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 void kvm_define_user_return_msr(unsigned slot, u32 msr)
344 BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
345 user_return_msrs_global.msrs[slot] = msr;
346 if (slot >= user_return_msrs_global.nr)
347 user_return_msrs_global.nr = slot + 1;
349 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
351 static void kvm_user_return_msr_cpu_online(void)
353 unsigned int cpu = smp_processor_id();
354 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
358 for (i = 0; i < user_return_msrs_global.nr; ++i) {
359 rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
360 msrs->values[i].host = value;
361 msrs->values[i].curr = value;
365 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
367 unsigned int cpu = smp_processor_id();
368 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
371 value = (value & mask) | (msrs->values[slot].host & ~mask);
372 if (value == msrs->values[slot].curr)
374 err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
378 msrs->values[slot].curr = value;
379 if (!msrs->registered) {
380 msrs->urn.on_user_return = kvm_on_user_return;
381 user_return_notifier_register(&msrs->urn);
382 msrs->registered = true;
386 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
388 static void drop_user_return_notifiers(void)
390 unsigned int cpu = smp_processor_id();
391 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
393 if (msrs->registered)
394 kvm_on_user_return(&msrs->urn);
397 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
399 return vcpu->arch.apic_base;
401 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
403 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
405 return kvm_apic_mode(kvm_get_apic_base(vcpu));
407 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
409 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
411 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
412 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
413 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
414 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
416 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
418 if (!msr_info->host_initiated) {
419 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
421 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
425 kvm_lapic_set_base(vcpu, msr_info->data);
426 kvm_recalculate_apic_map(vcpu->kvm);
429 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
431 asmlinkage __visible noinstr void kvm_spurious_fault(void)
433 /* Fault while not rebooting. We want the trace. */
434 BUG_ON(!kvm_rebooting);
436 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
438 #define EXCPT_BENIGN 0
439 #define EXCPT_CONTRIBUTORY 1
442 static int exception_class(int vector)
452 return EXCPT_CONTRIBUTORY;
459 #define EXCPT_FAULT 0
461 #define EXCPT_ABORT 2
462 #define EXCPT_INTERRUPT 3
464 static int exception_type(int vector)
468 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
469 return EXCPT_INTERRUPT;
473 /* #DB is trap, as instruction watchpoints are handled elsewhere */
474 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
477 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
480 /* Reserved exceptions will result in fault */
484 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
486 unsigned nr = vcpu->arch.exception.nr;
487 bool has_payload = vcpu->arch.exception.has_payload;
488 unsigned long payload = vcpu->arch.exception.payload;
496 * "Certain debug exceptions may clear bit 0-3. The
497 * remaining contents of the DR6 register are never
498 * cleared by the processor".
500 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
502 * In order to reflect the #DB exception payload in guest
503 * dr6, three components need to be considered: active low
504 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
506 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
507 * In the target guest dr6:
508 * FIXED_1 bits should always be set.
509 * Active low bits should be cleared if 1-setting in payload.
510 * Active high bits should be set if 1-setting in payload.
512 * Note, the payload is compatible with the pending debug
513 * exceptions/exit qualification under VMX, that active_low bits
514 * are active high in payload.
515 * So they need to be flipped for DR6.
517 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
518 vcpu->arch.dr6 |= payload;
519 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
522 * The #DB payload is defined as compatible with the 'pending
523 * debug exceptions' field under VMX, not DR6. While bit 12 is
524 * defined in the 'pending debug exceptions' field (enabled
525 * breakpoint), it is reserved and must be zero in DR6.
527 vcpu->arch.dr6 &= ~BIT(12);
530 vcpu->arch.cr2 = payload;
534 vcpu->arch.exception.has_payload = false;
535 vcpu->arch.exception.payload = 0;
537 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
539 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
540 unsigned nr, bool has_error, u32 error_code,
541 bool has_payload, unsigned long payload, bool reinject)
546 kvm_make_request(KVM_REQ_EVENT, vcpu);
548 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
552 * On vmentry, vcpu->arch.exception.pending is only
553 * true if an event injection was blocked by
554 * nested_run_pending. In that case, however,
555 * vcpu_enter_guest requests an immediate exit,
556 * and the guest shouldn't proceed far enough to
559 WARN_ON_ONCE(vcpu->arch.exception.pending);
560 vcpu->arch.exception.injected = true;
561 if (WARN_ON_ONCE(has_payload)) {
563 * A reinjected event has already
564 * delivered its payload.
570 vcpu->arch.exception.pending = true;
571 vcpu->arch.exception.injected = false;
573 vcpu->arch.exception.has_error_code = has_error;
574 vcpu->arch.exception.nr = nr;
575 vcpu->arch.exception.error_code = error_code;
576 vcpu->arch.exception.has_payload = has_payload;
577 vcpu->arch.exception.payload = payload;
578 if (!is_guest_mode(vcpu))
579 kvm_deliver_exception_payload(vcpu);
583 /* to check exception */
584 prev_nr = vcpu->arch.exception.nr;
585 if (prev_nr == DF_VECTOR) {
586 /* triple fault -> shutdown */
587 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
590 class1 = exception_class(prev_nr);
591 class2 = exception_class(nr);
592 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
593 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
595 * Generate double fault per SDM Table 5-5. Set
596 * exception.pending = true so that the double fault
597 * can trigger a nested vmexit.
599 vcpu->arch.exception.pending = true;
600 vcpu->arch.exception.injected = false;
601 vcpu->arch.exception.has_error_code = true;
602 vcpu->arch.exception.nr = DF_VECTOR;
603 vcpu->arch.exception.error_code = 0;
604 vcpu->arch.exception.has_payload = false;
605 vcpu->arch.exception.payload = 0;
607 /* replace previous exception with a new one in a hope
608 that instruction re-execution will regenerate lost
613 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
615 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
617 EXPORT_SYMBOL_GPL(kvm_queue_exception);
619 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
621 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
623 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
625 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
626 unsigned long payload)
628 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
630 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
632 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
633 u32 error_code, unsigned long payload)
635 kvm_multiple_exception(vcpu, nr, true, error_code,
636 true, payload, false);
639 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
642 kvm_inject_gp(vcpu, 0);
644 return kvm_skip_emulated_instruction(vcpu);
648 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
650 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
652 ++vcpu->stat.pf_guest;
653 vcpu->arch.exception.nested_apf =
654 is_guest_mode(vcpu) && fault->async_page_fault;
655 if (vcpu->arch.exception.nested_apf) {
656 vcpu->arch.apf.nested_apf_token = fault->address;
657 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
659 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
663 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
665 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
666 struct x86_exception *fault)
668 struct kvm_mmu *fault_mmu;
669 WARN_ON_ONCE(fault->vector != PF_VECTOR);
671 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
675 * Invalidate the TLB entry for the faulting address, if it exists,
676 * else the access will fault indefinitely (and to emulate hardware).
678 if ((fault->error_code & PFERR_PRESENT_MASK) &&
679 !(fault->error_code & PFERR_RSVD_MASK))
680 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
681 fault_mmu->root_hpa);
683 fault_mmu->inject_page_fault(vcpu, fault);
684 return fault->nested_page_fault;
686 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
688 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
690 atomic_inc(&vcpu->arch.nmi_queued);
691 kvm_make_request(KVM_REQ_NMI, vcpu);
693 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
695 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
697 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
699 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
701 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
703 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
705 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
708 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
709 * a #GP and return false.
711 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
713 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
715 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
718 EXPORT_SYMBOL_GPL(kvm_require_cpl);
720 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
722 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
725 kvm_queue_exception(vcpu, UD_VECTOR);
728 EXPORT_SYMBOL_GPL(kvm_require_dr);
731 * This function will be used to read from the physical memory of the currently
732 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
733 * can read from guest physical or from the guest's guest physical memory.
