2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
59 #include <trace/events/kvm.h>
61 #include <asm/debugreg.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
72 #include <asm/intel_pt.h>
74 #define CREATE_TRACE_POINTS
77 #define MAX_IO_MSRS 256
78 #define KVM_MAX_MCE_BANKS 32
79 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
80 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
82 #define emul_to_vcpu(ctxt) \
83 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
86 * - enable syscall per default because its emulated by KVM
87 * - enable LME and LMA per default on 64 bit KVM
91 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
93 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
96 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
97 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
99 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
100 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
102 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
103 static void process_nmi(struct kvm_vcpu *vcpu);
104 static void enter_smm(struct kvm_vcpu *vcpu);
105 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
106 static void store_regs(struct kvm_vcpu *vcpu);
107 static int sync_regs(struct kvm_vcpu *vcpu);
109 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
110 EXPORT_SYMBOL_GPL(kvm_x86_ops);
112 static bool __read_mostly ignore_msrs = 0;
113 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
115 static bool __read_mostly report_ignored_msrs = true;
116 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
118 unsigned int min_timer_period_us = 200;
119 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
121 static bool __read_mostly kvmclock_periodic_sync = true;
122 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
124 bool __read_mostly kvm_has_tsc_control;
125 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
126 u32 __read_mostly kvm_max_guest_tsc_khz;
127 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
128 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
129 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
130 u64 __read_mostly kvm_max_tsc_scaling_ratio;
131 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
132 u64 __read_mostly kvm_default_tsc_scaling_ratio;
133 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
135 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
136 static u32 __read_mostly tsc_tolerance_ppm = 250;
137 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
139 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
140 unsigned int __read_mostly lapic_timer_advance_ns = 1000;
141 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
142 EXPORT_SYMBOL_GPL(lapic_timer_advance_ns);
144 static bool __read_mostly vector_hashing = true;
145 module_param(vector_hashing, bool, S_IRUGO);
147 bool __read_mostly enable_vmware_backdoor = false;
148 module_param(enable_vmware_backdoor, bool, S_IRUGO);
149 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
151 static bool __read_mostly force_emulation_prefix = false;
152 module_param(force_emulation_prefix, bool, S_IRUGO);
154 #define KVM_NR_SHARED_MSRS 16
156 struct kvm_shared_msrs_global {
158 u32 msrs[KVM_NR_SHARED_MSRS];
161 struct kvm_shared_msrs {
162 struct user_return_notifier urn;
164 struct kvm_shared_msr_values {
167 } values[KVM_NR_SHARED_MSRS];
170 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
171 static struct kvm_shared_msrs __percpu *shared_msrs;
173 struct kvm_stats_debugfs_item debugfs_entries[] = {
174 { "pf_fixed", VCPU_STAT(pf_fixed) },
175 { "pf_guest", VCPU_STAT(pf_guest) },
176 { "tlb_flush", VCPU_STAT(tlb_flush) },
177 { "invlpg", VCPU_STAT(invlpg) },
178 { "exits", VCPU_STAT(exits) },
179 { "io_exits", VCPU_STAT(io_exits) },
180 { "mmio_exits", VCPU_STAT(mmio_exits) },
181 { "signal_exits", VCPU_STAT(signal_exits) },
182 { "irq_window", VCPU_STAT(irq_window_exits) },
183 { "nmi_window", VCPU_STAT(nmi_window_exits) },
184 { "halt_exits", VCPU_STAT(halt_exits) },
185 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
186 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
187 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
188 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
189 { "hypercalls", VCPU_STAT(hypercalls) },
190 { "request_irq", VCPU_STAT(request_irq_exits) },
191 { "irq_exits", VCPU_STAT(irq_exits) },
192 { "host_state_reload", VCPU_STAT(host_state_reload) },
193 { "fpu_reload", VCPU_STAT(fpu_reload) },
194 { "insn_emulation", VCPU_STAT(insn_emulation) },
195 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
196 { "irq_injections", VCPU_STAT(irq_injections) },
197 { "nmi_injections", VCPU_STAT(nmi_injections) },
198 { "req_event", VCPU_STAT(req_event) },
199 { "l1d_flush", VCPU_STAT(l1d_flush) },
200 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
201 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
202 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
203 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
204 { "mmu_flooded", VM_STAT(mmu_flooded) },
205 { "mmu_recycled", VM_STAT(mmu_recycled) },
206 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
207 { "mmu_unsync", VM_STAT(mmu_unsync) },
208 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
209 { "largepages", VM_STAT(lpages) },
210 { "max_mmu_page_hash_collisions",
211 VM_STAT(max_mmu_page_hash_collisions) },
215 u64 __read_mostly host_xcr0;
217 struct kmem_cache *x86_fpu_cache;
218 EXPORT_SYMBOL_GPL(x86_fpu_cache);
220 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
222 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
225 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
226 vcpu->arch.apf.gfns[i] = ~0;
229 static void kvm_on_user_return(struct user_return_notifier *urn)
232 struct kvm_shared_msrs *locals
233 = container_of(urn, struct kvm_shared_msrs, urn);
234 struct kvm_shared_msr_values *values;
238 * Disabling irqs at this point since the following code could be
239 * interrupted and executed through kvm_arch_hardware_disable()
241 local_irq_save(flags);
242 if (locals->registered) {
243 locals->registered = false;
244 user_return_notifier_unregister(urn);
246 local_irq_restore(flags);
247 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
248 values = &locals->values[slot];
249 if (values->host != values->curr) {
250 wrmsrl(shared_msrs_global.msrs[slot], values->host);
251 values->curr = values->host;
256 static void shared_msr_update(unsigned slot, u32 msr)
259 unsigned int cpu = smp_processor_id();
260 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
262 /* only read, and nobody should modify it at this time,
263 * so don't need lock */
264 if (slot >= shared_msrs_global.nr) {
265 printk(KERN_ERR "kvm: invalid MSR slot!");
268 rdmsrl_safe(msr, &value);
269 smsr->values[slot].host = value;
270 smsr->values[slot].curr = value;
273 void kvm_define_shared_msr(unsigned slot, u32 msr)
275 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
276 shared_msrs_global.msrs[slot] = msr;
277 if (slot >= shared_msrs_global.nr)
278 shared_msrs_global.nr = slot + 1;
280 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
282 static void kvm_shared_msr_cpu_online(void)
286 for (i = 0; i < shared_msrs_global.nr; ++i)
287 shared_msr_update(i, shared_msrs_global.msrs[i]);
290 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
292 unsigned int cpu = smp_processor_id();
293 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
296 if (((value ^ smsr->values[slot].curr) & mask) == 0)
298 smsr->values[slot].curr = value;
299 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
303 if (!smsr->registered) {
304 smsr->urn.on_user_return = kvm_on_user_return;
305 user_return_notifier_register(&smsr->urn);
306 smsr->registered = true;
310 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
312 static void drop_user_return_notifiers(void)
314 unsigned int cpu = smp_processor_id();
315 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
317 if (smsr->registered)
318 kvm_on_user_return(&smsr->urn);
321 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
323 return vcpu->arch.apic_base;
325 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
327 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
329 return kvm_apic_mode(kvm_get_apic_base(vcpu));
331 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
333 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
335 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
336 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
337 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
338 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
340 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
342 if (!msr_info->host_initiated) {
343 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
345 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
349 kvm_lapic_set_base(vcpu, msr_info->data);
352 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
354 asmlinkage __visible void kvm_spurious_fault(void)
356 /* Fault while not rebooting. We want the trace. */
359 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
361 #define EXCPT_BENIGN 0
362 #define EXCPT_CONTRIBUTORY 1
365 static int exception_class(int vector)
375 return EXCPT_CONTRIBUTORY;
382 #define EXCPT_FAULT 0
384 #define EXCPT_ABORT 2
385 #define EXCPT_INTERRUPT 3
387 static int exception_type(int vector)
391 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
392 return EXCPT_INTERRUPT;
396 /* #DB is trap, as instruction watchpoints are handled elsewhere */
397 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
400 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
403 /* Reserved exceptions will result in fault */
407 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
409 unsigned nr = vcpu->arch.exception.nr;
410 bool has_payload = vcpu->arch.exception.has_payload;
411 unsigned long payload = vcpu->arch.exception.payload;
419 * "Certain debug exceptions may clear bit 0-3. The
420 * remaining contents of the DR6 register are never
421 * cleared by the processor".
423 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
425 * DR6.RTM is set by all #DB exceptions that don't clear it.
427 vcpu->arch.dr6 |= DR6_RTM;
428 vcpu->arch.dr6 |= payload;
430 * Bit 16 should be set in the payload whenever the #DB
431 * exception should clear DR6.RTM. This makes the payload
432 * compatible with the pending debug exceptions under VMX.
433 * Though not currently documented in the SDM, this also
434 * makes the payload compatible with the exit qualification
435 * for #DB exceptions under VMX.
437 vcpu->arch.dr6 ^= payload & DR6_RTM;
440 vcpu->arch.cr2 = payload;
444 vcpu->arch.exception.has_payload = false;
445 vcpu->arch.exception.payload = 0;
447 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
449 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
450 unsigned nr, bool has_error, u32 error_code,
451 bool has_payload, unsigned long payload, bool reinject)
456 kvm_make_request(KVM_REQ_EVENT, vcpu);
458 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
460 if (has_error && !is_protmode(vcpu))
464 * On vmentry, vcpu->arch.exception.pending is only
465 * true if an event injection was blocked by
466 * nested_run_pending. In that case, however,
467 * vcpu_enter_guest requests an immediate exit,
468 * and the guest shouldn't proceed far enough to
471 WARN_ON_ONCE(vcpu->arch.exception.pending);
472 vcpu->arch.exception.injected = true;
473 if (WARN_ON_ONCE(has_payload)) {
475 * A reinjected event has already
476 * delivered its payload.
482 vcpu->arch.exception.pending = true;
483 vcpu->arch.exception.injected = false;
485 vcpu->arch.exception.has_error_code = has_error;
486 vcpu->arch.exception.nr = nr;
487 vcpu->arch.exception.error_code = error_code;
488 vcpu->arch.exception.has_payload = has_payload;
489 vcpu->arch.exception.payload = payload;
491 * In guest mode, payload delivery should be deferred,
492 * so that the L1 hypervisor can intercept #PF before
493 * CR2 is modified (or intercept #DB before DR6 is
494 * modified under nVMX). However, for ABI
495 * compatibility with KVM_GET_VCPU_EVENTS and
496 * KVM_SET_VCPU_EVENTS, we can't delay payload
497 * delivery unless userspace has enabled this
498 * functionality via the per-VM capability,
499 * KVM_CAP_EXCEPTION_PAYLOAD.
501 if (!vcpu->kvm->arch.exception_payload_enabled ||
502 !is_guest_mode(vcpu))
503 kvm_deliver_exception_payload(vcpu);
507 /* to check exception */
508 prev_nr = vcpu->arch.exception.nr;
509 if (prev_nr == DF_VECTOR) {
510 /* triple fault -> shutdown */
511 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
514 class1 = exception_class(prev_nr);
515 class2 = exception_class(nr);
516 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
517 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
519 * Generate double fault per SDM Table 5-5. Set
520 * exception.pending = true so that the double fault
521 * can trigger a nested vmexit.
523 vcpu->arch.exception.pending = true;
524 vcpu->arch.exception.injected = false;
525 vcpu->arch.exception.has_error_code = true;
526 vcpu->arch.exception.nr = DF_VECTOR;
527 vcpu->arch.exception.error_code = 0;
528 vcpu->arch.exception.has_payload = false;
529 vcpu->arch.exception.payload = 0;
531 /* replace previous exception with a new one in a hope
532 that instruction re-execution will regenerate lost
537 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
539 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
541 EXPORT_SYMBOL_GPL(kvm_queue_exception);
543 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
545 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
547 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
549 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
550 unsigned long payload)
552 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
555 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
556 u32 error_code, unsigned long payload)
558 kvm_multiple_exception(vcpu, nr, true, error_code,
559 true, payload, false);
562 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
565 kvm_inject_gp(vcpu, 0);
567 return kvm_skip_emulated_instruction(vcpu);
571 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
573 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
575 ++vcpu->stat.pf_guest;
576 vcpu->arch.exception.nested_apf =
577 is_guest_mode(vcpu) && fault->async_page_fault;
578 if (vcpu->arch.exception.nested_apf) {
579 vcpu->arch.apf.nested_apf_token = fault->address;
580 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
582 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
586 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
588 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
590 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
591 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
593 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
595 return fault->nested_page_fault;
598 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
600 atomic_inc(&vcpu->arch.nmi_queued);
601 kvm_make_request(KVM_REQ_NMI, vcpu);
603 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
605 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
607 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
609 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
611 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
613 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
615 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
618 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
619 * a #GP and return false.
621 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
623 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
625 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
628 EXPORT_SYMBOL_GPL(kvm_require_cpl);
630 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
632 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
635 kvm_queue_exception(vcpu, UD_VECTOR);
638 EXPORT_SYMBOL_GPL(kvm_require_dr);
641 * This function will be used to read from the physical memory of the currently
642 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
643 * can read from guest physical or from the guest's guest physical memory.
645 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
646 gfn_t ngfn, void *data, int offset, int len,
649 struct x86_exception exception;
653 ngpa = gfn_to_gpa(ngfn);
654 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
655 if (real_gfn == UNMAPPED_GVA)
658 real_gfn = gpa_to_gfn(real_gfn);
660 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
662 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
664 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
665 void *data, int offset, int len, u32 access)
667 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
668 data, offset, len, access);
672 * Load the pae pdptrs. Return true is they are all valid.
674 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
676 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
677 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
680 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
682 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
683 offset * sizeof(u64), sizeof(pdpte),
684 PFERR_USER_MASK|PFERR_WRITE_MASK);
689 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
690 if ((pdpte[i] & PT_PRESENT_MASK) &&
692 vcpu->arch.mmu->guest_rsvd_check.rsvd_bits_mask[0][2])) {
699 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
700 __set_bit(VCPU_EXREG_PDPTR,
701 (unsigned long *)&vcpu->arch.regs_avail);
702 __set_bit(VCPU_EXREG_PDPTR,
703 (unsigned long *)&vcpu->arch.regs_dirty);
708 EXPORT_SYMBOL_GPL(load_pdptrs);
710 bool pdptrs_changed(struct kvm_vcpu *vcpu)
712 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
718 if (is_long_mode(vcpu) || !is_pae(vcpu) || !is_paging(vcpu))
721 if (!test_bit(VCPU_EXREG_PDPTR,
722 (unsigned long *)&vcpu->arch.regs_avail))
725 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
726 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
727 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
728 PFERR_USER_MASK | PFERR_WRITE_MASK);
731 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
736 EXPORT_SYMBOL_GPL(pdptrs_changed);
738 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
740 unsigned long old_cr0 = kvm_read_cr0(vcpu);
741 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
746 if (cr0 & 0xffffffff00000000UL)
750 cr0 &= ~CR0_RESERVED_BITS;
752 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
755 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
758 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
760 if ((vcpu->arch.efer & EFER_LME)) {
765 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
770 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
775 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
778 kvm_x86_ops->set_cr0(vcpu, cr0);
780 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
781 kvm_clear_async_pf_completion_queue(vcpu);
782 kvm_async_pf_hash_reset(vcpu);
785 if ((cr0 ^ old_cr0) & update_bits)
786 kvm_mmu_reset_context(vcpu);
788 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
789 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
790 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
791 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
795 EXPORT_SYMBOL_GPL(kvm_set_cr0);
797 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
799 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
801 EXPORT_SYMBOL_GPL(kvm_lmsw);
803 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
805 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
806 !vcpu->guest_xcr0_loaded) {
807 /* kvm_set_xcr() also depends on this */
808 if (vcpu->arch.xcr0 != host_xcr0)
809 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
810 vcpu->guest_xcr0_loaded = 1;
814 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
816 if (vcpu->guest_xcr0_loaded) {
817 if (vcpu->arch.xcr0 != host_xcr0)
818 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
819 vcpu->guest_xcr0_loaded = 0;
823 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
826 u64 old_xcr0 = vcpu->arch.xcr0;
829 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
830 if (index != XCR_XFEATURE_ENABLED_MASK)
832 if (!(xcr0 & XFEATURE_MASK_FP))
834 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
838 * Do not allow the guest to set bits that we do not support
839 * saving. However, xcr0 bit 0 is always set, even if the
840 * emulated CPU does not support XSAVE (see fx_init).
842 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
843 if (xcr0 & ~valid_bits)
846 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
847 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
850 if (xcr0 & XFEATURE_MASK_AVX512) {
851 if (!(xcr0 & XFEATURE_MASK_YMM))
853 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
856 vcpu->arch.xcr0 = xcr0;
858 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
859 kvm_update_cpuid(vcpu);
863 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
865 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
866 __kvm_set_xcr(vcpu, index, xcr)) {
867 kvm_inject_gp(vcpu, 0);
872 EXPORT_SYMBOL_GPL(kvm_set_xcr);
874 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
876 unsigned long old_cr4 = kvm_read_cr4(vcpu);
877 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
878 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
880 if (cr4 & CR4_RESERVED_BITS)
883 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
886 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
889 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
892 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
895 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
898 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
901 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
904 if (is_long_mode(vcpu)) {
905 if (!(cr4 & X86_CR4_PAE))
907 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
908 && ((cr4 ^ old_cr4) & pdptr_bits)
909 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
913 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
914 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
917 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
918 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
922 if (kvm_x86_ops->set_cr4(vcpu, cr4))
925 if (((cr4 ^ old_cr4) & pdptr_bits) ||
926 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
927 kvm_mmu_reset_context(vcpu);
929 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
930 kvm_update_cpuid(vcpu);
934 EXPORT_SYMBOL_GPL(kvm_set_cr4);
936 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
938 bool skip_tlb_flush = false;
940 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
943 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
944 cr3 &= ~X86_CR3_PCID_NOFLUSH;
948 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
949 if (!skip_tlb_flush) {
950 kvm_mmu_sync_roots(vcpu);
951 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
956 if (is_long_mode(vcpu) &&
957 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
959 else if (is_pae(vcpu) && is_paging(vcpu) &&
960 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
963 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
964 vcpu->arch.cr3 = cr3;
965 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
969 EXPORT_SYMBOL_GPL(kvm_set_cr3);
971 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
973 if (cr8 & CR8_RESERVED_BITS)
975 if (lapic_in_kernel(vcpu))
976 kvm_lapic_set_tpr(vcpu, cr8);
978 vcpu->arch.cr8 = cr8;
981 EXPORT_SYMBOL_GPL(kvm_set_cr8);
983 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
985 if (lapic_in_kernel(vcpu))
986 return kvm_lapic_get_cr8(vcpu);
988 return vcpu->arch.cr8;
990 EXPORT_SYMBOL_GPL(kvm_get_cr8);
992 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
996 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
997 for (i = 0; i < KVM_NR_DB_REGS; i++)
998 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
999 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1003 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1005 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1006 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1009 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1013 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1014 dr7 = vcpu->arch.guest_debug_dr7;
1016 dr7 = vcpu->arch.dr7;
1017 kvm_x86_ops->set_dr7(vcpu, dr7);
1018 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1019 if (dr7 & DR7_BP_EN_MASK)
1020 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1023 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1025 u64 fixed = DR6_FIXED_1;
1027 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1032 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1036 vcpu->arch.db[dr] = val;
1037 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1038 vcpu->arch.eff_db[dr] = val;
1043 if (val & 0xffffffff00000000ULL)
1044 return -1; /* #GP */
1045 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1046 kvm_update_dr6(vcpu);
1051 if (val & 0xffffffff00000000ULL)
1052 return -1; /* #GP */
1053 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1054 kvm_update_dr7(vcpu);
1061 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1063 if (__kvm_set_dr(vcpu, dr, val)) {
1064 kvm_inject_gp(vcpu, 0);
1069 EXPORT_SYMBOL_GPL(kvm_set_dr);
1071 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1075 *val = vcpu->arch.db[dr];
1080 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1081 *val = vcpu->arch.dr6;
1083 *val = kvm_x86_ops->get_dr6(vcpu);
1088 *val = vcpu->arch.dr7;
1093 EXPORT_SYMBOL_GPL(kvm_get_dr);
1095 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1097 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
1101 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1104 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1105 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1108 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1111 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1112 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1114 * This list is modified at module load time to reflect the
1115 * capabilities of the host cpu. This capabilities test skips MSRs that are
1116 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1117 * may depend on host virtualization features rather than host cpu features.
1120 static u32 msrs_to_save[] = {
1121 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1123 #ifdef CONFIG_X86_64
1124 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1126 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1127 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1128 MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES,
1129 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1130 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1131 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1132 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1133 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1134 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1137 static unsigned num_msrs_to_save;
1139 static u32 emulated_msrs[] = {
1140 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1141 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1142 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1143 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1144 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1145 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1146 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1148 HV_X64_MSR_VP_INDEX,
1149 HV_X64_MSR_VP_RUNTIME,
1150 HV_X64_MSR_SCONTROL,
1151 HV_X64_MSR_STIMER0_CONFIG,
1152 HV_X64_MSR_VP_ASSIST_PAGE,
1153 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1154 HV_X64_MSR_TSC_EMULATION_STATUS,
1156 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1159 MSR_IA32_TSC_ADJUST,
1160 MSR_IA32_TSCDEADLINE,
1161 MSR_IA32_MISC_ENABLE,
1162 MSR_IA32_MCG_STATUS,
1164 MSR_IA32_MCG_EXT_CTL,
1168 MSR_MISC_FEATURES_ENABLES,
1169 MSR_AMD64_VIRT_SPEC_CTRL,
1172 static unsigned num_emulated_msrs;
1175 * List of msr numbers which are used to expose MSR-based features that
1176 * can be used by a hypervisor to validate requested CPU features.
1178 static u32 msr_based_features[] = {
1180 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1181 MSR_IA32_VMX_PINBASED_CTLS,
1182 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1183 MSR_IA32_VMX_PROCBASED_CTLS,
1184 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1185 MSR_IA32_VMX_EXIT_CTLS,
1186 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1187 MSR_IA32_VMX_ENTRY_CTLS,
1189 MSR_IA32_VMX_CR0_FIXED0,
1190 MSR_IA32_VMX_CR0_FIXED1,
1191 MSR_IA32_VMX_CR4_FIXED0,
1192 MSR_IA32_VMX_CR4_FIXED1,
1193 MSR_IA32_VMX_VMCS_ENUM,
1194 MSR_IA32_VMX_PROCBASED_CTLS2,
1195 MSR_IA32_VMX_EPT_VPID_CAP,
1196 MSR_IA32_VMX_VMFUNC,
1200 MSR_IA32_ARCH_CAPABILITIES,
1203 static unsigned int num_msr_based_features;
1205 u64 kvm_get_arch_capabilities(void)
1209 rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1212 * If we're doing cache flushes (either "always" or "cond")
1213 * we will do one whenever the guest does a vmlaunch/vmresume.
1214 * If an outer hypervisor is doing the cache flush for us
1215 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1216 * capability to the guest too, and if EPT is disabled we're not
1217 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1218 * require a nested hypervisor to do a flush of its own.
1220 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1221 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1225 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1227 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1229 switch (msr->index) {
1230 case MSR_IA32_ARCH_CAPABILITIES:
1231 msr->data = kvm_get_arch_capabilities();
1233 case MSR_IA32_UCODE_REV:
1234 rdmsrl_safe(msr->index, &msr->data);
1237 if (kvm_x86_ops->get_msr_feature(msr))
1243 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1245 struct kvm_msr_entry msr;
1249 r = kvm_get_msr_feature(&msr);
1258 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1260 if (efer & efer_reserved_bits)
1263 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1266 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1271 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1273 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1275 u64 old_efer = vcpu->arch.efer;
1277 if (!kvm_valid_efer(vcpu, efer))
1281 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1285 efer |= vcpu->arch.efer & EFER_LMA;
1287 kvm_x86_ops->set_efer(vcpu, efer);
1289 /* Update reserved bits */
1290 if ((efer ^ old_efer) & EFER_NX)
1291 kvm_mmu_reset_context(vcpu);
1296 void kvm_enable_efer_bits(u64 mask)
1298 efer_reserved_bits &= ~mask;
1300 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1303 * Writes msr value into into the appropriate "register".
1304 * Returns 0 on success, non-0 otherwise.
1305 * Assumes vcpu_load() was already called.
1307 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1309 switch (msr->index) {
1312 case MSR_KERNEL_GS_BASE:
1315 if (is_noncanonical_address(msr->data, vcpu))
1318 case MSR_IA32_SYSENTER_EIP:
1319 case MSR_IA32_SYSENTER_ESP:
1321 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1322 * non-canonical address is written on Intel but not on
1323 * AMD (which ignores the top 32-bits, because it does
1324 * not implement 64-bit SYSENTER).
