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>
62 #include <trace/events/kvm.h>
64 #include <asm/debugreg.h>
68 #include <linux/kernel_stat.h>
69 #include <asm/fpu/internal.h> /* Ugh! */
70 #include <asm/pvclock.h>
71 #include <asm/div64.h>
72 #include <asm/irq_remapping.h>
73 #include <asm/mshyperv.h>
74 #include <asm/hypervisor.h>
75 #include <asm/tlbflush.h>
76 #include <asm/intel_pt.h>
77 #include <asm/emulate_prefix.h>
79 #include <clocksource/hyperv_timer.h>
81 #define CREATE_TRACE_POINTS
84 #define MAX_IO_MSRS 256
85 #define KVM_MAX_MCE_BANKS 32
86 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
87 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
89 #define emul_to_vcpu(ctxt) \
90 ((struct kvm_vcpu *)(ctxt)->vcpu)
93 * - enable syscall per default because its emulated by KVM
94 * - enable LME and LMA per default on 64 bit KVM
98 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
100 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
103 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
105 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
106 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
108 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
109 static void process_nmi(struct kvm_vcpu *vcpu);
110 static void process_smi(struct kvm_vcpu *vcpu);
111 static void enter_smm(struct kvm_vcpu *vcpu);
112 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
113 static void store_regs(struct kvm_vcpu *vcpu);
114 static int sync_regs(struct kvm_vcpu *vcpu);
116 struct kvm_x86_ops kvm_x86_ops __read_mostly;
117 EXPORT_SYMBOL_GPL(kvm_x86_ops);
119 #define KVM_X86_OP(func) \
120 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
121 *(((struct kvm_x86_ops *)0)->func));
122 #define KVM_X86_OP_NULL KVM_X86_OP
123 #include <asm/kvm-x86-ops.h>
124 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
125 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
126 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
128 static bool __read_mostly ignore_msrs = 0;
129 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
131 bool __read_mostly report_ignored_msrs = true;
132 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
133 EXPORT_SYMBOL_GPL(report_ignored_msrs);
135 unsigned int min_timer_period_us = 200;
136 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
138 static bool __read_mostly kvmclock_periodic_sync = true;
139 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
141 bool __read_mostly kvm_has_tsc_control;
142 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
143 u32 __read_mostly kvm_max_guest_tsc_khz;
144 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
145 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
146 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
147 u64 __read_mostly kvm_max_tsc_scaling_ratio;
148 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
149 u64 __read_mostly kvm_default_tsc_scaling_ratio;
150 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
151 bool __read_mostly kvm_has_bus_lock_exit;
152 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
154 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
155 static u32 __read_mostly tsc_tolerance_ppm = 250;
156 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
159 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
160 * adaptive tuning starting from default advancement of 1000ns. '0' disables
161 * advancement entirely. Any other value is used as-is and disables adaptive
162 * tuning, i.e. allows privileged userspace to set an exact advancement time.
164 static int __read_mostly lapic_timer_advance_ns = -1;
165 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
167 static bool __read_mostly vector_hashing = true;
168 module_param(vector_hashing, bool, S_IRUGO);
170 bool __read_mostly enable_vmware_backdoor = false;
171 module_param(enable_vmware_backdoor, bool, S_IRUGO);
172 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
174 static bool __read_mostly force_emulation_prefix = false;
175 module_param(force_emulation_prefix, bool, S_IRUGO);
177 int __read_mostly pi_inject_timer = -1;
178 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
181 * Restoring the host value for MSRs that are only consumed when running in
182 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
183 * returns to userspace, i.e. the kernel can run with the guest's value.
185 #define KVM_MAX_NR_USER_RETURN_MSRS 16
187 struct kvm_user_return_msrs_global {
189 u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS];
192 struct kvm_user_return_msrs {
193 struct user_return_notifier urn;
195 struct kvm_user_return_msr_values {
198 } values[KVM_MAX_NR_USER_RETURN_MSRS];
201 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global;
202 static struct kvm_user_return_msrs __percpu *user_return_msrs;
204 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
205 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
206 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
207 | XFEATURE_MASK_PKRU)
209 u64 __read_mostly host_efer;
210 EXPORT_SYMBOL_GPL(host_efer);
212 bool __read_mostly allow_smaller_maxphyaddr = 0;
213 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
215 u64 __read_mostly host_xss;
216 EXPORT_SYMBOL_GPL(host_xss);
217 u64 __read_mostly supported_xss;
218 EXPORT_SYMBOL_GPL(supported_xss);
220 struct kvm_stats_debugfs_item debugfs_entries[] = {
221 VCPU_STAT("pf_fixed", pf_fixed),
222 VCPU_STAT("pf_guest", pf_guest),
223 VCPU_STAT("tlb_flush", tlb_flush),
224 VCPU_STAT("invlpg", invlpg),
225 VCPU_STAT("exits", exits),
226 VCPU_STAT("io_exits", io_exits),
227 VCPU_STAT("mmio_exits", mmio_exits),
228 VCPU_STAT("signal_exits", signal_exits),
229 VCPU_STAT("irq_window", irq_window_exits),
230 VCPU_STAT("nmi_window", nmi_window_exits),
231 VCPU_STAT("halt_exits", halt_exits),
232 VCPU_STAT("halt_successful_poll", halt_successful_poll),
233 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
234 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
235 VCPU_STAT("halt_wakeup", halt_wakeup),
236 VCPU_STAT("hypercalls", hypercalls),
237 VCPU_STAT("request_irq", request_irq_exits),
238 VCPU_STAT("irq_exits", irq_exits),
239 VCPU_STAT("host_state_reload", host_state_reload),
240 VCPU_STAT("fpu_reload", fpu_reload),
241 VCPU_STAT("insn_emulation", insn_emulation),
242 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
243 VCPU_STAT("irq_injections", irq_injections),
244 VCPU_STAT("nmi_injections", nmi_injections),
245 VCPU_STAT("req_event", req_event),
246 VCPU_STAT("l1d_flush", l1d_flush),
247 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
248 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
249 VCPU_STAT("nested_run", nested_run),
250 VCPU_STAT("directed_yield_attempted", directed_yield_attempted),
251 VCPU_STAT("directed_yield_successful", directed_yield_successful),
252 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
253 VM_STAT("mmu_pte_write", mmu_pte_write),
254 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
255 VM_STAT("mmu_flooded", mmu_flooded),
256 VM_STAT("mmu_recycled", mmu_recycled),
257 VM_STAT("mmu_cache_miss", mmu_cache_miss),
258 VM_STAT("mmu_unsync", mmu_unsync),
259 VM_STAT("remote_tlb_flush", remote_tlb_flush),
260 VM_STAT("largepages", lpages, .mode = 0444),
261 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
262 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
266 u64 __read_mostly host_xcr0;
267 u64 __read_mostly supported_xcr0;
268 EXPORT_SYMBOL_GPL(supported_xcr0);
270 static struct kmem_cache *x86_fpu_cache;
272 static struct kmem_cache *x86_emulator_cache;
275 * When called, it means the previous get/set msr reached an invalid msr.
276 * Return true if we want to ignore/silent this failed msr access.
278 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
280 const char *op = write ? "wrmsr" : "rdmsr";
283 if (report_ignored_msrs)
284 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
289 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
295 static struct kmem_cache *kvm_alloc_emulator_cache(void)
297 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
298 unsigned int size = sizeof(struct x86_emulate_ctxt);
300 return kmem_cache_create_usercopy("x86_emulator", size,
301 __alignof__(struct x86_emulate_ctxt),
302 SLAB_ACCOUNT, useroffset,
303 size - useroffset, NULL);
306 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
308 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
311 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
312 vcpu->arch.apf.gfns[i] = ~0;
315 static void kvm_on_user_return(struct user_return_notifier *urn)
318 struct kvm_user_return_msrs *msrs
319 = container_of(urn, struct kvm_user_return_msrs, urn);
320 struct kvm_user_return_msr_values *values;
324 * Disabling irqs at this point since the following code could be
325 * interrupted and executed through kvm_arch_hardware_disable()
327 local_irq_save(flags);
328 if (msrs->registered) {
329 msrs->registered = false;
330 user_return_notifier_unregister(urn);
332 local_irq_restore(flags);
333 for (slot = 0; slot < user_return_msrs_global.nr; ++slot) {
334 values = &msrs->values[slot];
335 if (values->host != values->curr) {
336 wrmsrl(user_return_msrs_global.msrs[slot], values->host);
337 values->curr = values->host;
342 int kvm_probe_user_return_msr(u32 msr)
348 ret = rdmsrl_safe(msr, &val);
351 ret = wrmsrl_safe(msr, val);
356 EXPORT_SYMBOL_GPL(kvm_probe_user_return_msr);
358 void kvm_define_user_return_msr(unsigned slot, u32 msr)
360 BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
361 user_return_msrs_global.msrs[slot] = msr;
362 if (slot >= user_return_msrs_global.nr)
363 user_return_msrs_global.nr = slot + 1;
365 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
367 static void kvm_user_return_msr_cpu_online(void)
369 unsigned int cpu = smp_processor_id();
370 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
374 for (i = 0; i < user_return_msrs_global.nr; ++i) {
375 rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
376 msrs->values[i].host = value;
377 msrs->values[i].curr = value;
381 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
383 unsigned int cpu = smp_processor_id();
384 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
387 value = (value & mask) | (msrs->values[slot].host & ~mask);
388 if (value == msrs->values[slot].curr)
390 err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
394 msrs->values[slot].curr = value;
395 if (!msrs->registered) {
396 msrs->urn.on_user_return = kvm_on_user_return;
397 user_return_notifier_register(&msrs->urn);
398 msrs->registered = true;
402 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
404 static void drop_user_return_notifiers(void)
406 unsigned int cpu = smp_processor_id();
407 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
409 if (msrs->registered)
410 kvm_on_user_return(&msrs->urn);
413 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
415 return vcpu->arch.apic_base;
417 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
419 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
421 return kvm_apic_mode(kvm_get_apic_base(vcpu));
423 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
425 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
427 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
428 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
429 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
430 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
432 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
434 if (!msr_info->host_initiated) {
435 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
437 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
441 kvm_lapic_set_base(vcpu, msr_info->data);
442 kvm_recalculate_apic_map(vcpu->kvm);
445 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
447 asmlinkage __visible noinstr void kvm_spurious_fault(void)
449 /* Fault while not rebooting. We want the trace. */
450 BUG_ON(!kvm_rebooting);
452 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
454 #define EXCPT_BENIGN 0
455 #define EXCPT_CONTRIBUTORY 1
458 static int exception_class(int vector)
468 return EXCPT_CONTRIBUTORY;
475 #define EXCPT_FAULT 0
477 #define EXCPT_ABORT 2
478 #define EXCPT_INTERRUPT 3
480 static int exception_type(int vector)
484 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
485 return EXCPT_INTERRUPT;
489 /* #DB is trap, as instruction watchpoints are handled elsewhere */
490 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
493 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
496 /* Reserved exceptions will result in fault */
500 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
502 unsigned nr = vcpu->arch.exception.nr;
503 bool has_payload = vcpu->arch.exception.has_payload;
504 unsigned long payload = vcpu->arch.exception.payload;
512 * "Certain debug exceptions may clear bit 0-3. The
513 * remaining contents of the DR6 register are never
514 * cleared by the processor".
516 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
518 * In order to reflect the #DB exception payload in guest
519 * dr6, three components need to be considered: active low
520 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
522 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
523 * In the target guest dr6:
524 * FIXED_1 bits should always be set.
525 * Active low bits should be cleared if 1-setting in payload.
526 * Active high bits should be set if 1-setting in payload.
528 * Note, the payload is compatible with the pending debug
529 * exceptions/exit qualification under VMX, that active_low bits
530 * are active high in payload.
531 * So they need to be flipped for DR6.
533 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
534 vcpu->arch.dr6 |= payload;
535 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
538 * The #DB payload is defined as compatible with the 'pending
539 * debug exceptions' field under VMX, not DR6. While bit 12 is
540 * defined in the 'pending debug exceptions' field (enabled
541 * breakpoint), it is reserved and must be zero in DR6.
543 vcpu->arch.dr6 &= ~BIT(12);
546 vcpu->arch.cr2 = payload;
550 vcpu->arch.exception.has_payload = false;
551 vcpu->arch.exception.payload = 0;
553 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
555 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
556 unsigned nr, bool has_error, u32 error_code,
557 bool has_payload, unsigned long payload, bool reinject)
562 kvm_make_request(KVM_REQ_EVENT, vcpu);
564 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
568 * On vmentry, vcpu->arch.exception.pending is only
569 * true if an event injection was blocked by
570 * nested_run_pending. In that case, however,
571 * vcpu_enter_guest requests an immediate exit,
572 * and the guest shouldn't proceed far enough to
575 WARN_ON_ONCE(vcpu->arch.exception.pending);
576 vcpu->arch.exception.injected = true;
577 if (WARN_ON_ONCE(has_payload)) {
579 * A reinjected event has already
580 * delivered its payload.
586 vcpu->arch.exception.pending = true;
587 vcpu->arch.exception.injected = false;
589 vcpu->arch.exception.has_error_code = has_error;
590 vcpu->arch.exception.nr = nr;
591 vcpu->arch.exception.error_code = error_code;
592 vcpu->arch.exception.has_payload = has_payload;
593 vcpu->arch.exception.payload = payload;
594 if (!is_guest_mode(vcpu))
595 kvm_deliver_exception_payload(vcpu);
599 /* to check exception */
600 prev_nr = vcpu->arch.exception.nr;
601 if (prev_nr == DF_VECTOR) {
602 /* triple fault -> shutdown */
603 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
606 class1 = exception_class(prev_nr);
607 class2 = exception_class(nr);
608 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
609 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
611 * Generate double fault per SDM Table 5-5. Set
612 * exception.pending = true so that the double fault
613 * can trigger a nested vmexit.
615 vcpu->arch.exception.pending = true;
616 vcpu->arch.exception.injected = false;
617 vcpu->arch.exception.has_error_code = true;
618 vcpu->arch.exception.nr = DF_VECTOR;
619 vcpu->arch.exception.error_code = 0;
620 vcpu->arch.exception.has_payload = false;
621 vcpu->arch.exception.payload = 0;
623 /* replace previous exception with a new one in a hope
624 that instruction re-execution will regenerate lost
629 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
631 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
633 EXPORT_SYMBOL_GPL(kvm_queue_exception);
635 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
637 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
639 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
641 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
642 unsigned long payload)
644 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
646 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
648 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
649 u32 error_code, unsigned long payload)
651 kvm_multiple_exception(vcpu, nr, true, error_code,
652 true, payload, false);
655 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
658 kvm_inject_gp(vcpu, 0);
660 return kvm_skip_emulated_instruction(vcpu);
664 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
666 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
668 ++vcpu->stat.pf_guest;
669 vcpu->arch.exception.nested_apf =
670 is_guest_mode(vcpu) && fault->async_page_fault;
671 if (vcpu->arch.exception.nested_apf) {
672 vcpu->arch.apf.nested_apf_token = fault->address;
673 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
675 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
679 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
681 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
682 struct x86_exception *fault)
684 struct kvm_mmu *fault_mmu;
685 WARN_ON_ONCE(fault->vector != PF_VECTOR);
687 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
691 * Invalidate the TLB entry for the faulting address, if it exists,
692 * else the access will fault indefinitely (and to emulate hardware).
694 if ((fault->error_code & PFERR_PRESENT_MASK) &&
695 !(fault->error_code & PFERR_RSVD_MASK))
696 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
697 fault_mmu->root_hpa);
699 fault_mmu->inject_page_fault(vcpu, fault);
700 return fault->nested_page_fault;
702 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
704 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
706 atomic_inc(&vcpu->arch.nmi_queued);
707 kvm_make_request(KVM_REQ_NMI, vcpu);
709 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
711 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
713 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
715 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
717 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
719 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
721 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
724 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
725 * a #GP and return false.
727 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
729 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
731 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
734 EXPORT_SYMBOL_GPL(kvm_require_cpl);
736 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
738 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
741 kvm_queue_exception(vcpu, UD_VECTOR);
744 EXPORT_SYMBOL_GPL(kvm_require_dr);
747 * This function will be used to read from the physical memory of the currently
748 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
749 * can read from guest physical or from the guest's guest physical memory.
751 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
752 gfn_t ngfn, void *data, int offset, int len,
755 struct x86_exception exception;
759 ngpa = gfn_to_gpa(ngfn);
760 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
761 if (real_gfn == UNMAPPED_GVA)
764 real_gfn = gpa_to_gfn(real_gfn);
766 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
768 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
770 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
771 void *data, int offset, int len, u32 access)
773 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
774 data, offset, len, access);
777 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
779 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
783 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
785 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
787 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
788 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
791 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
793 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
794 offset * sizeof(u64), sizeof(pdpte),
795 PFERR_USER_MASK|PFERR_WRITE_MASK);
800 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
801 if ((pdpte[i] & PT_PRESENT_MASK) &&
802 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
809 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
810 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
816 EXPORT_SYMBOL_GPL(load_pdptrs);
818 bool pdptrs_changed(struct kvm_vcpu *vcpu)
820 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
825 if (!is_pae_paging(vcpu))
828 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
831 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
832 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
833 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
834 PFERR_USER_MASK | PFERR_WRITE_MASK);
838 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
840 EXPORT_SYMBOL_GPL(pdptrs_changed);
842 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
844 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
846 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
847 kvm_clear_async_pf_completion_queue(vcpu);
848 kvm_async_pf_hash_reset(vcpu);
851 if ((cr0 ^ old_cr0) & update_bits)
852 kvm_mmu_reset_context(vcpu);
854 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
855 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
856 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
857 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
859 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
861 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
863 unsigned long old_cr0 = kvm_read_cr0(vcpu);
864 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
869 if (cr0 & 0xffffffff00000000UL)
873 cr0 &= ~CR0_RESERVED_BITS;
875 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
878 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
882 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
883 (cr0 & X86_CR0_PG)) {
888 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
893 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
894 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
895 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
898 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
901 static_call(kvm_x86_set_cr0)(vcpu, cr0);
903 kvm_post_set_cr0(vcpu, old_cr0, cr0);
907 EXPORT_SYMBOL_GPL(kvm_set_cr0);
909 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
911 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
913 EXPORT_SYMBOL_GPL(kvm_lmsw);
915 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
917 if (vcpu->arch.guest_state_protected)
920 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
922 if (vcpu->arch.xcr0 != host_xcr0)
923 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
925 if (vcpu->arch.xsaves_enabled &&
926 vcpu->arch.ia32_xss != host_xss)
927 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
930 if (static_cpu_has(X86_FEATURE_PKU) &&
931 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
932 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
933 vcpu->arch.pkru != vcpu->arch.host_pkru)
934 __write_pkru(vcpu->arch.pkru);
936 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
938 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
940 if (vcpu->arch.guest_state_protected)
943 if (static_cpu_has(X86_FEATURE_PKU) &&
944 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
945 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
946 vcpu->arch.pkru = rdpkru();
947 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
948 __write_pkru(vcpu->arch.host_pkru);
951 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
953 if (vcpu->arch.xcr0 != host_xcr0)
954 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
956 if (vcpu->arch.xsaves_enabled &&
957 vcpu->arch.ia32_xss != host_xss)
958 wrmsrl(MSR_IA32_XSS, host_xss);
962 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
964 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
967 u64 old_xcr0 = vcpu->arch.xcr0;
970 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
971 if (index != XCR_XFEATURE_ENABLED_MASK)
973 if (!(xcr0 & XFEATURE_MASK_FP))
975 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
979 * Do not allow the guest to set bits that we do not support
980 * saving. However, xcr0 bit 0 is always set, even if the
981 * emulated CPU does not support XSAVE (see fx_init).
983 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
984 if (xcr0 & ~valid_bits)
987 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
988 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
991 if (xcr0 & XFEATURE_MASK_AVX512) {
992 if (!(xcr0 & XFEATURE_MASK_YMM))
994 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
997 vcpu->arch.xcr0 = xcr0;
999 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1000 kvm_update_cpuid_runtime(vcpu);
1004 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1006 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1007 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1008 kvm_inject_gp(vcpu, 0);
1012 return kvm_skip_emulated_instruction(vcpu);
1014 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1016 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1018 if (cr4 & cr4_reserved_bits)
1021 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1024 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1026 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1028 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1030 unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1031 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
1033 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1034 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1035 kvm_mmu_reset_context(vcpu);
1037 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1039 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1041 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1042 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1045 if (!kvm_is_valid_cr4(vcpu, cr4))
1048 if (is_long_mode(vcpu)) {
1049 if (!(cr4 & X86_CR4_PAE))
1051 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1053 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1054 && ((cr4 ^ old_cr4) & pdptr_bits)
1055 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1056 kvm_read_cr3(vcpu)))
1059 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1060 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1063 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1064 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1068 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1070 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1074 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1076 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1078 bool skip_tlb_flush = false;
1079 #ifdef CONFIG_X86_64
1080 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1083 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1084 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1088 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1089 if (!skip_tlb_flush) {
1090 kvm_mmu_sync_roots(vcpu);
1091 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1097 * Do not condition the GPA check on long mode, this helper is used to
1098 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1099 * the current vCPU mode is accurate.
1101 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1104 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1107 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1108 vcpu->arch.cr3 = cr3;
1109 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1113 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1115 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1117 if (cr8 & CR8_RESERVED_BITS)
1119 if (lapic_in_kernel(vcpu))
1120 kvm_lapic_set_tpr(vcpu, cr8);
1122 vcpu->arch.cr8 = cr8;
1125 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1127 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1129 if (lapic_in_kernel(vcpu))
1130 return kvm_lapic_get_cr8(vcpu);
1132 return vcpu->arch.cr8;
1134 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1136 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1140 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1141 for (i = 0; i < KVM_NR_DB_REGS; i++)
1142 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1143 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1147 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1151 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1152 dr7 = vcpu->arch.guest_debug_dr7;
1154 dr7 = vcpu->arch.dr7;
1155 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1156 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1157 if (dr7 & DR7_BP_EN_MASK)
1158 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1160 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1162 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1164 u64 fixed = DR6_FIXED_1;
1166 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1171 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1173 size_t size = ARRAY_SIZE(vcpu->arch.db);
1177 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1178 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1179 vcpu->arch.eff_db[dr] = val;
1183 if (!kvm_dr6_valid(val))
1185 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1189 if (!kvm_dr7_valid(val))
1191 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1192 kvm_update_dr7(vcpu);
1198 EXPORT_SYMBOL_GPL(kvm_set_dr);
1200 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1202 size_t size = ARRAY_SIZE(vcpu->arch.db);
1206 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1210 *val = vcpu->arch.dr6;
1214 *val = vcpu->arch.dr7;
1218 EXPORT_SYMBOL_GPL(kvm_get_dr);
1220 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1222 u32 ecx = kvm_rcx_read(vcpu);
1225 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1226 kvm_inject_gp(vcpu, 0);
1230 kvm_rax_write(vcpu, (u32)data);
1231 kvm_rdx_write(vcpu, data >> 32);
1232 return kvm_skip_emulated_instruction(vcpu);
1234 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1237 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1238 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1240 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1241 * extract the supported MSRs from the related const lists.
1242 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1243 * capabilities of the host cpu. This capabilities test skips MSRs that are
1244 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1245 * may depend on host virtualization features rather than host cpu features.
1248 static const u32 msrs_to_save_all[] = {
1249 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1251 #ifdef CONFIG_X86_64
1252 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1254 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1255 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1257 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1258 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1259 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1260 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1261 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1262 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1263 MSR_IA32_UMWAIT_CONTROL,
1265 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1266 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1267 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1268 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1269 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1270 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1271 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1272 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1273 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1274 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1275 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1276 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1277 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1278 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1279 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1280 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1281 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1282 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1283 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1284 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1285 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1286 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1289 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1290 static unsigned num_msrs_to_save;
1292 static const u32 emulated_msrs_all[] = {
1293 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1294 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1295 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1296 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1297 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1298 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1299 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1301 HV_X64_MSR_VP_INDEX,
1302 HV_X64_MSR_VP_RUNTIME,
1303 HV_X64_MSR_SCONTROL,
1304 HV_X64_MSR_STIMER0_CONFIG,
1305 HV_X64_MSR_VP_ASSIST_PAGE,
1306 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1307 HV_X64_MSR_TSC_EMULATION_STATUS,
1308 HV_X64_MSR_SYNDBG_OPTIONS,
1309 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1310 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1311 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1313 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1314 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1316 MSR_IA32_TSC_ADJUST,
1317 MSR_IA32_TSC_DEADLINE,
1318 MSR_IA32_ARCH_CAPABILITIES,
1319 MSR_IA32_PERF_CAPABILITIES,
1320 MSR_IA32_MISC_ENABLE,
1321 MSR_IA32_MCG_STATUS,
1323 MSR_IA32_MCG_EXT_CTL,
1327 MSR_MISC_FEATURES_ENABLES,
1328 MSR_AMD64_VIRT_SPEC_CTRL,
1333 * The following list leaves out MSRs whose values are determined
1334 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1335 * We always support the "true" VMX control MSRs, even if the host
1336 * processor does not, so I am putting these registers here rather
1337 * than in msrs_to_save_all.
1340 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1341 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1342 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1343 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1345 MSR_IA32_VMX_CR0_FIXED0,
1346 MSR_IA32_VMX_CR4_FIXED0,
1347 MSR_IA32_VMX_VMCS_ENUM,
1348 MSR_IA32_VMX_PROCBASED_CTLS2,
1349 MSR_IA32_VMX_EPT_VPID_CAP,
1350 MSR_IA32_VMX_VMFUNC,
1353 MSR_KVM_POLL_CONTROL,
1356 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1357 static unsigned num_emulated_msrs;
1360 * List of msr numbers which are used to expose MSR-based features that
1361 * can be used by a hypervisor to validate requested CPU features.
1363 static const u32 msr_based_features_all[] = {
1365 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1366 MSR_IA32_VMX_PINBASED_CTLS,
1367 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1368 MSR_IA32_VMX_PROCBASED_CTLS,
1369 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1370 MSR_IA32_VMX_EXIT_CTLS,
1371 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1372 MSR_IA32_VMX_ENTRY_CTLS,
1374 MSR_IA32_VMX_CR0_FIXED0,
1375 MSR_IA32_VMX_CR0_FIXED1,
1376 MSR_IA32_VMX_CR4_FIXED0,
1377 MSR_IA32_VMX_CR4_FIXED1,
1378 MSR_IA32_VMX_VMCS_ENUM,
1379 MSR_IA32_VMX_PROCBASED_CTLS2,
1380 MSR_IA32_VMX_EPT_VPID_CAP,
1381 MSR_IA32_VMX_VMFUNC,
1385 MSR_IA32_ARCH_CAPABILITIES,
1386 MSR_IA32_PERF_CAPABILITIES,
1389 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1390 static unsigned int num_msr_based_features;
1392 static u64 kvm_get_arch_capabilities(void)
1396 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1397 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1400 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1401 * the nested hypervisor runs with NX huge pages. If it is not,
1402 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1403 * L1 guests, so it need not worry about its own (L2) guests.
1405 data |= ARCH_CAP_PSCHANGE_MC_NO;
1408 * If we're doing cache flushes (either "always" or "cond")
1409 * we will do one whenever the guest does a vmlaunch/vmresume.
1410 * If an outer hypervisor is doing the cache flush for us
1411 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1412 * capability to the guest too, and if EPT is disabled we're not
1413 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1414 * require a nested hypervisor to do a flush of its own.
1416 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1417 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1419 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1420 data |= ARCH_CAP_RDCL_NO;
1421 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1422 data |= ARCH_CAP_SSB_NO;
1423 if (!boot_cpu_has_bug(X86_BUG_MDS))
1424 data |= ARCH_CAP_MDS_NO;
1426 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1428 * If RTM=0 because the kernel has disabled TSX, the host might
1429 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1430 * and therefore knows that there cannot be TAA) but keep
1431 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1432 * and we want to allow migrating those guests to tsx=off hosts.
1434 data &= ~ARCH_CAP_TAA_NO;
1435 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1436 data |= ARCH_CAP_TAA_NO;
1439 * Nothing to do here; we emulate TSX_CTRL if present on the
1440 * host so the guest can choose between disabling TSX or
1441 * using VERW to clear CPU buffers.
1448 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1450 switch (msr->index) {
1451 case MSR_IA32_ARCH_CAPABILITIES:
1452 msr->data = kvm_get_arch_capabilities();
1454 case MSR_IA32_UCODE_REV:
1455 rdmsrl_safe(msr->index, &msr->data);
1458 return static_call(kvm_x86_get_msr_feature)(msr);
1463 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1465 struct kvm_msr_entry msr;
1469 r = kvm_get_msr_feature(&msr);
1471 if (r == KVM_MSR_RET_INVALID) {
1472 /* Unconditionally clear the output for simplicity */
1474 if (kvm_msr_ignored_check(index, 0, false))
1486 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1488 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1491 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1494 if (efer & (EFER_LME | EFER_LMA) &&
1495 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1498 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1504 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1506 if (efer & efer_reserved_bits)
1509 return __kvm_valid_efer(vcpu, efer);
1511 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1513 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1515 u64 old_efer = vcpu->arch.efer;
1516 u64 efer = msr_info->data;
1519 if (efer & efer_reserved_bits)
1522 if (!msr_info->host_initiated) {
1523 if (!__kvm_valid_efer(vcpu, efer))
1526 if (is_paging(vcpu) &&
1527 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1532 efer |= vcpu->arch.efer & EFER_LMA;
1534 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1540 /* Update reserved bits */
1541 if ((efer ^ old_efer) & EFER_NX)
1542 kvm_mmu_reset_context(vcpu);
1547 void kvm_enable_efer_bits(u64 mask)
1549 efer_reserved_bits &= ~mask;
1551 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1553 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1555 struct kvm_x86_msr_filter *msr_filter;
1556 struct msr_bitmap_range *ranges;
1557 struct kvm *kvm = vcpu->kvm;
1562 /* x2APIC MSRs do not support filtering. */
1563 if (index >= 0x800 && index <= 0x8ff)
1566 idx = srcu_read_lock(&kvm->srcu);
1568 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1574 allowed = msr_filter->default_allow;
1575 ranges = msr_filter->ranges;
1577 for (i = 0; i < msr_filter->count; i++) {
1578 u32 start = ranges[i].base;
1579 u32 end = start + ranges[i].nmsrs;
1580 u32 flags = ranges[i].flags;
1581 unsigned long *bitmap = ranges[i].bitmap;
1583 if ((index >= start) && (index < end) && (flags & type)) {
1584 allowed = !!test_bit(index - start, bitmap);
1590 srcu_read_unlock(&kvm->srcu, idx);
1594 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1597 * Write @data into the MSR specified by @index. Select MSR specific fault
1598 * checks are bypassed if @host_initiated is %true.