735 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
736 gfn_t ngfn, void *data, int offset, int len,
739 struct x86_exception exception;
743 ngpa = gfn_to_gpa(ngfn);
744 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
745 if (real_gfn == UNMAPPED_GVA)
748 real_gfn = gpa_to_gfn(real_gfn);
750 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
752 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
754 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
755 void *data, int offset, int len, u32 access)
757 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
758 data, offset, len, access);
761 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
763 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
767 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
769 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
771 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
772 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
775 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
777 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
778 offset * sizeof(u64), sizeof(pdpte),
779 PFERR_USER_MASK|PFERR_WRITE_MASK);
784 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
785 if ((pdpte[i] & PT_PRESENT_MASK) &&
786 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
793 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
794 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
800 EXPORT_SYMBOL_GPL(load_pdptrs);
802 bool pdptrs_changed(struct kvm_vcpu *vcpu)
804 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
809 if (!is_pae_paging(vcpu))
812 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
815 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
816 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
817 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
818 PFERR_USER_MASK | PFERR_WRITE_MASK);
822 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
824 EXPORT_SYMBOL_GPL(pdptrs_changed);
826 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
828 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
830 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
831 kvm_clear_async_pf_completion_queue(vcpu);
832 kvm_async_pf_hash_reset(vcpu);
835 if ((cr0 ^ old_cr0) & update_bits)
836 kvm_mmu_reset_context(vcpu);
838 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
839 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
840 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
841 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
843 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
845 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
847 unsigned long old_cr0 = kvm_read_cr0(vcpu);
848 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
853 if (cr0 & 0xffffffff00000000UL)
857 cr0 &= ~CR0_RESERVED_BITS;
859 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
862 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
866 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
867 (cr0 & X86_CR0_PG)) {
872 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
877 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
878 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
879 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
882 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
885 static_call(kvm_x86_set_cr0)(vcpu, cr0);
887 kvm_post_set_cr0(vcpu, old_cr0, cr0);
891 EXPORT_SYMBOL_GPL(kvm_set_cr0);
893 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
895 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
897 EXPORT_SYMBOL_GPL(kvm_lmsw);
899 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
901 if (vcpu->arch.guest_state_protected)
904 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
906 if (vcpu->arch.xcr0 != host_xcr0)
907 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
909 if (vcpu->arch.xsaves_enabled &&
910 vcpu->arch.ia32_xss != host_xss)
911 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
914 if (static_cpu_has(X86_FEATURE_PKU) &&
915 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
916 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
917 vcpu->arch.pkru != vcpu->arch.host_pkru)
918 __write_pkru(vcpu->arch.pkru);
920 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
922 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
924 if (vcpu->arch.guest_state_protected)
927 if (static_cpu_has(X86_FEATURE_PKU) &&
928 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
929 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
930 vcpu->arch.pkru = rdpkru();
931 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
932 __write_pkru(vcpu->arch.host_pkru);
935 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
937 if (vcpu->arch.xcr0 != host_xcr0)
938 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
940 if (vcpu->arch.xsaves_enabled &&
941 vcpu->arch.ia32_xss != host_xss)
942 wrmsrl(MSR_IA32_XSS, host_xss);
946 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
948 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
951 u64 old_xcr0 = vcpu->arch.xcr0;
954 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
955 if (index != XCR_XFEATURE_ENABLED_MASK)
957 if (!(xcr0 & XFEATURE_MASK_FP))
959 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
963 * Do not allow the guest to set bits that we do not support
964 * saving. However, xcr0 bit 0 is always set, even if the
965 * emulated CPU does not support XSAVE (see fx_init).
967 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
968 if (xcr0 & ~valid_bits)
971 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
972 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
975 if (xcr0 & XFEATURE_MASK_AVX512) {
976 if (!(xcr0 & XFEATURE_MASK_YMM))
978 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
981 vcpu->arch.xcr0 = xcr0;
983 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
984 kvm_update_cpuid_runtime(vcpu);
988 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
990 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
991 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
992 kvm_inject_gp(vcpu, 0);
996 return kvm_skip_emulated_instruction(vcpu);
998 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1000 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1002 if (cr4 & cr4_reserved_bits)
1005 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1008 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1010 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1012 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1014 unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1015 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
1017 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1018 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1019 kvm_mmu_reset_context(vcpu);
1021 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1023 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1025 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1026 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1029 if (!kvm_is_valid_cr4(vcpu, cr4))
1032 if (is_long_mode(vcpu)) {
1033 if (!(cr4 & X86_CR4_PAE))
1035 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1037 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1038 && ((cr4 ^ old_cr4) & pdptr_bits)
1039 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1040 kvm_read_cr3(vcpu)))
1043 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1044 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1047 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1048 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1052 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1054 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1058 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1060 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1062 bool skip_tlb_flush = false;
1063 #ifdef CONFIG_X86_64
1064 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1067 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1068 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1072 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1073 if (!skip_tlb_flush) {
1074 kvm_mmu_sync_roots(vcpu);
1075 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1081 * Do not condition the GPA check on long mode, this helper is used to
1082 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1083 * the current vCPU mode is accurate.
1085 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1088 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1091 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1092 vcpu->arch.cr3 = cr3;
1093 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1097 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1099 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1101 if (cr8 & CR8_RESERVED_BITS)
1103 if (lapic_in_kernel(vcpu))
1104 kvm_lapic_set_tpr(vcpu, cr8);
1106 vcpu->arch.cr8 = cr8;
1109 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1111 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1113 if (lapic_in_kernel(vcpu))
1114 return kvm_lapic_get_cr8(vcpu);
1116 return vcpu->arch.cr8;
1118 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1120 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1124 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1125 for (i = 0; i < KVM_NR_DB_REGS; i++)
1126 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1127 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1131 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1135 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1136 dr7 = vcpu->arch.guest_debug_dr7;
1138 dr7 = vcpu->arch.dr7;
1139 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1140 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1141 if (dr7 & DR7_BP_EN_MASK)
1142 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1144 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1146 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1148 u64 fixed = DR6_FIXED_1;
1150 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1155 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1157 size_t size = ARRAY_SIZE(vcpu->arch.db);
1161 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1162 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1163 vcpu->arch.eff_db[dr] = val;
1167 if (!kvm_dr6_valid(val))
1169 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1173 if (!kvm_dr7_valid(val))
1175 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1176 kvm_update_dr7(vcpu);
1182 EXPORT_SYMBOL_GPL(kvm_set_dr);
1184 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1186 size_t size = ARRAY_SIZE(vcpu->arch.db);
1190 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1194 *val = vcpu->arch.dr6;
1198 *val = vcpu->arch.dr7;
1202 EXPORT_SYMBOL_GPL(kvm_get_dr);
1204 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1206 u32 ecx = kvm_rcx_read(vcpu);
1209 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1210 kvm_inject_gp(vcpu, 0);
1214 kvm_rax_write(vcpu, (u32)data);
1215 kvm_rdx_write(vcpu, data >> 32);
1216 return kvm_skip_emulated_instruction(vcpu);
1218 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1221 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1222 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1224 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1225 * extract the supported MSRs from the related const lists.
1226 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1227 * capabilities of the host cpu. This capabilities test skips MSRs that are
1228 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1229 * may depend on host virtualization features rather than host cpu features.
1232 static const u32 msrs_to_save_all[] = {
1233 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1235 #ifdef CONFIG_X86_64
1236 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1238 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1239 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1241 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1242 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1243 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1244 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1245 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1246 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1247 MSR_IA32_UMWAIT_CONTROL,
1249 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1250 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1251 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1252 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1253 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1254 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1255 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1256 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1257 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1258 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1259 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1260 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1261 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1262 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1263 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1264 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1265 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1266 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1267 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1268 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1269 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1270 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1273 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1274 static unsigned num_msrs_to_save;
1276 static const u32 emulated_msrs_all[] = {
1277 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1278 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1279 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1280 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1281 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1282 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1283 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1285 HV_X64_MSR_VP_INDEX,
1286 HV_X64_MSR_VP_RUNTIME,
1287 HV_X64_MSR_SCONTROL,
1288 HV_X64_MSR_STIMER0_CONFIG,
1289 HV_X64_MSR_VP_ASSIST_PAGE,
1290 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1291 HV_X64_MSR_TSC_EMULATION_STATUS,
1292 HV_X64_MSR_SYNDBG_OPTIONS,
1293 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1294 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1295 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1297 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1298 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1300 MSR_IA32_TSC_ADJUST,
1301 MSR_IA32_TSC_DEADLINE,
1302 MSR_IA32_ARCH_CAPABILITIES,
1303 MSR_IA32_PERF_CAPABILITIES,
1304 MSR_IA32_MISC_ENABLE,
1305 MSR_IA32_MCG_STATUS,
1307 MSR_IA32_MCG_EXT_CTL,
1311 MSR_MISC_FEATURES_ENABLES,
1312 MSR_AMD64_VIRT_SPEC_CTRL,
1317 * The following list leaves out MSRs whose values are determined
1318 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1319 * We always support the "true" VMX control MSRs, even if the host
1320 * processor does not, so I am putting these registers here rather
1321 * than in msrs_to_save_all.
1324 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1325 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1326 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1327 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1329 MSR_IA32_VMX_CR0_FIXED0,
1330 MSR_IA32_VMX_CR4_FIXED0,
1331 MSR_IA32_VMX_VMCS_ENUM,
1332 MSR_IA32_VMX_PROCBASED_CTLS2,
1333 MSR_IA32_VMX_EPT_VPID_CAP,
1334 MSR_IA32_VMX_VMFUNC,
1337 MSR_KVM_POLL_CONTROL,
1340 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1341 static unsigned num_emulated_msrs;
1344 * List of msr numbers which are used to expose MSR-based features that
1345 * can be used by a hypervisor to validate requested CPU features.