1326 * 64-bit code should hence be able to write a non-canonical
1327 * value on AMD. Making the address canonical ensures that
1328 * vmentry does not fail on Intel after writing a non-canonical
1329 * value, and that something deterministic happens if the guest
1330 * invokes 64-bit SYSENTER.
1332 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1334 return kvm_x86_ops->set_msr(vcpu, msr);
1336 EXPORT_SYMBOL_GPL(kvm_set_msr);
1339 * Adapt set_msr() to msr_io()'s calling convention
1341 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1343 struct msr_data msr;
1347 msr.host_initiated = true;
1348 r = kvm_get_msr(vcpu, &msr);
1356 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1358 struct msr_data msr;
1362 msr.host_initiated = true;
1363 return kvm_set_msr(vcpu, &msr);
1366 #ifdef CONFIG_X86_64
1367 struct pvclock_gtod_data {
1370 struct { /* extract of a clocksource struct */
1383 static struct pvclock_gtod_data pvclock_gtod_data;
1385 static void update_pvclock_gtod(struct timekeeper *tk)
1387 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1390 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1392 write_seqcount_begin(&vdata->seq);
1394 /* copy pvclock gtod data */
1395 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1396 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1397 vdata->clock.mask = tk->tkr_mono.mask;
1398 vdata->clock.mult = tk->tkr_mono.mult;
1399 vdata->clock.shift = tk->tkr_mono.shift;
1401 vdata->boot_ns = boot_ns;
1402 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1404 vdata->wall_time_sec = tk->xtime_sec;
1406 write_seqcount_end(&vdata->seq);
1410 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1413 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1414 * vcpu_enter_guest. This function is only called from
1415 * the physical CPU that is running vcpu.
1417 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1420 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1424 struct pvclock_wall_clock wc;
1425 struct timespec64 boot;
1430 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1435 ++version; /* first time write, random junk */
1439 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1443 * The guest calculates current wall clock time by adding
1444 * system time (updated by kvm_guest_time_update below) to the
1445 * wall clock specified here. guest system time equals host
1446 * system time for us, thus we must fill in host boot time here.
1448 getboottime64(&boot);
1450 if (kvm->arch.kvmclock_offset) {
1451 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1452 boot = timespec64_sub(boot, ts);
1454 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1455 wc.nsec = boot.tv_nsec;
1456 wc.version = version;
1458 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1461 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1464 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1466 do_shl32_div32(dividend, divisor);
1470 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1471 s8 *pshift, u32 *pmultiplier)
1479 scaled64 = scaled_hz;
1480 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1485 tps32 = (uint32_t)tps64;
1486 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1487 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1495 *pmultiplier = div_frac(scaled64, tps32);
1497 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1498 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1501 #ifdef CONFIG_X86_64
1502 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1505 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1506 static unsigned long max_tsc_khz;
1508 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1510 u64 v = (u64)khz * (1000000 + ppm);
1515 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1519 /* Guest TSC same frequency as host TSC? */
1521 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1525 /* TSC scaling supported? */
1526 if (!kvm_has_tsc_control) {
1527 if (user_tsc_khz > tsc_khz) {
1528 vcpu->arch.tsc_catchup = 1;
1529 vcpu->arch.tsc_always_catchup = 1;
1532 WARN(1, "user requested TSC rate below hardware speed\n");
1537 /* TSC scaling required - calculate ratio */
1538 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1539 user_tsc_khz, tsc_khz);
1541 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1542 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1547 vcpu->arch.tsc_scaling_ratio = ratio;
1551 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1553 u32 thresh_lo, thresh_hi;
1554 int use_scaling = 0;
1556 /* tsc_khz can be zero if TSC calibration fails */
1557 if (user_tsc_khz == 0) {
1558 /* set tsc_scaling_ratio to a safe value */
1559 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1563 /* Compute a scale to convert nanoseconds in TSC cycles */
1564 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1565 &vcpu->arch.virtual_tsc_shift,
1566 &vcpu->arch.virtual_tsc_mult);
1567 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1570 * Compute the variation in TSC rate which is acceptable
1571 * within the range of tolerance and decide if the
1572 * rate being applied is within that bounds of the hardware
1573 * rate. If so, no scaling or compensation need be done.
1575 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1576 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1577 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1578 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1581 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1584 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1586 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1587 vcpu->arch.virtual_tsc_mult,
1588 vcpu->arch.virtual_tsc_shift);
1589 tsc += vcpu->arch.this_tsc_write;
1593 static inline int gtod_is_based_on_tsc(int mode)
1595 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1598 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1600 #ifdef CONFIG_X86_64
1602 struct kvm_arch *ka = &vcpu->kvm->arch;
1603 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1605 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1606 atomic_read(&vcpu->kvm->online_vcpus));
1609 * Once the masterclock is enabled, always perform request in
1610 * order to update it.
1612 * In order to enable masterclock, the host clocksource must be TSC
1613 * and the vcpus need to have matched TSCs. When that happens,
1614 * perform request to enable masterclock.
1616 if (ka->use_master_clock ||
1617 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1618 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1620 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1621 atomic_read(&vcpu->kvm->online_vcpus),
1622 ka->use_master_clock, gtod->clock.vclock_mode);
1626 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1628 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1629 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1633 * Multiply tsc by a fixed point number represented by ratio.
1635 * The most significant 64-N bits (mult) of ratio represent the
1636 * integral part of the fixed point number; the remaining N bits
1637 * (frac) represent the fractional part, ie. ratio represents a fixed
1638 * point number (mult + frac * 2^(-N)).
1640 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1642 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1644 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1647 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1650 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1652 if (ratio != kvm_default_tsc_scaling_ratio)
1653 _tsc = __scale_tsc(ratio, tsc);
1657 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1659 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1663 tsc = kvm_scale_tsc(vcpu, rdtsc());
1665 return target_tsc - tsc;
1668 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1670 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1672 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1674 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1676 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1678 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1681 static inline bool kvm_check_tsc_unstable(void)
1683 #ifdef CONFIG_X86_64
1685 * TSC is marked unstable when we're running on Hyper-V,
1686 * 'TSC page' clocksource is good.
1688 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1691 return check_tsc_unstable();
1694 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1696 struct kvm *kvm = vcpu->kvm;
1697 u64 offset, ns, elapsed;
1698 unsigned long flags;
1700 bool already_matched;
1701 u64 data = msr->data;
1702 bool synchronizing = false;
1704 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1705 offset = kvm_compute_tsc_offset(vcpu, data);
1706 ns = ktime_get_boot_ns();
1707 elapsed = ns - kvm->arch.last_tsc_nsec;
1709 if (vcpu->arch.virtual_tsc_khz) {
1710 if (data == 0 && msr->host_initiated) {
1712 * detection of vcpu initialization -- need to sync
1713 * with other vCPUs. This particularly helps to keep
1714 * kvm_clock stable after CPU hotplug
1716 synchronizing = true;
1718 u64 tsc_exp = kvm->arch.last_tsc_write +
1719 nsec_to_cycles(vcpu, elapsed);
1720 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1722 * Special case: TSC write with a small delta (1 second)
1723 * of virtual cycle time against real time is
1724 * interpreted as an attempt to synchronize the CPU.
1726 synchronizing = data < tsc_exp + tsc_hz &&
1727 data + tsc_hz > tsc_exp;
1732 * For a reliable TSC, we can match TSC offsets, and for an unstable
1733 * TSC, we add elapsed time in this computation. We could let the
1734 * compensation code attempt to catch up if we fall behind, but
1735 * it's better to try to match offsets from the beginning.
1737 if (synchronizing &&
1738 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1739 if (!kvm_check_tsc_unstable()) {
1740 offset = kvm->arch.cur_tsc_offset;
1741 pr_debug("kvm: matched tsc offset for %llu\n", data);
1743 u64 delta = nsec_to_cycles(vcpu, elapsed);
1745 offset = kvm_compute_tsc_offset(vcpu, data);
1746 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1749 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1752 * We split periods of matched TSC writes into generations.
1753 * For each generation, we track the original measured
1754 * nanosecond time, offset, and write, so if TSCs are in
1755 * sync, we can match exact offset, and if not, we can match
1756 * exact software computation in compute_guest_tsc()
1758 * These values are tracked in kvm->arch.cur_xxx variables.
1760 kvm->arch.cur_tsc_generation++;
1761 kvm->arch.cur_tsc_nsec = ns;
1762 kvm->arch.cur_tsc_write = data;
1763 kvm->arch.cur_tsc_offset = offset;
1765 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1766 kvm->arch.cur_tsc_generation, data);
1770 * We also track th most recent recorded KHZ, write and time to
1771 * allow the matching interval to be extended at each write.
1773 kvm->arch.last_tsc_nsec = ns;
1774 kvm->arch.last_tsc_write = data;
1775 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1777 vcpu->arch.last_guest_tsc = data;
1779 /* Keep track of which generation this VCPU has synchronized to */
1780 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1781 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1782 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1784 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1785 update_ia32_tsc_adjust_msr(vcpu, offset);
1787 kvm_vcpu_write_tsc_offset(vcpu, offset);
1788 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1790 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1792 kvm->arch.nr_vcpus_matched_tsc = 0;
1793 } else if (!already_matched) {
1794 kvm->arch.nr_vcpus_matched_tsc++;
1797 kvm_track_tsc_matching(vcpu);
1798 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1801 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1803 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1806 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1807 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1810 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1812 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1813 WARN_ON(adjustment < 0);
1814 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1815 adjust_tsc_offset_guest(vcpu, adjustment);
1818 #ifdef CONFIG_X86_64
1820 static u64 read_tsc(void)
1822 u64 ret = (u64)rdtsc_ordered();
1823 u64 last = pvclock_gtod_data.clock.cycle_last;
1825 if (likely(ret >= last))
1829 * GCC likes to generate cmov here, but this branch is extremely
1830 * predictable (it's just a function of time and the likely is
1831 * very likely) and there's a data dependence, so force GCC
1832 * to generate a branch instead. I don't barrier() because
1833 * we don't actually need a barrier, and if this function
1834 * ever gets inlined it will generate worse code.
1840 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1843 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1846 switch (gtod->clock.vclock_mode) {
1847 case VCLOCK_HVCLOCK:
1848 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1850 if (tsc_pg_val != U64_MAX) {
1851 /* TSC page valid */
1852 *mode = VCLOCK_HVCLOCK;
1853 v = (tsc_pg_val - gtod->clock.cycle_last) &
1856 /* TSC page invalid */
1857 *mode = VCLOCK_NONE;
1862 *tsc_timestamp = read_tsc();
1863 v = (*tsc_timestamp - gtod->clock.cycle_last) &
1867 *mode = VCLOCK_NONE;
1870 if (*mode == VCLOCK_NONE)
1871 *tsc_timestamp = v = 0;
1873 return v * gtod->clock.mult;
1876 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1878 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1884 seq = read_seqcount_begin(>od->seq);
1885 ns = gtod->nsec_base;
1886 ns += vgettsc(tsc_timestamp, &mode);
1887 ns >>= gtod->clock.shift;
1888 ns += gtod->boot_ns;
1889 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1895 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1897 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1903 seq = read_seqcount_begin(>od->seq);
1904 ts->tv_sec = gtod->wall_time_sec;
1905 ns = gtod->nsec_base;
1906 ns += vgettsc(tsc_timestamp, &mode);
1907 ns >>= gtod->clock.shift;
1908 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1910 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1916 /* returns true if host is using TSC based clocksource */
1917 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1919 /* checked again under seqlock below */
1920 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1923 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1927 /* returns true if host is using TSC based clocksource */
1928 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1931 /* checked again under seqlock below */
1932 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1935 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1941 * Assuming a stable TSC across physical CPUS, and a stable TSC
1942 * across virtual CPUs, the following condition is possible.
1943 * Each numbered line represents an event visible to both
1944 * CPUs at the next numbered event.
1946 * "timespecX" represents host monotonic time. "tscX" represents
1949 * VCPU0 on CPU0 | VCPU1 on CPU1
1951 * 1. read timespec0,tsc0
1952 * 2. | timespec1 = timespec0 + N
1954 * 3. transition to guest | transition to guest
1955 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1956 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1957 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1959 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1962 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1964 * - 0 < N - M => M < N
1966 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1967 * always the case (the difference between two distinct xtime instances
1968 * might be smaller then the difference between corresponding TSC reads,
1969 * when updating guest vcpus pvclock areas).
1971 * To avoid that problem, do not allow visibility of distinct
1972 * system_timestamp/tsc_timestamp values simultaneously: use a master
1973 * copy of host monotonic time values. Update that master copy
1976 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1980 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1982 #ifdef CONFIG_X86_64
1983 struct kvm_arch *ka = &kvm->arch;
1985 bool host_tsc_clocksource, vcpus_matched;
1987 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1988 atomic_read(&kvm->online_vcpus));
1991 * If the host uses TSC clock, then passthrough TSC as stable
1994 host_tsc_clocksource = kvm_get_time_and_clockread(
1995 &ka->master_kernel_ns,
1996 &ka->master_cycle_now);
1998 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1999 && !ka->backwards_tsc_observed
2000 && !ka->boot_vcpu_runs_old_kvmclock;
2002 if (ka->use_master_clock)
2003 atomic_set(&kvm_guest_has_master_clock, 1);
2005 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2006 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2011 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2013 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2016 static void kvm_gen_update_masterclock(struct kvm *kvm)
2018 #ifdef CONFIG_X86_64
2020 struct kvm_vcpu *vcpu;
2021 struct kvm_arch *ka = &kvm->arch;
2023 spin_lock(&ka->pvclock_gtod_sync_lock);
2024 kvm_make_mclock_inprogress_request(kvm);
2025 /* no guest entries from this point */
2026 pvclock_update_vm_gtod_copy(kvm);
2028 kvm_for_each_vcpu(i, vcpu, kvm)
2029 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2031 /* guest entries allowed */
2032 kvm_for_each_vcpu(i, vcpu, kvm)
2033 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2035 spin_unlock(&ka->pvclock_gtod_sync_lock);
2039 u64 get_kvmclock_ns(struct kvm *kvm)
2041 struct kvm_arch *ka = &kvm->arch;
2042 struct pvclock_vcpu_time_info hv_clock;
2045 spin_lock(&ka->pvclock_gtod_sync_lock);
2046 if (!ka->use_master_clock) {
2047 spin_unlock(&ka->pvclock_gtod_sync_lock);
2048 return ktime_get_boot_ns() + ka->kvmclock_offset;
2051 hv_clock.tsc_timestamp = ka->master_cycle_now;
2052 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2053 spin_unlock(&ka->pvclock_gtod_sync_lock);
2055 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2058 if (__this_cpu_read(cpu_tsc_khz)) {
2059 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2060 &hv_clock.tsc_shift,
2061 &hv_clock.tsc_to_system_mul);
2062 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2064 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2071 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2073 struct kvm_vcpu_arch *vcpu = &v->arch;
2074 struct pvclock_vcpu_time_info guest_hv_clock;
2076 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2077 &guest_hv_clock, sizeof(guest_hv_clock))))
2080 /* This VCPU is paused, but it's legal for a guest to read another
2081 * VCPU's kvmclock, so we really have to follow the specification where
2082 * it says that version is odd if data is being modified, and even after
2085 * Version field updates must be kept separate. This is because
2086 * kvm_write_guest_cached might use a "rep movs" instruction, and
2087 * writes within a string instruction are weakly ordered. So there
2088 * are three writes overall.
2090 * As a small optimization, only write the version field in the first
2091 * and third write. The vcpu->pv_time cache is still valid, because the
2092 * version field is the first in the struct.
2094 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2096 if (guest_hv_clock.version & 1)
2097 ++guest_hv_clock.version; /* first time write, random junk */
2099 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2100 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2102 sizeof(vcpu->hv_clock.version));
2106 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2107 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2109 if (vcpu->pvclock_set_guest_stopped_request) {
2110 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2111 vcpu->pvclock_set_guest_stopped_request = false;
2114 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2116 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2118 sizeof(vcpu->hv_clock));
2122 vcpu->hv_clock.version++;
2123 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2125 sizeof(vcpu->hv_clock.version));
2128 static int kvm_guest_time_update(struct kvm_vcpu *v)
2130 unsigned long flags, tgt_tsc_khz;
2131 struct kvm_vcpu_arch *vcpu = &v->arch;
2132 struct kvm_arch *ka = &v->kvm->arch;
2134 u64 tsc_timestamp, host_tsc;
2136 bool use_master_clock;
2142 * If the host uses TSC clock, then passthrough TSC as stable
2145 spin_lock(&ka->pvclock_gtod_sync_lock);
2146 use_master_clock = ka->use_master_clock;
2147 if (use_master_clock) {
2148 host_tsc = ka->master_cycle_now;
2149 kernel_ns = ka->master_kernel_ns;
2151 spin_unlock(&ka->pvclock_gtod_sync_lock);
2153 /* Keep irq disabled to prevent changes to the clock */
2154 local_irq_save(flags);
2155 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2156 if (unlikely(tgt_tsc_khz == 0)) {
2157 local_irq_restore(flags);
2158 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2161 if (!use_master_clock) {
2163 kernel_ns = ktime_get_boot_ns();
2166 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2169 * We may have to catch up the TSC to match elapsed wall clock
2170 * time for two reasons, even if kvmclock is used.
2171 * 1) CPU could have been running below the maximum TSC rate
2172 * 2) Broken TSC compensation resets the base at each VCPU
2173 * entry to avoid unknown leaps of TSC even when running
2174 * again on the same CPU. This may cause apparent elapsed
2175 * time to disappear, and the guest to stand still or run
2178 if (vcpu->tsc_catchup) {
2179 u64 tsc = compute_guest_tsc(v, kernel_ns);
2180 if (tsc > tsc_timestamp) {
2181 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2182 tsc_timestamp = tsc;
2186 local_irq_restore(flags);
2188 /* With all the info we got, fill in the values */
2190 if (kvm_has_tsc_control)
2191 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2193 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2194 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2195 &vcpu->hv_clock.tsc_shift,
2196 &vcpu->hv_clock.tsc_to_system_mul);
2197 vcpu->hw_tsc_khz = tgt_tsc_khz;
2200 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2201 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2202 vcpu->last_guest_tsc = tsc_timestamp;
2204 /* If the host uses TSC clocksource, then it is stable */
2206 if (use_master_clock)
2207 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2209 vcpu->hv_clock.flags = pvclock_flags;
2211 if (vcpu->pv_time_enabled)
2212 kvm_setup_pvclock_page(v);
2213 if (v == kvm_get_vcpu(v->kvm, 0))
2214 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2219 * kvmclock updates which are isolated to a given vcpu, such as
2220 * vcpu->cpu migration, should not allow system_timestamp from
2221 * the rest of the vcpus to remain static. Otherwise ntp frequency
2222 * correction applies to one vcpu's system_timestamp but not
2225 * So in those cases, request a kvmclock update for all vcpus.
2226 * We need to rate-limit these requests though, as they can
2227 * considerably slow guests that have a large number of vcpus.
2228 * The time for a remote vcpu to update its kvmclock is bound
2229 * by the delay we use to rate-limit the updates.
2232 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2234 static void kvmclock_update_fn(struct work_struct *work)
2237 struct delayed_work *dwork = to_delayed_work(work);
2238 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2239 kvmclock_update_work);
2240 struct kvm *kvm = container_of(ka, struct kvm, arch);
2241 struct kvm_vcpu *vcpu;
2243 kvm_for_each_vcpu(i, vcpu, kvm) {
2244 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2245 kvm_vcpu_kick(vcpu);
2249 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2251 struct kvm *kvm = v->kvm;
2253 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2254 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2255 KVMCLOCK_UPDATE_DELAY);
2258 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2260 static void kvmclock_sync_fn(struct work_struct *work)
2262 struct delayed_work *dwork = to_delayed_work(work);
2263 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2264 kvmclock_sync_work);
2265 struct kvm *kvm = container_of(ka, struct kvm, arch);
2267 if (!kvmclock_periodic_sync)
2270 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2271 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2272 KVMCLOCK_SYNC_PERIOD);
2275 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2277 u64 mcg_cap = vcpu->arch.mcg_cap;
2278 unsigned bank_num = mcg_cap & 0xff;
2279 u32 msr = msr_info->index;
2280 u64 data = msr_info->data;
2283 case MSR_IA32_MCG_STATUS:
2284 vcpu->arch.mcg_status = data;
2286 case MSR_IA32_MCG_CTL:
2287 if (!(mcg_cap & MCG_CTL_P) &&
2288 (data || !msr_info->host_initiated))
2290 if (data != 0 && data != ~(u64)0)
2292 vcpu->arch.mcg_ctl = data;
2295 if (msr >= MSR_IA32_MC0_CTL &&
2296 msr < MSR_IA32_MCx_CTL(bank_num)) {
2297 u32 offset = msr - MSR_IA32_MC0_CTL;
2298 /* only 0 or all 1s can be written to IA32_MCi_CTL
2299 * some Linux kernels though clear bit 10 in bank 4 to
2300 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2301 * this to avoid an uncatched #GP in the guest
2303 if ((offset & 0x3) == 0 &&
2304 data != 0 && (data | (1 << 10)) != ~(u64)0)
2306 if (!msr_info->host_initiated &&
2307 (offset & 0x3) == 1 && data != 0)
2309 vcpu->arch.mce_banks[offset] = data;
2317 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2319 struct kvm *kvm = vcpu->kvm;
2320 int lm = is_long_mode(vcpu);
2321 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2322 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2323 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2324 : kvm->arch.xen_hvm_config.blob_size_32;
2325 u32 page_num = data & ~PAGE_MASK;
2326 u64 page_addr = data & PAGE_MASK;
2331 if (page_num >= blob_size)
2334 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2339 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2348 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2350 gpa_t gpa = data & ~0x3f;
2352 /* Bits 3:5 are reserved, Should be zero */
2356 vcpu->arch.apf.msr_val = data;
2358 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2359 kvm_clear_async_pf_completion_queue(vcpu);
2360 kvm_async_pf_hash_reset(vcpu);
2364 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2368 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2369 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2370 kvm_async_pf_wakeup_all(vcpu);
2374 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2376 vcpu->arch.pv_time_enabled = false;
2379 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2381 ++vcpu->stat.tlb_flush;
2382 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2385 static void record_steal_time(struct kvm_vcpu *vcpu)
2387 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2390 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2391 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2395 * Doing a TLB flush here, on the guest's behalf, can avoid
2398 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2399 kvm_vcpu_flush_tlb(vcpu, false);
2401 if (vcpu->arch.st.steal.version & 1)
2402 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2404 vcpu->arch.st.steal.version += 1;
2406 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2407 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2411 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2412 vcpu->arch.st.last_steal;
2413 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2415 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2416 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2420 vcpu->arch.st.steal.version += 1;
2422 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2423 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2426 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2429 u32 msr = msr_info->index;
2430 u64 data = msr_info->data;
2433 case MSR_AMD64_NB_CFG:
2434 case MSR_IA32_UCODE_WRITE:
2435 case MSR_VM_HSAVE_PA:
2436 case MSR_AMD64_PATCH_LOADER:
2437 case MSR_AMD64_BU_CFG2:
2438 case MSR_AMD64_DC_CFG:
2439 case MSR_F15H_EX_CFG:
2442 case MSR_IA32_UCODE_REV:
2443 if (msr_info->host_initiated)
2444 vcpu->arch.microcode_version = data;
2447 return set_efer(vcpu, data);
2449 data &= ~(u64)0x40; /* ignore flush filter disable */
2450 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2451 data &= ~(u64)0x8; /* ignore TLB cache disable */
2452 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2454 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2459 case MSR_FAM10H_MMIO_CONF_BASE:
2461 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2466 case MSR_IA32_DEBUGCTLMSR:
2468 /* We support the non-activated case already */
2470 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2471 /* Values other than LBR and BTF are vendor-specific,
2472 thus reserved and should throw a #GP */
2475 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2478 case 0x200 ... 0x2ff:
2479 return kvm_mtrr_set_msr(vcpu, msr, data);
2480 case MSR_IA32_APICBASE:
2481 return kvm_set_apic_base(vcpu, msr_info);
2482 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2483 return kvm_x2apic_msr_write(vcpu, msr, data);
2484 case MSR_IA32_TSCDEADLINE:
2485 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2487 case MSR_IA32_TSC_ADJUST:
2488 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2489 if (!msr_info->host_initiated) {
2490 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2491 adjust_tsc_offset_guest(vcpu, adj);
2493 vcpu->arch.ia32_tsc_adjust_msr = data;
2496 case MSR_IA32_MISC_ENABLE:
2497 vcpu->arch.ia32_misc_enable_msr = data;
2499 case MSR_IA32_SMBASE:
2500 if (!msr_info->host_initiated)
2502 vcpu->arch.smbase = data;
2505 kvm_write_tsc(vcpu, msr_info);
2508 if (!msr_info->host_initiated)
2510 vcpu->arch.smi_count = data;
2512 case MSR_KVM_WALL_CLOCK_NEW:
2513 case MSR_KVM_WALL_CLOCK:
2514 vcpu->kvm->arch.wall_clock = data;
2515 kvm_write_wall_clock(vcpu->kvm, data);
2517 case MSR_KVM_SYSTEM_TIME_NEW:
2518 case MSR_KVM_SYSTEM_TIME: {
2519 struct kvm_arch *ka = &vcpu->kvm->arch;
2521 kvmclock_reset(vcpu);
2523 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2524 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2526 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2527 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2529 ka->boot_vcpu_runs_old_kvmclock = tmp;
2532 vcpu->arch.time = data;
2533 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2535 /* we verify if the enable bit is set... */
2539 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2540 &vcpu->arch.pv_time, data & ~1ULL,
2541 sizeof(struct pvclock_vcpu_time_info)))
2542 vcpu->arch.pv_time_enabled = false;
2544 vcpu->arch.pv_time_enabled = true;
2548 case MSR_KVM_ASYNC_PF_EN:
2549 if (kvm_pv_enable_async_pf(vcpu, data))
2552 case MSR_KVM_STEAL_TIME:
2554 if (unlikely(!sched_info_on()))
2557 if (data & KVM_STEAL_RESERVED_MASK)
2560 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2561 data & KVM_STEAL_VALID_BITS,
2562 sizeof(struct kvm_steal_time)))
2565 vcpu->arch.st.msr_val = data;
2567 if (!(data & KVM_MSR_ENABLED))
2570 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2573 case MSR_KVM_PV_EOI_EN:
2574 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2578 case MSR_IA32_MCG_CTL:
2579 case MSR_IA32_MCG_STATUS:
2580 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2581 return set_msr_mce(vcpu, msr_info);
2583 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2584 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2585 pr = true; /* fall through */
2586 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2587 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2588 if (kvm_pmu_is_valid_msr(vcpu, msr))
2589 return kvm_pmu_set_msr(vcpu, msr_info);
2591 if (pr || data != 0)
2592 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2593 "0x%x data 0x%llx\n", msr, data);
2595 case MSR_K7_CLK_CTL:
2597 * Ignore all writes to this no longer documented MSR.
