1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 #include <linux/suspend.h>
63 #include <trace/events/kvm.h>
65 #include <asm/debugreg.h>
69 #include <linux/kernel_stat.h>
70 #include <asm/fpu/internal.h> /* Ugh! */
71 #include <asm/pvclock.h>
72 #include <asm/div64.h>
73 #include <asm/irq_remapping.h>
74 #include <asm/mshyperv.h>
75 #include <asm/hypervisor.h>
76 #include <asm/tlbflush.h>
77 #include <asm/intel_pt.h>
78 #include <asm/emulate_prefix.h>
80 #include <clocksource/hyperv_timer.h>
82 #define CREATE_TRACE_POINTS
85 #define MAX_IO_MSRS 256
86 #define KVM_MAX_MCE_BANKS 32
87 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
88 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
90 #define emul_to_vcpu(ctxt) \
91 ((struct kvm_vcpu *)(ctxt)->vcpu)
94 * - enable syscall per default because its emulated by KVM
95 * - enable LME and LMA per default on 64 bit KVM
99 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
101 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
104 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
106 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
108 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
109 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
111 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
112 static void process_nmi(struct kvm_vcpu *vcpu);
113 static void process_smi(struct kvm_vcpu *vcpu);
114 static void enter_smm(struct kvm_vcpu *vcpu);
115 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
116 static void store_regs(struct kvm_vcpu *vcpu);
117 static int sync_regs(struct kvm_vcpu *vcpu);
119 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
120 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
122 struct kvm_x86_ops kvm_x86_ops __read_mostly;
123 EXPORT_SYMBOL_GPL(kvm_x86_ops);
125 #define KVM_X86_OP(func) \
126 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
127 *(((struct kvm_x86_ops *)0)->func));
128 #define KVM_X86_OP_NULL KVM_X86_OP
129 #include <asm/kvm-x86-ops.h>
130 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
131 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
132 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
134 static bool __read_mostly ignore_msrs = 0;
135 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
137 bool __read_mostly report_ignored_msrs = true;
138 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
139 EXPORT_SYMBOL_GPL(report_ignored_msrs);
141 unsigned int min_timer_period_us = 200;
142 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
144 static bool __read_mostly kvmclock_periodic_sync = true;
145 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
147 bool __read_mostly kvm_has_tsc_control;
148 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
149 u32 __read_mostly kvm_max_guest_tsc_khz;
150 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
151 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
152 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
153 u64 __read_mostly kvm_max_tsc_scaling_ratio;
154 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
155 u64 __read_mostly kvm_default_tsc_scaling_ratio;
156 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
157 bool __read_mostly kvm_has_bus_lock_exit;
158 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
160 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
161 static u32 __read_mostly tsc_tolerance_ppm = 250;
162 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
165 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
166 * adaptive tuning starting from default advancement of 1000ns. '0' disables
167 * advancement entirely. Any other value is used as-is and disables adaptive
168 * tuning, i.e. allows privileged userspace to set an exact advancement time.
170 static int __read_mostly lapic_timer_advance_ns = -1;
171 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
173 static bool __read_mostly vector_hashing = true;
174 module_param(vector_hashing, bool, S_IRUGO);
176 bool __read_mostly enable_vmware_backdoor = false;
177 module_param(enable_vmware_backdoor, bool, S_IRUGO);
178 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
180 static bool __read_mostly force_emulation_prefix = false;
181 module_param(force_emulation_prefix, bool, S_IRUGO);
183 int __read_mostly pi_inject_timer = -1;
184 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
187 * Restoring the host value for MSRs that are only consumed when running in
188 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
189 * returns to userspace, i.e. the kernel can run with the guest's value.
191 #define KVM_MAX_NR_USER_RETURN_MSRS 16
193 struct kvm_user_return_msrs {
194 struct user_return_notifier urn;
196 struct kvm_user_return_msr_values {
199 } values[KVM_MAX_NR_USER_RETURN_MSRS];
202 u32 __read_mostly kvm_nr_uret_msrs;
203 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
204 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
205 static struct kvm_user_return_msrs __percpu *user_return_msrs;
207 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
208 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
209 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
210 | XFEATURE_MASK_PKRU)
212 u64 __read_mostly host_efer;
213 EXPORT_SYMBOL_GPL(host_efer);
215 bool __read_mostly allow_smaller_maxphyaddr = 0;
216 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
218 bool __read_mostly enable_apicv = true;
219 EXPORT_SYMBOL_GPL(enable_apicv);
221 u64 __read_mostly host_xss;
222 EXPORT_SYMBOL_GPL(host_xss);
223 u64 __read_mostly supported_xss;
224 EXPORT_SYMBOL_GPL(supported_xss);
226 struct kvm_stats_debugfs_item debugfs_entries[] = {
227 VCPU_STAT("pf_fixed", pf_fixed),
228 VCPU_STAT("pf_guest", pf_guest),
229 VCPU_STAT("tlb_flush", tlb_flush),
230 VCPU_STAT("invlpg", invlpg),
231 VCPU_STAT("exits", exits),
232 VCPU_STAT("io_exits", io_exits),
233 VCPU_STAT("mmio_exits", mmio_exits),
234 VCPU_STAT("signal_exits", signal_exits),
235 VCPU_STAT("irq_window", irq_window_exits),
236 VCPU_STAT("nmi_window", nmi_window_exits),
237 VCPU_STAT("halt_exits", halt_exits),
238 VCPU_STAT("halt_successful_poll", generic.halt_successful_poll),
239 VCPU_STAT("halt_attempted_poll", generic.halt_attempted_poll),
240 VCPU_STAT("halt_poll_invalid", generic.halt_poll_invalid),
241 VCPU_STAT("halt_wakeup", generic.halt_wakeup),
242 VCPU_STAT("hypercalls", hypercalls),
243 VCPU_STAT("request_irq", request_irq_exits),
244 VCPU_STAT("irq_exits", irq_exits),
245 VCPU_STAT("host_state_reload", host_state_reload),
246 VCPU_STAT("fpu_reload", fpu_reload),
247 VCPU_STAT("insn_emulation", insn_emulation),
248 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
249 VCPU_STAT("irq_injections", irq_injections),
250 VCPU_STAT("nmi_injections", nmi_injections),
251 VCPU_STAT("req_event", req_event),
252 VCPU_STAT("l1d_flush", l1d_flush),
253 VCPU_STAT("halt_poll_success_ns", generic.halt_poll_success_ns),
254 VCPU_STAT("halt_poll_fail_ns", generic.halt_poll_fail_ns),
255 VCPU_STAT("nested_run", nested_run),
256 VCPU_STAT("directed_yield_attempted", directed_yield_attempted),
257 VCPU_STAT("directed_yield_successful", directed_yield_successful),
258 VCPU_STAT("guest_mode", guest_mode),
259 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
260 VM_STAT("mmu_pte_write", mmu_pte_write),
261 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
262 VM_STAT("mmu_flooded", mmu_flooded),
263 VM_STAT("mmu_recycled", mmu_recycled),
264 VM_STAT("mmu_cache_miss", mmu_cache_miss),
265 VM_STAT("mmu_unsync", mmu_unsync),
266 VM_STAT("remote_tlb_flush", generic.remote_tlb_flush),
267 VM_STAT("largepages", lpages, .mode = 0444),
268 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
269 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
273 u64 __read_mostly host_xcr0;
274 u64 __read_mostly supported_xcr0;
275 EXPORT_SYMBOL_GPL(supported_xcr0);
277 static struct kmem_cache *x86_fpu_cache;
279 static struct kmem_cache *x86_emulator_cache;
282 * When called, it means the previous get/set msr reached an invalid msr.
283 * Return true if we want to ignore/silent this failed msr access.
285 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
287 const char *op = write ? "wrmsr" : "rdmsr";
290 if (report_ignored_msrs)
291 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
296 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
302 static struct kmem_cache *kvm_alloc_emulator_cache(void)
304 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
305 unsigned int size = sizeof(struct x86_emulate_ctxt);
307 return kmem_cache_create_usercopy("x86_emulator", size,
308 __alignof__(struct x86_emulate_ctxt),
309 SLAB_ACCOUNT, useroffset,
310 size - useroffset, NULL);
313 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
315 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
318 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
319 vcpu->arch.apf.gfns[i] = ~0;
322 static void kvm_on_user_return(struct user_return_notifier *urn)
325 struct kvm_user_return_msrs *msrs
326 = container_of(urn, struct kvm_user_return_msrs, urn);
327 struct kvm_user_return_msr_values *values;
331 * Disabling irqs at this point since the following code could be
332 * interrupted and executed through kvm_arch_hardware_disable()
334 local_irq_save(flags);
335 if (msrs->registered) {
336 msrs->registered = false;
337 user_return_notifier_unregister(urn);
339 local_irq_restore(flags);
340 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
341 values = &msrs->values[slot];
342 if (values->host != values->curr) {
343 wrmsrl(kvm_uret_msrs_list[slot], values->host);
344 values->curr = values->host;
349 static int kvm_probe_user_return_msr(u32 msr)
355 ret = rdmsrl_safe(msr, &val);
358 ret = wrmsrl_safe(msr, val);
364 int kvm_add_user_return_msr(u32 msr)
366 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
368 if (kvm_probe_user_return_msr(msr))
371 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
372 return kvm_nr_uret_msrs++;
374 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
376 int kvm_find_user_return_msr(u32 msr)
380 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
381 if (kvm_uret_msrs_list[i] == msr)
386 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
388 static void kvm_user_return_msr_cpu_online(void)
390 unsigned int cpu = smp_processor_id();
391 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
395 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
396 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
397 msrs->values[i].host = value;
398 msrs->values[i].curr = value;
402 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
404 unsigned int cpu = smp_processor_id();
405 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
408 value = (value & mask) | (msrs->values[slot].host & ~mask);
409 if (value == msrs->values[slot].curr)
411 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
415 msrs->values[slot].curr = value;
416 if (!msrs->registered) {
417 msrs->urn.on_user_return = kvm_on_user_return;
418 user_return_notifier_register(&msrs->urn);
419 msrs->registered = true;
423 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
425 static void drop_user_return_notifiers(void)
427 unsigned int cpu = smp_processor_id();
428 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
430 if (msrs->registered)
431 kvm_on_user_return(&msrs->urn);
434 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
436 return vcpu->arch.apic_base;
438 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
440 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
442 return kvm_apic_mode(kvm_get_apic_base(vcpu));
444 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
446 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
448 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
449 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
450 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
451 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
453 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
455 if (!msr_info->host_initiated) {
456 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
458 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
462 kvm_lapic_set_base(vcpu, msr_info->data);
463 kvm_recalculate_apic_map(vcpu->kvm);
466 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
468 asmlinkage __visible noinstr void kvm_spurious_fault(void)
470 /* Fault while not rebooting. We want the trace. */
471 BUG_ON(!kvm_rebooting);
473 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
475 #define EXCPT_BENIGN 0
476 #define EXCPT_CONTRIBUTORY 1
479 static int exception_class(int vector)
489 return EXCPT_CONTRIBUTORY;
496 #define EXCPT_FAULT 0
498 #define EXCPT_ABORT 2
499 #define EXCPT_INTERRUPT 3
501 static int exception_type(int vector)
505 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
506 return EXCPT_INTERRUPT;
510 /* #DB is trap, as instruction watchpoints are handled elsewhere */
511 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
514 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
517 /* Reserved exceptions will result in fault */
521 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
523 unsigned nr = vcpu->arch.exception.nr;
524 bool has_payload = vcpu->arch.exception.has_payload;
525 unsigned long payload = vcpu->arch.exception.payload;
533 * "Certain debug exceptions may clear bit 0-3. The
534 * remaining contents of the DR6 register are never
535 * cleared by the processor".
537 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
539 * In order to reflect the #DB exception payload in guest
540 * dr6, three components need to be considered: active low
541 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
543 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
544 * In the target guest dr6:
545 * FIXED_1 bits should always be set.
546 * Active low bits should be cleared if 1-setting in payload.
547 * Active high bits should be set if 1-setting in payload.
549 * Note, the payload is compatible with the pending debug
550 * exceptions/exit qualification under VMX, that active_low bits
551 * are active high in payload.
552 * So they need to be flipped for DR6.
554 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
555 vcpu->arch.dr6 |= payload;
556 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
559 * The #DB payload is defined as compatible with the 'pending
560 * debug exceptions' field under VMX, not DR6. While bit 12 is
561 * defined in the 'pending debug exceptions' field (enabled
562 * breakpoint), it is reserved and must be zero in DR6.
564 vcpu->arch.dr6 &= ~BIT(12);
567 vcpu->arch.cr2 = payload;
571 vcpu->arch.exception.has_payload = false;
572 vcpu->arch.exception.payload = 0;
574 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
576 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
577 unsigned nr, bool has_error, u32 error_code,
578 bool has_payload, unsigned long payload, bool reinject)
583 kvm_make_request(KVM_REQ_EVENT, vcpu);
585 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
589 * On vmentry, vcpu->arch.exception.pending is only
590 * true if an event injection was blocked by
591 * nested_run_pending. In that case, however,
592 * vcpu_enter_guest requests an immediate exit,
593 * and the guest shouldn't proceed far enough to
596 WARN_ON_ONCE(vcpu->arch.exception.pending);
597 vcpu->arch.exception.injected = true;
598 if (WARN_ON_ONCE(has_payload)) {
600 * A reinjected event has already
601 * delivered its payload.
607 vcpu->arch.exception.pending = true;
608 vcpu->arch.exception.injected = false;
610 vcpu->arch.exception.has_error_code = has_error;
611 vcpu->arch.exception.nr = nr;
612 vcpu->arch.exception.error_code = error_code;
613 vcpu->arch.exception.has_payload = has_payload;
614 vcpu->arch.exception.payload = payload;
615 if (!is_guest_mode(vcpu))
616 kvm_deliver_exception_payload(vcpu);
620 /* to check exception */
621 prev_nr = vcpu->arch.exception.nr;
622 if (prev_nr == DF_VECTOR) {
623 /* triple fault -> shutdown */
624 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
627 class1 = exception_class(prev_nr);
628 class2 = exception_class(nr);
629 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
630 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
632 * Generate double fault per SDM Table 5-5. Set
633 * exception.pending = true so that the double fault
634 * can trigger a nested vmexit.
636 vcpu->arch.exception.pending = true;
637 vcpu->arch.exception.injected = false;
638 vcpu->arch.exception.has_error_code = true;
639 vcpu->arch.exception.nr = DF_VECTOR;
640 vcpu->arch.exception.error_code = 0;
641 vcpu->arch.exception.has_payload = false;
642 vcpu->arch.exception.payload = 0;
644 /* replace previous exception with a new one in a hope
645 that instruction re-execution will regenerate lost
650 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
652 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
654 EXPORT_SYMBOL_GPL(kvm_queue_exception);
656 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
658 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
660 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
662 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
663 unsigned long payload)
665 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
667 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
669 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
670 u32 error_code, unsigned long payload)
672 kvm_multiple_exception(vcpu, nr, true, error_code,
673 true, payload, false);
676 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
679 kvm_inject_gp(vcpu, 0);
681 return kvm_skip_emulated_instruction(vcpu);
685 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
687 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
689 ++vcpu->stat.pf_guest;
690 vcpu->arch.exception.nested_apf =
691 is_guest_mode(vcpu) && fault->async_page_fault;
692 if (vcpu->arch.exception.nested_apf) {
693 vcpu->arch.apf.nested_apf_token = fault->address;
694 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
696 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
700 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
702 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
703 struct x86_exception *fault)
705 struct kvm_mmu *fault_mmu;
706 WARN_ON_ONCE(fault->vector != PF_VECTOR);
708 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
712 * Invalidate the TLB entry for the faulting address, if it exists,
713 * else the access will fault indefinitely (and to emulate hardware).
715 if ((fault->error_code & PFERR_PRESENT_MASK) &&
716 !(fault->error_code & PFERR_RSVD_MASK))
717 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
718 fault_mmu->root_hpa);
720 fault_mmu->inject_page_fault(vcpu, fault);
721 return fault->nested_page_fault;
723 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
725 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
727 atomic_inc(&vcpu->arch.nmi_queued);
728 kvm_make_request(KVM_REQ_NMI, vcpu);
730 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
732 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
734 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
736 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
738 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
740 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
742 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
745 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
746 * a #GP and return false.
748 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
750 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
752 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
755 EXPORT_SYMBOL_GPL(kvm_require_cpl);
757 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
759 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
762 kvm_queue_exception(vcpu, UD_VECTOR);
765 EXPORT_SYMBOL_GPL(kvm_require_dr);
768 * This function will be used to read from the physical memory of the currently
769 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
770 * can read from guest physical or from the guest's guest physical memory.
772 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
773 gfn_t ngfn, void *data, int offset, int len,
776 struct x86_exception exception;
780 ngpa = gfn_to_gpa(ngfn);
781 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
782 if (real_gfn == UNMAPPED_GVA)
785 real_gfn = gpa_to_gfn(real_gfn);
787 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
789 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
791 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
793 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
797 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
799 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
801 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
802 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
805 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
807 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
808 offset * sizeof(u64), sizeof(pdpte),
809 PFERR_USER_MASK|PFERR_WRITE_MASK);
814 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
815 if ((pdpte[i] & PT_PRESENT_MASK) &&
816 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
823 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
824 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
825 vcpu->arch.pdptrs_from_userspace = false;
831 EXPORT_SYMBOL_GPL(load_pdptrs);
833 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
835 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
837 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
838 kvm_clear_async_pf_completion_queue(vcpu);
839 kvm_async_pf_hash_reset(vcpu);
842 if ((cr0 ^ old_cr0) & update_bits)
843 kvm_mmu_reset_context(vcpu);
845 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
846 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
847 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
848 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
850 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
852 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
854 unsigned long old_cr0 = kvm_read_cr0(vcpu);
855 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
860 if (cr0 & 0xffffffff00000000UL)
864 cr0 &= ~CR0_RESERVED_BITS;
866 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
869 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
873 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
874 (cr0 & X86_CR0_PG)) {
879 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
884 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
885 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
886 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
889 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
892 static_call(kvm_x86_set_cr0)(vcpu, cr0);
894 kvm_post_set_cr0(vcpu, old_cr0, cr0);
898 EXPORT_SYMBOL_GPL(kvm_set_cr0);
900 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
902 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
904 EXPORT_SYMBOL_GPL(kvm_lmsw);
906 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
908 if (vcpu->arch.guest_state_protected)
911 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
913 if (vcpu->arch.xcr0 != host_xcr0)
914 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
916 if (vcpu->arch.xsaves_enabled &&
917 vcpu->arch.ia32_xss != host_xss)
918 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
921 if (static_cpu_has(X86_FEATURE_PKU) &&
922 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
923 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
924 vcpu->arch.pkru != vcpu->arch.host_pkru)
925 __write_pkru(vcpu->arch.pkru);
927 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
929 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
931 if (vcpu->arch.guest_state_protected)
934 if (static_cpu_has(X86_FEATURE_PKU) &&
935 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
936 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
937 vcpu->arch.pkru = rdpkru();
938 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
939 __write_pkru(vcpu->arch.host_pkru);
942 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
944 if (vcpu->arch.xcr0 != host_xcr0)
945 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
947 if (vcpu->arch.xsaves_enabled &&
948 vcpu->arch.ia32_xss != host_xss)
949 wrmsrl(MSR_IA32_XSS, host_xss);
953 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
955 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
958 u64 old_xcr0 = vcpu->arch.xcr0;
961 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
962 if (index != XCR_XFEATURE_ENABLED_MASK)
964 if (!(xcr0 & XFEATURE_MASK_FP))
966 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
970 * Do not allow the guest to set bits that we do not support
971 * saving. However, xcr0 bit 0 is always set, even if the
972 * emulated CPU does not support XSAVE (see fx_init).
974 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
975 if (xcr0 & ~valid_bits)
978 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
979 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
982 if (xcr0 & XFEATURE_MASK_AVX512) {
983 if (!(xcr0 & XFEATURE_MASK_YMM))
985 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
988 vcpu->arch.xcr0 = xcr0;
990 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
991 kvm_update_cpuid_runtime(vcpu);
995 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
997 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
998 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
999 kvm_inject_gp(vcpu, 0);
1003 return kvm_skip_emulated_instruction(vcpu);
1005 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1007 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1009 if (cr4 & cr4_reserved_bits)
1012 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1015 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1017 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1019 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1021 unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1022 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
1024 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1025 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1026 kvm_mmu_reset_context(vcpu);
1028 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1030 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1032 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1033 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1036 if (!kvm_is_valid_cr4(vcpu, cr4))
1039 if (is_long_mode(vcpu)) {
1040 if (!(cr4 & X86_CR4_PAE))
1042 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1044 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1045 && ((cr4 ^ old_cr4) & pdptr_bits)
1046 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1047 kvm_read_cr3(vcpu)))
1050 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1051 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1054 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1055 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1059 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1061 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1065 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1067 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1069 struct kvm_mmu *mmu = vcpu->arch.mmu;
1070 unsigned long roots_to_free = 0;
1074 * If neither the current CR3 nor any of the prev_roots use the given
1075 * PCID, then nothing needs to be done here because a resync will
1076 * happen anyway before switching to any other CR3.
1078 if (kvm_get_active_pcid(vcpu) == pcid) {
1079 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1080 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1083 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1084 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1085 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1087 kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1090 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1092 bool skip_tlb_flush = false;
1093 unsigned long pcid = 0;
1094 #ifdef CONFIG_X86_64
1095 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1098 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1099 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1100 pcid = cr3 & X86_CR3_PCID_MASK;
1104 /* PDPTRs are always reloaded for PAE paging. */
1105 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1106 goto handle_tlb_flush;
1109 * Do not condition the GPA check on long mode, this helper is used to
1110 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1111 * the current vCPU mode is accurate.
1113 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1116 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1119 if (cr3 != kvm_read_cr3(vcpu))
1120 kvm_mmu_new_pgd(vcpu, cr3);
1122 vcpu->arch.cr3 = cr3;
1123 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1127 * A load of CR3 that flushes the TLB flushes only the current PCID,
1128 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1129 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1130 * and it's impossible to use a non-zero PCID when PCID is disabled,
1131 * i.e. only PCID=0 can be relevant.
1133 if (!skip_tlb_flush)
1134 kvm_invalidate_pcid(vcpu, pcid);
1138 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1140 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1142 if (cr8 & CR8_RESERVED_BITS)
1144 if (lapic_in_kernel(vcpu))
1145 kvm_lapic_set_tpr(vcpu, cr8);
1147 vcpu->arch.cr8 = cr8;
1150 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1152 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1154 if (lapic_in_kernel(vcpu))
1155 return kvm_lapic_get_cr8(vcpu);
1157 return vcpu->arch.cr8;
1159 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1161 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1165 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1166 for (i = 0; i < KVM_NR_DB_REGS; i++)
1167 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1168 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1172 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1176 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1177 dr7 = vcpu->arch.guest_debug_dr7;
1179 dr7 = vcpu->arch.dr7;
1180 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1181 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1182 if (dr7 & DR7_BP_EN_MASK)
1183 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1185 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1187 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1189 u64 fixed = DR6_FIXED_1;
1191 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1194 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1195 fixed |= DR6_BUS_LOCK;
1199 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1201 size_t size = ARRAY_SIZE(vcpu->arch.db);
1205 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1206 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1207 vcpu->arch.eff_db[dr] = val;
1211 if (!kvm_dr6_valid(val))
1213 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1217 if (!kvm_dr7_valid(val))
1219 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1220 kvm_update_dr7(vcpu);
1226 EXPORT_SYMBOL_GPL(kvm_set_dr);
1228 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1230 size_t size = ARRAY_SIZE(vcpu->arch.db);
1234 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1238 *val = vcpu->arch.dr6;
1242 *val = vcpu->arch.dr7;
1246 EXPORT_SYMBOL_GPL(kvm_get_dr);
1248 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1250 u32 ecx = kvm_rcx_read(vcpu);
1253 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1254 kvm_inject_gp(vcpu, 0);
1258 kvm_rax_write(vcpu, (u32)data);
1259 kvm_rdx_write(vcpu, data >> 32);
1260 return kvm_skip_emulated_instruction(vcpu);
1262 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1265 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1266 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1268 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1269 * extract the supported MSRs from the related const lists.
1270 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1271 * capabilities of the host cpu. This capabilities test skips MSRs that are
1272 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1273 * may depend on host virtualization features rather than host cpu features.
1276 static const u32 msrs_to_save_all[] = {
1277 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1279 #ifdef CONFIG_X86_64
1280 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1282 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1283 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1285 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1286 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1287 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1288 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1289 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1290 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1291 MSR_IA32_UMWAIT_CONTROL,
1293 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1294 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1295 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1296 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1297 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1298 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1299 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1300 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1301 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1302 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1303 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1304 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1305 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1306 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1307 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1308 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1309 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1310 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1311 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1312 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1313 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1314 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1317 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1318 static unsigned num_msrs_to_save;
1320 static const u32 emulated_msrs_all[] = {
1321 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1322 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1323 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1324 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1325 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1326 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1327 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1329 HV_X64_MSR_VP_INDEX,
1330 HV_X64_MSR_VP_RUNTIME,
1331 HV_X64_MSR_SCONTROL,
1332 HV_X64_MSR_STIMER0_CONFIG,
1333 HV_X64_MSR_VP_ASSIST_PAGE,
1334 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1335 HV_X64_MSR_TSC_EMULATION_STATUS,
1336 HV_X64_MSR_SYNDBG_OPTIONS,
1337 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1338 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1339 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1341 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1342 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1344 MSR_IA32_TSC_ADJUST,
1345 MSR_IA32_TSC_DEADLINE,
1346 MSR_IA32_ARCH_CAPABILITIES,
1347 MSR_IA32_PERF_CAPABILITIES,
1348 MSR_IA32_MISC_ENABLE,
1349 MSR_IA32_MCG_STATUS,
1351 MSR_IA32_MCG_EXT_CTL,
1355 MSR_MISC_FEATURES_ENABLES,
1356 MSR_AMD64_VIRT_SPEC_CTRL,
1361 * The following list leaves out MSRs whose values are determined
1362 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1363 * We always support the "true" VMX control MSRs, even if the host
1364 * processor does not, so I am putting these registers here rather
1365 * than in msrs_to_save_all.
1368 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1369 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1370 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1371 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1373 MSR_IA32_VMX_CR0_FIXED0,
1374 MSR_IA32_VMX_CR4_FIXED0,
1375 MSR_IA32_VMX_VMCS_ENUM,
1376 MSR_IA32_VMX_PROCBASED_CTLS2,
1377 MSR_IA32_VMX_EPT_VPID_CAP,
1378 MSR_IA32_VMX_VMFUNC,
1381 MSR_KVM_POLL_CONTROL,
1384 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1385 static unsigned num_emulated_msrs;
1388 * List of msr numbers which are used to expose MSR-based features that
1389 * can be used by a hypervisor to validate requested CPU features.
1391 static const u32 msr_based_features_all[] = {
1393 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1394 MSR_IA32_VMX_PINBASED_CTLS,
1395 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1396 MSR_IA32_VMX_PROCBASED_CTLS,
1397 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1398 MSR_IA32_VMX_EXIT_CTLS,
1399 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1400 MSR_IA32_VMX_ENTRY_CTLS,
1402 MSR_IA32_VMX_CR0_FIXED0,
1403 MSR_IA32_VMX_CR0_FIXED1,
1404 MSR_IA32_VMX_CR4_FIXED0,
1405 MSR_IA32_VMX_CR4_FIXED1,
1406 MSR_IA32_VMX_VMCS_ENUM,
1407 MSR_IA32_VMX_PROCBASED_CTLS2,
1408 MSR_IA32_VMX_EPT_VPID_CAP,
1409 MSR_IA32_VMX_VMFUNC,
1413 MSR_IA32_ARCH_CAPABILITIES,
1414 MSR_IA32_PERF_CAPABILITIES,
1417 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1418 static unsigned int num_msr_based_features;
1420 static u64 kvm_get_arch_capabilities(void)
1424 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1425 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1428 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1429 * the nested hypervisor runs with NX huge pages. If it is not,
1430 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1431 * L1 guests, so it need not worry about its own (L2) guests.
1433 data |= ARCH_CAP_PSCHANGE_MC_NO;
1436 * If we're doing cache flushes (either "always" or "cond")
1437 * we will do one whenever the guest does a vmlaunch/vmresume.
1438 * If an outer hypervisor is doing the cache flush for us
1439 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1440 * capability to the guest too, and if EPT is disabled we're not
1441 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1442 * require a nested hypervisor to do a flush of its own.
1444 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1445 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1447 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1448 data |= ARCH_CAP_RDCL_NO;
1449 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1450 data |= ARCH_CAP_SSB_NO;
1451 if (!boot_cpu_has_bug(X86_BUG_MDS))
1452 data |= ARCH_CAP_MDS_NO;
1454 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1456 * If RTM=0 because the kernel has disabled TSX, the host might
1457 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1458 * and therefore knows that there cannot be TAA) but keep
1459 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1460 * and we want to allow migrating those guests to tsx=off hosts.
1462 data &= ~ARCH_CAP_TAA_NO;
1463 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1464 data |= ARCH_CAP_TAA_NO;
1467 * Nothing to do here; we emulate TSX_CTRL if present on the
1468 * host so the guest can choose between disabling TSX or
1469 * using VERW to clear CPU buffers.
1476 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1478 switch (msr->index) {
1479 case MSR_IA32_ARCH_CAPABILITIES:
1480 msr->data = kvm_get_arch_capabilities();
1482 case MSR_IA32_UCODE_REV:
1483 rdmsrl_safe(msr->index, &msr->data);
1486 return static_call(kvm_x86_get_msr_feature)(msr);
1491 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1493 struct kvm_msr_entry msr;
1497 r = kvm_get_msr_feature(&msr);
1499 if (r == KVM_MSR_RET_INVALID) {
1500 /* Unconditionally clear the output for simplicity */
1502 if (kvm_msr_ignored_check(index, 0, false))
1514 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1516 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1519 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1522 if (efer & (EFER_LME | EFER_LMA) &&
1523 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1526 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1532 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1534 if (efer & efer_reserved_bits)
1537 return __kvm_valid_efer(vcpu, efer);
1539 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1541 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1543 u64 old_efer = vcpu->arch.efer;
1544 u64 efer = msr_info->data;
1547 if (efer & efer_reserved_bits)
1550 if (!msr_info->host_initiated) {
1551 if (!__kvm_valid_efer(vcpu, efer))
1554 if (is_paging(vcpu) &&
1555 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1560 efer |= vcpu->arch.efer & EFER_LMA;
1562 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1568 /* Update reserved bits */
1569 if ((efer ^ old_efer) & EFER_NX)
1570 kvm_mmu_reset_context(vcpu);
1575 void kvm_enable_efer_bits(u64 mask)
1577 efer_reserved_bits &= ~mask;
1579 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1581 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1583 struct kvm_x86_msr_filter *msr_filter;
1584 struct msr_bitmap_range *ranges;
1585 struct kvm *kvm = vcpu->kvm;
1590 /* x2APIC MSRs do not support filtering. */
1591 if (index >= 0x800 && index <= 0x8ff)
1594 idx = srcu_read_lock(&kvm->srcu);
1596 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1602 allowed = msr_filter->default_allow;
1603 ranges = msr_filter->ranges;
1605 for (i = 0; i < msr_filter->count; i++) {
1606 u32 start = ranges[i].base;
1607 u32 end = start + ranges[i].nmsrs;
1608 u32 flags = ranges[i].flags;
1609 unsigned long *bitmap = ranges[i].bitmap;
1611 if ((index >= start) && (index < end) && (flags & type)) {
1612 allowed = !!test_bit(index - start, bitmap);
1618 srcu_read_unlock(&kvm->srcu, idx);
1622 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1625 * Write @data into the MSR specified by @index. Select MSR specific fault
1626 * checks are bypassed if @host_initiated is %true.