1347 static const u32 msr_based_features_all[] = {
1349 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1350 MSR_IA32_VMX_PINBASED_CTLS,
1351 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1352 MSR_IA32_VMX_PROCBASED_CTLS,
1353 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1354 MSR_IA32_VMX_EXIT_CTLS,
1355 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1356 MSR_IA32_VMX_ENTRY_CTLS,
1358 MSR_IA32_VMX_CR0_FIXED0,
1359 MSR_IA32_VMX_CR0_FIXED1,
1360 MSR_IA32_VMX_CR4_FIXED0,
1361 MSR_IA32_VMX_CR4_FIXED1,
1362 MSR_IA32_VMX_VMCS_ENUM,
1363 MSR_IA32_VMX_PROCBASED_CTLS2,
1364 MSR_IA32_VMX_EPT_VPID_CAP,
1365 MSR_IA32_VMX_VMFUNC,
1369 MSR_IA32_ARCH_CAPABILITIES,
1370 MSR_IA32_PERF_CAPABILITIES,
1373 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1374 static unsigned int num_msr_based_features;
1376 static u64 kvm_get_arch_capabilities(void)
1380 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1381 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1384 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1385 * the nested hypervisor runs with NX huge pages. If it is not,
1386 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1387 * L1 guests, so it need not worry about its own (L2) guests.
1389 data |= ARCH_CAP_PSCHANGE_MC_NO;
1392 * If we're doing cache flushes (either "always" or "cond")
1393 * we will do one whenever the guest does a vmlaunch/vmresume.
1394 * If an outer hypervisor is doing the cache flush for us
1395 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1396 * capability to the guest too, and if EPT is disabled we're not
1397 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1398 * require a nested hypervisor to do a flush of its own.
1400 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1401 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1403 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1404 data |= ARCH_CAP_RDCL_NO;
1405 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1406 data |= ARCH_CAP_SSB_NO;
1407 if (!boot_cpu_has_bug(X86_BUG_MDS))
1408 data |= ARCH_CAP_MDS_NO;
1410 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1412 * If RTM=0 because the kernel has disabled TSX, the host might
1413 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1414 * and therefore knows that there cannot be TAA) but keep
1415 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1416 * and we want to allow migrating those guests to tsx=off hosts.
1418 data &= ~ARCH_CAP_TAA_NO;
1419 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1420 data |= ARCH_CAP_TAA_NO;
1423 * Nothing to do here; we emulate TSX_CTRL if present on the
1424 * host so the guest can choose between disabling TSX or
1425 * using VERW to clear CPU buffers.
1432 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1434 switch (msr->index) {
1435 case MSR_IA32_ARCH_CAPABILITIES:
1436 msr->data = kvm_get_arch_capabilities();
1438 case MSR_IA32_UCODE_REV:
1439 rdmsrl_safe(msr->index, &msr->data);
1442 return static_call(kvm_x86_get_msr_feature)(msr);
1447 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1449 struct kvm_msr_entry msr;
1453 r = kvm_get_msr_feature(&msr);
1455 if (r == KVM_MSR_RET_INVALID) {
1456 /* Unconditionally clear the output for simplicity */
1458 if (kvm_msr_ignored_check(index, 0, false))
1470 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1472 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1475 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1478 if (efer & (EFER_LME | EFER_LMA) &&
1479 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1482 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1488 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1490 if (efer & efer_reserved_bits)
1493 return __kvm_valid_efer(vcpu, efer);
1495 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1497 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1499 u64 old_efer = vcpu->arch.efer;
1500 u64 efer = msr_info->data;
1503 if (efer & efer_reserved_bits)
1506 if (!msr_info->host_initiated) {
1507 if (!__kvm_valid_efer(vcpu, efer))
1510 if (is_paging(vcpu) &&
1511 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1516 efer |= vcpu->arch.efer & EFER_LMA;
1518 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1524 /* Update reserved bits */
1525 if ((efer ^ old_efer) & EFER_NX)
1526 kvm_mmu_reset_context(vcpu);
1531 void kvm_enable_efer_bits(u64 mask)
1533 efer_reserved_bits &= ~mask;
1535 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1537 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1539 struct kvm_x86_msr_filter *msr_filter;
1540 struct msr_bitmap_range *ranges;
1541 struct kvm *kvm = vcpu->kvm;
1546 /* x2APIC MSRs do not support filtering. */
1547 if (index >= 0x800 && index <= 0x8ff)
1550 idx = srcu_read_lock(&kvm->srcu);
1552 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1558 allowed = msr_filter->default_allow;
1559 ranges = msr_filter->ranges;
1561 for (i = 0; i < msr_filter->count; i++) {
1562 u32 start = ranges[i].base;
1563 u32 end = start + ranges[i].nmsrs;
1564 u32 flags = ranges[i].flags;
1565 unsigned long *bitmap = ranges[i].bitmap;
1567 if ((index >= start) && (index < end) && (flags & type)) {
1568 allowed = !!test_bit(index - start, bitmap);
1574 srcu_read_unlock(&kvm->srcu, idx);
1578 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1581 * Write @data into the MSR specified by @index. Select MSR specific fault
1582 * checks are bypassed if @host_initiated is %true.
1583 * Returns 0 on success, non-0 otherwise.
1584 * Assumes vcpu_load() was already called.
1586 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1587 bool host_initiated)
1589 struct msr_data msr;
1591 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1592 return KVM_MSR_RET_FILTERED;
1597 case MSR_KERNEL_GS_BASE:
1600 if (is_noncanonical_address(data, vcpu))
1603 case MSR_IA32_SYSENTER_EIP:
1604 case MSR_IA32_SYSENTER_ESP:
1606 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1607 * non-canonical address is written on Intel but not on
1608 * AMD (which ignores the top 32-bits, because it does
1609 * not implement 64-bit SYSENTER).
1611 * 64-bit code should hence be able to write a non-canonical
1612 * value on AMD. Making the address canonical ensures that
1613 * vmentry does not fail on Intel after writing a non-canonical
1614 * value, and that something deterministic happens if the guest
1615 * invokes 64-bit SYSENTER.
1617 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1622 msr.host_initiated = host_initiated;
1624 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1627 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1628 u32 index, u64 data, bool host_initiated)
1630 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1632 if (ret == KVM_MSR_RET_INVALID)
1633 if (kvm_msr_ignored_check(index, data, true))
1640 * Read the MSR specified by @index into @data. Select MSR specific fault
1641 * checks are bypassed if @host_initiated is %true.
1642 * Returns 0 on success, non-0 otherwise.
1643 * Assumes vcpu_load() was already called.
1645 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1646 bool host_initiated)
1648 struct msr_data msr;
1651 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1652 return KVM_MSR_RET_FILTERED;
1655 msr.host_initiated = host_initiated;
1657 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1663 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1664 u32 index, u64 *data, bool host_initiated)
1666 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1668 if (ret == KVM_MSR_RET_INVALID) {
1669 /* Unconditionally clear *data for simplicity */
1671 if (kvm_msr_ignored_check(index, 0, false))
1678 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1680 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1682 EXPORT_SYMBOL_GPL(kvm_get_msr);
1684 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1686 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1688 EXPORT_SYMBOL_GPL(kvm_set_msr);
1690 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1692 int err = vcpu->run->msr.error;
1694 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1695 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1698 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1701 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1703 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1706 static u64 kvm_msr_reason(int r)
1709 case KVM_MSR_RET_INVALID:
1710 return KVM_MSR_EXIT_REASON_UNKNOWN;
1711 case KVM_MSR_RET_FILTERED:
1712 return KVM_MSR_EXIT_REASON_FILTER;
1714 return KVM_MSR_EXIT_REASON_INVAL;
1718 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1719 u32 exit_reason, u64 data,
1720 int (*completion)(struct kvm_vcpu *vcpu),
1723 u64 msr_reason = kvm_msr_reason(r);
1725 /* Check if the user wanted to know about this MSR fault */
1726 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1729 vcpu->run->exit_reason = exit_reason;
1730 vcpu->run->msr.error = 0;
1731 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1732 vcpu->run->msr.reason = msr_reason;
1733 vcpu->run->msr.index = index;
1734 vcpu->run->msr.data = data;
1735 vcpu->arch.complete_userspace_io = completion;
1740 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1742 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1743 complete_emulated_rdmsr, r);
1746 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1748 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1749 complete_emulated_wrmsr, r);
1752 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1754 u32 ecx = kvm_rcx_read(vcpu);
1758 r = kvm_get_msr(vcpu, ecx, &data);
1760 /* MSR read failed? See if we should ask user space */
1761 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1762 /* Bounce to user space */
1767 trace_kvm_msr_read(ecx, data);
1769 kvm_rax_write(vcpu, data & -1u);
1770 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1772 trace_kvm_msr_read_ex(ecx);
1775 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1777 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1779 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1781 u32 ecx = kvm_rcx_read(vcpu);
1782 u64 data = kvm_read_edx_eax(vcpu);
1785 r = kvm_set_msr(vcpu, ecx, data);
1787 /* MSR write failed? See if we should ask user space */
1788 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1789 /* Bounce to user space */
1792 /* Signal all other negative errors to userspace */
1797 trace_kvm_msr_write(ecx, data);
1799 trace_kvm_msr_write_ex(ecx, data);
1801 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1803 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1805 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1807 return kvm_skip_emulated_instruction(vcpu);
1809 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1811 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1813 /* Treat an INVD instruction as a NOP and just skip it. */
1814 return kvm_emulate_as_nop(vcpu);
1816 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1818 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1820 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1821 return kvm_emulate_as_nop(vcpu);
1823 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1825 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1827 kvm_queue_exception(vcpu, UD_VECTOR);
1830 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1832 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1834 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1835 return kvm_emulate_as_nop(vcpu);
1837 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1839 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1841 xfer_to_guest_mode_prepare();
1842 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1843 xfer_to_guest_mode_work_pending();
1847 * The fast path for frequent and performance sensitive wrmsr emulation,
1848 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1849 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1850 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1851 * other cases which must be called after interrupts are enabled on the host.