1599 * Returns 0 on success, non-0 otherwise.
1600 * Assumes vcpu_load() was already called.
1602 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1603 bool host_initiated)
1605 struct msr_data msr;
1607 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1608 return KVM_MSR_RET_FILTERED;
1613 case MSR_KERNEL_GS_BASE:
1616 if (is_noncanonical_address(data, vcpu))
1619 case MSR_IA32_SYSENTER_EIP:
1620 case MSR_IA32_SYSENTER_ESP:
1622 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1623 * non-canonical address is written on Intel but not on
1624 * AMD (which ignores the top 32-bits, because it does
1625 * not implement 64-bit SYSENTER).
1627 * 64-bit code should hence be able to write a non-canonical
1628 * value on AMD. Making the address canonical ensures that
1629 * vmentry does not fail on Intel after writing a non-canonical
1630 * value, and that something deterministic happens if the guest
1631 * invokes 64-bit SYSENTER.
1633 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1638 msr.host_initiated = host_initiated;
1640 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1643 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1644 u32 index, u64 data, bool host_initiated)
1646 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1648 if (ret == KVM_MSR_RET_INVALID)
1649 if (kvm_msr_ignored_check(index, data, true))
1656 * Read the MSR specified by @index into @data. Select MSR specific fault
1657 * checks are bypassed if @host_initiated is %true.
1658 * Returns 0 on success, non-0 otherwise.
1659 * Assumes vcpu_load() was already called.
1661 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1662 bool host_initiated)
1664 struct msr_data msr;
1667 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1668 return KVM_MSR_RET_FILTERED;
1671 msr.host_initiated = host_initiated;
1673 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1679 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1680 u32 index, u64 *data, bool host_initiated)
1682 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1684 if (ret == KVM_MSR_RET_INVALID) {
1685 /* Unconditionally clear *data for simplicity */
1687 if (kvm_msr_ignored_check(index, 0, false))
1694 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1696 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1698 EXPORT_SYMBOL_GPL(kvm_get_msr);
1700 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1702 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1704 EXPORT_SYMBOL_GPL(kvm_set_msr);
1706 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1708 int err = vcpu->run->msr.error;
1710 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1711 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1714 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1717 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1719 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1722 static u64 kvm_msr_reason(int r)
1725 case KVM_MSR_RET_INVALID:
1726 return KVM_MSR_EXIT_REASON_UNKNOWN;
1727 case KVM_MSR_RET_FILTERED:
1728 return KVM_MSR_EXIT_REASON_FILTER;
1730 return KVM_MSR_EXIT_REASON_INVAL;
1734 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1735 u32 exit_reason, u64 data,
1736 int (*completion)(struct kvm_vcpu *vcpu),
1739 u64 msr_reason = kvm_msr_reason(r);
1741 /* Check if the user wanted to know about this MSR fault */
1742 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1745 vcpu->run->exit_reason = exit_reason;
1746 vcpu->run->msr.error = 0;
1747 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1748 vcpu->run->msr.reason = msr_reason;
1749 vcpu->run->msr.index = index;
1750 vcpu->run->msr.data = data;
1751 vcpu->arch.complete_userspace_io = completion;
1756 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1758 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1759 complete_emulated_rdmsr, r);
1762 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1764 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1765 complete_emulated_wrmsr, r);
1768 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1770 u32 ecx = kvm_rcx_read(vcpu);
1774 r = kvm_get_msr(vcpu, ecx, &data);
1776 /* MSR read failed? See if we should ask user space */
1777 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1778 /* Bounce to user space */
1783 trace_kvm_msr_read(ecx, data);
1785 kvm_rax_write(vcpu, data & -1u);
1786 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1788 trace_kvm_msr_read_ex(ecx);
1791 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1793 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1795 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1797 u32 ecx = kvm_rcx_read(vcpu);
1798 u64 data = kvm_read_edx_eax(vcpu);
1801 r = kvm_set_msr(vcpu, ecx, data);
1803 /* MSR write failed? See if we should ask user space */
1804 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1805 /* Bounce to user space */
1808 /* Signal all other negative errors to userspace */
1813 trace_kvm_msr_write(ecx, data);
1815 trace_kvm_msr_write_ex(ecx, data);
1817 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1819 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1821 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1823 return kvm_skip_emulated_instruction(vcpu);
1825 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1827 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1829 /* Treat an INVD instruction as a NOP and just skip it. */
1830 return kvm_emulate_as_nop(vcpu);
1832 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1834 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1836 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1837 return kvm_emulate_as_nop(vcpu);
1839 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1841 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1843 kvm_queue_exception(vcpu, UD_VECTOR);
1846 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1848 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1850 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1851 return kvm_emulate_as_nop(vcpu);
1853 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1855 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1857 xfer_to_guest_mode_prepare();
1858 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1859 xfer_to_guest_mode_work_pending();
1863 * The fast path for frequent and performance sensitive wrmsr emulation,
1864 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1865 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1866 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1867 * other cases which must be called after interrupts are enabled on the host.
1869 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1871 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1874 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1875 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1876 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1877 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1880 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1881 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1882 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1883 trace_kvm_apic_write(APIC_ICR, (u32)data);
1890 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1892 if (!kvm_can_use_hv_timer(vcpu))
1895 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1899 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1901 u32 msr = kvm_rcx_read(vcpu);
1903 fastpath_t ret = EXIT_FASTPATH_NONE;
1906 case APIC_BASE_MSR + (APIC_ICR >> 4):
1907 data = kvm_read_edx_eax(vcpu);
1908 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1909 kvm_skip_emulated_instruction(vcpu);
1910 ret = EXIT_FASTPATH_EXIT_HANDLED;
1913 case MSR_IA32_TSC_DEADLINE:
1914 data = kvm_read_edx_eax(vcpu);
1915 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1916 kvm_skip_emulated_instruction(vcpu);
1917 ret = EXIT_FASTPATH_REENTER_GUEST;
1924 if (ret != EXIT_FASTPATH_NONE)
1925 trace_kvm_msr_write(msr, data);
1929 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1932 * Adapt set_msr() to msr_io()'s calling convention
1934 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1936 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1939 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1941 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1944 #ifdef CONFIG_X86_64
1945 struct pvclock_clock {
1955 struct pvclock_gtod_data {
1958 struct pvclock_clock clock; /* extract of a clocksource struct */
1959 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1965 static struct pvclock_gtod_data pvclock_gtod_data;
1967 static void update_pvclock_gtod(struct timekeeper *tk)
1969 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1971 write_seqcount_begin(&vdata->seq);
1973 /* copy pvclock gtod data */
1974 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1975 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1976 vdata->clock.mask = tk->tkr_mono.mask;
1977 vdata->clock.mult = tk->tkr_mono.mult;
1978 vdata->clock.shift = tk->tkr_mono.shift;
1979 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1980 vdata->clock.offset = tk->tkr_mono.base;
1982 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1983 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1984 vdata->raw_clock.mask = tk->tkr_raw.mask;
1985 vdata->raw_clock.mult = tk->tkr_raw.mult;
1986 vdata->raw_clock.shift = tk->tkr_raw.shift;
1987 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1988 vdata->raw_clock.offset = tk->tkr_raw.base;
1990 vdata->wall_time_sec = tk->xtime_sec;
1992 vdata->offs_boot = tk->offs_boot;
1994 write_seqcount_end(&vdata->seq);
1997 static s64 get_kvmclock_base_ns(void)
1999 /* Count up from boot time, but with the frequency of the raw clock. */
2000 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2003 static s64 get_kvmclock_base_ns(void)
2005 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2006 return ktime_get_boottime_ns();
2010 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2014 struct pvclock_wall_clock wc;
2021 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2026 ++version; /* first time write, random junk */
2030 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2034 * The guest calculates current wall clock time by adding
2035 * system time (updated by kvm_guest_time_update below) to the
2036 * wall clock specified here. We do the reverse here.
2038 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2040 wc.nsec = do_div(wall_nsec, 1000000000);
2041 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2042 wc.version = version;
2044 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2047 wc_sec_hi = wall_nsec >> 32;
2048 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2049 &wc_sec_hi, sizeof(wc_sec_hi));
2053 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2056 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2057 bool old_msr, bool host_initiated)
2059 struct kvm_arch *ka = &vcpu->kvm->arch;
2061 if (vcpu->vcpu_id == 0 && !host_initiated) {
2062 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2063 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2065 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2068 vcpu->arch.time = system_time;
2069 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2071 /* we verify if the enable bit is set... */
2072 vcpu->arch.pv_time_enabled = false;
2073 if (!(system_time & 1))
2076 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2077 &vcpu->arch.pv_time, system_time & ~1ULL,
2078 sizeof(struct pvclock_vcpu_time_info)))
2079 vcpu->arch.pv_time_enabled = true;
2084 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2086 do_shl32_div32(dividend, divisor);
2090 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2091 s8 *pshift, u32 *pmultiplier)
2099 scaled64 = scaled_hz;
2100 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2105 tps32 = (uint32_t)tps64;
2106 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2107 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2115 *pmultiplier = div_frac(scaled64, tps32);
2118 #ifdef CONFIG_X86_64
2119 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2122 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2123 static unsigned long max_tsc_khz;
2125 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2127 u64 v = (u64)khz * (1000000 + ppm);
2132 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2136 /* Guest TSC same frequency as host TSC? */
2138 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2142 /* TSC scaling supported? */
2143 if (!kvm_has_tsc_control) {
2144 if (user_tsc_khz > tsc_khz) {
2145 vcpu->arch.tsc_catchup = 1;
2146 vcpu->arch.tsc_always_catchup = 1;
2149 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2154 /* TSC scaling required - calculate ratio */
2155 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2156 user_tsc_khz, tsc_khz);
2158 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2159 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2164 vcpu->arch.tsc_scaling_ratio = ratio;
2168 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2170 u32 thresh_lo, thresh_hi;
2171 int use_scaling = 0;
2173 /* tsc_khz can be zero if TSC calibration fails */
2174 if (user_tsc_khz == 0) {
2175 /* set tsc_scaling_ratio to a safe value */
2176 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2180 /* Compute a scale to convert nanoseconds in TSC cycles */
2181 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2182 &vcpu->arch.virtual_tsc_shift,
2183 &vcpu->arch.virtual_tsc_mult);
2184 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2187 * Compute the variation in TSC rate which is acceptable
2188 * within the range of tolerance and decide if the
2189 * rate being applied is within that bounds of the hardware
2190 * rate. If so, no scaling or compensation need be done.
2192 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2193 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2194 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2195 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2198 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2201 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2203 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2204 vcpu->arch.virtual_tsc_mult,
2205 vcpu->arch.virtual_tsc_shift);
2206 tsc += vcpu->arch.this_tsc_write;
2210 static inline int gtod_is_based_on_tsc(int mode)
2212 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2215 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2217 #ifdef CONFIG_X86_64
2219 struct kvm_arch *ka = &vcpu->kvm->arch;
2220 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2222 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2223 atomic_read(&vcpu->kvm->online_vcpus));
2226 * Once the masterclock is enabled, always perform request in
2227 * order to update it.
2229 * In order to enable masterclock, the host clocksource must be TSC
2230 * and the vcpus need to have matched TSCs. When that happens,
2231 * perform request to enable masterclock.
2233 if (ka->use_master_clock ||
2234 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2235 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2237 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2238 atomic_read(&vcpu->kvm->online_vcpus),
2239 ka->use_master_clock, gtod->clock.vclock_mode);
2244 * Multiply tsc by a fixed point number represented by ratio.
2246 * The most significant 64-N bits (mult) of ratio represent the
2247 * integral part of the fixed point number; the remaining N bits
2248 * (frac) represent the fractional part, ie. ratio represents a fixed
2249 * point number (mult + frac * 2^(-N)).
2251 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2253 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2255 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2258 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2261 u64 ratio = vcpu->arch.tsc_scaling_ratio;
2263 if (ratio != kvm_default_tsc_scaling_ratio)
2264 _tsc = __scale_tsc(ratio, tsc);
2268 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2270 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2274 tsc = kvm_scale_tsc(vcpu, rdtsc());
2276 return target_tsc - tsc;
2279 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2281 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2283 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2285 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2287 vcpu->arch.l1_tsc_offset = offset;
2288 vcpu->arch.tsc_offset = static_call(kvm_x86_write_l1_tsc_offset)(vcpu, offset);
2291 static inline bool kvm_check_tsc_unstable(void)
2293 #ifdef CONFIG_X86_64
2295 * TSC is marked unstable when we're running on Hyper-V,
2296 * 'TSC page' clocksource is good.
2298 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2301 return check_tsc_unstable();
2304 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2306 struct kvm *kvm = vcpu->kvm;
2307 u64 offset, ns, elapsed;
2308 unsigned long flags;
2310 bool already_matched;
2311 bool synchronizing = false;
2313 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2314 offset = kvm_compute_tsc_offset(vcpu, data);
2315 ns = get_kvmclock_base_ns();
2316 elapsed = ns - kvm->arch.last_tsc_nsec;
2318 if (vcpu->arch.virtual_tsc_khz) {
2321 * detection of vcpu initialization -- need to sync
2322 * with other vCPUs. This particularly helps to keep
2323 * kvm_clock stable after CPU hotplug
2325 synchronizing = true;
2327 u64 tsc_exp = kvm->arch.last_tsc_write +
2328 nsec_to_cycles(vcpu, elapsed);
2329 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2331 * Special case: TSC write with a small delta (1 second)
2332 * of virtual cycle time against real time is
2333 * interpreted as an attempt to synchronize the CPU.
2335 synchronizing = data < tsc_exp + tsc_hz &&
2336 data + tsc_hz > tsc_exp;
2341 * For a reliable TSC, we can match TSC offsets, and for an unstable
2342 * TSC, we add elapsed time in this computation. We could let the
2343 * compensation code attempt to catch up if we fall behind, but
2344 * it's better to try to match offsets from the beginning.
2346 if (synchronizing &&
2347 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2348 if (!kvm_check_tsc_unstable()) {
2349 offset = kvm->arch.cur_tsc_offset;
2351 u64 delta = nsec_to_cycles(vcpu, elapsed);
2353 offset = kvm_compute_tsc_offset(vcpu, data);
2356 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2359 * We split periods of matched TSC writes into generations.
2360 * For each generation, we track the original measured
2361 * nanosecond time, offset, and write, so if TSCs are in
2362 * sync, we can match exact offset, and if not, we can match
2363 * exact software computation in compute_guest_tsc()
2365 * These values are tracked in kvm->arch.cur_xxx variables.
2367 kvm->arch.cur_tsc_generation++;
2368 kvm->arch.cur_tsc_nsec = ns;
2369 kvm->arch.cur_tsc_write = data;
2370 kvm->arch.cur_tsc_offset = offset;
2375 * We also track th most recent recorded KHZ, write and time to
2376 * allow the matching interval to be extended at each write.
2378 kvm->arch.last_tsc_nsec = ns;
2379 kvm->arch.last_tsc_write = data;
2380 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2382 vcpu->arch.last_guest_tsc = data;
2384 /* Keep track of which generation this VCPU has synchronized to */
2385 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2386 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2387 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2389 kvm_vcpu_write_tsc_offset(vcpu, offset);
2390 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2392 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2394 kvm->arch.nr_vcpus_matched_tsc = 0;
2395 } else if (!already_matched) {
2396 kvm->arch.nr_vcpus_matched_tsc++;
2399 kvm_track_tsc_matching(vcpu);
2400 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2403 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2406 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2407 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2410 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2412 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2413 WARN_ON(adjustment < 0);
2414 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2415 adjust_tsc_offset_guest(vcpu, adjustment);
2418 #ifdef CONFIG_X86_64
2420 static u64 read_tsc(void)
2422 u64 ret = (u64)rdtsc_ordered();
2423 u64 last = pvclock_gtod_data.clock.cycle_last;
2425 if (likely(ret >= last))
2429 * GCC likes to generate cmov here, but this branch is extremely
2430 * predictable (it's just a function of time and the likely is
2431 * very likely) and there's a data dependence, so force GCC
2432 * to generate a branch instead. I don't barrier() because
2433 * we don't actually need a barrier, and if this function
2434 * ever gets inlined it will generate worse code.
2440 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2446 switch (clock->vclock_mode) {
2447 case VDSO_CLOCKMODE_HVCLOCK:
2448 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2450 if (tsc_pg_val != U64_MAX) {
2451 /* TSC page valid */
2452 *mode = VDSO_CLOCKMODE_HVCLOCK;
2453 v = (tsc_pg_val - clock->cycle_last) &
2456 /* TSC page invalid */
2457 *mode = VDSO_CLOCKMODE_NONE;
2460 case VDSO_CLOCKMODE_TSC:
2461 *mode = VDSO_CLOCKMODE_TSC;
2462 *tsc_timestamp = read_tsc();
2463 v = (*tsc_timestamp - clock->cycle_last) &
2467 *mode = VDSO_CLOCKMODE_NONE;
2470 if (*mode == VDSO_CLOCKMODE_NONE)
2471 *tsc_timestamp = v = 0;
2473 return v * clock->mult;
2476 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2478 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2484 seq = read_seqcount_begin(>od->seq);
2485 ns = gtod->raw_clock.base_cycles;
2486 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2487 ns >>= gtod->raw_clock.shift;
2488 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2489 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2495 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2497 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2503 seq = read_seqcount_begin(>od->seq);
2504 ts->tv_sec = gtod->wall_time_sec;
2505 ns = gtod->clock.base_cycles;
2506 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2507 ns >>= gtod->clock.shift;
2508 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2510 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2516 /* returns true if host is using TSC based clocksource */
2517 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2519 /* checked again under seqlock below */
2520 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2523 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2527 /* returns true if host is using TSC based clocksource */
2528 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2531 /* checked again under seqlock below */
2532 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2535 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2541 * Assuming a stable TSC across physical CPUS, and a stable TSC
2542 * across virtual CPUs, the following condition is possible.
2543 * Each numbered line represents an event visible to both
2544 * CPUs at the next numbered event.
2546 * "timespecX" represents host monotonic time. "tscX" represents
2549 * VCPU0 on CPU0 | VCPU1 on CPU1
2551 * 1. read timespec0,tsc0
2552 * 2. | timespec1 = timespec0 + N
2554 * 3. transition to guest | transition to guest
2555 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2556 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2557 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2559 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2562 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2564 * - 0 < N - M => M < N
2566 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2567 * always the case (the difference between two distinct xtime instances
2568 * might be smaller then the difference between corresponding TSC reads,
2569 * when updating guest vcpus pvclock areas).
2571 * To avoid that problem, do not allow visibility of distinct
2572 * system_timestamp/tsc_timestamp values simultaneously: use a master
2573 * copy of host monotonic time values. Update that master copy
2576 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2580 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2582 #ifdef CONFIG_X86_64
2583 struct kvm_arch *ka = &kvm->arch;
2585 bool host_tsc_clocksource, vcpus_matched;
2587 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2588 atomic_read(&kvm->online_vcpus));
2591 * If the host uses TSC clock, then passthrough TSC as stable
2594 host_tsc_clocksource = kvm_get_time_and_clockread(
2595 &ka->master_kernel_ns,
2596 &ka->master_cycle_now);
2598 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2599 && !ka->backwards_tsc_observed
2600 && !ka->boot_vcpu_runs_old_kvmclock;
2602 if (ka->use_master_clock)
2603 atomic_set(&kvm_guest_has_master_clock, 1);
2605 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2606 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2611 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2613 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2616 static void kvm_gen_update_masterclock(struct kvm *kvm)
2618 #ifdef CONFIG_X86_64
2620 struct kvm_vcpu *vcpu;
2621 struct kvm_arch *ka = &kvm->arch;
2622 unsigned long flags;
2624 kvm_hv_invalidate_tsc_page(kvm);
2626 kvm_make_mclock_inprogress_request(kvm);
2628 /* no guest entries from this point */
2629 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2630 pvclock_update_vm_gtod_copy(kvm);
2631 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2633 kvm_for_each_vcpu(i, vcpu, kvm)
2634 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2636 /* guest entries allowed */
2637 kvm_for_each_vcpu(i, vcpu, kvm)
2638 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2642 u64 get_kvmclock_ns(struct kvm *kvm)
2644 struct kvm_arch *ka = &kvm->arch;
2645 struct pvclock_vcpu_time_info hv_clock;
2646 unsigned long flags;
2649 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2650 if (!ka->use_master_clock) {
2651 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2652 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2655 hv_clock.tsc_timestamp = ka->master_cycle_now;
2656 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2657 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2659 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2662 if (__this_cpu_read(cpu_tsc_khz)) {
2663 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2664 &hv_clock.tsc_shift,
2665 &hv_clock.tsc_to_system_mul);
2666 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2668 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2675 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2676 struct gfn_to_hva_cache *cache,
2677 unsigned int offset)
2679 struct kvm_vcpu_arch *vcpu = &v->arch;
2680 struct pvclock_vcpu_time_info guest_hv_clock;
2682 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2683 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2686 /* This VCPU is paused, but it's legal for a guest to read another
2687 * VCPU's kvmclock, so we really have to follow the specification where
2688 * it says that version is odd if data is being modified, and even after
2691 * Version field updates must be kept separate. This is because
2692 * kvm_write_guest_cached might use a "rep movs" instruction, and
2693 * writes within a string instruction are weakly ordered. So there
2694 * are three writes overall.
2696 * As a small optimization, only write the version field in the first
2697 * and third write. The vcpu->pv_time cache is still valid, because the
2698 * version field is the first in the struct.
2700 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2702 if (guest_hv_clock.version & 1)
2703 ++guest_hv_clock.version; /* first time write, random junk */
2705 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2706 kvm_write_guest_offset_cached(v->kvm, cache,
2707 &vcpu->hv_clock, offset,
2708 sizeof(vcpu->hv_clock.version));
2712 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2713 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2715 if (vcpu->pvclock_set_guest_stopped_request) {
2716 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2717 vcpu->pvclock_set_guest_stopped_request = false;
2720 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2722 kvm_write_guest_offset_cached(v->kvm, cache,
2723 &vcpu->hv_clock, offset,
2724 sizeof(vcpu->hv_clock));
2728 vcpu->hv_clock.version++;
2729 kvm_write_guest_offset_cached(v->kvm, cache,
2730 &vcpu->hv_clock, offset,
2731 sizeof(vcpu->hv_clock.version));
2734 static int kvm_guest_time_update(struct kvm_vcpu *v)
2736 unsigned long flags, tgt_tsc_khz;
2737 struct kvm_vcpu_arch *vcpu = &v->arch;
2738 struct kvm_arch *ka = &v->kvm->arch;
2740 u64 tsc_timestamp, host_tsc;
2742 bool use_master_clock;
2748 * If the host uses TSC clock, then passthrough TSC as stable
2751 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2752 use_master_clock = ka->use_master_clock;
2753 if (use_master_clock) {
2754 host_tsc = ka->master_cycle_now;
2755 kernel_ns = ka->master_kernel_ns;
2757 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2759 /* Keep irq disabled to prevent changes to the clock */
2760 local_irq_save(flags);
2761 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2762 if (unlikely(tgt_tsc_khz == 0)) {
2763 local_irq_restore(flags);
2764 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2767 if (!use_master_clock) {
2769 kernel_ns = get_kvmclock_base_ns();
2772 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2775 * We may have to catch up the TSC to match elapsed wall clock
2776 * time for two reasons, even if kvmclock is used.
2777 * 1) CPU could have been running below the maximum TSC rate
2778 * 2) Broken TSC compensation resets the base at each VCPU
2779 * entry to avoid unknown leaps of TSC even when running
2780 * again on the same CPU. This may cause apparent elapsed
2781 * time to disappear, and the guest to stand still or run
2784 if (vcpu->tsc_catchup) {
2785 u64 tsc = compute_guest_tsc(v, kernel_ns);
2786 if (tsc > tsc_timestamp) {
2787 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2788 tsc_timestamp = tsc;
2792 local_irq_restore(flags);
2794 /* With all the info we got, fill in the values */
2796 if (kvm_has_tsc_control)
2797 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2799 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2800 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2801 &vcpu->hv_clock.tsc_shift,
2802 &vcpu->hv_clock.tsc_to_system_mul);
2803 vcpu->hw_tsc_khz = tgt_tsc_khz;
2806 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2807 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2808 vcpu->last_guest_tsc = tsc_timestamp;
2810 /* If the host uses TSC clocksource, then it is stable */
2812 if (use_master_clock)
2813 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2815 vcpu->hv_clock.flags = pvclock_flags;
2817 if (vcpu->pv_time_enabled)
2818 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2819 if (vcpu->xen.vcpu_info_set)
2820 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2821 offsetof(struct compat_vcpu_info, time));
2822 if (vcpu->xen.vcpu_time_info_set)
2823 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2824 if (v == kvm_get_vcpu(v->kvm, 0))
2825 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2830 * kvmclock updates which are isolated to a given vcpu, such as
2831 * vcpu->cpu migration, should not allow system_timestamp from
2832 * the rest of the vcpus to remain static. Otherwise ntp frequency
2833 * correction applies to one vcpu's system_timestamp but not
2836 * So in those cases, request a kvmclock update for all vcpus.
2837 * We need to rate-limit these requests though, as they can
2838 * considerably slow guests that have a large number of vcpus.
2839 * The time for a remote vcpu to update its kvmclock is bound
2840 * by the delay we use to rate-limit the updates.