1627 * Returns 0 on success, non-0 otherwise.
1628 * Assumes vcpu_load() was already called.
1630 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1631 bool host_initiated)
1633 struct msr_data msr;
1635 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1636 return KVM_MSR_RET_FILTERED;
1641 case MSR_KERNEL_GS_BASE:
1644 if (is_noncanonical_address(data, vcpu))
1647 case MSR_IA32_SYSENTER_EIP:
1648 case MSR_IA32_SYSENTER_ESP:
1650 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1651 * non-canonical address is written on Intel but not on
1652 * AMD (which ignores the top 32-bits, because it does
1653 * not implement 64-bit SYSENTER).
1655 * 64-bit code should hence be able to write a non-canonical
1656 * value on AMD. Making the address canonical ensures that
1657 * vmentry does not fail on Intel after writing a non-canonical
1658 * value, and that something deterministic happens if the guest
1659 * invokes 64-bit SYSENTER.
1661 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1664 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1667 if (!host_initiated &&
1668 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1669 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1673 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1674 * incomplete and conflicting architectural behavior. Current
1675 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1676 * reserved and always read as zeros. Enforce Intel's reserved
1677 * bits check if and only if the guest CPU is Intel, and clear
1678 * the bits in all other cases. This ensures cross-vendor
1679 * migration will provide consistent behavior for the guest.
1681 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1690 msr.host_initiated = host_initiated;
1692 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1695 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1696 u32 index, u64 data, bool host_initiated)
1698 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1700 if (ret == KVM_MSR_RET_INVALID)
1701 if (kvm_msr_ignored_check(index, data, true))
1708 * Read the MSR specified by @index into @data. Select MSR specific fault
1709 * checks are bypassed if @host_initiated is %true.
1710 * Returns 0 on success, non-0 otherwise.
1711 * Assumes vcpu_load() was already called.
1713 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1714 bool host_initiated)
1716 struct msr_data msr;
1719 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1720 return KVM_MSR_RET_FILTERED;
1724 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1727 if (!host_initiated &&
1728 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1729 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1735 msr.host_initiated = host_initiated;
1737 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1743 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1744 u32 index, u64 *data, bool host_initiated)
1746 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1748 if (ret == KVM_MSR_RET_INVALID) {
1749 /* Unconditionally clear *data for simplicity */
1751 if (kvm_msr_ignored_check(index, 0, false))
1758 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1760 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1762 EXPORT_SYMBOL_GPL(kvm_get_msr);
1764 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1766 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1768 EXPORT_SYMBOL_GPL(kvm_set_msr);
1770 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1772 int err = vcpu->run->msr.error;
1774 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1775 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1778 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1781 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1783 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1786 static u64 kvm_msr_reason(int r)
1789 case KVM_MSR_RET_INVALID:
1790 return KVM_MSR_EXIT_REASON_UNKNOWN;
1791 case KVM_MSR_RET_FILTERED:
1792 return KVM_MSR_EXIT_REASON_FILTER;
1794 return KVM_MSR_EXIT_REASON_INVAL;
1798 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1799 u32 exit_reason, u64 data,
1800 int (*completion)(struct kvm_vcpu *vcpu),
1803 u64 msr_reason = kvm_msr_reason(r);
1805 /* Check if the user wanted to know about this MSR fault */
1806 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1809 vcpu->run->exit_reason = exit_reason;
1810 vcpu->run->msr.error = 0;
1811 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1812 vcpu->run->msr.reason = msr_reason;
1813 vcpu->run->msr.index = index;
1814 vcpu->run->msr.data = data;
1815 vcpu->arch.complete_userspace_io = completion;
1820 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1822 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1823 complete_emulated_rdmsr, r);
1826 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1828 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1829 complete_emulated_wrmsr, r);
1832 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1834 u32 ecx = kvm_rcx_read(vcpu);
1838 r = kvm_get_msr(vcpu, ecx, &data);
1840 /* MSR read failed? See if we should ask user space */
1841 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1842 /* Bounce to user space */
1847 trace_kvm_msr_read(ecx, data);
1849 kvm_rax_write(vcpu, data & -1u);
1850 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1852 trace_kvm_msr_read_ex(ecx);
1855 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1857 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1859 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1861 u32 ecx = kvm_rcx_read(vcpu);
1862 u64 data = kvm_read_edx_eax(vcpu);
1865 r = kvm_set_msr(vcpu, ecx, data);
1867 /* MSR write failed? See if we should ask user space */
1868 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1869 /* Bounce to user space */
1872 /* Signal all other negative errors to userspace */
1877 trace_kvm_msr_write(ecx, data);
1879 trace_kvm_msr_write_ex(ecx, data);
1881 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1883 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1885 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1887 return kvm_skip_emulated_instruction(vcpu);
1889 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1891 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1893 /* Treat an INVD instruction as a NOP and just skip it. */
1894 return kvm_emulate_as_nop(vcpu);
1896 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1898 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1900 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1901 return kvm_emulate_as_nop(vcpu);
1903 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1905 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1907 kvm_queue_exception(vcpu, UD_VECTOR);
1910 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1912 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1914 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1915 return kvm_emulate_as_nop(vcpu);
1917 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1919 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1921 xfer_to_guest_mode_prepare();
1922 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1923 xfer_to_guest_mode_work_pending();
1927 * The fast path for frequent and performance sensitive wrmsr emulation,
1928 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1929 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1930 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1931 * other cases which must be called after interrupts are enabled on the host.
1933 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1935 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1938 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1939 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1940 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1941 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1944 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1945 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1946 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1947 trace_kvm_apic_write(APIC_ICR, (u32)data);
1954 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1956 if (!kvm_can_use_hv_timer(vcpu))
1959 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1963 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1965 u32 msr = kvm_rcx_read(vcpu);
1967 fastpath_t ret = EXIT_FASTPATH_NONE;
1970 case APIC_BASE_MSR + (APIC_ICR >> 4):
1971 data = kvm_read_edx_eax(vcpu);
1972 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1973 kvm_skip_emulated_instruction(vcpu);
1974 ret = EXIT_FASTPATH_EXIT_HANDLED;
1977 case MSR_IA32_TSC_DEADLINE:
1978 data = kvm_read_edx_eax(vcpu);
1979 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1980 kvm_skip_emulated_instruction(vcpu);
1981 ret = EXIT_FASTPATH_REENTER_GUEST;
1988 if (ret != EXIT_FASTPATH_NONE)
1989 trace_kvm_msr_write(msr, data);
1993 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1996 * Adapt set_msr() to msr_io()'s calling convention
1998 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2000 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2003 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2005 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2008 #ifdef CONFIG_X86_64
2009 struct pvclock_clock {
2019 struct pvclock_gtod_data {
2022 struct pvclock_clock clock; /* extract of a clocksource struct */
2023 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2029 static struct pvclock_gtod_data pvclock_gtod_data;
2031 static void update_pvclock_gtod(struct timekeeper *tk)
2033 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2035 write_seqcount_begin(&vdata->seq);
2037 /* copy pvclock gtod data */
2038 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2039 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2040 vdata->clock.mask = tk->tkr_mono.mask;
2041 vdata->clock.mult = tk->tkr_mono.mult;
2042 vdata->clock.shift = tk->tkr_mono.shift;
2043 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2044 vdata->clock.offset = tk->tkr_mono.base;
2046 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2047 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2048 vdata->raw_clock.mask = tk->tkr_raw.mask;
2049 vdata->raw_clock.mult = tk->tkr_raw.mult;
2050 vdata->raw_clock.shift = tk->tkr_raw.shift;
2051 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2052 vdata->raw_clock.offset = tk->tkr_raw.base;
2054 vdata->wall_time_sec = tk->xtime_sec;
2056 vdata->offs_boot = tk->offs_boot;
2058 write_seqcount_end(&vdata->seq);
2061 static s64 get_kvmclock_base_ns(void)
2063 /* Count up from boot time, but with the frequency of the raw clock. */
2064 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2067 static s64 get_kvmclock_base_ns(void)
2069 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2070 return ktime_get_boottime_ns();
2074 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2078 struct pvclock_wall_clock wc;
2085 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2090 ++version; /* first time write, random junk */
2094 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2098 * The guest calculates current wall clock time by adding
2099 * system time (updated by kvm_guest_time_update below) to the
2100 * wall clock specified here. We do the reverse here.
2102 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2104 wc.nsec = do_div(wall_nsec, 1000000000);
2105 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2106 wc.version = version;
2108 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2111 wc_sec_hi = wall_nsec >> 32;
2112 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2113 &wc_sec_hi, sizeof(wc_sec_hi));
2117 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2120 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2121 bool old_msr, bool host_initiated)
2123 struct kvm_arch *ka = &vcpu->kvm->arch;
2125 if (vcpu->vcpu_id == 0 && !host_initiated) {
2126 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2127 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2129 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2132 vcpu->arch.time = system_time;
2133 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2135 /* we verify if the enable bit is set... */
2136 vcpu->arch.pv_time_enabled = false;
2137 if (!(system_time & 1))
2140 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2141 &vcpu->arch.pv_time, system_time & ~1ULL,
2142 sizeof(struct pvclock_vcpu_time_info)))
2143 vcpu->arch.pv_time_enabled = true;
2148 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2150 do_shl32_div32(dividend, divisor);
2154 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2155 s8 *pshift, u32 *pmultiplier)
2163 scaled64 = scaled_hz;
2164 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2169 tps32 = (uint32_t)tps64;
2170 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2171 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2179 *pmultiplier = div_frac(scaled64, tps32);
2182 #ifdef CONFIG_X86_64
2183 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2186 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2187 static unsigned long max_tsc_khz;
2189 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2191 u64 v = (u64)khz * (1000000 + ppm);
2196 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2198 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2202 /* Guest TSC same frequency as host TSC? */
2204 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2208 /* TSC scaling supported? */
2209 if (!kvm_has_tsc_control) {
2210 if (user_tsc_khz > tsc_khz) {
2211 vcpu->arch.tsc_catchup = 1;
2212 vcpu->arch.tsc_always_catchup = 1;
2215 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2220 /* TSC scaling required - calculate ratio */
2221 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2222 user_tsc_khz, tsc_khz);
2224 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2225 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2230 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2234 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2236 u32 thresh_lo, thresh_hi;
2237 int use_scaling = 0;
2239 /* tsc_khz can be zero if TSC calibration fails */
2240 if (user_tsc_khz == 0) {
2241 /* set tsc_scaling_ratio to a safe value */
2242 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2246 /* Compute a scale to convert nanoseconds in TSC cycles */
2247 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2248 &vcpu->arch.virtual_tsc_shift,
2249 &vcpu->arch.virtual_tsc_mult);
2250 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2253 * Compute the variation in TSC rate which is acceptable
2254 * within the range of tolerance and decide if the
2255 * rate being applied is within that bounds of the hardware
2256 * rate. If so, no scaling or compensation need be done.
2258 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2259 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2260 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2261 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2264 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2267 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2269 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2270 vcpu->arch.virtual_tsc_mult,
2271 vcpu->arch.virtual_tsc_shift);
2272 tsc += vcpu->arch.this_tsc_write;
2276 static inline int gtod_is_based_on_tsc(int mode)
2278 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2281 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2283 #ifdef CONFIG_X86_64
2285 struct kvm_arch *ka = &vcpu->kvm->arch;
2286 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2288 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2289 atomic_read(&vcpu->kvm->online_vcpus));
2292 * Once the masterclock is enabled, always perform request in
2293 * order to update it.
2295 * In order to enable masterclock, the host clocksource must be TSC
2296 * and the vcpus need to have matched TSCs. When that happens,
2297 * perform request to enable masterclock.
2299 if (ka->use_master_clock ||
2300 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2301 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2303 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2304 atomic_read(&vcpu->kvm->online_vcpus),
2305 ka->use_master_clock, gtod->clock.vclock_mode);
2310 * Multiply tsc by a fixed point number represented by ratio.
2312 * The most significant 64-N bits (mult) of ratio represent the
2313 * integral part of the fixed point number; the remaining N bits
2314 * (frac) represent the fractional part, ie. ratio represents a fixed
2315 * point number (mult + frac * 2^(-N)).
2317 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2319 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2321 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2324 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2328 if (ratio != kvm_default_tsc_scaling_ratio)
2329 _tsc = __scale_tsc(ratio, tsc);
2333 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2335 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2339 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2341 return target_tsc - tsc;
2344 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2346 return vcpu->arch.l1_tsc_offset +
2347 kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2349 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2351 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2355 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2356 nested_offset = l1_offset;
2358 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2359 kvm_tsc_scaling_ratio_frac_bits);
2361 nested_offset += l2_offset;
2362 return nested_offset;
2364 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2366 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2368 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2369 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2370 kvm_tsc_scaling_ratio_frac_bits);
2372 return l1_multiplier;
2374 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2376 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2378 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2379 vcpu->arch.l1_tsc_offset,
2382 vcpu->arch.l1_tsc_offset = l1_offset;
2385 * If we are here because L1 chose not to trap WRMSR to TSC then
2386 * according to the spec this should set L1's TSC (as opposed to
2387 * setting L1's offset for L2).
2389 if (is_guest_mode(vcpu))
2390 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2392 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2393 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2395 vcpu->arch.tsc_offset = l1_offset;
2397 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2400 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2402 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2404 /* Userspace is changing the multiplier while L2 is active */
2405 if (is_guest_mode(vcpu))
2406 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2408 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2410 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2412 if (kvm_has_tsc_control)
2413 static_call(kvm_x86_write_tsc_multiplier)(
2414 vcpu, vcpu->arch.tsc_scaling_ratio);
2417 static inline bool kvm_check_tsc_unstable(void)
2419 #ifdef CONFIG_X86_64
2421 * TSC is marked unstable when we're running on Hyper-V,
2422 * 'TSC page' clocksource is good.
2424 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2427 return check_tsc_unstable();
2430 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2432 struct kvm *kvm = vcpu->kvm;
2433 u64 offset, ns, elapsed;
2434 unsigned long flags;
2436 bool already_matched;
2437 bool synchronizing = false;
2439 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2440 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2441 ns = get_kvmclock_base_ns();
2442 elapsed = ns - kvm->arch.last_tsc_nsec;
2444 if (vcpu->arch.virtual_tsc_khz) {
2447 * detection of vcpu initialization -- need to sync
2448 * with other vCPUs. This particularly helps to keep
2449 * kvm_clock stable after CPU hotplug
2451 synchronizing = true;
2453 u64 tsc_exp = kvm->arch.last_tsc_write +
2454 nsec_to_cycles(vcpu, elapsed);
2455 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2457 * Special case: TSC write with a small delta (1 second)
2458 * of virtual cycle time against real time is
2459 * interpreted as an attempt to synchronize the CPU.
2461 synchronizing = data < tsc_exp + tsc_hz &&
2462 data + tsc_hz > tsc_exp;
2467 * For a reliable TSC, we can match TSC offsets, and for an unstable
2468 * TSC, we add elapsed time in this computation. We could let the
2469 * compensation code attempt to catch up if we fall behind, but
2470 * it's better to try to match offsets from the beginning.
2472 if (synchronizing &&
2473 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2474 if (!kvm_check_tsc_unstable()) {
2475 offset = kvm->arch.cur_tsc_offset;
2477 u64 delta = nsec_to_cycles(vcpu, elapsed);
2479 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2482 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2485 * We split periods of matched TSC writes into generations.
2486 * For each generation, we track the original measured
2487 * nanosecond time, offset, and write, so if TSCs are in
2488 * sync, we can match exact offset, and if not, we can match
2489 * exact software computation in compute_guest_tsc()
2491 * These values are tracked in kvm->arch.cur_xxx variables.
2493 kvm->arch.cur_tsc_generation++;
2494 kvm->arch.cur_tsc_nsec = ns;
2495 kvm->arch.cur_tsc_write = data;
2496 kvm->arch.cur_tsc_offset = offset;
2501 * We also track th most recent recorded KHZ, write and time to
2502 * allow the matching interval to be extended at each write.
2504 kvm->arch.last_tsc_nsec = ns;
2505 kvm->arch.last_tsc_write = data;
2506 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2508 vcpu->arch.last_guest_tsc = data;
2510 /* Keep track of which generation this VCPU has synchronized to */
2511 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2512 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2513 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2515 kvm_vcpu_write_tsc_offset(vcpu, offset);
2516 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2518 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2520 kvm->arch.nr_vcpus_matched_tsc = 0;
2521 } else if (!already_matched) {
2522 kvm->arch.nr_vcpus_matched_tsc++;
2525 kvm_track_tsc_matching(vcpu);
2526 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2529 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2532 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2533 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2536 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2538 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2539 WARN_ON(adjustment < 0);
2540 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2541 vcpu->arch.l1_tsc_scaling_ratio);
2542 adjust_tsc_offset_guest(vcpu, adjustment);
2545 #ifdef CONFIG_X86_64
2547 static u64 read_tsc(void)
2549 u64 ret = (u64)rdtsc_ordered();
2550 u64 last = pvclock_gtod_data.clock.cycle_last;
2552 if (likely(ret >= last))
2556 * GCC likes to generate cmov here, but this branch is extremely
2557 * predictable (it's just a function of time and the likely is
2558 * very likely) and there's a data dependence, so force GCC
2559 * to generate a branch instead. I don't barrier() because
2560 * we don't actually need a barrier, and if this function
2561 * ever gets inlined it will generate worse code.
2567 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2573 switch (clock->vclock_mode) {
2574 case VDSO_CLOCKMODE_HVCLOCK:
2575 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2577 if (tsc_pg_val != U64_MAX) {
2578 /* TSC page valid */
2579 *mode = VDSO_CLOCKMODE_HVCLOCK;
2580 v = (tsc_pg_val - clock->cycle_last) &
2583 /* TSC page invalid */
2584 *mode = VDSO_CLOCKMODE_NONE;
2587 case VDSO_CLOCKMODE_TSC:
2588 *mode = VDSO_CLOCKMODE_TSC;
2589 *tsc_timestamp = read_tsc();
2590 v = (*tsc_timestamp - clock->cycle_last) &
2594 *mode = VDSO_CLOCKMODE_NONE;
2597 if (*mode == VDSO_CLOCKMODE_NONE)
2598 *tsc_timestamp = v = 0;
2600 return v * clock->mult;
2603 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2605 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2611 seq = read_seqcount_begin(>od->seq);
2612 ns = gtod->raw_clock.base_cycles;
2613 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2614 ns >>= gtod->raw_clock.shift;
2615 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2616 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2622 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2624 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2630 seq = read_seqcount_begin(>od->seq);
2631 ts->tv_sec = gtod->wall_time_sec;
2632 ns = gtod->clock.base_cycles;
2633 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2634 ns >>= gtod->clock.shift;
2635 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2637 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2643 /* returns true if host is using TSC based clocksource */
2644 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2646 /* checked again under seqlock below */
2647 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2650 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2654 /* returns true if host is using TSC based clocksource */
2655 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2658 /* checked again under seqlock below */
2659 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2662 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2668 * Assuming a stable TSC across physical CPUS, and a stable TSC
2669 * across virtual CPUs, the following condition is possible.
2670 * Each numbered line represents an event visible to both
2671 * CPUs at the next numbered event.
2673 * "timespecX" represents host monotonic time. "tscX" represents
2676 * VCPU0 on CPU0 | VCPU1 on CPU1
2678 * 1. read timespec0,tsc0
2679 * 2. | timespec1 = timespec0 + N
2681 * 3. transition to guest | transition to guest
2682 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2683 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2684 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2686 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2689 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2691 * - 0 < N - M => M < N
2693 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2694 * always the case (the difference between two distinct xtime instances
2695 * might be smaller then the difference between corresponding TSC reads,
2696 * when updating guest vcpus pvclock areas).
2698 * To avoid that problem, do not allow visibility of distinct
2699 * system_timestamp/tsc_timestamp values simultaneously: use a master
2700 * copy of host monotonic time values. Update that master copy
2703 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2707 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2709 #ifdef CONFIG_X86_64
2710 struct kvm_arch *ka = &kvm->arch;
2712 bool host_tsc_clocksource, vcpus_matched;
2714 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2715 atomic_read(&kvm->online_vcpus));
2718 * If the host uses TSC clock, then passthrough TSC as stable
2721 host_tsc_clocksource = kvm_get_time_and_clockread(
2722 &ka->master_kernel_ns,
2723 &ka->master_cycle_now);
2725 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2726 && !ka->backwards_tsc_observed
2727 && !ka->boot_vcpu_runs_old_kvmclock;
2729 if (ka->use_master_clock)
2730 atomic_set(&kvm_guest_has_master_clock, 1);
2732 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2733 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2738 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2740 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2743 static void kvm_gen_update_masterclock(struct kvm *kvm)
2745 #ifdef CONFIG_X86_64
2747 struct kvm_vcpu *vcpu;
2748 struct kvm_arch *ka = &kvm->arch;
2749 unsigned long flags;
2751 kvm_hv_invalidate_tsc_page(kvm);
2753 kvm_make_mclock_inprogress_request(kvm);
2755 /* no guest entries from this point */
2756 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2757 pvclock_update_vm_gtod_copy(kvm);
2758 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2760 kvm_for_each_vcpu(i, vcpu, kvm)
2761 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2763 /* guest entries allowed */
2764 kvm_for_each_vcpu(i, vcpu, kvm)
2765 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2769 u64 get_kvmclock_ns(struct kvm *kvm)
2771 struct kvm_arch *ka = &kvm->arch;
2772 struct pvclock_vcpu_time_info hv_clock;
2773 unsigned long flags;
2776 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2777 if (!ka->use_master_clock) {
2778 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2779 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2782 hv_clock.tsc_timestamp = ka->master_cycle_now;
2783 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2784 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2786 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2789 if (__this_cpu_read(cpu_tsc_khz)) {
2790 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2791 &hv_clock.tsc_shift,
2792 &hv_clock.tsc_to_system_mul);
2793 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2795 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2802 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2803 struct gfn_to_hva_cache *cache,
2804 unsigned int offset)
2806 struct kvm_vcpu_arch *vcpu = &v->arch;
2807 struct pvclock_vcpu_time_info guest_hv_clock;
2809 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2810 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2813 /* This VCPU is paused, but it's legal for a guest to read another
2814 * VCPU's kvmclock, so we really have to follow the specification where
2815 * it says that version is odd if data is being modified, and even after
2818 * Version field updates must be kept separate. This is because
2819 * kvm_write_guest_cached might use a "rep movs" instruction, and
2820 * writes within a string instruction are weakly ordered. So there
2821 * are three writes overall.
2823 * As a small optimization, only write the version field in the first
2824 * and third write. The vcpu->pv_time cache is still valid, because the
2825 * version field is the first in the struct.
2827 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2829 if (guest_hv_clock.version & 1)
2830 ++guest_hv_clock.version; /* first time write, random junk */
2832 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2833 kvm_write_guest_offset_cached(v->kvm, cache,
2834 &vcpu->hv_clock, offset,
2835 sizeof(vcpu->hv_clock.version));
2839 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2840 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2842 if (vcpu->pvclock_set_guest_stopped_request) {
2843 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2844 vcpu->pvclock_set_guest_stopped_request = false;
2847 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2849 kvm_write_guest_offset_cached(v->kvm, cache,
2850 &vcpu->hv_clock, offset,
2851 sizeof(vcpu->hv_clock));
2855 vcpu->hv_clock.version++;
2856 kvm_write_guest_offset_cached(v->kvm, cache,
2857 &vcpu->hv_clock, offset,
2858 sizeof(vcpu->hv_clock.version));
2861 static int kvm_guest_time_update(struct kvm_vcpu *v)
2863 unsigned long flags, tgt_tsc_khz;
2864 struct kvm_vcpu_arch *vcpu = &v->arch;
2865 struct kvm_arch *ka = &v->kvm->arch;
2867 u64 tsc_timestamp, host_tsc;
2869 bool use_master_clock;
2875 * If the host uses TSC clock, then passthrough TSC as stable
2878 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2879 use_master_clock = ka->use_master_clock;
2880 if (use_master_clock) {
2881 host_tsc = ka->master_cycle_now;
2882 kernel_ns = ka->master_kernel_ns;
2884 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2886 /* Keep irq disabled to prevent changes to the clock */
2887 local_irq_save(flags);
2888 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2889 if (unlikely(tgt_tsc_khz == 0)) {
2890 local_irq_restore(flags);
2891 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2894 if (!use_master_clock) {
2896 kernel_ns = get_kvmclock_base_ns();
2899 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2902 * We may have to catch up the TSC to match elapsed wall clock
2903 * time for two reasons, even if kvmclock is used.
2904 * 1) CPU could have been running below the maximum TSC rate
2905 * 2) Broken TSC compensation resets the base at each VCPU
2906 * entry to avoid unknown leaps of TSC even when running
2907 * again on the same CPU. This may cause apparent elapsed
2908 * time to disappear, and the guest to stand still or run
2911 if (vcpu->tsc_catchup) {
2912 u64 tsc = compute_guest_tsc(v, kernel_ns);
2913 if (tsc > tsc_timestamp) {
2914 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2915 tsc_timestamp = tsc;
2919 local_irq_restore(flags);
2921 /* With all the info we got, fill in the values */
2923 if (kvm_has_tsc_control)
2924 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
2925 v->arch.l1_tsc_scaling_ratio);
2927 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2928 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2929 &vcpu->hv_clock.tsc_shift,
2930 &vcpu->hv_clock.tsc_to_system_mul);
2931 vcpu->hw_tsc_khz = tgt_tsc_khz;
2934 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2935 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2936 vcpu->last_guest_tsc = tsc_timestamp;
2938 /* If the host uses TSC clocksource, then it is stable */
2940 if (use_master_clock)
2941 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2943 vcpu->hv_clock.flags = pvclock_flags;
2945 if (vcpu->pv_time_enabled)
2946 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2947 if (vcpu->xen.vcpu_info_set)
2948 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2949 offsetof(struct compat_vcpu_info, time));
2950 if (vcpu->xen.vcpu_time_info_set)
2951 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2952 if (v == kvm_get_vcpu(v->kvm, 0))
2953 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2958 * kvmclock updates which are isolated to a given vcpu, such as
2959 * vcpu->cpu migration, should not allow system_timestamp from
2960 * the rest of the vcpus to remain static. Otherwise ntp frequency
2961 * correction applies to one vcpu's system_timestamp but not
2964 * So in those cases, request a kvmclock update for all vcpus.
2965 * We need to rate-limit these requests though, as they can
2966 * considerably slow guests that have a large number of vcpus.
2967 * The time for a remote vcpu to update its kvmclock is bound
2968 * by the delay we use to rate-limit the updates.
2971 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2973 static void kvmclock_update_fn(struct work_struct *work)
2976 struct delayed_work *dwork = to_delayed_work(work);
2977 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2978 kvmclock_update_work);
2979 struct kvm *kvm = container_of(ka, struct kvm, arch);
2980 struct kvm_vcpu *vcpu;
2982 kvm_for_each_vcpu(i, vcpu, kvm) {
2983 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2984 kvm_vcpu_kick(vcpu);
2988 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2990 struct kvm *kvm = v->kvm;
2992 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2993 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2994 KVMCLOCK_UPDATE_DELAY);
2997 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2999 static void kvmclock_sync_fn(struct work_struct *work)
3001 struct delayed_work *dwork = to_delayed_work(work);
3002 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3003 kvmclock_sync_work);
3004 struct kvm *kvm = container_of(ka, struct kvm, arch);
3006 if (!kvmclock_periodic_sync)
3009 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3010 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3011 KVMCLOCK_SYNC_PERIOD);
3015 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3017 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3019 /* McStatusWrEn enabled? */
3020 if (guest_cpuid_is_amd_or_hygon(vcpu))
3021 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3026 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3028 u64 mcg_cap = vcpu->arch.mcg_cap;
3029 unsigned bank_num = mcg_cap & 0xff;
3030 u32 msr = msr_info->index;
3031 u64 data = msr_info->data;
3034 case MSR_IA32_MCG_STATUS:
3035 vcpu->arch.mcg_status = data;
3037 case MSR_IA32_MCG_CTL:
3038 if (!(mcg_cap & MCG_CTL_P) &&
3039 (data || !msr_info->host_initiated))
3041 if (data != 0 && data != ~(u64)0)
3043 vcpu->arch.mcg_ctl = data;
3046 if (msr >= MSR_IA32_MC0_CTL &&
3047 msr < MSR_IA32_MCx_CTL(bank_num)) {
3048 u32 offset = array_index_nospec(
3049 msr - MSR_IA32_MC0_CTL,
3050 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3052 /* only 0 or all 1s can be written to IA32_MCi_CTL
3053 * some Linux kernels though clear bit 10 in bank 4 to
3054 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3055 * this to avoid an uncatched #GP in the guest
3057 if ((offset & 0x3) == 0 &&
3058 data != 0 && (data | (1 << 10)) != ~(u64)0)
3062 if (!msr_info->host_initiated &&
3063 (offset & 0x3) == 1 && data != 0) {
3064 if (!can_set_mci_status(vcpu))
3068 vcpu->arch.mce_banks[offset] = data;
3076 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3078 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3080 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3083 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3085 gpa_t gpa = data & ~0x3f;
3087 /* Bits 4:5 are reserved, Should be zero */
3091 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3092 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3095 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3096 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3099 if (!lapic_in_kernel(vcpu))
3100 return data ? 1 : 0;
3102 vcpu->arch.apf.msr_en_val = data;
3104 if (!kvm_pv_async_pf_enabled(vcpu)) {
3105 kvm_clear_async_pf_completion_queue(vcpu);
3106 kvm_async_pf_hash_reset(vcpu);
3110 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3114 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3115 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3117 kvm_async_pf_wakeup_all(vcpu);
3122 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3124 /* Bits 8-63 are reserved */
3128 if (!lapic_in_kernel(vcpu))
3131 vcpu->arch.apf.msr_int_val = data;
3133 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3138 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3140 vcpu->arch.pv_time_enabled = false;
3141 vcpu->arch.time = 0;
3144 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3146 ++vcpu->stat.tlb_flush;
3147 static_call(kvm_x86_tlb_flush_all)(vcpu);
3150 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3152 ++vcpu->stat.tlb_flush;
3156 * A TLB flush on behalf of the guest is equivalent to
3157 * INVPCID(all), toggling CR4.PGE, etc., which requires
3158 * a forced sync of the shadow page tables. Unload the
3159 * entire MMU here and the subsequent load will sync the
3160 * shadow page tables, and also flush the TLB.