1853 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1855 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1858 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1859 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1860 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1861 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1864 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1865 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1866 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1867 trace_kvm_apic_write(APIC_ICR, (u32)data);
1874 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1876 if (!kvm_can_use_hv_timer(vcpu))
1879 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1883 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1885 u32 msr = kvm_rcx_read(vcpu);
1887 fastpath_t ret = EXIT_FASTPATH_NONE;
1890 case APIC_BASE_MSR + (APIC_ICR >> 4):
1891 data = kvm_read_edx_eax(vcpu);
1892 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1893 kvm_skip_emulated_instruction(vcpu);
1894 ret = EXIT_FASTPATH_EXIT_HANDLED;
1897 case MSR_IA32_TSC_DEADLINE:
1898 data = kvm_read_edx_eax(vcpu);
1899 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1900 kvm_skip_emulated_instruction(vcpu);
1901 ret = EXIT_FASTPATH_REENTER_GUEST;
1908 if (ret != EXIT_FASTPATH_NONE)
1909 trace_kvm_msr_write(msr, data);
1913 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1916 * Adapt set_msr() to msr_io()'s calling convention
1918 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1920 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1923 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1925 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1928 #ifdef CONFIG_X86_64
1929 struct pvclock_clock {
1939 struct pvclock_gtod_data {
1942 struct pvclock_clock clock; /* extract of a clocksource struct */
1943 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1949 static struct pvclock_gtod_data pvclock_gtod_data;
1951 static void update_pvclock_gtod(struct timekeeper *tk)
1953 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1955 write_seqcount_begin(&vdata->seq);
1957 /* copy pvclock gtod data */
1958 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1959 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1960 vdata->clock.mask = tk->tkr_mono.mask;
1961 vdata->clock.mult = tk->tkr_mono.mult;
1962 vdata->clock.shift = tk->tkr_mono.shift;
1963 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1964 vdata->clock.offset = tk->tkr_mono.base;
1966 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1967 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1968 vdata->raw_clock.mask = tk->tkr_raw.mask;
1969 vdata->raw_clock.mult = tk->tkr_raw.mult;
1970 vdata->raw_clock.shift = tk->tkr_raw.shift;
1971 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1972 vdata->raw_clock.offset = tk->tkr_raw.base;
1974 vdata->wall_time_sec = tk->xtime_sec;
1976 vdata->offs_boot = tk->offs_boot;
1978 write_seqcount_end(&vdata->seq);
1981 static s64 get_kvmclock_base_ns(void)
1983 /* Count up from boot time, but with the frequency of the raw clock. */
1984 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1987 static s64 get_kvmclock_base_ns(void)
1989 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1990 return ktime_get_boottime_ns();
1994 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
1998 struct pvclock_wall_clock wc;
2005 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2010 ++version; /* first time write, random junk */
2014 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2018 * The guest calculates current wall clock time by adding
2019 * system time (updated by kvm_guest_time_update below) to the
2020 * wall clock specified here. We do the reverse here.
2022 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2024 wc.nsec = do_div(wall_nsec, 1000000000);
2025 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2026 wc.version = version;
2028 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2031 wc_sec_hi = wall_nsec >> 32;
2032 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2033 &wc_sec_hi, sizeof(wc_sec_hi));
2037 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2040 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2041 bool old_msr, bool host_initiated)
2043 struct kvm_arch *ka = &vcpu->kvm->arch;
2045 if (vcpu->vcpu_id == 0 && !host_initiated) {
2046 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2047 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2049 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2052 vcpu->arch.time = system_time;
2053 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2055 /* we verify if the enable bit is set... */
2056 vcpu->arch.pv_time_enabled = false;
2057 if (!(system_time & 1))
2060 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2061 &vcpu->arch.pv_time, system_time & ~1ULL,
2062 sizeof(struct pvclock_vcpu_time_info)))
2063 vcpu->arch.pv_time_enabled = true;
2068 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2070 do_shl32_div32(dividend, divisor);
2074 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2075 s8 *pshift, u32 *pmultiplier)
2083 scaled64 = scaled_hz;
2084 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2089 tps32 = (uint32_t)tps64;
2090 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2091 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2099 *pmultiplier = div_frac(scaled64, tps32);
2102 #ifdef CONFIG_X86_64
2103 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2106 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2107 static unsigned long max_tsc_khz;
2109 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2111 u64 v = (u64)khz * (1000000 + ppm);
2116 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2120 /* Guest TSC same frequency as host TSC? */
2122 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2126 /* TSC scaling supported? */
2127 if (!kvm_has_tsc_control) {
2128 if (user_tsc_khz > tsc_khz) {
2129 vcpu->arch.tsc_catchup = 1;
2130 vcpu->arch.tsc_always_catchup = 1;
2133 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2138 /* TSC scaling required - calculate ratio */
2139 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2140 user_tsc_khz, tsc_khz);
2142 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2143 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2148 vcpu->arch.tsc_scaling_ratio = ratio;
2152 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2154 u32 thresh_lo, thresh_hi;
2155 int use_scaling = 0;
2157 /* tsc_khz can be zero if TSC calibration fails */
2158 if (user_tsc_khz == 0) {
2159 /* set tsc_scaling_ratio to a safe value */
2160 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2164 /* Compute a scale to convert nanoseconds in TSC cycles */
2165 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2166 &vcpu->arch.virtual_tsc_shift,
2167 &vcpu->arch.virtual_tsc_mult);
2168 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2171 * Compute the variation in TSC rate which is acceptable
2172 * within the range of tolerance and decide if the
2173 * rate being applied is within that bounds of the hardware
2174 * rate. If so, no scaling or compensation need be done.
2176 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2177 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2178 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2179 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2182 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2185 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2187 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2188 vcpu->arch.virtual_tsc_mult,
2189 vcpu->arch.virtual_tsc_shift);
2190 tsc += vcpu->arch.this_tsc_write;
2194 static inline int gtod_is_based_on_tsc(int mode)
2196 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2199 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2201 #ifdef CONFIG_X86_64
2203 struct kvm_arch *ka = &vcpu->kvm->arch;
2204 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2206 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2207 atomic_read(&vcpu->kvm->online_vcpus));
2210 * Once the masterclock is enabled, always perform request in
2211 * order to update it.
2213 * In order to enable masterclock, the host clocksource must be TSC
2214 * and the vcpus need to have matched TSCs. When that happens,
2215 * perform request to enable masterclock.
2217 if (ka->use_master_clock ||
2218 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2219 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2221 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2222 atomic_read(&vcpu->kvm->online_vcpus),
2223 ka->use_master_clock, gtod->clock.vclock_mode);
2228 * Multiply tsc by a fixed point number represented by ratio.
2230 * The most significant 64-N bits (mult) of ratio represent the
2231 * integral part of the fixed point number; the remaining N bits
2232 * (frac) represent the fractional part, ie. ratio represents a fixed
2233 * point number (mult + frac * 2^(-N)).
2235 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2237 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2239 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2242 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2245 u64 ratio = vcpu->arch.tsc_scaling_ratio;
2247 if (ratio != kvm_default_tsc_scaling_ratio)
2248 _tsc = __scale_tsc(ratio, tsc);
2252 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2254 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2258 tsc = kvm_scale_tsc(vcpu, rdtsc());
2260 return target_tsc - tsc;
2263 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2265 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2267 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2269 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2271 vcpu->arch.l1_tsc_offset = offset;
2272 vcpu->arch.tsc_offset = static_call(kvm_x86_write_l1_tsc_offset)(vcpu, offset);
2275 static inline bool kvm_check_tsc_unstable(void)
2277 #ifdef CONFIG_X86_64
2279 * TSC is marked unstable when we're running on Hyper-V,
2280 * 'TSC page' clocksource is good.