2843 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2845 static void kvmclock_update_fn(struct work_struct *work)
2848 struct delayed_work *dwork = to_delayed_work(work);
2849 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2850 kvmclock_update_work);
2851 struct kvm *kvm = container_of(ka, struct kvm, arch);
2852 struct kvm_vcpu *vcpu;
2854 kvm_for_each_vcpu(i, vcpu, kvm) {
2855 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2856 kvm_vcpu_kick(vcpu);
2860 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2862 struct kvm *kvm = v->kvm;
2864 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2865 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2866 KVMCLOCK_UPDATE_DELAY);
2869 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2871 static void kvmclock_sync_fn(struct work_struct *work)
2873 struct delayed_work *dwork = to_delayed_work(work);
2874 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2875 kvmclock_sync_work);
2876 struct kvm *kvm = container_of(ka, struct kvm, arch);
2878 if (!kvmclock_periodic_sync)
2881 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2882 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2883 KVMCLOCK_SYNC_PERIOD);
2887 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2889 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2891 /* McStatusWrEn enabled? */
2892 if (guest_cpuid_is_amd_or_hygon(vcpu))
2893 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2898 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2900 u64 mcg_cap = vcpu->arch.mcg_cap;
2901 unsigned bank_num = mcg_cap & 0xff;
2902 u32 msr = msr_info->index;
2903 u64 data = msr_info->data;
2906 case MSR_IA32_MCG_STATUS:
2907 vcpu->arch.mcg_status = data;
2909 case MSR_IA32_MCG_CTL:
2910 if (!(mcg_cap & MCG_CTL_P) &&
2911 (data || !msr_info->host_initiated))
2913 if (data != 0 && data != ~(u64)0)
2915 vcpu->arch.mcg_ctl = data;
2918 if (msr >= MSR_IA32_MC0_CTL &&
2919 msr < MSR_IA32_MCx_CTL(bank_num)) {
2920 u32 offset = array_index_nospec(
2921 msr - MSR_IA32_MC0_CTL,
2922 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2924 /* only 0 or all 1s can be written to IA32_MCi_CTL
2925 * some Linux kernels though clear bit 10 in bank 4 to
2926 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2927 * this to avoid an uncatched #GP in the guest
2929 if ((offset & 0x3) == 0 &&
2930 data != 0 && (data | (1 << 10)) != ~(u64)0)
2934 if (!msr_info->host_initiated &&
2935 (offset & 0x3) == 1 && data != 0) {
2936 if (!can_set_mci_status(vcpu))
2940 vcpu->arch.mce_banks[offset] = data;
2948 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2950 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2952 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2955 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2957 gpa_t gpa = data & ~0x3f;
2959 /* Bits 4:5 are reserved, Should be zero */
2963 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2964 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2967 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2968 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2971 if (!lapic_in_kernel(vcpu))
2972 return data ? 1 : 0;
2974 vcpu->arch.apf.msr_en_val = data;
2976 if (!kvm_pv_async_pf_enabled(vcpu)) {
2977 kvm_clear_async_pf_completion_queue(vcpu);
2978 kvm_async_pf_hash_reset(vcpu);
2982 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2986 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2987 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2989 kvm_async_pf_wakeup_all(vcpu);
2994 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2996 /* Bits 8-63 are reserved */
3000 if (!lapic_in_kernel(vcpu))
3003 vcpu->arch.apf.msr_int_val = data;
3005 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3010 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3012 vcpu->arch.pv_time_enabled = false;
3013 vcpu->arch.time = 0;
3016 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3018 ++vcpu->stat.tlb_flush;
3019 static_call(kvm_x86_tlb_flush_all)(vcpu);
3022 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3024 ++vcpu->stat.tlb_flush;
3025 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3028 static void record_steal_time(struct kvm_vcpu *vcpu)
3030 struct kvm_host_map map;
3031 struct kvm_steal_time *st;
3033 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3034 kvm_xen_runstate_set_running(vcpu);
3038 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3041 /* -EAGAIN is returned in atomic context so we can just return. */
3042 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3043 &map, &vcpu->arch.st.cache, false))
3047 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3050 * Doing a TLB flush here, on the guest's behalf, can avoid
3053 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3054 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3055 st->preempted & KVM_VCPU_FLUSH_TLB);
3056 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
3057 kvm_vcpu_flush_tlb_guest(vcpu);
3060 vcpu->arch.st.preempted = 0;
3062 if (st->version & 1)
3063 st->version += 1; /* first time write, random junk */
3069 st->steal += current->sched_info.run_delay -
3070 vcpu->arch.st.last_steal;
3071 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3077 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3080 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3083 u32 msr = msr_info->index;
3084 u64 data = msr_info->data;
3086 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3087 return kvm_xen_write_hypercall_page(vcpu, data);
3090 case MSR_AMD64_NB_CFG:
3091 case MSR_IA32_UCODE_WRITE:
3092 case MSR_VM_HSAVE_PA:
3093 case MSR_AMD64_PATCH_LOADER:
3094 case MSR_AMD64_BU_CFG2:
3095 case MSR_AMD64_DC_CFG:
3096 case MSR_F15H_EX_CFG:
3099 case MSR_IA32_UCODE_REV:
3100 if (msr_info->host_initiated)
3101 vcpu->arch.microcode_version = data;
3103 case MSR_IA32_ARCH_CAPABILITIES:
3104 if (!msr_info->host_initiated)
3106 vcpu->arch.arch_capabilities = data;
3108 case MSR_IA32_PERF_CAPABILITIES: {
3109 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3111 if (!msr_info->host_initiated)
3113 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3115 if (data & ~msr_ent.data)
3118 vcpu->arch.perf_capabilities = data;
3123 return set_efer(vcpu, msr_info);
3125 data &= ~(u64)0x40; /* ignore flush filter disable */
3126 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3127 data &= ~(u64)0x8; /* ignore TLB cache disable */
3129 /* Handle McStatusWrEn */
3130 if (data == BIT_ULL(18)) {
3131 vcpu->arch.msr_hwcr = data;
3132 } else if (data != 0) {
3133 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3138 case MSR_FAM10H_MMIO_CONF_BASE:
3140 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3145 case 0x200 ... 0x2ff:
3146 return kvm_mtrr_set_msr(vcpu, msr, data);
3147 case MSR_IA32_APICBASE:
3148 return kvm_set_apic_base(vcpu, msr_info);
3149 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3150 return kvm_x2apic_msr_write(vcpu, msr, data);
3151 case MSR_IA32_TSC_DEADLINE:
3152 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3154 case MSR_IA32_TSC_ADJUST:
3155 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3156 if (!msr_info->host_initiated) {
3157 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3158 adjust_tsc_offset_guest(vcpu, adj);
3160 vcpu->arch.ia32_tsc_adjust_msr = data;
3163 case MSR_IA32_MISC_ENABLE:
3164 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3165 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3166 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3168 vcpu->arch.ia32_misc_enable_msr = data;
3169 kvm_update_cpuid_runtime(vcpu);
3171 vcpu->arch.ia32_misc_enable_msr = data;
3174 case MSR_IA32_SMBASE:
3175 if (!msr_info->host_initiated)
3177 vcpu->arch.smbase = data;
3179 case MSR_IA32_POWER_CTL:
3180 vcpu->arch.msr_ia32_power_ctl = data;
3183 if (msr_info->host_initiated) {
3184 kvm_synchronize_tsc(vcpu, data);
3186 u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3187 adjust_tsc_offset_guest(vcpu, adj);
3188 vcpu->arch.ia32_tsc_adjust_msr += adj;
3192 if (!msr_info->host_initiated &&
3193 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3196 * KVM supports exposing PT to the guest, but does not support
3197 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3198 * XSAVES/XRSTORS to save/restore PT MSRs.
3200 if (data & ~supported_xss)
3202 vcpu->arch.ia32_xss = data;
3205 if (!msr_info->host_initiated)
3207 vcpu->arch.smi_count = data;
3209 case MSR_KVM_WALL_CLOCK_NEW:
3210 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3213 vcpu->kvm->arch.wall_clock = data;
3214 kvm_write_wall_clock(vcpu->kvm, data, 0);
3216 case MSR_KVM_WALL_CLOCK:
3217 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3220 vcpu->kvm->arch.wall_clock = data;
3221 kvm_write_wall_clock(vcpu->kvm, data, 0);
3223 case MSR_KVM_SYSTEM_TIME_NEW:
3224 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3227 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3229 case MSR_KVM_SYSTEM_TIME:
3230 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3233 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3235 case MSR_KVM_ASYNC_PF_EN:
3236 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3239 if (kvm_pv_enable_async_pf(vcpu, data))
3242 case MSR_KVM_ASYNC_PF_INT:
3243 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3246 if (kvm_pv_enable_async_pf_int(vcpu, data))
3249 case MSR_KVM_ASYNC_PF_ACK:
3250 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3253 vcpu->arch.apf.pageready_pending = false;
3254 kvm_check_async_pf_completion(vcpu);
3257 case MSR_KVM_STEAL_TIME:
3258 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3261 if (unlikely(!sched_info_on()))
3264 if (data & KVM_STEAL_RESERVED_MASK)
3267 vcpu->arch.st.msr_val = data;
3269 if (!(data & KVM_MSR_ENABLED))
3272 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3275 case MSR_KVM_PV_EOI_EN:
3276 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3279 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3283 case MSR_KVM_POLL_CONTROL:
3284 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3287 /* only enable bit supported */
3288 if (data & (-1ULL << 1))
3291 vcpu->arch.msr_kvm_poll_control = data;
3294 case MSR_IA32_MCG_CTL:
3295 case MSR_IA32_MCG_STATUS:
3296 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3297 return set_msr_mce(vcpu, msr_info);
3299 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3300 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3303 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3304 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3305 if (kvm_pmu_is_valid_msr(vcpu, msr))
3306 return kvm_pmu_set_msr(vcpu, msr_info);
3308 if (pr || data != 0)
3309 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3310 "0x%x data 0x%llx\n", msr, data);
3312 case MSR_K7_CLK_CTL:
3314 * Ignore all writes to this no longer documented MSR.
3315 * Writes are only relevant for old K7 processors,
3316 * all pre-dating SVM, but a recommended workaround from
3317 * AMD for these chips. It is possible to specify the
3318 * affected processor models on the command line, hence
3319 * the need to ignore the workaround.
3322 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3323 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3324 case HV_X64_MSR_SYNDBG_OPTIONS:
3325 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3326 case HV_X64_MSR_CRASH_CTL:
3327 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3328 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3329 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3330 case HV_X64_MSR_TSC_EMULATION_STATUS:
3331 return kvm_hv_set_msr_common(vcpu, msr, data,
3332 msr_info->host_initiated);
3333 case MSR_IA32_BBL_CR_CTL3:
3334 /* Drop writes to this legacy MSR -- see rdmsr
3335 * counterpart for further detail.
3337 if (report_ignored_msrs)
3338 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3341 case MSR_AMD64_OSVW_ID_LENGTH:
3342 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3344 vcpu->arch.osvw.length = data;
3346 case MSR_AMD64_OSVW_STATUS:
3347 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3349 vcpu->arch.osvw.status = data;
3351 case MSR_PLATFORM_INFO:
3352 if (!msr_info->host_initiated ||
3353 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3354 cpuid_fault_enabled(vcpu)))
3356 vcpu->arch.msr_platform_info = data;
3358 case MSR_MISC_FEATURES_ENABLES:
3359 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3360 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3361 !supports_cpuid_fault(vcpu)))
3363 vcpu->arch.msr_misc_features_enables = data;
3366 if (kvm_pmu_is_valid_msr(vcpu, msr))
3367 return kvm_pmu_set_msr(vcpu, msr_info);
3368 return KVM_MSR_RET_INVALID;
3372 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3374 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3377 u64 mcg_cap = vcpu->arch.mcg_cap;
3378 unsigned bank_num = mcg_cap & 0xff;
3381 case MSR_IA32_P5_MC_ADDR:
3382 case MSR_IA32_P5_MC_TYPE:
3385 case MSR_IA32_MCG_CAP:
3386 data = vcpu->arch.mcg_cap;
3388 case MSR_IA32_MCG_CTL:
3389 if (!(mcg_cap & MCG_CTL_P) && !host)
3391 data = vcpu->arch.mcg_ctl;
3393 case MSR_IA32_MCG_STATUS:
3394 data = vcpu->arch.mcg_status;
3397 if (msr >= MSR_IA32_MC0_CTL &&
3398 msr < MSR_IA32_MCx_CTL(bank_num)) {
3399 u32 offset = array_index_nospec(
3400 msr - MSR_IA32_MC0_CTL,
3401 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3403 data = vcpu->arch.mce_banks[offset];
3412 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3414 switch (msr_info->index) {
3415 case MSR_IA32_PLATFORM_ID:
3416 case MSR_IA32_EBL_CR_POWERON:
3417 case MSR_IA32_LASTBRANCHFROMIP:
3418 case MSR_IA32_LASTBRANCHTOIP:
3419 case MSR_IA32_LASTINTFROMIP:
3420 case MSR_IA32_LASTINTTOIP:
3422 case MSR_K8_TSEG_ADDR:
3423 case MSR_K8_TSEG_MASK:
3424 case MSR_VM_HSAVE_PA:
3425 case MSR_K8_INT_PENDING_MSG:
3426 case MSR_AMD64_NB_CFG:
3427 case MSR_FAM10H_MMIO_CONF_BASE:
3428 case MSR_AMD64_BU_CFG2:
3429 case MSR_IA32_PERF_CTL:
3430 case MSR_AMD64_DC_CFG:
3431 case MSR_F15H_EX_CFG:
3433 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3434 * limit) MSRs. Just return 0, as we do not want to expose the host
3435 * data here. Do not conditionalize this on CPUID, as KVM does not do
3436 * so for existing CPU-specific MSRs.
3438 case MSR_RAPL_POWER_UNIT:
3439 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3440 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3441 case MSR_PKG_ENERGY_STATUS: /* Total package */
3442 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3445 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3446 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3447 return kvm_pmu_get_msr(vcpu, msr_info);
3448 if (!msr_info->host_initiated)
3452 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3453 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3454 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3455 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3456 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3457 return kvm_pmu_get_msr(vcpu, msr_info);
3460 case MSR_IA32_UCODE_REV:
3461 msr_info->data = vcpu->arch.microcode_version;
3463 case MSR_IA32_ARCH_CAPABILITIES:
3464 if (!msr_info->host_initiated &&
3465 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3467 msr_info->data = vcpu->arch.arch_capabilities;
3469 case MSR_IA32_PERF_CAPABILITIES:
3470 if (!msr_info->host_initiated &&
3471 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3473 msr_info->data = vcpu->arch.perf_capabilities;
3475 case MSR_IA32_POWER_CTL:
3476 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3478 case MSR_IA32_TSC: {
3480 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3481 * even when not intercepted. AMD manual doesn't explicitly
3482 * state this but appears to behave the same.
3484 * On userspace reads and writes, however, we unconditionally
3485 * return L1's TSC value to ensure backwards-compatible
3486 * behavior for migration.
3488 u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3489 vcpu->arch.tsc_offset;
3491 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3495 case 0x200 ... 0x2ff:
3496 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3497 case 0xcd: /* fsb frequency */
3501 * MSR_EBC_FREQUENCY_ID
3502 * Conservative value valid for even the basic CPU models.
3503 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3504 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3505 * and 266MHz for model 3, or 4. Set Core Clock
3506 * Frequency to System Bus Frequency Ratio to 1 (bits
3507 * 31:24) even though these are only valid for CPU
3508 * models > 2, however guests may end up dividing or
3509 * multiplying by zero otherwise.
3511 case MSR_EBC_FREQUENCY_ID:
3512 msr_info->data = 1 << 24;
3514 case MSR_IA32_APICBASE:
3515 msr_info->data = kvm_get_apic_base(vcpu);
3517 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3518 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3519 case MSR_IA32_TSC_DEADLINE:
3520 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3522 case MSR_IA32_TSC_ADJUST:
3523 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3525 case MSR_IA32_MISC_ENABLE:
3526 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3528 case MSR_IA32_SMBASE:
3529 if (!msr_info->host_initiated)
3531 msr_info->data = vcpu->arch.smbase;
3534 msr_info->data = vcpu->arch.smi_count;
3536 case MSR_IA32_PERF_STATUS:
3537 /* TSC increment by tick */
3538 msr_info->data = 1000ULL;
3539 /* CPU multiplier */
3540 msr_info->data |= (((uint64_t)4ULL) << 40);
3543 msr_info->data = vcpu->arch.efer;
3545 case MSR_KVM_WALL_CLOCK:
3546 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3549 msr_info->data = vcpu->kvm->arch.wall_clock;
3551 case MSR_KVM_WALL_CLOCK_NEW:
3552 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3555 msr_info->data = vcpu->kvm->arch.wall_clock;
3557 case MSR_KVM_SYSTEM_TIME:
3558 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3561 msr_info->data = vcpu->arch.time;
3563 case MSR_KVM_SYSTEM_TIME_NEW:
3564 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3567 msr_info->data = vcpu->arch.time;
3569 case MSR_KVM_ASYNC_PF_EN:
3570 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3573 msr_info->data = vcpu->arch.apf.msr_en_val;
3575 case MSR_KVM_ASYNC_PF_INT:
3576 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3579 msr_info->data = vcpu->arch.apf.msr_int_val;
3581 case MSR_KVM_ASYNC_PF_ACK:
3582 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3587 case MSR_KVM_STEAL_TIME:
3588 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3591 msr_info->data = vcpu->arch.st.msr_val;
3593 case MSR_KVM_PV_EOI_EN:
3594 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3597 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3599 case MSR_KVM_POLL_CONTROL:
3600 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3603 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3605 case MSR_IA32_P5_MC_ADDR:
3606 case MSR_IA32_P5_MC_TYPE:
3607 case MSR_IA32_MCG_CAP:
3608 case MSR_IA32_MCG_CTL:
3609 case MSR_IA32_MCG_STATUS:
3610 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3611 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3612 msr_info->host_initiated);
3614 if (!msr_info->host_initiated &&
3615 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3617 msr_info->data = vcpu->arch.ia32_xss;
3619 case MSR_K7_CLK_CTL:
3621 * Provide expected ramp-up count for K7. All other
3622 * are set to zero, indicating minimum divisors for
3625 * This prevents guest kernels on AMD host with CPU
3626 * type 6, model 8 and higher from exploding due to
3627 * the rdmsr failing.
3629 msr_info->data = 0x20000000;
3631 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3632 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3633 case HV_X64_MSR_SYNDBG_OPTIONS:
3634 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3635 case HV_X64_MSR_CRASH_CTL:
3636 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3637 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3638 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3639 case HV_X64_MSR_TSC_EMULATION_STATUS:
3640 return kvm_hv_get_msr_common(vcpu,
3641 msr_info->index, &msr_info->data,
3642 msr_info->host_initiated);
3643 case MSR_IA32_BBL_CR_CTL3:
3644 /* This legacy MSR exists but isn't fully documented in current
3645 * silicon. It is however accessed by winxp in very narrow
3646 * scenarios where it sets bit #19, itself documented as
3647 * a "reserved" bit. Best effort attempt to source coherent
3648 * read data here should the balance of the register be
3649 * interpreted by the guest:
3651 * L2 cache control register 3: 64GB range, 256KB size,
3652 * enabled, latency 0x1, configured
3654 msr_info->data = 0xbe702111;
3656 case MSR_AMD64_OSVW_ID_LENGTH:
3657 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3659 msr_info->data = vcpu->arch.osvw.length;
3661 case MSR_AMD64_OSVW_STATUS:
3662 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3664 msr_info->data = vcpu->arch.osvw.status;
3666 case MSR_PLATFORM_INFO:
3667 if (!msr_info->host_initiated &&
3668 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3670 msr_info->data = vcpu->arch.msr_platform_info;
3672 case MSR_MISC_FEATURES_ENABLES:
3673 msr_info->data = vcpu->arch.msr_misc_features_enables;
3676 msr_info->data = vcpu->arch.msr_hwcr;
3679 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3680 return kvm_pmu_get_msr(vcpu, msr_info);
3681 return KVM_MSR_RET_INVALID;
3685 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3688 * Read or write a bunch of msrs. All parameters are kernel addresses.
3690 * @return number of msrs set successfully.
3692 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3693 struct kvm_msr_entry *entries,
3694 int (*do_msr)(struct kvm_vcpu *vcpu,
3695 unsigned index, u64 *data))
3699 for (i = 0; i < msrs->nmsrs; ++i)
3700 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3707 * Read or write a bunch of msrs. Parameters are user addresses.
3709 * @return number of msrs set successfully.
3711 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3712 int (*do_msr)(struct kvm_vcpu *vcpu,
3713 unsigned index, u64 *data),
3716 struct kvm_msrs msrs;
3717 struct kvm_msr_entry *entries;
3722 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3726 if (msrs.nmsrs >= MAX_IO_MSRS)
3729 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3730 entries = memdup_user(user_msrs->entries, size);
3731 if (IS_ERR(entries)) {
3732 r = PTR_ERR(entries);
3736 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3741 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3752 static inline bool kvm_can_mwait_in_guest(void)
3754 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3755 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3756 boot_cpu_has(X86_FEATURE_ARAT);
3759 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3760 struct kvm_cpuid2 __user *cpuid_arg)
3762 struct kvm_cpuid2 cpuid;
3766 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3769 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3774 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3780 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3785 case KVM_CAP_IRQCHIP:
3787 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3788 case KVM_CAP_SET_TSS_ADDR:
3789 case KVM_CAP_EXT_CPUID:
3790 case KVM_CAP_EXT_EMUL_CPUID:
3791 case KVM_CAP_CLOCKSOURCE:
3793 case KVM_CAP_NOP_IO_DELAY:
3794 case KVM_CAP_MP_STATE:
3795 case KVM_CAP_SYNC_MMU:
3796 case KVM_CAP_USER_NMI:
3797 case KVM_CAP_REINJECT_CONTROL:
3798 case KVM_CAP_IRQ_INJECT_STATUS:
3799 case KVM_CAP_IOEVENTFD:
3800 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3802 case KVM_CAP_PIT_STATE2:
3803 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3804 case KVM_CAP_VCPU_EVENTS:
3805 case KVM_CAP_HYPERV:
3806 case KVM_CAP_HYPERV_VAPIC:
3807 case KVM_CAP_HYPERV_SPIN:
3808 case KVM_CAP_HYPERV_SYNIC:
3809 case KVM_CAP_HYPERV_SYNIC2:
3810 case KVM_CAP_HYPERV_VP_INDEX:
3811 case KVM_CAP_HYPERV_EVENTFD:
3812 case KVM_CAP_HYPERV_TLBFLUSH:
3813 case KVM_CAP_HYPERV_SEND_IPI:
3814 case KVM_CAP_HYPERV_CPUID:
3815 case KVM_CAP_SYS_HYPERV_CPUID:
3816 case KVM_CAP_PCI_SEGMENT:
3817 case KVM_CAP_DEBUGREGS:
3818 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3820 case KVM_CAP_ASYNC_PF:
3821 case KVM_CAP_ASYNC_PF_INT:
3822 case KVM_CAP_GET_TSC_KHZ:
3823 case KVM_CAP_KVMCLOCK_CTRL:
3824 case KVM_CAP_READONLY_MEM:
3825 case KVM_CAP_HYPERV_TIME:
3826 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3827 case KVM_CAP_TSC_DEADLINE_TIMER:
3828 case KVM_CAP_DISABLE_QUIRKS:
3829 case KVM_CAP_SET_BOOT_CPU_ID:
3830 case KVM_CAP_SPLIT_IRQCHIP:
3831 case KVM_CAP_IMMEDIATE_EXIT:
3832 case KVM_CAP_PMU_EVENT_FILTER:
3833 case KVM_CAP_GET_MSR_FEATURES:
3834 case KVM_CAP_MSR_PLATFORM_INFO:
3835 case KVM_CAP_EXCEPTION_PAYLOAD:
3836 case KVM_CAP_SET_GUEST_DEBUG:
3837 case KVM_CAP_LAST_CPU:
3838 case KVM_CAP_X86_USER_SPACE_MSR:
3839 case KVM_CAP_X86_MSR_FILTER:
3840 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3841 #ifdef CONFIG_X86_SGX_KVM
3842 case KVM_CAP_SGX_ATTRIBUTE:
3844 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
3847 case KVM_CAP_SET_GUEST_DEBUG2:
3848 return KVM_GUESTDBG_VALID_MASK;
3849 #ifdef CONFIG_KVM_XEN
3850 case KVM_CAP_XEN_HVM:
3851 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
3852 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
3853 KVM_XEN_HVM_CONFIG_SHARED_INFO;
3854 if (sched_info_on())
3855 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
3858 case KVM_CAP_SYNC_REGS:
3859 r = KVM_SYNC_X86_VALID_FIELDS;
3861 case KVM_CAP_ADJUST_CLOCK:
3862 r = KVM_CLOCK_TSC_STABLE;
3864 case KVM_CAP_X86_DISABLE_EXITS:
3865 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3866 KVM_X86_DISABLE_EXITS_CSTATE;
3867 if(kvm_can_mwait_in_guest())
3868 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3870 case KVM_CAP_X86_SMM:
3871 /* SMBASE is usually relocated above 1M on modern chipsets,
3872 * and SMM handlers might indeed rely on 4G segment limits,
3873 * so do not report SMM to be available if real mode is
3874 * emulated via vm86 mode. Still, do not go to great lengths
3875 * to avoid userspace's usage of the feature, because it is a
3876 * fringe case that is not enabled except via specific settings
3877 * of the module parameters.
3879 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
3882 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
3884 case KVM_CAP_NR_VCPUS:
3885 r = KVM_SOFT_MAX_VCPUS;
3887 case KVM_CAP_MAX_VCPUS:
3890 case KVM_CAP_MAX_VCPU_ID:
3891 r = KVM_MAX_VCPU_ID;
3893 case KVM_CAP_PV_MMU: /* obsolete */
3897 r = KVM_MAX_MCE_BANKS;
3900 r = boot_cpu_has(X86_FEATURE_XSAVE);
3902 case KVM_CAP_TSC_CONTROL:
3903 r = kvm_has_tsc_control;
3905 case KVM_CAP_X2APIC_API:
3906 r = KVM_X2APIC_API_VALID_FLAGS;
3908 case KVM_CAP_NESTED_STATE:
3909 r = kvm_x86_ops.nested_ops->get_state ?
3910 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3912 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3913 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3915 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3916 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3918 case KVM_CAP_SMALLER_MAXPHYADDR:
3919 r = (int) allow_smaller_maxphyaddr;
3921 case KVM_CAP_STEAL_TIME:
3922 r = sched_info_on();
3924 case KVM_CAP_X86_BUS_LOCK_EXIT:
3925 if (kvm_has_bus_lock_exit)
3926 r = KVM_BUS_LOCK_DETECTION_OFF |
3927 KVM_BUS_LOCK_DETECTION_EXIT;
3938 long kvm_arch_dev_ioctl(struct file *filp,
3939 unsigned int ioctl, unsigned long arg)
3941 void __user *argp = (void __user *)arg;
3945 case KVM_GET_MSR_INDEX_LIST: {
3946 struct kvm_msr_list __user *user_msr_list = argp;
3947 struct kvm_msr_list msr_list;
3951 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3954 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3955 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3958 if (n < msr_list.nmsrs)
3961 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3962 num_msrs_to_save * sizeof(u32)))
3964 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3966 num_emulated_msrs * sizeof(u32)))
3971 case KVM_GET_SUPPORTED_CPUID:
3972 case KVM_GET_EMULATED_CPUID: {
3973 struct kvm_cpuid2 __user *cpuid_arg = argp;
3974 struct kvm_cpuid2 cpuid;
3977 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3980 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3986 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3991 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3993 if (copy_to_user(argp, &kvm_mce_cap_supported,
3994 sizeof(kvm_mce_cap_supported)))
3998 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3999 struct kvm_msr_list __user *user_msr_list = argp;
4000 struct kvm_msr_list msr_list;
4004 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4007 msr_list.nmsrs = num_msr_based_features;
4008 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4011 if (n < msr_list.nmsrs)
4014 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4015 num_msr_based_features * sizeof(u32)))
4021 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4023 case KVM_GET_SUPPORTED_HV_CPUID:
4024 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4034 static void wbinvd_ipi(void *garbage)
4039 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4041 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4044 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4046 /* Address WBINVD may be executed by guest */
4047 if (need_emulate_wbinvd(vcpu)) {
4048 if (static_call(kvm_x86_has_wbinvd_exit)())
4049 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4050 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4051 smp_call_function_single(vcpu->cpu,
4052 wbinvd_ipi, NULL, 1);
4055 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4057 /* Save host pkru register if supported */
4058 vcpu->arch.host_pkru = read_pkru();
4060 /* Apply any externally detected TSC adjustments (due to suspend) */
4061 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4062 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4063 vcpu->arch.tsc_offset_adjustment = 0;
4064 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4067 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4068 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4069 rdtsc() - vcpu->arch.last_host_tsc;
4071 mark_tsc_unstable("KVM discovered backwards TSC");
4073 if (kvm_check_tsc_unstable()) {
4074 u64 offset = kvm_compute_tsc_offset(vcpu,
4075 vcpu->arch.last_guest_tsc);
4076 kvm_vcpu_write_tsc_offset(vcpu, offset);
4077 vcpu->arch.tsc_catchup = 1;
4080 if (kvm_lapic_hv_timer_in_use(vcpu))
4081 kvm_lapic_restart_hv_timer(vcpu);
4084 * On a host with synchronized TSC, there is no need to update
4085 * kvmclock on vcpu->cpu migration
4087 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4088 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4089 if (vcpu->cpu != cpu)
4090 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4094 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4097 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4099 struct kvm_host_map map;
4100 struct kvm_steal_time *st;
4102 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4105 if (vcpu->arch.st.preempted)
4108 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4109 &vcpu->arch.st.cache, true))
4113 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4115 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4117 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4120 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4124 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4125 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4128 * Take the srcu lock as memslots will be accessed to check the gfn
4129 * cache generation against the memslots generation.
4131 idx = srcu_read_lock(&vcpu->kvm->srcu);
4132 if (kvm_xen_msr_enabled(vcpu->kvm))
4133 kvm_xen_runstate_set_preempted(vcpu);
4135 kvm_steal_time_set_preempted(vcpu);
4136 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4138 static_call(kvm_x86_vcpu_put)(vcpu);
4139 vcpu->arch.last_host_tsc = rdtsc();
4141 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4142 * on every vmexit, but if not, we might have a stale dr6 from the
4143 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4148 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4149 struct kvm_lapic_state *s)
4151 if (vcpu->arch.apicv_active)
4152 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4154 return kvm_apic_get_state(vcpu, s);
4157 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4158 struct kvm_lapic_state *s)
4162 r = kvm_apic_set_state(vcpu, s);
4165 update_cr8_intercept(vcpu);
4170 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4173 * We can accept userspace's request for interrupt injection
4174 * as long as we have a place to store the interrupt number.
4175 * The actual injection will happen when the CPU is able to
4176 * deliver the interrupt.
4178 if (kvm_cpu_has_extint(vcpu))
4181 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4182 return (!lapic_in_kernel(vcpu) ||
4183 kvm_apic_accept_pic_intr(vcpu));
4186 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4188 return kvm_arch_interrupt_allowed(vcpu) &&
4189 kvm_cpu_accept_dm_intr(vcpu);
4192 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4193 struct kvm_interrupt *irq)
4195 if (irq->irq >= KVM_NR_INTERRUPTS)
4198 if (!irqchip_in_kernel(vcpu->kvm)) {
4199 kvm_queue_interrupt(vcpu, irq->irq, false);
4200 kvm_make_request(KVM_REQ_EVENT, vcpu);
4205 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4206 * fail for in-kernel 8259.