3162 kvm_mmu_unload(vcpu);
3166 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3169 static void record_steal_time(struct kvm_vcpu *vcpu)
3171 struct kvm_host_map map;
3172 struct kvm_steal_time *st;
3174 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3175 kvm_xen_runstate_set_running(vcpu);
3179 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3182 /* -EAGAIN is returned in atomic context so we can just return. */
3183 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3184 &map, &vcpu->arch.st.cache, false))
3188 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3191 * Doing a TLB flush here, on the guest's behalf, can avoid
3194 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3195 u8 st_preempted = xchg(&st->preempted, 0);
3197 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3198 st_preempted & KVM_VCPU_FLUSH_TLB);
3199 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3200 kvm_vcpu_flush_tlb_guest(vcpu);
3205 vcpu->arch.st.preempted = 0;
3207 if (st->version & 1)
3208 st->version += 1; /* first time write, random junk */
3214 st->steal += current->sched_info.run_delay -
3215 vcpu->arch.st.last_steal;
3216 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3222 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3225 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3228 u32 msr = msr_info->index;
3229 u64 data = msr_info->data;
3231 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3232 return kvm_xen_write_hypercall_page(vcpu, data);
3235 case MSR_AMD64_NB_CFG:
3236 case MSR_IA32_UCODE_WRITE:
3237 case MSR_VM_HSAVE_PA:
3238 case MSR_AMD64_PATCH_LOADER:
3239 case MSR_AMD64_BU_CFG2:
3240 case MSR_AMD64_DC_CFG:
3241 case MSR_F15H_EX_CFG:
3244 case MSR_IA32_UCODE_REV:
3245 if (msr_info->host_initiated)
3246 vcpu->arch.microcode_version = data;
3248 case MSR_IA32_ARCH_CAPABILITIES:
3249 if (!msr_info->host_initiated)
3251 vcpu->arch.arch_capabilities = data;
3253 case MSR_IA32_PERF_CAPABILITIES: {
3254 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3256 if (!msr_info->host_initiated)
3258 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3260 if (data & ~msr_ent.data)
3263 vcpu->arch.perf_capabilities = data;
3268 return set_efer(vcpu, msr_info);
3270 data &= ~(u64)0x40; /* ignore flush filter disable */
3271 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3272 data &= ~(u64)0x8; /* ignore TLB cache disable */
3274 /* Handle McStatusWrEn */
3275 if (data == BIT_ULL(18)) {
3276 vcpu->arch.msr_hwcr = data;
3277 } else if (data != 0) {
3278 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3283 case MSR_FAM10H_MMIO_CONF_BASE:
3285 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3290 case 0x200 ... 0x2ff:
3291 return kvm_mtrr_set_msr(vcpu, msr, data);
3292 case MSR_IA32_APICBASE:
3293 return kvm_set_apic_base(vcpu, msr_info);
3294 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3295 return kvm_x2apic_msr_write(vcpu, msr, data);
3296 case MSR_IA32_TSC_DEADLINE:
3297 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3299 case MSR_IA32_TSC_ADJUST:
3300 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3301 if (!msr_info->host_initiated) {
3302 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3303 adjust_tsc_offset_guest(vcpu, adj);
3305 vcpu->arch.ia32_tsc_adjust_msr = data;
3308 case MSR_IA32_MISC_ENABLE:
3309 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3310 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3311 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3313 vcpu->arch.ia32_misc_enable_msr = data;
3314 kvm_update_cpuid_runtime(vcpu);
3316 vcpu->arch.ia32_misc_enable_msr = data;
3319 case MSR_IA32_SMBASE:
3320 if (!msr_info->host_initiated)
3322 vcpu->arch.smbase = data;
3324 case MSR_IA32_POWER_CTL:
3325 vcpu->arch.msr_ia32_power_ctl = data;
3328 if (msr_info->host_initiated) {
3329 kvm_synchronize_tsc(vcpu, data);
3331 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3332 adjust_tsc_offset_guest(vcpu, adj);
3333 vcpu->arch.ia32_tsc_adjust_msr += adj;
3337 if (!msr_info->host_initiated &&
3338 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3341 * KVM supports exposing PT to the guest, but does not support
3342 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3343 * XSAVES/XRSTORS to save/restore PT MSRs.
3345 if (data & ~supported_xss)
3347 vcpu->arch.ia32_xss = data;
3350 if (!msr_info->host_initiated)
3352 vcpu->arch.smi_count = data;
3354 case MSR_KVM_WALL_CLOCK_NEW:
3355 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3358 vcpu->kvm->arch.wall_clock = data;
3359 kvm_write_wall_clock(vcpu->kvm, data, 0);
3361 case MSR_KVM_WALL_CLOCK:
3362 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3365 vcpu->kvm->arch.wall_clock = data;
3366 kvm_write_wall_clock(vcpu->kvm, data, 0);
3368 case MSR_KVM_SYSTEM_TIME_NEW:
3369 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3372 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3374 case MSR_KVM_SYSTEM_TIME:
3375 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3378 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3380 case MSR_KVM_ASYNC_PF_EN:
3381 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3384 if (kvm_pv_enable_async_pf(vcpu, data))
3387 case MSR_KVM_ASYNC_PF_INT:
3388 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3391 if (kvm_pv_enable_async_pf_int(vcpu, data))
3394 case MSR_KVM_ASYNC_PF_ACK:
3395 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3398 vcpu->arch.apf.pageready_pending = false;
3399 kvm_check_async_pf_completion(vcpu);
3402 case MSR_KVM_STEAL_TIME:
3403 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3406 if (unlikely(!sched_info_on()))
3409 if (data & KVM_STEAL_RESERVED_MASK)
3412 vcpu->arch.st.msr_val = data;
3414 if (!(data & KVM_MSR_ENABLED))
3417 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3420 case MSR_KVM_PV_EOI_EN:
3421 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3424 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3428 case MSR_KVM_POLL_CONTROL:
3429 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3432 /* only enable bit supported */
3433 if (data & (-1ULL << 1))
3436 vcpu->arch.msr_kvm_poll_control = data;
3439 case MSR_IA32_MCG_CTL:
3440 case MSR_IA32_MCG_STATUS:
3441 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3442 return set_msr_mce(vcpu, msr_info);
3444 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3445 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3448 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3449 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3450 if (kvm_pmu_is_valid_msr(vcpu, msr))
3451 return kvm_pmu_set_msr(vcpu, msr_info);
3453 if (pr || data != 0)
3454 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3455 "0x%x data 0x%llx\n", msr, data);
3457 case MSR_K7_CLK_CTL:
3459 * Ignore all writes to this no longer documented MSR.
3460 * Writes are only relevant for old K7 processors,
3461 * all pre-dating SVM, but a recommended workaround from
3462 * AMD for these chips. It is possible to specify the
3463 * affected processor models on the command line, hence
3464 * the need to ignore the workaround.
3467 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3468 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3469 case HV_X64_MSR_SYNDBG_OPTIONS:
3470 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3471 case HV_X64_MSR_CRASH_CTL:
3472 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3473 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3474 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3475 case HV_X64_MSR_TSC_EMULATION_STATUS:
3476 return kvm_hv_set_msr_common(vcpu, msr, data,
3477 msr_info->host_initiated);
3478 case MSR_IA32_BBL_CR_CTL3:
3479 /* Drop writes to this legacy MSR -- see rdmsr
3480 * counterpart for further detail.
3482 if (report_ignored_msrs)
3483 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3486 case MSR_AMD64_OSVW_ID_LENGTH:
3487 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3489 vcpu->arch.osvw.length = data;
3491 case MSR_AMD64_OSVW_STATUS:
3492 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3494 vcpu->arch.osvw.status = data;
3496 case MSR_PLATFORM_INFO:
3497 if (!msr_info->host_initiated ||
3498 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3499 cpuid_fault_enabled(vcpu)))
3501 vcpu->arch.msr_platform_info = data;
3503 case MSR_MISC_FEATURES_ENABLES:
3504 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3505 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3506 !supports_cpuid_fault(vcpu)))
3508 vcpu->arch.msr_misc_features_enables = data;
3511 if (kvm_pmu_is_valid_msr(vcpu, msr))
3512 return kvm_pmu_set_msr(vcpu, msr_info);
3513 return KVM_MSR_RET_INVALID;
3517 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3519 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3522 u64 mcg_cap = vcpu->arch.mcg_cap;
3523 unsigned bank_num = mcg_cap & 0xff;
3526 case MSR_IA32_P5_MC_ADDR:
3527 case MSR_IA32_P5_MC_TYPE:
3530 case MSR_IA32_MCG_CAP:
3531 data = vcpu->arch.mcg_cap;
3533 case MSR_IA32_MCG_CTL:
3534 if (!(mcg_cap & MCG_CTL_P) && !host)
3536 data = vcpu->arch.mcg_ctl;
3538 case MSR_IA32_MCG_STATUS:
3539 data = vcpu->arch.mcg_status;
3542 if (msr >= MSR_IA32_MC0_CTL &&
3543 msr < MSR_IA32_MCx_CTL(bank_num)) {
3544 u32 offset = array_index_nospec(
3545 msr - MSR_IA32_MC0_CTL,
3546 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3548 data = vcpu->arch.mce_banks[offset];
3557 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3559 switch (msr_info->index) {
3560 case MSR_IA32_PLATFORM_ID:
3561 case MSR_IA32_EBL_CR_POWERON:
3562 case MSR_IA32_LASTBRANCHFROMIP:
3563 case MSR_IA32_LASTBRANCHTOIP:
3564 case MSR_IA32_LASTINTFROMIP:
3565 case MSR_IA32_LASTINTTOIP:
3566 case MSR_AMD64_SYSCFG:
3567 case MSR_K8_TSEG_ADDR:
3568 case MSR_K8_TSEG_MASK:
3569 case MSR_VM_HSAVE_PA:
3570 case MSR_K8_INT_PENDING_MSG:
3571 case MSR_AMD64_NB_CFG:
3572 case MSR_FAM10H_MMIO_CONF_BASE:
3573 case MSR_AMD64_BU_CFG2:
3574 case MSR_IA32_PERF_CTL:
3575 case MSR_AMD64_DC_CFG:
3576 case MSR_F15H_EX_CFG:
3578 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3579 * limit) MSRs. Just return 0, as we do not want to expose the host
3580 * data here. Do not conditionalize this on CPUID, as KVM does not do
3581 * so for existing CPU-specific MSRs.
3583 case MSR_RAPL_POWER_UNIT:
3584 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3585 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3586 case MSR_PKG_ENERGY_STATUS: /* Total package */
3587 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3590 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3591 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3592 return kvm_pmu_get_msr(vcpu, msr_info);
3593 if (!msr_info->host_initiated)
3597 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3598 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3599 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3600 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3601 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3602 return kvm_pmu_get_msr(vcpu, msr_info);
3605 case MSR_IA32_UCODE_REV:
3606 msr_info->data = vcpu->arch.microcode_version;
3608 case MSR_IA32_ARCH_CAPABILITIES:
3609 if (!msr_info->host_initiated &&
3610 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3612 msr_info->data = vcpu->arch.arch_capabilities;
3614 case MSR_IA32_PERF_CAPABILITIES:
3615 if (!msr_info->host_initiated &&
3616 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3618 msr_info->data = vcpu->arch.perf_capabilities;
3620 case MSR_IA32_POWER_CTL:
3621 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3623 case MSR_IA32_TSC: {
3625 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3626 * even when not intercepted. AMD manual doesn't explicitly
3627 * state this but appears to behave the same.
3629 * On userspace reads and writes, however, we unconditionally
3630 * return L1's TSC value to ensure backwards-compatible
3631 * behavior for migration.
3635 if (msr_info->host_initiated) {
3636 offset = vcpu->arch.l1_tsc_offset;
3637 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3639 offset = vcpu->arch.tsc_offset;
3640 ratio = vcpu->arch.tsc_scaling_ratio;
3643 msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3647 case 0x200 ... 0x2ff:
3648 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3649 case 0xcd: /* fsb frequency */
3653 * MSR_EBC_FREQUENCY_ID
3654 * Conservative value valid for even the basic CPU models.
3655 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3656 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3657 * and 266MHz for model 3, or 4. Set Core Clock
3658 * Frequency to System Bus Frequency Ratio to 1 (bits
3659 * 31:24) even though these are only valid for CPU
3660 * models > 2, however guests may end up dividing or
3661 * multiplying by zero otherwise.
3663 case MSR_EBC_FREQUENCY_ID:
3664 msr_info->data = 1 << 24;
3666 case MSR_IA32_APICBASE:
3667 msr_info->data = kvm_get_apic_base(vcpu);
3669 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3670 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3671 case MSR_IA32_TSC_DEADLINE:
3672 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3674 case MSR_IA32_TSC_ADJUST:
3675 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3677 case MSR_IA32_MISC_ENABLE:
3678 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3680 case MSR_IA32_SMBASE:
3681 if (!msr_info->host_initiated)
3683 msr_info->data = vcpu->arch.smbase;
3686 msr_info->data = vcpu->arch.smi_count;
3688 case MSR_IA32_PERF_STATUS:
3689 /* TSC increment by tick */
3690 msr_info->data = 1000ULL;
3691 /* CPU multiplier */
3692 msr_info->data |= (((uint64_t)4ULL) << 40);
3695 msr_info->data = vcpu->arch.efer;
3697 case MSR_KVM_WALL_CLOCK:
3698 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3701 msr_info->data = vcpu->kvm->arch.wall_clock;
3703 case MSR_KVM_WALL_CLOCK_NEW:
3704 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3707 msr_info->data = vcpu->kvm->arch.wall_clock;
3709 case MSR_KVM_SYSTEM_TIME:
3710 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3713 msr_info->data = vcpu->arch.time;
3715 case MSR_KVM_SYSTEM_TIME_NEW:
3716 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3719 msr_info->data = vcpu->arch.time;
3721 case MSR_KVM_ASYNC_PF_EN:
3722 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3725 msr_info->data = vcpu->arch.apf.msr_en_val;
3727 case MSR_KVM_ASYNC_PF_INT:
3728 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3731 msr_info->data = vcpu->arch.apf.msr_int_val;
3733 case MSR_KVM_ASYNC_PF_ACK:
3734 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3739 case MSR_KVM_STEAL_TIME:
3740 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3743 msr_info->data = vcpu->arch.st.msr_val;
3745 case MSR_KVM_PV_EOI_EN:
3746 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3749 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3751 case MSR_KVM_POLL_CONTROL:
3752 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3755 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3757 case MSR_IA32_P5_MC_ADDR:
3758 case MSR_IA32_P5_MC_TYPE:
3759 case MSR_IA32_MCG_CAP:
3760 case MSR_IA32_MCG_CTL:
3761 case MSR_IA32_MCG_STATUS:
3762 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3763 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3764 msr_info->host_initiated);
3766 if (!msr_info->host_initiated &&
3767 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3769 msr_info->data = vcpu->arch.ia32_xss;
3771 case MSR_K7_CLK_CTL:
3773 * Provide expected ramp-up count for K7. All other
3774 * are set to zero, indicating minimum divisors for
3777 * This prevents guest kernels on AMD host with CPU
3778 * type 6, model 8 and higher from exploding due to
3779 * the rdmsr failing.
3781 msr_info->data = 0x20000000;
3783 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3784 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3785 case HV_X64_MSR_SYNDBG_OPTIONS:
3786 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3787 case HV_X64_MSR_CRASH_CTL:
3788 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3789 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3790 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3791 case HV_X64_MSR_TSC_EMULATION_STATUS:
3792 return kvm_hv_get_msr_common(vcpu,
3793 msr_info->index, &msr_info->data,
3794 msr_info->host_initiated);
3795 case MSR_IA32_BBL_CR_CTL3:
3796 /* This legacy MSR exists but isn't fully documented in current
3797 * silicon. It is however accessed by winxp in very narrow
3798 * scenarios where it sets bit #19, itself documented as
3799 * a "reserved" bit. Best effort attempt to source coherent
3800 * read data here should the balance of the register be
3801 * interpreted by the guest:
3803 * L2 cache control register 3: 64GB range, 256KB size,
3804 * enabled, latency 0x1, configured
3806 msr_info->data = 0xbe702111;
3808 case MSR_AMD64_OSVW_ID_LENGTH:
3809 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3811 msr_info->data = vcpu->arch.osvw.length;
3813 case MSR_AMD64_OSVW_STATUS:
3814 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3816 msr_info->data = vcpu->arch.osvw.status;
3818 case MSR_PLATFORM_INFO:
3819 if (!msr_info->host_initiated &&
3820 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3822 msr_info->data = vcpu->arch.msr_platform_info;
3824 case MSR_MISC_FEATURES_ENABLES:
3825 msr_info->data = vcpu->arch.msr_misc_features_enables;
3828 msr_info->data = vcpu->arch.msr_hwcr;
3831 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3832 return kvm_pmu_get_msr(vcpu, msr_info);
3833 return KVM_MSR_RET_INVALID;
3837 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3840 * Read or write a bunch of msrs. All parameters are kernel addresses.
3842 * @return number of msrs set successfully.
3844 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3845 struct kvm_msr_entry *entries,
3846 int (*do_msr)(struct kvm_vcpu *vcpu,
3847 unsigned index, u64 *data))
3851 for (i = 0; i < msrs->nmsrs; ++i)
3852 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3859 * Read or write a bunch of msrs. Parameters are user addresses.
3861 * @return number of msrs set successfully.
3863 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3864 int (*do_msr)(struct kvm_vcpu *vcpu,
3865 unsigned index, u64 *data),
3868 struct kvm_msrs msrs;
3869 struct kvm_msr_entry *entries;
3874 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3878 if (msrs.nmsrs >= MAX_IO_MSRS)
3881 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3882 entries = memdup_user(user_msrs->entries, size);
3883 if (IS_ERR(entries)) {
3884 r = PTR_ERR(entries);
3888 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3893 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3904 static inline bool kvm_can_mwait_in_guest(void)
3906 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3907 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3908 boot_cpu_has(X86_FEATURE_ARAT);
3911 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3912 struct kvm_cpuid2 __user *cpuid_arg)
3914 struct kvm_cpuid2 cpuid;
3918 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3921 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3926 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3932 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3937 case KVM_CAP_IRQCHIP:
3939 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3940 case KVM_CAP_SET_TSS_ADDR:
3941 case KVM_CAP_EXT_CPUID:
3942 case KVM_CAP_EXT_EMUL_CPUID:
3943 case KVM_CAP_CLOCKSOURCE:
3945 case KVM_CAP_NOP_IO_DELAY:
3946 case KVM_CAP_MP_STATE:
3947 case KVM_CAP_SYNC_MMU:
3948 case KVM_CAP_USER_NMI:
3949 case KVM_CAP_REINJECT_CONTROL:
3950 case KVM_CAP_IRQ_INJECT_STATUS:
3951 case KVM_CAP_IOEVENTFD:
3952 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3954 case KVM_CAP_PIT_STATE2:
3955 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3956 case KVM_CAP_VCPU_EVENTS:
3957 case KVM_CAP_HYPERV:
3958 case KVM_CAP_HYPERV_VAPIC:
3959 case KVM_CAP_HYPERV_SPIN:
3960 case KVM_CAP_HYPERV_SYNIC:
3961 case KVM_CAP_HYPERV_SYNIC2:
3962 case KVM_CAP_HYPERV_VP_INDEX:
3963 case KVM_CAP_HYPERV_EVENTFD:
3964 case KVM_CAP_HYPERV_TLBFLUSH:
3965 case KVM_CAP_HYPERV_SEND_IPI:
3966 case KVM_CAP_HYPERV_CPUID:
3967 case KVM_CAP_HYPERV_ENFORCE_CPUID:
3968 case KVM_CAP_SYS_HYPERV_CPUID:
3969 case KVM_CAP_PCI_SEGMENT:
3970 case KVM_CAP_DEBUGREGS:
3971 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3973 case KVM_CAP_ASYNC_PF:
3974 case KVM_CAP_ASYNC_PF_INT:
3975 case KVM_CAP_GET_TSC_KHZ:
3976 case KVM_CAP_KVMCLOCK_CTRL:
3977 case KVM_CAP_READONLY_MEM:
3978 case KVM_CAP_HYPERV_TIME:
3979 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3980 case KVM_CAP_TSC_DEADLINE_TIMER:
3981 case KVM_CAP_DISABLE_QUIRKS:
3982 case KVM_CAP_SET_BOOT_CPU_ID:
3983 case KVM_CAP_SPLIT_IRQCHIP:
3984 case KVM_CAP_IMMEDIATE_EXIT:
3985 case KVM_CAP_PMU_EVENT_FILTER:
3986 case KVM_CAP_GET_MSR_FEATURES:
3987 case KVM_CAP_MSR_PLATFORM_INFO:
3988 case KVM_CAP_EXCEPTION_PAYLOAD:
3989 case KVM_CAP_SET_GUEST_DEBUG:
3990 case KVM_CAP_LAST_CPU:
3991 case KVM_CAP_X86_USER_SPACE_MSR:
3992 case KVM_CAP_X86_MSR_FILTER:
3993 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3994 #ifdef CONFIG_X86_SGX_KVM
3995 case KVM_CAP_SGX_ATTRIBUTE:
3997 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
3998 case KVM_CAP_SREGS2:
4001 case KVM_CAP_EXIT_HYPERCALL:
4002 r = KVM_EXIT_HYPERCALL_VALID_MASK;
4004 case KVM_CAP_SET_GUEST_DEBUG2:
4005 return KVM_GUESTDBG_VALID_MASK;
4006 #ifdef CONFIG_KVM_XEN
4007 case KVM_CAP_XEN_HVM:
4008 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4009 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4010 KVM_XEN_HVM_CONFIG_SHARED_INFO;
4011 if (sched_info_on())
4012 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4015 case KVM_CAP_SYNC_REGS:
4016 r = KVM_SYNC_X86_VALID_FIELDS;
4018 case KVM_CAP_ADJUST_CLOCK:
4019 r = KVM_CLOCK_TSC_STABLE;
4021 case KVM_CAP_X86_DISABLE_EXITS:
4022 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4023 KVM_X86_DISABLE_EXITS_CSTATE;
4024 if(kvm_can_mwait_in_guest())
4025 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4027 case KVM_CAP_X86_SMM:
4028 /* SMBASE is usually relocated above 1M on modern chipsets,
4029 * and SMM handlers might indeed rely on 4G segment limits,
4030 * so do not report SMM to be available if real mode is
4031 * emulated via vm86 mode. Still, do not go to great lengths
4032 * to avoid userspace's usage of the feature, because it is a
4033 * fringe case that is not enabled except via specific settings
4034 * of the module parameters.
4036 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4039 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4041 case KVM_CAP_NR_VCPUS:
4042 r = KVM_SOFT_MAX_VCPUS;
4044 case KVM_CAP_MAX_VCPUS:
4047 case KVM_CAP_MAX_VCPU_ID:
4048 r = KVM_MAX_VCPU_ID;
4050 case KVM_CAP_PV_MMU: /* obsolete */
4054 r = KVM_MAX_MCE_BANKS;
4057 r = boot_cpu_has(X86_FEATURE_XSAVE);
4059 case KVM_CAP_TSC_CONTROL:
4060 r = kvm_has_tsc_control;
4062 case KVM_CAP_X2APIC_API:
4063 r = KVM_X2APIC_API_VALID_FLAGS;
4065 case KVM_CAP_NESTED_STATE:
4066 r = kvm_x86_ops.nested_ops->get_state ?
4067 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4069 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4070 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4072 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4073 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4075 case KVM_CAP_SMALLER_MAXPHYADDR:
4076 r = (int) allow_smaller_maxphyaddr;
4078 case KVM_CAP_STEAL_TIME:
4079 r = sched_info_on();
4081 case KVM_CAP_X86_BUS_LOCK_EXIT:
4082 if (kvm_has_bus_lock_exit)
4083 r = KVM_BUS_LOCK_DETECTION_OFF |
4084 KVM_BUS_LOCK_DETECTION_EXIT;
4095 long kvm_arch_dev_ioctl(struct file *filp,
4096 unsigned int ioctl, unsigned long arg)
4098 void __user *argp = (void __user *)arg;
4102 case KVM_GET_MSR_INDEX_LIST: {
4103 struct kvm_msr_list __user *user_msr_list = argp;
4104 struct kvm_msr_list msr_list;
4108 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4111 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4112 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4115 if (n < msr_list.nmsrs)
4118 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4119 num_msrs_to_save * sizeof(u32)))
4121 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4123 num_emulated_msrs * sizeof(u32)))
4128 case KVM_GET_SUPPORTED_CPUID:
4129 case KVM_GET_EMULATED_CPUID: {
4130 struct kvm_cpuid2 __user *cpuid_arg = argp;
4131 struct kvm_cpuid2 cpuid;
4134 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4137 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4143 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4148 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4150 if (copy_to_user(argp, &kvm_mce_cap_supported,
4151 sizeof(kvm_mce_cap_supported)))
4155 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4156 struct kvm_msr_list __user *user_msr_list = argp;
4157 struct kvm_msr_list msr_list;
4161 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4164 msr_list.nmsrs = num_msr_based_features;
4165 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4168 if (n < msr_list.nmsrs)
4171 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4172 num_msr_based_features * sizeof(u32)))
4178 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4180 case KVM_GET_SUPPORTED_HV_CPUID:
4181 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4191 static void wbinvd_ipi(void *garbage)
4196 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4198 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4201 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4203 /* Address WBINVD may be executed by guest */
4204 if (need_emulate_wbinvd(vcpu)) {
4205 if (static_call(kvm_x86_has_wbinvd_exit)())
4206 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4207 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4208 smp_call_function_single(vcpu->cpu,
4209 wbinvd_ipi, NULL, 1);
4212 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4214 /* Save host pkru register if supported */
4215 vcpu->arch.host_pkru = read_pkru();
4217 /* Apply any externally detected TSC adjustments (due to suspend) */
4218 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4219 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4220 vcpu->arch.tsc_offset_adjustment = 0;
4221 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4224 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4225 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4226 rdtsc() - vcpu->arch.last_host_tsc;
4228 mark_tsc_unstable("KVM discovered backwards TSC");
4230 if (kvm_check_tsc_unstable()) {
4231 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4232 vcpu->arch.last_guest_tsc);
4233 kvm_vcpu_write_tsc_offset(vcpu, offset);
4234 vcpu->arch.tsc_catchup = 1;
4237 if (kvm_lapic_hv_timer_in_use(vcpu))
4238 kvm_lapic_restart_hv_timer(vcpu);
4241 * On a host with synchronized TSC, there is no need to update
4242 * kvmclock on vcpu->cpu migration
4244 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4245 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4246 if (vcpu->cpu != cpu)
4247 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4251 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4254 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4256 struct kvm_host_map map;
4257 struct kvm_steal_time *st;
4259 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4262 if (vcpu->arch.st.preempted)
4265 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4266 &vcpu->arch.st.cache, true))
4270 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4272 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4274 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4277 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4281 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4282 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4285 * Take the srcu lock as memslots will be accessed to check the gfn
4286 * cache generation against the memslots generation.
4288 idx = srcu_read_lock(&vcpu->kvm->srcu);
4289 if (kvm_xen_msr_enabled(vcpu->kvm))
4290 kvm_xen_runstate_set_preempted(vcpu);
4292 kvm_steal_time_set_preempted(vcpu);
4293 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4295 static_call(kvm_x86_vcpu_put)(vcpu);
4296 vcpu->arch.last_host_tsc = rdtsc();
4298 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4299 * on every vmexit, but if not, we might have a stale dr6 from the
4300 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4305 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4306 struct kvm_lapic_state *s)
4308 if (vcpu->arch.apicv_active)
4309 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4311 return kvm_apic_get_state(vcpu, s);
4314 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4315 struct kvm_lapic_state *s)
4319 r = kvm_apic_set_state(vcpu, s);
4322 update_cr8_intercept(vcpu);
4327 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4330 * We can accept userspace's request for interrupt injection
4331 * as long as we have a place to store the interrupt number.
4332 * The actual injection will happen when the CPU is able to
4333 * deliver the interrupt.
4335 if (kvm_cpu_has_extint(vcpu))
4338 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4339 return (!lapic_in_kernel(vcpu) ||
4340 kvm_apic_accept_pic_intr(vcpu));
4343 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4345 return kvm_arch_interrupt_allowed(vcpu) &&
4346 kvm_cpu_accept_dm_intr(vcpu);
4349 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4350 struct kvm_interrupt *irq)
4352 if (irq->irq >= KVM_NR_INTERRUPTS)
4355 if (!irqchip_in_kernel(vcpu->kvm)) {
4356 kvm_queue_interrupt(vcpu, irq->irq, false);
4357 kvm_make_request(KVM_REQ_EVENT, vcpu);
4362 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4363 * fail for in-kernel 8259.