2282 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2285 return check_tsc_unstable();
2288 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2290 struct kvm *kvm = vcpu->kvm;
2291 u64 offset, ns, elapsed;
2292 unsigned long flags;
2294 bool already_matched;
2295 bool synchronizing = false;
2297 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2298 offset = kvm_compute_tsc_offset(vcpu, data);
2299 ns = get_kvmclock_base_ns();
2300 elapsed = ns - kvm->arch.last_tsc_nsec;
2302 if (vcpu->arch.virtual_tsc_khz) {
2305 * detection of vcpu initialization -- need to sync
2306 * with other vCPUs. This particularly helps to keep
2307 * kvm_clock stable after CPU hotplug
2309 synchronizing = true;
2311 u64 tsc_exp = kvm->arch.last_tsc_write +
2312 nsec_to_cycles(vcpu, elapsed);
2313 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2315 * Special case: TSC write with a small delta (1 second)
2316 * of virtual cycle time against real time is
2317 * interpreted as an attempt to synchronize the CPU.
2319 synchronizing = data < tsc_exp + tsc_hz &&
2320 data + tsc_hz > tsc_exp;
2325 * For a reliable TSC, we can match TSC offsets, and for an unstable
2326 * TSC, we add elapsed time in this computation. We could let the
2327 * compensation code attempt to catch up if we fall behind, but
2328 * it's better to try to match offsets from the beginning.
2330 if (synchronizing &&
2331 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2332 if (!kvm_check_tsc_unstable()) {
2333 offset = kvm->arch.cur_tsc_offset;
2335 u64 delta = nsec_to_cycles(vcpu, elapsed);
2337 offset = kvm_compute_tsc_offset(vcpu, data);
2340 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2343 * We split periods of matched TSC writes into generations.
2344 * For each generation, we track the original measured
2345 * nanosecond time, offset, and write, so if TSCs are in
2346 * sync, we can match exact offset, and if not, we can match
2347 * exact software computation in compute_guest_tsc()
2349 * These values are tracked in kvm->arch.cur_xxx variables.
2351 kvm->arch.cur_tsc_generation++;
2352 kvm->arch.cur_tsc_nsec = ns;
2353 kvm->arch.cur_tsc_write = data;
2354 kvm->arch.cur_tsc_offset = offset;
2359 * We also track th most recent recorded KHZ, write and time to
2360 * allow the matching interval to be extended at each write.
2362 kvm->arch.last_tsc_nsec = ns;
2363 kvm->arch.last_tsc_write = data;
2364 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2366 vcpu->arch.last_guest_tsc = data;
2368 /* Keep track of which generation this VCPU has synchronized to */
2369 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2370 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2371 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2373 kvm_vcpu_write_tsc_offset(vcpu, offset);
2374 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2376 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2378 kvm->arch.nr_vcpus_matched_tsc = 0;
2379 } else if (!already_matched) {
2380 kvm->arch.nr_vcpus_matched_tsc++;
2383 kvm_track_tsc_matching(vcpu);
2384 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2387 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2390 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2391 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2394 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2396 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2397 WARN_ON(adjustment < 0);
2398 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2399 adjust_tsc_offset_guest(vcpu, adjustment);
2402 #ifdef CONFIG_X86_64
2404 static u64 read_tsc(void)
2406 u64 ret = (u64)rdtsc_ordered();
2407 u64 last = pvclock_gtod_data.clock.cycle_last;
2409 if (likely(ret >= last))
2413 * GCC likes to generate cmov here, but this branch is extremely
2414 * predictable (it's just a function of time and the likely is
2415 * very likely) and there's a data dependence, so force GCC
2416 * to generate a branch instead. I don't barrier() because
2417 * we don't actually need a barrier, and if this function
2418 * ever gets inlined it will generate worse code.
2424 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2430 switch (clock->vclock_mode) {
2431 case VDSO_CLOCKMODE_HVCLOCK:
2432 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2434 if (tsc_pg_val != U64_MAX) {
2435 /* TSC page valid */
2436 *mode = VDSO_CLOCKMODE_HVCLOCK;
2437 v = (tsc_pg_val - clock->cycle_last) &
2440 /* TSC page invalid */
2441 *mode = VDSO_CLOCKMODE_NONE;
2444 case VDSO_CLOCKMODE_TSC:
2445 *mode = VDSO_CLOCKMODE_TSC;
2446 *tsc_timestamp = read_tsc();
2447 v = (*tsc_timestamp - clock->cycle_last) &
2451 *mode = VDSO_CLOCKMODE_NONE;
2454 if (*mode == VDSO_CLOCKMODE_NONE)
2455 *tsc_timestamp = v = 0;
2457 return v * clock->mult;
2460 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2462 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2468 seq = read_seqcount_begin(>od->seq);
2469 ns = gtod->raw_clock.base_cycles;
2470 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2471 ns >>= gtod->raw_clock.shift;
2472 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2473 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2479 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2481 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2487 seq = read_seqcount_begin(>od->seq);
2488 ts->tv_sec = gtod->wall_time_sec;
2489 ns = gtod->clock.base_cycles;
2490 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2491 ns >>= gtod->clock.shift;
2492 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2494 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2500 /* returns true if host is using TSC based clocksource */
2501 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2503 /* checked again under seqlock below */
2504 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2507 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2511 /* returns true if host is using TSC based clocksource */
2512 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2515 /* checked again under seqlock below */
2516 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2519 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2525 * Assuming a stable TSC across physical CPUS, and a stable TSC
2526 * across virtual CPUs, the following condition is possible.
2527 * Each numbered line represents an event visible to both
2528 * CPUs at the next numbered event.
2530 * "timespecX" represents host monotonic time. "tscX" represents
2533 * VCPU0 on CPU0 | VCPU1 on CPU1
2535 * 1. read timespec0,tsc0
2536 * 2. | timespec1 = timespec0 + N
2538 * 3. transition to guest | transition to guest
2539 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2540 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2541 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2543 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2546 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2548 * - 0 < N - M => M < N
2550 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2551 * always the case (the difference between two distinct xtime instances
2552 * might be smaller then the difference between corresponding TSC reads,
2553 * when updating guest vcpus pvclock areas).
2555 * To avoid that problem, do not allow visibility of distinct
2556 * system_timestamp/tsc_timestamp values simultaneously: use a master
2557 * copy of host monotonic time values. Update that master copy
2560 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2564 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2566 #ifdef CONFIG_X86_64
2567 struct kvm_arch *ka = &kvm->arch;
2569 bool host_tsc_clocksource, vcpus_matched;
2571 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2572 atomic_read(&kvm->online_vcpus));
2575 * If the host uses TSC clock, then passthrough TSC as stable
2578 host_tsc_clocksource = kvm_get_time_and_clockread(
2579 &ka->master_kernel_ns,
2580 &ka->master_cycle_now);
2582 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2583 && !ka->backwards_tsc_observed
2584 && !ka->boot_vcpu_runs_old_kvmclock;
2586 if (ka->use_master_clock)
2587 atomic_set(&kvm_guest_has_master_clock, 1);
2589 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2590 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2595 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2597 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2600 static void kvm_gen_update_masterclock(struct kvm *kvm)
2602 #ifdef CONFIG_X86_64
2604 struct kvm_vcpu *vcpu;
2605 struct kvm_arch *ka = &kvm->arch;
2606 unsigned long flags;
2608 kvm_hv_invalidate_tsc_page(kvm);
2610 kvm_make_mclock_inprogress_request(kvm);
2612 /* no guest entries from this point */
2613 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2614 pvclock_update_vm_gtod_copy(kvm);
2615 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2617 kvm_for_each_vcpu(i, vcpu, kvm)
2618 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2620 /* guest entries allowed */
2621 kvm_for_each_vcpu(i, vcpu, kvm)
2622 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2626 u64 get_kvmclock_ns(struct kvm *kvm)
2628 struct kvm_arch *ka = &kvm->arch;
2629 struct pvclock_vcpu_time_info hv_clock;
2630 unsigned long flags;
2633 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2634 if (!ka->use_master_clock) {
2635 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2636 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2639 hv_clock.tsc_timestamp = ka->master_cycle_now;
2640 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2641 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2643 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2646 if (__this_cpu_read(cpu_tsc_khz)) {
2647 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2648 &hv_clock.tsc_shift,
2649 &hv_clock.tsc_to_system_mul);
2650 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2652 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2659 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2660 struct gfn_to_hva_cache *cache,
2661 unsigned int offset)
2663 struct kvm_vcpu_arch *vcpu = &v->arch;
2664 struct pvclock_vcpu_time_info guest_hv_clock;
2666 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2667 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2670 /* This VCPU is paused, but it's legal for a guest to read another
2671 * VCPU's kvmclock, so we really have to follow the specification where
2672 * it says that version is odd if data is being modified, and even after
2675 * Version field updates must be kept separate. This is because
2676 * kvm_write_guest_cached might use a "rep movs" instruction, and
2677 * writes within a string instruction are weakly ordered. So there
2678 * are three writes overall.