2598 * Writes are only relevant for old K7 processors,
2599 * all pre-dating SVM, but a recommended workaround from
2600 * AMD for these chips. It is possible to specify the
2601 * affected processor models on the command line, hence
2602 * the need to ignore the workaround.
2605 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2606 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2607 case HV_X64_MSR_CRASH_CTL:
2608 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2609 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2610 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2611 case HV_X64_MSR_TSC_EMULATION_STATUS:
2612 return kvm_hv_set_msr_common(vcpu, msr, data,
2613 msr_info->host_initiated);
2614 case MSR_IA32_BBL_CR_CTL3:
2615 /* Drop writes to this legacy MSR -- see rdmsr
2616 * counterpart for further detail.
2618 if (report_ignored_msrs)
2619 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2622 case MSR_AMD64_OSVW_ID_LENGTH:
2623 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2625 vcpu->arch.osvw.length = data;
2627 case MSR_AMD64_OSVW_STATUS:
2628 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2630 vcpu->arch.osvw.status = data;
2632 case MSR_PLATFORM_INFO:
2633 if (!msr_info->host_initiated ||
2634 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2635 cpuid_fault_enabled(vcpu)))
2637 vcpu->arch.msr_platform_info = data;
2639 case MSR_MISC_FEATURES_ENABLES:
2640 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2641 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2642 !supports_cpuid_fault(vcpu)))
2644 vcpu->arch.msr_misc_features_enables = data;
2647 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2648 return xen_hvm_config(vcpu, data);
2649 if (kvm_pmu_is_valid_msr(vcpu, msr))
2650 return kvm_pmu_set_msr(vcpu, msr_info);
2652 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2656 if (report_ignored_msrs)
2658 "ignored wrmsr: 0x%x data 0x%llx\n",
2665 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2669 * Reads an msr value (of 'msr_index') into 'pdata'.
2670 * Returns 0 on success, non-0 otherwise.
2671 * Assumes vcpu_load() was already called.
2673 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2675 return kvm_x86_ops->get_msr(vcpu, msr);
2677 EXPORT_SYMBOL_GPL(kvm_get_msr);
2679 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2682 u64 mcg_cap = vcpu->arch.mcg_cap;
2683 unsigned bank_num = mcg_cap & 0xff;
2686 case MSR_IA32_P5_MC_ADDR:
2687 case MSR_IA32_P5_MC_TYPE:
2690 case MSR_IA32_MCG_CAP:
2691 data = vcpu->arch.mcg_cap;
2693 case MSR_IA32_MCG_CTL:
2694 if (!(mcg_cap & MCG_CTL_P) && !host)
2696 data = vcpu->arch.mcg_ctl;
2698 case MSR_IA32_MCG_STATUS:
2699 data = vcpu->arch.mcg_status;
2702 if (msr >= MSR_IA32_MC0_CTL &&
2703 msr < MSR_IA32_MCx_CTL(bank_num)) {
2704 u32 offset = msr - MSR_IA32_MC0_CTL;
2705 data = vcpu->arch.mce_banks[offset];
2714 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2716 switch (msr_info->index) {
2717 case MSR_IA32_PLATFORM_ID:
2718 case MSR_IA32_EBL_CR_POWERON:
2719 case MSR_IA32_DEBUGCTLMSR:
2720 case MSR_IA32_LASTBRANCHFROMIP:
2721 case MSR_IA32_LASTBRANCHTOIP:
2722 case MSR_IA32_LASTINTFROMIP:
2723 case MSR_IA32_LASTINTTOIP:
2725 case MSR_K8_TSEG_ADDR:
2726 case MSR_K8_TSEG_MASK:
2728 case MSR_VM_HSAVE_PA:
2729 case MSR_K8_INT_PENDING_MSG:
2730 case MSR_AMD64_NB_CFG:
2731 case MSR_FAM10H_MMIO_CONF_BASE:
2732 case MSR_AMD64_BU_CFG2:
2733 case MSR_IA32_PERF_CTL:
2734 case MSR_AMD64_DC_CFG:
2735 case MSR_F15H_EX_CFG:
2738 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2739 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2740 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2741 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2742 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2743 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2744 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2747 case MSR_IA32_UCODE_REV:
2748 msr_info->data = vcpu->arch.microcode_version;
2751 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2754 case 0x200 ... 0x2ff:
2755 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2756 case 0xcd: /* fsb frequency */
2760 * MSR_EBC_FREQUENCY_ID
2761 * Conservative value valid for even the basic CPU models.
2762 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2763 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2764 * and 266MHz for model 3, or 4. Set Core Clock
2765 * Frequency to System Bus Frequency Ratio to 1 (bits
2766 * 31:24) even though these are only valid for CPU
2767 * models > 2, however guests may end up dividing or
2768 * multiplying by zero otherwise.
2770 case MSR_EBC_FREQUENCY_ID:
2771 msr_info->data = 1 << 24;
2773 case MSR_IA32_APICBASE:
2774 msr_info->data = kvm_get_apic_base(vcpu);
2776 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2777 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2779 case MSR_IA32_TSCDEADLINE:
2780 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2782 case MSR_IA32_TSC_ADJUST:
2783 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2785 case MSR_IA32_MISC_ENABLE:
2786 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2788 case MSR_IA32_SMBASE:
2789 if (!msr_info->host_initiated)
2791 msr_info->data = vcpu->arch.smbase;
2794 msr_info->data = vcpu->arch.smi_count;
2796 case MSR_IA32_PERF_STATUS:
2797 /* TSC increment by tick */
2798 msr_info->data = 1000ULL;
2799 /* CPU multiplier */
2800 msr_info->data |= (((uint64_t)4ULL) << 40);
2803 msr_info->data = vcpu->arch.efer;
2805 case MSR_KVM_WALL_CLOCK:
2806 case MSR_KVM_WALL_CLOCK_NEW:
2807 msr_info->data = vcpu->kvm->arch.wall_clock;
2809 case MSR_KVM_SYSTEM_TIME:
2810 case MSR_KVM_SYSTEM_TIME_NEW:
2811 msr_info->data = vcpu->arch.time;
2813 case MSR_KVM_ASYNC_PF_EN:
2814 msr_info->data = vcpu->arch.apf.msr_val;
2816 case MSR_KVM_STEAL_TIME:
2817 msr_info->data = vcpu->arch.st.msr_val;
2819 case MSR_KVM_PV_EOI_EN:
2820 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2822 case MSR_IA32_P5_MC_ADDR:
2823 case MSR_IA32_P5_MC_TYPE:
2824 case MSR_IA32_MCG_CAP:
2825 case MSR_IA32_MCG_CTL:
2826 case MSR_IA32_MCG_STATUS:
2827 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2828 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2829 msr_info->host_initiated);
2830 case MSR_K7_CLK_CTL:
2832 * Provide expected ramp-up count for K7. All other
2833 * are set to zero, indicating minimum divisors for
2836 * This prevents guest kernels on AMD host with CPU
2837 * type 6, model 8 and higher from exploding due to
2838 * the rdmsr failing.
2840 msr_info->data = 0x20000000;
2842 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2843 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2844 case HV_X64_MSR_CRASH_CTL:
2845 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2846 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2847 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2848 case HV_X64_MSR_TSC_EMULATION_STATUS:
2849 return kvm_hv_get_msr_common(vcpu,
2850 msr_info->index, &msr_info->data,
2851 msr_info->host_initiated);
2853 case MSR_IA32_BBL_CR_CTL3:
2854 /* This legacy MSR exists but isn't fully documented in current
2855 * silicon. It is however accessed by winxp in very narrow
2856 * scenarios where it sets bit #19, itself documented as
2857 * a "reserved" bit. Best effort attempt to source coherent
2858 * read data here should the balance of the register be
2859 * interpreted by the guest:
2861 * L2 cache control register 3: 64GB range, 256KB size,
2862 * enabled, latency 0x1, configured
2864 msr_info->data = 0xbe702111;
2866 case MSR_AMD64_OSVW_ID_LENGTH:
2867 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2869 msr_info->data = vcpu->arch.osvw.length;
2871 case MSR_AMD64_OSVW_STATUS:
2872 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2874 msr_info->data = vcpu->arch.osvw.status;
2876 case MSR_PLATFORM_INFO:
2877 if (!msr_info->host_initiated &&
2878 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2880 msr_info->data = vcpu->arch.msr_platform_info;
2882 case MSR_MISC_FEATURES_ENABLES:
2883 msr_info->data = vcpu->arch.msr_misc_features_enables;
2886 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2887 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2889 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2893 if (report_ignored_msrs)
2894 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2902 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2905 * Read or write a bunch of msrs. All parameters are kernel addresses.
2907 * @return number of msrs set successfully.
2909 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2910 struct kvm_msr_entry *entries,
2911 int (*do_msr)(struct kvm_vcpu *vcpu,
2912 unsigned index, u64 *data))
2916 for (i = 0; i < msrs->nmsrs; ++i)
2917 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2924 * Read or write a bunch of msrs. Parameters are user addresses.
2926 * @return number of msrs set successfully.
2928 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2929 int (*do_msr)(struct kvm_vcpu *vcpu,
2930 unsigned index, u64 *data),
2933 struct kvm_msrs msrs;
2934 struct kvm_msr_entry *entries;
2939 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
2943 if (msrs.nmsrs >= MAX_IO_MSRS)
2946 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2947 entries = memdup_user(user_msrs->entries, size);
2948 if (IS_ERR(entries)) {
2949 r = PTR_ERR(entries);
2953 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2958 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2969 static inline bool kvm_can_mwait_in_guest(void)
2971 return boot_cpu_has(X86_FEATURE_MWAIT) &&
2972 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
2973 boot_cpu_has(X86_FEATURE_ARAT);
2976 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2981 case KVM_CAP_IRQCHIP:
2983 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2984 case KVM_CAP_SET_TSS_ADDR:
2985 case KVM_CAP_EXT_CPUID:
2986 case KVM_CAP_EXT_EMUL_CPUID:
2987 case KVM_CAP_CLOCKSOURCE:
2989 case KVM_CAP_NOP_IO_DELAY:
2990 case KVM_CAP_MP_STATE:
2991 case KVM_CAP_SYNC_MMU:
2992 case KVM_CAP_USER_NMI:
2993 case KVM_CAP_REINJECT_CONTROL:
2994 case KVM_CAP_IRQ_INJECT_STATUS:
2995 case KVM_CAP_IOEVENTFD:
2996 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2998 case KVM_CAP_PIT_STATE2:
2999 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3000 case KVM_CAP_XEN_HVM:
3001 case KVM_CAP_VCPU_EVENTS:
3002 case KVM_CAP_HYPERV:
3003 case KVM_CAP_HYPERV_VAPIC:
3004 case KVM_CAP_HYPERV_SPIN:
3005 case KVM_CAP_HYPERV_SYNIC:
3006 case KVM_CAP_HYPERV_SYNIC2:
3007 case KVM_CAP_HYPERV_VP_INDEX:
3008 case KVM_CAP_HYPERV_EVENTFD:
3009 case KVM_CAP_HYPERV_TLBFLUSH:
3010 case KVM_CAP_HYPERV_SEND_IPI:
3011 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3012 case KVM_CAP_HYPERV_CPUID:
3013 case KVM_CAP_PCI_SEGMENT:
3014 case KVM_CAP_DEBUGREGS:
3015 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3017 case KVM_CAP_ASYNC_PF:
3018 case KVM_CAP_GET_TSC_KHZ:
3019 case KVM_CAP_KVMCLOCK_CTRL:
3020 case KVM_CAP_READONLY_MEM:
3021 case KVM_CAP_HYPERV_TIME:
3022 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3023 case KVM_CAP_TSC_DEADLINE_TIMER:
3024 case KVM_CAP_DISABLE_QUIRKS:
3025 case KVM_CAP_SET_BOOT_CPU_ID:
3026 case KVM_CAP_SPLIT_IRQCHIP:
3027 case KVM_CAP_IMMEDIATE_EXIT:
3028 case KVM_CAP_GET_MSR_FEATURES:
3029 case KVM_CAP_MSR_PLATFORM_INFO:
3030 case KVM_CAP_EXCEPTION_PAYLOAD:
3033 case KVM_CAP_SYNC_REGS:
3034 r = KVM_SYNC_X86_VALID_FIELDS;
3036 case KVM_CAP_ADJUST_CLOCK:
3037 r = KVM_CLOCK_TSC_STABLE;
3039 case KVM_CAP_X86_DISABLE_EXITS:
3040 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE;
3041 if(kvm_can_mwait_in_guest())
3042 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3044 case KVM_CAP_X86_SMM:
3045 /* SMBASE is usually relocated above 1M on modern chipsets,
3046 * and SMM handlers might indeed rely on 4G segment limits,
3047 * so do not report SMM to be available if real mode is
3048 * emulated via vm86 mode. Still, do not go to great lengths
3049 * to avoid userspace's usage of the feature, because it is a
3050 * fringe case that is not enabled except via specific settings
3051 * of the module parameters.
3053 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3056 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3058 case KVM_CAP_NR_VCPUS:
3059 r = KVM_SOFT_MAX_VCPUS;
3061 case KVM_CAP_MAX_VCPUS:
3064 case KVM_CAP_NR_MEMSLOTS:
3065 r = KVM_USER_MEM_SLOTS;
3067 case KVM_CAP_PV_MMU: /* obsolete */
3071 r = KVM_MAX_MCE_BANKS;
3074 r = boot_cpu_has(X86_FEATURE_XSAVE);
3076 case KVM_CAP_TSC_CONTROL:
3077 r = kvm_has_tsc_control;
3079 case KVM_CAP_X2APIC_API:
3080 r = KVM_X2APIC_API_VALID_FLAGS;
3082 case KVM_CAP_NESTED_STATE:
3083 r = kvm_x86_ops->get_nested_state ?
3084 kvm_x86_ops->get_nested_state(NULL, 0, 0) : 0;
3093 long kvm_arch_dev_ioctl(struct file *filp,
3094 unsigned int ioctl, unsigned long arg)
3096 void __user *argp = (void __user *)arg;
3100 case KVM_GET_MSR_INDEX_LIST: {
3101 struct kvm_msr_list __user *user_msr_list = argp;
3102 struct kvm_msr_list msr_list;
3106 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3109 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3110 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3113 if (n < msr_list.nmsrs)
3116 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3117 num_msrs_to_save * sizeof(u32)))
3119 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3121 num_emulated_msrs * sizeof(u32)))
3126 case KVM_GET_SUPPORTED_CPUID:
3127 case KVM_GET_EMULATED_CPUID: {
3128 struct kvm_cpuid2 __user *cpuid_arg = argp;
3129 struct kvm_cpuid2 cpuid;
3132 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3135 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3141 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3146 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3148 if (copy_to_user(argp, &kvm_mce_cap_supported,
3149 sizeof(kvm_mce_cap_supported)))
3153 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3154 struct kvm_msr_list __user *user_msr_list = argp;
3155 struct kvm_msr_list msr_list;
3159 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3162 msr_list.nmsrs = num_msr_based_features;
3163 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3166 if (n < msr_list.nmsrs)
3169 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3170 num_msr_based_features * sizeof(u32)))
3176 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3186 static void wbinvd_ipi(void *garbage)
3191 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3193 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3196 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3198 /* Address WBINVD may be executed by guest */
3199 if (need_emulate_wbinvd(vcpu)) {
3200 if (kvm_x86_ops->has_wbinvd_exit())
3201 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3202 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3203 smp_call_function_single(vcpu->cpu,
3204 wbinvd_ipi, NULL, 1);
3207 kvm_x86_ops->vcpu_load(vcpu, cpu);
3209 /* Apply any externally detected TSC adjustments (due to suspend) */
3210 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3211 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3212 vcpu->arch.tsc_offset_adjustment = 0;
3213 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3216 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3217 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3218 rdtsc() - vcpu->arch.last_host_tsc;
3220 mark_tsc_unstable("KVM discovered backwards TSC");
3222 if (kvm_check_tsc_unstable()) {
3223 u64 offset = kvm_compute_tsc_offset(vcpu,
3224 vcpu->arch.last_guest_tsc);
3225 kvm_vcpu_write_tsc_offset(vcpu, offset);
3226 vcpu->arch.tsc_catchup = 1;
3229 if (kvm_lapic_hv_timer_in_use(vcpu))
3230 kvm_lapic_restart_hv_timer(vcpu);
3233 * On a host with synchronized TSC, there is no need to update
3234 * kvmclock on vcpu->cpu migration
3236 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3237 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3238 if (vcpu->cpu != cpu)
3239 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3243 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3246 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3248 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3251 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3253 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3254 &vcpu->arch.st.steal.preempted,
3255 offsetof(struct kvm_steal_time, preempted),
3256 sizeof(vcpu->arch.st.steal.preempted));
3259 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3263 if (vcpu->preempted)
3264 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3267 * Disable page faults because we're in atomic context here.
3268 * kvm_write_guest_offset_cached() would call might_fault()
3269 * that relies on pagefault_disable() to tell if there's a
3270 * bug. NOTE: the write to guest memory may not go through if
3271 * during postcopy live migration or if there's heavy guest
3274 pagefault_disable();
3276 * kvm_memslots() will be called by
3277 * kvm_write_guest_offset_cached() so take the srcu lock.
3279 idx = srcu_read_lock(&vcpu->kvm->srcu);
3280 kvm_steal_time_set_preempted(vcpu);
3281 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3283 kvm_x86_ops->vcpu_put(vcpu);
3284 vcpu->arch.last_host_tsc = rdtsc();
3286 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3287 * on every vmexit, but if not, we might have a stale dr6 from the
3288 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3293 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3294 struct kvm_lapic_state *s)
3296 if (vcpu->arch.apicv_active)
3297 kvm_x86_ops->sync_pir_to_irr(vcpu);
3299 return kvm_apic_get_state(vcpu, s);
3302 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3303 struct kvm_lapic_state *s)
3307 r = kvm_apic_set_state(vcpu, s);
3310 update_cr8_intercept(vcpu);
3315 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3317 return (!lapic_in_kernel(vcpu) ||
3318 kvm_apic_accept_pic_intr(vcpu));
3322 * if userspace requested an interrupt window, check that the
3323 * interrupt window is open.
3325 * No need to exit to userspace if we already have an interrupt queued.
3327 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3329 return kvm_arch_interrupt_allowed(vcpu) &&
3330 !kvm_cpu_has_interrupt(vcpu) &&
3331 !kvm_event_needs_reinjection(vcpu) &&
3332 kvm_cpu_accept_dm_intr(vcpu);
3335 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3336 struct kvm_interrupt *irq)
3338 if (irq->irq >= KVM_NR_INTERRUPTS)
3341 if (!irqchip_in_kernel(vcpu->kvm)) {
3342 kvm_queue_interrupt(vcpu, irq->irq, false);
3343 kvm_make_request(KVM_REQ_EVENT, vcpu);
3348 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3349 * fail for in-kernel 8259.