4365 if (pic_in_kernel(vcpu->kvm))
4368 if (vcpu->arch.pending_external_vector != -1)
4371 vcpu->arch.pending_external_vector = irq->irq;
4372 kvm_make_request(KVM_REQ_EVENT, vcpu);
4376 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4378 kvm_inject_nmi(vcpu);
4383 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4385 kvm_make_request(KVM_REQ_SMI, vcpu);
4390 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4391 struct kvm_tpr_access_ctl *tac)
4395 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4399 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4403 unsigned bank_num = mcg_cap & 0xff, bank;
4406 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4408 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4411 vcpu->arch.mcg_cap = mcg_cap;
4412 /* Init IA32_MCG_CTL to all 1s */
4413 if (mcg_cap & MCG_CTL_P)
4414 vcpu->arch.mcg_ctl = ~(u64)0;
4415 /* Init IA32_MCi_CTL to all 1s */
4416 for (bank = 0; bank < bank_num; bank++)
4417 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4419 static_call(kvm_x86_setup_mce)(vcpu);
4424 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4425 struct kvm_x86_mce *mce)
4427 u64 mcg_cap = vcpu->arch.mcg_cap;
4428 unsigned bank_num = mcg_cap & 0xff;
4429 u64 *banks = vcpu->arch.mce_banks;
4431 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4434 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4435 * reporting is disabled
4437 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4438 vcpu->arch.mcg_ctl != ~(u64)0)
4440 banks += 4 * mce->bank;
4442 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4443 * reporting is disabled for the bank
4445 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4447 if (mce->status & MCI_STATUS_UC) {
4448 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4449 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4450 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4453 if (banks[1] & MCI_STATUS_VAL)
4454 mce->status |= MCI_STATUS_OVER;
4455 banks[2] = mce->addr;
4456 banks[3] = mce->misc;
4457 vcpu->arch.mcg_status = mce->mcg_status;
4458 banks[1] = mce->status;
4459 kvm_queue_exception(vcpu, MC_VECTOR);
4460 } else if (!(banks[1] & MCI_STATUS_VAL)
4461 || !(banks[1] & MCI_STATUS_UC)) {
4462 if (banks[1] & MCI_STATUS_VAL)
4463 mce->status |= MCI_STATUS_OVER;
4464 banks[2] = mce->addr;
4465 banks[3] = mce->misc;
4466 banks[1] = mce->status;
4468 banks[1] |= MCI_STATUS_OVER;
4472 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4473 struct kvm_vcpu_events *events)
4477 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4481 * In guest mode, payload delivery should be deferred,
4482 * so that the L1 hypervisor can intercept #PF before
4483 * CR2 is modified (or intercept #DB before DR6 is
4484 * modified under nVMX). Unless the per-VM capability,
4485 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4486 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4487 * opportunistically defer the exception payload, deliver it if the
4488 * capability hasn't been requested before processing a
4489 * KVM_GET_VCPU_EVENTS.
4491 if (!vcpu->kvm->arch.exception_payload_enabled &&
4492 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4493 kvm_deliver_exception_payload(vcpu);
4496 * The API doesn't provide the instruction length for software
4497 * exceptions, so don't report them. As long as the guest RIP
4498 * isn't advanced, we should expect to encounter the exception
4501 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4502 events->exception.injected = 0;
4503 events->exception.pending = 0;
4505 events->exception.injected = vcpu->arch.exception.injected;
4506 events->exception.pending = vcpu->arch.exception.pending;
4508 * For ABI compatibility, deliberately conflate
4509 * pending and injected exceptions when
4510 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4512 if (!vcpu->kvm->arch.exception_payload_enabled)
4513 events->exception.injected |=
4514 vcpu->arch.exception.pending;
4516 events->exception.nr = vcpu->arch.exception.nr;
4517 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4518 events->exception.error_code = vcpu->arch.exception.error_code;
4519 events->exception_has_payload = vcpu->arch.exception.has_payload;
4520 events->exception_payload = vcpu->arch.exception.payload;
4522 events->interrupt.injected =
4523 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4524 events->interrupt.nr = vcpu->arch.interrupt.nr;
4525 events->interrupt.soft = 0;
4526 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4528 events->nmi.injected = vcpu->arch.nmi_injected;
4529 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4530 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4531 events->nmi.pad = 0;
4533 events->sipi_vector = 0; /* never valid when reporting to user space */
4535 events->smi.smm = is_smm(vcpu);
4536 events->smi.pending = vcpu->arch.smi_pending;
4537 events->smi.smm_inside_nmi =
4538 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4539 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4541 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4542 | KVM_VCPUEVENT_VALID_SHADOW
4543 | KVM_VCPUEVENT_VALID_SMM);
4544 if (vcpu->kvm->arch.exception_payload_enabled)
4545 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4547 memset(&events->reserved, 0, sizeof(events->reserved));
4550 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4552 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4553 struct kvm_vcpu_events *events)
4555 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4556 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4557 | KVM_VCPUEVENT_VALID_SHADOW
4558 | KVM_VCPUEVENT_VALID_SMM
4559 | KVM_VCPUEVENT_VALID_PAYLOAD))
4562 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4563 if (!vcpu->kvm->arch.exception_payload_enabled)
4565 if (events->exception.pending)
4566 events->exception.injected = 0;
4568 events->exception_has_payload = 0;
4570 events->exception.pending = 0;
4571 events->exception_has_payload = 0;
4574 if ((events->exception.injected || events->exception.pending) &&
4575 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4578 /* INITs are latched while in SMM */
4579 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4580 (events->smi.smm || events->smi.pending) &&
4581 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4585 vcpu->arch.exception.injected = events->exception.injected;
4586 vcpu->arch.exception.pending = events->exception.pending;
4587 vcpu->arch.exception.nr = events->exception.nr;
4588 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4589 vcpu->arch.exception.error_code = events->exception.error_code;
4590 vcpu->arch.exception.has_payload = events->exception_has_payload;
4591 vcpu->arch.exception.payload = events->exception_payload;
4593 vcpu->arch.interrupt.injected = events->interrupt.injected;
4594 vcpu->arch.interrupt.nr = events->interrupt.nr;
4595 vcpu->arch.interrupt.soft = events->interrupt.soft;
4596 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4597 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4598 events->interrupt.shadow);
4600 vcpu->arch.nmi_injected = events->nmi.injected;
4601 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4602 vcpu->arch.nmi_pending = events->nmi.pending;
4603 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4605 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4606 lapic_in_kernel(vcpu))
4607 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4609 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4610 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4611 kvm_smm_changed(vcpu, events->smi.smm);
4613 vcpu->arch.smi_pending = events->smi.pending;
4615 if (events->smi.smm) {
4616 if (events->smi.smm_inside_nmi)
4617 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4619 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4622 if (lapic_in_kernel(vcpu)) {
4623 if (events->smi.latched_init)
4624 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4626 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4630 kvm_make_request(KVM_REQ_EVENT, vcpu);
4635 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4636 struct kvm_debugregs *dbgregs)
4640 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4641 kvm_get_dr(vcpu, 6, &val);
4643 dbgregs->dr7 = vcpu->arch.dr7;
4645 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4648 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4649 struct kvm_debugregs *dbgregs)
4654 if (!kvm_dr6_valid(dbgregs->dr6))
4656 if (!kvm_dr7_valid(dbgregs->dr7))
4659 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4660 kvm_update_dr0123(vcpu);
4661 vcpu->arch.dr6 = dbgregs->dr6;
4662 vcpu->arch.dr7 = dbgregs->dr7;
4663 kvm_update_dr7(vcpu);
4668 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4670 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4672 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4673 u64 xstate_bv = xsave->header.xfeatures;
4677 * Copy legacy XSAVE area, to avoid complications with CPUID
4678 * leaves 0 and 1 in the loop below.
4680 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4683 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4684 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4687 * Copy each region from the possibly compacted offset to the
4688 * non-compacted offset.
4690 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4692 u64 xfeature_mask = valid & -valid;
4693 int xfeature_nr = fls64(xfeature_mask) - 1;
4694 void *src = get_xsave_addr(xsave, xfeature_nr);
4697 u32 size, offset, ecx, edx;
4698 cpuid_count(XSTATE_CPUID, xfeature_nr,
4699 &size, &offset, &ecx, &edx);
4700 if (xfeature_nr == XFEATURE_PKRU)
4701 memcpy(dest + offset, &vcpu->arch.pkru,
4702 sizeof(vcpu->arch.pkru));
4704 memcpy(dest + offset, src, size);
4708 valid -= xfeature_mask;
4712 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4714 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4715 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4719 * Copy legacy XSAVE area, to avoid complications with CPUID
4720 * leaves 0 and 1 in the loop below.
4722 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4724 /* Set XSTATE_BV and possibly XCOMP_BV. */
4725 xsave->header.xfeatures = xstate_bv;
4726 if (boot_cpu_has(X86_FEATURE_XSAVES))
4727 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4730 * Copy each region from the non-compacted offset to the
4731 * possibly compacted offset.
4733 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4735 u64 xfeature_mask = valid & -valid;
4736 int xfeature_nr = fls64(xfeature_mask) - 1;
4737 void *dest = get_xsave_addr(xsave, xfeature_nr);
4740 u32 size, offset, ecx, edx;
4741 cpuid_count(XSTATE_CPUID, xfeature_nr,
4742 &size, &offset, &ecx, &edx);
4743 if (xfeature_nr == XFEATURE_PKRU)
4744 memcpy(&vcpu->arch.pkru, src + offset,
4745 sizeof(vcpu->arch.pkru));
4747 memcpy(dest, src + offset, size);
4750 valid -= xfeature_mask;
4754 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4755 struct kvm_xsave *guest_xsave)
4757 if (!vcpu->arch.guest_fpu)
4760 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4761 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4762 fill_xsave((u8 *) guest_xsave->region, vcpu);
4764 memcpy(guest_xsave->region,
4765 &vcpu->arch.guest_fpu->state.fxsave,
4766 sizeof(struct fxregs_state));
4767 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4768 XFEATURE_MASK_FPSSE;
4772 #define XSAVE_MXCSR_OFFSET 24
4774 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4775 struct kvm_xsave *guest_xsave)
4780 if (!vcpu->arch.guest_fpu)
4783 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4784 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4786 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4788 * Here we allow setting states that are not present in
4789 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4790 * with old userspace.
4792 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4794 load_xsave(vcpu, (u8 *)guest_xsave->region);
4796 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4797 mxcsr & ~mxcsr_feature_mask)
4799 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4800 guest_xsave->region, sizeof(struct fxregs_state));
4805 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4806 struct kvm_xcrs *guest_xcrs)
4808 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4809 guest_xcrs->nr_xcrs = 0;
4813 guest_xcrs->nr_xcrs = 1;
4814 guest_xcrs->flags = 0;
4815 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4816 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4819 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4820 struct kvm_xcrs *guest_xcrs)
4824 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4827 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4830 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4831 /* Only support XCR0 currently */
4832 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4833 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4834 guest_xcrs->xcrs[i].value);
4843 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4844 * stopped by the hypervisor. This function will be called from the host only.
4845 * EINVAL is returned when the host attempts to set the flag for a guest that
4846 * does not support pv clocks.
4848 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4850 if (!vcpu->arch.pv_time_enabled)
4852 vcpu->arch.pvclock_set_guest_stopped_request = true;
4853 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4857 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4858 struct kvm_enable_cap *cap)
4861 uint16_t vmcs_version;
4862 void __user *user_ptr;
4868 case KVM_CAP_HYPERV_SYNIC2:
4873 case KVM_CAP_HYPERV_SYNIC:
4874 if (!irqchip_in_kernel(vcpu->kvm))
4876 return kvm_hv_activate_synic(vcpu, cap->cap ==
4877 KVM_CAP_HYPERV_SYNIC2);
4878 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4879 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4881 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4883 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4884 if (copy_to_user(user_ptr, &vmcs_version,
4885 sizeof(vmcs_version)))
4889 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4890 if (!kvm_x86_ops.enable_direct_tlbflush)
4893 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4895 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4896 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
4898 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4899 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4900 if (vcpu->arch.pv_cpuid.enforce)
4901 kvm_update_pv_runtime(vcpu);
4909 long kvm_arch_vcpu_ioctl(struct file *filp,
4910 unsigned int ioctl, unsigned long arg)
4912 struct kvm_vcpu *vcpu = filp->private_data;
4913 void __user *argp = (void __user *)arg;
4916 struct kvm_sregs2 *sregs2;
4917 struct kvm_lapic_state *lapic;
4918 struct kvm_xsave *xsave;
4919 struct kvm_xcrs *xcrs;
4927 case KVM_GET_LAPIC: {
4929 if (!lapic_in_kernel(vcpu))
4931 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4932 GFP_KERNEL_ACCOUNT);
4937 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4941 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4946 case KVM_SET_LAPIC: {
4948 if (!lapic_in_kernel(vcpu))
4950 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4951 if (IS_ERR(u.lapic)) {
4952 r = PTR_ERR(u.lapic);
4956 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4959 case KVM_INTERRUPT: {
4960 struct kvm_interrupt irq;
4963 if (copy_from_user(&irq, argp, sizeof(irq)))
4965 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4969 r = kvm_vcpu_ioctl_nmi(vcpu);
4973 r = kvm_vcpu_ioctl_smi(vcpu);
4976 case KVM_SET_CPUID: {
4977 struct kvm_cpuid __user *cpuid_arg = argp;
4978 struct kvm_cpuid cpuid;
4981 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4983 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4986 case KVM_SET_CPUID2: {
4987 struct kvm_cpuid2 __user *cpuid_arg = argp;
4988 struct kvm_cpuid2 cpuid;
4991 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4993 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4994 cpuid_arg->entries);
4997 case KVM_GET_CPUID2: {
4998 struct kvm_cpuid2 __user *cpuid_arg = argp;
4999 struct kvm_cpuid2 cpuid;
5002 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5004 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5005 cpuid_arg->entries);
5009 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5014 case KVM_GET_MSRS: {
5015 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5016 r = msr_io(vcpu, argp, do_get_msr, 1);
5017 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5020 case KVM_SET_MSRS: {
5021 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5022 r = msr_io(vcpu, argp, do_set_msr, 0);
5023 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5026 case KVM_TPR_ACCESS_REPORTING: {
5027 struct kvm_tpr_access_ctl tac;
5030 if (copy_from_user(&tac, argp, sizeof(tac)))
5032 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5036 if (copy_to_user(argp, &tac, sizeof(tac)))
5041 case KVM_SET_VAPIC_ADDR: {
5042 struct kvm_vapic_addr va;
5046 if (!lapic_in_kernel(vcpu))
5049 if (copy_from_user(&va, argp, sizeof(va)))
5051 idx = srcu_read_lock(&vcpu->kvm->srcu);
5052 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5053 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5056 case KVM_X86_SETUP_MCE: {
5060 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5062 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5065 case KVM_X86_SET_MCE: {
5066 struct kvm_x86_mce mce;
5069 if (copy_from_user(&mce, argp, sizeof(mce)))
5071 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5074 case KVM_GET_VCPU_EVENTS: {
5075 struct kvm_vcpu_events events;
5077 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5080 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5085 case KVM_SET_VCPU_EVENTS: {
5086 struct kvm_vcpu_events events;
5089 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5092 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5095 case KVM_GET_DEBUGREGS: {
5096 struct kvm_debugregs dbgregs;
5098 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5101 if (copy_to_user(argp, &dbgregs,
5102 sizeof(struct kvm_debugregs)))
5107 case KVM_SET_DEBUGREGS: {
5108 struct kvm_debugregs dbgregs;
5111 if (copy_from_user(&dbgregs, argp,
5112 sizeof(struct kvm_debugregs)))
5115 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5118 case KVM_GET_XSAVE: {
5119 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5124 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5127 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5132 case KVM_SET_XSAVE: {
5133 u.xsave = memdup_user(argp, sizeof(*u.xsave));
5134 if (IS_ERR(u.xsave)) {
5135 r = PTR_ERR(u.xsave);
5139 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5142 case KVM_GET_XCRS: {
5143 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5148 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5151 if (copy_to_user(argp, u.xcrs,
5152 sizeof(struct kvm_xcrs)))
5157 case KVM_SET_XCRS: {
5158 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5159 if (IS_ERR(u.xcrs)) {
5160 r = PTR_ERR(u.xcrs);
5164 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5167 case KVM_SET_TSC_KHZ: {
5171 user_tsc_khz = (u32)arg;
5173 if (kvm_has_tsc_control &&
5174 user_tsc_khz >= kvm_max_guest_tsc_khz)
5177 if (user_tsc_khz == 0)
5178 user_tsc_khz = tsc_khz;
5180 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5185 case KVM_GET_TSC_KHZ: {
5186 r = vcpu->arch.virtual_tsc_khz;
5189 case KVM_KVMCLOCK_CTRL: {
5190 r = kvm_set_guest_paused(vcpu);
5193 case KVM_ENABLE_CAP: {
5194 struct kvm_enable_cap cap;
5197 if (copy_from_user(&cap, argp, sizeof(cap)))
5199 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5202 case KVM_GET_NESTED_STATE: {
5203 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5207 if (!kvm_x86_ops.nested_ops->get_state)
5210 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5212 if (get_user(user_data_size, &user_kvm_nested_state->size))
5215 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5220 if (r > user_data_size) {
5221 if (put_user(r, &user_kvm_nested_state->size))
5231 case KVM_SET_NESTED_STATE: {
5232 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5233 struct kvm_nested_state kvm_state;
5237 if (!kvm_x86_ops.nested_ops->set_state)
5241 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5245 if (kvm_state.size < sizeof(kvm_state))
5248 if (kvm_state.flags &
5249 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5250 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5251 | KVM_STATE_NESTED_GIF_SET))
5254 /* nested_run_pending implies guest_mode. */
5255 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5256 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5259 idx = srcu_read_lock(&vcpu->kvm->srcu);
5260 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5261 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5264 case KVM_GET_SUPPORTED_HV_CPUID:
5265 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5267 #ifdef CONFIG_KVM_XEN
5268 case KVM_XEN_VCPU_GET_ATTR: {
5269 struct kvm_xen_vcpu_attr xva;
5272 if (copy_from_user(&xva, argp, sizeof(xva)))
5274 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5275 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5279 case KVM_XEN_VCPU_SET_ATTR: {
5280 struct kvm_xen_vcpu_attr xva;
5283 if (copy_from_user(&xva, argp, sizeof(xva)))
5285 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5289 case KVM_GET_SREGS2: {
5290 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5294 __get_sregs2(vcpu, u.sregs2);
5296 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5301 case KVM_SET_SREGS2: {
5302 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5303 if (IS_ERR(u.sregs2)) {
5304 r = PTR_ERR(u.sregs2);
5308 r = __set_sregs2(vcpu, u.sregs2);
5321 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5323 return VM_FAULT_SIGBUS;
5326 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5330 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5332 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5336 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5339 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5342 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5343 unsigned long kvm_nr_mmu_pages)
5345 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5348 mutex_lock(&kvm->slots_lock);
5350 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5351 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5353 mutex_unlock(&kvm->slots_lock);
5357 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5359 return kvm->arch.n_max_mmu_pages;
5362 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5364 struct kvm_pic *pic = kvm->arch.vpic;
5368 switch (chip->chip_id) {
5369 case KVM_IRQCHIP_PIC_MASTER:
5370 memcpy(&chip->chip.pic, &pic->pics[0],
5371 sizeof(struct kvm_pic_state));
5373 case KVM_IRQCHIP_PIC_SLAVE:
5374 memcpy(&chip->chip.pic, &pic->pics[1],
5375 sizeof(struct kvm_pic_state));
5377 case KVM_IRQCHIP_IOAPIC:
5378 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5387 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5389 struct kvm_pic *pic = kvm->arch.vpic;
5393 switch (chip->chip_id) {
5394 case KVM_IRQCHIP_PIC_MASTER:
5395 spin_lock(&pic->lock);
5396 memcpy(&pic->pics[0], &chip->chip.pic,
5397 sizeof(struct kvm_pic_state));
5398 spin_unlock(&pic->lock);
5400 case KVM_IRQCHIP_PIC_SLAVE:
5401 spin_lock(&pic->lock);
5402 memcpy(&pic->pics[1], &chip->chip.pic,
5403 sizeof(struct kvm_pic_state));
5404 spin_unlock(&pic->lock);
5406 case KVM_IRQCHIP_IOAPIC:
5407 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5413 kvm_pic_update_irq(pic);
5417 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5419 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5421 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5423 mutex_lock(&kps->lock);
5424 memcpy(ps, &kps->channels, sizeof(*ps));
5425 mutex_unlock(&kps->lock);
5429 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5432 struct kvm_pit *pit = kvm->arch.vpit;
5434 mutex_lock(&pit->pit_state.lock);
5435 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5436 for (i = 0; i < 3; i++)
5437 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5438 mutex_unlock(&pit->pit_state.lock);
5442 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5444 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5445 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5446 sizeof(ps->channels));
5447 ps->flags = kvm->arch.vpit->pit_state.flags;
5448 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5449 memset(&ps->reserved, 0, sizeof(ps->reserved));
5453 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5457 u32 prev_legacy, cur_legacy;
5458 struct kvm_pit *pit = kvm->arch.vpit;
5460 mutex_lock(&pit->pit_state.lock);
5461 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5462 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5463 if (!prev_legacy && cur_legacy)
5465 memcpy(&pit->pit_state.channels, &ps->channels,
5466 sizeof(pit->pit_state.channels));
5467 pit->pit_state.flags = ps->flags;
5468 for (i = 0; i < 3; i++)
5469 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5471 mutex_unlock(&pit->pit_state.lock);
5475 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5476 struct kvm_reinject_control *control)
5478 struct kvm_pit *pit = kvm->arch.vpit;
5480 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5481 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5482 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5484 mutex_lock(&pit->pit_state.lock);
5485 kvm_pit_set_reinject(pit, control->pit_reinject);
5486 mutex_unlock(&pit->pit_state.lock);
5491 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5495 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5496 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5497 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5500 struct kvm_vcpu *vcpu;
5503 kvm_for_each_vcpu(i, vcpu, kvm)
5504 kvm_vcpu_kick(vcpu);
5507 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5510 if (!irqchip_in_kernel(kvm))
5513 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5514 irq_event->irq, irq_event->level,
5519 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5520 struct kvm_enable_cap *cap)
5528 case KVM_CAP_DISABLE_QUIRKS:
5529 kvm->arch.disabled_quirks = cap->args[0];
5532 case KVM_CAP_SPLIT_IRQCHIP: {
5533 mutex_lock(&kvm->lock);
5535 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5536 goto split_irqchip_unlock;
5538 if (irqchip_in_kernel(kvm))
5539 goto split_irqchip_unlock;
5540 if (kvm->created_vcpus)
5541 goto split_irqchip_unlock;
5542 r = kvm_setup_empty_irq_routing(kvm);
5544 goto split_irqchip_unlock;
5545 /* Pairs with irqchip_in_kernel. */
5547 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5548 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5550 split_irqchip_unlock:
5551 mutex_unlock(&kvm->lock);
5554 case KVM_CAP_X2APIC_API:
5556 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5559 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5560 kvm->arch.x2apic_format = true;
5561 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5562 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5566 case KVM_CAP_X86_DISABLE_EXITS:
5568 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5571 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5572 kvm_can_mwait_in_guest())
5573 kvm->arch.mwait_in_guest = true;
5574 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5575 kvm->arch.hlt_in_guest = true;
5576 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5577 kvm->arch.pause_in_guest = true;
5578 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5579 kvm->arch.cstate_in_guest = true;
5582 case KVM_CAP_MSR_PLATFORM_INFO:
5583 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5586 case KVM_CAP_EXCEPTION_PAYLOAD:
5587 kvm->arch.exception_payload_enabled = cap->args[0];
5590 case KVM_CAP_X86_USER_SPACE_MSR:
5591 kvm->arch.user_space_msr_mask = cap->args[0];
5594 case KVM_CAP_X86_BUS_LOCK_EXIT:
5596 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5599 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5600 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5603 if (kvm_has_bus_lock_exit &&
5604 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5605 kvm->arch.bus_lock_detection_enabled = true;
5608 #ifdef CONFIG_X86_SGX_KVM
5609 case KVM_CAP_SGX_ATTRIBUTE: {
5610 unsigned long allowed_attributes = 0;
5612 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5616 /* KVM only supports the PROVISIONKEY privileged attribute. */
5617 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5618 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5619 kvm->arch.sgx_provisioning_allowed = true;
5625 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5627 if (kvm_x86_ops.vm_copy_enc_context_from)
5628 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5630 case KVM_CAP_EXIT_HYPERCALL:
5631 if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5635 kvm->arch.hypercall_exit_enabled = cap->args[0];
5645 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5647 struct kvm_x86_msr_filter *msr_filter;
5649 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5653 msr_filter->default_allow = default_allow;
5657 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5664 for (i = 0; i < msr_filter->count; i++)
5665 kfree(msr_filter->ranges[i].bitmap);
5670 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5671 struct kvm_msr_filter_range *user_range)
5673 unsigned long *bitmap = NULL;
5676 if (!user_range->nmsrs)
5679 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5682 if (!user_range->flags)
5685 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5686 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5689 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5691 return PTR_ERR(bitmap);
5693 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5694 .flags = user_range->flags,
5695 .base = user_range->base,
5696 .nmsrs = user_range->nmsrs,
5700 msr_filter->count++;
5704 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5706 struct kvm_msr_filter __user *user_msr_filter = argp;
5707 struct kvm_x86_msr_filter *new_filter, *old_filter;
5708 struct kvm_msr_filter filter;
5714 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5717 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5718 empty &= !filter.ranges[i].nmsrs;
5720 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5721 if (empty && !default_allow)
5724 new_filter = kvm_alloc_msr_filter(default_allow);
5728 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5729 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5731 kvm_free_msr_filter(new_filter);
5736 mutex_lock(&kvm->lock);
5738 /* The per-VM filter is protected by kvm->lock... */
5739 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5741 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5742 synchronize_srcu(&kvm->srcu);
5744 kvm_free_msr_filter(old_filter);
5746 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5747 mutex_unlock(&kvm->lock);
5752 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5753 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5755 struct kvm_vcpu *vcpu;
5758 mutex_lock(&kvm->lock);
5759 kvm_for_each_vcpu(i, vcpu, kvm) {
5760 if (!vcpu->arch.pv_time_enabled)
5763 ret = kvm_set_guest_paused(vcpu);
5765 kvm_err("Failed to pause guest VCPU%d: %d\n",
5766 vcpu->vcpu_id, ret);
5770 mutex_unlock(&kvm->lock);
5772 return ret ? NOTIFY_BAD : NOTIFY_DONE;
5775 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5778 case PM_HIBERNATION_PREPARE:
5779 case PM_SUSPEND_PREPARE:
5780 return kvm_arch_suspend_notifier(kvm);
5785 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5787 long kvm_arch_vm_ioctl(struct file *filp,
5788 unsigned int ioctl, unsigned long arg)
5790 struct kvm *kvm = filp->private_data;
5791 void __user *argp = (void __user *)arg;
5794 * This union makes it completely explicit to gcc-3.x
5795 * that these two variables' stack usage should be
5796 * combined, not added together.
5799 struct kvm_pit_state ps;
5800 struct kvm_pit_state2 ps2;
5801 struct kvm_pit_config pit_config;
5805 case KVM_SET_TSS_ADDR:
5806 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5808 case KVM_SET_IDENTITY_MAP_ADDR: {
5811 mutex_lock(&kvm->lock);
5813 if (kvm->created_vcpus)
5814 goto set_identity_unlock;
5816 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5817 goto set_identity_unlock;
5818 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5819 set_identity_unlock:
5820 mutex_unlock(&kvm->lock);
5823 case KVM_SET_NR_MMU_PAGES:
5824 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5826 case KVM_GET_NR_MMU_PAGES:
5827 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5829 case KVM_CREATE_IRQCHIP: {
5830 mutex_lock(&kvm->lock);
5833 if (irqchip_in_kernel(kvm))
5834 goto create_irqchip_unlock;
5837 if (kvm->created_vcpus)
5838 goto create_irqchip_unlock;
5840 r = kvm_pic_init(kvm);
5842 goto create_irqchip_unlock;
5844 r = kvm_ioapic_init(kvm);
5846 kvm_pic_destroy(kvm);
5847 goto create_irqchip_unlock;
5850 r = kvm_setup_default_irq_routing(kvm);
5852 kvm_ioapic_destroy(kvm);
5853 kvm_pic_destroy(kvm);
5854 goto create_irqchip_unlock;
5856 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5858 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5859 create_irqchip_unlock:
5860 mutex_unlock(&kvm->lock);
5863 case KVM_CREATE_PIT:
5864 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5866 case KVM_CREATE_PIT2:
5868 if (copy_from_user(&u.pit_config, argp,
5869 sizeof(struct kvm_pit_config)))
5872 mutex_lock(&kvm->lock);
5875 goto create_pit_unlock;
5877 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5881 mutex_unlock(&kvm->lock);
5883 case KVM_GET_IRQCHIP: {
5884 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5885 struct kvm_irqchip *chip;
5887 chip = memdup_user(argp, sizeof(*chip));
5894 if (!irqchip_kernel(kvm))
5895 goto get_irqchip_out;
5896 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5898 goto get_irqchip_out;
5900 if (copy_to_user(argp, chip, sizeof(*chip)))
5901 goto get_irqchip_out;
5907 case KVM_SET_IRQCHIP: {
5908 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5909 struct kvm_irqchip *chip;
5911 chip = memdup_user(argp, sizeof(*chip));
5918 if (!irqchip_kernel(kvm))
5919 goto set_irqchip_out;
5920 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5927 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5930 if (!kvm->arch.vpit)
5932 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5936 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5943 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5945 mutex_lock(&kvm->lock);
5947 if (!kvm->arch.vpit)
5949 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5951 mutex_unlock(&kvm->lock);
5954 case KVM_GET_PIT2: {
5956 if (!kvm->arch.vpit)
5958 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5962 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5967 case KVM_SET_PIT2: {
5969 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5971 mutex_lock(&kvm->lock);
5973 if (!kvm->arch.vpit)
5975 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5977 mutex_unlock(&kvm->lock);
5980 case KVM_REINJECT_CONTROL: {
5981 struct kvm_reinject_control control;
5983 if (copy_from_user(&control, argp, sizeof(control)))
5986 if (!kvm->arch.vpit)
5988 r = kvm_vm_ioctl_reinject(kvm, &control);
5991 case KVM_SET_BOOT_CPU_ID:
5993 mutex_lock(&kvm->lock);
5994 if (kvm->created_vcpus)
5997 kvm->arch.bsp_vcpu_id = arg;
5998 mutex_unlock(&kvm->lock);
6000 #ifdef CONFIG_KVM_XEN
6001 case KVM_XEN_HVM_CONFIG: {
6002 struct kvm_xen_hvm_config xhc;
6004 if (copy_from_user(&xhc, argp, sizeof(xhc)))
6006 r = kvm_xen_hvm_config(kvm, &xhc);
6009 case KVM_XEN_HVM_GET_ATTR: {
6010 struct kvm_xen_hvm_attr xha;
6013 if (copy_from_user(&xha, argp, sizeof(xha)))
6015 r = kvm_xen_hvm_get_attr(kvm, &xha);
6016 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6020 case KVM_XEN_HVM_SET_ATTR: {
6021 struct kvm_xen_hvm_attr xha;
6024 if (copy_from_user(&xha, argp, sizeof(xha)))
6026 r = kvm_xen_hvm_set_attr(kvm, &xha);
6030 case KVM_SET_CLOCK: {
6031 struct kvm_arch *ka = &kvm->arch;
6032 struct kvm_clock_data user_ns;
6036 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
6045 * TODO: userspace has to take care of races with VCPU_RUN, so
6046 * kvm_gen_update_masterclock() can be cut down to locked
6047 * pvclock_update_vm_gtod_copy().