2680 * As a small optimization, only write the version field in the first
2681 * and third write. The vcpu->pv_time cache is still valid, because the
2682 * version field is the first in the struct.
2684 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2686 if (guest_hv_clock.version & 1)
2687 ++guest_hv_clock.version; /* first time write, random junk */
2689 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2690 kvm_write_guest_offset_cached(v->kvm, cache,
2691 &vcpu->hv_clock, offset,
2692 sizeof(vcpu->hv_clock.version));
2696 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2697 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2699 if (vcpu->pvclock_set_guest_stopped_request) {
2700 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2701 vcpu->pvclock_set_guest_stopped_request = false;
2704 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2706 kvm_write_guest_offset_cached(v->kvm, cache,
2707 &vcpu->hv_clock, offset,
2708 sizeof(vcpu->hv_clock));
2712 vcpu->hv_clock.version++;
2713 kvm_write_guest_offset_cached(v->kvm, cache,
2714 &vcpu->hv_clock, offset,
2715 sizeof(vcpu->hv_clock.version));
2718 static int kvm_guest_time_update(struct kvm_vcpu *v)
2720 unsigned long flags, tgt_tsc_khz;
2721 struct kvm_vcpu_arch *vcpu = &v->arch;
2722 struct kvm_arch *ka = &v->kvm->arch;
2724 u64 tsc_timestamp, host_tsc;
2726 bool use_master_clock;
2732 * If the host uses TSC clock, then passthrough TSC as stable
2735 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2736 use_master_clock = ka->use_master_clock;
2737 if (use_master_clock) {
2738 host_tsc = ka->master_cycle_now;
2739 kernel_ns = ka->master_kernel_ns;
2741 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2743 /* Keep irq disabled to prevent changes to the clock */
2744 local_irq_save(flags);
2745 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2746 if (unlikely(tgt_tsc_khz == 0)) {
2747 local_irq_restore(flags);
2748 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2751 if (!use_master_clock) {
2753 kernel_ns = get_kvmclock_base_ns();
2756 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2759 * We may have to catch up the TSC to match elapsed wall clock
2760 * time for two reasons, even if kvmclock is used.
2761 * 1) CPU could have been running below the maximum TSC rate
2762 * 2) Broken TSC compensation resets the base at each VCPU
2763 * entry to avoid unknown leaps of TSC even when running
2764 * again on the same CPU. This may cause apparent elapsed
2765 * time to disappear, and the guest to stand still or run
2768 if (vcpu->tsc_catchup) {
2769 u64 tsc = compute_guest_tsc(v, kernel_ns);
2770 if (tsc > tsc_timestamp) {
2771 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2772 tsc_timestamp = tsc;
2776 local_irq_restore(flags);
2778 /* With all the info we got, fill in the values */
2780 if (kvm_has_tsc_control)
2781 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2783 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2784 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2785 &vcpu->hv_clock.tsc_shift,
2786 &vcpu->hv_clock.tsc_to_system_mul);
2787 vcpu->hw_tsc_khz = tgt_tsc_khz;
2790 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2791 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2792 vcpu->last_guest_tsc = tsc_timestamp;
2794 /* If the host uses TSC clocksource, then it is stable */
2796 if (use_master_clock)
2797 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2799 vcpu->hv_clock.flags = pvclock_flags;
2801 if (vcpu->pv_time_enabled)
2802 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2803 if (vcpu->xen.vcpu_info_set)
2804 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2805 offsetof(struct compat_vcpu_info, time));
2806 if (vcpu->xen.vcpu_time_info_set)
2807 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2808 if (v == kvm_get_vcpu(v->kvm, 0))
2809 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2814 * kvmclock updates which are isolated to a given vcpu, such as
2815 * vcpu->cpu migration, should not allow system_timestamp from
2816 * the rest of the vcpus to remain static. Otherwise ntp frequency
2817 * correction applies to one vcpu's system_timestamp but not
2820 * So in those cases, request a kvmclock update for all vcpus.
2821 * We need to rate-limit these requests though, as they can
2822 * considerably slow guests that have a large number of vcpus.
2823 * The time for a remote vcpu to update its kvmclock is bound
2824 * by the delay we use to rate-limit the updates.
2827 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2829 static void kvmclock_update_fn(struct work_struct *work)
2832 struct delayed_work *dwork = to_delayed_work(work);
2833 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2834 kvmclock_update_work);
2835 struct kvm *kvm = container_of(ka, struct kvm, arch);
2836 struct kvm_vcpu *vcpu;
2838 kvm_for_each_vcpu(i, vcpu, kvm) {
2839 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2840 kvm_vcpu_kick(vcpu);
2844 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2846 struct kvm *kvm = v->kvm;
2848 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2849 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2850 KVMCLOCK_UPDATE_DELAY);
2853 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2855 static void kvmclock_sync_fn(struct work_struct *work)
2857 struct delayed_work *dwork = to_delayed_work(work);
2858 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2859 kvmclock_sync_work);
2860 struct kvm *kvm = container_of(ka, struct kvm, arch);
2862 if (!kvmclock_periodic_sync)
2865 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2866 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2867 KVMCLOCK_SYNC_PERIOD);
2871 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2873 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2875 /* McStatusWrEn enabled? */
2876 if (guest_cpuid_is_amd_or_hygon(vcpu))
2877 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2882 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2884 u64 mcg_cap = vcpu->arch.mcg_cap;
2885 unsigned bank_num = mcg_cap & 0xff;
2886 u32 msr = msr_info->index;
2887 u64 data = msr_info->data;
2890 case MSR_IA32_MCG_STATUS:
2891 vcpu->arch.mcg_status = data;
2893 case MSR_IA32_MCG_CTL:
2894 if (!(mcg_cap & MCG_CTL_P) &&
2895 (data || !msr_info->host_initiated))
2897 if (data != 0 && data != ~(u64)0)
2899 vcpu->arch.mcg_ctl = data;
2902 if (msr >= MSR_IA32_MC0_CTL &&
2903 msr < MSR_IA32_MCx_CTL(bank_num)) {
2904 u32 offset = array_index_nospec(
2905 msr - MSR_IA32_MC0_CTL,
2906 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2908 /* only 0 or all 1s can be written to IA32_MCi_CTL
2909 * some Linux kernels though clear bit 10 in bank 4 to
2910 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2911 * this to avoid an uncatched #GP in the guest
2913 if ((offset & 0x3) == 0 &&
2914 data != 0 && (data | (1 << 10)) != ~(u64)0)
2918 if (!msr_info->host_initiated &&
2919 (offset & 0x3) == 1 && data != 0) {
2920 if (!can_set_mci_status(vcpu))
2924 vcpu->arch.mce_banks[offset] = data;
2932 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2934 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2936 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2939 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2941 gpa_t gpa = data & ~0x3f;
2943 /* Bits 4:5 are reserved, Should be zero */
2947 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2948 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2951 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2952 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2955 if (!lapic_in_kernel(vcpu))
2956 return data ? 1 : 0;
2958 vcpu->arch.apf.msr_en_val = data;
2960 if (!kvm_pv_async_pf_enabled(vcpu)) {
2961 kvm_clear_async_pf_completion_queue(vcpu);
2962 kvm_async_pf_hash_reset(vcpu);
2966 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2970 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2971 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2973 kvm_async_pf_wakeup_all(vcpu);
2978 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2980 /* Bits 8-63 are reserved */
2984 if (!lapic_in_kernel(vcpu))
2987 vcpu->arch.apf.msr_int_val = data;
2989 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2994 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2996 vcpu->arch.pv_time_enabled = false;
2997 vcpu->arch.time = 0;
3000 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3002 ++vcpu->stat.tlb_flush;
3003 static_call(kvm_x86_tlb_flush_all)(vcpu);
3006 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3008 ++vcpu->stat.tlb_flush;
3009 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3012 static void record_steal_time(struct kvm_vcpu *vcpu)
3014 struct kvm_host_map map;
3015 struct kvm_steal_time *st;
3017 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3018 kvm_xen_runstate_set_running(vcpu);
3022 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3025 /* -EAGAIN is returned in atomic context so we can just return. */
3026 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3027 &map, &vcpu->arch.st.cache, false))
3031 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3034 * Doing a TLB flush here, on the guest's behalf, can avoid
3037 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3038 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3039 st->preempted & KVM_VCPU_FLUSH_TLB);