4208 if (pic_in_kernel(vcpu->kvm))
4211 if (vcpu->arch.pending_external_vector != -1)
4214 vcpu->arch.pending_external_vector = irq->irq;
4215 kvm_make_request(KVM_REQ_EVENT, vcpu);
4219 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4221 kvm_inject_nmi(vcpu);
4226 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4228 kvm_make_request(KVM_REQ_SMI, vcpu);
4233 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4234 struct kvm_tpr_access_ctl *tac)
4238 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4242 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4246 unsigned bank_num = mcg_cap & 0xff, bank;
4249 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4251 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4254 vcpu->arch.mcg_cap = mcg_cap;
4255 /* Init IA32_MCG_CTL to all 1s */
4256 if (mcg_cap & MCG_CTL_P)
4257 vcpu->arch.mcg_ctl = ~(u64)0;
4258 /* Init IA32_MCi_CTL to all 1s */
4259 for (bank = 0; bank < bank_num; bank++)
4260 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4262 static_call(kvm_x86_setup_mce)(vcpu);
4267 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4268 struct kvm_x86_mce *mce)
4270 u64 mcg_cap = vcpu->arch.mcg_cap;
4271 unsigned bank_num = mcg_cap & 0xff;
4272 u64 *banks = vcpu->arch.mce_banks;
4274 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4277 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4278 * reporting is disabled
4280 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4281 vcpu->arch.mcg_ctl != ~(u64)0)
4283 banks += 4 * mce->bank;
4285 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4286 * reporting is disabled for the bank
4288 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4290 if (mce->status & MCI_STATUS_UC) {
4291 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4292 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4293 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4296 if (banks[1] & MCI_STATUS_VAL)
4297 mce->status |= MCI_STATUS_OVER;
4298 banks[2] = mce->addr;
4299 banks[3] = mce->misc;
4300 vcpu->arch.mcg_status = mce->mcg_status;
4301 banks[1] = mce->status;
4302 kvm_queue_exception(vcpu, MC_VECTOR);
4303 } else if (!(banks[1] & MCI_STATUS_VAL)
4304 || !(banks[1] & MCI_STATUS_UC)) {
4305 if (banks[1] & MCI_STATUS_VAL)
4306 mce->status |= MCI_STATUS_OVER;
4307 banks[2] = mce->addr;
4308 banks[3] = mce->misc;
4309 banks[1] = mce->status;
4311 banks[1] |= MCI_STATUS_OVER;
4315 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4316 struct kvm_vcpu_events *events)
4320 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4324 * In guest mode, payload delivery should be deferred,
4325 * so that the L1 hypervisor can intercept #PF before
4326 * CR2 is modified (or intercept #DB before DR6 is
4327 * modified under nVMX). Unless the per-VM capability,
4328 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4329 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4330 * opportunistically defer the exception payload, deliver it if the
4331 * capability hasn't been requested before processing a
4332 * KVM_GET_VCPU_EVENTS.
4334 if (!vcpu->kvm->arch.exception_payload_enabled &&
4335 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4336 kvm_deliver_exception_payload(vcpu);
4339 * The API doesn't provide the instruction length for software
4340 * exceptions, so don't report them. As long as the guest RIP
4341 * isn't advanced, we should expect to encounter the exception
4344 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4345 events->exception.injected = 0;
4346 events->exception.pending = 0;
4348 events->exception.injected = vcpu->arch.exception.injected;
4349 events->exception.pending = vcpu->arch.exception.pending;
4351 * For ABI compatibility, deliberately conflate
4352 * pending and injected exceptions when
4353 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4355 if (!vcpu->kvm->arch.exception_payload_enabled)
4356 events->exception.injected |=
4357 vcpu->arch.exception.pending;
4359 events->exception.nr = vcpu->arch.exception.nr;
4360 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4361 events->exception.error_code = vcpu->arch.exception.error_code;
4362 events->exception_has_payload = vcpu->arch.exception.has_payload;
4363 events->exception_payload = vcpu->arch.exception.payload;
4365 events->interrupt.injected =
4366 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4367 events->interrupt.nr = vcpu->arch.interrupt.nr;
4368 events->interrupt.soft = 0;
4369 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4371 events->nmi.injected = vcpu->arch.nmi_injected;
4372 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4373 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4374 events->nmi.pad = 0;
4376 events->sipi_vector = 0; /* never valid when reporting to user space */
4378 events->smi.smm = is_smm(vcpu);
4379 events->smi.pending = vcpu->arch.smi_pending;
4380 events->smi.smm_inside_nmi =
4381 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4382 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4384 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4385 | KVM_VCPUEVENT_VALID_SHADOW
4386 | KVM_VCPUEVENT_VALID_SMM);
4387 if (vcpu->kvm->arch.exception_payload_enabled)
4388 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4390 memset(&events->reserved, 0, sizeof(events->reserved));
4393 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4395 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4396 struct kvm_vcpu_events *events)
4398 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4399 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4400 | KVM_VCPUEVENT_VALID_SHADOW
4401 | KVM_VCPUEVENT_VALID_SMM
4402 | KVM_VCPUEVENT_VALID_PAYLOAD))
4405 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4406 if (!vcpu->kvm->arch.exception_payload_enabled)
4408 if (events->exception.pending)
4409 events->exception.injected = 0;
4411 events->exception_has_payload = 0;
4413 events->exception.pending = 0;
4414 events->exception_has_payload = 0;
4417 if ((events->exception.injected || events->exception.pending) &&
4418 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4421 /* INITs are latched while in SMM */
4422 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4423 (events->smi.smm || events->smi.pending) &&
4424 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4428 vcpu->arch.exception.injected = events->exception.injected;
4429 vcpu->arch.exception.pending = events->exception.pending;
4430 vcpu->arch.exception.nr = events->exception.nr;
4431 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4432 vcpu->arch.exception.error_code = events->exception.error_code;
4433 vcpu->arch.exception.has_payload = events->exception_has_payload;
4434 vcpu->arch.exception.payload = events->exception_payload;
4436 vcpu->arch.interrupt.injected = events->interrupt.injected;
4437 vcpu->arch.interrupt.nr = events->interrupt.nr;
4438 vcpu->arch.interrupt.soft = events->interrupt.soft;
4439 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4440 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4441 events->interrupt.shadow);
4443 vcpu->arch.nmi_injected = events->nmi.injected;
4444 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4445 vcpu->arch.nmi_pending = events->nmi.pending;
4446 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4448 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4449 lapic_in_kernel(vcpu))
4450 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4452 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4453 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4454 if (events->smi.smm)
4455 vcpu->arch.hflags |= HF_SMM_MASK;
4457 vcpu->arch.hflags &= ~HF_SMM_MASK;
4458 kvm_smm_changed(vcpu);
4461 vcpu->arch.smi_pending = events->smi.pending;
4463 if (events->smi.smm) {
4464 if (events->smi.smm_inside_nmi)
4465 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4467 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4470 if (lapic_in_kernel(vcpu)) {
4471 if (events->smi.latched_init)
4472 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4474 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4478 kvm_make_request(KVM_REQ_EVENT, vcpu);
4483 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4484 struct kvm_debugregs *dbgregs)
4488 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4489 kvm_get_dr(vcpu, 6, &val);
4491 dbgregs->dr7 = vcpu->arch.dr7;
4493 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4496 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4497 struct kvm_debugregs *dbgregs)
4502 if (!kvm_dr6_valid(dbgregs->dr6))
4504 if (!kvm_dr7_valid(dbgregs->dr7))
4507 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4508 kvm_update_dr0123(vcpu);
4509 vcpu->arch.dr6 = dbgregs->dr6;
4510 vcpu->arch.dr7 = dbgregs->dr7;
4511 kvm_update_dr7(vcpu);
4516 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4518 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4520 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4521 u64 xstate_bv = xsave->header.xfeatures;
4525 * Copy legacy XSAVE area, to avoid complications with CPUID
4526 * leaves 0 and 1 in the loop below.
4528 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4531 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4532 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4535 * Copy each region from the possibly compacted offset to the
4536 * non-compacted offset.
4538 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4540 u64 xfeature_mask = valid & -valid;
4541 int xfeature_nr = fls64(xfeature_mask) - 1;
4542 void *src = get_xsave_addr(xsave, xfeature_nr);
4545 u32 size, offset, ecx, edx;
4546 cpuid_count(XSTATE_CPUID, xfeature_nr,
4547 &size, &offset, &ecx, &edx);
4548 if (xfeature_nr == XFEATURE_PKRU)
4549 memcpy(dest + offset, &vcpu->arch.pkru,
4550 sizeof(vcpu->arch.pkru));
4552 memcpy(dest + offset, src, size);
4556 valid -= xfeature_mask;
4560 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4562 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4563 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4567 * Copy legacy XSAVE area, to avoid complications with CPUID
4568 * leaves 0 and 1 in the loop below.
4570 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4572 /* Set XSTATE_BV and possibly XCOMP_BV. */
4573 xsave->header.xfeatures = xstate_bv;
4574 if (boot_cpu_has(X86_FEATURE_XSAVES))
4575 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4578 * Copy each region from the non-compacted offset to the
4579 * possibly compacted offset.
4581 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4583 u64 xfeature_mask = valid & -valid;
4584 int xfeature_nr = fls64(xfeature_mask) - 1;
4585 void *dest = get_xsave_addr(xsave, xfeature_nr);
4588 u32 size, offset, ecx, edx;
4589 cpuid_count(XSTATE_CPUID, xfeature_nr,
4590 &size, &offset, &ecx, &edx);
4591 if (xfeature_nr == XFEATURE_PKRU)
4592 memcpy(&vcpu->arch.pkru, src + offset,
4593 sizeof(vcpu->arch.pkru));
4595 memcpy(dest, src + offset, size);
4598 valid -= xfeature_mask;
4602 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4603 struct kvm_xsave *guest_xsave)
4605 if (!vcpu->arch.guest_fpu)
4608 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4609 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4610 fill_xsave((u8 *) guest_xsave->region, vcpu);
4612 memcpy(guest_xsave->region,
4613 &vcpu->arch.guest_fpu->state.fxsave,
4614 sizeof(struct fxregs_state));
4615 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4616 XFEATURE_MASK_FPSSE;
4620 #define XSAVE_MXCSR_OFFSET 24
4622 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4623 struct kvm_xsave *guest_xsave)
4628 if (!vcpu->arch.guest_fpu)
4631 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4632 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4634 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4636 * Here we allow setting states that are not present in
4637 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4638 * with old userspace.
4640 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4642 load_xsave(vcpu, (u8 *)guest_xsave->region);
4644 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4645 mxcsr & ~mxcsr_feature_mask)
4647 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4648 guest_xsave->region, sizeof(struct fxregs_state));
4653 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4654 struct kvm_xcrs *guest_xcrs)
4656 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4657 guest_xcrs->nr_xcrs = 0;
4661 guest_xcrs->nr_xcrs = 1;
4662 guest_xcrs->flags = 0;
4663 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4664 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4667 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4668 struct kvm_xcrs *guest_xcrs)
4672 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4675 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4678 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4679 /* Only support XCR0 currently */
4680 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4681 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4682 guest_xcrs->xcrs[i].value);
4691 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4692 * stopped by the hypervisor. This function will be called from the host only.
4693 * EINVAL is returned when the host attempts to set the flag for a guest that
4694 * does not support pv clocks.
4696 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4698 if (!vcpu->arch.pv_time_enabled)
4700 vcpu->arch.pvclock_set_guest_stopped_request = true;
4701 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4705 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4706 struct kvm_enable_cap *cap)
4709 uint16_t vmcs_version;
4710 void __user *user_ptr;
4716 case KVM_CAP_HYPERV_SYNIC2:
4721 case KVM_CAP_HYPERV_SYNIC:
4722 if (!irqchip_in_kernel(vcpu->kvm))
4724 return kvm_hv_activate_synic(vcpu, cap->cap ==
4725 KVM_CAP_HYPERV_SYNIC2);
4726 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4727 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4729 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4731 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4732 if (copy_to_user(user_ptr, &vmcs_version,
4733 sizeof(vmcs_version)))
4737 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4738 if (!kvm_x86_ops.enable_direct_tlbflush)
4741 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4743 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4744 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4745 if (vcpu->arch.pv_cpuid.enforce)
4746 kvm_update_pv_runtime(vcpu);
4754 long kvm_arch_vcpu_ioctl(struct file *filp,
4755 unsigned int ioctl, unsigned long arg)
4757 struct kvm_vcpu *vcpu = filp->private_data;
4758 void __user *argp = (void __user *)arg;
4761 struct kvm_lapic_state *lapic;
4762 struct kvm_xsave *xsave;
4763 struct kvm_xcrs *xcrs;
4771 case KVM_GET_LAPIC: {
4773 if (!lapic_in_kernel(vcpu))
4775 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4776 GFP_KERNEL_ACCOUNT);
4781 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4785 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4790 case KVM_SET_LAPIC: {
4792 if (!lapic_in_kernel(vcpu))
4794 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4795 if (IS_ERR(u.lapic)) {
4796 r = PTR_ERR(u.lapic);
4800 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4803 case KVM_INTERRUPT: {
4804 struct kvm_interrupt irq;
4807 if (copy_from_user(&irq, argp, sizeof(irq)))
4809 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4813 r = kvm_vcpu_ioctl_nmi(vcpu);
4817 r = kvm_vcpu_ioctl_smi(vcpu);
4820 case KVM_SET_CPUID: {
4821 struct kvm_cpuid __user *cpuid_arg = argp;
4822 struct kvm_cpuid cpuid;
4825 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4827 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4830 case KVM_SET_CPUID2: {
4831 struct kvm_cpuid2 __user *cpuid_arg = argp;
4832 struct kvm_cpuid2 cpuid;
4835 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4837 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4838 cpuid_arg->entries);
4841 case KVM_GET_CPUID2: {
4842 struct kvm_cpuid2 __user *cpuid_arg = argp;
4843 struct kvm_cpuid2 cpuid;
4846 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4848 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4849 cpuid_arg->entries);
4853 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4858 case KVM_GET_MSRS: {
4859 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4860 r = msr_io(vcpu, argp, do_get_msr, 1);
4861 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4864 case KVM_SET_MSRS: {
4865 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4866 r = msr_io(vcpu, argp, do_set_msr, 0);
4867 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4870 case KVM_TPR_ACCESS_REPORTING: {
4871 struct kvm_tpr_access_ctl tac;
4874 if (copy_from_user(&tac, argp, sizeof(tac)))
4876 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4880 if (copy_to_user(argp, &tac, sizeof(tac)))
4885 case KVM_SET_VAPIC_ADDR: {
4886 struct kvm_vapic_addr va;
4890 if (!lapic_in_kernel(vcpu))
4893 if (copy_from_user(&va, argp, sizeof(va)))
4895 idx = srcu_read_lock(&vcpu->kvm->srcu);
4896 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4897 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4900 case KVM_X86_SETUP_MCE: {
4904 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4906 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4909 case KVM_X86_SET_MCE: {
4910 struct kvm_x86_mce mce;
4913 if (copy_from_user(&mce, argp, sizeof(mce)))
4915 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4918 case KVM_GET_VCPU_EVENTS: {
4919 struct kvm_vcpu_events events;
4921 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4924 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4929 case KVM_SET_VCPU_EVENTS: {
4930 struct kvm_vcpu_events events;
4933 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4936 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4939 case KVM_GET_DEBUGREGS: {
4940 struct kvm_debugregs dbgregs;
4942 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4945 if (copy_to_user(argp, &dbgregs,
4946 sizeof(struct kvm_debugregs)))
4951 case KVM_SET_DEBUGREGS: {
4952 struct kvm_debugregs dbgregs;
4955 if (copy_from_user(&dbgregs, argp,
4956 sizeof(struct kvm_debugregs)))
4959 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4962 case KVM_GET_XSAVE: {
4963 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4968 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4971 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4976 case KVM_SET_XSAVE: {
4977 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4978 if (IS_ERR(u.xsave)) {
4979 r = PTR_ERR(u.xsave);
4983 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4986 case KVM_GET_XCRS: {
4987 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4992 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4995 if (copy_to_user(argp, u.xcrs,
4996 sizeof(struct kvm_xcrs)))
5001 case KVM_SET_XCRS: {
5002 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5003 if (IS_ERR(u.xcrs)) {
5004 r = PTR_ERR(u.xcrs);
5008 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5011 case KVM_SET_TSC_KHZ: {
5015 user_tsc_khz = (u32)arg;
5017 if (kvm_has_tsc_control &&
5018 user_tsc_khz >= kvm_max_guest_tsc_khz)
5021 if (user_tsc_khz == 0)
5022 user_tsc_khz = tsc_khz;
5024 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5029 case KVM_GET_TSC_KHZ: {
5030 r = vcpu->arch.virtual_tsc_khz;
5033 case KVM_KVMCLOCK_CTRL: {
5034 r = kvm_set_guest_paused(vcpu);
5037 case KVM_ENABLE_CAP: {
5038 struct kvm_enable_cap cap;
5041 if (copy_from_user(&cap, argp, sizeof(cap)))
5043 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5046 case KVM_GET_NESTED_STATE: {
5047 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5051 if (!kvm_x86_ops.nested_ops->get_state)
5054 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5056 if (get_user(user_data_size, &user_kvm_nested_state->size))
5059 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5064 if (r > user_data_size) {
5065 if (put_user(r, &user_kvm_nested_state->size))
5075 case KVM_SET_NESTED_STATE: {
5076 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5077 struct kvm_nested_state kvm_state;
5081 if (!kvm_x86_ops.nested_ops->set_state)
5085 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5089 if (kvm_state.size < sizeof(kvm_state))
5092 if (kvm_state.flags &
5093 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5094 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5095 | KVM_STATE_NESTED_GIF_SET))
5098 /* nested_run_pending implies guest_mode. */
5099 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5100 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5103 idx = srcu_read_lock(&vcpu->kvm->srcu);
5104 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5105 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5108 case KVM_GET_SUPPORTED_HV_CPUID:
5109 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5111 #ifdef CONFIG_KVM_XEN
5112 case KVM_XEN_VCPU_GET_ATTR: {
5113 struct kvm_xen_vcpu_attr xva;
5116 if (copy_from_user(&xva, argp, sizeof(xva)))
5118 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5119 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5123 case KVM_XEN_VCPU_SET_ATTR: {
5124 struct kvm_xen_vcpu_attr xva;
5127 if (copy_from_user(&xva, argp, sizeof(xva)))
5129 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5143 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5145 return VM_FAULT_SIGBUS;
5148 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5152 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5154 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5158 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5161 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5164 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5165 unsigned long kvm_nr_mmu_pages)
5167 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5170 mutex_lock(&kvm->slots_lock);
5172 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5173 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5175 mutex_unlock(&kvm->slots_lock);
5179 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5181 return kvm->arch.n_max_mmu_pages;
5184 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5186 struct kvm_pic *pic = kvm->arch.vpic;
5190 switch (chip->chip_id) {
5191 case KVM_IRQCHIP_PIC_MASTER:
5192 memcpy(&chip->chip.pic, &pic->pics[0],
5193 sizeof(struct kvm_pic_state));
5195 case KVM_IRQCHIP_PIC_SLAVE:
5196 memcpy(&chip->chip.pic, &pic->pics[1],
5197 sizeof(struct kvm_pic_state));
5199 case KVM_IRQCHIP_IOAPIC:
5200 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5209 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5211 struct kvm_pic *pic = kvm->arch.vpic;
5215 switch (chip->chip_id) {
5216 case KVM_IRQCHIP_PIC_MASTER:
5217 spin_lock(&pic->lock);
5218 memcpy(&pic->pics[0], &chip->chip.pic,
5219 sizeof(struct kvm_pic_state));
5220 spin_unlock(&pic->lock);
5222 case KVM_IRQCHIP_PIC_SLAVE:
5223 spin_lock(&pic->lock);
5224 memcpy(&pic->pics[1], &chip->chip.pic,
5225 sizeof(struct kvm_pic_state));
5226 spin_unlock(&pic->lock);
5228 case KVM_IRQCHIP_IOAPIC:
5229 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5235 kvm_pic_update_irq(pic);
5239 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5241 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5243 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5245 mutex_lock(&kps->lock);
5246 memcpy(ps, &kps->channels, sizeof(*ps));
5247 mutex_unlock(&kps->lock);
5251 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5254 struct kvm_pit *pit = kvm->arch.vpit;
5256 mutex_lock(&pit->pit_state.lock);
5257 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5258 for (i = 0; i < 3; i++)
5259 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5260 mutex_unlock(&pit->pit_state.lock);
5264 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5266 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5267 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5268 sizeof(ps->channels));
5269 ps->flags = kvm->arch.vpit->pit_state.flags;
5270 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5271 memset(&ps->reserved, 0, sizeof(ps->reserved));
5275 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5279 u32 prev_legacy, cur_legacy;
5280 struct kvm_pit *pit = kvm->arch.vpit;
5282 mutex_lock(&pit->pit_state.lock);
5283 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5284 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5285 if (!prev_legacy && cur_legacy)
5287 memcpy(&pit->pit_state.channels, &ps->channels,
5288 sizeof(pit->pit_state.channels));
5289 pit->pit_state.flags = ps->flags;
5290 for (i = 0; i < 3; i++)
5291 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5293 mutex_unlock(&pit->pit_state.lock);
5297 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5298 struct kvm_reinject_control *control)
5300 struct kvm_pit *pit = kvm->arch.vpit;
5302 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5303 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5304 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5306 mutex_lock(&pit->pit_state.lock);
5307 kvm_pit_set_reinject(pit, control->pit_reinject);
5308 mutex_unlock(&pit->pit_state.lock);
5313 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5317 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5318 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5319 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5322 struct kvm_vcpu *vcpu;
5325 kvm_for_each_vcpu(i, vcpu, kvm)
5326 kvm_vcpu_kick(vcpu);
5329 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5332 if (!irqchip_in_kernel(kvm))
5335 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5336 irq_event->irq, irq_event->level,
5341 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5342 struct kvm_enable_cap *cap)
5350 case KVM_CAP_DISABLE_QUIRKS:
5351 kvm->arch.disabled_quirks = cap->args[0];
5354 case KVM_CAP_SPLIT_IRQCHIP: {
5355 mutex_lock(&kvm->lock);
5357 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5358 goto split_irqchip_unlock;
5360 if (irqchip_in_kernel(kvm))
5361 goto split_irqchip_unlock;
5362 if (kvm->created_vcpus)
5363 goto split_irqchip_unlock;
5364 r = kvm_setup_empty_irq_routing(kvm);
5366 goto split_irqchip_unlock;
5367 /* Pairs with irqchip_in_kernel. */
5369 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5370 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5372 split_irqchip_unlock:
5373 mutex_unlock(&kvm->lock);
5376 case KVM_CAP_X2APIC_API:
5378 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5381 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5382 kvm->arch.x2apic_format = true;
5383 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5384 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5388 case KVM_CAP_X86_DISABLE_EXITS:
5390 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5393 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5394 kvm_can_mwait_in_guest())
5395 kvm->arch.mwait_in_guest = true;
5396 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5397 kvm->arch.hlt_in_guest = true;
5398 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5399 kvm->arch.pause_in_guest = true;
5400 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5401 kvm->arch.cstate_in_guest = true;
5404 case KVM_CAP_MSR_PLATFORM_INFO:
5405 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5408 case KVM_CAP_EXCEPTION_PAYLOAD:
5409 kvm->arch.exception_payload_enabled = cap->args[0];
5412 case KVM_CAP_X86_USER_SPACE_MSR:
5413 kvm->arch.user_space_msr_mask = cap->args[0];
5416 case KVM_CAP_X86_BUS_LOCK_EXIT:
5418 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5421 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5422 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5425 if (kvm_has_bus_lock_exit &&
5426 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5427 kvm->arch.bus_lock_detection_enabled = true;
5430 #ifdef CONFIG_X86_SGX_KVM
5431 case KVM_CAP_SGX_ATTRIBUTE: {
5432 unsigned long allowed_attributes = 0;
5434 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5438 /* KVM only supports the PROVISIONKEY privileged attribute. */
5439 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5440 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5441 kvm->arch.sgx_provisioning_allowed = true;
5447 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5449 if (kvm_x86_ops.vm_copy_enc_context_from)
5450 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5459 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5461 struct kvm_x86_msr_filter *msr_filter;
5463 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5467 msr_filter->default_allow = default_allow;
5471 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5478 for (i = 0; i < msr_filter->count; i++)
5479 kfree(msr_filter->ranges[i].bitmap);
5484 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5485 struct kvm_msr_filter_range *user_range)
5487 unsigned long *bitmap = NULL;
5490 if (!user_range->nmsrs)
5493 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5496 if (!user_range->flags)
5499 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5500 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5503 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5505 return PTR_ERR(bitmap);
5507 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5508 .flags = user_range->flags,
5509 .base = user_range->base,
5510 .nmsrs = user_range->nmsrs,
5514 msr_filter->count++;
5518 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5520 struct kvm_msr_filter __user *user_msr_filter = argp;
5521 struct kvm_x86_msr_filter *new_filter, *old_filter;
5522 struct kvm_msr_filter filter;
5528 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5531 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5532 empty &= !filter.ranges[i].nmsrs;
5534 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5535 if (empty && !default_allow)
5538 new_filter = kvm_alloc_msr_filter(default_allow);
5542 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5543 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5545 kvm_free_msr_filter(new_filter);
5550 mutex_lock(&kvm->lock);
5552 /* The per-VM filter is protected by kvm->lock... */
5553 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5555 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5556 synchronize_srcu(&kvm->srcu);
5558 kvm_free_msr_filter(old_filter);
5560 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5561 mutex_unlock(&kvm->lock);
5566 long kvm_arch_vm_ioctl(struct file *filp,
5567 unsigned int ioctl, unsigned long arg)
5569 struct kvm *kvm = filp->private_data;
5570 void __user *argp = (void __user *)arg;
5573 * This union makes it completely explicit to gcc-3.x
5574 * that these two variables' stack usage should be
5575 * combined, not added together.
5578 struct kvm_pit_state ps;
5579 struct kvm_pit_state2 ps2;
5580 struct kvm_pit_config pit_config;
5584 case KVM_SET_TSS_ADDR:
5585 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5587 case KVM_SET_IDENTITY_MAP_ADDR: {
5590 mutex_lock(&kvm->lock);
5592 if (kvm->created_vcpus)
5593 goto set_identity_unlock;
5595 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5596 goto set_identity_unlock;
5597 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5598 set_identity_unlock:
5599 mutex_unlock(&kvm->lock);
5602 case KVM_SET_NR_MMU_PAGES:
5603 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5605 case KVM_GET_NR_MMU_PAGES:
5606 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5608 case KVM_CREATE_IRQCHIP: {
5609 mutex_lock(&kvm->lock);
5612 if (irqchip_in_kernel(kvm))
5613 goto create_irqchip_unlock;
5616 if (kvm->created_vcpus)
5617 goto create_irqchip_unlock;
5619 r = kvm_pic_init(kvm);
5621 goto create_irqchip_unlock;
5623 r = kvm_ioapic_init(kvm);
5625 kvm_pic_destroy(kvm);
5626 goto create_irqchip_unlock;
5629 r = kvm_setup_default_irq_routing(kvm);
5631 kvm_ioapic_destroy(kvm);
5632 kvm_pic_destroy(kvm);
5633 goto create_irqchip_unlock;
5635 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5637 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5638 create_irqchip_unlock:
5639 mutex_unlock(&kvm->lock);
5642 case KVM_CREATE_PIT:
5643 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5645 case KVM_CREATE_PIT2:
5647 if (copy_from_user(&u.pit_config, argp,
5648 sizeof(struct kvm_pit_config)))
5651 mutex_lock(&kvm->lock);
5654 goto create_pit_unlock;
5656 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5660 mutex_unlock(&kvm->lock);
5662 case KVM_GET_IRQCHIP: {
5663 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5664 struct kvm_irqchip *chip;
5666 chip = memdup_user(argp, sizeof(*chip));
5673 if (!irqchip_kernel(kvm))
5674 goto get_irqchip_out;
5675 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5677 goto get_irqchip_out;
5679 if (copy_to_user(argp, chip, sizeof(*chip)))
5680 goto get_irqchip_out;
5686 case KVM_SET_IRQCHIP: {
5687 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5688 struct kvm_irqchip *chip;
5690 chip = memdup_user(argp, sizeof(*chip));
5697 if (!irqchip_kernel(kvm))
5698 goto set_irqchip_out;
5699 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5706 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5709 if (!kvm->arch.vpit)
5711 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5715 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5722 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5724 mutex_lock(&kvm->lock);
5726 if (!kvm->arch.vpit)
5728 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5730 mutex_unlock(&kvm->lock);
5733 case KVM_GET_PIT2: {
5735 if (!kvm->arch.vpit)
5737 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5741 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5746 case KVM_SET_PIT2: {
5748 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5750 mutex_lock(&kvm->lock);
5752 if (!kvm->arch.vpit)
5754 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5756 mutex_unlock(&kvm->lock);
5759 case KVM_REINJECT_CONTROL: {
5760 struct kvm_reinject_control control;
5762 if (copy_from_user(&control, argp, sizeof(control)))
5765 if (!kvm->arch.vpit)
5767 r = kvm_vm_ioctl_reinject(kvm, &control);
5770 case KVM_SET_BOOT_CPU_ID:
5772 mutex_lock(&kvm->lock);
5773 if (kvm->created_vcpus)
5776 kvm->arch.bsp_vcpu_id = arg;
5777 mutex_unlock(&kvm->lock);
5779 #ifdef CONFIG_KVM_XEN
5780 case KVM_XEN_HVM_CONFIG: {
5781 struct kvm_xen_hvm_config xhc;
5783 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5785 r = kvm_xen_hvm_config(kvm, &xhc);
5788 case KVM_XEN_HVM_GET_ATTR: {
5789 struct kvm_xen_hvm_attr xha;
5792 if (copy_from_user(&xha, argp, sizeof(xha)))
5794 r = kvm_xen_hvm_get_attr(kvm, &xha);
5795 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5799 case KVM_XEN_HVM_SET_ATTR: {
5800 struct kvm_xen_hvm_attr xha;
5803 if (copy_from_user(&xha, argp, sizeof(xha)))
5805 r = kvm_xen_hvm_set_attr(kvm, &xha);
5809 case KVM_SET_CLOCK: {
5810 struct kvm_arch *ka = &kvm->arch;
5811 struct kvm_clock_data user_ns;
5815 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5824 * TODO: userspace has to take care of races with VCPU_RUN, so
5825 * kvm_gen_update_masterclock() can be cut down to locked
5826 * pvclock_update_vm_gtod_copy().