6049 kvm_gen_update_masterclock(kvm);
6052 * This pairs with kvm_guest_time_update(): when masterclock is
6053 * in use, we use master_kernel_ns + kvmclock_offset to set
6054 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6055 * is slightly ahead) here we risk going negative on unsigned
6056 * 'system_time' when 'user_ns.clock' is very small.
6058 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
6059 if (kvm->arch.use_master_clock)
6060 now_ns = ka->master_kernel_ns;
6062 now_ns = get_kvmclock_base_ns();
6063 ka->kvmclock_offset = user_ns.clock - now_ns;
6064 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
6066 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
6069 case KVM_GET_CLOCK: {
6070 struct kvm_clock_data user_ns;
6073 now_ns = get_kvmclock_ns(kvm);
6074 user_ns.clock = now_ns;
6075 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
6076 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
6079 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
6084 case KVM_MEMORY_ENCRYPT_OP: {
6086 if (kvm_x86_ops.mem_enc_op)
6087 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6090 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6091 struct kvm_enc_region region;
6094 if (copy_from_user(®ion, argp, sizeof(region)))
6098 if (kvm_x86_ops.mem_enc_reg_region)
6099 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
6102 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6103 struct kvm_enc_region region;
6106 if (copy_from_user(®ion, argp, sizeof(region)))
6110 if (kvm_x86_ops.mem_enc_unreg_region)
6111 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
6114 case KVM_HYPERV_EVENTFD: {
6115 struct kvm_hyperv_eventfd hvevfd;
6118 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6120 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6123 case KVM_SET_PMU_EVENT_FILTER:
6124 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6126 case KVM_X86_SET_MSR_FILTER:
6127 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6136 static void kvm_init_msr_list(void)
6138 struct x86_pmu_capability x86_pmu;
6142 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6143 "Please update the fixed PMCs in msrs_to_saved_all[]");
6145 perf_get_x86_pmu_capability(&x86_pmu);
6147 num_msrs_to_save = 0;
6148 num_emulated_msrs = 0;
6149 num_msr_based_features = 0;
6151 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6152 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6156 * Even MSRs that are valid in the host may not be exposed
6157 * to the guests in some cases.
6159 switch (msrs_to_save_all[i]) {
6160 case MSR_IA32_BNDCFGS:
6161 if (!kvm_mpx_supported())
6165 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6166 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6169 case MSR_IA32_UMWAIT_CONTROL:
6170 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6173 case MSR_IA32_RTIT_CTL:
6174 case MSR_IA32_RTIT_STATUS:
6175 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6178 case MSR_IA32_RTIT_CR3_MATCH:
6179 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6180 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6183 case MSR_IA32_RTIT_OUTPUT_BASE:
6184 case MSR_IA32_RTIT_OUTPUT_MASK:
6185 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6186 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6187 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6190 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6191 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6192 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6193 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6196 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6197 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6198 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6201 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6202 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6203 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6210 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6213 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6214 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6217 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6220 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6221 struct kvm_msr_entry msr;
6223 msr.index = msr_based_features_all[i];
6224 if (kvm_get_msr_feature(&msr))
6227 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6231 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6239 if (!(lapic_in_kernel(vcpu) &&
6240 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6241 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6252 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6259 if (!(lapic_in_kernel(vcpu) &&
6260 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6262 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6264 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6274 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6275 struct kvm_segment *var, int seg)
6277 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6280 void kvm_get_segment(struct kvm_vcpu *vcpu,
6281 struct kvm_segment *var, int seg)
6283 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6286 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6287 struct x86_exception *exception)
6291 BUG_ON(!mmu_is_nested(vcpu));
6293 /* NPT walks are always user-walks */
6294 access |= PFERR_USER_MASK;
6295 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6300 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6301 struct x86_exception *exception)
6303 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6304 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6306 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6308 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6309 struct x86_exception *exception)
6311 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6312 access |= PFERR_FETCH_MASK;
6313 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6316 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6317 struct x86_exception *exception)
6319 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6320 access |= PFERR_WRITE_MASK;
6321 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6323 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6325 /* uses this to access any guest's mapped memory without checking CPL */
6326 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6327 struct x86_exception *exception)
6329 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6332 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6333 struct kvm_vcpu *vcpu, u32 access,
6334 struct x86_exception *exception)
6337 int r = X86EMUL_CONTINUE;
6340 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6342 unsigned offset = addr & (PAGE_SIZE-1);
6343 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6346 if (gpa == UNMAPPED_GVA)
6347 return X86EMUL_PROPAGATE_FAULT;
6348 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6351 r = X86EMUL_IO_NEEDED;
6363 /* used for instruction fetching */
6364 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6365 gva_t addr, void *val, unsigned int bytes,
6366 struct x86_exception *exception)
6368 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6369 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6373 /* Inline kvm_read_guest_virt_helper for speed. */
6374 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6376 if (unlikely(gpa == UNMAPPED_GVA))
6377 return X86EMUL_PROPAGATE_FAULT;
6379 offset = addr & (PAGE_SIZE-1);
6380 if (WARN_ON(offset + bytes > PAGE_SIZE))
6381 bytes = (unsigned)PAGE_SIZE - offset;
6382 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6384 if (unlikely(ret < 0))
6385 return X86EMUL_IO_NEEDED;
6387 return X86EMUL_CONTINUE;
6390 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6391 gva_t addr, void *val, unsigned int bytes,
6392 struct x86_exception *exception)
6394 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6397 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6398 * is returned, but our callers are not ready for that and they blindly
6399 * call kvm_inject_page_fault. Ensure that they at least do not leak
6400 * uninitialized kernel stack memory into cr2 and error code.
6402 memset(exception, 0, sizeof(*exception));
6403 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6406 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6408 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6409 gva_t addr, void *val, unsigned int bytes,
6410 struct x86_exception *exception, bool system)
6412 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6415 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6416 access |= PFERR_USER_MASK;
6418 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6421 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6422 unsigned long addr, void *val, unsigned int bytes)
6424 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6425 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6427 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6430 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6431 struct kvm_vcpu *vcpu, u32 access,
6432 struct x86_exception *exception)
6435 int r = X86EMUL_CONTINUE;
6438 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6441 unsigned offset = addr & (PAGE_SIZE-1);
6442 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6445 if (gpa == UNMAPPED_GVA)
6446 return X86EMUL_PROPAGATE_FAULT;
6447 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6449 r = X86EMUL_IO_NEEDED;
6461 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6462 unsigned int bytes, struct x86_exception *exception,
6465 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6466 u32 access = PFERR_WRITE_MASK;
6468 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6469 access |= PFERR_USER_MASK;
6471 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6475 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6476 unsigned int bytes, struct x86_exception *exception)
6478 /* kvm_write_guest_virt_system can pull in tons of pages. */
6479 vcpu->arch.l1tf_flush_l1d = true;
6481 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6482 PFERR_WRITE_MASK, exception);
6484 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6486 int handle_ud(struct kvm_vcpu *vcpu)
6488 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6489 int emul_type = EMULTYPE_TRAP_UD;
6490 char sig[5]; /* ud2; .ascii "kvm" */
6491 struct x86_exception e;
6493 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6496 if (force_emulation_prefix &&
6497 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6498 sig, sizeof(sig), &e) == 0 &&
6499 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6500 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6501 emul_type = EMULTYPE_TRAP_UD_FORCED;
6504 return kvm_emulate_instruction(vcpu, emul_type);
6506 EXPORT_SYMBOL_GPL(handle_ud);
6508 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6509 gpa_t gpa, bool write)
6511 /* For APIC access vmexit */
6512 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6515 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6516 trace_vcpu_match_mmio(gva, gpa, write, true);
6523 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6524 gpa_t *gpa, struct x86_exception *exception,
6527 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6528 | (write ? PFERR_WRITE_MASK : 0);
6531 * currently PKRU is only applied to ept enabled guest so
6532 * there is no pkey in EPT page table for L1 guest or EPT
6533 * shadow page table for L2 guest.
6535 if (vcpu_match_mmio_gva(vcpu, gva)
6536 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6537 vcpu->arch.mmio_access, 0, access)) {
6538 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6539 (gva & (PAGE_SIZE - 1));
6540 trace_vcpu_match_mmio(gva, *gpa, write, false);
6544 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6546 if (*gpa == UNMAPPED_GVA)
6549 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6552 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6553 const void *val, int bytes)
6557 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6560 kvm_page_track_write(vcpu, gpa, val, bytes);
6564 struct read_write_emulator_ops {
6565 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6567 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6568 void *val, int bytes);
6569 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6570 int bytes, void *val);
6571 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6572 void *val, int bytes);
6576 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6578 if (vcpu->mmio_read_completed) {
6579 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6580 vcpu->mmio_fragments[0].gpa, val);
6581 vcpu->mmio_read_completed = 0;
6588 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6589 void *val, int bytes)
6591 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6594 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6595 void *val, int bytes)
6597 return emulator_write_phys(vcpu, gpa, val, bytes);
6600 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6602 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6603 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6606 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6607 void *val, int bytes)
6609 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6610 return X86EMUL_IO_NEEDED;
6613 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6614 void *val, int bytes)
6616 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6618 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6619 return X86EMUL_CONTINUE;
6622 static const struct read_write_emulator_ops read_emultor = {
6623 .read_write_prepare = read_prepare,
6624 .read_write_emulate = read_emulate,
6625 .read_write_mmio = vcpu_mmio_read,
6626 .read_write_exit_mmio = read_exit_mmio,
6629 static const struct read_write_emulator_ops write_emultor = {
6630 .read_write_emulate = write_emulate,
6631 .read_write_mmio = write_mmio,
6632 .read_write_exit_mmio = write_exit_mmio,
6636 static int emulator_read_write_onepage(unsigned long addr, void *val,
6638 struct x86_exception *exception,
6639 struct kvm_vcpu *vcpu,
6640 const struct read_write_emulator_ops *ops)
6644 bool write = ops->write;
6645 struct kvm_mmio_fragment *frag;
6646 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6649 * If the exit was due to a NPF we may already have a GPA.
6650 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6651 * Note, this cannot be used on string operations since string
6652 * operation using rep will only have the initial GPA from the NPF
6655 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6656 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6657 gpa = ctxt->gpa_val;
6658 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6660 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6662 return X86EMUL_PROPAGATE_FAULT;
6665 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6666 return X86EMUL_CONTINUE;
6669 * Is this MMIO handled locally?
6671 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6672 if (handled == bytes)
6673 return X86EMUL_CONTINUE;
6679 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6680 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6684 return X86EMUL_CONTINUE;
6687 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6689 void *val, unsigned int bytes,
6690 struct x86_exception *exception,
6691 const struct read_write_emulator_ops *ops)
6693 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6697 if (ops->read_write_prepare &&
6698 ops->read_write_prepare(vcpu, val, bytes))
6699 return X86EMUL_CONTINUE;
6701 vcpu->mmio_nr_fragments = 0;
6703 /* Crossing a page boundary? */
6704 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6707 now = -addr & ~PAGE_MASK;
6708 rc = emulator_read_write_onepage(addr, val, now, exception,
6711 if (rc != X86EMUL_CONTINUE)
6714 if (ctxt->mode != X86EMUL_MODE_PROT64)
6720 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6722 if (rc != X86EMUL_CONTINUE)
6725 if (!vcpu->mmio_nr_fragments)
6728 gpa = vcpu->mmio_fragments[0].gpa;
6730 vcpu->mmio_needed = 1;
6731 vcpu->mmio_cur_fragment = 0;
6733 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6734 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6735 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6736 vcpu->run->mmio.phys_addr = gpa;
6738 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6741 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6745 struct x86_exception *exception)
6747 return emulator_read_write(ctxt, addr, val, bytes,
6748 exception, &read_emultor);
6751 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6755 struct x86_exception *exception)
6757 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6758 exception, &write_emultor);
6761 #define CMPXCHG_TYPE(t, ptr, old, new) \
6762 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6764 #ifdef CONFIG_X86_64
6765 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6767 # define CMPXCHG64(ptr, old, new) \
6768 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6771 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6776 struct x86_exception *exception)
6778 struct kvm_host_map map;
6779 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6785 /* guests cmpxchg8b have to be emulated atomically */
6786 if (bytes > 8 || (bytes & (bytes - 1)))
6789 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6791 if (gpa == UNMAPPED_GVA ||
6792 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6796 * Emulate the atomic as a straight write to avoid #AC if SLD is
6797 * enabled in the host and the access splits a cache line.
6799 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6800 page_line_mask = ~(cache_line_size() - 1);
6802 page_line_mask = PAGE_MASK;
6804 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6807 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6810 kaddr = map.hva + offset_in_page(gpa);
6814 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6817 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6820 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6823 exchanged = CMPXCHG64(kaddr, old, new);
6829 kvm_vcpu_unmap(vcpu, &map, true);
6832 return X86EMUL_CMPXCHG_FAILED;
6834 kvm_page_track_write(vcpu, gpa, new, bytes);
6836 return X86EMUL_CONTINUE;
6839 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6841 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6844 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6848 for (i = 0; i < vcpu->arch.pio.count; i++) {
6849 if (vcpu->arch.pio.in)
6850 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6851 vcpu->arch.pio.size, pd);
6853 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6854 vcpu->arch.pio.port, vcpu->arch.pio.size,
6858 pd += vcpu->arch.pio.size;
6863 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6864 unsigned short port, void *val,
6865 unsigned int count, bool in)
6867 vcpu->arch.pio.port = port;
6868 vcpu->arch.pio.in = in;
6869 vcpu->arch.pio.count = count;
6870 vcpu->arch.pio.size = size;
6872 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6873 vcpu->arch.pio.count = 0;
6877 vcpu->run->exit_reason = KVM_EXIT_IO;
6878 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6879 vcpu->run->io.size = size;
6880 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6881 vcpu->run->io.count = count;
6882 vcpu->run->io.port = port;
6887 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6888 unsigned short port, void *val, unsigned int count)
6892 if (vcpu->arch.pio.count)
6895 memset(vcpu->arch.pio_data, 0, size * count);
6897 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6900 memcpy(val, vcpu->arch.pio_data, size * count);
6901 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6902 vcpu->arch.pio.count = 0;
6909 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6910 int size, unsigned short port, void *val,
6913 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6917 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6918 unsigned short port, const void *val,
6921 memcpy(vcpu->arch.pio_data, val, size * count);
6922 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6923 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6926 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6927 int size, unsigned short port,
6928 const void *val, unsigned int count)
6930 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6933 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6935 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6938 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6940 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6943 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6945 if (!need_emulate_wbinvd(vcpu))
6946 return X86EMUL_CONTINUE;
6948 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6949 int cpu = get_cpu();
6951 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6952 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6953 wbinvd_ipi, NULL, 1);
6955 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6958 return X86EMUL_CONTINUE;
6961 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6963 kvm_emulate_wbinvd_noskip(vcpu);
6964 return kvm_skip_emulated_instruction(vcpu);
6966 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6970 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6972 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6975 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6976 unsigned long *dest)
6978 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6981 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6982 unsigned long value)
6985 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6988 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6990 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6993 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6995 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6996 unsigned long value;
7000 value = kvm_read_cr0(vcpu);
7003 value = vcpu->arch.cr2;
7006 value = kvm_read_cr3(vcpu);
7009 value = kvm_read_cr4(vcpu);
7012 value = kvm_get_cr8(vcpu);
7015 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7022 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7024 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7029 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7032 vcpu->arch.cr2 = val;
7035 res = kvm_set_cr3(vcpu, val);
7038 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7041 res = kvm_set_cr8(vcpu, val);
7044 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7051 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7053 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7056 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7058 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7061 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7063 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7066 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7068 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7071 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7073 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7076 static unsigned long emulator_get_cached_segment_base(
7077 struct x86_emulate_ctxt *ctxt, int seg)
7079 return get_segment_base(emul_to_vcpu(ctxt), seg);
7082 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7083 struct desc_struct *desc, u32 *base3,
7086 struct kvm_segment var;
7088 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7089 *selector = var.selector;
7092 memset(desc, 0, sizeof(*desc));
7100 set_desc_limit(desc, var.limit);
7101 set_desc_base(desc, (unsigned long)var.base);
7102 #ifdef CONFIG_X86_64
7104 *base3 = var.base >> 32;
7106 desc->type = var.type;
7108 desc->dpl = var.dpl;
7109 desc->p = var.present;
7110 desc->avl = var.avl;
7118 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7119 struct desc_struct *desc, u32 base3,
7122 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7123 struct kvm_segment var;
7125 var.selector = selector;
7126 var.base = get_desc_base(desc);
7127 #ifdef CONFIG_X86_64
7128 var.base |= ((u64)base3) << 32;
7130 var.limit = get_desc_limit(desc);
7132 var.limit = (var.limit << 12) | 0xfff;
7133 var.type = desc->type;
7134 var.dpl = desc->dpl;
7139 var.avl = desc->avl;
7140 var.present = desc->p;
7141 var.unusable = !var.present;
7144 kvm_set_segment(vcpu, &var, seg);
7148 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7149 u32 msr_index, u64 *pdata)
7151 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7154 r = kvm_get_msr(vcpu, msr_index, pdata);
7156 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7157 /* Bounce to user space */
7158 return X86EMUL_IO_NEEDED;
7164 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7165 u32 msr_index, u64 data)
7167 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7170 r = kvm_set_msr(vcpu, msr_index, data);
7172 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7173 /* Bounce to user space */
7174 return X86EMUL_IO_NEEDED;
7180 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7182 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7184 return vcpu->arch.smbase;
7187 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7189 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7191 vcpu->arch.smbase = smbase;
7194 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7197 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7200 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7201 u32 pmc, u64 *pdata)
7203 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7206 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7208 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7211 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7212 struct x86_instruction_info *info,
7213 enum x86_intercept_stage stage)
7215 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7219 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7220 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7223 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7226 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7228 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7231 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7233 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7236 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7238 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7241 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7243 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7246 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7248 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7251 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7253 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7256 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7258 return emul_to_vcpu(ctxt)->arch.hflags;
7261 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7263 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7265 kvm_smm_changed(vcpu, false);
7268 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7269 const char *smstate)
7271 return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7274 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7276 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7279 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7281 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7284 static const struct x86_emulate_ops emulate_ops = {
7285 .read_gpr = emulator_read_gpr,
7286 .write_gpr = emulator_write_gpr,
7287 .read_std = emulator_read_std,
7288 .write_std = emulator_write_std,
7289 .read_phys = kvm_read_guest_phys_system,
7290 .fetch = kvm_fetch_guest_virt,
7291 .read_emulated = emulator_read_emulated,
7292 .write_emulated = emulator_write_emulated,
7293 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7294 .invlpg = emulator_invlpg,
7295 .pio_in_emulated = emulator_pio_in_emulated,
7296 .pio_out_emulated = emulator_pio_out_emulated,
7297 .get_segment = emulator_get_segment,
7298 .set_segment = emulator_set_segment,
7299 .get_cached_segment_base = emulator_get_cached_segment_base,
7300 .get_gdt = emulator_get_gdt,
7301 .get_idt = emulator_get_idt,
7302 .set_gdt = emulator_set_gdt,
7303 .set_idt = emulator_set_idt,
7304 .get_cr = emulator_get_cr,
7305 .set_cr = emulator_set_cr,
7306 .cpl = emulator_get_cpl,
7307 .get_dr = emulator_get_dr,
7308 .set_dr = emulator_set_dr,
7309 .get_smbase = emulator_get_smbase,
7310 .set_smbase = emulator_set_smbase,
7311 .set_msr = emulator_set_msr,
7312 .get_msr = emulator_get_msr,
7313 .check_pmc = emulator_check_pmc,
7314 .read_pmc = emulator_read_pmc,
7315 .halt = emulator_halt,
7316 .wbinvd = emulator_wbinvd,
7317 .fix_hypercall = emulator_fix_hypercall,
7318 .intercept = emulator_intercept,
7319 .get_cpuid = emulator_get_cpuid,
7320 .guest_has_long_mode = emulator_guest_has_long_mode,
7321 .guest_has_movbe = emulator_guest_has_movbe,
7322 .guest_has_fxsr = emulator_guest_has_fxsr,
7323 .set_nmi_mask = emulator_set_nmi_mask,
7324 .get_hflags = emulator_get_hflags,
7325 .exiting_smm = emulator_exiting_smm,
7326 .leave_smm = emulator_leave_smm,
7327 .triple_fault = emulator_triple_fault,
7328 .set_xcr = emulator_set_xcr,
7331 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7333 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7335 * an sti; sti; sequence only disable interrupts for the first
7336 * instruction. So, if the last instruction, be it emulated or
7337 * not, left the system with the INT_STI flag enabled, it
7338 * means that the last instruction is an sti. We should not
7339 * leave the flag on in this case. The same goes for mov ss
7341 if (int_shadow & mask)
7343 if (unlikely(int_shadow || mask)) {
7344 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7346 kvm_make_request(KVM_REQ_EVENT, vcpu);
7350 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7352 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7353 if (ctxt->exception.vector == PF_VECTOR)
7354 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7356 if (ctxt->exception.error_code_valid)
7357 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7358 ctxt->exception.error_code);
7360 kvm_queue_exception(vcpu, ctxt->exception.vector);
7364 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7366 struct x86_emulate_ctxt *ctxt;
7368 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7370 pr_err("kvm: failed to allocate vcpu's emulator\n");
7375 ctxt->ops = &emulate_ops;
7376 vcpu->arch.emulate_ctxt = ctxt;
7381 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7383 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7386 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7388 ctxt->gpa_available = false;
7389 ctxt->eflags = kvm_get_rflags(vcpu);
7390 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7392 ctxt->eip = kvm_rip_read(vcpu);
7393 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7394 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7395 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7396 cs_db ? X86EMUL_MODE_PROT32 :
7397 X86EMUL_MODE_PROT16;
7398 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7399 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7400 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7402 ctxt->interruptibility = 0;
7403 ctxt->have_exception = false;
7404 ctxt->exception.vector = -1;
7405 ctxt->perm_ok = false;
7407 init_decode_cache(ctxt);
7408 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7411 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7413 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7416 init_emulate_ctxt(vcpu);
7420 ctxt->_eip = ctxt->eip + inc_eip;
7421 ret = emulate_int_real(ctxt, irq);
7423 if (ret != X86EMUL_CONTINUE) {
7424 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7426 ctxt->eip = ctxt->_eip;
7427 kvm_rip_write(vcpu, ctxt->eip);
7428 kvm_set_rflags(vcpu, ctxt->eflags);
7431 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7433 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7435 ++vcpu->stat.insn_emulation_fail;
7436 trace_kvm_emulate_insn_failed(vcpu);
7438 if (emulation_type & EMULTYPE_VMWARE_GP) {
7439 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7443 if (emulation_type & EMULTYPE_SKIP) {
7444 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7445 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7446 vcpu->run->internal.ndata = 0;
7450 kvm_queue_exception(vcpu, UD_VECTOR);
7452 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7453 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7454 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7455 vcpu->run->internal.ndata = 0;
7462 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7463 bool write_fault_to_shadow_pgtable,
7466 gpa_t gpa = cr2_or_gpa;
7469 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7472 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7473 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7476 if (!vcpu->arch.mmu->direct_map) {
7478 * Write permission should be allowed since only
7479 * write access need to be emulated.
7481 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7484 * If the mapping is invalid in guest, let cpu retry
7485 * it to generate fault.
7487 if (gpa == UNMAPPED_GVA)
7492 * Do not retry the unhandleable instruction if it faults on the
7493 * readonly host memory, otherwise it will goto a infinite loop:
7494 * retry instruction -> write #PF -> emulation fail -> retry
7495 * instruction -> ...
7497 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7500 * If the instruction failed on the error pfn, it can not be fixed,
7501 * report the error to userspace.
7503 if (is_error_noslot_pfn(pfn))
7506 kvm_release_pfn_clean(pfn);
7508 /* The instructions are well-emulated on direct mmu. */
7509 if (vcpu->arch.mmu->direct_map) {
7510 unsigned int indirect_shadow_pages;
7512 write_lock(&vcpu->kvm->mmu_lock);
7513 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7514 write_unlock(&vcpu->kvm->mmu_lock);
7516 if (indirect_shadow_pages)
7517 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7523 * if emulation was due to access to shadowed page table
7524 * and it failed try to unshadow page and re-enter the
7525 * guest to let CPU execute the instruction.
7527 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7530 * If the access faults on its page table, it can not
7531 * be fixed by unprotecting shadow page and it should
7532 * be reported to userspace.
7534 return !write_fault_to_shadow_pgtable;
7537 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7538 gpa_t cr2_or_gpa, int emulation_type)
7540 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7541 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7543 last_retry_eip = vcpu->arch.last_retry_eip;
7544 last_retry_addr = vcpu->arch.last_retry_addr;
7547 * If the emulation is caused by #PF and it is non-page_table
7548 * writing instruction, it means the VM-EXIT is caused by shadow
7549 * page protected, we can zap the shadow page and retry this
7550 * instruction directly.
7552 * Note: if the guest uses a non-page-table modifying instruction
7553 * on the PDE that points to the instruction, then we will unmap
7554 * the instruction and go to an infinite loop. So, we cache the
7555 * last retried eip and the last fault address, if we meet the eip
7556 * and the address again, we can break out of the potential infinite
7559 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7561 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7564 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7565 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7568 if (x86_page_table_writing_insn(ctxt))
7571 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7574 vcpu->arch.last_retry_eip = ctxt->eip;
7575 vcpu->arch.last_retry_addr = cr2_or_gpa;
7577 if (!vcpu->arch.mmu->direct_map)
7578 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7580 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7585 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7586 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7588 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7590 trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7593 vcpu->arch.hflags |= HF_SMM_MASK;
7595 vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7597 /* Process a latched INIT or SMI, if any. */
7598 kvm_make_request(KVM_REQ_EVENT, vcpu);
7601 kvm_mmu_reset_context(vcpu);
7604 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7613 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7614 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7619 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7621 struct kvm_run *kvm_run = vcpu->run;
7623 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7624 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7625 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7626 kvm_run->debug.arch.exception = DB_VECTOR;
7627 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7630 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7634 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7636 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7639 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7644 * rflags is the old, "raw" value of the flags. The new value has
7645 * not been saved yet.
7647 * This is correct even for TF set by the guest, because "the
7648 * processor will not generate this exception after the instruction
7649 * that sets the TF flag".
7651 if (unlikely(rflags & X86_EFLAGS_TF))
7652 r = kvm_vcpu_do_singlestep(vcpu);
7655 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7657 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7659 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7660 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7661 struct kvm_run *kvm_run = vcpu->run;
7662 unsigned long eip = kvm_get_linear_rip(vcpu);
7663 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7664 vcpu->arch.guest_debug_dr7,
7668 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7669 kvm_run->debug.arch.pc = eip;
7670 kvm_run->debug.arch.exception = DB_VECTOR;
7671 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7677 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7678 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7679 unsigned long eip = kvm_get_linear_rip(vcpu);
7680 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7685 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7694 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7696 switch (ctxt->opcode_len) {
7703 case 0xe6: /* OUT */
7707 case 0x6c: /* INS */
7709 case 0x6e: /* OUTS */
7716 case 0x33: /* RDPMC */
7726 * Decode to be emulated instruction. Return EMULATION_OK if success.
7728 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7729 void *insn, int insn_len)
7731 int r = EMULATION_OK;
7732 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7734 init_emulate_ctxt(vcpu);
7737 * We will reenter on the same instruction since we do not set
7738 * complete_userspace_io. This does not handle watchpoints yet,
7739 * those would be handled in the emulate_ops.
7741 if (!(emulation_type & EMULTYPE_SKIP) &&
7742 kvm_vcpu_check_breakpoint(vcpu, &r))
7745 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7747 trace_kvm_emulate_insn_start(vcpu);
7748 ++vcpu->stat.insn_emulation;
7752 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7754 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7755 int emulation_type, void *insn, int insn_len)
7758 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7759 bool writeback = true;
7760 bool write_fault_to_spt;
7762 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7765 vcpu->arch.l1tf_flush_l1d = true;
7768 * Clear write_fault_to_shadow_pgtable here to ensure it is
7771 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7772 vcpu->arch.write_fault_to_shadow_pgtable = false;
7774 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7775 kvm_clear_exception_queue(vcpu);
7777 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7779 if (r != EMULATION_OK) {
7780 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7781 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7782 kvm_queue_exception(vcpu, UD_VECTOR);
7785 if (reexecute_instruction(vcpu, cr2_or_gpa,
7789 if (ctxt->have_exception) {
7791 * #UD should result in just EMULATION_FAILED, and trap-like
7792 * exception should not be encountered during decode.
7794 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7795 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7796 inject_emulated_exception(vcpu);
7799 return handle_emulation_failure(vcpu, emulation_type);
7803 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7804 !is_vmware_backdoor_opcode(ctxt)) {
7805 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7810 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7811 * for kvm_skip_emulated_instruction(). The caller is responsible for
7812 * updating interruptibility state and injecting single-step #DBs.
7814 if (emulation_type & EMULTYPE_SKIP) {
7815 kvm_rip_write(vcpu, ctxt->_eip);
7816 if (ctxt->eflags & X86_EFLAGS_RF)
7817 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7821 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7824 /* this is needed for vmware backdoor interface to work since it
7825 changes registers values during IO operation */
7826 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7827 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7828 emulator_invalidate_register_cache(ctxt);
7832 if (emulation_type & EMULTYPE_PF) {
7833 /* Save the faulting GPA (cr2) in the address field */
7834 ctxt->exception.address = cr2_or_gpa;
7836 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7837 if (vcpu->arch.mmu->direct_map) {
7838 ctxt->gpa_available = true;
7839 ctxt->gpa_val = cr2_or_gpa;
7842 /* Sanitize the address out of an abundance of paranoia. */
7843 ctxt->exception.address = 0;
7846 r = x86_emulate_insn(ctxt);
7848 if (r == EMULATION_INTERCEPTED)
7851 if (r == EMULATION_FAILED) {
7852 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7856 return handle_emulation_failure(vcpu, emulation_type);
7859 if (ctxt->have_exception) {
7861 if (inject_emulated_exception(vcpu))
7863 } else if (vcpu->arch.pio.count) {
7864 if (!vcpu->arch.pio.in) {
7865 /* FIXME: return into emulator if single-stepping. */
7866 vcpu->arch.pio.count = 0;
7869 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7872 } else if (vcpu->mmio_needed) {
7873 ++vcpu->stat.mmio_exits;
7875 if (!vcpu->mmio_is_write)
7878 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7879 } else if (r == EMULATION_RESTART)
7885 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7886 toggle_interruptibility(vcpu, ctxt->interruptibility);
7887 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7888 if (!ctxt->have_exception ||
7889 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7890 kvm_rip_write(vcpu, ctxt->eip);
7891 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7892 r = kvm_vcpu_do_singlestep(vcpu);
7893 if (kvm_x86_ops.update_emulated_instruction)
7894 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7895 __kvm_set_rflags(vcpu, ctxt->eflags);
7899 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7900 * do nothing, and it will be requested again as soon as
7901 * the shadow expires. But we still need to check here,
7902 * because POPF has no interrupt shadow.
