No dirty mappings will already exist.
10175 * MOVE/DELETE: The old mappings will already have been cleaned up by
10176 * kvm_arch_flush_shadow_memslot()
10178 if (change == KVM_MR_FLAGS_ONLY &&
10179 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10180 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10181 kvm_mmu_zap_collapsible_sptes(kvm, new);
10184 * Set up write protection and/or dirty logging for the new slot.
10186 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
10187 * been zapped so no dirty logging staff is needed for old slot. For
10188 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
10189 * new and it's also covered when dealing with the new slot.
10191 * FIXME: const-ify all uses of struct kvm_memory_slot.
10193 if (change != KVM_MR_DELETE)
10194 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
10196 /* Free the arrays associated with the old memslot. */
10197 if (change == KVM_MR_MOVE)
10198 kvm_arch_free_memslot(kvm, old);
10201 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10203 kvm_mmu_zap_all(kvm);
10206 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10207 struct kvm_memory_slot *slot)
10209 kvm_page_track_flush_slot(kvm, slot);
10212 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10214 return (is_guest_mode(vcpu) &&
10215 kvm_x86_ops.guest_apic_has_interrupt &&
10216 kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10219 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10221 if (!list_empty_careful(&vcpu->async_pf.done))
10224 if (kvm_apic_has_events(vcpu))
10227 if (vcpu->arch.pv.pv_unhalted)
10230 if (vcpu->arch.exception.pending)
10233 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10234 (vcpu->arch.nmi_pending &&
10235 kvm_x86_ops.nmi_allowed(vcpu, false)))
10238 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10239 (vcpu->arch.smi_pending &&
10240 kvm_x86_ops.smi_allowed(vcpu, false)))
10243 if (kvm_arch_interrupt_allowed(vcpu) &&
10244 (kvm_cpu_has_interrupt(vcpu) ||
10245 kvm_guest_apic_has_interrupt(vcpu)))
10248 if (kvm_hv_has_stimer_pending(vcpu))
10251 if (is_guest_mode(vcpu) &&
10252 kvm_x86_ops.nested_ops->hv_timer_pending &&
10253 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10259 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10261 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10264 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10266 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10269 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10270 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10271 kvm_test_request(KVM_REQ_EVENT, vcpu))
10274 if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10280 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10282 return vcpu->arch.preempted_in_kernel;
10285 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10287 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10290 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10292 return kvm_x86_ops.interrupt_allowed(vcpu, false);
10295 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10297 if (is_64_bit_mode(vcpu))
10298 return kvm_rip_read(vcpu);
10299 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10300 kvm_rip_read(vcpu));
10302 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10304 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10306 return kvm_get_linear_rip(vcpu) == linear_rip;
10308 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10310 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10312 unsigned long rflags;
10314 rflags = kvm_x86_ops.get_rflags(vcpu);
10315 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10316 rflags &= ~X86_EFLAGS_TF;
10319 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10321 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10323 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10324 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10325 rflags |= X86_EFLAGS_TF;
10326 kvm_x86_ops.set_rflags(vcpu, rflags);
10329 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10331 __kvm_set_rflags(vcpu, rflags);
10332 kvm_make_request(KVM_REQ_EVENT, vcpu);
10334 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10336 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10340 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10344 r = kvm_mmu_reload(vcpu);
10348 if (!vcpu->arch.mmu->direct_map &&
10349 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10352 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10355 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10357 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
10359 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10362 static inline u32 kvm_async_pf_next_probe(u32 key)
10364 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
10367 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10369 u32 key = kvm_async_pf_hash_fn(gfn);
10371 while (vcpu->arch.apf.gfns[key] != ~0)
10372 key = kvm_async_pf_next_probe(key);
10374 vcpu->arch.apf.gfns[key] = gfn;
10377 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10380 u32 key = kvm_async_pf_hash_fn(gfn);
10382 for (i = 0; i < ASYNC_PF_PER_VCPU &&
10383 (vcpu->arch.apf.gfns[key] != gfn &&
10384 vcpu->arch.apf.gfns[key] != ~0); i++)
10385 key = kvm_async_pf_next_probe(key);
10390 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10392 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10395 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10399 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10401 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
10405 vcpu->arch.apf.gfns[i] = ~0;
10407 j = kvm_async_pf_next_probe(j);
10408 if (vcpu->arch.apf.gfns[j] == ~0)
10410 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10412 * k lies cyclically in ]i,j]
10414 * |....j i.k.| or |.k..j i...|
10416 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10417 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10422 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10425 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10429 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10432 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10436 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10438 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10441 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10442 (vcpu->arch.apf.send_user_only &&
10443 kvm_x86_ops.get_cpl(vcpu) == 0))
10449 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10451 if (unlikely(!lapic_in_kernel(vcpu) ||
10452 kvm_event_needs_reinjection(vcpu) ||
10453 vcpu->arch.exception.pending))
10456 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10460 * If interrupts are off we cannot even use an artificial
10463 return kvm_arch_interrupt_allowed(vcpu);
10466 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10467 struct kvm_async_pf *work)
10469 struct x86_exception fault;
10471 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10472 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10474 if (kvm_can_deliver_async_pf(vcpu) &&
10475 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10476 fault.vector = PF_VECTOR;
10477 fault.error_code_valid = true;
10478 fault.error_code = 0;
10479 fault.nested_page_fault = false;
10480 fault.address = work->arch.token;
10481 fault.async_page_fault = true;
10482 kvm_inject_page_fault(vcpu, &fault);
10485 * It is not possible to deliver a paravirtualized asynchronous
10486 * page fault, but putting the guest in an artificial halt state
10487 * can be beneficial nevertheless: if an interrupt arrives, we
10488 * can deliver it timely and perhaps the guest will schedule
10489 * another process. When the instruction that triggered a page
10490 * fault is retried, hopefully the page will be ready in the host.