3351 if (pic_in_kernel(vcpu->kvm))
3354 if (vcpu->arch.pending_external_vector != -1)
3357 vcpu->arch.pending_external_vector = irq->irq;
3358 kvm_make_request(KVM_REQ_EVENT, vcpu);
3362 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3364 kvm_inject_nmi(vcpu);
3369 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3371 kvm_make_request(KVM_REQ_SMI, vcpu);
3376 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3377 struct kvm_tpr_access_ctl *tac)
3381 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3385 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3389 unsigned bank_num = mcg_cap & 0xff, bank;
3392 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3394 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3397 vcpu->arch.mcg_cap = mcg_cap;
3398 /* Init IA32_MCG_CTL to all 1s */
3399 if (mcg_cap & MCG_CTL_P)
3400 vcpu->arch.mcg_ctl = ~(u64)0;
3401 /* Init IA32_MCi_CTL to all 1s */
3402 for (bank = 0; bank < bank_num; bank++)
3403 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3405 if (kvm_x86_ops->setup_mce)
3406 kvm_x86_ops->setup_mce(vcpu);
3411 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3412 struct kvm_x86_mce *mce)
3414 u64 mcg_cap = vcpu->arch.mcg_cap;
3415 unsigned bank_num = mcg_cap & 0xff;
3416 u64 *banks = vcpu->arch.mce_banks;
3418 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3421 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3422 * reporting is disabled
3424 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3425 vcpu->arch.mcg_ctl != ~(u64)0)
3427 banks += 4 * mce->bank;
3429 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3430 * reporting is disabled for the bank
3432 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3434 if (mce->status & MCI_STATUS_UC) {
3435 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3436 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3437 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3440 if (banks[1] & MCI_STATUS_VAL)
3441 mce->status |= MCI_STATUS_OVER;
3442 banks[2] = mce->addr;
3443 banks[3] = mce->misc;
3444 vcpu->arch.mcg_status = mce->mcg_status;
3445 banks[1] = mce->status;
3446 kvm_queue_exception(vcpu, MC_VECTOR);
3447 } else if (!(banks[1] & MCI_STATUS_VAL)
3448 || !(banks[1] & MCI_STATUS_UC)) {
3449 if (banks[1] & MCI_STATUS_VAL)
3450 mce->status |= MCI_STATUS_OVER;
3451 banks[2] = mce->addr;
3452 banks[3] = mce->misc;
3453 banks[1] = mce->status;
3455 banks[1] |= MCI_STATUS_OVER;
3459 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3460 struct kvm_vcpu_events *events)
3465 * The API doesn't provide the instruction length for software
3466 * exceptions, so don't report them. As long as the guest RIP
3467 * isn't advanced, we should expect to encounter the exception
3470 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3471 events->exception.injected = 0;
3472 events->exception.pending = 0;
3474 events->exception.injected = vcpu->arch.exception.injected;
3475 events->exception.pending = vcpu->arch.exception.pending;
3477 * For ABI compatibility, deliberately conflate
3478 * pending and injected exceptions when
3479 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3481 if (!vcpu->kvm->arch.exception_payload_enabled)
3482 events->exception.injected |=
3483 vcpu->arch.exception.pending;
3485 events->exception.nr = vcpu->arch.exception.nr;
3486 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3487 events->exception.error_code = vcpu->arch.exception.error_code;
3488 events->exception_has_payload = vcpu->arch.exception.has_payload;
3489 events->exception_payload = vcpu->arch.exception.payload;
3491 events->interrupt.injected =
3492 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3493 events->interrupt.nr = vcpu->arch.interrupt.nr;
3494 events->interrupt.soft = 0;
3495 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3497 events->nmi.injected = vcpu->arch.nmi_injected;
3498 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3499 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3500 events->nmi.pad = 0;
3502 events->sipi_vector = 0; /* never valid when reporting to user space */
3504 events->smi.smm = is_smm(vcpu);
3505 events->smi.pending = vcpu->arch.smi_pending;
3506 events->smi.smm_inside_nmi =
3507 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3508 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3510 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3511 | KVM_VCPUEVENT_VALID_SHADOW
3512 | KVM_VCPUEVENT_VALID_SMM);
3513 if (vcpu->kvm->arch.exception_payload_enabled)
3514 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3516 memset(&events->reserved, 0, sizeof(events->reserved));
3519 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3521 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3522 struct kvm_vcpu_events *events)
3524 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3525 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3526 | KVM_VCPUEVENT_VALID_SHADOW
3527 | KVM_VCPUEVENT_VALID_SMM
3528 | KVM_VCPUEVENT_VALID_PAYLOAD))
3531 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3532 if (!vcpu->kvm->arch.exception_payload_enabled)
3534 if (events->exception.pending)
3535 events->exception.injected = 0;
3537 events->exception_has_payload = 0;
3539 events->exception.pending = 0;
3540 events->exception_has_payload = 0;
3543 if ((events->exception.injected || events->exception.pending) &&
3544 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3547 /* INITs are latched while in SMM */
3548 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3549 (events->smi.smm || events->smi.pending) &&
3550 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3554 vcpu->arch.exception.injected = events->exception.injected;
3555 vcpu->arch.exception.pending = events->exception.pending;
3556 vcpu->arch.exception.nr = events->exception.nr;
3557 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3558 vcpu->arch.exception.error_code = events->exception.error_code;
3559 vcpu->arch.exception.has_payload = events->exception_has_payload;
3560 vcpu->arch.exception.payload = events->exception_payload;
3562 vcpu->arch.interrupt.injected = events->interrupt.injected;
3563 vcpu->arch.interrupt.nr = events->interrupt.nr;
3564 vcpu->arch.interrupt.soft = events->interrupt.soft;
3565 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3566 kvm_x86_ops->set_interrupt_shadow(vcpu,
3567 events->interrupt.shadow);
3569 vcpu->arch.nmi_injected = events->nmi.injected;
3570 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3571 vcpu->arch.nmi_pending = events->nmi.pending;
3572 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3574 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3575 lapic_in_kernel(vcpu))
3576 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3578 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3579 u32 hflags = vcpu->arch.hflags;
3580 if (events->smi.smm)
3581 hflags |= HF_SMM_MASK;
3583 hflags &= ~HF_SMM_MASK;
3584 kvm_set_hflags(vcpu, hflags);
3586 vcpu->arch.smi_pending = events->smi.pending;
3588 if (events->smi.smm) {
3589 if (events->smi.smm_inside_nmi)
3590 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3592 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3593 if (lapic_in_kernel(vcpu)) {
3594 if (events->smi.latched_init)
3595 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3597 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3602 kvm_make_request(KVM_REQ_EVENT, vcpu);
3607 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3608 struct kvm_debugregs *dbgregs)
3612 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3613 kvm_get_dr(vcpu, 6, &val);
3615 dbgregs->dr7 = vcpu->arch.dr7;
3617 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3620 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3621 struct kvm_debugregs *dbgregs)
3626 if (dbgregs->dr6 & ~0xffffffffull)
3628 if (dbgregs->dr7 & ~0xffffffffull)
3631 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3632 kvm_update_dr0123(vcpu);
3633 vcpu->arch.dr6 = dbgregs->dr6;
3634 kvm_update_dr6(vcpu);
3635 vcpu->arch.dr7 = dbgregs->dr7;
3636 kvm_update_dr7(vcpu);
3641 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3643 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3645 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3646 u64 xstate_bv = xsave->header.xfeatures;
3650 * Copy legacy XSAVE area, to avoid complications with CPUID
3651 * leaves 0 and 1 in the loop below.
3653 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3656 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3657 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3660 * Copy each region from the possibly compacted offset to the
3661 * non-compacted offset.
3663 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3665 u64 feature = valid & -valid;
3666 int index = fls64(feature) - 1;
3667 void *src = get_xsave_addr(xsave, feature);
3670 u32 size, offset, ecx, edx;
3671 cpuid_count(XSTATE_CPUID, index,
3672 &size, &offset, &ecx, &edx);
3673 if (feature == XFEATURE_MASK_PKRU)
3674 memcpy(dest + offset, &vcpu->arch.pkru,
3675 sizeof(vcpu->arch.pkru));
3677 memcpy(dest + offset, src, size);
3685 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3687 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3688 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3692 * Copy legacy XSAVE area, to avoid complications with CPUID
3693 * leaves 0 and 1 in the loop below.
3695 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3697 /* Set XSTATE_BV and possibly XCOMP_BV. */
3698 xsave->header.xfeatures = xstate_bv;
3699 if (boot_cpu_has(X86_FEATURE_XSAVES))
3700 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3703 * Copy each region from the non-compacted offset to the
3704 * possibly compacted offset.
3706 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3708 u64 feature = valid & -valid;
3709 int index = fls64(feature) - 1;
3710 void *dest = get_xsave_addr(xsave, feature);
3713 u32 size, offset, ecx, edx;
3714 cpuid_count(XSTATE_CPUID, index,
3715 &size, &offset, &ecx, &edx);
3716 if (feature == XFEATURE_MASK_PKRU)
3717 memcpy(&vcpu->arch.pkru, src + offset,
3718 sizeof(vcpu->arch.pkru));
3720 memcpy(dest, src + offset, size);
3727 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3728 struct kvm_xsave *guest_xsave)
3730 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3731 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3732 fill_xsave((u8 *) guest_xsave->region, vcpu);
3734 memcpy(guest_xsave->region,
3735 &vcpu->arch.guest_fpu->state.fxsave,
3736 sizeof(struct fxregs_state));
3737 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3738 XFEATURE_MASK_FPSSE;
3742 #define XSAVE_MXCSR_OFFSET 24
3744 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3745 struct kvm_xsave *guest_xsave)
3748 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3749 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3751 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3753 * Here we allow setting states that are not present in
3754 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3755 * with old userspace.
3757 if (xstate_bv & ~kvm_supported_xcr0() ||
3758 mxcsr & ~mxcsr_feature_mask)
3760 load_xsave(vcpu, (u8 *)guest_xsave->region);
3762 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3763 mxcsr & ~mxcsr_feature_mask)
3765 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3766 guest_xsave->region, sizeof(struct fxregs_state));
3771 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3772 struct kvm_xcrs *guest_xcrs)
3774 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3775 guest_xcrs->nr_xcrs = 0;
3779 guest_xcrs->nr_xcrs = 1;
3780 guest_xcrs->flags = 0;
3781 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3782 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3785 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3786 struct kvm_xcrs *guest_xcrs)
3790 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3793 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3796 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3797 /* Only support XCR0 currently */
3798 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3799 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3800 guest_xcrs->xcrs[i].value);
3809 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3810 * stopped by the hypervisor. This function will be called from the host only.
3811 * EINVAL is returned when the host attempts to set the flag for a guest that
3812 * does not support pv clocks.
3814 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3816 if (!vcpu->arch.pv_time_enabled)
3818 vcpu->arch.pvclock_set_guest_stopped_request = true;
3819 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3823 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3824 struct kvm_enable_cap *cap)
3827 uint16_t vmcs_version;
3828 void __user *user_ptr;
3834 case KVM_CAP_HYPERV_SYNIC2:
3839 case KVM_CAP_HYPERV_SYNIC:
3840 if (!irqchip_in_kernel(vcpu->kvm))
3842 return kvm_hv_activate_synic(vcpu, cap->cap ==
3843 KVM_CAP_HYPERV_SYNIC2);
3844 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3845 if (!kvm_x86_ops->nested_enable_evmcs)
3847 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3849 user_ptr = (void __user *)(uintptr_t)cap->args[0];
3850 if (copy_to_user(user_ptr, &vmcs_version,
3851 sizeof(vmcs_version)))
3861 long kvm_arch_vcpu_ioctl(struct file *filp,
3862 unsigned int ioctl, unsigned long arg)
3864 struct kvm_vcpu *vcpu = filp->private_data;
3865 void __user *argp = (void __user *)arg;
3868 struct kvm_lapic_state *lapic;
3869 struct kvm_xsave *xsave;
3870 struct kvm_xcrs *xcrs;
3878 case KVM_GET_LAPIC: {
3880 if (!lapic_in_kernel(vcpu))
3882 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3887 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3891 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3896 case KVM_SET_LAPIC: {
3898 if (!lapic_in_kernel(vcpu))
3900 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3901 if (IS_ERR(u.lapic)) {
3902 r = PTR_ERR(u.lapic);
3906 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3909 case KVM_INTERRUPT: {
3910 struct kvm_interrupt irq;
3913 if (copy_from_user(&irq, argp, sizeof(irq)))
3915 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3919 r = kvm_vcpu_ioctl_nmi(vcpu);
3923 r = kvm_vcpu_ioctl_smi(vcpu);
3926 case KVM_SET_CPUID: {
3927 struct kvm_cpuid __user *cpuid_arg = argp;
3928 struct kvm_cpuid cpuid;
3931 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3933 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3936 case KVM_SET_CPUID2: {
3937 struct kvm_cpuid2 __user *cpuid_arg = argp;
3938 struct kvm_cpuid2 cpuid;
3941 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3943 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3944 cpuid_arg->entries);
3947 case KVM_GET_CPUID2: {
3948 struct kvm_cpuid2 __user *cpuid_arg = argp;
3949 struct kvm_cpuid2 cpuid;
3952 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3954 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3955 cpuid_arg->entries);
3959 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3964 case KVM_GET_MSRS: {
3965 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3966 r = msr_io(vcpu, argp, do_get_msr, 1);
3967 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3970 case KVM_SET_MSRS: {
3971 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3972 r = msr_io(vcpu, argp, do_set_msr, 0);
3973 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3976 case KVM_TPR_ACCESS_REPORTING: {
3977 struct kvm_tpr_access_ctl tac;
3980 if (copy_from_user(&tac, argp, sizeof(tac)))
3982 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3986 if (copy_to_user(argp, &tac, sizeof(tac)))
3991 case KVM_SET_VAPIC_ADDR: {
3992 struct kvm_vapic_addr va;
3996 if (!lapic_in_kernel(vcpu))
3999 if (copy_from_user(&va, argp, sizeof(va)))
4001 idx = srcu_read_lock(&vcpu->kvm->srcu);
4002 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4003 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4006 case KVM_X86_SETUP_MCE: {
4010 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4012 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4015 case KVM_X86_SET_MCE: {
4016 struct kvm_x86_mce mce;
4019 if (copy_from_user(&mce, argp, sizeof(mce)))
4021 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4024 case KVM_GET_VCPU_EVENTS: {
4025 struct kvm_vcpu_events events;
4027 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4030 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4035 case KVM_SET_VCPU_EVENTS: {
4036 struct kvm_vcpu_events events;
4039 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4042 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4045 case KVM_GET_DEBUGREGS: {
4046 struct kvm_debugregs dbgregs;
4048 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4051 if (copy_to_user(argp, &dbgregs,
4052 sizeof(struct kvm_debugregs)))
4057 case KVM_SET_DEBUGREGS: {
4058 struct kvm_debugregs dbgregs;
4061 if (copy_from_user(&dbgregs, argp,
4062 sizeof(struct kvm_debugregs)))
4065 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4068 case KVM_GET_XSAVE: {
4069 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
4074 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4077 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4082 case KVM_SET_XSAVE: {
4083 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4084 if (IS_ERR(u.xsave)) {
4085 r = PTR_ERR(u.xsave);
4089 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4092 case KVM_GET_XCRS: {
4093 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
4098 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4101 if (copy_to_user(argp, u.xcrs,
4102 sizeof(struct kvm_xcrs)))
4107 case KVM_SET_XCRS: {
4108 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4109 if (IS_ERR(u.xcrs)) {
4110 r = PTR_ERR(u.xcrs);
4114 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4117 case KVM_SET_TSC_KHZ: {
4121 user_tsc_khz = (u32)arg;
4123 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4126 if (user_tsc_khz == 0)
4127 user_tsc_khz = tsc_khz;
4129 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4134 case KVM_GET_TSC_KHZ: {
4135 r = vcpu->arch.virtual_tsc_khz;
4138 case KVM_KVMCLOCK_CTRL: {
4139 r = kvm_set_guest_paused(vcpu);
4142 case KVM_ENABLE_CAP: {
4143 struct kvm_enable_cap cap;
4146 if (copy_from_user(&cap, argp, sizeof(cap)))
4148 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4151 case KVM_GET_NESTED_STATE: {
4152 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4156 if (!kvm_x86_ops->get_nested_state)
4159 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4161 if (get_user(user_data_size, &user_kvm_nested_state->size))
4164 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4169 if (r > user_data_size) {
4170 if (put_user(r, &user_kvm_nested_state->size))
4180 case KVM_SET_NESTED_STATE: {
4181 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4182 struct kvm_nested_state kvm_state;
4185 if (!kvm_x86_ops->set_nested_state)
4189 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4193 if (kvm_state.size < sizeof(kvm_state))
4196 if (kvm_state.flags &
4197 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4198 | KVM_STATE_NESTED_EVMCS))
4201 /* nested_run_pending implies guest_mode. */
4202 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4203 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4206 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4209 case KVM_GET_SUPPORTED_HV_CPUID: {
4210 struct kvm_cpuid2 __user *cpuid_arg = argp;
4211 struct kvm_cpuid2 cpuid;
4214 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4217 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4218 cpuid_arg->entries);
4223 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4238 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4240 return VM_FAULT_SIGBUS;
4243 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4247 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4249 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4253 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4256 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4259 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4260 u32 kvm_nr_mmu_pages)
4262 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4265 mutex_lock(&kvm->slots_lock);
4267 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4268 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4270 mutex_unlock(&kvm->slots_lock);
4274 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4276 return kvm->arch.n_max_mmu_pages;
4279 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4281 struct kvm_pic *pic = kvm->arch.vpic;
4285 switch (chip->chip_id) {
4286 case KVM_IRQCHIP_PIC_MASTER:
4287 memcpy(&chip->chip.pic, &pic->pics[0],
4288 sizeof(struct kvm_pic_state));
4290 case KVM_IRQCHIP_PIC_SLAVE:
4291 memcpy(&chip->chip.pic, &pic->pics[1],
4292 sizeof(struct kvm_pic_state));
4294 case KVM_IRQCHIP_IOAPIC:
4295 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4304 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4306 struct kvm_pic *pic = kvm->arch.vpic;
4310 switch (chip->chip_id) {
4311 case KVM_IRQCHIP_PIC_MASTER:
4312 spin_lock(&pic->lock);
4313 memcpy(&pic->pics[0], &chip->chip.pic,
4314 sizeof(struct kvm_pic_state));
4315 spin_unlock(&pic->lock);
4317 case KVM_IRQCHIP_PIC_SLAVE:
4318 spin_lock(&pic->lock);
4319 memcpy(&pic->pics[1], &chip->chip.pic,
4320 sizeof(struct kvm_pic_state));
4321 spin_unlock(&pic->lock);
4323 case KVM_IRQCHIP_IOAPIC:
4324 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4330 kvm_pic_update_irq(pic);
4334 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4336 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4338 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4340 mutex_lock(&kps->lock);
4341 memcpy(ps, &kps->channels, sizeof(*ps));
4342 mutex_unlock(&kps->lock);
4346 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4349 struct kvm_pit *pit = kvm->arch.vpit;
4351 mutex_lock(&pit->pit_state.lock);
4352 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4353 for (i = 0; i < 3; i++)
4354 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4355 mutex_unlock(&pit->pit_state.lock);
4359 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4361 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4362 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4363 sizeof(ps->channels));
4364 ps->flags = kvm->arch.vpit->pit_state.flags;
4365 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4366 memset(&ps->reserved, 0, sizeof(ps->reserved));
4370 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4374 u32 prev_legacy, cur_legacy;
4375 struct kvm_pit *pit = kvm->arch.vpit;
4377 mutex_lock(&pit->pit_state.lock);
4378 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4379 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4380 if (!prev_legacy && cur_legacy)
4382 memcpy(&pit->pit_state.channels, &ps->channels,
4383 sizeof(pit->pit_state.channels));
4384 pit->pit_state.flags = ps->flags;
4385 for (i = 0; i < 3; i++)
4386 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4388 mutex_unlock(&pit->pit_state.lock);
4392 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4393 struct kvm_reinject_control *control)
4395 struct kvm_pit *pit = kvm->arch.vpit;
4400 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4401 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4402 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4404 mutex_lock(&pit->pit_state.lock);
4405 kvm_pit_set_reinject(pit, control->pit_reinject);
4406 mutex_unlock(&pit->pit_state.lock);
4412 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4413 * @kvm: kvm instance
4414 * @log: slot id and address to which we copy the log
4416 * Steps 1-4 below provide general overview of dirty page logging. See
4417 * kvm_get_dirty_log_protect() function description for additional details.
4419 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4420 * always flush the TLB (step 4) even if previous step failed and the dirty
4421 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4422 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4423 * writes will be marked dirty for next log read.
4425 * 1. Take a snapshot of the bit and clear it if needed.
4426 * 2. Write protect the corresponding page.
4427 * 3. Copy the snapshot to the userspace.
4428 * 4. Flush TLB's if needed.
4430 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4435 mutex_lock(&kvm->slots_lock);
4438 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4440 if (kvm_x86_ops->flush_log_dirty)
4441 kvm_x86_ops->flush_log_dirty(kvm);
4443 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4446 * All the TLBs can be flushed out of mmu lock, see the comments in
4447 * kvm_mmu_slot_remove_write_access().
4449 lockdep_assert_held(&kvm->slots_lock);
4451 kvm_flush_remote_tlbs(kvm);
4453 mutex_unlock(&kvm->slots_lock);
4457 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4462 mutex_lock(&kvm->slots_lock);
4465 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4467 if (kvm_x86_ops->flush_log_dirty)
4468 kvm_x86_ops->flush_log_dirty(kvm);
4470 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4473 * All the TLBs can be flushed out of mmu lock, see the comments in
4474 * kvm_mmu_slot_remove_write_access().
4476 lockdep_assert_held(&kvm->slots_lock);
4478 kvm_flush_remote_tlbs(kvm);
4480 mutex_unlock(&kvm->slots_lock);
4484 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4487 if (!irqchip_in_kernel(kvm))
4490 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4491 irq_event->irq, irq_event->level,
4496 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4497 struct kvm_enable_cap *cap)
4505 case KVM_CAP_DISABLE_QUIRKS:
4506 kvm->arch.disabled_quirks = cap->args[0];
4509 case KVM_CAP_SPLIT_IRQCHIP: {
4510 mutex_lock(&kvm->lock);
4512 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4513 goto split_irqchip_unlock;
4515 if (irqchip_in_kernel(kvm))
4516 goto split_irqchip_unlock;
4517 if (kvm->created_vcpus)
4518 goto split_irqchip_unlock;
4519 r = kvm_setup_empty_irq_routing(kvm);
4521 goto split_irqchip_unlock;
4522 /* Pairs with irqchip_in_kernel. */
4524 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4525 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4527 split_irqchip_unlock:
4528 mutex_unlock(&kvm->lock);
4531 case KVM_CAP_X2APIC_API:
4533 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4536 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4537 kvm->arch.x2apic_format = true;
4538 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4539 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4543 case KVM_CAP_X86_DISABLE_EXITS:
4545 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4548 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4549 kvm_can_mwait_in_guest())
4550 kvm->arch.mwait_in_guest = true;
4551 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4552 kvm->arch.hlt_in_guest = true;
4553 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4554 kvm->arch.pause_in_guest = true;
4557 case KVM_CAP_MSR_PLATFORM_INFO:
4558 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4561 case KVM_CAP_EXCEPTION_PAYLOAD:
4562 kvm->arch.exception_payload_enabled = cap->args[0];
4572 long kvm_arch_vm_ioctl(struct file *filp,
4573 unsigned int ioctl, unsigned long arg)
4575 struct kvm *kvm = filp->private_data;
4576 void __user *argp = (void __user *)arg;
4579 * This union makes it completely explicit to gcc-3.x
4580 * that these two variables' stack usage should be
4581 * combined, not added together.
4584 struct kvm_pit_state ps;
4585 struct kvm_pit_state2 ps2;
4586 struct kvm_pit_config pit_config;
4590 case KVM_SET_TSS_ADDR:
4591 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4593 case KVM_SET_IDENTITY_MAP_ADDR: {
4596 mutex_lock(&kvm->lock);
4598 if (kvm->created_vcpus)
4599 goto set_identity_unlock;
4601 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4602 goto set_identity_unlock;
4603 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4604 set_identity_unlock:
4605 mutex_unlock(&kvm->lock);
4608 case KVM_SET_NR_MMU_PAGES:
4609 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4611 case KVM_GET_NR_MMU_PAGES:
4612 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4614 case KVM_CREATE_IRQCHIP: {
4615 mutex_lock(&kvm->lock);
4618 if (irqchip_in_kernel(kvm))
4619 goto create_irqchip_unlock;
4622 if (kvm->created_vcpus)
4623 goto create_irqchip_unlock;
4625 r = kvm_pic_init(kvm);
4627 goto create_irqchip_unlock;
4629 r = kvm_ioapic_init(kvm);
4631 kvm_pic_destroy(kvm);
4632 goto create_irqchip_unlock;
4635 r = kvm_setup_default_irq_routing(kvm);
4637 kvm_ioapic_destroy(kvm);
4638 kvm_pic_destroy(kvm);
4639 goto create_irqchip_unlock;
4641 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4643 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4644 create_irqchip_unlock:
4645 mutex_unlock(&kvm->lock);
4648 case KVM_CREATE_PIT:
4649 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4651 case KVM_CREATE_PIT2:
4653 if (copy_from_user(&u.pit_config, argp,
4654 sizeof(struct kvm_pit_config)))
4657 mutex_lock(&kvm->lock);
4660 goto create_pit_unlock;
4662 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4666 mutex_unlock(&kvm->lock);
4668 case KVM_GET_IRQCHIP: {
4669 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4670 struct kvm_irqchip *chip;
4672 chip = memdup_user(argp, sizeof(*chip));
4679 if (!irqchip_kernel(kvm))
4680 goto get_irqchip_out;
4681 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4683 goto get_irqchip_out;
4685 if (copy_to_user(argp, chip, sizeof(*chip)))
4686 goto get_irqchip_out;
4692 case KVM_SET_IRQCHIP: {
4693 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4694 struct kvm_irqchip *chip;
4696 chip = memdup_user(argp, sizeof(*chip));
4703 if (!irqchip_kernel(kvm))
4704 goto set_irqchip_out;
4705 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4707 goto set_irqchip_out;
4715 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4718 if (!kvm->arch.vpit)
4720 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4724 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4731 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4734 if (!kvm->arch.vpit)
4736 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4739 case KVM_GET_PIT2: {
4741 if (!kvm->arch.vpit)
4743 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4747 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4752 case KVM_SET_PIT2: {
4754 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4757 if (!kvm->arch.vpit)
4759 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4762 case KVM_REINJECT_CONTROL: {
4763 struct kvm_reinject_control control;
4765 if (copy_from_user(&control, argp, sizeof(control)))
4767 r = kvm_vm_ioctl_reinject(kvm, &control);
4770 case KVM_SET_BOOT_CPU_ID:
4772 mutex_lock(&kvm->lock);
4773 if (kvm->created_vcpus)
4776 kvm->arch.bsp_vcpu_id = arg;
4777 mutex_unlock(&kvm->lock);
4779 case KVM_XEN_HVM_CONFIG: {
4780 struct kvm_xen_hvm_config xhc;
4782 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4787 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4791 case KVM_SET_CLOCK: {
4792 struct kvm_clock_data user_ns;
4796 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4805 * TODO: userspace has to take care of races with VCPU_RUN, so
4806 * kvm_gen_update_masterclock() can be cut down to locked
4807 * pvclock_update_vm_gtod_copy().
4809 kvm_gen_update_masterclock(kvm);
4810 now_ns = get_kvmclock_ns(kvm);
4811 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4812 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4815 case KVM_GET_CLOCK: {
4816 struct kvm_clock_data user_ns;
4819 now_ns = get_kvmclock_ns(kvm);
4820 user_ns.clock = now_ns;
4821 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4822 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4825 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4830 case KVM_MEMORY_ENCRYPT_OP: {
4832 if (kvm_x86_ops->mem_enc_op)
4833 r = kvm_x86_ops->mem_enc_op(kvm, argp);
4836 case KVM_MEMORY_ENCRYPT_REG_REGION: {
4837 struct kvm_enc_region region;
4840 if (copy_from_user(®ion, argp, sizeof(region)))
4844 if (kvm_x86_ops->mem_enc_reg_region)
4845 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
4848 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4849 struct kvm_enc_region region;
4852 if (copy_from_user(®ion, argp, sizeof(region)))
4856 if (kvm_x86_ops->mem_enc_unreg_region)
4857 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
4860 case KVM_HYPERV_EVENTFD: {
4861 struct kvm_hyperv_eventfd hvevfd;
4864 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4866 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4876 static void kvm_init_msr_list(void)
4881 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4882 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4886 * Even MSRs that are valid in the host may not be exposed
4887 * to the guests in some cases.