7904 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7905 kvm_make_request(KVM_REQ_EVENT, vcpu);
7907 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7912 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7914 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7916 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7918 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7919 void *insn, int insn_len)
7921 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7923 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7925 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7927 vcpu->arch.pio.count = 0;
7931 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7933 vcpu->arch.pio.count = 0;
7935 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7938 return kvm_skip_emulated_instruction(vcpu);
7941 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7942 unsigned short port)
7944 unsigned long val = kvm_rax_read(vcpu);
7945 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7951 * Workaround userspace that relies on old KVM behavior of %rip being
7952 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7955 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7956 vcpu->arch.complete_userspace_io =
7957 complete_fast_pio_out_port_0x7e;
7958 kvm_skip_emulated_instruction(vcpu);
7960 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7961 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7966 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7970 /* We should only ever be called with arch.pio.count equal to 1 */
7971 BUG_ON(vcpu->arch.pio.count != 1);
7973 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7974 vcpu->arch.pio.count = 0;
7978 /* For size less than 4 we merge, else we zero extend */
7979 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7982 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7983 * the copy and tracing
7985 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7986 kvm_rax_write(vcpu, val);
7988 return kvm_skip_emulated_instruction(vcpu);
7991 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7992 unsigned short port)
7997 /* For size less than 4 we merge, else we zero extend */
7998 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8000 ret = emulator_pio_in(vcpu, size, port, &val, 1);
8002 kvm_rax_write(vcpu, val);
8006 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8007 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8012 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8017 ret = kvm_fast_pio_in(vcpu, size, port);
8019 ret = kvm_fast_pio_out(vcpu, size, port);
8020 return ret && kvm_skip_emulated_instruction(vcpu);
8022 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8024 static int kvmclock_cpu_down_prep(unsigned int cpu)
8026 __this_cpu_write(cpu_tsc_khz, 0);
8030 static void tsc_khz_changed(void *data)
8032 struct cpufreq_freqs *freq = data;
8033 unsigned long khz = 0;
8037 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8038 khz = cpufreq_quick_get(raw_smp_processor_id());
8041 __this_cpu_write(cpu_tsc_khz, khz);
8044 #ifdef CONFIG_X86_64
8045 static void kvm_hyperv_tsc_notifier(void)
8048 struct kvm_vcpu *vcpu;
8050 unsigned long flags;
8052 mutex_lock(&kvm_lock);
8053 list_for_each_entry(kvm, &vm_list, vm_list)
8054 kvm_make_mclock_inprogress_request(kvm);
8056 hyperv_stop_tsc_emulation();
8058 /* TSC frequency always matches when on Hyper-V */
8059 for_each_present_cpu(cpu)
8060 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8061 kvm_max_guest_tsc_khz = tsc_khz;
8063 list_for_each_entry(kvm, &vm_list, vm_list) {
8064 struct kvm_arch *ka = &kvm->arch;
8066 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
8067 pvclock_update_vm_gtod_copy(kvm);
8068 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
8070 kvm_for_each_vcpu(cpu, vcpu, kvm)
8071 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8073 kvm_for_each_vcpu(cpu, vcpu, kvm)
8074 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
8076 mutex_unlock(&kvm_lock);
8080 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8083 struct kvm_vcpu *vcpu;
8084 int i, send_ipi = 0;
8087 * We allow guests to temporarily run on slowing clocks,
8088 * provided we notify them after, or to run on accelerating
8089 * clocks, provided we notify them before. Thus time never
8092 * However, we have a problem. We can't atomically update
8093 * the frequency of a given CPU from this function; it is
8094 * merely a notifier, which can be called from any CPU.
8095 * Changing the TSC frequency at arbitrary points in time
8096 * requires a recomputation of local variables related to
8097 * the TSC for each VCPU. We must flag these local variables
8098 * to be updated and be sure the update takes place with the
8099 * new frequency before any guests proceed.
8101 * Unfortunately, the combination of hotplug CPU and frequency
8102 * change creates an intractable locking scenario; the order
8103 * of when these callouts happen is undefined with respect to
8104 * CPU hotplug, and they can race with each other. As such,
8105 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8106 * undefined; you can actually have a CPU frequency change take
8107 * place in between the computation of X and the setting of the
8108 * variable. To protect against this problem, all updates of
8109 * the per_cpu tsc_khz variable are done in an interrupt
8110 * protected IPI, and all callers wishing to update the value
8111 * must wait for a synchronous IPI to complete (which is trivial
8112 * if the caller is on the CPU already). This establishes the
8113 * necessary total order on variable updates.
8115 * Note that because a guest time update may take place
8116 * anytime after the setting of the VCPU's request bit, the
8117 * correct TSC value must be set before the request. However,
8118 * to ensure the update actually makes it to any guest which
8119 * starts running in hardware virtualization between the set
8120 * and the acquisition of the spinlock, we must also ping the
8121 * CPU after setting the request bit.
8125 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8127 mutex_lock(&kvm_lock);
8128 list_for_each_entry(kvm, &vm_list, vm_list) {
8129 kvm_for_each_vcpu(i, vcpu, kvm) {
8130 if (vcpu->cpu != cpu)
8132 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8133 if (vcpu->cpu != raw_smp_processor_id())
8137 mutex_unlock(&kvm_lock);
8139 if (freq->old < freq->new && send_ipi) {
8141 * We upscale the frequency. Must make the guest
8142 * doesn't see old kvmclock values while running with
8143 * the new frequency, otherwise we risk the guest sees
8144 * time go backwards.
8146 * In case we update the frequency for another cpu
8147 * (which might be in guest context) send an interrupt
8148 * to kick the cpu out of guest context. Next time
8149 * guest context is entered kvmclock will be updated,
8150 * so the guest will not see stale values.
8152 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8156 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8159 struct cpufreq_freqs *freq = data;
8162 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8164 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8167 for_each_cpu(cpu, freq->policy->cpus)
8168 __kvmclock_cpufreq_notifier(freq, cpu);
8173 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8174 .notifier_call = kvmclock_cpufreq_notifier
8177 static int kvmclock_cpu_online(unsigned int cpu)
8179 tsc_khz_changed(NULL);
8183 static void kvm_timer_init(void)
8185 max_tsc_khz = tsc_khz;
8187 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8188 #ifdef CONFIG_CPU_FREQ
8189 struct cpufreq_policy *policy;
8193 policy = cpufreq_cpu_get(cpu);
8195 if (policy->cpuinfo.max_freq)
8196 max_tsc_khz = policy->cpuinfo.max_freq;
8197 cpufreq_cpu_put(policy);
8201 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8202 CPUFREQ_TRANSITION_NOTIFIER);
8205 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8206 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8209 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8210 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8212 int kvm_is_in_guest(void)
8214 return __this_cpu_read(current_vcpu) != NULL;
8217 static int kvm_is_user_mode(void)
8221 if (__this_cpu_read(current_vcpu))
8222 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8224 return user_mode != 0;
8227 static unsigned long kvm_get_guest_ip(void)
8229 unsigned long ip = 0;
8231 if (__this_cpu_read(current_vcpu))
8232 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8237 static void kvm_handle_intel_pt_intr(void)
8239 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8241 kvm_make_request(KVM_REQ_PMI, vcpu);
8242 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8243 (unsigned long *)&vcpu->arch.pmu.global_status);
8246 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8247 .is_in_guest = kvm_is_in_guest,
8248 .is_user_mode = kvm_is_user_mode,
8249 .get_guest_ip = kvm_get_guest_ip,
8250 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8253 #ifdef CONFIG_X86_64
8254 static void pvclock_gtod_update_fn(struct work_struct *work)
8258 struct kvm_vcpu *vcpu;
8261 mutex_lock(&kvm_lock);
8262 list_for_each_entry(kvm, &vm_list, vm_list)
8263 kvm_for_each_vcpu(i, vcpu, kvm)
8264 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8265 atomic_set(&kvm_guest_has_master_clock, 0);
8266 mutex_unlock(&kvm_lock);
8269 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8272 * Indirection to move queue_work() out of the tk_core.seq write held
8273 * region to prevent possible deadlocks against time accessors which
8274 * are invoked with work related locks held.
8276 static void pvclock_irq_work_fn(struct irq_work *w)
8278 queue_work(system_long_wq, &pvclock_gtod_work);
8281 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8284 * Notification about pvclock gtod data update.
8286 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8289 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8290 struct timekeeper *tk = priv;
8292 update_pvclock_gtod(tk);
8295 * Disable master clock if host does not trust, or does not use,
8296 * TSC based clocksource. Delegate queue_work() to irq_work as
8297 * this is invoked with tk_core.seq write held.
8299 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8300 atomic_read(&kvm_guest_has_master_clock) != 0)
8301 irq_work_queue(&pvclock_irq_work);
8305 static struct notifier_block pvclock_gtod_notifier = {
8306 .notifier_call = pvclock_gtod_notify,
8310 int kvm_arch_init(void *opaque)
8312 struct kvm_x86_init_ops *ops = opaque;
8315 if (kvm_x86_ops.hardware_enable) {
8316 printk(KERN_ERR "kvm: already loaded the other module\n");
8321 if (!ops->cpu_has_kvm_support()) {
8322 pr_err_ratelimited("kvm: no hardware support\n");
8326 if (ops->disabled_by_bios()) {
8327 pr_err_ratelimited("kvm: disabled by bios\n");
8333 * KVM explicitly assumes that the guest has an FPU and
8334 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8335 * vCPU's FPU state as a fxregs_state struct.
8337 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8338 printk(KERN_ERR "kvm: inadequate fpu\n");
8344 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8345 __alignof__(struct fpu), SLAB_ACCOUNT,
8347 if (!x86_fpu_cache) {
8348 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8352 x86_emulator_cache = kvm_alloc_emulator_cache();
8353 if (!x86_emulator_cache) {
8354 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8355 goto out_free_x86_fpu_cache;
8358 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8359 if (!user_return_msrs) {
8360 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8361 goto out_free_x86_emulator_cache;
8363 kvm_nr_uret_msrs = 0;
8365 r = kvm_mmu_module_init();
8367 goto out_free_percpu;
8371 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8373 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8374 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8375 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8378 if (pi_inject_timer == -1)
8379 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8380 #ifdef CONFIG_X86_64
8381 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8383 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8384 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8390 free_percpu(user_return_msrs);
8391 out_free_x86_emulator_cache:
8392 kmem_cache_destroy(x86_emulator_cache);
8393 out_free_x86_fpu_cache:
8394 kmem_cache_destroy(x86_fpu_cache);
8399 void kvm_arch_exit(void)
8401 #ifdef CONFIG_X86_64
8402 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8403 clear_hv_tscchange_cb();
8406 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8408 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8409 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8410 CPUFREQ_TRANSITION_NOTIFIER);
8411 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8412 #ifdef CONFIG_X86_64
8413 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8414 irq_work_sync(&pvclock_irq_work);
8415 cancel_work_sync(&pvclock_gtod_work);
8417 kvm_x86_ops.hardware_enable = NULL;
8418 kvm_mmu_module_exit();
8419 free_percpu(user_return_msrs);
8420 kmem_cache_destroy(x86_emulator_cache);
8421 kmem_cache_destroy(x86_fpu_cache);
8422 #ifdef CONFIG_KVM_XEN
8423 static_key_deferred_flush(&kvm_xen_enabled);
8424 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8428 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8430 ++vcpu->stat.halt_exits;
8431 if (lapic_in_kernel(vcpu)) {
8432 vcpu->arch.mp_state = state;
8435 vcpu->run->exit_reason = reason;
8440 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8442 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8444 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8446 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8448 int ret = kvm_skip_emulated_instruction(vcpu);
8450 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8451 * KVM_EXIT_DEBUG here.
8453 return kvm_vcpu_halt(vcpu) && ret;
8455 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8457 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8459 int ret = kvm_skip_emulated_instruction(vcpu);
8461 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8463 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8465 #ifdef CONFIG_X86_64
8466 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8467 unsigned long clock_type)
8469 struct kvm_clock_pairing clock_pairing;
8470 struct timespec64 ts;
8474 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8475 return -KVM_EOPNOTSUPP;
8477 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8478 return -KVM_EOPNOTSUPP;
8480 clock_pairing.sec = ts.tv_sec;
8481 clock_pairing.nsec = ts.tv_nsec;
8482 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8483 clock_pairing.flags = 0;
8484 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8487 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8488 sizeof(struct kvm_clock_pairing)))
8496 * kvm_pv_kick_cpu_op: Kick a vcpu.
8498 * @apicid - apicid of vcpu to be kicked.
8500 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8502 struct kvm_lapic_irq lapic_irq;
8504 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8505 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8506 lapic_irq.level = 0;
8507 lapic_irq.dest_id = apicid;
8508 lapic_irq.msi_redir_hint = false;
8510 lapic_irq.delivery_mode = APIC_DM_REMRD;
8511 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8514 bool kvm_apicv_activated(struct kvm *kvm)
8516 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8518 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8520 static void kvm_apicv_init(struct kvm *kvm)
8523 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8524 &kvm->arch.apicv_inhibit_reasons);
8526 set_bit(APICV_INHIBIT_REASON_DISABLE,
8527 &kvm->arch.apicv_inhibit_reasons);
8530 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8532 struct kvm_vcpu *target = NULL;
8533 struct kvm_apic_map *map;
8535 vcpu->stat.directed_yield_attempted++;
8537 if (single_task_running())
8541 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8543 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8544 target = map->phys_map[dest_id]->vcpu;
8548 if (!target || !READ_ONCE(target->ready))
8551 /* Ignore requests to yield to self */
8555 if (kvm_vcpu_yield_to(target) <= 0)
8558 vcpu->stat.directed_yield_successful++;
8564 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8566 u64 ret = vcpu->run->hypercall.ret;
8568 if (!is_64_bit_mode(vcpu))
8570 kvm_rax_write(vcpu, ret);
8571 ++vcpu->stat.hypercalls;
8572 return kvm_skip_emulated_instruction(vcpu);
8575 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8577 unsigned long nr, a0, a1, a2, a3, ret;
8580 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8581 return kvm_xen_hypercall(vcpu);
8583 if (kvm_hv_hypercall_enabled(vcpu))
8584 return kvm_hv_hypercall(vcpu);
8586 nr = kvm_rax_read(vcpu);
8587 a0 = kvm_rbx_read(vcpu);
8588 a1 = kvm_rcx_read(vcpu);
8589 a2 = kvm_rdx_read(vcpu);
8590 a3 = kvm_rsi_read(vcpu);
8592 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8594 op_64_bit = is_64_bit_mode(vcpu);
8603 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8611 case KVM_HC_VAPIC_POLL_IRQ:
8614 case KVM_HC_KICK_CPU:
8615 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8618 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8619 kvm_sched_yield(vcpu, a1);
8622 #ifdef CONFIG_X86_64
8623 case KVM_HC_CLOCK_PAIRING:
8624 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8627 case KVM_HC_SEND_IPI:
8628 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8631 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8633 case KVM_HC_SCHED_YIELD:
8634 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8637 kvm_sched_yield(vcpu, a0);
8640 case KVM_HC_MAP_GPA_RANGE: {
8641 u64 gpa = a0, npages = a1, attrs = a2;
8644 if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
8647 if (!PAGE_ALIGNED(gpa) || !npages ||
8648 gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
8653 vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
8654 vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
8655 vcpu->run->hypercall.args[0] = gpa;
8656 vcpu->run->hypercall.args[1] = npages;
8657 vcpu->run->hypercall.args[2] = attrs;
8658 vcpu->run->hypercall.longmode = op_64_bit;
8659 vcpu->arch.complete_userspace_io = complete_hypercall_exit;
8669 kvm_rax_write(vcpu, ret);
8671 ++vcpu->stat.hypercalls;
8672 return kvm_skip_emulated_instruction(vcpu);
8674 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8676 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8678 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8679 char instruction[3];
8680 unsigned long rip = kvm_rip_read(vcpu);
8682 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8684 return emulator_write_emulated(ctxt, rip, instruction, 3,
8688 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8690 return vcpu->run->request_interrupt_window &&
8691 likely(!pic_in_kernel(vcpu->kvm));
8694 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8696 struct kvm_run *kvm_run = vcpu->run;
8699 * if_flag is obsolete and useless, so do not bother
8700 * setting it for SEV-ES guests. Userspace can just
8701 * use kvm_run->ready_for_interrupt_injection.
8703 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8704 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8706 kvm_run->cr8 = kvm_get_cr8(vcpu);
8707 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8708 kvm_run->ready_for_interrupt_injection =
8709 pic_in_kernel(vcpu->kvm) ||
8710 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8713 kvm_run->flags |= KVM_RUN_X86_SMM;
8716 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8720 if (!kvm_x86_ops.update_cr8_intercept)
8723 if (!lapic_in_kernel(vcpu))
8726 if (vcpu->arch.apicv_active)
8729 if (!vcpu->arch.apic->vapic_addr)
8730 max_irr = kvm_lapic_find_highest_irr(vcpu);
8737 tpr = kvm_lapic_get_cr8(vcpu);
8739 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8743 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8745 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8746 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8750 return kvm_x86_ops.nested_ops->check_events(vcpu);
8753 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8755 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8756 vcpu->arch.exception.error_code = false;
8757 static_call(kvm_x86_queue_exception)(vcpu);
8760 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8763 bool can_inject = true;
8765 /* try to reinject previous events if any */
8767 if (vcpu->arch.exception.injected) {
8768 kvm_inject_exception(vcpu);
8772 * Do not inject an NMI or interrupt if there is a pending
8773 * exception. Exceptions and interrupts are recognized at
8774 * instruction boundaries, i.e. the start of an instruction.
8775 * Trap-like exceptions, e.g. #DB, have higher priority than
8776 * NMIs and interrupts, i.e. traps are recognized before an
8777 * NMI/interrupt that's pending on the same instruction.
8778 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8779 * priority, but are only generated (pended) during instruction
8780 * execution, i.e. a pending fault-like exception means the
8781 * fault occurred on the *previous* instruction and must be
8782 * serviced prior to recognizing any new events in order to
8783 * fully complete the previous instruction.
8785 else if (!vcpu->arch.exception.pending) {
8786 if (vcpu->arch.nmi_injected) {
8787 static_call(kvm_x86_set_nmi)(vcpu);
8789 } else if (vcpu->arch.interrupt.injected) {
8790 static_call(kvm_x86_set_irq)(vcpu);
8795 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8796 vcpu->arch.exception.pending);
8799 * Call check_nested_events() even if we reinjected a previous event
8800 * in order for caller to determine if it should require immediate-exit
8801 * from L2 to L1 due to pending L1 events which require exit
8804 if (is_guest_mode(vcpu)) {
8805 r = kvm_check_nested_events(vcpu);
8810 /* try to inject new event if pending */
8811 if (vcpu->arch.exception.pending) {
8812 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8813 vcpu->arch.exception.has_error_code,
8814 vcpu->arch.exception.error_code);
8816 vcpu->arch.exception.pending = false;
8817 vcpu->arch.exception.injected = true;
8819 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8820 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8823 if (vcpu->arch.exception.nr == DB_VECTOR) {
8824 kvm_deliver_exception_payload(vcpu);
8825 if (vcpu->arch.dr7 & DR7_GD) {
8826 vcpu->arch.dr7 &= ~DR7_GD;
8827 kvm_update_dr7(vcpu);
8831 kvm_inject_exception(vcpu);
8836 * Finally, inject interrupt events. If an event cannot be injected
8837 * due to architectural conditions (e.g. IF=0) a window-open exit
8838 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8839 * and can architecturally be injected, but we cannot do it right now:
8840 * an interrupt could have arrived just now and we have to inject it
8841 * as a vmexit, or there could already an event in the queue, which is
8842 * indicated by can_inject. In that case we request an immediate exit
8843 * in order to make progress and get back here for another iteration.
8844 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8846 if (vcpu->arch.smi_pending) {
8847 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8851 vcpu->arch.smi_pending = false;
8852 ++vcpu->arch.smi_count;
8856 static_call(kvm_x86_enable_smi_window)(vcpu);
8859 if (vcpu->arch.nmi_pending) {
8860 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8864 --vcpu->arch.nmi_pending;
8865 vcpu->arch.nmi_injected = true;
8866 static_call(kvm_x86_set_nmi)(vcpu);
8868 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8870 if (vcpu->arch.nmi_pending)
8871 static_call(kvm_x86_enable_nmi_window)(vcpu);
8874 if (kvm_cpu_has_injectable_intr(vcpu)) {
8875 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8879 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8880 static_call(kvm_x86_set_irq)(vcpu);
8881 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8883 if (kvm_cpu_has_injectable_intr(vcpu))
8884 static_call(kvm_x86_enable_irq_window)(vcpu);
8887 if (is_guest_mode(vcpu) &&
8888 kvm_x86_ops.nested_ops->hv_timer_pending &&
8889 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8890 *req_immediate_exit = true;
8892 WARN_ON(vcpu->arch.exception.pending);
8897 *req_immediate_exit = true;
8903 static void process_nmi(struct kvm_vcpu *vcpu)
8908 * x86 is limited to one NMI running, and one NMI pending after it.
8909 * If an NMI is already in progress, limit further NMIs to just one.
8910 * Otherwise, allow two (and we'll inject the first one immediately).
8912 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8915 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8916 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8917 kvm_make_request(KVM_REQ_EVENT, vcpu);
8920 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8923 flags |= seg->g << 23;
8924 flags |= seg->db << 22;
8925 flags |= seg->l << 21;
8926 flags |= seg->avl << 20;
8927 flags |= seg->present << 15;
8928 flags |= seg->dpl << 13;
8929 flags |= seg->s << 12;
8930 flags |= seg->type << 8;
8934 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8936 struct kvm_segment seg;
8939 kvm_get_segment(vcpu, &seg, n);
8940 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8943 offset = 0x7f84 + n * 12;
8945 offset = 0x7f2c + (n - 3) * 12;
8947 put_smstate(u32, buf, offset + 8, seg.base);
8948 put_smstate(u32, buf, offset + 4, seg.limit);
8949 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8952 #ifdef CONFIG_X86_64
8953 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8955 struct kvm_segment seg;
8959 kvm_get_segment(vcpu, &seg, n);
8960 offset = 0x7e00 + n * 16;
8962 flags = enter_smm_get_segment_flags(&seg) >> 8;
8963 put_smstate(u16, buf, offset, seg.selector);
8964 put_smstate(u16, buf, offset + 2, flags);
8965 put_smstate(u32, buf, offset + 4, seg.limit);
8966 put_smstate(u64, buf, offset + 8, seg.base);
8970 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8973 struct kvm_segment seg;
8977 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8978 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8979 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8980 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8982 for (i = 0; i < 8; i++)
8983 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
8985 kvm_get_dr(vcpu, 6, &val);
8986 put_smstate(u32, buf, 0x7fcc, (u32)val);
8987 kvm_get_dr(vcpu, 7, &val);
8988 put_smstate(u32, buf, 0x7fc8, (u32)val);
8990 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8991 put_smstate(u32, buf, 0x7fc4, seg.selector);
8992 put_smstate(u32, buf, 0x7f64, seg.base);
8993 put_smstate(u32, buf, 0x7f60, seg.limit);
8994 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8996 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8997 put_smstate(u32, buf, 0x7fc0, seg.selector);
8998 put_smstate(u32, buf, 0x7f80, seg.base);
8999 put_smstate(u32, buf, 0x7f7c, seg.limit);
9000 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9002 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9003 put_smstate(u32, buf, 0x7f74, dt.address);
9004 put_smstate(u32, buf, 0x7f70, dt.size);
9006 static_call(kvm_x86_get_idt)(vcpu, &dt);
9007 put_smstate(u32, buf, 0x7f58, dt.address);
9008 put_smstate(u32, buf, 0x7f54, dt.size);
9010 for (i = 0; i < 6; i++)
9011 enter_smm_save_seg_32(vcpu, buf, i);
9013 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9016 put_smstate(u32, buf, 0x7efc, 0x00020000);
9017 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9020 #ifdef CONFIG_X86_64
9021 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9024 struct kvm_segment seg;
9028 for (i = 0; i < 16; i++)
9029 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9031 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9032 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9034 kvm_get_dr(vcpu, 6, &val);
9035 put_smstate(u64, buf, 0x7f68, val);
9036 kvm_get_dr(vcpu, 7, &val);
9037 put_smstate(u64, buf, 0x7f60, val);
9039 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9040 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9041 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9043 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9046 put_smstate(u32, buf, 0x7efc, 0x00020064);
9048 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9050 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9051 put_smstate(u16, buf, 0x7e90, seg.selector);
9052 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9053 put_smstate(u32, buf, 0x7e94, seg.limit);
9054 put_smstate(u64, buf, 0x7e98, seg.base);
9056 static_call(kvm_x86_get_idt)(vcpu, &dt);
9057 put_smstate(u32, buf, 0x7e84, dt.size);
9058 put_smstate(u64, buf, 0x7e88, dt.address);
9060 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9061 put_smstate(u16, buf, 0x7e70, seg.selector);
9062 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9063 put_smstate(u32, buf, 0x7e74, seg.limit);
9064 put_smstate(u64, buf, 0x7e78, seg.base);
9066 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9067 put_smstate(u32, buf, 0x7e64, dt.size);
9068 put_smstate(u64, buf, 0x7e68, dt.address);
9070 for (i = 0; i < 6; i++)
9071 enter_smm_save_seg_64(vcpu, buf, i);
9075 static void enter_smm(struct kvm_vcpu *vcpu)
9077 struct kvm_segment cs, ds;
9082 memset(buf, 0, 512);
9083 #ifdef CONFIG_X86_64
9084 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9085 enter_smm_save_state_64(vcpu, buf);
9088 enter_smm_save_state_32(vcpu, buf);
9091 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9092 * state (e.g. leave guest mode) after we've saved the state into the
9093 * SMM state-save area.
9095 static_call(kvm_x86_enter_smm)(vcpu, buf);
9097 kvm_smm_changed(vcpu, true);
9098 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9100 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9101 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9103 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9105 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9106 kvm_rip_write(vcpu, 0x8000);
9108 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9109 static_call(kvm_x86_set_cr0)(vcpu, cr0);
9110 vcpu->arch.cr0 = cr0;
9112 static_call(kvm_x86_set_cr4)(vcpu, 0);
9114 /* Undocumented: IDT limit is set to zero on entry to SMM. */
9115 dt.address = dt.size = 0;
9116 static_call(kvm_x86_set_idt)(vcpu, &dt);
9118 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9120 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9121 cs.base = vcpu->arch.smbase;
9126 cs.limit = ds.limit = 0xffffffff;
9127 cs.type = ds.type = 0x3;
9128 cs.dpl = ds.dpl = 0;
9133 cs.avl = ds.avl = 0;
9134 cs.present = ds.present = 1;
9135 cs.unusable = ds.unusable = 0;
9136 cs.padding = ds.padding = 0;
9138 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9139 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9140 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9141 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9142 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9143 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9145 #ifdef CONFIG_X86_64
9146 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9147 static_call(kvm_x86_set_efer)(vcpu, 0);
9150 kvm_update_cpuid_runtime(vcpu);
9151 kvm_mmu_reset_context(vcpu);
9154 static void process_smi(struct kvm_vcpu *vcpu)
9156 vcpu->arch.smi_pending = true;
9157 kvm_make_request(KVM_REQ_EVENT, vcpu);
9160 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9161 unsigned long *vcpu_bitmap)
9165 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
9167 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
9168 NULL, vcpu_bitmap, cpus);
9170 free_cpumask_var(cpus);
9173 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9175 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9178 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9180 if (!lapic_in_kernel(vcpu))
9183 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
9184 kvm_apic_update_apicv(vcpu);
9185 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9188 * When APICv gets disabled, we may still have injected interrupts
9189 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9190 * still active when the interrupt got accepted. Make sure
9191 * inject_pending_event() is called to check for that.
9193 if (!vcpu->arch.apicv_active)
9194 kvm_make_request(KVM_REQ_EVENT, vcpu);
9196 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9199 * NOTE: Do not hold any lock prior to calling this.
9201 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
9202 * locked, because it calls __x86_set_memory_region() which does
9203 * synchronize_srcu(&kvm->srcu).
9205 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9207 struct kvm_vcpu *except;
9208 unsigned long old, new, expected;
9210 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9211 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9214 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
9216 expected = new = old;
9218 __clear_bit(bit, &new);
9220 __set_bit(bit, &new);
9223 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
9224 } while (old != expected);
9229 trace_kvm_apicv_update_request(activate, bit);
9230 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
9231 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
9234 * Sending request to update APICV for all other vcpus,
9235 * while update the calling vcpu immediately instead of
9236 * waiting for another #VMEXIT to handle the request.
9238 except = kvm_get_running_vcpu();
9239 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
9242 kvm_vcpu_update_apicv(except);
9244 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9246 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9248 if (!kvm_apic_present(vcpu))
9251 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9253 if (irqchip_split(vcpu->kvm))
9254 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9256 if (vcpu->arch.apicv_active)
9257 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9258 if (ioapic_in_kernel(vcpu->kvm))
9259 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9262 if (is_guest_mode(vcpu))
9263 vcpu->arch.load_eoi_exitmap_pending = true;
9265 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9268 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9270 u64 eoi_exit_bitmap[4];
9272 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9275 if (to_hv_vcpu(vcpu))
9276 bitmap_or((ulong *)eoi_exit_bitmap,
9277 vcpu->arch.ioapic_handled_vectors,
9278 to_hv_synic(vcpu)->vec_bitmap, 256);
9280 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9283 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9284 unsigned long start, unsigned long end)
9286 unsigned long apic_address;
9289 * The physical address of apic access page is stored in the VMCS.
9290 * Update it when it becomes invalid.