10492 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10496 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10497 struct kvm_async_pf *work)
10499 struct x86_exception fault;
10502 if (work->wakeup_all)
10503 work->arch.token = ~0; /* broadcast wakeup */
10505 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10506 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10508 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10509 !apf_get_user(vcpu, &val)) {
10510 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10511 vcpu->arch.exception.pending &&
10512 vcpu->arch.exception.nr == PF_VECTOR &&
10513 !apf_put_user(vcpu, 0)) {
10514 vcpu->arch.exception.injected = false;
10515 vcpu->arch.exception.pending = false;
10516 vcpu->arch.exception.nr = 0;
10517 vcpu->arch.exception.has_error_code = false;
10518 vcpu->arch.exception.error_code = 0;
10519 vcpu->arch.exception.has_payload = false;
10520 vcpu->arch.exception.payload = 0;
10521 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10522 fault.vector = PF_VECTOR;
10523 fault.error_code_valid = true;
10524 fault.error_code = 0;
10525 fault.nested_page_fault = false;
10526 fault.address = work->arch.token;
10527 fault.async_page_fault = true;
10528 kvm_inject_page_fault(vcpu, &fault);
10531 vcpu->arch.apf.halted = false;
10532 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10535 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10537 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10540 return kvm_can_do_async_pf(vcpu);
10543 void kvm_arch_start_assignment(struct kvm *kvm)
10545 atomic_inc(&kvm->arch.assigned_device_count);
10547 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10549 void kvm_arch_end_assignment(struct kvm *kvm)
10551 atomic_dec(&kvm->arch.assigned_device_count);
10553 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10555 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10557 return atomic_read(&kvm->arch.assigned_device_count);
10559 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10561 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10563 atomic_inc(&kvm->arch.noncoherent_dma_count);
10565 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10567 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10569 atomic_dec(&kvm->arch.noncoherent_dma_count);
10571 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10573 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10575 return atomic_read(&kvm->arch.noncoherent_dma_count);
10577 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10579 bool kvm_arch_has_irq_bypass(void)
10584 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10585 struct irq_bypass_producer *prod)
10587 struct kvm_kernel_irqfd *irqfd =
10588 container_of(cons, struct kvm_kernel_irqfd, consumer);
10590 irqfd->producer = prod;
10592 return kvm_x86_ops.update_pi_irte(irqfd->kvm,
10593 prod->irq, irqfd->gsi, 1);
10596 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10597 struct irq_bypass_producer *prod)
10600 struct kvm_kernel_irqfd *irqfd =
10601 container_of(cons, struct kvm_kernel_irqfd, consumer);
10603 WARN_ON(irqfd->producer != prod);
10604 irqfd->producer = NULL;
10607 * When producer of consumer is unregistered, we change back to
10608 * remapped mode, so we can re-use the current implementation
10609 * when the irq is masked/disabled or the consumer side (KVM
10610 * int this case doesn't want to receive the interrupts.
10612 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10614 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10615 " fails: %d\n", irqfd->consumer.token, ret);
10618 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10619 uint32_t guest_irq, bool set)
10621 return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
10624 bool kvm_vector_hashing_enabled(void)
10626 return vector_hashing;
10629 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10631 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10633 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10635 u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu)
10637 uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD;
10639 /* The STIBP bit doesn't fault even if it's not advertised */
10640 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
10641 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
10642 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10643 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
10644 !boot_cpu_has(X86_FEATURE_AMD_IBRS))
10645 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10647 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) &&
10648 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
10649 bits &= ~SPEC_CTRL_SSBD;
10650 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
10651 !boot_cpu_has(X86_FEATURE_AMD_SSBD))
10652 bits &= ~SPEC_CTRL_SSBD;
10656 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits);
10658 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10659 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10660 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10661 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10662 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10663 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10664 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10665 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10666 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10667 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10668 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10669 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10670 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10671 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10672 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10673 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10674 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10675 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10676 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10677 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10678 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
10679 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);