4889 switch (msrs_to_save[i]) {
4890 case MSR_IA32_BNDCFGS:
4891 if (!kvm_mpx_supported())
4895 if (!kvm_x86_ops->rdtscp_supported())
4898 case MSR_IA32_RTIT_CTL:
4899 case MSR_IA32_RTIT_STATUS:
4900 if (!kvm_x86_ops->pt_supported())
4903 case MSR_IA32_RTIT_CR3_MATCH:
4904 if (!kvm_x86_ops->pt_supported() ||
4905 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
4908 case MSR_IA32_RTIT_OUTPUT_BASE:
4909 case MSR_IA32_RTIT_OUTPUT_MASK:
4910 if (!kvm_x86_ops->pt_supported() ||
4911 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
4912 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
4915 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
4916 if (!kvm_x86_ops->pt_supported() ||
4917 msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
4918 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
4927 msrs_to_save[j] = msrs_to_save[i];
4930 num_msrs_to_save = j;
4932 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4933 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4937 emulated_msrs[j] = emulated_msrs[i];
4940 num_emulated_msrs = j;
4942 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4943 struct kvm_msr_entry msr;
4945 msr.index = msr_based_features[i];
4946 if (kvm_get_msr_feature(&msr))
4950 msr_based_features[j] = msr_based_features[i];
4953 num_msr_based_features = j;
4956 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4964 if (!(lapic_in_kernel(vcpu) &&
4965 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4966 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4977 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4984 if (!(lapic_in_kernel(vcpu) &&
4985 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4987 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4989 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4999 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5000 struct kvm_segment *var, int seg)
5002 kvm_x86_ops->set_segment(vcpu, var, seg);
5005 void kvm_get_segment(struct kvm_vcpu *vcpu,
5006 struct kvm_segment *var, int seg)
5008 kvm_x86_ops->get_segment(vcpu, var, seg);
5011 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5012 struct x86_exception *exception)
5016 BUG_ON(!mmu_is_nested(vcpu));
5018 /* NPT walks are always user-walks */
5019 access |= PFERR_USER_MASK;
5020 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5025 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5026 struct x86_exception *exception)
5028 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5029 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5032 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5033 struct x86_exception *exception)
5035 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5036 access |= PFERR_FETCH_MASK;
5037 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5040 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5041 struct x86_exception *exception)
5043 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5044 access |= PFERR_WRITE_MASK;
5045 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5048 /* uses this to access any guest's mapped memory without checking CPL */
5049 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5050 struct x86_exception *exception)
5052 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5055 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5056 struct kvm_vcpu *vcpu, u32 access,
5057 struct x86_exception *exception)
5060 int r = X86EMUL_CONTINUE;
5063 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5065 unsigned offset = addr & (PAGE_SIZE-1);
5066 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5069 if (gpa == UNMAPPED_GVA)
5070 return X86EMUL_PROPAGATE_FAULT;
5071 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5074 r = X86EMUL_IO_NEEDED;
5086 /* used for instruction fetching */
5087 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5088 gva_t addr, void *val, unsigned int bytes,
5089 struct x86_exception *exception)
5091 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5092 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5096 /* Inline kvm_read_guest_virt_helper for speed. */
5097 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5099 if (unlikely(gpa == UNMAPPED_GVA))
5100 return X86EMUL_PROPAGATE_FAULT;
5102 offset = addr & (PAGE_SIZE-1);
5103 if (WARN_ON(offset + bytes > PAGE_SIZE))
5104 bytes = (unsigned)PAGE_SIZE - offset;
5105 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5107 if (unlikely(ret < 0))
5108 return X86EMUL_IO_NEEDED;
5110 return X86EMUL_CONTINUE;
5113 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5114 gva_t addr, void *val, unsigned int bytes,
5115 struct x86_exception *exception)
5117 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5119 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5122 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5124 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5125 gva_t addr, void *val, unsigned int bytes,
5126 struct x86_exception *exception, bool system)
5128 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5131 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5132 access |= PFERR_USER_MASK;
5134 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5137 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5138 unsigned long addr, void *val, unsigned int bytes)
5140 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5141 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5143 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5146 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5147 struct kvm_vcpu *vcpu, u32 access,
5148 struct x86_exception *exception)
5151 int r = X86EMUL_CONTINUE;
5154 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5157 unsigned offset = addr & (PAGE_SIZE-1);
5158 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5161 if (gpa == UNMAPPED_GVA)
5162 return X86EMUL_PROPAGATE_FAULT;
5163 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5165 r = X86EMUL_IO_NEEDED;
5177 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5178 unsigned int bytes, struct x86_exception *exception,
5181 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5182 u32 access = PFERR_WRITE_MASK;
5184 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5185 access |= PFERR_USER_MASK;
5187 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5191 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5192 unsigned int bytes, struct x86_exception *exception)
5194 /* kvm_write_guest_virt_system can pull in tons of pages. */
5195 vcpu->arch.l1tf_flush_l1d = true;
5197 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5198 PFERR_WRITE_MASK, exception);
5200 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5202 int handle_ud(struct kvm_vcpu *vcpu)
5204 int emul_type = EMULTYPE_TRAP_UD;
5205 enum emulation_result er;
5206 char sig[5]; /* ud2; .ascii "kvm" */
5207 struct x86_exception e;
5209 if (force_emulation_prefix &&
5210 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5211 sig, sizeof(sig), &e) == 0 &&
5212 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5213 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5217 er = kvm_emulate_instruction(vcpu, emul_type);
5218 if (er == EMULATE_USER_EXIT)
5220 if (er != EMULATE_DONE)
5221 kvm_queue_exception(vcpu, UD_VECTOR);
5224 EXPORT_SYMBOL_GPL(handle_ud);
5226 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5227 gpa_t gpa, bool write)
5229 /* For APIC access vmexit */
5230 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5233 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5234 trace_vcpu_match_mmio(gva, gpa, write, true);
5241 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5242 gpa_t *gpa, struct x86_exception *exception,
5245 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5246 | (write ? PFERR_WRITE_MASK : 0);
5249 * currently PKRU is only applied to ept enabled guest so
5250 * there is no pkey in EPT page table for L1 guest or EPT
5251 * shadow page table for L2 guest.
5253 if (vcpu_match_mmio_gva(vcpu, gva)
5254 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5255 vcpu->arch.access, 0, access)) {
5256 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5257 (gva & (PAGE_SIZE - 1));
5258 trace_vcpu_match_mmio(gva, *gpa, write, false);
5262 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5264 if (*gpa == UNMAPPED_GVA)
5267 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5270 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5271 const void *val, int bytes)
5275 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5278 kvm_page_track_write(vcpu, gpa, val, bytes);
5282 struct read_write_emulator_ops {
5283 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5285 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5286 void *val, int bytes);
5287 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5288 int bytes, void *val);
5289 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5290 void *val, int bytes);
5294 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5296 if (vcpu->mmio_read_completed) {
5297 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5298 vcpu->mmio_fragments[0].gpa, val);
5299 vcpu->mmio_read_completed = 0;
5306 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5307 void *val, int bytes)
5309 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5312 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5313 void *val, int bytes)
5315 return emulator_write_phys(vcpu, gpa, val, bytes);
5318 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5320 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5321 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5324 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5325 void *val, int bytes)
5327 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5328 return X86EMUL_IO_NEEDED;
5331 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5332 void *val, int bytes)
5334 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5336 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5337 return X86EMUL_CONTINUE;
5340 static const struct read_write_emulator_ops read_emultor = {
5341 .read_write_prepare = read_prepare,
5342 .read_write_emulate = read_emulate,
5343 .read_write_mmio = vcpu_mmio_read,
5344 .read_write_exit_mmio = read_exit_mmio,
5347 static const struct read_write_emulator_ops write_emultor = {
5348 .read_write_emulate = write_emulate,
5349 .read_write_mmio = write_mmio,
5350 .read_write_exit_mmio = write_exit_mmio,
5354 static int emulator_read_write_onepage(unsigned long addr, void *val,
5356 struct x86_exception *exception,
5357 struct kvm_vcpu *vcpu,
5358 const struct read_write_emulator_ops *ops)
5362 bool write = ops->write;
5363 struct kvm_mmio_fragment *frag;
5364 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5367 * If the exit was due to a NPF we may already have a GPA.
5368 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5369 * Note, this cannot be used on string operations since string
5370 * operation using rep will only have the initial GPA from the NPF
5373 if (vcpu->arch.gpa_available &&
5374 emulator_can_use_gpa(ctxt) &&
5375 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5376 gpa = vcpu->arch.gpa_val;
5377 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5379 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5381 return X86EMUL_PROPAGATE_FAULT;
5384 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5385 return X86EMUL_CONTINUE;
5388 * Is this MMIO handled locally?
5390 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5391 if (handled == bytes)
5392 return X86EMUL_CONTINUE;
5398 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5399 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5403 return X86EMUL_CONTINUE;
5406 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5408 void *val, unsigned int bytes,
5409 struct x86_exception *exception,
5410 const struct read_write_emulator_ops *ops)
5412 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5416 if (ops->read_write_prepare &&
5417 ops->read_write_prepare(vcpu, val, bytes))
5418 return X86EMUL_CONTINUE;
5420 vcpu->mmio_nr_fragments = 0;
5422 /* Crossing a page boundary? */
5423 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5426 now = -addr & ~PAGE_MASK;
5427 rc = emulator_read_write_onepage(addr, val, now, exception,
5430 if (rc != X86EMUL_CONTINUE)
5433 if (ctxt->mode != X86EMUL_MODE_PROT64)
5439 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5441 if (rc != X86EMUL_CONTINUE)
5444 if (!vcpu->mmio_nr_fragments)
5447 gpa = vcpu->mmio_fragments[0].gpa;
5449 vcpu->mmio_needed = 1;
5450 vcpu->mmio_cur_fragment = 0;
5452 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5453 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5454 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5455 vcpu->run->mmio.phys_addr = gpa;
5457 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5460 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5464 struct x86_exception *exception)
5466 return emulator_read_write(ctxt, addr, val, bytes,
5467 exception, &read_emultor);
5470 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5474 struct x86_exception *exception)
5476 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5477 exception, &write_emultor);
5480 #define CMPXCHG_TYPE(t, ptr, old, new) \
5481 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5483 #ifdef CONFIG_X86_64
5484 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5486 # define CMPXCHG64(ptr, old, new) \
5487 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5490 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5495 struct x86_exception *exception)
5497 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5503 /* guests cmpxchg8b have to be emulated atomically */
5504 if (bytes > 8 || (bytes & (bytes - 1)))
5507 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5509 if (gpa == UNMAPPED_GVA ||
5510 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5513 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5516 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5517 if (is_error_page(page))
5520 kaddr = kmap_atomic(page);
5521 kaddr += offset_in_page(gpa);
5524 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5527 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5530 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5533 exchanged = CMPXCHG64(kaddr, old, new);
5538 kunmap_atomic(kaddr);
5539 kvm_release_page_dirty(page);
5542 return X86EMUL_CMPXCHG_FAILED;
5544 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5545 kvm_page_track_write(vcpu, gpa, new, bytes);
5547 return X86EMUL_CONTINUE;
5550 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5552 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5555 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5559 for (i = 0; i < vcpu->arch.pio.count; i++) {
5560 if (vcpu->arch.pio.in)
5561 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5562 vcpu->arch.pio.size, pd);
5564 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5565 vcpu->arch.pio.port, vcpu->arch.pio.size,
5569 pd += vcpu->arch.pio.size;
5574 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5575 unsigned short port, void *val,
5576 unsigned int count, bool in)
5578 vcpu->arch.pio.port = port;
5579 vcpu->arch.pio.in = in;
5580 vcpu->arch.pio.count = count;
5581 vcpu->arch.pio.size = size;
5583 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5584 vcpu->arch.pio.count = 0;
5588 vcpu->run->exit_reason = KVM_EXIT_IO;
5589 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5590 vcpu->run->io.size = size;
5591 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5592 vcpu->run->io.count = count;
5593 vcpu->run->io.port = port;
5598 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5599 int size, unsigned short port, void *val,
5602 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5605 if (vcpu->arch.pio.count)
5608 memset(vcpu->arch.pio_data, 0, size * count);
5610 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5613 memcpy(val, vcpu->arch.pio_data, size * count);
5614 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5615 vcpu->arch.pio.count = 0;
5622 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5623 int size, unsigned short port,
5624 const void *val, unsigned int count)
5626 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5628 memcpy(vcpu->arch.pio_data, val, size * count);
5629 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5630 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5633 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5635 return kvm_x86_ops->get_segment_base(vcpu, seg);
5638 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5640 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5643 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5645 if (!need_emulate_wbinvd(vcpu))
5646 return X86EMUL_CONTINUE;
5648 if (kvm_x86_ops->has_wbinvd_exit()) {
5649 int cpu = get_cpu();
5651 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5652 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5653 wbinvd_ipi, NULL, 1);
5655 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5658 return X86EMUL_CONTINUE;
5661 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5663 kvm_emulate_wbinvd_noskip(vcpu);
5664 return kvm_skip_emulated_instruction(vcpu);
5666 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5670 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5672 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5675 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5676 unsigned long *dest)
5678 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5681 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5682 unsigned long value)
5685 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5688 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5690 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5693 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5695 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5696 unsigned long value;
5700 value = kvm_read_cr0(vcpu);
5703 value = vcpu->arch.cr2;
5706 value = kvm_read_cr3(vcpu);
5709 value = kvm_read_cr4(vcpu);
5712 value = kvm_get_cr8(vcpu);
5715 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5722 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5724 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5729 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5732 vcpu->arch.cr2 = val;
5735 res = kvm_set_cr3(vcpu, val);
5738 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5741 res = kvm_set_cr8(vcpu, val);
5744 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5751 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5753 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5756 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5758 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5761 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5763 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5766 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5768 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5771 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5773 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5776 static unsigned long emulator_get_cached_segment_base(
5777 struct x86_emulate_ctxt *ctxt, int seg)
5779 return get_segment_base(emul_to_vcpu(ctxt), seg);
5782 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5783 struct desc_struct *desc, u32 *base3,
5786 struct kvm_segment var;
5788 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5789 *selector = var.selector;
5792 memset(desc, 0, sizeof(*desc));
5800 set_desc_limit(desc, var.limit);
5801 set_desc_base(desc, (unsigned long)var.base);
5802 #ifdef CONFIG_X86_64
5804 *base3 = var.base >> 32;
5806 desc->type = var.type;
5808 desc->dpl = var.dpl;
5809 desc->p = var.present;
5810 desc->avl = var.avl;
5818 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5819 struct desc_struct *desc, u32 base3,
5822 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5823 struct kvm_segment var;
5825 var.selector = selector;
5826 var.base = get_desc_base(desc);
5827 #ifdef CONFIG_X86_64
5828 var.base |= ((u64)base3) << 32;
5830 var.limit = get_desc_limit(desc);
5832 var.limit = (var.limit << 12) | 0xfff;
5833 var.type = desc->type;
5834 var.dpl = desc->dpl;
5839 var.avl = desc->avl;
5840 var.present = desc->p;
5841 var.unusable = !var.present;
5844 kvm_set_segment(vcpu, &var, seg);
5848 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5849 u32 msr_index, u64 *pdata)
5851 struct msr_data msr;
5854 msr.index = msr_index;
5855 msr.host_initiated = false;
5856 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5864 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5865 u32 msr_index, u64 data)
5867 struct msr_data msr;
5870 msr.index = msr_index;
5871 msr.host_initiated = false;
5872 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5875 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5877 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5879 return vcpu->arch.smbase;
5882 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5884 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5886 vcpu->arch.smbase = smbase;
5889 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5892 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5895 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5896 u32 pmc, u64 *pdata)
5898 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5901 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5903 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5906 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5907 struct x86_instruction_info *info,
5908 enum x86_intercept_stage stage)
5910 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5913 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5914 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5916 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5919 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5921 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5924 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5926 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5929 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5931 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5934 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5936 return emul_to_vcpu(ctxt)->arch.hflags;
5939 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5941 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5944 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5946 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5949 static const struct x86_emulate_ops emulate_ops = {
5950 .read_gpr = emulator_read_gpr,
5951 .write_gpr = emulator_write_gpr,
5952 .read_std = emulator_read_std,
5953 .write_std = emulator_write_std,
5954 .read_phys = kvm_read_guest_phys_system,
5955 .fetch = kvm_fetch_guest_virt,
5956 .read_emulated = emulator_read_emulated,
5957 .write_emulated = emulator_write_emulated,
5958 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5959 .invlpg = emulator_invlpg,
5960 .pio_in_emulated = emulator_pio_in_emulated,
5961 .pio_out_emulated = emulator_pio_out_emulated,
5962 .get_segment = emulator_get_segment,
5963 .set_segment = emulator_set_segment,
5964 .get_cached_segment_base = emulator_get_cached_segment_base,
5965 .get_gdt = emulator_get_gdt,
5966 .get_idt = emulator_get_idt,
5967 .set_gdt = emulator_set_gdt,
5968 .set_idt = emulator_set_idt,
5969 .get_cr = emulator_get_cr,
5970 .set_cr = emulator_set_cr,
5971 .cpl = emulator_get_cpl,
5972 .get_dr = emulator_get_dr,
5973 .set_dr = emulator_set_dr,
5974 .get_smbase = emulator_get_smbase,
5975 .set_smbase = emulator_set_smbase,
5976 .set_msr = emulator_set_msr,
5977 .get_msr = emulator_get_msr,
5978 .check_pmc = emulator_check_pmc,
5979 .read_pmc = emulator_read_pmc,
5980 .halt = emulator_halt,
5981 .wbinvd = emulator_wbinvd,
5982 .fix_hypercall = emulator_fix_hypercall,
5983 .intercept = emulator_intercept,
5984 .get_cpuid = emulator_get_cpuid,
5985 .set_nmi_mask = emulator_set_nmi_mask,
5986 .get_hflags = emulator_get_hflags,
5987 .set_hflags = emulator_set_hflags,
5988 .pre_leave_smm = emulator_pre_leave_smm,
5991 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5993 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5995 * an sti; sti; sequence only disable interrupts for the first
5996 * instruction. So, if the last instruction, be it emulated or
5997 * not, left the system with the INT_STI flag enabled, it
5998 * means that the last instruction is an sti. We should not
5999 * leave the flag on in this case. The same goes for mov ss
6001 if (int_shadow & mask)
6003 if (unlikely(int_shadow || mask)) {
6004 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6006 kvm_make_request(KVM_REQ_EVENT, vcpu);
6010 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6012 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6013 if (ctxt->exception.vector == PF_VECTOR)
6014 return kvm_propagate_fault(vcpu, &ctxt->exception);
6016 if (ctxt->exception.error_code_valid)
6017 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6018 ctxt->exception.error_code);
6020 kvm_queue_exception(vcpu, ctxt->exception.vector);
6024 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6026 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6029 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6031 ctxt->eflags = kvm_get_rflags(vcpu);
6032 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6034 ctxt->eip = kvm_rip_read(vcpu);
6035 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6036 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6037 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6038 cs_db ? X86EMUL_MODE_PROT32 :
6039 X86EMUL_MODE_PROT16;
6040 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6041 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6042 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6044 init_decode_cache(ctxt);
6045 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6048 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6050 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6053 init_emulate_ctxt(vcpu);
6057 ctxt->_eip = ctxt->eip + inc_eip;
6058 ret = emulate_int_real(ctxt, irq);
6060 if (ret != X86EMUL_CONTINUE)
6061 return EMULATE_FAIL;
6063 ctxt->eip = ctxt->_eip;
6064 kvm_rip_write(vcpu, ctxt->eip);
6065 kvm_set_rflags(vcpu, ctxt->eflags);
6067 return EMULATE_DONE;
6069 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6071 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6073 int r = EMULATE_DONE;
6075 ++vcpu->stat.insn_emulation_fail;
6076 trace_kvm_emulate_insn_failed(vcpu);
6078 if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6079 return EMULATE_FAIL;
6081 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6082 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6083 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6084 vcpu->run->internal.ndata = 0;
6085 r = EMULATE_USER_EXIT;
6088 kvm_queue_exception(vcpu, UD_VECTOR);
6093 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6094 bool write_fault_to_shadow_pgtable,
6100 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6103 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6106 if (!vcpu->arch.mmu->direct_map) {
6108 * Write permission should be allowed since only
6109 * write access need to be emulated.
6111 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6114 * If the mapping is invalid in guest, let cpu retry
6115 * it to generate fault.
6117 if (gpa == UNMAPPED_GVA)
6122 * Do not retry the unhandleable instruction if it faults on the
6123 * readonly host memory, otherwise it will goto a infinite loop:
6124 * retry instruction -> write #PF -> emulation fail -> retry
6125 * instruction -> ...
6127 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6130 * If the instruction failed on the error pfn, it can not be fixed,
6131 * report the error to userspace.
6133 if (is_error_noslot_pfn(pfn))
6136 kvm_release_pfn_clean(pfn);
6138 /* The instructions are well-emulated on direct mmu. */
6139 if (vcpu->arch.mmu->direct_map) {
6140 unsigned int indirect_shadow_pages;
6142 spin_lock(&vcpu->kvm->mmu_lock);
6143 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6144 spin_unlock(&vcpu->kvm->mmu_lock);
6146 if (indirect_shadow_pages)
6147 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6153 * if emulation was due to access to shadowed page table
6154 * and it failed try to unshadow page and re-enter the
6155 * guest to let CPU execute the instruction.
6157 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6160 * If the access faults on its page table, it can not
6161 * be fixed by unprotecting shadow page and it should
6162 * be reported to userspace.
6164 return !write_fault_to_shadow_pgtable;
6167 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6168 unsigned long cr2, int emulation_type)
6170 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6171 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6173 last_retry_eip = vcpu->arch.last_retry_eip;
6174 last_retry_addr = vcpu->arch.last_retry_addr;
6177 * If the emulation is caused by #PF and it is non-page_table
6178 * writing instruction, it means the VM-EXIT is caused by shadow
6179 * page protected, we can zap the shadow page and retry this
6180 * instruction directly.
6182 * Note: if the guest uses a non-page-table modifying instruction
6183 * on the PDE that points to the instruction, then we will unmap
6184 * the instruction and go to an infinite loop. So, we cache the
6185 * last retried eip and the last fault address, if we meet the eip
6186 * and the address again, we can break out of the potential infinite
6189 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6191 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6194 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6197 if (x86_page_table_writing_insn(ctxt))
6200 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6203 vcpu->arch.last_retry_eip = ctxt->eip;
6204 vcpu->arch.last_retry_addr = cr2;
6206 if (!vcpu->arch.mmu->direct_map)
6207 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6209 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6214 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6215 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6217 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6219 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6220 /* This is a good place to trace that we are exiting SMM. */
6221 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6223 /* Process a latched INIT or SMI, if any. */
6224 kvm_make_request(KVM_REQ_EVENT, vcpu);
6227 kvm_mmu_reset_context(vcpu);
6230 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
6232 unsigned changed = vcpu->arch.hflags ^ emul_flags;
6234 vcpu->arch.hflags = emul_flags;
6236 if (changed & HF_SMM_MASK)
6237 kvm_smm_changed(vcpu);
6240 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6249 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6250 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6255 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6257 struct kvm_run *kvm_run = vcpu->run;
6259 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6260 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6261 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6262 kvm_run->debug.arch.exception = DB_VECTOR;
6263 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6264 *r = EMULATE_USER_EXIT;
6266 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6270 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6272 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6273 int r = EMULATE_DONE;
6275 kvm_x86_ops->skip_emulated_instruction(vcpu);
6278 * rflags is the old, "raw" value of the flags. The new value has
6279 * not been saved yet.
6281 * This is correct even for TF set by the guest, because "the
6282 * processor will not generate this exception after the instruction
6283 * that sets the TF flag".