9292 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9293 if (start <= apic_address && apic_address < end)
9294 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9297 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9299 if (!lapic_in_kernel(vcpu))
9302 if (!kvm_x86_ops.set_apic_access_page_addr)
9305 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9308 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9310 smp_send_reschedule(vcpu->cpu);
9312 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9315 * Returns 1 to let vcpu_run() continue the guest execution loop without
9316 * exiting to the userspace. Otherwise, the value will be returned to the
9319 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9323 dm_request_for_irq_injection(vcpu) &&
9324 kvm_cpu_accept_dm_intr(vcpu);
9325 fastpath_t exit_fastpath;
9327 bool req_immediate_exit = false;
9329 /* Forbid vmenter if vcpu dirty ring is soft-full */
9330 if (unlikely(vcpu->kvm->dirty_ring_size &&
9331 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9332 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9333 trace_kvm_dirty_ring_exit(vcpu);
9338 if (kvm_request_pending(vcpu)) {
9339 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9340 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9345 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9346 kvm_mmu_unload(vcpu);
9347 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9348 __kvm_migrate_timers(vcpu);
9349 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9350 kvm_gen_update_masterclock(vcpu->kvm);
9351 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9352 kvm_gen_kvmclock_update(vcpu);
9353 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9354 r = kvm_guest_time_update(vcpu);
9358 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9359 kvm_mmu_sync_roots(vcpu);
9360 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9361 kvm_mmu_load_pgd(vcpu);
9362 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9363 kvm_vcpu_flush_tlb_all(vcpu);
9365 /* Flushing all ASIDs flushes the current ASID... */
9366 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9368 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9369 kvm_vcpu_flush_tlb_current(vcpu);
9370 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9371 kvm_vcpu_flush_tlb_guest(vcpu);
9373 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9374 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9378 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9379 if (is_guest_mode(vcpu)) {
9380 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9382 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9383 vcpu->mmio_needed = 0;
9388 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9389 /* Page is swapped out. Do synthetic halt */
9390 vcpu->arch.apf.halted = true;
9394 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9395 record_steal_time(vcpu);
9396 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9398 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9400 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9401 kvm_pmu_handle_event(vcpu);
9402 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9403 kvm_pmu_deliver_pmi(vcpu);
9404 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9405 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9406 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9407 vcpu->arch.ioapic_handled_vectors)) {
9408 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9409 vcpu->run->eoi.vector =
9410 vcpu->arch.pending_ioapic_eoi;
9415 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9416 vcpu_scan_ioapic(vcpu);
9417 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9418 vcpu_load_eoi_exitmap(vcpu);
9419 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9420 kvm_vcpu_reload_apic_access_page(vcpu);
9421 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9422 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9423 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9427 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9428 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9429 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9433 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9434 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9436 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9437 vcpu->run->hyperv = hv_vcpu->exit;
9443 * KVM_REQ_HV_STIMER has to be processed after
9444 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9445 * depend on the guest clock being up-to-date
9447 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9448 kvm_hv_process_stimers(vcpu);
9449 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9450 kvm_vcpu_update_apicv(vcpu);
9451 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9452 kvm_check_async_pf_completion(vcpu);
9453 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9454 static_call(kvm_x86_msr_filter_changed)(vcpu);
9456 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9457 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9460 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9461 kvm_xen_has_interrupt(vcpu)) {
9462 ++vcpu->stat.req_event;
9463 r = kvm_apic_accept_events(vcpu);
9468 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9473 r = inject_pending_event(vcpu, &req_immediate_exit);
9479 static_call(kvm_x86_enable_irq_window)(vcpu);
9481 if (kvm_lapic_enabled(vcpu)) {
9482 update_cr8_intercept(vcpu);
9483 kvm_lapic_sync_to_vapic(vcpu);
9487 r = kvm_mmu_reload(vcpu);
9489 goto cancel_injection;
9494 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9497 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9498 * IPI are then delayed after guest entry, which ensures that they
9499 * result in virtual interrupt delivery.
9501 local_irq_disable();
9502 vcpu->mode = IN_GUEST_MODE;
9504 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9507 * 1) We should set ->mode before checking ->requests. Please see
9508 * the comment in kvm_vcpu_exiting_guest_mode().
9510 * 2) For APICv, we should set ->mode before checking PID.ON. This
9511 * pairs with the memory barrier implicit in pi_test_and_set_on
9512 * (see vmx_deliver_posted_interrupt).
9514 * 3) This also orders the write to mode from any reads to the page
9515 * tables done while the VCPU is running. Please see the comment
9516 * in kvm_flush_remote_tlbs.
9518 smp_mb__after_srcu_read_unlock();
9521 * This handles the case where a posted interrupt was
9522 * notified with kvm_vcpu_kick.
9524 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9525 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9527 if (kvm_vcpu_exit_request(vcpu)) {
9528 vcpu->mode = OUTSIDE_GUEST_MODE;
9532 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9534 goto cancel_injection;
9537 if (req_immediate_exit) {
9538 kvm_make_request(KVM_REQ_EVENT, vcpu);
9539 static_call(kvm_x86_request_immediate_exit)(vcpu);
9542 fpregs_assert_state_consistent();
9543 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9544 switch_fpu_return();
9546 if (unlikely(vcpu->arch.switch_db_regs)) {
9548 set_debugreg(vcpu->arch.eff_db[0], 0);
9549 set_debugreg(vcpu->arch.eff_db[1], 1);
9550 set_debugreg(vcpu->arch.eff_db[2], 2);
9551 set_debugreg(vcpu->arch.eff_db[3], 3);
9552 set_debugreg(vcpu->arch.dr6, 6);
9553 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9557 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9558 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9561 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9562 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9566 if (vcpu->arch.apicv_active)
9567 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9571 * Do this here before restoring debug registers on the host. And
9572 * since we do this before handling the vmexit, a DR access vmexit
9573 * can (a) read the correct value of the debug registers, (b) set
9574 * KVM_DEBUGREG_WONT_EXIT again.
9576 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9577 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9578 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9579 kvm_update_dr0123(vcpu);
9580 kvm_update_dr7(vcpu);
9581 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9585 * If the guest has used debug registers, at least dr7
9586 * will be disabled while returning to the host.
9587 * If we don't have active breakpoints in the host, we don't
9588 * care about the messed up debug address registers. But if
9589 * we have some of them active, restore the old state.
9591 if (hw_breakpoint_active())
9592 hw_breakpoint_restore();
9594 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9595 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9597 vcpu->mode = OUTSIDE_GUEST_MODE;
9600 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9603 * Consume any pending interrupts, including the possible source of
9604 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9605 * An instruction is required after local_irq_enable() to fully unblock
9606 * interrupts on processors that implement an interrupt shadow, the
9607 * stat.exits increment will do nicely.
9609 kvm_before_interrupt(vcpu);
9612 local_irq_disable();
9613 kvm_after_interrupt(vcpu);
9616 * Wait until after servicing IRQs to account guest time so that any
9617 * ticks that occurred while running the guest are properly accounted
9618 * to the guest. Waiting until IRQs are enabled degrades the accuracy
9619 * of accounting via context tracking, but the loss of accuracy is
9620 * acceptable for all known use cases.
9622 vtime_account_guest_exit();
9624 if (lapic_in_kernel(vcpu)) {
9625 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9626 if (delta != S64_MIN) {
9627 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9628 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9635 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9638 * Profile KVM exit RIPs:
9640 if (unlikely(prof_on == KVM_PROFILING)) {
9641 unsigned long rip = kvm_rip_read(vcpu);
9642 profile_hit(KVM_PROFILING, (void *)rip);
9645 if (unlikely(vcpu->arch.tsc_always_catchup))
9646 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9648 if (vcpu->arch.apic_attention)
9649 kvm_lapic_sync_from_vapic(vcpu);
9651 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9655 if (req_immediate_exit)
9656 kvm_make_request(KVM_REQ_EVENT, vcpu);
9657 static_call(kvm_x86_cancel_injection)(vcpu);
9658 if (unlikely(vcpu->arch.apic_attention))
9659 kvm_lapic_sync_from_vapic(vcpu);
9664 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9666 if (!kvm_arch_vcpu_runnable(vcpu) &&
9667 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9668 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9669 kvm_vcpu_block(vcpu);
9670 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9672 if (kvm_x86_ops.post_block)
9673 static_call(kvm_x86_post_block)(vcpu);
9675 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9679 if (kvm_apic_accept_events(vcpu) < 0)
9681 switch(vcpu->arch.mp_state) {
9682 case KVM_MP_STATE_HALTED:
9683 case KVM_MP_STATE_AP_RESET_HOLD:
9684 vcpu->arch.pv.pv_unhalted = false;
9685 vcpu->arch.mp_state =
9686 KVM_MP_STATE_RUNNABLE;
9688 case KVM_MP_STATE_RUNNABLE:
9689 vcpu->arch.apf.halted = false;
9691 case KVM_MP_STATE_INIT_RECEIVED:
9699 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9701 if (is_guest_mode(vcpu))
9702 kvm_check_nested_events(vcpu);
9704 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9705 !vcpu->arch.apf.halted);
9708 static int vcpu_run(struct kvm_vcpu *vcpu)
9711 struct kvm *kvm = vcpu->kvm;
9713 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9714 vcpu->arch.l1tf_flush_l1d = true;
9717 if (kvm_vcpu_running(vcpu)) {
9718 r = vcpu_enter_guest(vcpu);
9720 r = vcpu_block(kvm, vcpu);
9726 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9727 if (kvm_cpu_has_pending_timer(vcpu))
9728 kvm_inject_pending_timer_irqs(vcpu);
9730 if (dm_request_for_irq_injection(vcpu) &&
9731 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9733 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9734 ++vcpu->stat.request_irq_exits;
9738 if (__xfer_to_guest_mode_work_pending()) {
9739 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9740 r = xfer_to_guest_mode_handle_work(vcpu);
9743 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9747 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9752 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9756 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9757 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9758 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9762 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9764 BUG_ON(!vcpu->arch.pio.count);
9766 return complete_emulated_io(vcpu);
9770 * Implements the following, as a state machine:
9774 * for each mmio piece in the fragment
9782 * for each mmio piece in the fragment
9787 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9789 struct kvm_run *run = vcpu->run;
9790 struct kvm_mmio_fragment *frag;
9793 BUG_ON(!vcpu->mmio_needed);
9795 /* Complete previous fragment */
9796 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9797 len = min(8u, frag->len);
9798 if (!vcpu->mmio_is_write)
9799 memcpy(frag->data, run->mmio.data, len);
9801 if (frag->len <= 8) {
9802 /* Switch to the next fragment. */
9804 vcpu->mmio_cur_fragment++;
9806 /* Go forward to the next mmio piece. */
9812 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9813 vcpu->mmio_needed = 0;
9815 /* FIXME: return into emulator if single-stepping. */
9816 if (vcpu->mmio_is_write)
9818 vcpu->mmio_read_completed = 1;
9819 return complete_emulated_io(vcpu);
9822 run->exit_reason = KVM_EXIT_MMIO;
9823 run->mmio.phys_addr = frag->gpa;
9824 if (vcpu->mmio_is_write)
9825 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9826 run->mmio.len = min(8u, frag->len);
9827 run->mmio.is_write = vcpu->mmio_is_write;
9828 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9832 static void kvm_save_current_fpu(struct fpu *fpu)
9835 * If the target FPU state is not resident in the CPU registers, just
9836 * memcpy() from current, else save CPU state directly to the target.
9838 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9839 memcpy(&fpu->state, ¤t->thread.fpu.state,
9840 fpu_kernel_xstate_size);
9842 copy_fpregs_to_fpstate(fpu);
9845 /* Swap (qemu) user FPU context for the guest FPU context. */
9846 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9850 kvm_save_current_fpu(vcpu->arch.user_fpu);
9853 * Guests with protected state can't have it set by the hypervisor,
9854 * so skip trying to set it.
9856 if (vcpu->arch.guest_fpu)
9857 /* PKRU is separately restored in kvm_x86_ops.run. */
9858 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9859 ~XFEATURE_MASK_PKRU);
9861 fpregs_mark_activate();
9867 /* When vcpu_run ends, restore user space FPU context. */
9868 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9873 * Guests with protected state can't have it read by the hypervisor,
9874 * so skip trying to save it.
9876 if (vcpu->arch.guest_fpu)
9877 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9879 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9881 fpregs_mark_activate();
9884 ++vcpu->stat.fpu_reload;
9888 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9890 struct kvm_run *kvm_run = vcpu->run;
9894 kvm_sigset_activate(vcpu);
9896 kvm_load_guest_fpu(vcpu);
9898 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9899 if (kvm_run->immediate_exit) {
9903 kvm_vcpu_block(vcpu);
9904 if (kvm_apic_accept_events(vcpu) < 0) {
9908 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9910 if (signal_pending(current)) {
9912 kvm_run->exit_reason = KVM_EXIT_INTR;
9913 ++vcpu->stat.signal_exits;
9918 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9923 if (kvm_run->kvm_dirty_regs) {
9924 r = sync_regs(vcpu);
9929 /* re-sync apic's tpr */
9930 if (!lapic_in_kernel(vcpu)) {
9931 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9937 if (unlikely(vcpu->arch.complete_userspace_io)) {
9938 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9939 vcpu->arch.complete_userspace_io = NULL;
9944 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9946 if (kvm_run->immediate_exit)
9952 kvm_put_guest_fpu(vcpu);
9953 if (kvm_run->kvm_valid_regs)
9955 post_kvm_run_save(vcpu);
9956 kvm_sigset_deactivate(vcpu);
9962 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9964 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9966 * We are here if userspace calls get_regs() in the middle of
9967 * instruction emulation. Registers state needs to be copied
9968 * back from emulation context to vcpu. Userspace shouldn't do
9969 * that usually, but some bad designed PV devices (vmware
9970 * backdoor interface) need this to work
9972 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9973 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9975 regs->rax = kvm_rax_read(vcpu);
9976 regs->rbx = kvm_rbx_read(vcpu);
9977 regs->rcx = kvm_rcx_read(vcpu);
9978 regs->rdx = kvm_rdx_read(vcpu);
9979 regs->rsi = kvm_rsi_read(vcpu);
9980 regs->rdi = kvm_rdi_read(vcpu);
9981 regs->rsp = kvm_rsp_read(vcpu);
9982 regs->rbp = kvm_rbp_read(vcpu);
9983 #ifdef CONFIG_X86_64
9984 regs->r8 = kvm_r8_read(vcpu);
9985 regs->r9 = kvm_r9_read(vcpu);
9986 regs->r10 = kvm_r10_read(vcpu);
9987 regs->r11 = kvm_r11_read(vcpu);
9988 regs->r12 = kvm_r12_read(vcpu);
9989 regs->r13 = kvm_r13_read(vcpu);
9990 regs->r14 = kvm_r14_read(vcpu);
9991 regs->r15 = kvm_r15_read(vcpu);
9994 regs->rip = kvm_rip_read(vcpu);
9995 regs->rflags = kvm_get_rflags(vcpu);
9998 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10001 __get_regs(vcpu, regs);
10006 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10008 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10009 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10011 kvm_rax_write(vcpu, regs->rax);
10012 kvm_rbx_write(vcpu, regs->rbx);
10013 kvm_rcx_write(vcpu, regs->rcx);
10014 kvm_rdx_write(vcpu, regs->rdx);
10015 kvm_rsi_write(vcpu, regs->rsi);
10016 kvm_rdi_write(vcpu, regs->rdi);
10017 kvm_rsp_write(vcpu, regs->rsp);
10018 kvm_rbp_write(vcpu, regs->rbp);
10019 #ifdef CONFIG_X86_64
10020 kvm_r8_write(vcpu, regs->r8);
10021 kvm_r9_write(vcpu, regs->r9);
10022 kvm_r10_write(vcpu, regs->r10);
10023 kvm_r11_write(vcpu, regs->r11);
10024 kvm_r12_write(vcpu, regs->r12);
10025 kvm_r13_write(vcpu, regs->r13);
10026 kvm_r14_write(vcpu, regs->r14);
10027 kvm_r15_write(vcpu, regs->r15);
10030 kvm_rip_write(vcpu, regs->rip);
10031 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10033 vcpu->arch.exception.pending = false;
10035 kvm_make_request(KVM_REQ_EVENT, vcpu);
10038 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10041 __set_regs(vcpu, regs);
10046 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10048 struct kvm_segment cs;
10050 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10054 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10056 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10058 struct desc_ptr dt;
10060 if (vcpu->arch.guest_state_protected)
10061 goto skip_protected_regs;
10063 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10064 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10065 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10066 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10067 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10068 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10070 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10071 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10073 static_call(kvm_x86_get_idt)(vcpu, &dt);
10074 sregs->idt.limit = dt.size;
10075 sregs->idt.base = dt.address;
10076 static_call(kvm_x86_get_gdt)(vcpu, &dt);
10077 sregs->gdt.limit = dt.size;
10078 sregs->gdt.base = dt.address;
10080 sregs->cr2 = vcpu->arch.cr2;
10081 sregs->cr3 = kvm_read_cr3(vcpu);
10083 skip_protected_regs:
10084 sregs->cr0 = kvm_read_cr0(vcpu);
10085 sregs->cr4 = kvm_read_cr4(vcpu);
10086 sregs->cr8 = kvm_get_cr8(vcpu);
10087 sregs->efer = vcpu->arch.efer;
10088 sregs->apic_base = kvm_get_apic_base(vcpu);
10091 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10093 __get_sregs_common(vcpu, sregs);
10095 if (vcpu->arch.guest_state_protected)
10098 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10099 set_bit(vcpu->arch.interrupt.nr,
10100 (unsigned long *)sregs->interrupt_bitmap);
10103 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10107 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10109 if (vcpu->arch.guest_state_protected)
10112 if (is_pae_paging(vcpu)) {
10113 for (i = 0 ; i < 4 ; i++)
10114 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10115 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10119 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10120 struct kvm_sregs *sregs)
10123 __get_sregs(vcpu, sregs);
10128 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10129 struct kvm_mp_state *mp_state)
10134 if (kvm_mpx_supported())
10135 kvm_load_guest_fpu(vcpu);
10137 r = kvm_apic_accept_events(vcpu);
10142 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10143 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10144 vcpu->arch.pv.pv_unhalted)
10145 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10147 mp_state->mp_state = vcpu->arch.mp_state;
10150 if (kvm_mpx_supported())
10151 kvm_put_guest_fpu(vcpu);
10156 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10157 struct kvm_mp_state *mp_state)
10163 if (!lapic_in_kernel(vcpu) &&
10164 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10168 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10169 * INIT state; latched init should be reported using
10170 * KVM_SET_VCPU_EVENTS, so reject it here.
10172 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10173 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10174 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10177 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10178 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10179 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10181 vcpu->arch.mp_state = mp_state->mp_state;
10182 kvm_make_request(KVM_REQ_EVENT, vcpu);
10190 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10191 int reason, bool has_error_code, u32 error_code)
10193 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10196 init_emulate_ctxt(vcpu);
10198 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10199 has_error_code, error_code);
10201 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10202 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10203 vcpu->run->internal.ndata = 0;
10207 kvm_rip_write(vcpu, ctxt->eip);
10208 kvm_set_rflags(vcpu, ctxt->eflags);
10211 EXPORT_SYMBOL_GPL(kvm_task_switch);
10213 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10215 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10217 * When EFER.LME and CR0.PG are set, the processor is in
10218 * 64-bit mode (though maybe in a 32-bit code segment).
10219 * CR4.PAE and EFER.LMA must be set.
10221 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10223 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10227 * Not in 64-bit mode: EFER.LMA is clear and the code
10228 * segment cannot be 64-bit.
10230 if (sregs->efer & EFER_LMA || sregs->cs.l)
10234 return kvm_is_valid_cr4(vcpu, sregs->cr4);
10237 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10238 int *mmu_reset_needed, bool update_pdptrs)
10240 struct msr_data apic_base_msr;
10242 struct desc_ptr dt;
10244 if (!kvm_is_valid_sregs(vcpu, sregs))
10247 apic_base_msr.data = sregs->apic_base;
10248 apic_base_msr.host_initiated = true;
10249 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10252 if (vcpu->arch.guest_state_protected)
10255 dt.size = sregs->idt.limit;
10256 dt.address = sregs->idt.base;
10257 static_call(kvm_x86_set_idt)(vcpu, &dt);
10258 dt.size = sregs->gdt.limit;
10259 dt.address = sregs->gdt.base;
10260 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10262 vcpu->arch.cr2 = sregs->cr2;
10263 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10264 vcpu->arch.cr3 = sregs->cr3;
10265 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10267 kvm_set_cr8(vcpu, sregs->cr8);
10269 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10270 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10272 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10273 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10274 vcpu->arch.cr0 = sregs->cr0;
10276 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10277 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10279 if (update_pdptrs) {
10280 idx = srcu_read_lock(&vcpu->kvm->srcu);
10281 if (is_pae_paging(vcpu)) {
10282 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10283 *mmu_reset_needed = 1;
10285 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10288 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10289 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10290 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10291 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10292 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10293 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10295 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10296 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10298 update_cr8_intercept(vcpu);
10300 /* Older userspace won't unhalt the vcpu on reset. */
10301 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10302 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10303 !is_protmode(vcpu))
10304 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10309 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10311 int pending_vec, max_bits;
10312 int mmu_reset_needed = 0;
10313 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10318 if (mmu_reset_needed)
10319 kvm_mmu_reset_context(vcpu);
10321 max_bits = KVM_NR_INTERRUPTS;
10322 pending_vec = find_first_bit(
10323 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10325 if (pending_vec < max_bits) {
10326 kvm_queue_interrupt(vcpu, pending_vec, false);
10327 pr_debug("Set back pending irq %d\n", pending_vec);
10328 kvm_make_request(KVM_REQ_EVENT, vcpu);
10333 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10335 int mmu_reset_needed = 0;
10336 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10337 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10338 !(sregs2->efer & EFER_LMA);
10341 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10344 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10347 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10348 &mmu_reset_needed, !valid_pdptrs);
10352 if (valid_pdptrs) {
10353 for (i = 0; i < 4 ; i++)
10354 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10356 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10357 mmu_reset_needed = 1;
10358 vcpu->arch.pdptrs_from_userspace = true;
10360 if (mmu_reset_needed)
10361 kvm_mmu_reset_context(vcpu);
10365 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10366 struct kvm_sregs *sregs)
10371 ret = __set_sregs(vcpu, sregs);
10376 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10377 struct kvm_guest_debug *dbg)
10379 unsigned long rflags;
10382 if (vcpu->arch.guest_state_protected)
10387 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10389 if (vcpu->arch.exception.pending)
10391 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10392 kvm_queue_exception(vcpu, DB_VECTOR);
10394 kvm_queue_exception(vcpu, BP_VECTOR);
10398 * Read rflags as long as potentially injected trace flags are still
10401 rflags = kvm_get_rflags(vcpu);
10403 vcpu->guest_debug = dbg->control;
10404 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10405 vcpu->guest_debug = 0;
10407 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10408 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10409 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10410 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10412 for (i = 0; i < KVM_NR_DB_REGS; i++)
10413 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10415 kvm_update_dr7(vcpu);
10417 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10418 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10421 * Trigger an rflags update that will inject or remove the trace
10424 kvm_set_rflags(vcpu, rflags);
10426 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10436 * Translate a guest virtual address to a guest physical address.
10438 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10439 struct kvm_translation *tr)
10441 unsigned long vaddr = tr->linear_address;
10447 idx = srcu_read_lock(&vcpu->kvm->srcu);
10448 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10449 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10450 tr->physical_address = gpa;
10451 tr->valid = gpa != UNMAPPED_GVA;
10459 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10461 struct fxregs_state *fxsave;
10463 if (!vcpu->arch.guest_fpu)
10468 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10469 memcpy(fpu->fpr, fxsave->st_space, 128);
10470 fpu->fcw = fxsave->cwd;
10471 fpu->fsw = fxsave->swd;
10472 fpu->ftwx = fxsave->twd;
10473 fpu->last_opcode = fxsave->fop;
10474 fpu->last_ip = fxsave->rip;
10475 fpu->last_dp = fxsave->rdp;
10476 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10482 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10484 struct fxregs_state *fxsave;
10486 if (!vcpu->arch.guest_fpu)
10491 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10493 memcpy(fxsave->st_space, fpu->fpr, 128);
10494 fxsave->cwd = fpu->fcw;
10495 fxsave->swd = fpu->fsw;
10496 fxsave->twd = fpu->ftwx;
10497 fxsave->fop = fpu->last_opcode;
10498 fxsave->rip = fpu->last_ip;
10499 fxsave->rdp = fpu->last_dp;
10500 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10506 static void store_regs(struct kvm_vcpu *vcpu)
10508 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10510 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10511 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10513 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10514 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10516 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10517 kvm_vcpu_ioctl_x86_get_vcpu_events(
10518 vcpu, &vcpu->run->s.regs.events);
10521 static int sync_regs(struct kvm_vcpu *vcpu)
10523 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10526 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10527 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10528 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10530 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10531 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10533 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10535 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10536 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10537 vcpu, &vcpu->run->s.regs.events))
10539 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10545 static void fx_init(struct kvm_vcpu *vcpu)
10547 if (!vcpu->arch.guest_fpu)
10550 fpstate_init(&vcpu->arch.guest_fpu->state);
10551 if (boot_cpu_has(X86_FEATURE_XSAVES))
10552 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10553 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10556 * Ensure guest xcr0 is valid for loading
10558 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10560 vcpu->arch.cr0 |= X86_CR0_ET;
10563 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10565 if (vcpu->arch.guest_fpu) {
10566 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10567 vcpu->arch.guest_fpu = NULL;
10570 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10572 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10574 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10575 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10576 "guest TSC will not be reliable\n");
10581 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10586 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10587 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10589 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10591 r = kvm_mmu_create(vcpu);
10595 if (irqchip_in_kernel(vcpu->kvm)) {
10596 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10598 goto fail_mmu_destroy;
10599 if (kvm_apicv_activated(vcpu->kvm))
10600 vcpu->arch.apicv_active = true;
10602 static_branch_inc(&kvm_has_noapic_vcpu);
10606 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10608 goto fail_free_lapic;
10609 vcpu->arch.pio_data = page_address(page);
10611 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10612 GFP_KERNEL_ACCOUNT);
10613 if (!vcpu->arch.mce_banks)
10614 goto fail_free_pio_data;
10615 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10617 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10618 GFP_KERNEL_ACCOUNT))
10619 goto fail_free_mce_banks;
10621 if (!alloc_emulate_ctxt(vcpu))
10622 goto free_wbinvd_dirty_mask;
10624 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10625 GFP_KERNEL_ACCOUNT);
10626 if (!vcpu->arch.user_fpu) {
10627 pr_err("kvm: failed to allocate userspace's fpu\n");
10628 goto free_emulate_ctxt;
10631 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10632 GFP_KERNEL_ACCOUNT);
10633 if (!vcpu->arch.guest_fpu) {
10634 pr_err("kvm: failed to allocate vcpu's fpu\n");
10635 goto free_user_fpu;
10639 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10640 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10642 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10644 kvm_async_pf_hash_reset(vcpu);
10645 kvm_pmu_init(vcpu);
10647 vcpu->arch.pending_external_vector = -1;
10648 vcpu->arch.preempted_in_kernel = false;
10650 #if IS_ENABLED(CONFIG_HYPERV)
10651 vcpu->arch.hv_root_tdp = INVALID_PAGE;
10654 r = static_call(kvm_x86_vcpu_create)(vcpu);
10656 goto free_guest_fpu;
10658 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10659 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10660 kvm_vcpu_mtrr_init(vcpu);
10662 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10663 kvm_vcpu_reset(vcpu, false);
10664 kvm_init_mmu(vcpu);
10669 kvm_free_guest_fpu(vcpu);
10671 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10673 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10674 free_wbinvd_dirty_mask:
10675 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10676 fail_free_mce_banks:
10677 kfree(vcpu->arch.mce_banks);
10678 fail_free_pio_data:
10679 free_page((unsigned long)vcpu->arch.pio_data);
10681 kvm_free_lapic(vcpu);
10683 kvm_mmu_destroy(vcpu);
10687 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10689 struct kvm *kvm = vcpu->kvm;
10691 if (mutex_lock_killable(&vcpu->mutex))
10694 kvm_synchronize_tsc(vcpu, 0);
10697 /* poll control enabled by default */
10698 vcpu->arch.msr_kvm_poll_control = 1;
10700 mutex_unlock(&vcpu->mutex);
10702 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10703 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10704 KVMCLOCK_SYNC_PERIOD);
10707 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10709 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10712 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10714 kvmclock_reset(vcpu);
10716 static_call(kvm_x86_vcpu_free)(vcpu);
10718 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10719 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10720 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10721 kvm_free_guest_fpu(vcpu);
10723 kvm_hv_vcpu_uninit(vcpu);
10724 kvm_pmu_destroy(vcpu);
10725 kfree(vcpu->arch.mce_banks);
10726 kvm_free_lapic(vcpu);
10727 idx = srcu_read_lock(&vcpu->kvm->srcu);
10728 kvm_mmu_destroy(vcpu);
10729 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10730 free_page((unsigned long)vcpu->arch.pio_data);
10731 kvfree(vcpu->arch.cpuid_entries);
10732 if (!lapic_in_kernel(vcpu))
10733 static_branch_dec(&kvm_has_noapic_vcpu);
10736 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10738 kvm_lapic_reset(vcpu, init_event);
10740 vcpu->arch.hflags = 0;
10742 vcpu->arch.smi_pending = 0;
10743 vcpu->arch.smi_count = 0;
10744 atomic_set(&vcpu->arch.nmi_queued, 0);
10745 vcpu->arch.nmi_pending = 0;
10746 vcpu->arch.nmi_injected = false;
10747 kvm_clear_interrupt_queue(vcpu);
10748 kvm_clear_exception_queue(vcpu);
10750 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10751 kvm_update_dr0123(vcpu);
10752 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10753 vcpu->arch.dr7 = DR7_FIXED_1;
10754 kvm_update_dr7(vcpu);
10756 vcpu->arch.cr2 = 0;
10758 kvm_make_request(KVM_REQ_EVENT, vcpu);
10759 vcpu->arch.apf.msr_en_val = 0;
10760 vcpu->arch.apf.msr_int_val = 0;
10761 vcpu->arch.st.msr_val = 0;
10763 kvmclock_reset(vcpu);
10765 kvm_clear_async_pf_completion_queue(vcpu);
10766 kvm_async_pf_hash_reset(vcpu);
10767 vcpu->arch.apf.halted = false;
10769 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10770 void *mpx_state_buffer;
10773 * To avoid have the INIT path from kvm_apic_has_events() that be
10774 * called with loaded FPU and does not let userspace fix the state.