3040 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
3041 kvm_vcpu_flush_tlb_guest(vcpu);
3044 vcpu->arch.st.preempted = 0;
3046 if (st->version & 1)
3047 st->version += 1; /* first time write, random junk */
3053 st->steal += current->sched_info.run_delay -
3054 vcpu->arch.st.last_steal;
3055 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3061 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3064 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3067 u32 msr = msr_info->index;
3068 u64 data = msr_info->data;
3070 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3071 return kvm_xen_write_hypercall_page(vcpu, data);
3074 case MSR_AMD64_NB_CFG:
3075 case MSR_IA32_UCODE_WRITE:
3076 case MSR_VM_HSAVE_PA:
3077 case MSR_AMD64_PATCH_LOADER:
3078 case MSR_AMD64_BU_CFG2:
3079 case MSR_AMD64_DC_CFG:
3080 case MSR_F15H_EX_CFG:
3083 case MSR_IA32_UCODE_REV:
3084 if (msr_info->host_initiated)
3085 vcpu->arch.microcode_version = data;
3087 case MSR_IA32_ARCH_CAPABILITIES:
3088 if (!msr_info->host_initiated)
3090 vcpu->arch.arch_capabilities = data;
3092 case MSR_IA32_PERF_CAPABILITIES: {
3093 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3095 if (!msr_info->host_initiated)
3097 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3099 if (data & ~msr_ent.data)
3102 vcpu->arch.perf_capabilities = data;
3107 return set_efer(vcpu, msr_info);
3109 data &= ~(u64)0x40; /* ignore flush filter disable */
3110 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3111 data &= ~(u64)0x8; /* ignore TLB cache disable */
3113 /* Handle McStatusWrEn */
3114 if (data == BIT_ULL(18)) {
3115 vcpu->arch.msr_hwcr = data;
3116 } else if (data != 0) {
3117 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3122 case MSR_FAM10H_MMIO_CONF_BASE:
3124 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3129 case 0x200 ... 0x2ff:
3130 return kvm_mtrr_set_msr(vcpu, msr, data);
3131 case MSR_IA32_APICBASE:
3132 return kvm_set_apic_base(vcpu, msr_info);
3133 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3134 return kvm_x2apic_msr_write(vcpu, msr, data);
3135 case MSR_IA32_TSC_DEADLINE:
3136 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3138 case MSR_IA32_TSC_ADJUST:
3139 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3140 if (!msr_info->host_initiated) {
3141 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3142 adjust_tsc_offset_guest(vcpu, adj);
3144 vcpu->arch.ia32_tsc_adjust_msr = data;
3147 case MSR_IA32_MISC_ENABLE:
3148 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3149 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3150 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3152 vcpu->arch.ia32_misc_enable_msr = data;
3153 kvm_update_cpuid_runtime(vcpu);
3155 vcpu->arch.ia32_misc_enable_msr = data;
3158 case MSR_IA32_SMBASE:
3159 if (!msr_info->host_initiated)
3161 vcpu->arch.smbase = data;
3163 case MSR_IA32_POWER_CTL:
3164 vcpu->arch.msr_ia32_power_ctl = data;
3167 if (msr_info->host_initiated) {
3168 kvm_synchronize_tsc(vcpu, data);
3170 u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3171 adjust_tsc_offset_guest(vcpu, adj);
3172 vcpu->arch.ia32_tsc_adjust_msr += adj;
3176 if (!msr_info->host_initiated &&
3177 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3180 * KVM supports exposing PT to the guest, but does not support
3181 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3182 * XSAVES/XRSTORS to save/restore PT MSRs.
3184 if (data & ~supported_xss)
3186 vcpu->arch.ia32_xss = data;
3189 if (!msr_info->host_initiated)
3191 vcpu->arch.smi_count = data;
3193 case MSR_KVM_WALL_CLOCK_NEW:
3194 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3197 vcpu->kvm->arch.wall_clock = data;
3198 kvm_write_wall_clock(vcpu->kvm, data, 0);
3200 case MSR_KVM_WALL_CLOCK:
3201 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3204 vcpu->kvm->arch.wall_clock = data;
3205 kvm_write_wall_clock(vcpu->kvm, data, 0);
3207 case MSR_KVM_SYSTEM_TIME_NEW:
3208 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3211 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3213 case MSR_KVM_SYSTEM_TIME:
3214 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3217 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3219 case MSR_KVM_ASYNC_PF_EN:
3220 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3223 if (kvm_pv_enable_async_pf(vcpu, data))
3226 case MSR_KVM_ASYNC_PF_INT:
3227 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3230 if (kvm_pv_enable_async_pf_int(vcpu, data))
3233 case MSR_KVM_ASYNC_PF_ACK:
3234 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3237 vcpu->arch.apf.pageready_pending = false;
3238 kvm_check_async_pf_completion(vcpu);
3241 case MSR_KVM_STEAL_TIME:
3242 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3245 if (unlikely(!sched_info_on()))
3248 if (data & KVM_STEAL_RESERVED_MASK)
3251 vcpu->arch.st.msr_val = data;
3253 if (!(data & KVM_MSR_ENABLED))
3256 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3259 case MSR_KVM_PV_EOI_EN:
3260 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3263 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3267 case MSR_KVM_POLL_CONTROL:
3268 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3271 /* only enable bit supported */
3272 if (data & (-1ULL << 1))
3275 vcpu->arch.msr_kvm_poll_control = data;
3278 case MSR_IA32_MCG_CTL:
3279 case MSR_IA32_MCG_STATUS:
3280 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3281 return set_msr_mce(vcpu, msr_info);
3283 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3284 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3287 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3288 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3289 if (kvm_pmu_is_valid_msr(vcpu, msr))
3290 return kvm_pmu_set_msr(vcpu, msr_info);
3292 if (pr || data != 0)
3293 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3294 "0x%x data 0x%llx\n", msr, data);
3296 case MSR_K7_CLK_CTL:
3298 * Ignore all writes to this no longer documented MSR.
3299 * Writes are only relevant for old K7 processors,
3300 * all pre-dating SVM, but a recommended workaround from
3301 * AMD for these chips. It is possible to specify the
3302 * affected processor models on the command line, hence
3303 * the need to ignore the workaround.
3306 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3307 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3308 case HV_X64_MSR_SYNDBG_OPTIONS:
3309 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3310 case HV_X64_MSR_CRASH_CTL:
3311 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3312 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3313 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3314 case HV_X64_MSR_TSC_EMULATION_STATUS:
3315 return kvm_hv_set_msr_common(vcpu, msr, data,
3316 msr_info->host_initiated);
3317 case MSR_IA32_BBL_CR_CTL3:
3318 /* Drop writes to this legacy MSR -- see rdmsr
3319 * counterpart for further detail.
3321 if (report_ignored_msrs)
3322 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3325 case MSR_AMD64_OSVW_ID_LENGTH:
3326 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3328 vcpu->arch.osvw.length = data;
3330 case MSR_AMD64_OSVW_STATUS:
3331 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3333 vcpu->arch.osvw.status = data;
3335 case MSR_PLATFORM_INFO:
3336 if (!msr_info->host_initiated ||
3337 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3338 cpuid_fault_enabled(vcpu)))
3340 vcpu->arch.msr_platform_info = data;
3342 case MSR_MISC_FEATURES_ENABLES:
3343 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3344 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3345 !supports_cpuid_fault(vcpu)))
3347 vcpu->arch.msr_misc_features_enables = data;
3350 if (kvm_pmu_is_valid_msr(vcpu, msr))
3351 return kvm_pmu_set_msr(vcpu, msr_info);
3352 return KVM_MSR_RET_INVALID;
3356 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3358 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3361 u64 mcg_cap = vcpu->arch.mcg_cap;
3362 unsigned bank_num = mcg_cap & 0xff;
3365 case MSR_IA32_P5_MC_ADDR:
3366 case MSR_IA32_P5_MC_TYPE:
3369 case MSR_IA32_MCG_CAP:
3370 data = vcpu->arch.mcg_cap;
3372 case MSR_IA32_MCG_CTL:
3373 if (!(mcg_cap & MCG_CTL_P) && !host)
3375 data = vcpu->arch.mcg_ctl;
3377 case MSR_IA32_MCG_STATUS:
3378 data = vcpu->arch.mcg_status;
3381 if (msr >= MSR_IA32_MC0_CTL &&
3382 msr < MSR_IA32_MCx_CTL(bank_num)) {
3383 u32 offset = array_index_nospec(
3384 msr - MSR_IA32_MC0_CTL,
3385 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3387 data = vcpu->arch.mce_banks[offset];
3396 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3398 switch (msr_info->index) {
3399 case MSR_IA32_PLATFORM_ID:
3400 case MSR_IA32_EBL_CR_POWERON:
3401 case MSR_IA32_LASTBRANCHFROMIP:
3402 case MSR_IA32_LASTBRANCHTOIP:
3403 case MSR_IA32_LASTINTFROMIP:
3404 case MSR_IA32_LASTINTTOIP:
3406 case MSR_K8_TSEG_ADDR:
3407 case MSR_K8_TSEG_MASK:
3408 case MSR_VM_HSAVE_PA:
3409 case MSR_K8_INT_PENDING_MSG:
3410 case MSR_AMD64_NB_CFG:
3411 case MSR_FAM10H_MMIO_CONF_BASE:
3412 case MSR_AMD64_BU_CFG2:
3413 case MSR_IA32_PERF_CTL:
3414 case MSR_AMD64_DC_CFG:
3415 case MSR_F15H_EX_CFG:
3417 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3418 * limit) MSRs. Just return 0, as we do not want to expose the host
3419 * data here. Do not conditionalize this on CPUID, as KVM does not do
3420 * so for existing CPU-specific MSRs.