6285 if (unlikely(rflags & X86_EFLAGS_TF))
6286 kvm_vcpu_do_singlestep(vcpu, &r);
6287 return r == EMULATE_DONE;
6289 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6291 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6293 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6294 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6295 struct kvm_run *kvm_run = vcpu->run;
6296 unsigned long eip = kvm_get_linear_rip(vcpu);
6297 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6298 vcpu->arch.guest_debug_dr7,
6302 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6303 kvm_run->debug.arch.pc = eip;
6304 kvm_run->debug.arch.exception = DB_VECTOR;
6305 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6306 *r = EMULATE_USER_EXIT;
6311 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6312 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6313 unsigned long eip = kvm_get_linear_rip(vcpu);
6314 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6319 vcpu->arch.dr6 &= ~15;
6320 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6321 kvm_queue_exception(vcpu, DB_VECTOR);
6330 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6332 switch (ctxt->opcode_len) {
6339 case 0xe6: /* OUT */
6343 case 0x6c: /* INS */
6345 case 0x6e: /* OUTS */
6352 case 0x33: /* RDPMC */
6361 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6368 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6369 bool writeback = true;
6370 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6372 vcpu->arch.l1tf_flush_l1d = true;
6375 * Clear write_fault_to_shadow_pgtable here to ensure it is
6378 vcpu->arch.write_fault_to_shadow_pgtable = false;
6379 kvm_clear_exception_queue(vcpu);
6381 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6382 init_emulate_ctxt(vcpu);
6385 * We will reenter on the same instruction since
6386 * we do not set complete_userspace_io. This does not
6387 * handle watchpoints yet, those would be handled in
6390 if (!(emulation_type & EMULTYPE_SKIP) &&
6391 kvm_vcpu_check_breakpoint(vcpu, &r))
6394 ctxt->interruptibility = 0;
6395 ctxt->have_exception = false;
6396 ctxt->exception.vector = -1;
6397 ctxt->perm_ok = false;
6399 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6401 r = x86_decode_insn(ctxt, insn, insn_len);
6403 trace_kvm_emulate_insn_start(vcpu);
6404 ++vcpu->stat.insn_emulation;
6405 if (r != EMULATION_OK) {
6406 if (emulation_type & EMULTYPE_TRAP_UD)
6407 return EMULATE_FAIL;
6408 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6410 return EMULATE_DONE;
6411 if (ctxt->have_exception && inject_emulated_exception(vcpu))
6412 return EMULATE_DONE;
6413 if (emulation_type & EMULTYPE_SKIP)
6414 return EMULATE_FAIL;
6415 return handle_emulation_failure(vcpu, emulation_type);
6419 if ((emulation_type & EMULTYPE_VMWARE) &&
6420 !is_vmware_backdoor_opcode(ctxt))
6421 return EMULATE_FAIL;
6423 if (emulation_type & EMULTYPE_SKIP) {
6424 kvm_rip_write(vcpu, ctxt->_eip);
6425 if (ctxt->eflags & X86_EFLAGS_RF)
6426 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6427 return EMULATE_DONE;
6430 if (retry_instruction(ctxt, cr2, emulation_type))
6431 return EMULATE_DONE;
6433 /* this is needed for vmware backdoor interface to work since it
6434 changes registers values during IO operation */
6435 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6436 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6437 emulator_invalidate_register_cache(ctxt);
6441 /* Save the faulting GPA (cr2) in the address field */
6442 ctxt->exception.address = cr2;
6444 r = x86_emulate_insn(ctxt);
6446 if (r == EMULATION_INTERCEPTED)
6447 return EMULATE_DONE;
6449 if (r == EMULATION_FAILED) {
6450 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6452 return EMULATE_DONE;
6454 return handle_emulation_failure(vcpu, emulation_type);
6457 if (ctxt->have_exception) {
6459 if (inject_emulated_exception(vcpu))
6461 } else if (vcpu->arch.pio.count) {
6462 if (!vcpu->arch.pio.in) {
6463 /* FIXME: return into emulator if single-stepping. */
6464 vcpu->arch.pio.count = 0;
6467 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6469 r = EMULATE_USER_EXIT;
6470 } else if (vcpu->mmio_needed) {
6471 if (!vcpu->mmio_is_write)
6473 r = EMULATE_USER_EXIT;
6474 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6475 } else if (r == EMULATION_RESTART)
6481 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6482 toggle_interruptibility(vcpu, ctxt->interruptibility);
6483 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6484 kvm_rip_write(vcpu, ctxt->eip);
6485 if (r == EMULATE_DONE && ctxt->tf)
6486 kvm_vcpu_do_singlestep(vcpu, &r);
6487 if (!ctxt->have_exception ||
6488 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6489 __kvm_set_rflags(vcpu, ctxt->eflags);
6492 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6493 * do nothing, and it will be requested again as soon as
6494 * the shadow expires. But we still need to check here,
6495 * because POPF has no interrupt shadow.
6497 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6498 kvm_make_request(KVM_REQ_EVENT, vcpu);
6500 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6505 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6507 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6509 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6511 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6512 void *insn, int insn_len)
6514 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6516 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6518 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6519 unsigned short port)
6521 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6522 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6523 size, port, &val, 1);
6524 /* do not return to emulator after return from userspace */
6525 vcpu->arch.pio.count = 0;
6529 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6533 /* We should only ever be called with arch.pio.count equal to 1 */
6534 BUG_ON(vcpu->arch.pio.count != 1);
6536 /* For size less than 4 we merge, else we zero extend */
6537 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6541 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6542 * the copy and tracing
6544 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6545 vcpu->arch.pio.port, &val, 1);
6546 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6551 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6552 unsigned short port)
6557 /* For size less than 4 we merge, else we zero extend */
6558 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6560 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6563 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6567 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6572 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6574 int ret = kvm_skip_emulated_instruction(vcpu);
6577 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6578 * KVM_EXIT_DEBUG here.
6581 return kvm_fast_pio_in(vcpu, size, port) && ret;
6583 return kvm_fast_pio_out(vcpu, size, port) && ret;
6585 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6587 static int kvmclock_cpu_down_prep(unsigned int cpu)
6589 __this_cpu_write(cpu_tsc_khz, 0);
6593 static void tsc_khz_changed(void *data)
6595 struct cpufreq_freqs *freq = data;
6596 unsigned long khz = 0;
6600 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6601 khz = cpufreq_quick_get(raw_smp_processor_id());
6604 __this_cpu_write(cpu_tsc_khz, khz);
6607 #ifdef CONFIG_X86_64
6608 static void kvm_hyperv_tsc_notifier(void)
6611 struct kvm_vcpu *vcpu;
6614 spin_lock(&kvm_lock);
6615 list_for_each_entry(kvm, &vm_list, vm_list)
6616 kvm_make_mclock_inprogress_request(kvm);
6618 hyperv_stop_tsc_emulation();
6620 /* TSC frequency always matches when on Hyper-V */
6621 for_each_present_cpu(cpu)
6622 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6623 kvm_max_guest_tsc_khz = tsc_khz;
6625 list_for_each_entry(kvm, &vm_list, vm_list) {
6626 struct kvm_arch *ka = &kvm->arch;
6628 spin_lock(&ka->pvclock_gtod_sync_lock);
6630 pvclock_update_vm_gtod_copy(kvm);
6632 kvm_for_each_vcpu(cpu, vcpu, kvm)
6633 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6635 kvm_for_each_vcpu(cpu, vcpu, kvm)
6636 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6638 spin_unlock(&ka->pvclock_gtod_sync_lock);
6640 spin_unlock(&kvm_lock);
6644 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6647 struct cpufreq_freqs *freq = data;
6649 struct kvm_vcpu *vcpu;
6650 int i, send_ipi = 0;
6653 * We allow guests to temporarily run on slowing clocks,
6654 * provided we notify them after, or to run on accelerating
6655 * clocks, provided we notify them before. Thus time never
6658 * However, we have a problem. We can't atomically update
6659 * the frequency of a given CPU from this function; it is
6660 * merely a notifier, which can be called from any CPU.
6661 * Changing the TSC frequency at arbitrary points in time
6662 * requires a recomputation of local variables related to
6663 * the TSC for each VCPU. We must flag these local variables
6664 * to be updated and be sure the update takes place with the
6665 * new frequency before any guests proceed.
6667 * Unfortunately, the combination of hotplug CPU and frequency
6668 * change creates an intractable locking scenario; the order
6669 * of when these callouts happen is undefined with respect to
6670 * CPU hotplug, and they can race with each other. As such,
6671 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6672 * undefined; you can actually have a CPU frequency change take
6673 * place in between the computation of X and the setting of the
6674 * variable. To protect against this problem, all updates of
6675 * the per_cpu tsc_khz variable are done in an interrupt
6676 * protected IPI, and all callers wishing to update the value
6677 * must wait for a synchronous IPI to complete (which is trivial
6678 * if the caller is on the CPU already). This establishes the
6679 * necessary total order on variable updates.
6681 * Note that because a guest time update may take place
6682 * anytime after the setting of the VCPU's request bit, the
6683 * correct TSC value must be set before the request. However,
6684 * to ensure the update actually makes it to any guest which
6685 * starts running in hardware virtualization between the set
6686 * and the acquisition of the spinlock, we must also ping the
6687 * CPU after setting the request bit.
6691 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6693 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6696 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6698 spin_lock(&kvm_lock);
6699 list_for_each_entry(kvm, &vm_list, vm_list) {
6700 kvm_for_each_vcpu(i, vcpu, kvm) {
6701 if (vcpu->cpu != freq->cpu)
6703 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6704 if (vcpu->cpu != smp_processor_id())
6708 spin_unlock(&kvm_lock);
6710 if (freq->old < freq->new && send_ipi) {
6712 * We upscale the frequency. Must make the guest
6713 * doesn't see old kvmclock values while running with
6714 * the new frequency, otherwise we risk the guest sees
6715 * time go backwards.
6717 * In case we update the frequency for another cpu
6718 * (which might be in guest context) send an interrupt
6719 * to kick the cpu out of guest context. Next time
6720 * guest context is entered kvmclock will be updated,
6721 * so the guest will not see stale values.
6723 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6728 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6729 .notifier_call = kvmclock_cpufreq_notifier
6732 static int kvmclock_cpu_online(unsigned int cpu)
6734 tsc_khz_changed(NULL);
6738 static void kvm_timer_init(void)
6740 max_tsc_khz = tsc_khz;
6742 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6743 #ifdef CONFIG_CPU_FREQ
6744 struct cpufreq_policy policy;
6747 memset(&policy, 0, sizeof(policy));
6749 cpufreq_get_policy(&policy, cpu);
6750 if (policy.cpuinfo.max_freq)
6751 max_tsc_khz = policy.cpuinfo.max_freq;
6754 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6755 CPUFREQ_TRANSITION_NOTIFIER);
6757 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6759 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6760 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6763 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6764 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6766 int kvm_is_in_guest(void)
6768 return __this_cpu_read(current_vcpu) != NULL;
6771 static int kvm_is_user_mode(void)
6775 if (__this_cpu_read(current_vcpu))
6776 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6778 return user_mode != 0;
6781 static unsigned long kvm_get_guest_ip(void)
6783 unsigned long ip = 0;
6785 if (__this_cpu_read(current_vcpu))
6786 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6791 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6792 .is_in_guest = kvm_is_in_guest,
6793 .is_user_mode = kvm_is_user_mode,
6794 .get_guest_ip = kvm_get_guest_ip,
6797 static void kvm_set_mmio_spte_mask(void)
6800 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6803 * Set the reserved bits and the present bit of an paging-structure
6804 * entry to generate page fault with PFER.RSV = 1.
6808 * Mask the uppermost physical address bit, which would be reserved as
6809 * long as the supported physical address width is less than 52.
6813 /* Set the present bit. */
6817 * If reserved bit is not supported, clear the present bit to disable
6820 if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6823 kvm_mmu_set_mmio_spte_mask(mask, mask);
6826 #ifdef CONFIG_X86_64
6827 static void pvclock_gtod_update_fn(struct work_struct *work)
6831 struct kvm_vcpu *vcpu;
6834 spin_lock(&kvm_lock);
6835 list_for_each_entry(kvm, &vm_list, vm_list)
6836 kvm_for_each_vcpu(i, vcpu, kvm)
6837 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6838 atomic_set(&kvm_guest_has_master_clock, 0);
6839 spin_unlock(&kvm_lock);
6842 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6845 * Notification about pvclock gtod data update.
6847 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6850 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6851 struct timekeeper *tk = priv;
6853 update_pvclock_gtod(tk);
6855 /* disable master clock if host does not trust, or does not
6856 * use, TSC based clocksource.
6858 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6859 atomic_read(&kvm_guest_has_master_clock) != 0)
6860 queue_work(system_long_wq, &pvclock_gtod_work);
6865 static struct notifier_block pvclock_gtod_notifier = {
6866 .notifier_call = pvclock_gtod_notify,
6870 int kvm_arch_init(void *opaque)
6873 struct kvm_x86_ops *ops = opaque;
6876 printk(KERN_ERR "kvm: already loaded the other module\n");
6881 if (!ops->cpu_has_kvm_support()) {
6882 printk(KERN_ERR "kvm: no hardware support\n");
6886 if (ops->disabled_by_bios()) {
6887 printk(KERN_ERR "kvm: disabled by bios\n");
6893 * KVM explicitly assumes that the guest has an FPU and
6894 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
6895 * vCPU's FPU state as a fxregs_state struct.
6897 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
6898 printk(KERN_ERR "kvm: inadequate fpu\n");
6904 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
6905 __alignof__(struct fpu), SLAB_ACCOUNT,
6907 if (!x86_fpu_cache) {
6908 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
6912 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6914 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6915 goto out_free_x86_fpu_cache;
6918 r = kvm_mmu_module_init();
6920 goto out_free_percpu;
6922 kvm_set_mmio_spte_mask();
6926 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6927 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6928 PT_PRESENT_MASK, 0, sme_me_mask);
6931 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6933 if (boot_cpu_has(X86_FEATURE_XSAVE))
6934 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6937 #ifdef CONFIG_X86_64
6938 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6940 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6941 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6947 free_percpu(shared_msrs);
6948 out_free_x86_fpu_cache:
6949 kmem_cache_destroy(x86_fpu_cache);
6954 void kvm_arch_exit(void)
6956 #ifdef CONFIG_X86_64
6957 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6958 clear_hv_tscchange_cb();
6961 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6963 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6964 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6965 CPUFREQ_TRANSITION_NOTIFIER);
6966 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6967 #ifdef CONFIG_X86_64
6968 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6971 kvm_mmu_module_exit();
6972 free_percpu(shared_msrs);
6973 kmem_cache_destroy(x86_fpu_cache);
6976 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6978 ++vcpu->stat.halt_exits;
6979 if (lapic_in_kernel(vcpu)) {
6980 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6983 vcpu->run->exit_reason = KVM_EXIT_HLT;
6987 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6989 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6991 int ret = kvm_skip_emulated_instruction(vcpu);
6993 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6994 * KVM_EXIT_DEBUG here.
6996 return kvm_vcpu_halt(vcpu) && ret;
6998 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7000 #ifdef CONFIG_X86_64
7001 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7002 unsigned long clock_type)
7004 struct kvm_clock_pairing clock_pairing;
7005 struct timespec64 ts;
7009 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7010 return -KVM_EOPNOTSUPP;
7012 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7013 return -KVM_EOPNOTSUPP;
7015 clock_pairing.sec = ts.tv_sec;
7016 clock_pairing.nsec = ts.tv_nsec;
7017 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7018 clock_pairing.flags = 0;
7019 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7022 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7023 sizeof(struct kvm_clock_pairing)))
7031 * kvm_pv_kick_cpu_op: Kick a vcpu.
7033 * @apicid - apicid of vcpu to be kicked.
7035 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7037 struct kvm_lapic_irq lapic_irq;
7039 lapic_irq.shorthand = 0;
7040 lapic_irq.dest_mode = 0;
7041 lapic_irq.level = 0;
7042 lapic_irq.dest_id = apicid;
7043 lapic_irq.msi_redir_hint = false;
7045 lapic_irq.delivery_mode = APIC_DM_REMRD;
7046 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7049 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7051 vcpu->arch.apicv_active = false;
7052 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7055 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7057 unsigned long nr, a0, a1, a2, a3, ret;
7060 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7061 return kvm_hv_hypercall(vcpu);
7063 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
7064 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
7065 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
7066 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
7067 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
7069 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7071 op_64_bit = is_64_bit_mode(vcpu);
7080 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7086 case KVM_HC_VAPIC_POLL_IRQ:
7089 case KVM_HC_KICK_CPU:
7090 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7093 #ifdef CONFIG_X86_64
7094 case KVM_HC_CLOCK_PAIRING:
7095 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7098 case KVM_HC_SEND_IPI:
7099 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7108 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
7110 ++vcpu->stat.hypercalls;
7111 return kvm_skip_emulated_instruction(vcpu);
7113 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7115 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7117 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7118 char instruction[3];
7119 unsigned long rip = kvm_rip_read(vcpu);
7121 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7123 return emulator_write_emulated(ctxt, rip, instruction, 3,
7127 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7129 return vcpu->run->request_interrupt_window &&
7130 likely(!pic_in_kernel(vcpu->kvm));
7133 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7135 struct kvm_run *kvm_run = vcpu->run;
7137 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7138 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7139 kvm_run->cr8 = kvm_get_cr8(vcpu);
7140 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7141 kvm_run->ready_for_interrupt_injection =
7142 pic_in_kernel(vcpu->kvm) ||
7143 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7146 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7150 if (!kvm_x86_ops->update_cr8_intercept)
7153 if (!lapic_in_kernel(vcpu))
7156 if (vcpu->arch.apicv_active)
7159 if (!vcpu->arch.apic->vapic_addr)
7160 max_irr = kvm_lapic_find_highest_irr(vcpu);
7167 tpr = kvm_lapic_get_cr8(vcpu);
7169 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7172 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7176 /* try to reinject previous events if any */
7178 if (vcpu->arch.exception.injected)
7179 kvm_x86_ops->queue_exception(vcpu);
7181 * Do not inject an NMI or interrupt if there is a pending
7182 * exception. Exceptions and interrupts are recognized at
7183 * instruction boundaries, i.e. the start of an instruction.
7184 * Trap-like exceptions, e.g. #DB, have higher priority than
7185 * NMIs and interrupts, i.e. traps are recognized before an
7186 * NMI/interrupt that's pending on the same instruction.
7187 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7188 * priority, but are only generated (pended) during instruction
7189 * execution, i.e. a pending fault-like exception means the
7190 * fault occurred on the *previous* instruction and must be
7191 * serviced prior to recognizing any new events in order to
7192 * fully complete the previous instruction.
7194 else if (!vcpu->arch.exception.pending) {
7195 if (vcpu->arch.nmi_injected)
7196 kvm_x86_ops->set_nmi(vcpu);
7197 else if (vcpu->arch.interrupt.injected)
7198 kvm_x86_ops->set_irq(vcpu);
7202 * Call check_nested_events() even if we reinjected a previous event
7203 * in order for caller to determine if it should require immediate-exit
7204 * from L2 to L1 due to pending L1 events which require exit
7207 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7208 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7213 /* try to inject new event if pending */
7214 if (vcpu->arch.exception.pending) {
7215 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7216 vcpu->arch.exception.has_error_code,
7217 vcpu->arch.exception.error_code);
7219 WARN_ON_ONCE(vcpu->arch.exception.injected);
7220 vcpu->arch.exception.pending = false;
7221 vcpu->arch.exception.injected = true;
7223 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7224 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7227 if (vcpu->arch.exception.nr == DB_VECTOR) {
7229 * This code assumes that nSVM doesn't use
7230 * check_nested_events(). If it does, the
7231 * DR6/DR7 changes should happen before L1
7232 * gets a #VMEXIT for an intercepted #DB in
7233 * L2. (Under VMX, on the other hand, the
7234 * DR6/DR7 changes should not happen in the
7235 * event of a VM-exit to L1 for an intercepted
7238 kvm_deliver_exception_payload(vcpu);
7239 if (vcpu->arch.dr7 & DR7_GD) {
7240 vcpu->arch.dr7 &= ~DR7_GD;
7241 kvm_update_dr7(vcpu);
7245 kvm_x86_ops->queue_exception(vcpu);
7248 /* Don't consider new event if we re-injected an event */
7249 if (kvm_event_needs_reinjection(vcpu))
7252 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7253 kvm_x86_ops->smi_allowed(vcpu)) {
7254 vcpu->arch.smi_pending = false;
7255 ++vcpu->arch.smi_count;
7257 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7258 --vcpu->arch.nmi_pending;
7259 vcpu->arch.nmi_injected = true;
7260 kvm_x86_ops->set_nmi(vcpu);
7261 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7263 * Because interrupts can be injected asynchronously, we are
7264 * calling check_nested_events again here to avoid a race condition.
7265 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7266 * proposal and current concerns. Perhaps we should be setting
7267 * KVM_REQ_EVENT only on certain events and not unconditionally?
7269 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7270 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7274 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7275 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7277 kvm_x86_ops->set_irq(vcpu);
7284 static void process_nmi(struct kvm_vcpu *vcpu)
7289 * x86 is limited to one NMI running, and one NMI pending after it.
7290 * If an NMI is already in progress, limit further NMIs to just one.
7291 * Otherwise, allow two (and we'll inject the first one immediately).
7293 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7296 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7297 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7298 kvm_make_request(KVM_REQ_EVENT, vcpu);
7301 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7304 flags |= seg->g << 23;
7305 flags |= seg->db << 22;
7306 flags |= seg->l << 21;
7307 flags |= seg->avl << 20;
7308 flags |= seg->present << 15;
7309 flags |= seg->dpl << 13;
7310 flags |= seg->s << 12;
7311 flags |= seg->type << 8;
7315 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7317 struct kvm_segment seg;
7320 kvm_get_segment(vcpu, &seg, n);
7321 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7324 offset = 0x7f84 + n * 12;
7326 offset = 0x7f2c + (n - 3) * 12;
7328 put_smstate(u32, buf, offset + 8, seg.base);
7329 put_smstate(u32, buf, offset + 4, seg.limit);
7330 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7333 #ifdef CONFIG_X86_64
7334 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7336 struct kvm_segment seg;
7340 kvm_get_segment(vcpu, &seg, n);
7341 offset = 0x7e00 + n * 16;
7343 flags = enter_smm_get_segment_flags(&seg) >> 8;
7344 put_smstate(u16, buf, offset, seg.selector);
7345 put_smstate(u16, buf, offset + 2, flags);
7346 put_smstate(u32, buf, offset + 4, seg.limit);
7347 put_smstate(u64, buf, offset + 8, seg.base);
7351 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7354 struct kvm_segment seg;
7358 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7359 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7360 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7361 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7363 for (i = 0; i < 8; i++)
7364 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7366 kvm_get_dr(vcpu, 6, &val);
7367 put_smstate(u32, buf, 0x7fcc, (u32)val);
7368 kvm_get_dr(vcpu, 7, &val);
7369 put_smstate(u32, buf, 0x7fc8, (u32)val);
7371 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7372 put_smstate(u32, buf, 0x7fc4, seg.selector);
7373 put_smstate(u32, buf, 0x7f64, seg.base);
7374 put_smstate(u32, buf, 0x7f60, seg.limit);
7375 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7377 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7378 put_smstate(u32, buf, 0x7fc0, seg.selector);
7379 put_smstate(u32, buf, 0x7f80, seg.base);
7380 put_smstate(u32, buf, 0x7f7c, seg.limit);
7381 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7383 kvm_x86_ops->get_gdt(vcpu, &dt);
7384 put_smstate(u32, buf, 0x7f74, dt.address);
7385 put_smstate(u32, buf, 0x7f70, dt.size);
7387 kvm_x86_ops->get_idt(vcpu, &dt);
7388 put_smstate(u32, buf, 0x7f58, dt.address);
7389 put_smstate(u32, buf, 0x7f54, dt.size);
7391 for (i = 0; i < 6; i++)
7392 enter_smm_save_seg_32(vcpu, buf, i);
7394 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7397 put_smstate(u32, buf, 0x7efc, 0x00020000);
7398 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7401 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7403 #ifdef CONFIG_X86_64
7405 struct kvm_segment seg;
7409 for (i = 0; i < 16; i++)
7410 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7412 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7413 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7415 kvm_get_dr(vcpu, 6, &val);
7416 put_smstate(u64, buf, 0x7f68, val);
7417 kvm_get_dr(vcpu, 7, &val);
7418 put_smstate(u64, buf, 0x7f60, val);
7420 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7421 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7422 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7424 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7427 put_smstate(u32, buf, 0x7efc, 0x00020064);
7429 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7431 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7432 put_smstate(u16, buf, 0x7e90, seg.selector);
7433 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7434 put_smstate(u32, buf, 0x7e94, seg.limit);
7435 put_smstate(u64, buf, 0x7e98, seg.base);
7437 kvm_x86_ops->get_idt(vcpu, &dt);
7438 put_smstate(u32, buf, 0x7e84, dt.size);
7439 put_smstate(u64, buf, 0x7e88, dt.address);
7441 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7442 put_smstate(u16, buf, 0x7e70, seg.selector);
7443 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7444 put_smstate(u32, buf, 0x7e74, seg.limit);
7445 put_smstate(u64, buf, 0x7e78, seg.base);
7447 kvm_x86_ops->get_gdt(vcpu, &dt);
7448 put_smstate(u32, buf, 0x7e64, dt.size);
7449 put_smstate(u64, buf, 0x7e68, dt.address);
7451 for (i = 0; i < 6; i++)
7452 enter_smm_save_seg_64(vcpu, buf, i);
7458 static void enter_smm(struct kvm_vcpu *vcpu)
7460 struct kvm_segment cs, ds;
7465 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7466 memset(buf, 0, 512);
7467 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7468 enter_smm_save_state_64(vcpu, buf);
7470 enter_smm_save_state_32(vcpu, buf);
7473 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7474 * vCPU state (e.g. leave guest mode) after we've saved the state into
7475 * the SMM state-save area.