10777 kvm_put_guest_fpu(vcpu);
10778 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10780 if (mpx_state_buffer)
10781 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10782 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10784 if (mpx_state_buffer)
10785 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10787 kvm_load_guest_fpu(vcpu);
10791 kvm_pmu_reset(vcpu);
10792 vcpu->arch.smbase = 0x30000;
10794 vcpu->arch.msr_misc_features_enables = 0;
10796 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10799 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10800 vcpu->arch.regs_avail = ~0;
10801 vcpu->arch.regs_dirty = ~0;
10803 vcpu->arch.ia32_xss = 0;
10805 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10808 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10810 struct kvm_segment cs;
10812 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10813 cs.selector = vector << 8;
10814 cs.base = vector << 12;
10815 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10816 kvm_rip_write(vcpu, 0);
10818 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10820 int kvm_arch_hardware_enable(void)
10823 struct kvm_vcpu *vcpu;
10828 bool stable, backwards_tsc = false;
10830 kvm_user_return_msr_cpu_online();
10831 ret = static_call(kvm_x86_hardware_enable)();
10835 local_tsc = rdtsc();
10836 stable = !kvm_check_tsc_unstable();
10837 list_for_each_entry(kvm, &vm_list, vm_list) {
10838 kvm_for_each_vcpu(i, vcpu, kvm) {
10839 if (!stable && vcpu->cpu == smp_processor_id())
10840 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10841 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10842 backwards_tsc = true;
10843 if (vcpu->arch.last_host_tsc > max_tsc)
10844 max_tsc = vcpu->arch.last_host_tsc;
10850 * Sometimes, even reliable TSCs go backwards. This happens on
10851 * platforms that reset TSC during suspend or hibernate actions, but
10852 * maintain synchronization. We must compensate. Fortunately, we can
10853 * detect that condition here, which happens early in CPU bringup,
10854 * before any KVM threads can be running. Unfortunately, we can't
10855 * bring the TSCs fully up to date with real time, as we aren't yet far
10856 * enough into CPU bringup that we know how much real time has actually
10857 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10858 * variables that haven't been updated yet.
10860 * So we simply find the maximum observed TSC above, then record the
10861 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10862 * the adjustment will be applied. Note that we accumulate
10863 * adjustments, in case multiple suspend cycles happen before some VCPU
10864 * gets a chance to run again. In the event that no KVM threads get a
10865 * chance to run, we will miss the entire elapsed period, as we'll have
10866 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10867 * loose cycle time. This isn't too big a deal, since the loss will be
10868 * uniform across all VCPUs (not to mention the scenario is extremely
10869 * unlikely). It is possible that a second hibernate recovery happens
10870 * much faster than a first, causing the observed TSC here to be
10871 * smaller; this would require additional padding adjustment, which is
10872 * why we set last_host_tsc to the local tsc observed here.
10874 * N.B. - this code below runs only on platforms with reliable TSC,
10875 * as that is the only way backwards_tsc is set above. Also note
10876 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10877 * have the same delta_cyc adjustment applied if backwards_tsc
10878 * is detected. Note further, this adjustment is only done once,
10879 * as we reset last_host_tsc on all VCPUs to stop this from being
10880 * called multiple times (one for each physical CPU bringup).
10882 * Platforms with unreliable TSCs don't have to deal with this, they
10883 * will be compensated by the logic in vcpu_load, which sets the TSC to
10884 * catchup mode. This will catchup all VCPUs to real time, but cannot
10885 * guarantee that they stay in perfect synchronization.
10887 if (backwards_tsc) {
10888 u64 delta_cyc = max_tsc - local_tsc;
10889 list_for_each_entry(kvm, &vm_list, vm_list) {
10890 kvm->arch.backwards_tsc_observed = true;
10891 kvm_for_each_vcpu(i, vcpu, kvm) {
10892 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10893 vcpu->arch.last_host_tsc = local_tsc;
10894 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10898 * We have to disable TSC offset matching.. if you were
10899 * booting a VM while issuing an S4 host suspend....
10900 * you may have some problem. Solving this issue is
10901 * left as an exercise to the reader.
10903 kvm->arch.last_tsc_nsec = 0;
10904 kvm->arch.last_tsc_write = 0;
10911 void kvm_arch_hardware_disable(void)
10913 static_call(kvm_x86_hardware_disable)();
10914 drop_user_return_notifiers();
10917 int kvm_arch_hardware_setup(void *opaque)
10919 struct kvm_x86_init_ops *ops = opaque;
10922 rdmsrl_safe(MSR_EFER, &host_efer);
10923 if (WARN_ON_ONCE(boot_cpu_has(X86_FEATURE_NX) &&
10924 !(host_efer & EFER_NX)))
10927 if (boot_cpu_has(X86_FEATURE_XSAVES))
10928 rdmsrl(MSR_IA32_XSS, host_xss);
10930 r = ops->hardware_setup();
10934 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10935 kvm_ops_static_call_update();
10937 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10940 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10941 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10942 #undef __kvm_cpu_cap_has
10944 if (kvm_has_tsc_control) {
10946 * Make sure the user can only configure tsc_khz values that
10947 * fit into a signed integer.
10948 * A min value is not calculated because it will always
10949 * be 1 on all machines.
10951 u64 max = min(0x7fffffffULL,
10952 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10953 kvm_max_guest_tsc_khz = max;
10955 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10958 kvm_init_msr_list();
10962 void kvm_arch_hardware_unsetup(void)
10964 static_call(kvm_x86_hardware_unsetup)();
10967 int kvm_arch_check_processor_compat(void *opaque)
10969 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10970 struct kvm_x86_init_ops *ops = opaque;
10972 WARN_ON(!irqs_disabled());
10974 if (__cr4_reserved_bits(cpu_has, c) !=
10975 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10978 return ops->check_processor_compatibility();
10981 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10983 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10985 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10987 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10989 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10992 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10993 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10995 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10997 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10999 vcpu->arch.l1tf_flush_l1d = true;
11000 if (pmu->version && unlikely(pmu->event_count)) {
11001 pmu->need_cleanup = true;
11002 kvm_make_request(KVM_REQ_PMU, vcpu);
11004 static_call(kvm_x86_sched_in)(vcpu, cpu);
11007 void kvm_arch_free_vm(struct kvm *kvm)
11009 kfree(to_kvm_hv(kvm)->hv_pa_pg);
11014 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11019 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11020 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11021 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11022 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11023 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11024 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11026 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11027 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11028 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11029 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11030 &kvm->arch.irq_sources_bitmap);
11032 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11033 mutex_init(&kvm->arch.apic_map_lock);
11034 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
11036 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11037 pvclock_update_vm_gtod_copy(kvm);
11039 kvm->arch.guest_can_read_msr_platform_info = true;
11041 #if IS_ENABLED(CONFIG_HYPERV)
11042 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11043 kvm->arch.hv_root_tdp = INVALID_PAGE;
11046 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11047 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11049 kvm_apicv_init(kvm);
11050 kvm_hv_init_vm(kvm);
11051 kvm_page_track_init(kvm);
11052 kvm_mmu_init_vm(kvm);
11054 return static_call(kvm_x86_vm_init)(kvm);
11057 int kvm_arch_post_init_vm(struct kvm *kvm)
11059 return kvm_mmu_post_init_vm(kvm);
11062 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11065 kvm_mmu_unload(vcpu);
11069 static void kvm_free_vcpus(struct kvm *kvm)
11072 struct kvm_vcpu *vcpu;
11075 * Unpin any mmu pages first.
11077 kvm_for_each_vcpu(i, vcpu, kvm) {
11078 kvm_clear_async_pf_completion_queue(vcpu);
11079 kvm_unload_vcpu_mmu(vcpu);
11081 kvm_for_each_vcpu(i, vcpu, kvm)
11082 kvm_vcpu_destroy(vcpu);
11084 mutex_lock(&kvm->lock);
11085 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11086 kvm->vcpus[i] = NULL;
11088 atomic_set(&kvm->online_vcpus, 0);
11089 mutex_unlock(&kvm->lock);
11092 void kvm_arch_sync_events(struct kvm *kvm)
11094 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11095 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11099 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
11102 * __x86_set_memory_region: Setup KVM internal memory slot
11104 * @kvm: the kvm pointer to the VM.
11105 * @id: the slot ID to setup.
11106 * @gpa: the GPA to install the slot (unused when @size == 0).
11107 * @size: the size of the slot. Set to zero to uninstall a slot.
11109 * This function helps to setup a KVM internal memory slot. Specify
11110 * @size > 0 to install a new slot, while @size == 0 to uninstall a
11111 * slot. The return code can be one of the following:
11113 * HVA: on success (uninstall will return a bogus HVA)
11116 * The caller should always use IS_ERR() to check the return value
11117 * before use. Note, the KVM internal memory slots are guaranteed to
11118 * remain valid and unchanged until the VM is destroyed, i.e., the
11119 * GPA->HVA translation will not change. However, the HVA is a user
11120 * address, i.e. its accessibility is not guaranteed, and must be
11121 * accessed via __copy_{to,from}_user().
11123 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11127 unsigned long hva, old_npages;
11128 struct kvm_memslots *slots = kvm_memslots(kvm);
11129 struct kvm_memory_slot *slot;
11131 /* Called with kvm->slots_lock held. */
11132 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11133 return ERR_PTR_USR(-EINVAL);
11135 slot = id_to_memslot(slots, id);
11137 if (slot && slot->npages)
11138 return ERR_PTR_USR(-EEXIST);
11141 * MAP_SHARED to prevent internal slot pages from being moved
11144 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11145 MAP_SHARED | MAP_ANONYMOUS, 0);
11146 if (IS_ERR((void *)hva))
11147 return (void __user *)hva;
11149 if (!slot || !slot->npages)
11152 old_npages = slot->npages;
11153 hva = slot->userspace_addr;
11156 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11157 struct kvm_userspace_memory_region m;
11159 m.slot = id | (i << 16);
11161 m.guest_phys_addr = gpa;
11162 m.userspace_addr = hva;
11163 m.memory_size = size;
11164 r = __kvm_set_memory_region(kvm, &m);
11166 return ERR_PTR_USR(r);
11170 vm_munmap(hva, old_npages * PAGE_SIZE);
11172 return (void __user *)hva;
11174 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11176 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11178 kvm_mmu_pre_destroy_vm(kvm);
11181 void kvm_arch_destroy_vm(struct kvm *kvm)
11183 if (current->mm == kvm->mm) {
11185 * Free memory regions allocated on behalf of userspace,
11186 * unless the the memory map has changed due to process exit
11189 mutex_lock(&kvm->slots_lock);
11190 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11192 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11194 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11195 mutex_unlock(&kvm->slots_lock);
11197 static_call_cond(kvm_x86_vm_destroy)(kvm);
11198 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11199 kvm_pic_destroy(kvm);
11200 kvm_ioapic_destroy(kvm);
11201 kvm_free_vcpus(kvm);
11202 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11203 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11204 kvm_mmu_uninit_vm(kvm);
11205 kvm_page_track_cleanup(kvm);
11206 kvm_xen_destroy_vm(kvm);
11207 kvm_hv_destroy_vm(kvm);
11210 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11214 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11215 kvfree(slot->arch.rmap[i]);
11216 slot->arch.rmap[i] = NULL;
11220 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11224 memslot_rmap_free(slot);
11226 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11227 kvfree(slot->arch.lpage_info[i - 1]);
11228 slot->arch.lpage_info[i - 1] = NULL;
11231 kvm_page_track_free_memslot(slot);
11234 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
11235 unsigned long npages)
11237 const int sz = sizeof(*slot->arch.rmap[0]);
11240 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11242 int lpages = gfn_to_index(slot->base_gfn + npages - 1,
11243 slot->base_gfn, level) + 1;
11245 WARN_ON(slot->arch.rmap[i]);
11247 slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11248 if (!slot->arch.rmap[i]) {
11249 memslot_rmap_free(slot);
11257 int alloc_all_memslots_rmaps(struct kvm *kvm)
11259 struct kvm_memslots *slots;
11260 struct kvm_memory_slot *slot;
11264 * Check if memslots alreday have rmaps early before acquiring
11265 * the slots_arch_lock below.
11267 if (kvm_memslots_have_rmaps(kvm))
11270 mutex_lock(&kvm->slots_arch_lock);
11273 * Read memslots_have_rmaps again, under the slots arch lock,
11274 * before allocating the rmaps
11276 if (kvm_memslots_have_rmaps(kvm)) {
11277 mutex_unlock(&kvm->slots_arch_lock);
11281 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11282 slots = __kvm_memslots(kvm, i);
11283 kvm_for_each_memslot(slot, slots) {
11284 r = memslot_rmap_alloc(slot, slot->npages);
11286 mutex_unlock(&kvm->slots_arch_lock);
11293 * Ensure that memslots_have_rmaps becomes true strictly after
11294 * all the rmap pointers are set.
11296 smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11297 mutex_unlock(&kvm->slots_arch_lock);
11301 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11302 struct kvm_memory_slot *slot,
11303 unsigned long npages)
11308 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
11309 * old arrays will be freed by __kvm_set_memory_region() if installing
11310 * the new memslot is successful.
11312 memset(&slot->arch, 0, sizeof(slot->arch));
11314 if (kvm_memslots_have_rmaps(kvm)) {
11315 r = memslot_rmap_alloc(slot, npages);
11320 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11321 struct kvm_lpage_info *linfo;
11322 unsigned long ugfn;
11326 lpages = gfn_to_index(slot->base_gfn + npages - 1,
11327 slot->base_gfn, level) + 1;
11329 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11333 slot->arch.lpage_info[i - 1] = linfo;
11335 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11336 linfo[0].disallow_lpage = 1;
11337 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11338 linfo[lpages - 1].disallow_lpage = 1;
11339 ugfn = slot->userspace_addr >> PAGE_SHIFT;
11341 * If the gfn and userspace address are not aligned wrt each
11342 * other, disable large page support for this slot.
11344 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11347 for (j = 0; j < lpages; ++j)
11348 linfo[j].disallow_lpage = 1;
11352 if (kvm_page_track_create_memslot(slot, npages))
11358 memslot_rmap_free(slot);
11360 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11361 kvfree(slot->arch.lpage_info[i - 1]);
11362 slot->arch.lpage_info[i - 1] = NULL;
11367 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11369 struct kvm_vcpu *vcpu;
11373 * memslots->generation has been incremented.
11374 * mmio generation may have reached its maximum value.
11376 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11378 /* Force re-initialization of steal_time cache */
11379 kvm_for_each_vcpu(i, vcpu, kvm)
11380 kvm_vcpu_kick(vcpu);
11383 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11384 struct kvm_memory_slot *memslot,
11385 const struct kvm_userspace_memory_region *mem,
11386 enum kvm_mr_change change)
11388 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11389 return kvm_alloc_memslot_metadata(kvm, memslot,
11390 mem->memory_size >> PAGE_SHIFT);
11395 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11397 struct kvm_arch *ka = &kvm->arch;
11399 if (!kvm_x86_ops.cpu_dirty_log_size)
11402 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11403 (!enable && --ka->cpu_dirty_logging_count == 0))
11404 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11406 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11409 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11410 struct kvm_memory_slot *old,
11411 struct kvm_memory_slot *new,
11412 enum kvm_mr_change change)
11414 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11417 * Update CPU dirty logging if dirty logging is being toggled. This
11418 * applies to all operations.
11420 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11421 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11424 * Nothing more to do for RO slots (which can't be dirtied and can't be
11425 * made writable) or CREATE/MOVE/DELETE of a slot.
11427 * For a memslot with dirty logging disabled:
11428 * CREATE: No dirty mappings will already exist.
11429 * MOVE/DELETE: The old mappings will already have been cleaned up by
11430 * kvm_arch_flush_shadow_memslot()
11432 * For a memslot with dirty logging enabled:
11433 * CREATE: No shadow pages exist, thus nothing to write-protect
11434 * and no dirty bits to clear.
11435 * MOVE/DELETE: The old mappings will already have been cleaned up by
11436 * kvm_arch_flush_shadow_memslot().
11438 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11442 * READONLY and non-flags changes were filtered out above, and the only
11443 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11444 * logging isn't being toggled on or off.
11446 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11449 if (!log_dirty_pages) {
11451 * Dirty logging tracks sptes in 4k granularity, meaning that
11452 * large sptes have to be split. If live migration succeeds,
11453 * the guest in the source machine will be destroyed and large
11454 * sptes will be created in the destination. However, if the
11455 * guest continues to run in the source machine (for example if
11456 * live migration fails), small sptes will remain around and
11457 * cause bad performance.
11459 * Scan sptes if dirty logging has been stopped, dropping those
11460 * which can be collapsed into a single large-page spte. Later
11461 * page faults will create the large-page sptes.
11463 kvm_mmu_zap_collapsible_sptes(kvm, new);
11466 * Initially-all-set does not require write protecting any page,
11467 * because they're all assumed to be dirty.
11469 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11472 if (kvm_x86_ops.cpu_dirty_log_size) {
11473 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11474 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11476 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11481 void kvm_arch_commit_memory_region(struct kvm *kvm,
11482 const struct kvm_userspace_memory_region *mem,
11483 struct kvm_memory_slot *old,
11484 const struct kvm_memory_slot *new,
11485 enum kvm_mr_change change)
11487 if (!kvm->arch.n_requested_mmu_pages)
11488 kvm_mmu_change_mmu_pages(kvm,
11489 kvm_mmu_calculate_default_mmu_pages(kvm));
11492 * FIXME: const-ify all uses of struct kvm_memory_slot.
11494 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11496 /* Free the arrays associated with the old memslot. */
11497 if (change == KVM_MR_MOVE)
11498 kvm_arch_free_memslot(kvm, old);
11501 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11503 kvm_mmu_zap_all(kvm);
11506 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11507 struct kvm_memory_slot *slot)
11509 kvm_page_track_flush_slot(kvm, slot);
11512 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11514 return (is_guest_mode(vcpu) &&
11515 kvm_x86_ops.guest_apic_has_interrupt &&
11516 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11519 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11521 if (!list_empty_careful(&vcpu->async_pf.done))
11524 if (kvm_apic_has_events(vcpu))
11527 if (vcpu->arch.pv.pv_unhalted)
11530 if (vcpu->arch.exception.pending)
11533 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11534 (vcpu->arch.nmi_pending &&
11535 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11538 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11539 (vcpu->arch.smi_pending &&
11540 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11543 if (kvm_arch_interrupt_allowed(vcpu) &&
11544 (kvm_cpu_has_interrupt(vcpu) ||
11545 kvm_guest_apic_has_interrupt(vcpu)))
11548 if (kvm_hv_has_stimer_pending(vcpu))
11551 if (is_guest_mode(vcpu) &&
11552 kvm_x86_ops.nested_ops->hv_timer_pending &&
11553 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11559 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11561 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11564 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11566 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11572 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11574 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11577 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11578 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11579 kvm_test_request(KVM_REQ_EVENT, vcpu))
11582 return kvm_arch_dy_has_pending_interrupt(vcpu);
11585 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11587 if (vcpu->arch.guest_state_protected)
11590 return vcpu->arch.preempted_in_kernel;
11593 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11595 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11598 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11600 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11603 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11605 /* Can't read the RIP when guest state is protected, just return 0 */
11606 if (vcpu->arch.guest_state_protected)
11609 if (is_64_bit_mode(vcpu))
11610 return kvm_rip_read(vcpu);
11611 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11612 kvm_rip_read(vcpu));
11614 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11616 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11618 return kvm_get_linear_rip(vcpu) == linear_rip;
11620 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11622 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11624 unsigned long rflags;
11626 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11627 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11628 rflags &= ~X86_EFLAGS_TF;
11631 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11633 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11635 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11636 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11637 rflags |= X86_EFLAGS_TF;
11638 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11641 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11643 __kvm_set_rflags(vcpu, rflags);
11644 kvm_make_request(KVM_REQ_EVENT, vcpu);
11646 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11648 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11652 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11656 r = kvm_mmu_reload(vcpu);
11660 if (!vcpu->arch.mmu->direct_map &&
11661 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11664 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11667 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11669 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11671 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11674 static inline u32 kvm_async_pf_next_probe(u32 key)
11676 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11679 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11681 u32 key = kvm_async_pf_hash_fn(gfn);
11683 while (vcpu->arch.apf.gfns[key] != ~0)
11684 key = kvm_async_pf_next_probe(key);
11686 vcpu->arch.apf.gfns[key] = gfn;
11689 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11692 u32 key = kvm_async_pf_hash_fn(gfn);
11694 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11695 (vcpu->arch.apf.gfns[key] != gfn &&
11696 vcpu->arch.apf.gfns[key] != ~0); i++)
11697 key = kvm_async_pf_next_probe(key);
11702 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11704 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11707 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11711 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11713 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11717 vcpu->arch.apf.gfns[i] = ~0;
11719 j = kvm_async_pf_next_probe(j);
11720 if (vcpu->arch.apf.gfns[j] == ~0)
11722 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11724 * k lies cyclically in ]i,j]
11726 * |....j i.k.| or |.k..j i...|
11728 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11729 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11734 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11736 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11738 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11742 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11744 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11746 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11747 &token, offset, sizeof(token));
11750 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11752 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11755 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11756 &val, offset, sizeof(val)))
11762 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11764 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11767 if (!kvm_pv_async_pf_enabled(vcpu) ||
11768 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11774 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11776 if (unlikely(!lapic_in_kernel(vcpu) ||
11777 kvm_event_needs_reinjection(vcpu) ||
11778 vcpu->arch.exception.pending))
11781 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11785 * If interrupts are off we cannot even use an artificial
11788 return kvm_arch_interrupt_allowed(vcpu);
11791 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11792 struct kvm_async_pf *work)
11794 struct x86_exception fault;
11796 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11797 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11799 if (kvm_can_deliver_async_pf(vcpu) &&
11800 !apf_put_user_notpresent(vcpu)) {
11801 fault.vector = PF_VECTOR;
11802 fault.error_code_valid = true;
11803 fault.error_code = 0;
11804 fault.nested_page_fault = false;
11805 fault.address = work->arch.token;
11806 fault.async_page_fault = true;
11807 kvm_inject_page_fault(vcpu, &fault);
11811 * It is not possible to deliver a paravirtualized asynchronous
11812 * page fault, but putting the guest in an artificial halt state
11813 * can be beneficial nevertheless: if an interrupt arrives, we
11814 * can deliver it timely and perhaps the guest will schedule
11815 * another process. When the instruction that triggered a page
11816 * fault is retried, hopefully the page will be ready in the host.
11818 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11823 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11824 struct kvm_async_pf *work)
11826 struct kvm_lapic_irq irq = {
11827 .delivery_mode = APIC_DM_FIXED,
11828 .vector = vcpu->arch.apf.vec
11831 if (work->wakeup_all)
11832 work->arch.token = ~0; /* broadcast wakeup */
11834 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11835 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11837 if ((work->wakeup_all || work->notpresent_injected) &&
11838 kvm_pv_async_pf_enabled(vcpu) &&
11839 !apf_put_user_ready(vcpu, work->arch.token)) {
11840 vcpu->arch.apf.pageready_pending = true;
11841 kvm_apic_set_irq(vcpu, &irq, NULL);
11844 vcpu->arch.apf.halted = false;
11845 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11848 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11850 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11851 if (!vcpu->arch.apf.pageready_pending)
11852 kvm_vcpu_kick(vcpu);
11855 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11857 if (!kvm_pv_async_pf_enabled(vcpu))
11860 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11863 void kvm_arch_start_assignment(struct kvm *kvm)
11865 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
11866 static_call_cond(kvm_x86_start_assignment)(kvm);
11868 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11870 void kvm_arch_end_assignment(struct kvm *kvm)
11872 atomic_dec(&kvm->arch.assigned_device_count);
11874 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11876 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11878 return atomic_read(&kvm->arch.assigned_device_count);
11880 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11882 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11884 atomic_inc(&kvm->arch.noncoherent_dma_count);
11886 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11888 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11890 atomic_dec(&kvm->arch.noncoherent_dma_count);
11892 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11894 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11896 return atomic_read(&kvm->arch.noncoherent_dma_count);
11898 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11900 bool kvm_arch_has_irq_bypass(void)
11905 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11906 struct irq_bypass_producer *prod)
11908 struct kvm_kernel_irqfd *irqfd =
11909 container_of(cons, struct kvm_kernel_irqfd, consumer);
11912 irqfd->producer = prod;
11913 kvm_arch_start_assignment(irqfd->kvm);
11914 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11915 prod->irq, irqfd->gsi, 1);
11918 kvm_arch_end_assignment(irqfd->kvm);
11923 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11924 struct irq_bypass_producer *prod)
11927 struct kvm_kernel_irqfd *irqfd =
11928 container_of(cons, struct kvm_kernel_irqfd, consumer);
11930 WARN_ON(irqfd->producer != prod);
11931 irqfd->producer = NULL;
11934 * When producer of consumer is unregistered, we change back to
11935 * remapped mode, so we can re-use the current implementation
11936 * when the irq is masked/disabled or the consumer side (KVM
11937 * int this case doesn't want to receive the interrupts.
11939 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11941 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11942 " fails: %d\n", irqfd->consumer.token, ret);
11944 kvm_arch_end_assignment(irqfd->kvm);
11947 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11948 uint32_t guest_irq, bool set)
11950 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11953 bool kvm_vector_hashing_enabled(void)
11955 return vector_hashing;
11958 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11960 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11962 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11965 int kvm_spec_ctrl_test_value(u64 value)
11968 * test that setting IA32_SPEC_CTRL to given value
11969 * is allowed by the host processor
11973 unsigned long flags;
11976 local_irq_save(flags);
11978 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11980 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11983 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11985 local_irq_restore(flags);
11989 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11991 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11993 struct x86_exception fault;
11994 u32 access = error_code &
11995 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11997 if (!(error_code & PFERR_PRESENT_MASK) ||
11998 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12000 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12001 * tables probably do not match the TLB. Just proceed
12002 * with the error code that the processor gave.
12004 fault.vector = PF_VECTOR;
12005 fault.error_code_valid = true;
12006 fault.error_code = error_code;
12007 fault.nested_page_fault = false;
12008 fault.address = gva;
12010 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12012 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12015 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12016 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12017 * indicates whether exit to userspace is needed.
12019 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12020 struct x86_exception *e)
12022 if (r == X86EMUL_PROPAGATE_FAULT) {
12023 kvm_inject_emulated_page_fault(vcpu, e);
12028 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12029 * while handling a VMX instruction KVM could've handled the request
12030 * correctly by exiting to userspace and performing I/O but there
12031 * doesn't seem to be a real use-case behind such requests, just return
12032 * KVM_EXIT_INTERNAL_ERROR for now.
12034 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
12035 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
12036 vcpu->run->internal.ndata = 0;
12040 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12042 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12045 struct x86_exception e;
12052 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12053 if (r != X86EMUL_CONTINUE)
12054 return kvm_handle_memory_failure(vcpu, r, &e);
12056 if (operand.pcid >> 12 != 0) {
12057 kvm_inject_gp(vcpu, 0);
12061 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12064 case INVPCID_TYPE_INDIV_ADDR:
12065 if ((!pcid_enabled && (operand.pcid != 0)) ||
12066 is_noncanonical_address(operand.gla, vcpu)) {
12067 kvm_inject_gp(vcpu, 0);
12070 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12071 return kvm_skip_emulated_instruction(vcpu);
12073 case INVPCID_TYPE_SINGLE_CTXT:
12074 if (!pcid_enabled && (operand.pcid != 0)) {
12075 kvm_inject_gp(vcpu, 0);
12079 kvm_invalidate_pcid(vcpu, operand.pcid);
12080 return kvm_skip_emulated_instruction(vcpu);
12082 case INVPCID_TYPE_ALL_NON_GLOBAL:
12084 * Currently, KVM doesn't mark global entries in the shadow
12085 * page tables, so a non-global flush just degenerates to a
12086 * global flush. If needed, we could optimize this later by
12087 * keeping track of global entries in shadow page tables.
12091 case INVPCID_TYPE_ALL_INCL_GLOBAL:
12092 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12093 return kvm_skip_emulated_instruction(vcpu);
12096 BUG(); /* We have already checked above that type <= 3 */
12099 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12101 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12103 struct kvm_run *run = vcpu->run;
12104 struct kvm_mmio_fragment *frag;
12107 BUG_ON(!vcpu->mmio_needed);
12109 /* Complete previous fragment */
12110 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12111 len = min(8u, frag->len);
12112 if (!vcpu->mmio_is_write)
12113 memcpy(frag->data, run->mmio.data, len);
12115 if (frag->len <= 8) {
12116 /* Switch to the next fragment. */
12118 vcpu->mmio_cur_fragment++;
12120 /* Go forward to the next mmio piece. */
12126 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12127 vcpu->mmio_needed = 0;
12129 // VMG change, at this point, we're always done
12130 // RIP has already been advanced
12134 // More MMIO is needed
12135 run->mmio.phys_addr = frag->gpa;
12136 run->mmio.len = min(8u, frag->len);
12137 run->mmio.is_write = vcpu->mmio_is_write;
12138 if (run->mmio.is_write)
12139 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12140 run->exit_reason = KVM_EXIT_MMIO;
12142 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12147 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12151 struct kvm_mmio_fragment *frag;
12156 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12157 if (handled == bytes)
12164 /*TODO: Check if need to increment number of frags */
12165 frag = vcpu->mmio_fragments;
12166 vcpu->mmio_nr_fragments = 1;
12171 vcpu->mmio_needed = 1;
12172 vcpu->mmio_cur_fragment = 0;
12174 vcpu->run->mmio.phys_addr = gpa;
12175 vcpu->run->mmio.len = min(8u, frag->len);
12176 vcpu->run->mmio.is_write = 1;
12177 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12178 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12180 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12184 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12186 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12190 struct kvm_mmio_fragment *frag;
12195 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12196 if (handled == bytes)
12203 /*TODO: Check if need to increment number of frags */
12204 frag = vcpu->mmio_fragments;
12205 vcpu->mmio_nr_fragments = 1;
12210 vcpu->mmio_needed = 1;
12211 vcpu->mmio_cur_fragment = 0;
12213 vcpu->run->mmio.phys_addr = gpa;
12214 vcpu->run->mmio.len = min(8u, frag->len);
12215 vcpu->run->mmio.is_write = 0;
12216 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12218 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12222 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12224 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12226 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
12227 vcpu->arch.pio.count * vcpu->arch.pio.size);
12228 vcpu->arch.pio.count = 0;
12233 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12234 unsigned int port, void *data, unsigned int count)
12238 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
12243 vcpu->arch.pio.count = 0;
12248 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12249 unsigned int port, void *data, unsigned int count)
12253 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
12256 vcpu->arch.pio.count = 0;
12258 vcpu->arch.guest_ins_data = data;
12259 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12265 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12266 unsigned int port, void *data, unsigned int count,
12269 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
12270 : kvm_sev_es_outs(vcpu, size, port, data, count);
12272 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12274 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12275 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12276 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12277 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12278 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12279 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12280 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12281 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12282 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12283 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12284 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12285 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12286 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12287 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12288 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12289 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12290 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12291 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12292 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12293 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12294 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12295 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12296 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12297 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12298 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12299 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12300 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);