5828 kvm_gen_update_masterclock(kvm);
5831 * This pairs with kvm_guest_time_update(): when masterclock is
5832 * in use, we use master_kernel_ns + kvmclock_offset to set
5833 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
5834 * is slightly ahead) here we risk going negative on unsigned
5835 * 'system_time' when 'user_ns.clock' is very small.
5837 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
5838 if (kvm->arch.use_master_clock)
5839 now_ns = ka->master_kernel_ns;
5841 now_ns = get_kvmclock_base_ns();
5842 ka->kvmclock_offset = user_ns.clock - now_ns;
5843 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
5845 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5848 case KVM_GET_CLOCK: {
5849 struct kvm_clock_data user_ns;
5852 now_ns = get_kvmclock_ns(kvm);
5853 user_ns.clock = now_ns;
5854 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5855 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5858 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5863 case KVM_MEMORY_ENCRYPT_OP: {
5865 if (kvm_x86_ops.mem_enc_op)
5866 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
5869 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5870 struct kvm_enc_region region;
5873 if (copy_from_user(®ion, argp, sizeof(region)))
5877 if (kvm_x86_ops.mem_enc_reg_region)
5878 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
5881 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5882 struct kvm_enc_region region;
5885 if (copy_from_user(®ion, argp, sizeof(region)))
5889 if (kvm_x86_ops.mem_enc_unreg_region)
5890 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
5893 case KVM_HYPERV_EVENTFD: {
5894 struct kvm_hyperv_eventfd hvevfd;
5897 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5899 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5902 case KVM_SET_PMU_EVENT_FILTER:
5903 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5905 case KVM_X86_SET_MSR_FILTER:
5906 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5915 static void kvm_init_msr_list(void)
5917 struct x86_pmu_capability x86_pmu;
5921 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5922 "Please update the fixed PMCs in msrs_to_saved_all[]");
5924 perf_get_x86_pmu_capability(&x86_pmu);
5926 num_msrs_to_save = 0;
5927 num_emulated_msrs = 0;
5928 num_msr_based_features = 0;
5930 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5931 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5935 * Even MSRs that are valid in the host may not be exposed
5936 * to the guests in some cases.
5938 switch (msrs_to_save_all[i]) {
5939 case MSR_IA32_BNDCFGS:
5940 if (!kvm_mpx_supported())
5944 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5947 case MSR_IA32_UMWAIT_CONTROL:
5948 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5951 case MSR_IA32_RTIT_CTL:
5952 case MSR_IA32_RTIT_STATUS:
5953 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5956 case MSR_IA32_RTIT_CR3_MATCH:
5957 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5958 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5961 case MSR_IA32_RTIT_OUTPUT_BASE:
5962 case MSR_IA32_RTIT_OUTPUT_MASK:
5963 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5964 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5965 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5968 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5969 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5970 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5971 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5974 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5975 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5976 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5979 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5980 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5981 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5988 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5991 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5992 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
5995 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5998 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5999 struct kvm_msr_entry msr;
6001 msr.index = msr_based_features_all[i];
6002 if (kvm_get_msr_feature(&msr))
6005 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6009 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6017 if (!(lapic_in_kernel(vcpu) &&
6018 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6019 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6030 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6037 if (!(lapic_in_kernel(vcpu) &&
6038 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6040 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6042 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6052 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6053 struct kvm_segment *var, int seg)
6055 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6058 void kvm_get_segment(struct kvm_vcpu *vcpu,
6059 struct kvm_segment *var, int seg)
6061 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6064 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6065 struct x86_exception *exception)
6069 BUG_ON(!mmu_is_nested(vcpu));
6071 /* NPT walks are always user-walks */
6072 access |= PFERR_USER_MASK;
6073 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6078 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6079 struct x86_exception *exception)
6081 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6082 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6084 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6086 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6087 struct x86_exception *exception)
6089 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6090 access |= PFERR_FETCH_MASK;
6091 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6094 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6095 struct x86_exception *exception)
6097 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6098 access |= PFERR_WRITE_MASK;
6099 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6101 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6103 /* uses this to access any guest's mapped memory without checking CPL */
6104 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6105 struct x86_exception *exception)
6107 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6110 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6111 struct kvm_vcpu *vcpu, u32 access,
6112 struct x86_exception *exception)
6115 int r = X86EMUL_CONTINUE;
6118 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6120 unsigned offset = addr & (PAGE_SIZE-1);
6121 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6124 if (gpa == UNMAPPED_GVA)
6125 return X86EMUL_PROPAGATE_FAULT;
6126 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6129 r = X86EMUL_IO_NEEDED;
6141 /* used for instruction fetching */
6142 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6143 gva_t addr, void *val, unsigned int bytes,
6144 struct x86_exception *exception)
6146 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6147 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6151 /* Inline kvm_read_guest_virt_helper for speed. */
6152 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6154 if (unlikely(gpa == UNMAPPED_GVA))
6155 return X86EMUL_PROPAGATE_FAULT;
6157 offset = addr & (PAGE_SIZE-1);
6158 if (WARN_ON(offset + bytes > PAGE_SIZE))
6159 bytes = (unsigned)PAGE_SIZE - offset;
6160 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6162 if (unlikely(ret < 0))
6163 return X86EMUL_IO_NEEDED;
6165 return X86EMUL_CONTINUE;
6168 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6169 gva_t addr, void *val, unsigned int bytes,
6170 struct x86_exception *exception)
6172 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6175 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6176 * is returned, but our callers are not ready for that and they blindly
6177 * call kvm_inject_page_fault. Ensure that they at least do not leak
6178 * uninitialized kernel stack memory into cr2 and error code.
6180 memset(exception, 0, sizeof(*exception));
6181 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6184 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6186 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6187 gva_t addr, void *val, unsigned int bytes,
6188 struct x86_exception *exception, bool system)
6190 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6193 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6194 access |= PFERR_USER_MASK;
6196 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6199 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6200 unsigned long addr, void *val, unsigned int bytes)
6202 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6203 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6205 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6208 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6209 struct kvm_vcpu *vcpu, u32 access,
6210 struct x86_exception *exception)
6213 int r = X86EMUL_CONTINUE;
6216 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6219 unsigned offset = addr & (PAGE_SIZE-1);
6220 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6223 if (gpa == UNMAPPED_GVA)
6224 return X86EMUL_PROPAGATE_FAULT;
6225 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6227 r = X86EMUL_IO_NEEDED;
6239 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6240 unsigned int bytes, struct x86_exception *exception,
6243 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6244 u32 access = PFERR_WRITE_MASK;
6246 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6247 access |= PFERR_USER_MASK;
6249 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6253 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6254 unsigned int bytes, struct x86_exception *exception)
6256 /* kvm_write_guest_virt_system can pull in tons of pages. */
6257 vcpu->arch.l1tf_flush_l1d = true;
6259 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6260 PFERR_WRITE_MASK, exception);
6262 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6264 int handle_ud(struct kvm_vcpu *vcpu)
6266 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6267 int emul_type = EMULTYPE_TRAP_UD;
6268 char sig[5]; /* ud2; .ascii "kvm" */
6269 struct x86_exception e;
6271 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6274 if (force_emulation_prefix &&
6275 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6276 sig, sizeof(sig), &e) == 0 &&
6277 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6278 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6279 emul_type = EMULTYPE_TRAP_UD_FORCED;
6282 return kvm_emulate_instruction(vcpu, emul_type);
6284 EXPORT_SYMBOL_GPL(handle_ud);
6286 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6287 gpa_t gpa, bool write)
6289 /* For APIC access vmexit */
6290 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6293 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6294 trace_vcpu_match_mmio(gva, gpa, write, true);
6301 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6302 gpa_t *gpa, struct x86_exception *exception,
6305 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6306 | (write ? PFERR_WRITE_MASK : 0);
6309 * currently PKRU is only applied to ept enabled guest so
6310 * there is no pkey in EPT page table for L1 guest or EPT
6311 * shadow page table for L2 guest.
6313 if (vcpu_match_mmio_gva(vcpu, gva)
6314 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6315 vcpu->arch.mmio_access, 0, access)) {
6316 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6317 (gva & (PAGE_SIZE - 1));
6318 trace_vcpu_match_mmio(gva, *gpa, write, false);
6322 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6324 if (*gpa == UNMAPPED_GVA)
6327 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6330 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6331 const void *val, int bytes)
6335 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6338 kvm_page_track_write(vcpu, gpa, val, bytes);
6342 struct read_write_emulator_ops {
6343 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6345 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6346 void *val, int bytes);
6347 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6348 int bytes, void *val);
6349 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6350 void *val, int bytes);
6354 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6356 if (vcpu->mmio_read_completed) {
6357 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6358 vcpu->mmio_fragments[0].gpa, val);
6359 vcpu->mmio_read_completed = 0;
6366 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6367 void *val, int bytes)
6369 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6372 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6373 void *val, int bytes)
6375 return emulator_write_phys(vcpu, gpa, val, bytes);
6378 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6380 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6381 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6384 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6385 void *val, int bytes)
6387 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6388 return X86EMUL_IO_NEEDED;
6391 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6392 void *val, int bytes)
6394 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6396 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6397 return X86EMUL_CONTINUE;
6400 static const struct read_write_emulator_ops read_emultor = {
6401 .read_write_prepare = read_prepare,
6402 .read_write_emulate = read_emulate,
6403 .read_write_mmio = vcpu_mmio_read,
6404 .read_write_exit_mmio = read_exit_mmio,
6407 static const struct read_write_emulator_ops write_emultor = {
6408 .read_write_emulate = write_emulate,
6409 .read_write_mmio = write_mmio,
6410 .read_write_exit_mmio = write_exit_mmio,
6414 static int emulator_read_write_onepage(unsigned long addr, void *val,
6416 struct x86_exception *exception,
6417 struct kvm_vcpu *vcpu,
6418 const struct read_write_emulator_ops *ops)
6422 bool write = ops->write;
6423 struct kvm_mmio_fragment *frag;
6424 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6427 * If the exit was due to a NPF we may already have a GPA.
6428 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6429 * Note, this cannot be used on string operations since string
6430 * operation using rep will only have the initial GPA from the NPF
6433 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6434 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6435 gpa = ctxt->gpa_val;
6436 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6438 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6440 return X86EMUL_PROPAGATE_FAULT;
6443 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6444 return X86EMUL_CONTINUE;
6447 * Is this MMIO handled locally?
6449 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6450 if (handled == bytes)
6451 return X86EMUL_CONTINUE;
6457 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6458 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6462 return X86EMUL_CONTINUE;
6465 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6467 void *val, unsigned int bytes,
6468 struct x86_exception *exception,
6469 const struct read_write_emulator_ops *ops)
6471 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6475 if (ops->read_write_prepare &&
6476 ops->read_write_prepare(vcpu, val, bytes))
6477 return X86EMUL_CONTINUE;
6479 vcpu->mmio_nr_fragments = 0;
6481 /* Crossing a page boundary? */
6482 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6485 now = -addr & ~PAGE_MASK;
6486 rc = emulator_read_write_onepage(addr, val, now, exception,
6489 if (rc != X86EMUL_CONTINUE)
6492 if (ctxt->mode != X86EMUL_MODE_PROT64)
6498 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6500 if (rc != X86EMUL_CONTINUE)
6503 if (!vcpu->mmio_nr_fragments)
6506 gpa = vcpu->mmio_fragments[0].gpa;
6508 vcpu->mmio_needed = 1;
6509 vcpu->mmio_cur_fragment = 0;
6511 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6512 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6513 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6514 vcpu->run->mmio.phys_addr = gpa;
6516 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6519 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6523 struct x86_exception *exception)
6525 return emulator_read_write(ctxt, addr, val, bytes,
6526 exception, &read_emultor);
6529 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6533 struct x86_exception *exception)
6535 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6536 exception, &write_emultor);
6539 #define CMPXCHG_TYPE(t, ptr, old, new) \
6540 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6542 #ifdef CONFIG_X86_64
6543 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6545 # define CMPXCHG64(ptr, old, new) \
6546 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6549 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6554 struct x86_exception *exception)
6556 struct kvm_host_map map;
6557 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6563 /* guests cmpxchg8b have to be emulated atomically */
6564 if (bytes > 8 || (bytes & (bytes - 1)))
6567 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6569 if (gpa == UNMAPPED_GVA ||
6570 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6574 * Emulate the atomic as a straight write to avoid #AC if SLD is
6575 * enabled in the host and the access splits a cache line.
6577 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6578 page_line_mask = ~(cache_line_size() - 1);
6580 page_line_mask = PAGE_MASK;
6582 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6585 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6588 kaddr = map.hva + offset_in_page(gpa);
6592 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6595 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6598 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6601 exchanged = CMPXCHG64(kaddr, old, new);
6607 kvm_vcpu_unmap(vcpu, &map, true);
6610 return X86EMUL_CMPXCHG_FAILED;
6612 kvm_page_track_write(vcpu, gpa, new, bytes);
6614 return X86EMUL_CONTINUE;
6617 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6619 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6622 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6626 for (i = 0; i < vcpu->arch.pio.count; i++) {
6627 if (vcpu->arch.pio.in)
6628 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6629 vcpu->arch.pio.size, pd);
6631 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6632 vcpu->arch.pio.port, vcpu->arch.pio.size,
6636 pd += vcpu->arch.pio.size;
6641 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6642 unsigned short port, void *val,
6643 unsigned int count, bool in)
6645 vcpu->arch.pio.port = port;
6646 vcpu->arch.pio.in = in;
6647 vcpu->arch.pio.count = count;
6648 vcpu->arch.pio.size = size;
6650 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6651 vcpu->arch.pio.count = 0;
6655 vcpu->run->exit_reason = KVM_EXIT_IO;
6656 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6657 vcpu->run->io.size = size;
6658 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6659 vcpu->run->io.count = count;
6660 vcpu->run->io.port = port;
6665 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6666 unsigned short port, void *val, unsigned int count)
6670 if (vcpu->arch.pio.count)
6673 memset(vcpu->arch.pio_data, 0, size * count);
6675 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6678 memcpy(val, vcpu->arch.pio_data, size * count);
6679 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6680 vcpu->arch.pio.count = 0;
6687 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6688 int size, unsigned short port, void *val,
6691 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6695 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6696 unsigned short port, const void *val,
6699 memcpy(vcpu->arch.pio_data, val, size * count);
6700 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6701 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6704 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6705 int size, unsigned short port,
6706 const void *val, unsigned int count)
6708 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6711 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6713 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6716 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6718 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6721 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6723 if (!need_emulate_wbinvd(vcpu))
6724 return X86EMUL_CONTINUE;
6726 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6727 int cpu = get_cpu();
6729 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6730 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6731 wbinvd_ipi, NULL, 1);
6733 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6736 return X86EMUL_CONTINUE;
6739 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6741 kvm_emulate_wbinvd_noskip(vcpu);
6742 return kvm_skip_emulated_instruction(vcpu);
6744 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6748 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6750 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6753 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6754 unsigned long *dest)
6756 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6759 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6760 unsigned long value)
6763 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6766 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6768 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6771 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6773 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6774 unsigned long value;
6778 value = kvm_read_cr0(vcpu);
6781 value = vcpu->arch.cr2;
6784 value = kvm_read_cr3(vcpu);
6787 value = kvm_read_cr4(vcpu);
6790 value = kvm_get_cr8(vcpu);
6793 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6800 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6802 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6807 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6810 vcpu->arch.cr2 = val;
6813 res = kvm_set_cr3(vcpu, val);
6816 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6819 res = kvm_set_cr8(vcpu, val);
6822 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6829 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6831 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
6834 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6836 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
6839 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6841 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
6844 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6846 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
6849 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6851 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
6854 static unsigned long emulator_get_cached_segment_base(
6855 struct x86_emulate_ctxt *ctxt, int seg)
6857 return get_segment_base(emul_to_vcpu(ctxt), seg);
6860 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6861 struct desc_struct *desc, u32 *base3,
6864 struct kvm_segment var;
6866 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6867 *selector = var.selector;
6870 memset(desc, 0, sizeof(*desc));
6878 set_desc_limit(desc, var.limit);
6879 set_desc_base(desc, (unsigned long)var.base);
6880 #ifdef CONFIG_X86_64
6882 *base3 = var.base >> 32;
6884 desc->type = var.type;
6886 desc->dpl = var.dpl;
6887 desc->p = var.present;
6888 desc->avl = var.avl;
6896 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6897 struct desc_struct *desc, u32 base3,
6900 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6901 struct kvm_segment var;
6903 var.selector = selector;
6904 var.base = get_desc_base(desc);
6905 #ifdef CONFIG_X86_64
6906 var.base |= ((u64)base3) << 32;
6908 var.limit = get_desc_limit(desc);
6910 var.limit = (var.limit << 12) | 0xfff;
6911 var.type = desc->type;
6912 var.dpl = desc->dpl;
6917 var.avl = desc->avl;
6918 var.present = desc->p;
6919 var.unusable = !var.present;
6922 kvm_set_segment(vcpu, &var, seg);
6926 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6927 u32 msr_index, u64 *pdata)
6929 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6932 r = kvm_get_msr(vcpu, msr_index, pdata);
6934 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6935 /* Bounce to user space */
6936 return X86EMUL_IO_NEEDED;
6942 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6943 u32 msr_index, u64 data)
6945 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6948 r = kvm_set_msr(vcpu, msr_index, data);
6950 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6951 /* Bounce to user space */
6952 return X86EMUL_IO_NEEDED;
6958 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6960 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6962 return vcpu->arch.smbase;
6965 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6967 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6969 vcpu->arch.smbase = smbase;
6972 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6975 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6978 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6979 u32 pmc, u64 *pdata)
6981 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6984 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6986 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6989 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6990 struct x86_instruction_info *info,
6991 enum x86_intercept_stage stage)
6993 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
6997 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6998 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7001 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7004 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7006 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7009 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7011 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7014 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7016 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7019 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7021 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7024 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7026 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7029 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7031 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7034 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7036 return emul_to_vcpu(ctxt)->arch.hflags;
7039 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
7041 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
7044 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
7045 const char *smstate)
7047 return static_call(kvm_x86_pre_leave_smm)(emul_to_vcpu(ctxt), smstate);
7050 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
7052 kvm_smm_changed(emul_to_vcpu(ctxt));
7055 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7057 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7060 static const struct x86_emulate_ops emulate_ops = {
7061 .read_gpr = emulator_read_gpr,
7062 .write_gpr = emulator_write_gpr,
7063 .read_std = emulator_read_std,
7064 .write_std = emulator_write_std,
7065 .read_phys = kvm_read_guest_phys_system,
7066 .fetch = kvm_fetch_guest_virt,
7067 .read_emulated = emulator_read_emulated,
7068 .write_emulated = emulator_write_emulated,
7069 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7070 .invlpg = emulator_invlpg,
7071 .pio_in_emulated = emulator_pio_in_emulated,
7072 .pio_out_emulated = emulator_pio_out_emulated,
7073 .get_segment = emulator_get_segment,
7074 .set_segment = emulator_set_segment,
7075 .get_cached_segment_base = emulator_get_cached_segment_base,
7076 .get_gdt = emulator_get_gdt,
7077 .get_idt = emulator_get_idt,
7078 .set_gdt = emulator_set_gdt,
7079 .set_idt = emulator_set_idt,
7080 .get_cr = emulator_get_cr,
7081 .set_cr = emulator_set_cr,
7082 .cpl = emulator_get_cpl,
7083 .get_dr = emulator_get_dr,
7084 .set_dr = emulator_set_dr,
7085 .get_smbase = emulator_get_smbase,
7086 .set_smbase = emulator_set_smbase,
7087 .set_msr = emulator_set_msr,
7088 .get_msr = emulator_get_msr,
7089 .check_pmc = emulator_check_pmc,
7090 .read_pmc = emulator_read_pmc,
7091 .halt = emulator_halt,
7092 .wbinvd = emulator_wbinvd,
7093 .fix_hypercall = emulator_fix_hypercall,
7094 .intercept = emulator_intercept,
7095 .get_cpuid = emulator_get_cpuid,
7096 .guest_has_long_mode = emulator_guest_has_long_mode,
7097 .guest_has_movbe = emulator_guest_has_movbe,
7098 .guest_has_fxsr = emulator_guest_has_fxsr,
7099 .set_nmi_mask = emulator_set_nmi_mask,
7100 .get_hflags = emulator_get_hflags,
7101 .set_hflags = emulator_set_hflags,
7102 .pre_leave_smm = emulator_pre_leave_smm,
7103 .post_leave_smm = emulator_post_leave_smm,
7104 .set_xcr = emulator_set_xcr,
7107 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7109 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7111 * an sti; sti; sequence only disable interrupts for the first
7112 * instruction. So, if the last instruction, be it emulated or
7113 * not, left the system with the INT_STI flag enabled, it
7114 * means that the last instruction is an sti. We should not
7115 * leave the flag on in this case. The same goes for mov ss
7117 if (int_shadow & mask)
7119 if (unlikely(int_shadow || mask)) {
7120 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7122 kvm_make_request(KVM_REQ_EVENT, vcpu);
7126 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7128 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7129 if (ctxt->exception.vector == PF_VECTOR)
7130 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7132 if (ctxt->exception.error_code_valid)
7133 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7134 ctxt->exception.error_code);
7136 kvm_queue_exception(vcpu, ctxt->exception.vector);
7140 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7142 struct x86_emulate_ctxt *ctxt;
7144 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7146 pr_err("kvm: failed to allocate vcpu's emulator\n");
7151 ctxt->ops = &emulate_ops;
7152 vcpu->arch.emulate_ctxt = ctxt;
7157 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7159 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7162 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7164 ctxt->gpa_available = false;
7165 ctxt->eflags = kvm_get_rflags(vcpu);
7166 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7168 ctxt->eip = kvm_rip_read(vcpu);
7169 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7170 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7171 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7172 cs_db ? X86EMUL_MODE_PROT32 :
7173 X86EMUL_MODE_PROT16;
7174 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7175 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7176 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7178 init_decode_cache(ctxt);
7179 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7182 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7184 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7187 init_emulate_ctxt(vcpu);
7191 ctxt->_eip = ctxt->eip + inc_eip;
7192 ret = emulate_int_real(ctxt, irq);
7194 if (ret != X86EMUL_CONTINUE) {
7195 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7197 ctxt->eip = ctxt->_eip;
7198 kvm_rip_write(vcpu, ctxt->eip);
7199 kvm_set_rflags(vcpu, ctxt->eflags);
7202 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7204 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7206 ++vcpu->stat.insn_emulation_fail;
7207 trace_kvm_emulate_insn_failed(vcpu);
7209 if (emulation_type & EMULTYPE_VMWARE_GP) {
7210 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7214 if (emulation_type & EMULTYPE_SKIP) {
7215 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7216 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7217 vcpu->run->internal.ndata = 0;
7221 kvm_queue_exception(vcpu, UD_VECTOR);
7223 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7224 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7225 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7226 vcpu->run->internal.ndata = 0;
7233 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7234 bool write_fault_to_shadow_pgtable,
7237 gpa_t gpa = cr2_or_gpa;
7240 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7243 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7244 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7247 if (!vcpu->arch.mmu->direct_map) {
7249 * Write permission should be allowed since only
7250 * write access need to be emulated.
7252 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7255 * If the mapping is invalid in guest, let cpu retry
7256 * it to generate fault.
7258 if (gpa == UNMAPPED_GVA)
7263 * Do not retry the unhandleable instruction if it faults on the
7264 * readonly host memory, otherwise it will goto a infinite loop:
7265 * retry instruction -> write #PF -> emulation fail -> retry
7266 * instruction -> ...
7268 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7271 * If the instruction failed on the error pfn, it can not be fixed,
7272 * report the error to userspace.
7274 if (is_error_noslot_pfn(pfn))
7277 kvm_release_pfn_clean(pfn);
7279 /* The instructions are well-emulated on direct mmu. */
7280 if (vcpu->arch.mmu->direct_map) {
7281 unsigned int indirect_shadow_pages;
7283 write_lock(&vcpu->kvm->mmu_lock);
7284 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7285 write_unlock(&vcpu->kvm->mmu_lock);
7287 if (indirect_shadow_pages)
7288 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7294 * if emulation was due to access to shadowed page table
7295 * and it failed try to unshadow page and re-enter the
7296 * guest to let CPU execute the instruction.
7298 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7301 * If the access faults on its page table, it can not
7302 * be fixed by unprotecting shadow page and it should
7303 * be reported to userspace.
7305 return !write_fault_to_shadow_pgtable;
7308 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7309 gpa_t cr2_or_gpa, int emulation_type)
7311 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7312 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7314 last_retry_eip = vcpu->arch.last_retry_eip;
7315 last_retry_addr = vcpu->arch.last_retry_addr;
7318 * If the emulation is caused by #PF and it is non-page_table
7319 * writing instruction, it means the VM-EXIT is caused by shadow
7320 * page protected, we can zap the shadow page and retry this
7321 * instruction directly.
7323 * Note: if the guest uses a non-page-table modifying instruction
7324 * on the PDE that points to the instruction, then we will unmap
7325 * the instruction and go to an infinite loop. So, we cache the
7326 * last retried eip and the last fault address, if we meet the eip
7327 * and the address again, we can break out of the potential infinite
7330 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7332 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7335 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7336 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7339 if (x86_page_table_writing_insn(ctxt))
7342 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7345 vcpu->arch.last_retry_eip = ctxt->eip;
7346 vcpu->arch.last_retry_addr = cr2_or_gpa;
7348 if (!vcpu->arch.mmu->direct_map)
7349 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7351 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7356 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7357 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7359 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7361 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7362 /* This is a good place to trace that we are exiting SMM. */
7363 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7365 /* Process a latched INIT or SMI, if any. */
7366 kvm_make_request(KVM_REQ_EVENT, vcpu);
7369 kvm_mmu_reset_context(vcpu);
7372 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7381 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7382 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7387 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7389 struct kvm_run *kvm_run = vcpu->run;
7391 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7392 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7393 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7394 kvm_run->debug.arch.exception = DB_VECTOR;
7395 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7398 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7402 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7404 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7407 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7412 * rflags is the old, "raw" value of the flags. The new value has
7413 * not been saved yet.
7415 * This is correct even for TF set by the guest, because "the
7416 * processor will not generate this exception after the instruction
7417 * that sets the TF flag".
7419 if (unlikely(rflags & X86_EFLAGS_TF))
7420 r = kvm_vcpu_do_singlestep(vcpu);
7423 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7425 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7427 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7428 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7429 struct kvm_run *kvm_run = vcpu->run;
7430 unsigned long eip = kvm_get_linear_rip(vcpu);
7431 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7432 vcpu->arch.guest_debug_dr7,
7436 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7437 kvm_run->debug.arch.pc = eip;
7438 kvm_run->debug.arch.exception = DB_VECTOR;
7439 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7445 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7446 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7447 unsigned long eip = kvm_get_linear_rip(vcpu);
7448 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7453 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7462 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7464 switch (ctxt->opcode_len) {
7471 case 0xe6: /* OUT */
7475 case 0x6c: /* INS */
7477 case 0x6e: /* OUTS */
7484 case 0x33: /* RDPMC */
7494 * Decode to be emulated instruction. Return EMULATION_OK if success.
7496 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7497 void *insn, int insn_len)
7499 int r = EMULATION_OK;
7500 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7502 init_emulate_ctxt(vcpu);
7505 * We will reenter on the same instruction since we do not set
7506 * complete_userspace_io. This does not handle watchpoints yet,
7507 * those would be handled in the emulate_ops.
7509 if (!(emulation_type & EMULTYPE_SKIP) &&
7510 kvm_vcpu_check_breakpoint(vcpu, &r))
7513 ctxt->interruptibility = 0;
7514 ctxt->have_exception = false;
7515 ctxt->exception.vector = -1;
7516 ctxt->perm_ok = false;
7518 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7520 r = x86_decode_insn(ctxt, insn, insn_len);
7522 trace_kvm_emulate_insn_start(vcpu);
7523 ++vcpu->stat.insn_emulation;
7527 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7529 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7530 int emulation_type, void *insn, int insn_len)
7533 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7534 bool writeback = true;
7535 bool write_fault_to_spt;
7537 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7540 vcpu->arch.l1tf_flush_l1d = true;
7543 * Clear write_fault_to_shadow_pgtable here to ensure it is
7546 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7547 vcpu->arch.write_fault_to_shadow_pgtable = false;
7549 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7550 kvm_clear_exception_queue(vcpu);
7552 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7554 if (r != EMULATION_OK) {
7555 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7556 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7557 kvm_queue_exception(vcpu, UD_VECTOR);
7560 if (reexecute_instruction(vcpu, cr2_or_gpa,
7564 if (ctxt->have_exception) {
7566 * #UD should result in just EMULATION_FAILED, and trap-like
7567 * exception should not be encountered during decode.
7569 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7570 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7571 inject_emulated_exception(vcpu);
7574 return handle_emulation_failure(vcpu, emulation_type);
7578 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7579 !is_vmware_backdoor_opcode(ctxt)) {
7580 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7585 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7586 * for kvm_skip_emulated_instruction(). The caller is responsible for
7587 * updating interruptibility state and injecting single-step #DBs.