7477 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7479 vcpu->arch.hflags |= HF_SMM_MASK;
7480 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7482 if (kvm_x86_ops->get_nmi_mask(vcpu))
7483 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7485 kvm_x86_ops->set_nmi_mask(vcpu, true);
7487 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7488 kvm_rip_write(vcpu, 0x8000);
7490 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7491 kvm_x86_ops->set_cr0(vcpu, cr0);
7492 vcpu->arch.cr0 = cr0;
7494 kvm_x86_ops->set_cr4(vcpu, 0);
7496 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7497 dt.address = dt.size = 0;
7498 kvm_x86_ops->set_idt(vcpu, &dt);
7500 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7502 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7503 cs.base = vcpu->arch.smbase;
7508 cs.limit = ds.limit = 0xffffffff;
7509 cs.type = ds.type = 0x3;
7510 cs.dpl = ds.dpl = 0;
7515 cs.avl = ds.avl = 0;
7516 cs.present = ds.present = 1;
7517 cs.unusable = ds.unusable = 0;
7518 cs.padding = ds.padding = 0;
7520 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7521 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7522 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7523 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7524 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7525 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7527 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7528 kvm_x86_ops->set_efer(vcpu, 0);
7530 kvm_update_cpuid(vcpu);
7531 kvm_mmu_reset_context(vcpu);
7534 static void process_smi(struct kvm_vcpu *vcpu)
7536 vcpu->arch.smi_pending = true;
7537 kvm_make_request(KVM_REQ_EVENT, vcpu);
7540 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7542 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7545 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7547 if (!kvm_apic_present(vcpu))
7550 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7552 if (irqchip_split(vcpu->kvm))
7553 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7555 if (vcpu->arch.apicv_active)
7556 kvm_x86_ops->sync_pir_to_irr(vcpu);
7557 if (ioapic_in_kernel(vcpu->kvm))
7558 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7561 if (is_guest_mode(vcpu))
7562 vcpu->arch.load_eoi_exitmap_pending = true;
7564 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7567 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7569 u64 eoi_exit_bitmap[4];
7571 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7574 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7575 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7576 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7579 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7580 unsigned long start, unsigned long end,
7583 unsigned long apic_address;
7586 * The physical address of apic access page is stored in the VMCS.
7587 * Update it when it becomes invalid.
7589 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7590 if (start <= apic_address && apic_address < end)
7591 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7596 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7598 struct page *page = NULL;
7600 if (!lapic_in_kernel(vcpu))
7603 if (!kvm_x86_ops->set_apic_access_page_addr)
7606 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7607 if (is_error_page(page))
7609 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7612 * Do not pin apic access page in memory, the MMU notifier
7613 * will call us again if it is migrated or swapped out.
7617 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7619 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7621 smp_send_reschedule(vcpu->cpu);
7623 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7626 * Returns 1 to let vcpu_run() continue the guest execution loop without
7627 * exiting to the userspace. Otherwise, the value will be returned to the
7630 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7634 dm_request_for_irq_injection(vcpu) &&
7635 kvm_cpu_accept_dm_intr(vcpu);
7637 bool req_immediate_exit = false;
7639 if (kvm_request_pending(vcpu)) {
7640 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7641 kvm_x86_ops->get_vmcs12_pages(vcpu);
7642 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7643 kvm_mmu_unload(vcpu);
7644 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7645 __kvm_migrate_timers(vcpu);
7646 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7647 kvm_gen_update_masterclock(vcpu->kvm);
7648 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7649 kvm_gen_kvmclock_update(vcpu);
7650 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7651 r = kvm_guest_time_update(vcpu);
7655 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7656 kvm_mmu_sync_roots(vcpu);
7657 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7658 kvm_mmu_load_cr3(vcpu);
7659 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7660 kvm_vcpu_flush_tlb(vcpu, true);
7661 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7662 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7666 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7667 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7668 vcpu->mmio_needed = 0;
7672 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7673 /* Page is swapped out. Do synthetic halt */
7674 vcpu->arch.apf.halted = true;
7678 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7679 record_steal_time(vcpu);
7680 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7682 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7684 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7685 kvm_pmu_handle_event(vcpu);
7686 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7687 kvm_pmu_deliver_pmi(vcpu);
7688 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7689 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7690 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7691 vcpu->arch.ioapic_handled_vectors)) {
7692 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7693 vcpu->run->eoi.vector =
7694 vcpu->arch.pending_ioapic_eoi;
7699 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7700 vcpu_scan_ioapic(vcpu);
7701 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7702 vcpu_load_eoi_exitmap(vcpu);
7703 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7704 kvm_vcpu_reload_apic_access_page(vcpu);
7705 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7706 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7707 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7711 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7712 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7713 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7717 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7718 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7719 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7725 * KVM_REQ_HV_STIMER has to be processed after
7726 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7727 * depend on the guest clock being up-to-date
7729 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7730 kvm_hv_process_stimers(vcpu);
7733 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7734 ++vcpu->stat.req_event;
7735 kvm_apic_accept_events(vcpu);
7736 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7741 if (inject_pending_event(vcpu, req_int_win) != 0)
7742 req_immediate_exit = true;
7744 /* Enable SMI/NMI/IRQ window open exits if needed.
7746 * SMIs have three cases:
7747 * 1) They can be nested, and then there is nothing to
7748 * do here because RSM will cause a vmexit anyway.
7749 * 2) There is an ISA-specific reason why SMI cannot be
7750 * injected, and the moment when this changes can be
7752 * 3) Or the SMI can be pending because
7753 * inject_pending_event has completed the injection
7754 * of an IRQ or NMI from the previous vmexit, and
7755 * then we request an immediate exit to inject the
7758 if (vcpu->arch.smi_pending && !is_smm(vcpu))
7759 if (!kvm_x86_ops->enable_smi_window(vcpu))
7760 req_immediate_exit = true;
7761 if (vcpu->arch.nmi_pending)
7762 kvm_x86_ops->enable_nmi_window(vcpu);
7763 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7764 kvm_x86_ops->enable_irq_window(vcpu);
7765 WARN_ON(vcpu->arch.exception.pending);
7768 if (kvm_lapic_enabled(vcpu)) {
7769 update_cr8_intercept(vcpu);
7770 kvm_lapic_sync_to_vapic(vcpu);
7774 r = kvm_mmu_reload(vcpu);
7776 goto cancel_injection;
7781 kvm_x86_ops->prepare_guest_switch(vcpu);
7784 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7785 * IPI are then delayed after guest entry, which ensures that they
7786 * result in virtual interrupt delivery.
7788 local_irq_disable();
7789 vcpu->mode = IN_GUEST_MODE;
7791 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7794 * 1) We should set ->mode before checking ->requests. Please see
7795 * the comment in kvm_vcpu_exiting_guest_mode().
7797 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7798 * pairs with the memory barrier implicit in pi_test_and_set_on
7799 * (see vmx_deliver_posted_interrupt).
7801 * 3) This also orders the write to mode from any reads to the page
7802 * tables done while the VCPU is running. Please see the comment
7803 * in kvm_flush_remote_tlbs.
7805 smp_mb__after_srcu_read_unlock();
7808 * This handles the case where a posted interrupt was
7809 * notified with kvm_vcpu_kick.
7811 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7812 kvm_x86_ops->sync_pir_to_irr(vcpu);
7814 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7815 || need_resched() || signal_pending(current)) {
7816 vcpu->mode = OUTSIDE_GUEST_MODE;
7820 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7822 goto cancel_injection;
7825 kvm_load_guest_xcr0(vcpu);
7827 if (req_immediate_exit) {
7828 kvm_make_request(KVM_REQ_EVENT, vcpu);
7829 kvm_x86_ops->request_immediate_exit(vcpu);
7832 trace_kvm_entry(vcpu->vcpu_id);
7833 if (lapic_timer_advance_ns)
7834 wait_lapic_expire(vcpu);
7835 guest_enter_irqoff();
7837 if (unlikely(vcpu->arch.switch_db_regs)) {
7839 set_debugreg(vcpu->arch.eff_db[0], 0);
7840 set_debugreg(vcpu->arch.eff_db[1], 1);
7841 set_debugreg(vcpu->arch.eff_db[2], 2);
7842 set_debugreg(vcpu->arch.eff_db[3], 3);
7843 set_debugreg(vcpu->arch.dr6, 6);
7844 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7847 kvm_x86_ops->run(vcpu);
7850 * Do this here before restoring debug registers on the host. And
7851 * since we do this before handling the vmexit, a DR access vmexit
7852 * can (a) read the correct value of the debug registers, (b) set
7853 * KVM_DEBUGREG_WONT_EXIT again.
7855 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7856 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7857 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7858 kvm_update_dr0123(vcpu);
7859 kvm_update_dr6(vcpu);
7860 kvm_update_dr7(vcpu);
7861 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7865 * If the guest has used debug registers, at least dr7
7866 * will be disabled while returning to the host.
7867 * If we don't have active breakpoints in the host, we don't
7868 * care about the messed up debug address registers. But if
7869 * we have some of them active, restore the old state.
7871 if (hw_breakpoint_active())
7872 hw_breakpoint_restore();
7874 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7876 vcpu->mode = OUTSIDE_GUEST_MODE;
7879 kvm_put_guest_xcr0(vcpu);
7881 kvm_before_interrupt(vcpu);
7882 kvm_x86_ops->handle_external_intr(vcpu);
7883 kvm_after_interrupt(vcpu);
7887 guest_exit_irqoff();
7892 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7895 * Profile KVM exit RIPs:
7897 if (unlikely(prof_on == KVM_PROFILING)) {
7898 unsigned long rip = kvm_rip_read(vcpu);
7899 profile_hit(KVM_PROFILING, (void *)rip);
7902 if (unlikely(vcpu->arch.tsc_always_catchup))
7903 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7905 if (vcpu->arch.apic_attention)
7906 kvm_lapic_sync_from_vapic(vcpu);
7908 vcpu->arch.gpa_available = false;
7909 r = kvm_x86_ops->handle_exit(vcpu);
7913 kvm_x86_ops->cancel_injection(vcpu);
7914 if (unlikely(vcpu->arch.apic_attention))
7915 kvm_lapic_sync_from_vapic(vcpu);
7920 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7922 if (!kvm_arch_vcpu_runnable(vcpu) &&
7923 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7924 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7925 kvm_vcpu_block(vcpu);
7926 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7928 if (kvm_x86_ops->post_block)
7929 kvm_x86_ops->post_block(vcpu);
7931 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7935 kvm_apic_accept_events(vcpu);
7936 switch(vcpu->arch.mp_state) {
7937 case KVM_MP_STATE_HALTED:
7938 vcpu->arch.pv.pv_unhalted = false;
7939 vcpu->arch.mp_state =
7940 KVM_MP_STATE_RUNNABLE;
7942 case KVM_MP_STATE_RUNNABLE:
7943 vcpu->arch.apf.halted = false;
7945 case KVM_MP_STATE_INIT_RECEIVED:
7954 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7956 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7957 kvm_x86_ops->check_nested_events(vcpu, false);
7959 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7960 !vcpu->arch.apf.halted);
7963 static int vcpu_run(struct kvm_vcpu *vcpu)
7966 struct kvm *kvm = vcpu->kvm;
7968 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7969 vcpu->arch.l1tf_flush_l1d = true;
7972 if (kvm_vcpu_running(vcpu)) {
7973 r = vcpu_enter_guest(vcpu);
7975 r = vcpu_block(kvm, vcpu);
7981 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7982 if (kvm_cpu_has_pending_timer(vcpu))
7983 kvm_inject_pending_timer_irqs(vcpu);
7985 if (dm_request_for_irq_injection(vcpu) &&
7986 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7988 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7989 ++vcpu->stat.request_irq_exits;
7993 kvm_check_async_pf_completion(vcpu);
7995 if (signal_pending(current)) {
7997 vcpu->run->exit_reason = KVM_EXIT_INTR;
7998 ++vcpu->stat.signal_exits;
8001 if (need_resched()) {
8002 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8004 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8008 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8013 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8016 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8017 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8018 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8019 if (r != EMULATE_DONE)
8024 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8026 BUG_ON(!vcpu->arch.pio.count);
8028 return complete_emulated_io(vcpu);
8032 * Implements the following, as a state machine:
8036 * for each mmio piece in the fragment
8044 * for each mmio piece in the fragment
8049 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8051 struct kvm_run *run = vcpu->run;
8052 struct kvm_mmio_fragment *frag;
8055 BUG_ON(!vcpu->mmio_needed);
8057 /* Complete previous fragment */
8058 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8059 len = min(8u, frag->len);
8060 if (!vcpu->mmio_is_write)
8061 memcpy(frag->data, run->mmio.data, len);
8063 if (frag->len <= 8) {
8064 /* Switch to the next fragment. */
8066 vcpu->mmio_cur_fragment++;
8068 /* Go forward to the next mmio piece. */
8074 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8075 vcpu->mmio_needed = 0;
8077 /* FIXME: return into emulator if single-stepping. */
8078 if (vcpu->mmio_is_write)
8080 vcpu->mmio_read_completed = 1;
8081 return complete_emulated_io(vcpu);
8084 run->exit_reason = KVM_EXIT_MMIO;
8085 run->mmio.phys_addr = frag->gpa;
8086 if (vcpu->mmio_is_write)
8087 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8088 run->mmio.len = min(8u, frag->len);
8089 run->mmio.is_write = vcpu->mmio_is_write;
8090 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8094 /* Swap (qemu) user FPU context for the guest FPU context. */
8095 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8098 copy_fpregs_to_fpstate(¤t->thread.fpu);
8099 /* PKRU is separately restored in kvm_x86_ops->run. */
8100 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8101 ~XFEATURE_MASK_PKRU);
8106 /* When vcpu_run ends, restore user space FPU context. */
8107 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8110 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8111 copy_kernel_to_fpregs(¤t->thread.fpu.state);
8113 ++vcpu->stat.fpu_reload;
8117 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8122 kvm_sigset_activate(vcpu);
8123 kvm_load_guest_fpu(vcpu);
8125 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8126 if (kvm_run->immediate_exit) {
8130 kvm_vcpu_block(vcpu);
8131 kvm_apic_accept_events(vcpu);
8132 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8134 if (signal_pending(current)) {
8136 vcpu->run->exit_reason = KVM_EXIT_INTR;
8137 ++vcpu->stat.signal_exits;
8142 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8147 if (vcpu->run->kvm_dirty_regs) {
8148 r = sync_regs(vcpu);
8153 /* re-sync apic's tpr */
8154 if (!lapic_in_kernel(vcpu)) {
8155 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8161 if (unlikely(vcpu->arch.complete_userspace_io)) {
8162 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8163 vcpu->arch.complete_userspace_io = NULL;
8168 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8170 if (kvm_run->immediate_exit)
8176 kvm_put_guest_fpu(vcpu);
8177 if (vcpu->run->kvm_valid_regs)
8179 post_kvm_run_save(vcpu);
8180 kvm_sigset_deactivate(vcpu);
8186 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8188 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8190 * We are here if userspace calls get_regs() in the middle of
8191 * instruction emulation. Registers state needs to be copied
8192 * back from emulation context to vcpu. Userspace shouldn't do
8193 * that usually, but some bad designed PV devices (vmware
8194 * backdoor interface) need this to work
8196 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8197 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8199 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
8200 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
8201 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
8202 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
8203 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
8204 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
8205 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8206 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
8207 #ifdef CONFIG_X86_64
8208 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
8209 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
8210 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
8211 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
8212 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
8213 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
8214 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
8215 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
8218 regs->rip = kvm_rip_read(vcpu);
8219 regs->rflags = kvm_get_rflags(vcpu);
8222 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8225 __get_regs(vcpu, regs);
8230 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8232 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8233 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8235 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
8236 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
8237 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
8238 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
8239 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
8240 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
8241 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
8242 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
8243 #ifdef CONFIG_X86_64
8244 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
8245 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
8246 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
8247 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
8248 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
8249 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
8250 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
8251 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
8254 kvm_rip_write(vcpu, regs->rip);
8255 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8257 vcpu->arch.exception.pending = false;
8259 kvm_make_request(KVM_REQ_EVENT, vcpu);
8262 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8265 __set_regs(vcpu, regs);
8270 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8272 struct kvm_segment cs;
8274 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8278 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8280 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8284 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8285 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8286 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8287 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8288 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8289 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8291 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8292 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8294 kvm_x86_ops->get_idt(vcpu, &dt);
8295 sregs->idt.limit = dt.size;
8296 sregs->idt.base = dt.address;
8297 kvm_x86_ops->get_gdt(vcpu, &dt);
8298 sregs->gdt.limit = dt.size;
8299 sregs->gdt.base = dt.address;
8301 sregs->cr0 = kvm_read_cr0(vcpu);
8302 sregs->cr2 = vcpu->arch.cr2;
8303 sregs->cr3 = kvm_read_cr3(vcpu);
8304 sregs->cr4 = kvm_read_cr4(vcpu);
8305 sregs->cr8 = kvm_get_cr8(vcpu);
8306 sregs->efer = vcpu->arch.efer;
8307 sregs->apic_base = kvm_get_apic_base(vcpu);
8309 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8311 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8312 set_bit(vcpu->arch.interrupt.nr,
8313 (unsigned long *)sregs->interrupt_bitmap);
8316 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8317 struct kvm_sregs *sregs)
8320 __get_sregs(vcpu, sregs);
8325 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8326 struct kvm_mp_state *mp_state)
8330 kvm_apic_accept_events(vcpu);
8331 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8332 vcpu->arch.pv.pv_unhalted)
8333 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8335 mp_state->mp_state = vcpu->arch.mp_state;
8341 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8342 struct kvm_mp_state *mp_state)
8348 if (!lapic_in_kernel(vcpu) &&
8349 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8352 /* INITs are latched while in SMM */
8353 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8354 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8355 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8358 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8359 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8360 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8362 vcpu->arch.mp_state = mp_state->mp_state;
8363 kvm_make_request(KVM_REQ_EVENT, vcpu);
8371 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8372 int reason, bool has_error_code, u32 error_code)
8374 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8377 init_emulate_ctxt(vcpu);
8379 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8380 has_error_code, error_code);
8383 return EMULATE_FAIL;
8385 kvm_rip_write(vcpu, ctxt->eip);
8386 kvm_set_rflags(vcpu, ctxt->eflags);
8387 kvm_make_request(KVM_REQ_EVENT, vcpu);
8388 return EMULATE_DONE;
8390 EXPORT_SYMBOL_GPL(kvm_task_switch);
8392 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8394 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8395 (sregs->cr4 & X86_CR4_OSXSAVE))
8398 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8400 * When EFER.LME and CR0.PG are set, the processor is in
8401 * 64-bit mode (though maybe in a 32-bit code segment).
8402 * CR4.PAE and EFER.LMA must be set.
8404 if (!(sregs->cr4 & X86_CR4_PAE)
8405 || !(sregs->efer & EFER_LMA))
8409 * Not in 64-bit mode: EFER.LMA is clear and the code
8410 * segment cannot be 64-bit.
8412 if (sregs->efer & EFER_LMA || sregs->cs.l)
8419 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8421 struct msr_data apic_base_msr;
8422 int mmu_reset_needed = 0;
8423 int cpuid_update_needed = 0;
8424 int pending_vec, max_bits, idx;
8428 if (kvm_valid_sregs(vcpu, sregs))
8431 apic_base_msr.data = sregs->apic_base;
8432 apic_base_msr.host_initiated = true;
8433 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8436 dt.size = sregs->idt.limit;
8437 dt.address = sregs->idt.base;
8438 kvm_x86_ops->set_idt(vcpu, &dt);
8439 dt.size = sregs->gdt.limit;
8440 dt.address = sregs->gdt.base;
8441 kvm_x86_ops->set_gdt(vcpu, &dt);
8443 vcpu->arch.cr2 = sregs->cr2;
8444 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8445 vcpu->arch.cr3 = sregs->cr3;
8446 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8448 kvm_set_cr8(vcpu, sregs->cr8);
8450 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8451 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8453 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8454 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8455 vcpu->arch.cr0 = sregs->cr0;
8457 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8458 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8459 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8460 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8461 if (cpuid_update_needed)
8462 kvm_update_cpuid(vcpu);
8464 idx = srcu_read_lock(&vcpu->kvm->srcu);
8465 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8466 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8467 mmu_reset_needed = 1;
8469 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8471 if (mmu_reset_needed)
8472 kvm_mmu_reset_context(vcpu);
8474 max_bits = KVM_NR_INTERRUPTS;
8475 pending_vec = find_first_bit(
8476 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8477 if (pending_vec < max_bits) {
8478 kvm_queue_interrupt(vcpu, pending_vec, false);
8479 pr_debug("Set back pending irq %d\n", pending_vec);
8482 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8483 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8484 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8485 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8486 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8487 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8489 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8490 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8492 update_cr8_intercept(vcpu);
8494 /* Older userspace won't unhalt the vcpu on reset. */
8495 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8496 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8498 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8500 kvm_make_request(KVM_REQ_EVENT, vcpu);
8507 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8508 struct kvm_sregs *sregs)
8513 ret = __set_sregs(vcpu, sregs);
8518 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8519 struct kvm_guest_debug *dbg)
8521 unsigned long rflags;
8526 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8528 if (vcpu->arch.exception.pending)
8530 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8531 kvm_queue_exception(vcpu, DB_VECTOR);
8533 kvm_queue_exception(vcpu, BP_VECTOR);
8537 * Read rflags as long as potentially injected trace flags are still
8540 rflags = kvm_get_rflags(vcpu);
8542 vcpu->guest_debug = dbg->control;
8543 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8544 vcpu->guest_debug = 0;
8546 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8547 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8548 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8549 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8551 for (i = 0; i < KVM_NR_DB_REGS; i++)
8552 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8554 kvm_update_dr7(vcpu);
8556 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8557 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8558 get_segment_base(vcpu, VCPU_SREG_CS);
8561 * Trigger an rflags update that will inject or remove the trace
8564 kvm_set_rflags(vcpu, rflags);
8566 kvm_x86_ops->update_bp_intercept(vcpu);
8576 * Translate a guest virtual address to a guest physical address.