7589 if (emulation_type & EMULTYPE_SKIP) {
7590 kvm_rip_write(vcpu, ctxt->_eip);
7591 if (ctxt->eflags & X86_EFLAGS_RF)
7592 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7596 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7599 /* this is needed for vmware backdoor interface to work since it
7600 changes registers values during IO operation */
7601 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7602 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7603 emulator_invalidate_register_cache(ctxt);
7607 if (emulation_type & EMULTYPE_PF) {
7608 /* Save the faulting GPA (cr2) in the address field */
7609 ctxt->exception.address = cr2_or_gpa;
7611 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7612 if (vcpu->arch.mmu->direct_map) {
7613 ctxt->gpa_available = true;
7614 ctxt->gpa_val = cr2_or_gpa;
7617 /* Sanitize the address out of an abundance of paranoia. */
7618 ctxt->exception.address = 0;
7621 r = x86_emulate_insn(ctxt);
7623 if (r == EMULATION_INTERCEPTED)
7626 if (r == EMULATION_FAILED) {
7627 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7631 return handle_emulation_failure(vcpu, emulation_type);
7634 if (ctxt->have_exception) {
7636 if (inject_emulated_exception(vcpu))
7638 } else if (vcpu->arch.pio.count) {
7639 if (!vcpu->arch.pio.in) {
7640 /* FIXME: return into emulator if single-stepping. */
7641 vcpu->arch.pio.count = 0;
7644 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7647 } else if (vcpu->mmio_needed) {
7648 ++vcpu->stat.mmio_exits;
7650 if (!vcpu->mmio_is_write)
7653 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7654 } else if (r == EMULATION_RESTART)
7660 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7661 toggle_interruptibility(vcpu, ctxt->interruptibility);
7662 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7663 if (!ctxt->have_exception ||
7664 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7665 kvm_rip_write(vcpu, ctxt->eip);
7666 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7667 r = kvm_vcpu_do_singlestep(vcpu);
7668 if (kvm_x86_ops.update_emulated_instruction)
7669 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7670 __kvm_set_rflags(vcpu, ctxt->eflags);
7674 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7675 * do nothing, and it will be requested again as soon as
7676 * the shadow expires. But we still need to check here,
7677 * because POPF has no interrupt shadow.
7679 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7680 kvm_make_request(KVM_REQ_EVENT, vcpu);
7682 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7687 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7689 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7691 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7693 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7694 void *insn, int insn_len)
7696 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7698 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7700 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7702 vcpu->arch.pio.count = 0;
7706 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7708 vcpu->arch.pio.count = 0;
7710 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7713 return kvm_skip_emulated_instruction(vcpu);
7716 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7717 unsigned short port)
7719 unsigned long val = kvm_rax_read(vcpu);
7720 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7726 * Workaround userspace that relies on old KVM behavior of %rip being
7727 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7730 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7731 vcpu->arch.complete_userspace_io =
7732 complete_fast_pio_out_port_0x7e;
7733 kvm_skip_emulated_instruction(vcpu);
7735 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7736 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7741 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7745 /* We should only ever be called with arch.pio.count equal to 1 */
7746 BUG_ON(vcpu->arch.pio.count != 1);
7748 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7749 vcpu->arch.pio.count = 0;
7753 /* For size less than 4 we merge, else we zero extend */
7754 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7757 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7758 * the copy and tracing
7760 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7761 kvm_rax_write(vcpu, val);
7763 return kvm_skip_emulated_instruction(vcpu);
7766 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7767 unsigned short port)
7772 /* For size less than 4 we merge, else we zero extend */
7773 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7775 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7777 kvm_rax_write(vcpu, val);
7781 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7782 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7787 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7792 ret = kvm_fast_pio_in(vcpu, size, port);
7794 ret = kvm_fast_pio_out(vcpu, size, port);
7795 return ret && kvm_skip_emulated_instruction(vcpu);
7797 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7799 static int kvmclock_cpu_down_prep(unsigned int cpu)
7801 __this_cpu_write(cpu_tsc_khz, 0);
7805 static void tsc_khz_changed(void *data)
7807 struct cpufreq_freqs *freq = data;
7808 unsigned long khz = 0;
7812 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7813 khz = cpufreq_quick_get(raw_smp_processor_id());
7816 __this_cpu_write(cpu_tsc_khz, khz);
7819 #ifdef CONFIG_X86_64
7820 static void kvm_hyperv_tsc_notifier(void)
7823 struct kvm_vcpu *vcpu;
7825 unsigned long flags;
7827 mutex_lock(&kvm_lock);
7828 list_for_each_entry(kvm, &vm_list, vm_list)
7829 kvm_make_mclock_inprogress_request(kvm);
7831 hyperv_stop_tsc_emulation();
7833 /* TSC frequency always matches when on Hyper-V */
7834 for_each_present_cpu(cpu)
7835 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7836 kvm_max_guest_tsc_khz = tsc_khz;
7838 list_for_each_entry(kvm, &vm_list, vm_list) {
7839 struct kvm_arch *ka = &kvm->arch;
7841 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
7842 pvclock_update_vm_gtod_copy(kvm);
7843 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
7845 kvm_for_each_vcpu(cpu, vcpu, kvm)
7846 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7848 kvm_for_each_vcpu(cpu, vcpu, kvm)
7849 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7851 mutex_unlock(&kvm_lock);
7855 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7858 struct kvm_vcpu *vcpu;
7859 int i, send_ipi = 0;
7862 * We allow guests to temporarily run on slowing clocks,
7863 * provided we notify them after, or to run on accelerating
7864 * clocks, provided we notify them before. Thus time never
7867 * However, we have a problem. We can't atomically update
7868 * the frequency of a given CPU from this function; it is
7869 * merely a notifier, which can be called from any CPU.
7870 * Changing the TSC frequency at arbitrary points in time
7871 * requires a recomputation of local variables related to
7872 * the TSC for each VCPU. We must flag these local variables
7873 * to be updated and be sure the update takes place with the
7874 * new frequency before any guests proceed.
7876 * Unfortunately, the combination of hotplug CPU and frequency
7877 * change creates an intractable locking scenario; the order
7878 * of when these callouts happen is undefined with respect to
7879 * CPU hotplug, and they can race with each other. As such,
7880 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7881 * undefined; you can actually have a CPU frequency change take
7882 * place in between the computation of X and the setting of the
7883 * variable. To protect against this problem, all updates of
7884 * the per_cpu tsc_khz variable are done in an interrupt
7885 * protected IPI, and all callers wishing to update the value
7886 * must wait for a synchronous IPI to complete (which is trivial
7887 * if the caller is on the CPU already). This establishes the
7888 * necessary total order on variable updates.
7890 * Note that because a guest time update may take place
7891 * anytime after the setting of the VCPU's request bit, the
7892 * correct TSC value must be set before the request. However,
7893 * to ensure the update actually makes it to any guest which
7894 * starts running in hardware virtualization between the set
7895 * and the acquisition of the spinlock, we must also ping the
7896 * CPU after setting the request bit.
7900 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7902 mutex_lock(&kvm_lock);
7903 list_for_each_entry(kvm, &vm_list, vm_list) {
7904 kvm_for_each_vcpu(i, vcpu, kvm) {
7905 if (vcpu->cpu != cpu)
7907 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7908 if (vcpu->cpu != raw_smp_processor_id())
7912 mutex_unlock(&kvm_lock);
7914 if (freq->old < freq->new && send_ipi) {
7916 * We upscale the frequency. Must make the guest
7917 * doesn't see old kvmclock values while running with
7918 * the new frequency, otherwise we risk the guest sees
7919 * time go backwards.
7921 * In case we update the frequency for another cpu
7922 * (which might be in guest context) send an interrupt
7923 * to kick the cpu out of guest context. Next time
7924 * guest context is entered kvmclock will be updated,
7925 * so the guest will not see stale values.
7927 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7931 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7934 struct cpufreq_freqs *freq = data;
7937 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7939 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7942 for_each_cpu(cpu, freq->policy->cpus)
7943 __kvmclock_cpufreq_notifier(freq, cpu);
7948 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7949 .notifier_call = kvmclock_cpufreq_notifier
7952 static int kvmclock_cpu_online(unsigned int cpu)
7954 tsc_khz_changed(NULL);
7958 static void kvm_timer_init(void)
7960 max_tsc_khz = tsc_khz;
7962 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7963 #ifdef CONFIG_CPU_FREQ
7964 struct cpufreq_policy *policy;
7968 policy = cpufreq_cpu_get(cpu);
7970 if (policy->cpuinfo.max_freq)
7971 max_tsc_khz = policy->cpuinfo.max_freq;
7972 cpufreq_cpu_put(policy);
7976 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7977 CPUFREQ_TRANSITION_NOTIFIER);
7980 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7981 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7984 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7985 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7987 int kvm_is_in_guest(void)
7989 return __this_cpu_read(current_vcpu) != NULL;
7992 static int kvm_is_user_mode(void)
7996 if (__this_cpu_read(current_vcpu))
7997 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
7999 return user_mode != 0;
8002 static unsigned long kvm_get_guest_ip(void)
8004 unsigned long ip = 0;
8006 if (__this_cpu_read(current_vcpu))
8007 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8012 static void kvm_handle_intel_pt_intr(void)
8014 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8016 kvm_make_request(KVM_REQ_PMI, vcpu);
8017 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8018 (unsigned long *)&vcpu->arch.pmu.global_status);
8021 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8022 .is_in_guest = kvm_is_in_guest,
8023 .is_user_mode = kvm_is_user_mode,
8024 .get_guest_ip = kvm_get_guest_ip,
8025 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8028 #ifdef CONFIG_X86_64
8029 static void pvclock_gtod_update_fn(struct work_struct *work)
8033 struct kvm_vcpu *vcpu;
8036 mutex_lock(&kvm_lock);
8037 list_for_each_entry(kvm, &vm_list, vm_list)
8038 kvm_for_each_vcpu(i, vcpu, kvm)
8039 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8040 atomic_set(&kvm_guest_has_master_clock, 0);
8041 mutex_unlock(&kvm_lock);
8044 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8047 * Notification about pvclock gtod data update.
8049 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8052 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8053 struct timekeeper *tk = priv;
8055 update_pvclock_gtod(tk);
8057 /* disable master clock if host does not trust, or does not
8058 * use, TSC based clocksource.
8060 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8061 atomic_read(&kvm_guest_has_master_clock) != 0)
8062 queue_work(system_long_wq, &pvclock_gtod_work);
8067 static struct notifier_block pvclock_gtod_notifier = {
8068 .notifier_call = pvclock_gtod_notify,
8072 int kvm_arch_init(void *opaque)
8074 struct kvm_x86_init_ops *ops = opaque;
8077 if (kvm_x86_ops.hardware_enable) {
8078 printk(KERN_ERR "kvm: already loaded the other module\n");
8083 if (!ops->cpu_has_kvm_support()) {
8084 pr_err_ratelimited("kvm: no hardware support\n");
8088 if (ops->disabled_by_bios()) {
8089 pr_err_ratelimited("kvm: disabled by bios\n");
8095 * KVM explicitly assumes that the guest has an FPU and
8096 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8097 * vCPU's FPU state as a fxregs_state struct.
8099 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8100 printk(KERN_ERR "kvm: inadequate fpu\n");
8106 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8107 __alignof__(struct fpu), SLAB_ACCOUNT,
8109 if (!x86_fpu_cache) {
8110 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8114 x86_emulator_cache = kvm_alloc_emulator_cache();
8115 if (!x86_emulator_cache) {
8116 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8117 goto out_free_x86_fpu_cache;
8120 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8121 if (!user_return_msrs) {
8122 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8123 goto out_free_x86_emulator_cache;
8126 r = kvm_mmu_module_init();
8128 goto out_free_percpu;
8132 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8134 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8135 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8136 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8139 if (pi_inject_timer == -1)
8140 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8141 #ifdef CONFIG_X86_64
8142 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8144 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8145 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8151 free_percpu(user_return_msrs);
8152 out_free_x86_emulator_cache:
8153 kmem_cache_destroy(x86_emulator_cache);
8154 out_free_x86_fpu_cache:
8155 kmem_cache_destroy(x86_fpu_cache);
8160 void kvm_arch_exit(void)
8162 #ifdef CONFIG_X86_64
8163 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8164 clear_hv_tscchange_cb();
8167 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8169 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8170 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8171 CPUFREQ_TRANSITION_NOTIFIER);
8172 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8173 #ifdef CONFIG_X86_64
8174 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8176 kvm_x86_ops.hardware_enable = NULL;
8177 kvm_mmu_module_exit();
8178 free_percpu(user_return_msrs);
8179 kmem_cache_destroy(x86_fpu_cache);
8180 #ifdef CONFIG_KVM_XEN
8181 static_key_deferred_flush(&kvm_xen_enabled);
8182 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8186 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8188 ++vcpu->stat.halt_exits;
8189 if (lapic_in_kernel(vcpu)) {
8190 vcpu->arch.mp_state = state;
8193 vcpu->run->exit_reason = reason;
8198 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8200 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8202 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8204 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8206 int ret = kvm_skip_emulated_instruction(vcpu);
8208 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8209 * KVM_EXIT_DEBUG here.
8211 return kvm_vcpu_halt(vcpu) && ret;
8213 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8215 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8217 int ret = kvm_skip_emulated_instruction(vcpu);
8219 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8221 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8223 #ifdef CONFIG_X86_64
8224 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8225 unsigned long clock_type)
8227 struct kvm_clock_pairing clock_pairing;
8228 struct timespec64 ts;
8232 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8233 return -KVM_EOPNOTSUPP;
8235 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8236 return -KVM_EOPNOTSUPP;
8238 clock_pairing.sec = ts.tv_sec;
8239 clock_pairing.nsec = ts.tv_nsec;
8240 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8241 clock_pairing.flags = 0;
8242 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8245 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8246 sizeof(struct kvm_clock_pairing)))
8254 * kvm_pv_kick_cpu_op: Kick a vcpu.
8256 * @apicid - apicid of vcpu to be kicked.
8258 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8260 struct kvm_lapic_irq lapic_irq;
8262 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8263 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8264 lapic_irq.level = 0;
8265 lapic_irq.dest_id = apicid;
8266 lapic_irq.msi_redir_hint = false;
8268 lapic_irq.delivery_mode = APIC_DM_REMRD;
8269 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8272 bool kvm_apicv_activated(struct kvm *kvm)
8274 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8276 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8278 void kvm_apicv_init(struct kvm *kvm, bool enable)
8281 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8282 &kvm->arch.apicv_inhibit_reasons);
8284 set_bit(APICV_INHIBIT_REASON_DISABLE,
8285 &kvm->arch.apicv_inhibit_reasons);
8287 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8289 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8291 struct kvm_vcpu *target = NULL;
8292 struct kvm_apic_map *map;
8294 vcpu->stat.directed_yield_attempted++;
8297 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8299 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8300 target = map->phys_map[dest_id]->vcpu;
8304 if (!target || !READ_ONCE(target->ready))
8307 /* Ignore requests to yield to self */
8311 if (kvm_vcpu_yield_to(target) <= 0)
8314 vcpu->stat.directed_yield_successful++;
8320 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8322 unsigned long nr, a0, a1, a2, a3, ret;
8325 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8326 return kvm_xen_hypercall(vcpu);
8328 if (kvm_hv_hypercall_enabled(vcpu))
8329 return kvm_hv_hypercall(vcpu);
8331 nr = kvm_rax_read(vcpu);
8332 a0 = kvm_rbx_read(vcpu);
8333 a1 = kvm_rcx_read(vcpu);
8334 a2 = kvm_rdx_read(vcpu);
8335 a3 = kvm_rsi_read(vcpu);
8337 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8339 op_64_bit = is_64_bit_mode(vcpu);
8348 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8356 case KVM_HC_VAPIC_POLL_IRQ:
8359 case KVM_HC_KICK_CPU:
8360 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8363 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8364 kvm_sched_yield(vcpu, a1);
8367 #ifdef CONFIG_X86_64
8368 case KVM_HC_CLOCK_PAIRING:
8369 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8372 case KVM_HC_SEND_IPI:
8373 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8376 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8378 case KVM_HC_SCHED_YIELD:
8379 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8382 kvm_sched_yield(vcpu, a0);
8392 kvm_rax_write(vcpu, ret);
8394 ++vcpu->stat.hypercalls;
8395 return kvm_skip_emulated_instruction(vcpu);
8397 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8399 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8401 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8402 char instruction[3];
8403 unsigned long rip = kvm_rip_read(vcpu);
8405 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8407 return emulator_write_emulated(ctxt, rip, instruction, 3,
8411 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8413 return vcpu->run->request_interrupt_window &&
8414 likely(!pic_in_kernel(vcpu->kvm));
8417 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8419 struct kvm_run *kvm_run = vcpu->run;
8422 * if_flag is obsolete and useless, so do not bother
8423 * setting it for SEV-ES guests. Userspace can just
8424 * use kvm_run->ready_for_interrupt_injection.
8426 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8427 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8429 kvm_run->cr8 = kvm_get_cr8(vcpu);
8430 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8431 kvm_run->ready_for_interrupt_injection =
8432 pic_in_kernel(vcpu->kvm) ||
8433 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8436 kvm_run->flags |= KVM_RUN_X86_SMM;
8439 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8443 if (!kvm_x86_ops.update_cr8_intercept)
8446 if (!lapic_in_kernel(vcpu))
8449 if (vcpu->arch.apicv_active)
8452 if (!vcpu->arch.apic->vapic_addr)
8453 max_irr = kvm_lapic_find_highest_irr(vcpu);
8460 tpr = kvm_lapic_get_cr8(vcpu);
8462 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8466 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8468 if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
8471 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8472 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8476 return kvm_x86_ops.nested_ops->check_events(vcpu);
8479 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8481 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8482 vcpu->arch.exception.error_code = false;
8483 static_call(kvm_x86_queue_exception)(vcpu);
8486 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8489 bool can_inject = true;
8491 /* try to reinject previous events if any */
8493 if (vcpu->arch.exception.injected) {
8494 kvm_inject_exception(vcpu);
8498 * Do not inject an NMI or interrupt if there is a pending
8499 * exception. Exceptions and interrupts are recognized at
8500 * instruction boundaries, i.e. the start of an instruction.
8501 * Trap-like exceptions, e.g. #DB, have higher priority than
8502 * NMIs and interrupts, i.e. traps are recognized before an
8503 * NMI/interrupt that's pending on the same instruction.
8504 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8505 * priority, but are only generated (pended) during instruction
8506 * execution, i.e. a pending fault-like exception means the
8507 * fault occurred on the *previous* instruction and must be
8508 * serviced prior to recognizing any new events in order to
8509 * fully complete the previous instruction.
8511 else if (!vcpu->arch.exception.pending) {
8512 if (vcpu->arch.nmi_injected) {
8513 static_call(kvm_x86_set_nmi)(vcpu);
8515 } else if (vcpu->arch.interrupt.injected) {
8516 static_call(kvm_x86_set_irq)(vcpu);
8521 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8522 vcpu->arch.exception.pending);
8525 * Call check_nested_events() even if we reinjected a previous event
8526 * in order for caller to determine if it should require immediate-exit
8527 * from L2 to L1 due to pending L1 events which require exit
8530 if (is_guest_mode(vcpu)) {
8531 r = kvm_check_nested_events(vcpu);
8536 /* try to inject new event if pending */
8537 if (vcpu->arch.exception.pending) {
8538 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8539 vcpu->arch.exception.has_error_code,
8540 vcpu->arch.exception.error_code);
8542 vcpu->arch.exception.pending = false;
8543 vcpu->arch.exception.injected = true;
8545 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8546 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8549 if (vcpu->arch.exception.nr == DB_VECTOR) {
8550 kvm_deliver_exception_payload(vcpu);
8551 if (vcpu->arch.dr7 & DR7_GD) {
8552 vcpu->arch.dr7 &= ~DR7_GD;
8553 kvm_update_dr7(vcpu);
8557 kvm_inject_exception(vcpu);
8562 * Finally, inject interrupt events. If an event cannot be injected
8563 * due to architectural conditions (e.g. IF=0) a window-open exit
8564 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8565 * and can architecturally be injected, but we cannot do it right now:
8566 * an interrupt could have arrived just now and we have to inject it
8567 * as a vmexit, or there could already an event in the queue, which is
8568 * indicated by can_inject. In that case we request an immediate exit
8569 * in order to make progress and get back here for another iteration.
8570 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8572 if (vcpu->arch.smi_pending) {
8573 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8577 vcpu->arch.smi_pending = false;
8578 ++vcpu->arch.smi_count;
8582 static_call(kvm_x86_enable_smi_window)(vcpu);
8585 if (vcpu->arch.nmi_pending) {
8586 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8590 --vcpu->arch.nmi_pending;
8591 vcpu->arch.nmi_injected = true;
8592 static_call(kvm_x86_set_nmi)(vcpu);
8594 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8596 if (vcpu->arch.nmi_pending)
8597 static_call(kvm_x86_enable_nmi_window)(vcpu);
8600 if (kvm_cpu_has_injectable_intr(vcpu)) {
8601 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8605 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8606 static_call(kvm_x86_set_irq)(vcpu);
8607 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8609 if (kvm_cpu_has_injectable_intr(vcpu))
8610 static_call(kvm_x86_enable_irq_window)(vcpu);
8613 if (is_guest_mode(vcpu) &&
8614 kvm_x86_ops.nested_ops->hv_timer_pending &&
8615 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8616 *req_immediate_exit = true;
8618 WARN_ON(vcpu->arch.exception.pending);
8622 *req_immediate_exit = true;
8626 static void process_nmi(struct kvm_vcpu *vcpu)
8631 * x86 is limited to one NMI running, and one NMI pending after it.
8632 * If an NMI is already in progress, limit further NMIs to just one.
8633 * Otherwise, allow two (and we'll inject the first one immediately).
8635 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8638 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8639 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8640 kvm_make_request(KVM_REQ_EVENT, vcpu);
8643 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8646 flags |= seg->g << 23;
8647 flags |= seg->db << 22;
8648 flags |= seg->l << 21;
8649 flags |= seg->avl << 20;
8650 flags |= seg->present << 15;
8651 flags |= seg->dpl << 13;
8652 flags |= seg->s << 12;
8653 flags |= seg->type << 8;
8657 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8659 struct kvm_segment seg;
8662 kvm_get_segment(vcpu, &seg, n);
8663 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8666 offset = 0x7f84 + n * 12;
8668 offset = 0x7f2c + (n - 3) * 12;
8670 put_smstate(u32, buf, offset + 8, seg.base);
8671 put_smstate(u32, buf, offset + 4, seg.limit);
8672 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8675 #ifdef CONFIG_X86_64
8676 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8678 struct kvm_segment seg;
8682 kvm_get_segment(vcpu, &seg, n);
8683 offset = 0x7e00 + n * 16;
8685 flags = enter_smm_get_segment_flags(&seg) >> 8;
8686 put_smstate(u16, buf, offset, seg.selector);
8687 put_smstate(u16, buf, offset + 2, flags);
8688 put_smstate(u32, buf, offset + 4, seg.limit);
8689 put_smstate(u64, buf, offset + 8, seg.base);
8693 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8696 struct kvm_segment seg;
8700 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8701 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8702 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8703 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8705 for (i = 0; i < 8; i++)
8706 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
8708 kvm_get_dr(vcpu, 6, &val);
8709 put_smstate(u32, buf, 0x7fcc, (u32)val);
8710 kvm_get_dr(vcpu, 7, &val);
8711 put_smstate(u32, buf, 0x7fc8, (u32)val);
8713 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8714 put_smstate(u32, buf, 0x7fc4, seg.selector);
8715 put_smstate(u32, buf, 0x7f64, seg.base);
8716 put_smstate(u32, buf, 0x7f60, seg.limit);
8717 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8719 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8720 put_smstate(u32, buf, 0x7fc0, seg.selector);
8721 put_smstate(u32, buf, 0x7f80, seg.base);
8722 put_smstate(u32, buf, 0x7f7c, seg.limit);
8723 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8725 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8726 put_smstate(u32, buf, 0x7f74, dt.address);
8727 put_smstate(u32, buf, 0x7f70, dt.size);
8729 static_call(kvm_x86_get_idt)(vcpu, &dt);
8730 put_smstate(u32, buf, 0x7f58, dt.address);
8731 put_smstate(u32, buf, 0x7f54, dt.size);
8733 for (i = 0; i < 6; i++)
8734 enter_smm_save_seg_32(vcpu, buf, i);
8736 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8739 put_smstate(u32, buf, 0x7efc, 0x00020000);
8740 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8743 #ifdef CONFIG_X86_64
8744 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8747 struct kvm_segment seg;
8751 for (i = 0; i < 16; i++)
8752 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
8754 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8755 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8757 kvm_get_dr(vcpu, 6, &val);
8758 put_smstate(u64, buf, 0x7f68, val);
8759 kvm_get_dr(vcpu, 7, &val);
8760 put_smstate(u64, buf, 0x7f60, val);
8762 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8763 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8764 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8766 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8769 put_smstate(u32, buf, 0x7efc, 0x00020064);
8771 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8773 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8774 put_smstate(u16, buf, 0x7e90, seg.selector);
8775 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8776 put_smstate(u32, buf, 0x7e94, seg.limit);
8777 put_smstate(u64, buf, 0x7e98, seg.base);
8779 static_call(kvm_x86_get_idt)(vcpu, &dt);
8780 put_smstate(u32, buf, 0x7e84, dt.size);
8781 put_smstate(u64, buf, 0x7e88, dt.address);
8783 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8784 put_smstate(u16, buf, 0x7e70, seg.selector);
8785 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8786 put_smstate(u32, buf, 0x7e74, seg.limit);
8787 put_smstate(u64, buf, 0x7e78, seg.base);
8789 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8790 put_smstate(u32, buf, 0x7e64, dt.size);
8791 put_smstate(u64, buf, 0x7e68, dt.address);
8793 for (i = 0; i < 6; i++)
8794 enter_smm_save_seg_64(vcpu, buf, i);
8798 static void enter_smm(struct kvm_vcpu *vcpu)
8800 struct kvm_segment cs, ds;
8805 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8806 memset(buf, 0, 512);
8807 #ifdef CONFIG_X86_64
8808 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8809 enter_smm_save_state_64(vcpu, buf);
8812 enter_smm_save_state_32(vcpu, buf);
8815 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8816 * vCPU state (e.g. leave guest mode) after we've saved the state into
8817 * the SMM state-save area.
8819 static_call(kvm_x86_pre_enter_smm)(vcpu, buf);
8821 vcpu->arch.hflags |= HF_SMM_MASK;
8822 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8824 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
8825 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8827 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
8829 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8830 kvm_rip_write(vcpu, 0x8000);
8832 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8833 static_call(kvm_x86_set_cr0)(vcpu, cr0);
8834 vcpu->arch.cr0 = cr0;
8836 static_call(kvm_x86_set_cr4)(vcpu, 0);
8838 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8839 dt.address = dt.size = 0;
8840 static_call(kvm_x86_set_idt)(vcpu, &dt);
8842 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8844 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8845 cs.base = vcpu->arch.smbase;
8850 cs.limit = ds.limit = 0xffffffff;
8851 cs.type = ds.type = 0x3;
8852 cs.dpl = ds.dpl = 0;
8857 cs.avl = ds.avl = 0;
8858 cs.present = ds.present = 1;
8859 cs.unusable = ds.unusable = 0;
8860 cs.padding = ds.padding = 0;
8862 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8863 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8864 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8865 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8866 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8867 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8869 #ifdef CONFIG_X86_64
8870 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8871 static_call(kvm_x86_set_efer)(vcpu, 0);
8874 kvm_update_cpuid_runtime(vcpu);
8875 kvm_mmu_reset_context(vcpu);
8878 static void process_smi(struct kvm_vcpu *vcpu)
8880 vcpu->arch.smi_pending = true;
8881 kvm_make_request(KVM_REQ_EVENT, vcpu);
8884 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8885 unsigned long *vcpu_bitmap)
8889 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8891 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8892 NULL, vcpu_bitmap, cpus);
8894 free_cpumask_var(cpus);
8897 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8899 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8902 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8904 if (!lapic_in_kernel(vcpu))
8907 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8908 kvm_apic_update_apicv(vcpu);
8909 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
8911 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8914 * NOTE: Do not hold any lock prior to calling this.
8916 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8917 * locked, because it calls __x86_set_memory_region() which does
8918 * synchronize_srcu(&kvm->srcu).
8920 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8922 struct kvm_vcpu *except;
8923 unsigned long old, new, expected;
8925 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8926 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
8929 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8931 expected = new = old;
8933 __clear_bit(bit, &new);
8935 __set_bit(bit, &new);
8938 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8939 } while (old != expected);
8944 trace_kvm_apicv_update_request(activate, bit);
8945 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8946 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
8949 * Sending request to update APICV for all other vcpus,
8950 * while update the calling vcpu immediately instead of
8951 * waiting for another #VMEXIT to handle the request.
8953 except = kvm_get_running_vcpu();
8954 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8957 kvm_vcpu_update_apicv(except);
8959 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8961 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8963 if (!kvm_apic_present(vcpu))
8966 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8968 if (irqchip_split(vcpu->kvm))
8969 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8971 if (vcpu->arch.apicv_active)
8972 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
8973 if (ioapic_in_kernel(vcpu->kvm))
8974 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8977 if (is_guest_mode(vcpu))
8978 vcpu->arch.load_eoi_exitmap_pending = true;
8980 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8983 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8985 u64 eoi_exit_bitmap[4];
8987 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8990 if (to_hv_vcpu(vcpu))
8991 bitmap_or((ulong *)eoi_exit_bitmap,
8992 vcpu->arch.ioapic_handled_vectors,
8993 to_hv_synic(vcpu)->vec_bitmap, 256);
8995 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
8998 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8999 unsigned long start, unsigned long end)
9001 unsigned long apic_address;
9004 * The physical address of apic access page is stored in the VMCS.
9005 * Update it when it becomes invalid.
9007 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9008 if (start <= apic_address && apic_address < end)
9009 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9012 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9014 if (!lapic_in_kernel(vcpu))
9017 if (!kvm_x86_ops.set_apic_access_page_addr)
9020 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9023 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9025 smp_send_reschedule(vcpu->cpu);
9027 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9030 * Returns 1 to let vcpu_run() continue the guest execution loop without
9031 * exiting to the userspace. Otherwise, the value will be returned to the
9034 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9038 dm_request_for_irq_injection(vcpu) &&
9039 kvm_cpu_accept_dm_intr(vcpu);
9040 fastpath_t exit_fastpath;
9042 bool req_immediate_exit = false;
9044 /* Forbid vmenter if vcpu dirty ring is soft-full */
9045 if (unlikely(vcpu->kvm->dirty_ring_size &&
9046 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9047 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9048 trace_kvm_dirty_ring_exit(vcpu);
9053 if (kvm_request_pending(vcpu)) {
9054 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9055 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9060 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9061 kvm_mmu_unload(vcpu);
9062 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9063 __kvm_migrate_timers(vcpu);
9064 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9065 kvm_gen_update_masterclock(vcpu->kvm);
9066 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9067 kvm_gen_kvmclock_update(vcpu);
9068 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9069 r = kvm_guest_time_update(vcpu);
9073 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9074 kvm_mmu_sync_roots(vcpu);
9075 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9076 kvm_mmu_load_pgd(vcpu);
9077 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9078 kvm_vcpu_flush_tlb_all(vcpu);
9080 /* Flushing all ASIDs flushes the current ASID... */
9081 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9083 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9084 kvm_vcpu_flush_tlb_current(vcpu);
9085 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
9086 kvm_vcpu_flush_tlb_guest(vcpu);
9088 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9089 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9093 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9094 if (is_guest_mode(vcpu)) {
9095 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9097 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9098 vcpu->mmio_needed = 0;
9103 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9104 /* Page is swapped out. Do synthetic halt */
9105 vcpu->arch.apf.halted = true;
9109 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9110 record_steal_time(vcpu);
9111 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9113 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9115 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9116 kvm_pmu_handle_event(vcpu);
9117 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9118 kvm_pmu_deliver_pmi(vcpu);
9119 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9120 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9121 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9122 vcpu->arch.ioapic_handled_vectors)) {
9123 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9124 vcpu->run->eoi.vector =
9125 vcpu->arch.pending_ioapic_eoi;
9130 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9131 vcpu_scan_ioapic(vcpu);
9132 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9133 vcpu_load_eoi_exitmap(vcpu);
9134 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9135 kvm_vcpu_reload_apic_access_page(vcpu);
9136 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9137 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9138 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9142 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9143 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9144 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9148 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9149 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9151 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9152 vcpu->run->hyperv = hv_vcpu->exit;
9158 * KVM_REQ_HV_STIMER has to be processed after
9159 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9160 * depend on the guest clock being up-to-date
9162 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9163 kvm_hv_process_stimers(vcpu);
9164 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9165 kvm_vcpu_update_apicv(vcpu);
9166 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9167 kvm_check_async_pf_completion(vcpu);
9168 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9169 static_call(kvm_x86_msr_filter_changed)(vcpu);
9171 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9172 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9175 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9176 kvm_xen_has_interrupt(vcpu)) {
9177 ++vcpu->stat.req_event;
9178 kvm_apic_accept_events(vcpu);
9179 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9184 inject_pending_event(vcpu, &req_immediate_exit);
9186 static_call(kvm_x86_enable_irq_window)(vcpu);
9188 if (kvm_lapic_enabled(vcpu)) {
9189 update_cr8_intercept(vcpu);
9190 kvm_lapic_sync_to_vapic(vcpu);
9194 r = kvm_mmu_reload(vcpu);
9196 goto cancel_injection;
9201 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9204 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9205 * IPI are then delayed after guest entry, which ensures that they
9206 * result in virtual interrupt delivery.
9208 local_irq_disable();
9209 vcpu->mode = IN_GUEST_MODE;
9211 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9214 * 1) We should set ->mode before checking ->requests. Please see
9215 * the comment in kvm_vcpu_exiting_guest_mode().
9217 * 2) For APICv, we should set ->mode before checking PID.ON. This
9218 * pairs with the memory barrier implicit in pi_test_and_set_on
9219 * (see vmx_deliver_posted_interrupt).
9221 * 3) This also orders the write to mode from any reads to the page
9222 * tables done while the VCPU is running. Please see the comment
9223 * in kvm_flush_remote_tlbs.
9225 smp_mb__after_srcu_read_unlock();
9228 * This handles the case where a posted interrupt was
9229 * notified with kvm_vcpu_kick.
9231 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9232 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9234 if (kvm_vcpu_exit_request(vcpu)) {
9235 vcpu->mode = OUTSIDE_GUEST_MODE;
9239 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9241 goto cancel_injection;
9244 if (req_immediate_exit) {
9245 kvm_make_request(KVM_REQ_EVENT, vcpu);
9246 static_call(kvm_x86_request_immediate_exit)(vcpu);
9249 fpregs_assert_state_consistent();
9250 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9251 switch_fpu_return();
9253 if (unlikely(vcpu->arch.switch_db_regs)) {
9255 set_debugreg(vcpu->arch.eff_db[0], 0);
9256 set_debugreg(vcpu->arch.eff_db[1], 1);
9257 set_debugreg(vcpu->arch.eff_db[2], 2);
9258 set_debugreg(vcpu->arch.eff_db[3], 3);
9259 set_debugreg(vcpu->arch.dr6, 6);
9260 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9264 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9265 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9268 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9269 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9273 if (vcpu->arch.apicv_active)
9274 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9278 * Do this here before restoring debug registers on the host. And
9279 * since we do this before handling the vmexit, a DR access vmexit
9280 * can (a) read the correct value of the debug registers, (b) set
9281 * KVM_DEBUGREG_WONT_EXIT again.
9283 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9284 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9285 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9286 kvm_update_dr0123(vcpu);
9287 kvm_update_dr7(vcpu);
9288 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9292 * If the guest has used debug registers, at least dr7
9293 * will be disabled while returning to the host.
9294 * If we don't have active breakpoints in the host, we don't
9295 * care about the messed up debug address registers. But if
9296 * we have some of them active, restore the old state.
9298 if (hw_breakpoint_active())
9299 hw_breakpoint_restore();
9301 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9302 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9304 vcpu->mode = OUTSIDE_GUEST_MODE;
9307 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9310 * Consume any pending interrupts, including the possible source of
9311 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9312 * An instruction is required after local_irq_enable() to fully unblock
9313 * interrupts on processors that implement an interrupt shadow, the
9314 * stat.exits increment will do nicely.
9316 kvm_before_interrupt(vcpu);
9319 local_irq_disable();
9320 kvm_after_interrupt(vcpu);
9322 if (lapic_in_kernel(vcpu)) {
9323 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9324 if (delta != S64_MIN) {
9325 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9326 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9333 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9336 * Profile KVM exit RIPs:
9338 if (unlikely(prof_on == KVM_PROFILING)) {
9339 unsigned long rip = kvm_rip_read(vcpu);
9340 profile_hit(KVM_PROFILING, (void *)rip);
9343 if (unlikely(vcpu->arch.tsc_always_catchup))
9344 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9346 if (vcpu->arch.apic_attention)
9347 kvm_lapic_sync_from_vapic(vcpu);
9349 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9353 if (req_immediate_exit)
9354 kvm_make_request(KVM_REQ_EVENT, vcpu);
9355 static_call(kvm_x86_cancel_injection)(vcpu);
9356 if (unlikely(vcpu->arch.apic_attention))
9357 kvm_lapic_sync_from_vapic(vcpu);
9362 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9364 if (!kvm_arch_vcpu_runnable(vcpu) &&
9365 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9366 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9367 kvm_vcpu_block(vcpu);
9368 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9370 if (kvm_x86_ops.post_block)
9371 static_call(kvm_x86_post_block)(vcpu);
9373 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9377 kvm_apic_accept_events(vcpu);
9378 switch(vcpu->arch.mp_state) {
9379 case KVM_MP_STATE_HALTED:
9380 case KVM_MP_STATE_AP_RESET_HOLD:
9381 vcpu->arch.pv.pv_unhalted = false;
9382 vcpu->arch.mp_state =
9383 KVM_MP_STATE_RUNNABLE;
9385 case KVM_MP_STATE_RUNNABLE:
9386 vcpu->arch.apf.halted = false;
9388 case KVM_MP_STATE_INIT_RECEIVED:
9396 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9398 if (is_guest_mode(vcpu))
9399 kvm_check_nested_events(vcpu);
9401 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9402 !vcpu->arch.apf.halted);
9405 static int vcpu_run(struct kvm_vcpu *vcpu)
9408 struct kvm *kvm = vcpu->kvm;
9410 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9411 vcpu->arch.l1tf_flush_l1d = true;
9414 if (kvm_vcpu_running(vcpu)) {
9415 r = vcpu_enter_guest(vcpu);
9417 r = vcpu_block(kvm, vcpu);
9423 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9424 if (kvm_cpu_has_pending_timer(vcpu))
9425 kvm_inject_pending_timer_irqs(vcpu);
9427 if (dm_request_for_irq_injection(vcpu) &&
9428 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9430 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9431 ++vcpu->stat.request_irq_exits;
9435 if (__xfer_to_guest_mode_work_pending()) {
9436 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9437 r = xfer_to_guest_mode_handle_work(vcpu);
9440 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9444 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9449 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9453 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9454 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9455 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9459 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9461 BUG_ON(!vcpu->arch.pio.count);
9463 return complete_emulated_io(vcpu);
9467 * Implements the following, as a state machine:
9471 * for each mmio piece in the fragment
9479 * for each mmio piece in the fragment
9484 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9486 struct kvm_run *run = vcpu->run;
9487 struct kvm_mmio_fragment *frag;
9490 BUG_ON(!vcpu->mmio_needed);
9492 /* Complete previous fragment */
9493 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9494 len = min(8u, frag->len);
9495 if (!vcpu->mmio_is_write)
9496 memcpy(frag->data, run->mmio.data, len);
9498 if (frag->len <= 8) {
9499 /* Switch to the next fragment. */
9501 vcpu->mmio_cur_fragment++;
9503 /* Go forward to the next mmio piece. */
9509 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9510 vcpu->mmio_needed = 0;
9512 /* FIXME: return into emulator if single-stepping. */
9513 if (vcpu->mmio_is_write)
9515 vcpu->mmio_read_completed = 1;
9516 return complete_emulated_io(vcpu);
9519 run->exit_reason = KVM_EXIT_MMIO;
9520 run->mmio.phys_addr = frag->gpa;
9521 if (vcpu->mmio_is_write)
9522 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9523 run->mmio.len = min(8u, frag->len);
9524 run->mmio.is_write = vcpu->mmio_is_write;
9525 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9529 static void kvm_save_current_fpu(struct fpu *fpu)
9532 * If the target FPU state is not resident in the CPU registers, just
9533 * memcpy() from current, else save CPU state directly to the target.
9535 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9536 memcpy(&fpu->state, ¤t->thread.fpu.state,
9537 fpu_kernel_xstate_size);
9539 copy_fpregs_to_fpstate(fpu);
9542 /* Swap (qemu) user FPU context for the guest FPU context. */
9543 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9547 kvm_save_current_fpu(vcpu->arch.user_fpu);
9550 * Guests with protected state can't have it set by the hypervisor,
9551 * so skip trying to set it.
9553 if (vcpu->arch.guest_fpu)
9554 /* PKRU is separately restored in kvm_x86_ops.run. */
9555 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9556 ~XFEATURE_MASK_PKRU);
9558 fpregs_mark_activate();
9564 /* When vcpu_run ends, restore user space FPU context. */
9565 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9570 * Guests with protected state can't have it read by the hypervisor,
9571 * so skip trying to save it.
9573 if (vcpu->arch.guest_fpu)
9574 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9576 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9578 fpregs_mark_activate();
9581 ++vcpu->stat.fpu_reload;
9585 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9587 struct kvm_run *kvm_run = vcpu->run;
9591 kvm_sigset_activate(vcpu);
9593 kvm_load_guest_fpu(vcpu);
9595 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9596 if (kvm_run->immediate_exit) {
9600 kvm_vcpu_block(vcpu);
9601 kvm_apic_accept_events(vcpu);
9602 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9604 if (signal_pending(current)) {
9606 kvm_run->exit_reason = KVM_EXIT_INTR;
9607 ++vcpu->stat.signal_exits;
9612 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9617 if (kvm_run->kvm_dirty_regs) {
9618 r = sync_regs(vcpu);
9623 /* re-sync apic's tpr */
9624 if (!lapic_in_kernel(vcpu)) {
9625 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9631 if (unlikely(vcpu->arch.complete_userspace_io)) {
9632 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9633 vcpu->arch.complete_userspace_io = NULL;
9638 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9640 if (kvm_run->immediate_exit)
9646 kvm_put_guest_fpu(vcpu);
9647 if (kvm_run->kvm_valid_regs)
9649 post_kvm_run_save(vcpu);
9650 kvm_sigset_deactivate(vcpu);
9656 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9658 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9660 * We are here if userspace calls get_regs() in the middle of
9661 * instruction emulation. Registers state needs to be copied
9662 * back from emulation context to vcpu. Userspace shouldn't do
9663 * that usually, but some bad designed PV devices (vmware
9664 * backdoor interface) need this to work
9666 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9667 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9669 regs->rax = kvm_rax_read(vcpu);
9670 regs->rbx = kvm_rbx_read(vcpu);
9671 regs->rcx = kvm_rcx_read(vcpu);
9672 regs->rdx = kvm_rdx_read(vcpu);
9673 regs->rsi = kvm_rsi_read(vcpu);
9674 regs->rdi = kvm_rdi_read(vcpu);
9675 regs->rsp = kvm_rsp_read(vcpu);
9676 regs->rbp = kvm_rbp_read(vcpu);
9677 #ifdef CONFIG_X86_64
9678 regs->r8 = kvm_r8_read(vcpu);
9679 regs->r9 = kvm_r9_read(vcpu);
9680 regs->r10 = kvm_r10_read(vcpu);
9681 regs->r11 = kvm_r11_read(vcpu);
9682 regs->r12 = kvm_r12_read(vcpu);
9683 regs->r13 = kvm_r13_read(vcpu);
9684 regs->r14 = kvm_r14_read(vcpu);
9685 regs->r15 = kvm_r15_read(vcpu);
9688 regs->rip = kvm_rip_read(vcpu);
9689 regs->rflags = kvm_get_rflags(vcpu);
9692 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9695 __get_regs(vcpu, regs);
9700 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9702 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9703 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9705 kvm_rax_write(vcpu, regs->rax);
9706 kvm_rbx_write(vcpu, regs->rbx);
9707 kvm_rcx_write(vcpu, regs->rcx);
9708 kvm_rdx_write(vcpu, regs->rdx);
9709 kvm_rsi_write(vcpu, regs->rsi);
9710 kvm_rdi_write(vcpu, regs->rdi);
9711 kvm_rsp_write(vcpu, regs->rsp);
9712 kvm_rbp_write(vcpu, regs->rbp);
9713 #ifdef CONFIG_X86_64
9714 kvm_r8_write(vcpu, regs->r8);
9715 kvm_r9_write(vcpu, regs->r9);
9716 kvm_r10_write(vcpu, regs->r10);
9717 kvm_r11_write(vcpu, regs->r11);
9718 kvm_r12_write(vcpu, regs->r12);
9719 kvm_r13_write(vcpu, regs->r13);
9720 kvm_r14_write(vcpu, regs->r14);
9721 kvm_r15_write(vcpu, regs->r15);
9724 kvm_rip_write(vcpu, regs->rip);
9725 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9727 vcpu->arch.exception.pending = false;
9729 kvm_make_request(KVM_REQ_EVENT, vcpu);
9732 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9735 __set_regs(vcpu, regs);
9740 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9742 struct kvm_segment cs;
9744 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9748 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9750 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9754 if (vcpu->arch.guest_state_protected)
9755 goto skip_protected_regs;
9757 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9758 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9759 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9760 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9761 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9762 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9764 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9765 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9767 static_call(kvm_x86_get_idt)(vcpu, &dt);
9768 sregs->idt.limit = dt.size;
9769 sregs->idt.base = dt.address;
9770 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9771 sregs->gdt.limit = dt.size;
9772 sregs->gdt.base = dt.address;
9774 sregs->cr2 = vcpu->arch.cr2;
9775 sregs->cr3 = kvm_read_cr3(vcpu);
9777 skip_protected_regs:
9778 sregs->cr0 = kvm_read_cr0(vcpu);
9779 sregs->cr4 = kvm_read_cr4(vcpu);
9780 sregs->cr8 = kvm_get_cr8(vcpu);
9781 sregs->efer = vcpu->arch.efer;
9782 sregs->apic_base = kvm_get_apic_base(vcpu);
9784 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9786 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9787 set_bit(vcpu->arch.interrupt.nr,
9788 (unsigned long *)sregs->interrupt_bitmap);
9791 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9792 struct kvm_sregs *sregs)
9795 __get_sregs(vcpu, sregs);
9800 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9801 struct kvm_mp_state *mp_state)
9804 if (kvm_mpx_supported())
9805 kvm_load_guest_fpu(vcpu);
9807 kvm_apic_accept_events(vcpu);
9808 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9809 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9810 vcpu->arch.pv.pv_unhalted)
9811 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9813 mp_state->mp_state = vcpu->arch.mp_state;
9815 if (kvm_mpx_supported())
9816 kvm_put_guest_fpu(vcpu);
9821 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9822 struct kvm_mp_state *mp_state)
9828 if (!lapic_in_kernel(vcpu) &&
9829 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9833 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9834 * INIT state; latched init should be reported using
9835 * KVM_SET_VCPU_EVENTS, so reject it here.
9837 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9838 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9839 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9842 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9843 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9844 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9846 vcpu->arch.mp_state = mp_state->mp_state;
9847 kvm_make_request(KVM_REQ_EVENT, vcpu);
9855 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9856 int reason, bool has_error_code, u32 error_code)
9858 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9861 init_emulate_ctxt(vcpu);
9863 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9864 has_error_code, error_code);
9866 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9867 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9868 vcpu->run->internal.ndata = 0;
9872 kvm_rip_write(vcpu, ctxt->eip);
9873 kvm_set_rflags(vcpu, ctxt->eflags);
9876 EXPORT_SYMBOL_GPL(kvm_task_switch);
9878 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9880 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9882 * When EFER.LME and CR0.PG are set, the processor is in
9883 * 64-bit mode (though maybe in a 32-bit code segment).
9884 * CR4.PAE and EFER.LMA must be set.
9886 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9888 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
9892 * Not in 64-bit mode: EFER.LMA is clear and the code
9893 * segment cannot be 64-bit.
9895 if (sregs->efer & EFER_LMA || sregs->cs.l)
9899 return kvm_is_valid_cr4(vcpu, sregs->cr4);
9902 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9904 struct msr_data apic_base_msr;
9905 int mmu_reset_needed = 0;
9906 int pending_vec, max_bits, idx;
9910 if (!kvm_is_valid_sregs(vcpu, sregs))
9913 apic_base_msr.data = sregs->apic_base;
9914 apic_base_msr.host_initiated = true;
9915 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9918 if (vcpu->arch.guest_state_protected)
9919 goto skip_protected_regs;
9921 dt.size = sregs->idt.limit;
9922 dt.address = sregs->idt.base;
9923 static_call(kvm_x86_set_idt)(vcpu, &dt);
9924 dt.size = sregs->gdt.limit;
9925 dt.address = sregs->gdt.base;
9926 static_call(kvm_x86_set_gdt)(vcpu, &dt);
9928 vcpu->arch.cr2 = sregs->cr2;
9929 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9930 vcpu->arch.cr3 = sregs->cr3;
9931 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9933 kvm_set_cr8(vcpu, sregs->cr8);
9935 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9936 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
9938 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9939 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
9940 vcpu->arch.cr0 = sregs->cr0;
9942 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9943 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
9945 idx = srcu_read_lock(&vcpu->kvm->srcu);
9946 if (is_pae_paging(vcpu)) {
9947 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9948 mmu_reset_needed = 1;
9950 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9952 if (mmu_reset_needed)
9953 kvm_mmu_reset_context(vcpu);
9955 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9956 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9957 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9958 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9959 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9960 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9962 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9963 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9965 update_cr8_intercept(vcpu);
9967 /* Older userspace won't unhalt the vcpu on reset. */
9968 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9969 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9971 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9973 skip_protected_regs:
9974 max_bits = KVM_NR_INTERRUPTS;
9975 pending_vec = find_first_bit(
9976 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9977 if (pending_vec < max_bits) {
9978 kvm_queue_interrupt(vcpu, pending_vec, false);
9979 pr_debug("Set back pending irq %d\n", pending_vec);
9982 kvm_make_request(KVM_REQ_EVENT, vcpu);
9989 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9990 struct kvm_sregs *sregs)
9995 ret = __set_sregs(vcpu, sregs);
10000 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10001 struct kvm_guest_debug *dbg)
10003 unsigned long rflags;
10006 if (vcpu->arch.guest_state_protected)
10011 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10013 if (vcpu->arch.exception.pending)
10015 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10016 kvm_queue_exception(vcpu, DB_VECTOR);
10018 kvm_queue_exception(vcpu, BP_VECTOR);
10022 * Read rflags as long as potentially injected trace flags are still
10025 rflags = kvm_get_rflags(vcpu);
10027 vcpu->guest_debug = dbg->control;
10028 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10029 vcpu->guest_debug = 0;
10031 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10032 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10033 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10034 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10036 for (i = 0; i < KVM_NR_DB_REGS; i++)
10037 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10039 kvm_update_dr7(vcpu);
10041 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10042 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
10043 get_segment_base(vcpu, VCPU_SREG_CS);
10046 * Trigger an rflags update that will inject or remove the trace
10049 kvm_set_rflags(vcpu, rflags);
10051 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10061 * Translate a guest virtual address to a guest physical address.
10063 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10064 struct kvm_translation *tr)
10066 unsigned long vaddr = tr->linear_address;
10072 idx = srcu_read_lock(&vcpu->kvm->srcu);
10073 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10074 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10075 tr->physical_address = gpa;
10076 tr->valid = gpa != UNMAPPED_GVA;
10084 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10086 struct fxregs_state *fxsave;
10088 if (!vcpu->arch.guest_fpu)
10093 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10094 memcpy(fpu->fpr, fxsave->st_space, 128);
10095 fpu->fcw = fxsave->cwd;
10096 fpu->fsw = fxsave->swd;
10097 fpu->ftwx = fxsave->twd;
10098 fpu->last_opcode = fxsave->fop;
10099 fpu->last_ip = fxsave->rip;
10100 fpu->last_dp = fxsave->rdp;
10101 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10107 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10109 struct fxregs_state *fxsave;
10111 if (!vcpu->arch.guest_fpu)
10116 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10118 memcpy(fxsave->st_space, fpu->fpr, 128);
10119 fxsave->cwd = fpu->fcw;
10120 fxsave->swd = fpu->fsw;
10121 fxsave->twd = fpu->ftwx;
10122 fxsave->fop = fpu->last_opcode;
10123 fxsave->rip = fpu->last_ip;
10124 fxsave->rdp = fpu->last_dp;
10125 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10131 static void store_regs(struct kvm_vcpu *vcpu)
10133 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10135 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10136 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10138 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10139 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10141 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10142 kvm_vcpu_ioctl_x86_get_vcpu_events(
10143 vcpu, &vcpu->run->s.regs.events);
10146 static int sync_regs(struct kvm_vcpu *vcpu)
10148 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10151 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10152 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10153 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10155 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10156 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10158 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10160 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10161 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10162 vcpu, &vcpu->run->s.regs.events))
10164 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10170 static void fx_init(struct kvm_vcpu *vcpu)
10172 if (!vcpu->arch.guest_fpu)
10175 fpstate_init(&vcpu->arch.guest_fpu->state);
10176 if (boot_cpu_has(X86_FEATURE_XSAVES))
10177 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10178 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10181 * Ensure guest xcr0 is valid for loading
10183 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10185 vcpu->arch.cr0 |= X86_CR0_ET;
10188 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10190 if (vcpu->arch.guest_fpu) {
10191 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10192 vcpu->arch.guest_fpu = NULL;
10195 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10197 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10199 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10200 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10201 "guest TSC will not be reliable\n");
10206 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10211 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10212 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10214 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10216 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10218 r = kvm_mmu_create(vcpu);
10222 if (irqchip_in_kernel(vcpu->kvm)) {
10223 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10225 goto fail_mmu_destroy;
10226 if (kvm_apicv_activated(vcpu->kvm))
10227 vcpu->arch.apicv_active = true;
10229 static_branch_inc(&kvm_has_noapic_vcpu);
10233 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10235 goto fail_free_lapic;
10236 vcpu->arch.pio_data = page_address(page);
10238 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10239 GFP_KERNEL_ACCOUNT);
10240 if (!vcpu->arch.mce_banks)
10241 goto fail_free_pio_data;
10242 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10244 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10245 GFP_KERNEL_ACCOUNT))
10246 goto fail_free_mce_banks;
10248 if (!alloc_emulate_ctxt(vcpu))
10249 goto free_wbinvd_dirty_mask;
10251 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10252 GFP_KERNEL_ACCOUNT);
10253 if (!vcpu->arch.user_fpu) {
10254 pr_err("kvm: failed to allocate userspace's fpu\n");
10255 goto free_emulate_ctxt;
10258 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10259 GFP_KERNEL_ACCOUNT);
10260 if (!vcpu->arch.guest_fpu) {
10261 pr_err("kvm: failed to allocate vcpu's fpu\n");
10262 goto free_user_fpu;
10266 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10267 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10269 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10271 kvm_async_pf_hash_reset(vcpu);
10272 kvm_pmu_init(vcpu);
10274 vcpu->arch.pending_external_vector = -1;
10275 vcpu->arch.preempted_in_kernel = false;
10277 r = static_call(kvm_x86_vcpu_create)(vcpu);
10279 goto free_guest_fpu;
10281 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10282 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10283 kvm_vcpu_mtrr_init(vcpu);
10285 kvm_vcpu_reset(vcpu, false);
10286 kvm_init_mmu(vcpu, false);
10291 kvm_free_guest_fpu(vcpu);
10293 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10295 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10296 free_wbinvd_dirty_mask:
10297 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10298 fail_free_mce_banks:
10299 kfree(vcpu->arch.mce_banks);
10300 fail_free_pio_data:
10301 free_page((unsigned long)vcpu->arch.pio_data);
10303 kvm_free_lapic(vcpu);
10305 kvm_mmu_destroy(vcpu);
10309 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10311 struct kvm *kvm = vcpu->kvm;
10313 if (mutex_lock_killable(&vcpu->mutex))
10316 kvm_synchronize_tsc(vcpu, 0);
10319 /* poll control enabled by default */
10320 vcpu->arch.msr_kvm_poll_control = 1;
10322 mutex_unlock(&vcpu->mutex);
10324 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10325 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10326 KVMCLOCK_SYNC_PERIOD);
10329 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10331 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10334 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10336 kvmclock_reset(vcpu);
10338 static_call(kvm_x86_vcpu_free)(vcpu);
10340 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10341 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10342 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10343 kvm_free_guest_fpu(vcpu);
10345 kvm_hv_vcpu_uninit(vcpu);
10346 kvm_pmu_destroy(vcpu);
10347 kfree(vcpu->arch.mce_banks);
10348 kvm_free_lapic(vcpu);
10349 idx = srcu_read_lock(&vcpu->kvm->srcu);
10350 kvm_mmu_destroy(vcpu);
10351 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10352 free_page((unsigned long)vcpu->arch.pio_data);
10353 kvfree(vcpu->arch.cpuid_entries);
10354 if (!lapic_in_kernel(vcpu))
10355 static_branch_dec(&kvm_has_noapic_vcpu);
10358 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10360 kvm_lapic_reset(vcpu, init_event);
10362 vcpu->arch.hflags = 0;
10364 vcpu->arch.smi_pending = 0;
10365 vcpu->arch.smi_count = 0;
10366 atomic_set(&vcpu->arch.nmi_queued, 0);
10367 vcpu->arch.nmi_pending = 0;
10368 vcpu->arch.nmi_injected = false;
10369 kvm_clear_interrupt_queue(vcpu);
10370 kvm_clear_exception_queue(vcpu);
10372 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10373 kvm_update_dr0123(vcpu);
10374 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10375 vcpu->arch.dr7 = DR7_FIXED_1;
10376 kvm_update_dr7(vcpu);
10378 vcpu->arch.cr2 = 0;
10380 kvm_make_request(KVM_REQ_EVENT, vcpu);
10381 vcpu->arch.apf.msr_en_val = 0;
10382 vcpu->arch.apf.msr_int_val = 0;
10383 vcpu->arch.st.msr_val = 0;
10385 kvmclock_reset(vcpu);
10387 kvm_clear_async_pf_completion_queue(vcpu);
10388 kvm_async_pf_hash_reset(vcpu);
10389 vcpu->arch.apf.halted = false;
10391 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10392 void *mpx_state_buffer;
10395 * To avoid have the INIT path from kvm_apic_has_events() that be
10396 * called with loaded FPU and does not let userspace fix the state.