8578 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8579 struct kvm_translation *tr)
8581 unsigned long vaddr = tr->linear_address;
8587 idx = srcu_read_lock(&vcpu->kvm->srcu);
8588 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8589 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8590 tr->physical_address = gpa;
8591 tr->valid = gpa != UNMAPPED_GVA;
8599 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8601 struct fxregs_state *fxsave;
8605 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8606 memcpy(fpu->fpr, fxsave->st_space, 128);
8607 fpu->fcw = fxsave->cwd;
8608 fpu->fsw = fxsave->swd;
8609 fpu->ftwx = fxsave->twd;
8610 fpu->last_opcode = fxsave->fop;
8611 fpu->last_ip = fxsave->rip;
8612 fpu->last_dp = fxsave->rdp;
8613 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8619 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8621 struct fxregs_state *fxsave;
8625 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8627 memcpy(fxsave->st_space, fpu->fpr, 128);
8628 fxsave->cwd = fpu->fcw;
8629 fxsave->swd = fpu->fsw;
8630 fxsave->twd = fpu->ftwx;
8631 fxsave->fop = fpu->last_opcode;
8632 fxsave->rip = fpu->last_ip;
8633 fxsave->rdp = fpu->last_dp;
8634 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8640 static void store_regs(struct kvm_vcpu *vcpu)
8642 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8644 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8645 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8647 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8648 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8650 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8651 kvm_vcpu_ioctl_x86_get_vcpu_events(
8652 vcpu, &vcpu->run->s.regs.events);
8655 static int sync_regs(struct kvm_vcpu *vcpu)
8657 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8660 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8661 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8662 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8664 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8665 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8667 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8669 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8670 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8671 vcpu, &vcpu->run->s.regs.events))
8673 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8679 static void fx_init(struct kvm_vcpu *vcpu)
8681 fpstate_init(&vcpu->arch.guest_fpu->state);
8682 if (boot_cpu_has(X86_FEATURE_XSAVES))
8683 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
8684 host_xcr0 | XSTATE_COMPACTION_ENABLED;
8687 * Ensure guest xcr0 is valid for loading
8689 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8691 vcpu->arch.cr0 |= X86_CR0_ET;
8694 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8696 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8698 kvmclock_reset(vcpu);
8700 kvm_x86_ops->vcpu_free(vcpu);
8701 free_cpumask_var(wbinvd_dirty_mask);
8704 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8707 struct kvm_vcpu *vcpu;
8709 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8710 printk_once(KERN_WARNING
8711 "kvm: SMP vm created on host with unstable TSC; "
8712 "guest TSC will not be reliable\n");
8714 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8719 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8721 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8722 kvm_vcpu_mtrr_init(vcpu);
8724 kvm_vcpu_reset(vcpu, false);
8725 kvm_init_mmu(vcpu, false);
8730 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8732 struct msr_data msr;
8733 struct kvm *kvm = vcpu->kvm;
8735 kvm_hv_vcpu_postcreate(vcpu);
8737 if (mutex_lock_killable(&vcpu->mutex))
8741 msr.index = MSR_IA32_TSC;
8742 msr.host_initiated = true;
8743 kvm_write_tsc(vcpu, &msr);
8745 mutex_unlock(&vcpu->mutex);
8747 if (!kvmclock_periodic_sync)
8750 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8751 KVMCLOCK_SYNC_PERIOD);
8754 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8756 vcpu->arch.apf.msr_val = 0;
8759 kvm_mmu_unload(vcpu);
8762 kvm_x86_ops->vcpu_free(vcpu);
8765 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8767 kvm_lapic_reset(vcpu, init_event);
8769 vcpu->arch.hflags = 0;
8771 vcpu->arch.smi_pending = 0;
8772 vcpu->arch.smi_count = 0;
8773 atomic_set(&vcpu->arch.nmi_queued, 0);
8774 vcpu->arch.nmi_pending = 0;
8775 vcpu->arch.nmi_injected = false;
8776 kvm_clear_interrupt_queue(vcpu);
8777 kvm_clear_exception_queue(vcpu);
8778 vcpu->arch.exception.pending = false;
8780 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8781 kvm_update_dr0123(vcpu);
8782 vcpu->arch.dr6 = DR6_INIT;
8783 kvm_update_dr6(vcpu);
8784 vcpu->arch.dr7 = DR7_FIXED_1;
8785 kvm_update_dr7(vcpu);
8789 kvm_make_request(KVM_REQ_EVENT, vcpu);
8790 vcpu->arch.apf.msr_val = 0;
8791 vcpu->arch.st.msr_val = 0;
8793 kvmclock_reset(vcpu);
8795 kvm_clear_async_pf_completion_queue(vcpu);
8796 kvm_async_pf_hash_reset(vcpu);
8797 vcpu->arch.apf.halted = false;
8799 if (kvm_mpx_supported()) {
8800 void *mpx_state_buffer;
8803 * To avoid have the INIT path from kvm_apic_has_events() that be
8804 * called with loaded FPU and does not let userspace fix the state.
8807 kvm_put_guest_fpu(vcpu);
8808 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8809 XFEATURE_MASK_BNDREGS);
8810 if (mpx_state_buffer)
8811 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8812 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8813 XFEATURE_MASK_BNDCSR);
8814 if (mpx_state_buffer)
8815 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8817 kvm_load_guest_fpu(vcpu);
8821 kvm_pmu_reset(vcpu);
8822 vcpu->arch.smbase = 0x30000;
8824 vcpu->arch.msr_misc_features_enables = 0;
8826 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8829 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8830 vcpu->arch.regs_avail = ~0;
8831 vcpu->arch.regs_dirty = ~0;
8833 vcpu->arch.ia32_xss = 0;
8835 kvm_x86_ops->vcpu_reset(vcpu, init_event);
8838 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8840 struct kvm_segment cs;
8842 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8843 cs.selector = vector << 8;
8844 cs.base = vector << 12;
8845 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8846 kvm_rip_write(vcpu, 0);
8849 int kvm_arch_hardware_enable(void)
8852 struct kvm_vcpu *vcpu;
8857 bool stable, backwards_tsc = false;
8859 kvm_shared_msr_cpu_online();
8860 ret = kvm_x86_ops->hardware_enable();
8864 local_tsc = rdtsc();
8865 stable = !kvm_check_tsc_unstable();
8866 list_for_each_entry(kvm, &vm_list, vm_list) {
8867 kvm_for_each_vcpu(i, vcpu, kvm) {
8868 if (!stable && vcpu->cpu == smp_processor_id())
8869 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8870 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8871 backwards_tsc = true;
8872 if (vcpu->arch.last_host_tsc > max_tsc)
8873 max_tsc = vcpu->arch.last_host_tsc;
8879 * Sometimes, even reliable TSCs go backwards. This happens on
8880 * platforms that reset TSC during suspend or hibernate actions, but
8881 * maintain synchronization. We must compensate. Fortunately, we can
8882 * detect that condition here, which happens early in CPU bringup,
8883 * before any KVM threads can be running. Unfortunately, we can't
8884 * bring the TSCs fully up to date with real time, as we aren't yet far
8885 * enough into CPU bringup that we know how much real time has actually
8886 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8887 * variables that haven't been updated yet.
8889 * So we simply find the maximum observed TSC above, then record the
8890 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8891 * the adjustment will be applied. Note that we accumulate
8892 * adjustments, in case multiple suspend cycles happen before some VCPU
8893 * gets a chance to run again. In the event that no KVM threads get a
8894 * chance to run, we will miss the entire elapsed period, as we'll have
8895 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8896 * loose cycle time. This isn't too big a deal, since the loss will be
8897 * uniform across all VCPUs (not to mention the scenario is extremely
8898 * unlikely). It is possible that a second hibernate recovery happens
8899 * much faster than a first, causing the observed TSC here to be
8900 * smaller; this would require additional padding adjustment, which is
8901 * why we set last_host_tsc to the local tsc observed here.
8903 * N.B. - this code below runs only on platforms with reliable TSC,
8904 * as that is the only way backwards_tsc is set above. Also note
8905 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8906 * have the same delta_cyc adjustment applied if backwards_tsc
8907 * is detected. Note further, this adjustment is only done once,
8908 * as we reset last_host_tsc on all VCPUs to stop this from being
8909 * called multiple times (one for each physical CPU bringup).
8911 * Platforms with unreliable TSCs don't have to deal with this, they
8912 * will be compensated by the logic in vcpu_load, which sets the TSC to
8913 * catchup mode. This will catchup all VCPUs to real time, but cannot
8914 * guarantee that they stay in perfect synchronization.
8916 if (backwards_tsc) {
8917 u64 delta_cyc = max_tsc - local_tsc;
8918 list_for_each_entry(kvm, &vm_list, vm_list) {
8919 kvm->arch.backwards_tsc_observed = true;
8920 kvm_for_each_vcpu(i, vcpu, kvm) {
8921 vcpu->arch.tsc_offset_adjustment += delta_cyc;
8922 vcpu->arch.last_host_tsc = local_tsc;
8923 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8927 * We have to disable TSC offset matching.. if you were
8928 * booting a VM while issuing an S4 host suspend....
8929 * you may have some problem. Solving this issue is
8930 * left as an exercise to the reader.
8932 kvm->arch.last_tsc_nsec = 0;
8933 kvm->arch.last_tsc_write = 0;
8940 void kvm_arch_hardware_disable(void)
8942 kvm_x86_ops->hardware_disable();
8943 drop_user_return_notifiers();
8946 int kvm_arch_hardware_setup(void)
8950 r = kvm_x86_ops->hardware_setup();
8954 if (kvm_has_tsc_control) {
8956 * Make sure the user can only configure tsc_khz values that
8957 * fit into a signed integer.
8958 * A min value is not calculated because it will always
8959 * be 1 on all machines.
8961 u64 max = min(0x7fffffffULL,
8962 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8963 kvm_max_guest_tsc_khz = max;
8965 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8968 kvm_init_msr_list();
8972 void kvm_arch_hardware_unsetup(void)
8974 kvm_x86_ops->hardware_unsetup();
8977 void kvm_arch_check_processor_compat(void *rtn)
8979 kvm_x86_ops->check_processor_compatibility(rtn);
8982 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8984 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8986 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8988 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8990 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8993 struct static_key kvm_no_apic_vcpu __read_mostly;
8994 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8996 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9001 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9002 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9003 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9004 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9006 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9008 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9013 vcpu->arch.pio_data = page_address(page);
9015 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9017 r = kvm_mmu_create(vcpu);
9019 goto fail_free_pio_data;
9021 if (irqchip_in_kernel(vcpu->kvm)) {
9022 r = kvm_create_lapic(vcpu);
9024 goto fail_mmu_destroy;
9026 static_key_slow_inc(&kvm_no_apic_vcpu);
9028 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9030 if (!vcpu->arch.mce_banks) {
9032 goto fail_free_lapic;
9034 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9036 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
9038 goto fail_free_mce_banks;
9043 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9045 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9047 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9049 kvm_async_pf_hash_reset(vcpu);
9052 vcpu->arch.pending_external_vector = -1;
9053 vcpu->arch.preempted_in_kernel = false;
9055 kvm_hv_vcpu_init(vcpu);
9059 fail_free_mce_banks:
9060 kfree(vcpu->arch.mce_banks);
9062 kvm_free_lapic(vcpu);
9064 kvm_mmu_destroy(vcpu);
9066 free_page((unsigned long)vcpu->arch.pio_data);
9071 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9075 kvm_hv_vcpu_uninit(vcpu);
9076 kvm_pmu_destroy(vcpu);
9077 kfree(vcpu->arch.mce_banks);
9078 kvm_free_lapic(vcpu);
9079 idx = srcu_read_lock(&vcpu->kvm->srcu);
9080 kvm_mmu_destroy(vcpu);
9081 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9082 free_page((unsigned long)vcpu->arch.pio_data);
9083 if (!lapic_in_kernel(vcpu))
9084 static_key_slow_dec(&kvm_no_apic_vcpu);
9087 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9089 vcpu->arch.l1tf_flush_l1d = true;
9090 kvm_x86_ops->sched_in(vcpu, cpu);
9093 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9098 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9099 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9100 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9101 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9102 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9104 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9105 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9106 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9107 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9108 &kvm->arch.irq_sources_bitmap);
9110 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9111 mutex_init(&kvm->arch.apic_map_lock);
9112 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9114 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9115 pvclock_update_vm_gtod_copy(kvm);
9117 kvm->arch.guest_can_read_msr_platform_info = true;
9119 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9120 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9122 kvm_hv_init_vm(kvm);
9123 kvm_page_track_init(kvm);
9124 kvm_mmu_init_vm(kvm);
9126 if (kvm_x86_ops->vm_init)
9127 return kvm_x86_ops->vm_init(kvm);
9132 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9135 kvm_mmu_unload(vcpu);
9139 static void kvm_free_vcpus(struct kvm *kvm)
9142 struct kvm_vcpu *vcpu;
9145 * Unpin any mmu pages first.
9147 kvm_for_each_vcpu(i, vcpu, kvm) {
9148 kvm_clear_async_pf_completion_queue(vcpu);
9149 kvm_unload_vcpu_mmu(vcpu);
9151 kvm_for_each_vcpu(i, vcpu, kvm)
9152 kvm_arch_vcpu_free(vcpu);
9154 mutex_lock(&kvm->lock);
9155 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9156 kvm->vcpus[i] = NULL;
9158 atomic_set(&kvm->online_vcpus, 0);
9159 mutex_unlock(&kvm->lock);
9162 void kvm_arch_sync_events(struct kvm *kvm)
9164 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9165 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9169 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9173 struct kvm_memslots *slots = kvm_memslots(kvm);
9174 struct kvm_memory_slot *slot, old;
9176 /* Called with kvm->slots_lock held. */
9177 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9180 slot = id_to_memslot(slots, id);
9186 * MAP_SHARED to prevent internal slot pages from being moved
9189 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9190 MAP_SHARED | MAP_ANONYMOUS, 0);
9191 if (IS_ERR((void *)hva))
9192 return PTR_ERR((void *)hva);
9201 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9202 struct kvm_userspace_memory_region m;
9204 m.slot = id | (i << 16);
9206 m.guest_phys_addr = gpa;
9207 m.userspace_addr = hva;
9208 m.memory_size = size;
9209 r = __kvm_set_memory_region(kvm, &m);
9215 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9219 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9221 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9225 mutex_lock(&kvm->slots_lock);
9226 r = __x86_set_memory_region(kvm, id, gpa, size);
9227 mutex_unlock(&kvm->slots_lock);
9231 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9233 void kvm_arch_destroy_vm(struct kvm *kvm)
9235 if (current->mm == kvm->mm) {
9237 * Free memory regions allocated on behalf of userspace,
9238 * unless the the memory map has changed due to process exit
9241 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9242 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9243 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9245 if (kvm_x86_ops->vm_destroy)
9246 kvm_x86_ops->vm_destroy(kvm);
9247 kvm_pic_destroy(kvm);
9248 kvm_ioapic_destroy(kvm);
9249 kvm_free_vcpus(kvm);
9250 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9251 kvm_mmu_uninit_vm(kvm);
9252 kvm_page_track_cleanup(kvm);
9253 kvm_hv_destroy_vm(kvm);
9256 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9257 struct kvm_memory_slot *dont)
9261 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9262 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9263 kvfree(free->arch.rmap[i]);
9264 free->arch.rmap[i] = NULL;
9269 if (!dont || free->arch.lpage_info[i - 1] !=
9270 dont->arch.lpage_info[i - 1]) {
9271 kvfree(free->arch.lpage_info[i - 1]);
9272 free->arch.lpage_info[i - 1] = NULL;
9276 kvm_page_track_free_memslot(free, dont);
9279 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9280 unsigned long npages)
9284 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9285 struct kvm_lpage_info *linfo;
9290 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9291 slot->base_gfn, level) + 1;
9293 slot->arch.rmap[i] =
9294 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9296 if (!slot->arch.rmap[i])
9301 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL);
9305 slot->arch.lpage_info[i - 1] = linfo;
9307 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9308 linfo[0].disallow_lpage = 1;
9309 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9310 linfo[lpages - 1].disallow_lpage = 1;
9311 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9313 * If the gfn and userspace address are not aligned wrt each
9314 * other, or if explicitly asked to, disable large page
9315 * support for this slot
9317 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9318 !kvm_largepages_enabled()) {
9321 for (j = 0; j < lpages; ++j)
9322 linfo[j].disallow_lpage = 1;
9326 if (kvm_page_track_create_memslot(slot, npages))
9332 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9333 kvfree(slot->arch.rmap[i]);
9334 slot->arch.rmap[i] = NULL;
9338 kvfree(slot->arch.lpage_info[i - 1]);
9339 slot->arch.lpage_info[i - 1] = NULL;
9344 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
9347 * memslots->generation has been incremented.
9348 * mmio generation may have reached its maximum value.
9350 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
9353 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9354 struct kvm_memory_slot *memslot,
9355 const struct kvm_userspace_memory_region *mem,
9356 enum kvm_mr_change change)
9361 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9362 struct kvm_memory_slot *new)
9364 /* Still write protect RO slot */
9365 if (new->flags & KVM_MEM_READONLY) {
9366 kvm_mmu_slot_remove_write_access(kvm, new);
9371 * Call kvm_x86_ops dirty logging hooks when they are valid.
9373 * kvm_x86_ops->slot_disable_log_dirty is called when:
9375 * - KVM_MR_CREATE with dirty logging is disabled
9376 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9378 * The reason is, in case of PML, we need to set D-bit for any slots
9379 * with dirty logging disabled in order to eliminate unnecessary GPA
9380 * logging in PML buffer (and potential PML buffer full VMEXT). This
9381 * guarantees leaving PML enabled during guest's lifetime won't have
9382 * any additional overhead from PML when guest is running with dirty
9383 * logging disabled for memory slots.
9385 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9386 * to dirty logging mode.
9388 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9390 * In case of write protect:
9392 * Write protect all pages for dirty logging.
9394 * All the sptes including the large sptes which point to this
9395 * slot are set to readonly. We can not create any new large
9396 * spte on this slot until the end of the logging.
9398 * See the comments in fast_page_fault().
9400 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9401 if (kvm_x86_ops->slot_enable_log_dirty)
9402 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9404 kvm_mmu_slot_remove_write_access(kvm, new);
9406 if (kvm_x86_ops->slot_disable_log_dirty)
9407 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9411 void kvm_arch_commit_memory_region(struct kvm *kvm,
9412 const struct kvm_userspace_memory_region *mem,
9413 const struct kvm_memory_slot *old,
9414 const struct kvm_memory_slot *new,
9415 enum kvm_mr_change change)
9417 int nr_mmu_pages = 0;
9419 if (!kvm->arch.n_requested_mmu_pages)
9420 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9423 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9426 * Dirty logging tracks sptes in 4k granularity, meaning that large
9427 * sptes have to be split. If live migration is successful, the guest
9428 * in the source machine will be destroyed and large sptes will be
9429 * created in the destination. However, if the guest continues to run
9430 * in the source machine (for example if live migration fails), small
9431 * sptes will remain around and cause bad performance.
9433 * Scan sptes if dirty logging has been stopped, dropping those
9434 * which can be collapsed into a single large-page spte. Later
9435 * page faults will create the large-page sptes.
9437 if ((change != KVM_MR_DELETE) &&
9438 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9439 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9440 kvm_mmu_zap_collapsible_sptes(kvm, new);
9443 * Set up write protection and/or dirty logging for the new slot.
9445 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9446 * been zapped so no dirty logging staff is needed for old slot. For
9447 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9448 * new and it's also covered when dealing with the new slot.
9450 * FIXME: const-ify all uses of struct kvm_memory_slot.
9452 if (change != KVM_MR_DELETE)
9453 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9456 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9458 kvm_mmu_invalidate_zap_all_pages(kvm);
9461 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9462 struct kvm_memory_slot *slot)
9464 kvm_page_track_flush_slot(kvm, slot);
9467 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9469 return (is_guest_mode(vcpu) &&
9470 kvm_x86_ops->guest_apic_has_interrupt &&
9471 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9474 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9476 if (!list_empty_careful(&vcpu->async_pf.done))
9479 if (kvm_apic_has_events(vcpu))
9482 if (vcpu->arch.pv.pv_unhalted)
9485 if (vcpu->arch.exception.pending)
9488 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9489 (vcpu->arch.nmi_pending &&
9490 kvm_x86_ops->nmi_allowed(vcpu)))
9493 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9494 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9497 if (kvm_arch_interrupt_allowed(vcpu) &&
9498 (kvm_cpu_has_interrupt(vcpu) ||
9499 kvm_guest_apic_has_interrupt(vcpu)))
9502 if (kvm_hv_has_stimer_pending(vcpu))
9508 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9510 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9513 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9515 return vcpu->arch.preempted_in_kernel;
9518 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9520 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9523 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9525 return kvm_x86_ops->interrupt_allowed(vcpu);
9528 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9530 if (is_64_bit_mode(vcpu))
9531 return kvm_rip_read(vcpu);
9532 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9533 kvm_rip_read(vcpu));
9535 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9537 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9539 return kvm_get_linear_rip(vcpu) == linear_rip;
9541 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9543 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9545 unsigned long rflags;
9547 rflags = kvm_x86_ops->get_rflags(vcpu);
9548 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9549 rflags &= ~X86_EFLAGS_TF;
9552 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9554 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9556 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9557 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9558 rflags |= X86_EFLAGS_TF;
9559 kvm_x86_ops->set_rflags(vcpu, rflags);
9562 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9564 __kvm_set_rflags(vcpu, rflags);
9565 kvm_make_request(KVM_REQ_EVENT, vcpu);
9567 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9569 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9573 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9577 r = kvm_mmu_reload(vcpu);
9581 if (!vcpu->arch.mmu->direct_map &&
9582 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9585 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9588 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9590 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9593 static inline u32 kvm_async_pf_next_probe(u32 key)
9595 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9598 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9600 u32 key = kvm_async_pf_hash_fn(gfn);
9602 while (vcpu->arch.apf.gfns[key] != ~0)
9603 key = kvm_async_pf_next_probe(key);
9605 vcpu->arch.apf.gfns[key] = gfn;
9608 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9611 u32 key = kvm_async_pf_hash_fn(gfn);
9613 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9614 (vcpu->arch.apf.gfns[key] != gfn &&
9615 vcpu->arch.apf.gfns[key] != ~0); i++)
9616 key = kvm_async_pf_next_probe(key);
9621 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9623 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9626 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9630 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9632 vcpu->arch.apf.gfns[i] = ~0;
9634 j = kvm_async_pf_next_probe(j);
9635 if (vcpu->arch.apf.gfns[j] == ~0)
9637 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9639 * k lies cyclically in ]i,j]
9641 * |....j i.k.| or |.k..j i...|
9643 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9644 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9649 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9652 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9656 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9659 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9663 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9664 struct kvm_async_pf *work)
9666 struct x86_exception fault;
9668 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9669 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9671 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9672 (vcpu->arch.apf.send_user_only &&
9673 kvm_x86_ops->get_cpl(vcpu) == 0))
9674 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9675 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9676 fault.vector = PF_VECTOR;
9677 fault.error_code_valid = true;
9678 fault.error_code = 0;
9679 fault.nested_page_fault = false;
9680 fault.address = work->arch.token;
9681 fault.async_page_fault = true;
9682 kvm_inject_page_fault(vcpu, &fault);
9686 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9687 struct kvm_async_pf *work)
9689 struct x86_exception fault;
9692 if (work->wakeup_all)
9693 work->arch.token = ~0; /* broadcast wakeup */
9695 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9696 trace_kvm_async_pf_ready(work->arch.token, work->gva);
9698 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9699 !apf_get_user(vcpu, &val)) {
9700 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9701 vcpu->arch.exception.pending &&
9702 vcpu->arch.exception.nr == PF_VECTOR &&
9703 !apf_put_user(vcpu, 0)) {
9704 vcpu->arch.exception.injected = false;
9705 vcpu->arch.exception.pending = false;
9706 vcpu->arch.exception.nr = 0;
9707 vcpu->arch.exception.has_error_code = false;
9708 vcpu->arch.exception.error_code = 0;
9709 vcpu->arch.exception.has_payload = false;
9710 vcpu->arch.exception.payload = 0;
9711 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9712 fault.vector = PF_VECTOR;
9713 fault.error_code_valid = true;
9714 fault.error_code = 0;
9715 fault.nested_page_fault = false;
9716 fault.address = work->arch.token;
9717 fault.async_page_fault = true;
9718 kvm_inject_page_fault(vcpu, &fault);
9721 vcpu->arch.apf.halted = false;
9722 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9725 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9727 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9730 return kvm_can_do_async_pf(vcpu);
9733 void kvm_arch_start_assignment(struct kvm *kvm)
9735 atomic_inc(&kvm->arch.assigned_device_count);
9737 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9739 void kvm_arch_end_assignment(struct kvm *kvm)
9741 atomic_dec(&kvm->arch.assigned_device_count);
9743 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9745 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9747 return atomic_read(&kvm->arch.assigned_device_count);
9749 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9751 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9753 atomic_inc(&kvm->arch.noncoherent_dma_count);
9755 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9757 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9759 atomic_dec(&kvm->arch.noncoherent_dma_count);
9761 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9763 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9765 return atomic_read(&kvm->arch.noncoherent_dma_count);
9767 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9769 bool kvm_arch_has_irq_bypass(void)
9771 return kvm_x86_ops->update_pi_irte != NULL;
9774 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9775 struct irq_bypass_producer *prod)
9777 struct kvm_kernel_irqfd *irqfd =
9778 container_of(cons, struct kvm_kernel_irqfd, consumer);
9780 irqfd->producer = prod;
9782 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9783 prod->irq, irqfd->gsi, 1);
9786 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9787 struct irq_bypass_producer *prod)
9790 struct kvm_kernel_irqfd *irqfd =
9791 container_of(cons, struct kvm_kernel_irqfd, consumer);
9793 WARN_ON(irqfd->producer != prod);
9794 irqfd->producer = NULL;
9797 * When producer of consumer is unregistered, we change back to
9798 * remapped mode, so we can re-use the current implementation
9799 * when the irq is masked/disabled or the consumer side (KVM
9800 * int this case doesn't want to receive the interrupts.
9802 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9804 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9805 " fails: %d\n", irqfd->consumer.token, ret);
9808 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9809 uint32_t guest_irq, bool set)
9811 if (!kvm_x86_ops->update_pi_irte)
9814 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9817 bool kvm_vector_hashing_enabled(void)
9819 return vector_hashing;
9821 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9823 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9824 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9825 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9826 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9827 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9828 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9829 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9830 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9831 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9832 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9833 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9834 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9835 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9836 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9837 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9838 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9839 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9840 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9841 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);