10399 kvm_put_guest_fpu(vcpu);
10400 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10402 if (mpx_state_buffer)
10403 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10404 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10406 if (mpx_state_buffer)
10407 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10409 kvm_load_guest_fpu(vcpu);
10413 kvm_pmu_reset(vcpu);
10414 vcpu->arch.smbase = 0x30000;
10416 vcpu->arch.msr_misc_features_enables = 0;
10418 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10421 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10422 vcpu->arch.regs_avail = ~0;
10423 vcpu->arch.regs_dirty = ~0;
10425 vcpu->arch.ia32_xss = 0;
10427 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10430 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10432 struct kvm_segment cs;
10434 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10435 cs.selector = vector << 8;
10436 cs.base = vector << 12;
10437 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10438 kvm_rip_write(vcpu, 0);
10440 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10442 int kvm_arch_hardware_enable(void)
10445 struct kvm_vcpu *vcpu;
10450 bool stable, backwards_tsc = false;
10452 kvm_user_return_msr_cpu_online();
10453 ret = static_call(kvm_x86_hardware_enable)();
10457 local_tsc = rdtsc();
10458 stable = !kvm_check_tsc_unstable();
10459 list_for_each_entry(kvm, &vm_list, vm_list) {
10460 kvm_for_each_vcpu(i, vcpu, kvm) {
10461 if (!stable && vcpu->cpu == smp_processor_id())
10462 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10463 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10464 backwards_tsc = true;
10465 if (vcpu->arch.last_host_tsc > max_tsc)
10466 max_tsc = vcpu->arch.last_host_tsc;
10472 * Sometimes, even reliable TSCs go backwards. This happens on
10473 * platforms that reset TSC during suspend or hibernate actions, but
10474 * maintain synchronization. We must compensate. Fortunately, we can
10475 * detect that condition here, which happens early in CPU bringup,
10476 * before any KVM threads can be running. Unfortunately, we can't
10477 * bring the TSCs fully up to date with real time, as we aren't yet far
10478 * enough into CPU bringup that we know how much real time has actually
10479 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10480 * variables that haven't been updated yet.
10482 * So we simply find the maximum observed TSC above, then record the
10483 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10484 * the adjustment will be applied. Note that we accumulate
10485 * adjustments, in case multiple suspend cycles happen before some VCPU
10486 * gets a chance to run again. In the event that no KVM threads get a
10487 * chance to run, we will miss the entire elapsed period, as we'll have
10488 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10489 * loose cycle time. This isn't too big a deal, since the loss will be
10490 * uniform across all VCPUs (not to mention the scenario is extremely
10491 * unlikely). It is possible that a second hibernate recovery happens
10492 * much faster than a first, causing the observed TSC here to be
10493 * smaller; this would require additional padding adjustment, which is
10494 * why we set last_host_tsc to the local tsc observed here.
10496 * N.B. - this code below runs only on platforms with reliable TSC,
10497 * as that is the only way backwards_tsc is set above. Also note
10498 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10499 * have the same delta_cyc adjustment applied if backwards_tsc
10500 * is detected. Note further, this adjustment is only done once,
10501 * as we reset last_host_tsc on all VCPUs to stop this from being
10502 * called multiple times (one for each physical CPU bringup).
10504 * Platforms with unreliable TSCs don't have to deal with this, they
10505 * will be compensated by the logic in vcpu_load, which sets the TSC to
10506 * catchup mode. This will catchup all VCPUs to real time, but cannot
10507 * guarantee that they stay in perfect synchronization.
10509 if (backwards_tsc) {
10510 u64 delta_cyc = max_tsc - local_tsc;
10511 list_for_each_entry(kvm, &vm_list, vm_list) {
10512 kvm->arch.backwards_tsc_observed = true;
10513 kvm_for_each_vcpu(i, vcpu, kvm) {
10514 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10515 vcpu->arch.last_host_tsc = local_tsc;
10516 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10520 * We have to disable TSC offset matching.. if you were
10521 * booting a VM while issuing an S4 host suspend....
10522 * you may have some problem. Solving this issue is
10523 * left as an exercise to the reader.
10525 kvm->arch.last_tsc_nsec = 0;
10526 kvm->arch.last_tsc_write = 0;
10533 void kvm_arch_hardware_disable(void)
10535 static_call(kvm_x86_hardware_disable)();
10536 drop_user_return_notifiers();
10539 int kvm_arch_hardware_setup(void *opaque)
10541 struct kvm_x86_init_ops *ops = opaque;
10544 rdmsrl_safe(MSR_EFER, &host_efer);
10546 if (boot_cpu_has(X86_FEATURE_XSAVES))
10547 rdmsrl(MSR_IA32_XSS, host_xss);
10549 r = ops->hardware_setup();
10553 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10554 kvm_ops_static_call_update();
10556 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10559 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10560 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10561 #undef __kvm_cpu_cap_has
10563 if (kvm_has_tsc_control) {
10565 * Make sure the user can only configure tsc_khz values that
10566 * fit into a signed integer.
10567 * A min value is not calculated because it will always
10568 * be 1 on all machines.
10570 u64 max = min(0x7fffffffULL,
10571 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10572 kvm_max_guest_tsc_khz = max;
10574 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10577 kvm_init_msr_list();
10581 void kvm_arch_hardware_unsetup(void)
10583 static_call(kvm_x86_hardware_unsetup)();
10586 int kvm_arch_check_processor_compat(void *opaque)
10588 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10589 struct kvm_x86_init_ops *ops = opaque;
10591 WARN_ON(!irqs_disabled());
10593 if (__cr4_reserved_bits(cpu_has, c) !=
10594 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10597 return ops->check_processor_compatibility();
10600 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10602 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10604 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10606 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10608 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10611 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10612 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10614 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10616 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10618 vcpu->arch.l1tf_flush_l1d = true;
10619 if (pmu->version && unlikely(pmu->event_count)) {
10620 pmu->need_cleanup = true;
10621 kvm_make_request(KVM_REQ_PMU, vcpu);
10623 static_call(kvm_x86_sched_in)(vcpu, cpu);
10626 void kvm_arch_free_vm(struct kvm *kvm)
10628 kfree(to_kvm_hv(kvm)->hv_pa_pg);
10633 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10638 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10639 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10640 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10641 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10642 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10643 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10645 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10646 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10647 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10648 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10649 &kvm->arch.irq_sources_bitmap);
10651 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10652 mutex_init(&kvm->arch.apic_map_lock);
10653 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10655 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10656 pvclock_update_vm_gtod_copy(kvm);
10658 kvm->arch.guest_can_read_msr_platform_info = true;
10660 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10661 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10663 kvm_hv_init_vm(kvm);
10664 kvm_page_track_init(kvm);
10665 kvm_mmu_init_vm(kvm);
10667 return static_call(kvm_x86_vm_init)(kvm);
10670 int kvm_arch_post_init_vm(struct kvm *kvm)
10672 return kvm_mmu_post_init_vm(kvm);
10675 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10678 kvm_mmu_unload(vcpu);
10682 static void kvm_free_vcpus(struct kvm *kvm)
10685 struct kvm_vcpu *vcpu;
10688 * Unpin any mmu pages first.
10690 kvm_for_each_vcpu(i, vcpu, kvm) {
10691 kvm_clear_async_pf_completion_queue(vcpu);
10692 kvm_unload_vcpu_mmu(vcpu);
10694 kvm_for_each_vcpu(i, vcpu, kvm)
10695 kvm_vcpu_destroy(vcpu);
10697 mutex_lock(&kvm->lock);
10698 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10699 kvm->vcpus[i] = NULL;
10701 atomic_set(&kvm->online_vcpus, 0);
10702 mutex_unlock(&kvm->lock);
10705 void kvm_arch_sync_events(struct kvm *kvm)
10707 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10708 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10712 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
10715 * __x86_set_memory_region: Setup KVM internal memory slot
10717 * @kvm: the kvm pointer to the VM.
10718 * @id: the slot ID to setup.
10719 * @gpa: the GPA to install the slot (unused when @size == 0).
10720 * @size: the size of the slot. Set to zero to uninstall a slot.
10722 * This function helps to setup a KVM internal memory slot. Specify
10723 * @size > 0 to install a new slot, while @size == 0 to uninstall a
10724 * slot. The return code can be one of the following:
10726 * HVA: on success (uninstall will return a bogus HVA)
10729 * The caller should always use IS_ERR() to check the return value
10730 * before use. Note, the KVM internal memory slots are guaranteed to
10731 * remain valid and unchanged until the VM is destroyed, i.e., the
10732 * GPA->HVA translation will not change. However, the HVA is a user
10733 * address, i.e. its accessibility is not guaranteed, and must be
10734 * accessed via __copy_{to,from}_user().
10736 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10740 unsigned long hva, old_npages;
10741 struct kvm_memslots *slots = kvm_memslots(kvm);
10742 struct kvm_memory_slot *slot;
10744 /* Called with kvm->slots_lock held. */
10745 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10746 return ERR_PTR_USR(-EINVAL);
10748 slot = id_to_memslot(slots, id);
10750 if (slot && slot->npages)
10751 return ERR_PTR_USR(-EEXIST);
10754 * MAP_SHARED to prevent internal slot pages from being moved
10757 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10758 MAP_SHARED | MAP_ANONYMOUS, 0);
10759 if (IS_ERR((void *)hva))
10760 return (void __user *)hva;
10762 if (!slot || !slot->npages)
10765 old_npages = slot->npages;
10766 hva = slot->userspace_addr;
10769 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10770 struct kvm_userspace_memory_region m;
10772 m.slot = id | (i << 16);
10774 m.guest_phys_addr = gpa;
10775 m.userspace_addr = hva;
10776 m.memory_size = size;
10777 r = __kvm_set_memory_region(kvm, &m);
10779 return ERR_PTR_USR(r);
10783 vm_munmap(hva, old_npages * PAGE_SIZE);
10785 return (void __user *)hva;
10787 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10789 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10791 kvm_mmu_pre_destroy_vm(kvm);
10794 void kvm_arch_destroy_vm(struct kvm *kvm)
10796 if (current->mm == kvm->mm) {
10798 * Free memory regions allocated on behalf of userspace,
10799 * unless the the memory map has changed due to process exit
10802 mutex_lock(&kvm->slots_lock);
10803 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10805 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10807 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10808 mutex_unlock(&kvm->slots_lock);
10810 static_call_cond(kvm_x86_vm_destroy)(kvm);
10811 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
10812 kvm_pic_destroy(kvm);
10813 kvm_ioapic_destroy(kvm);
10814 kvm_free_vcpus(kvm);
10815 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10816 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10817 kvm_mmu_uninit_vm(kvm);
10818 kvm_page_track_cleanup(kvm);
10819 kvm_xen_destroy_vm(kvm);
10820 kvm_hv_destroy_vm(kvm);
10823 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10827 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10828 kvfree(slot->arch.rmap[i]);
10829 slot->arch.rmap[i] = NULL;
10834 kvfree(slot->arch.lpage_info[i - 1]);
10835 slot->arch.lpage_info[i - 1] = NULL;
10838 kvm_page_track_free_memslot(slot);
10841 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10842 unsigned long npages)
10847 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10848 * old arrays will be freed by __kvm_set_memory_region() if installing
10849 * the new memslot is successful.
10851 memset(&slot->arch, 0, sizeof(slot->arch));
10853 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10854 struct kvm_lpage_info *linfo;
10855 unsigned long ugfn;
10859 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10860 slot->base_gfn, level) + 1;
10862 slot->arch.rmap[i] =
10863 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10864 GFP_KERNEL_ACCOUNT);
10865 if (!slot->arch.rmap[i])
10870 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10874 slot->arch.lpage_info[i - 1] = linfo;
10876 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10877 linfo[0].disallow_lpage = 1;
10878 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10879 linfo[lpages - 1].disallow_lpage = 1;
10880 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10882 * If the gfn and userspace address are not aligned wrt each
10883 * other, disable large page support for this slot.
10885 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10888 for (j = 0; j < lpages; ++j)
10889 linfo[j].disallow_lpage = 1;
10893 if (kvm_page_track_create_memslot(slot, npages))
10899 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10900 kvfree(slot->arch.rmap[i]);
10901 slot->arch.rmap[i] = NULL;
10905 kvfree(slot->arch.lpage_info[i - 1]);
10906 slot->arch.lpage_info[i - 1] = NULL;
10911 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10913 struct kvm_vcpu *vcpu;
10917 * memslots->generation has been incremented.
10918 * mmio generation may have reached its maximum value.
10920 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10922 /* Force re-initialization of steal_time cache */
10923 kvm_for_each_vcpu(i, vcpu, kvm)
10924 kvm_vcpu_kick(vcpu);
10927 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10928 struct kvm_memory_slot *memslot,
10929 const struct kvm_userspace_memory_region *mem,
10930 enum kvm_mr_change change)
10932 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10933 return kvm_alloc_memslot_metadata(memslot,
10934 mem->memory_size >> PAGE_SHIFT);
10939 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
10941 struct kvm_arch *ka = &kvm->arch;
10943 if (!kvm_x86_ops.cpu_dirty_log_size)
10946 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
10947 (!enable && --ka->cpu_dirty_logging_count == 0))
10948 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
10950 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
10953 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10954 struct kvm_memory_slot *old,
10955 struct kvm_memory_slot *new,
10956 enum kvm_mr_change change)
10958 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
10961 * Update CPU dirty logging if dirty logging is being toggled. This
10962 * applies to all operations.
10964 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
10965 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
10968 * Nothing more to do for RO slots (which can't be dirtied and can't be
10969 * made writable) or CREATE/MOVE/DELETE of a slot.
10971 * For a memslot with dirty logging disabled:
10972 * CREATE: No dirty mappings will already exist.
10973 * MOVE/DELETE: The old mappings will already have been cleaned up by
10974 * kvm_arch_flush_shadow_memslot()
10976 * For a memslot with dirty logging enabled:
10977 * CREATE: No shadow pages exist, thus nothing to write-protect
10978 * and no dirty bits to clear.
10979 * MOVE/DELETE: The old mappings will already have been cleaned up by
10980 * kvm_arch_flush_shadow_memslot().
10982 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10986 * READONLY and non-flags changes were filtered out above, and the only
10987 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
10988 * logging isn't being toggled on or off.
10990 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
10993 if (!log_dirty_pages) {
10995 * Dirty logging tracks sptes in 4k granularity, meaning that
10996 * large sptes have to be split. If live migration succeeds,
10997 * the guest in the source machine will be destroyed and large
10998 * sptes will be created in the destination. However, if the
10999 * guest continues to run in the source machine (for example if
11000 * live migration fails), small sptes will remain around and
11001 * cause bad performance.
11003 * Scan sptes if dirty logging has been stopped, dropping those
11004 * which can be collapsed into a single large-page spte. Later
11005 * page faults will create the large-page sptes.
11007 kvm_mmu_zap_collapsible_sptes(kvm, new);
11009 /* By default, write-protect everything to log writes. */
11010 int level = PG_LEVEL_4K;
11012 if (kvm_x86_ops.cpu_dirty_log_size) {
11014 * Clear all dirty bits, unless pages are treated as
11015 * dirty from the get-go.
11017 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
11018 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11021 * Write-protect large pages on write so that dirty
11022 * logging happens at 4k granularity. No need to
11023 * write-protect small SPTEs since write accesses are
11024 * logged by the CPU via dirty bits.
11026 level = PG_LEVEL_2M;
11027 } else if (kvm_dirty_log_manual_protect_and_init_set(kvm)) {
11029 * If we're with initial-all-set, we don't need
11030 * to write protect any small page because
11031 * they're reported as dirty already. However
11032 * we still need to write-protect huge pages
11033 * so that the page split can happen lazily on
11034 * the first write to the huge page.
11036 level = PG_LEVEL_2M;
11038 kvm_mmu_slot_remove_write_access(kvm, new, level);
11042 void kvm_arch_commit_memory_region(struct kvm *kvm,
11043 const struct kvm_userspace_memory_region *mem,
11044 struct kvm_memory_slot *old,
11045 const struct kvm_memory_slot *new,
11046 enum kvm_mr_change change)
11048 if (!kvm->arch.n_requested_mmu_pages)
11049 kvm_mmu_change_mmu_pages(kvm,
11050 kvm_mmu_calculate_default_mmu_pages(kvm));
11053 * FIXME: const-ify all uses of struct kvm_memory_slot.
11055 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11057 /* Free the arrays associated with the old memslot. */
11058 if (change == KVM_MR_MOVE)
11059 kvm_arch_free_memslot(kvm, old);
11062 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11064 kvm_mmu_zap_all(kvm);
11067 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11068 struct kvm_memory_slot *slot)
11070 kvm_page_track_flush_slot(kvm, slot);
11073 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11075 return (is_guest_mode(vcpu) &&
11076 kvm_x86_ops.guest_apic_has_interrupt &&
11077 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11080 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11082 if (!list_empty_careful(&vcpu->async_pf.done))
11085 if (kvm_apic_has_events(vcpu))
11088 if (vcpu->arch.pv.pv_unhalted)
11091 if (vcpu->arch.exception.pending)
11094 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11095 (vcpu->arch.nmi_pending &&
11096 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11099 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11100 (vcpu->arch.smi_pending &&
11101 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11104 if (kvm_arch_interrupt_allowed(vcpu) &&
11105 (kvm_cpu_has_interrupt(vcpu) ||
11106 kvm_guest_apic_has_interrupt(vcpu)))
11109 if (kvm_hv_has_stimer_pending(vcpu))
11112 if (is_guest_mode(vcpu) &&
11113 kvm_x86_ops.nested_ops->hv_timer_pending &&
11114 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11120 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11122 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11125 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11127 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11133 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11135 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11138 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11139 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11140 kvm_test_request(KVM_REQ_EVENT, vcpu))
11143 return kvm_arch_dy_has_pending_interrupt(vcpu);
11146 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11148 if (vcpu->arch.guest_state_protected)
11151 return vcpu->arch.preempted_in_kernel;
11154 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11156 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11159 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11161 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11164 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11166 /* Can't read the RIP when guest state is protected, just return 0 */
11167 if (vcpu->arch.guest_state_protected)
11170 if (is_64_bit_mode(vcpu))
11171 return kvm_rip_read(vcpu);
11172 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11173 kvm_rip_read(vcpu));
11175 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11177 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11179 return kvm_get_linear_rip(vcpu) == linear_rip;
11181 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11183 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11185 unsigned long rflags;
11187 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11188 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11189 rflags &= ~X86_EFLAGS_TF;
11192 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11194 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11196 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11197 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11198 rflags |= X86_EFLAGS_TF;
11199 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11202 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11204 __kvm_set_rflags(vcpu, rflags);
11205 kvm_make_request(KVM_REQ_EVENT, vcpu);
11207 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11209 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11213 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11217 r = kvm_mmu_reload(vcpu);
11221 if (!vcpu->arch.mmu->direct_map &&
11222 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11225 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11228 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11230 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11232 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11235 static inline u32 kvm_async_pf_next_probe(u32 key)
11237 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11240 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11242 u32 key = kvm_async_pf_hash_fn(gfn);
11244 while (vcpu->arch.apf.gfns[key] != ~0)
11245 key = kvm_async_pf_next_probe(key);
11247 vcpu->arch.apf.gfns[key] = gfn;
11250 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11253 u32 key = kvm_async_pf_hash_fn(gfn);
11255 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11256 (vcpu->arch.apf.gfns[key] != gfn &&
11257 vcpu->arch.apf.gfns[key] != ~0); i++)
11258 key = kvm_async_pf_next_probe(key);
11263 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11265 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11268 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11272 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11274 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11278 vcpu->arch.apf.gfns[i] = ~0;
11280 j = kvm_async_pf_next_probe(j);
11281 if (vcpu->arch.apf.gfns[j] == ~0)
11283 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11285 * k lies cyclically in ]i,j]
11287 * |....j i.k.| or |.k..j i...|
11289 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11290 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11295 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11297 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11299 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11303 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11305 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11307 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11308 &token, offset, sizeof(token));
11311 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11313 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11316 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11317 &val, offset, sizeof(val)))
11323 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11325 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11328 if (!kvm_pv_async_pf_enabled(vcpu) ||
11329 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11335 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11337 if (unlikely(!lapic_in_kernel(vcpu) ||
11338 kvm_event_needs_reinjection(vcpu) ||
11339 vcpu->arch.exception.pending))
11342 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11346 * If interrupts are off we cannot even use an artificial
11349 return kvm_arch_interrupt_allowed(vcpu);
11352 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11353 struct kvm_async_pf *work)
11355 struct x86_exception fault;
11357 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11358 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11360 if (kvm_can_deliver_async_pf(vcpu) &&
11361 !apf_put_user_notpresent(vcpu)) {
11362 fault.vector = PF_VECTOR;
11363 fault.error_code_valid = true;
11364 fault.error_code = 0;
11365 fault.nested_page_fault = false;
11366 fault.address = work->arch.token;
11367 fault.async_page_fault = true;
11368 kvm_inject_page_fault(vcpu, &fault);
11372 * It is not possible to deliver a paravirtualized asynchronous
11373 * page fault, but putting the guest in an artificial halt state
11374 * can be beneficial nevertheless: if an interrupt arrives, we
11375 * can deliver it timely and perhaps the guest will schedule
11376 * another process. When the instruction that triggered a page
11377 * fault is retried, hopefully the page will be ready in the host.
11379 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11384 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11385 struct kvm_async_pf *work)
11387 struct kvm_lapic_irq irq = {
11388 .delivery_mode = APIC_DM_FIXED,
11389 .vector = vcpu->arch.apf.vec
11392 if (work->wakeup_all)
11393 work->arch.token = ~0; /* broadcast wakeup */
11395 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11396 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11398 if ((work->wakeup_all || work->notpresent_injected) &&
11399 kvm_pv_async_pf_enabled(vcpu) &&
11400 !apf_put_user_ready(vcpu, work->arch.token)) {
11401 vcpu->arch.apf.pageready_pending = true;
11402 kvm_apic_set_irq(vcpu, &irq, NULL);
11405 vcpu->arch.apf.halted = false;
11406 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11409 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11411 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11412 if (!vcpu->arch.apf.pageready_pending)
11413 kvm_vcpu_kick(vcpu);
11416 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11418 if (!kvm_pv_async_pf_enabled(vcpu))
11421 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11424 void kvm_arch_start_assignment(struct kvm *kvm)
11426 atomic_inc(&kvm->arch.assigned_device_count);
11428 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11430 void kvm_arch_end_assignment(struct kvm *kvm)
11432 atomic_dec(&kvm->arch.assigned_device_count);
11434 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11436 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11438 return atomic_read(&kvm->arch.assigned_device_count);
11440 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11442 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11444 atomic_inc(&kvm->arch.noncoherent_dma_count);
11446 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11448 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11450 atomic_dec(&kvm->arch.noncoherent_dma_count);
11452 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11454 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11456 return atomic_read(&kvm->arch.noncoherent_dma_count);
11458 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11460 bool kvm_arch_has_irq_bypass(void)
11465 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11466 struct irq_bypass_producer *prod)
11468 struct kvm_kernel_irqfd *irqfd =
11469 container_of(cons, struct kvm_kernel_irqfd, consumer);
11472 irqfd->producer = prod;
11473 kvm_arch_start_assignment(irqfd->kvm);
11474 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11475 prod->irq, irqfd->gsi, 1);
11478 kvm_arch_end_assignment(irqfd->kvm);
11483 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11484 struct irq_bypass_producer *prod)
11487 struct kvm_kernel_irqfd *irqfd =
11488 container_of(cons, struct kvm_kernel_irqfd, consumer);
11490 WARN_ON(irqfd->producer != prod);
11491 irqfd->producer = NULL;
11494 * When producer of consumer is unregistered, we change back to
11495 * remapped mode, so we can re-use the current implementation
11496 * when the irq is masked/disabled or the consumer side (KVM
11497 * int this case doesn't want to receive the interrupts.
11499 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11501 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11502 " fails: %d\n", irqfd->consumer.token, ret);
11504 kvm_arch_end_assignment(irqfd->kvm);
11507 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11508 uint32_t guest_irq, bool set)
11510 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11513 bool kvm_vector_hashing_enabled(void)
11515 return vector_hashing;
11518 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11520 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11522 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11525 int kvm_spec_ctrl_test_value(u64 value)
11528 * test that setting IA32_SPEC_CTRL to given value
11529 * is allowed by the host processor
11533 unsigned long flags;
11536 local_irq_save(flags);
11538 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11540 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11543 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11545 local_irq_restore(flags);
11549 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11551 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11553 struct x86_exception fault;
11554 u32 access = error_code &
11555 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11557 if (!(error_code & PFERR_PRESENT_MASK) ||
11558 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11560 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11561 * tables probably do not match the TLB. Just proceed
11562 * with the error code that the processor gave.
11564 fault.vector = PF_VECTOR;
11565 fault.error_code_valid = true;
11566 fault.error_code = error_code;
11567 fault.nested_page_fault = false;
11568 fault.address = gva;
11570 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11572 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11575 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11576 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11577 * indicates whether exit to userspace is needed.
11579 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11580 struct x86_exception *e)
11582 if (r == X86EMUL_PROPAGATE_FAULT) {
11583 kvm_inject_emulated_page_fault(vcpu, e);
11588 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11589 * while handling a VMX instruction KVM could've handled the request
11590 * correctly by exiting to userspace and performing I/O but there
11591 * doesn't seem to be a real use-case behind such requests, just return
11592 * KVM_EXIT_INTERNAL_ERROR for now.
11594 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11595 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11596 vcpu->run->internal.ndata = 0;
11600 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11602 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11605 struct x86_exception e;
11607 unsigned long roots_to_free = 0;
11614 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11615 if (r != X86EMUL_CONTINUE)
11616 return kvm_handle_memory_failure(vcpu, r, &e);
11618 if (operand.pcid >> 12 != 0) {
11619 kvm_inject_gp(vcpu, 0);
11623 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11626 case INVPCID_TYPE_INDIV_ADDR:
11627 if ((!pcid_enabled && (operand.pcid != 0)) ||
11628 is_noncanonical_address(operand.gla, vcpu)) {
11629 kvm_inject_gp(vcpu, 0);
11632 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11633 return kvm_skip_emulated_instruction(vcpu);
11635 case INVPCID_TYPE_SINGLE_CTXT:
11636 if (!pcid_enabled && (operand.pcid != 0)) {
11637 kvm_inject_gp(vcpu, 0);
11641 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11642 kvm_mmu_sync_roots(vcpu);
11643 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11646 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11647 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11649 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11651 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11653 * If neither the current cr3 nor any of the prev_roots use the
11654 * given PCID, then nothing needs to be done here because a
11655 * resync will happen anyway before switching to any other CR3.
11658 return kvm_skip_emulated_instruction(vcpu);
11660 case INVPCID_TYPE_ALL_NON_GLOBAL:
11662 * Currently, KVM doesn't mark global entries in the shadow
11663 * page tables, so a non-global flush just degenerates to a
11664 * global flush. If needed, we could optimize this later by
11665 * keeping track of global entries in shadow page tables.
11669 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11670 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
11671 return kvm_skip_emulated_instruction(vcpu);
11674 BUG(); /* We have already checked above that type <= 3 */
11677 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11679 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11681 struct kvm_run *run = vcpu->run;
11682 struct kvm_mmio_fragment *frag;
11685 BUG_ON(!vcpu->mmio_needed);
11687 /* Complete previous fragment */
11688 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11689 len = min(8u, frag->len);
11690 if (!vcpu->mmio_is_write)
11691 memcpy(frag->data, run->mmio.data, len);
11693 if (frag->len <= 8) {
11694 /* Switch to the next fragment. */
11696 vcpu->mmio_cur_fragment++;
11698 /* Go forward to the next mmio piece. */
11704 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11705 vcpu->mmio_needed = 0;
11707 // VMG change, at this point, we're always done
11708 // RIP has already been advanced
11712 // More MMIO is needed
11713 run->mmio.phys_addr = frag->gpa;
11714 run->mmio.len = min(8u, frag->len);
11715 run->mmio.is_write = vcpu->mmio_is_write;
11716 if (run->mmio.is_write)
11717 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11718 run->exit_reason = KVM_EXIT_MMIO;
11720 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11725 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11729 struct kvm_mmio_fragment *frag;
11734 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11735 if (handled == bytes)
11742 /*TODO: Check if need to increment number of frags */
11743 frag = vcpu->mmio_fragments;
11744 vcpu->mmio_nr_fragments = 1;
11749 vcpu->mmio_needed = 1;
11750 vcpu->mmio_cur_fragment = 0;
11752 vcpu->run->mmio.phys_addr = gpa;
11753 vcpu->run->mmio.len = min(8u, frag->len);
11754 vcpu->run->mmio.is_write = 1;
11755 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11756 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11758 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11762 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11764 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11768 struct kvm_mmio_fragment *frag;
11773 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11774 if (handled == bytes)
11781 /*TODO: Check if need to increment number of frags */
11782 frag = vcpu->mmio_fragments;
11783 vcpu->mmio_nr_fragments = 1;
11788 vcpu->mmio_needed = 1;
11789 vcpu->mmio_cur_fragment = 0;
11791 vcpu->run->mmio.phys_addr = gpa;
11792 vcpu->run->mmio.len = min(8u, frag->len);
11793 vcpu->run->mmio.is_write = 0;
11794 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11796 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11800 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11802 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11804 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11805 vcpu->arch.pio.count * vcpu->arch.pio.size);
11806 vcpu->arch.pio.count = 0;
11811 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11812 unsigned int port, void *data, unsigned int count)
11816 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11821 vcpu->arch.pio.count = 0;
11826 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11827 unsigned int port, void *data, unsigned int count)
11831 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11834 vcpu->arch.pio.count = 0;
11836 vcpu->arch.guest_ins_data = data;
11837 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11843 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11844 unsigned int port, void *data, unsigned int count,
11847 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11848 : kvm_sev_es_outs(vcpu, size, port, data, count);
11850 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11852 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11853 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11854 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11855 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11856 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11857 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11858 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11859 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11860 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11861 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11862 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11863 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11864 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11865 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11866 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11867 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11868 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11869 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11870 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11871 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11872 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11873 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11874 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11875 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11876 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11877 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11878 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);