3422 case MSR_RAPL_POWER_UNIT:
3423 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3424 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3425 case MSR_PKG_ENERGY_STATUS: /* Total package */
3426 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3429 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3430 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3431 return kvm_pmu_get_msr(vcpu, msr_info);
3432 if (!msr_info->host_initiated)
3436 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3437 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3438 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3439 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3440 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3441 return kvm_pmu_get_msr(vcpu, msr_info);
3444 case MSR_IA32_UCODE_REV:
3445 msr_info->data = vcpu->arch.microcode_version;
3447 case MSR_IA32_ARCH_CAPABILITIES:
3448 if (!msr_info->host_initiated &&
3449 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3451 msr_info->data = vcpu->arch.arch_capabilities;
3453 case MSR_IA32_PERF_CAPABILITIES:
3454 if (!msr_info->host_initiated &&
3455 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3457 msr_info->data = vcpu->arch.perf_capabilities;
3459 case MSR_IA32_POWER_CTL:
3460 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3462 case MSR_IA32_TSC: {
3464 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3465 * even when not intercepted. AMD manual doesn't explicitly
3466 * state this but appears to behave the same.
3468 * On userspace reads and writes, however, we unconditionally
3469 * return L1's TSC value to ensure backwards-compatible
3470 * behavior for migration.
3472 u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3473 vcpu->arch.tsc_offset;
3475 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3479 case 0x200 ... 0x2ff:
3480 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3481 case 0xcd: /* fsb frequency */
3485 * MSR_EBC_FREQUENCY_ID
3486 * Conservative value valid for even the basic CPU models.
3487 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3488 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3489 * and 266MHz for model 3, or 4. Set Core Clock
3490 * Frequency to System Bus Frequency Ratio to 1 (bits
3491 * 31:24) even though these are only valid for CPU
3492 * models > 2, however guests may end up dividing or
3493 * multiplying by zero otherwise.
3495 case MSR_EBC_FREQUENCY_ID:
3496 msr_info->data = 1 << 24;
3498 case MSR_IA32_APICBASE:
3499 msr_info->data = kvm_get_apic_base(vcpu);
3501 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3502 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3503 case MSR_IA32_TSC_DEADLINE:
3504 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3506 case MSR_IA32_TSC_ADJUST:
3507 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3509 case MSR_IA32_MISC_ENABLE:
3510 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3512 case MSR_IA32_SMBASE:
3513 if (!msr_info->host_initiated)
3515 msr_info->data = vcpu->arch.smbase;
3518 msr_info->data = vcpu->arch.smi_count;
3520 case MSR_IA32_PERF_STATUS:
3521 /* TSC increment by tick */
3522 msr_info->data = 1000ULL;
3523 /* CPU multiplier */
3524 msr_info->data |= (((uint64_t)4ULL) << 40);
3527 msr_info->data = vcpu->arch.efer;
3529 case MSR_KVM_WALL_CLOCK:
3530 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3533 msr_info->data = vcpu->kvm->arch.wall_clock;
3535 case MSR_KVM_WALL_CLOCK_NEW:
3536 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3539 msr_info->data = vcpu->kvm->arch.wall_clock;
3541 case MSR_KVM_SYSTEM_TIME:
3542 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3545 msr_info->data = vcpu->arch.time;
3547 case MSR_KVM_SYSTEM_TIME_NEW:
3548 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3551 msr_info->data = vcpu->arch.time;
3553 case MSR_KVM_ASYNC_PF_EN:
3554 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3557 msr_info->data = vcpu->arch.apf.msr_en_val;
3559 case MSR_KVM_ASYNC_PF_INT:
3560 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3563 msr_info->data = vcpu->arch.apf.msr_int_val;
3565 case MSR_KVM_ASYNC_PF_ACK:
3566 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3571 case MSR_KVM_STEAL_TIME:
3572 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3575 msr_info->data = vcpu->arch.st.msr_val;
3577 case MSR_KVM_PV_EOI_EN:
3578 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3581 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3583 case MSR_KVM_POLL_CONTROL:
3584 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3587 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3589 case MSR_IA32_P5_MC_ADDR:
3590 case MSR_IA32_P5_MC_TYPE:
3591 case MSR_IA32_MCG_CAP:
3592 case MSR_IA32_MCG_CTL:
3593 case MSR_IA32_MCG_STATUS:
3594 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3595 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3596 msr_info->host_initiated);
3598 if (!msr_info->host_initiated &&
3599 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3601 msr_info->data = vcpu->arch.ia32_xss;
3603 case MSR_K7_CLK_CTL:
3605 * Provide expected ramp-up count for K7. All other
3606 * are set to zero, indicating minimum divisors for
3609 * This prevents guest kernels on AMD host with CPU
3610 * type 6, model 8 and higher from exploding due to
3611 * the rdmsr failing.
3613 msr_info->data = 0x20000000;
3615 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3616 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3617 case HV_X64_MSR_SYNDBG_OPTIONS:
3618 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3619 case HV_X64_MSR_CRASH_CTL:
3620 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3621 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3622 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3623 case HV_X64_MSR_TSC_EMULATION_STATUS:
3624 return kvm_hv_get_msr_common(vcpu,
3625 msr_info->index, &msr_info->data,
3626 msr_info->host_initiated);
3627 case MSR_IA32_BBL_CR_CTL3:
3628 /* This legacy MSR exists but isn't fully documented in current
3629 * silicon. It is however accessed by winxp in very narrow
3630 * scenarios where it sets bit #19, itself documented as
3631 * a "reserved" bit. Best effort attempt to source coherent
3632 * read data here should the balance of the register be
3633 * interpreted by the guest:
3635 * L2 cache control register 3: 64GB range, 256KB size,
3636 * enabled, latency 0x1, configured
3638 msr_info->data = 0xbe702111;
3640 case MSR_AMD64_OSVW_ID_LENGTH:
3641 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3643 msr_info->data = vcpu->arch.osvw.length;
3645 case MSR_AMD64_OSVW_STATUS:
3646 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3648 msr_info->data = vcpu->arch.osvw.status;
3650 case MSR_PLATFORM_INFO:
3651 if (!msr_info->host_initiated &&
3652 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3654 msr_info->data = vcpu->arch.msr_platform_info;
3656 case MSR_MISC_FEATURES_ENABLES:
3657 msr_info->data = vcpu->arch.msr_misc_features_enables;
3660 msr_info->data = vcpu->arch.msr_hwcr;
3663 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3664 return kvm_pmu_get_msr(vcpu, msr_info);
3665 return KVM_MSR_RET_INVALID;
3669 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3672 * Read or write a bunch of msrs. All parameters are kernel addresses.
3674 * @return number of msrs set successfully.
3676 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3677 struct kvm_msr_entry *entries,
3678 int (*do_msr)(struct kvm_vcpu *vcpu,
3679 unsigned index, u64 *data))
3683 for (i = 0; i < msrs->nmsrs; ++i)
3684 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3691 * Read or write a bunch of msrs. Parameters are user addresses.
3693 * @return number of msrs set successfully.
3695 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3696 int (*do_msr)(struct kvm_vcpu *vcpu,
3697 unsigned index, u64 *data),
3700 struct kvm_msrs msrs;
3701 struct kvm_msr_entry *entries;
3706 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3710 if (msrs.nmsrs >= MAX_IO_MSRS)
3713 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3714 entries = memdup_user(user_msrs->entries, size);
3715 if (IS_ERR(entries)) {
3716 r = PTR_ERR(entries);
3720 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3725 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3736 static inline bool kvm_can_mwait_in_guest(void)
3738 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3739 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3740 boot_cpu_has(X86_FEATURE_ARAT);
3743 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3744 struct kvm_cpuid2 __user *cpuid_arg)
3746 struct kvm_cpuid2 cpuid;
3750 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3753 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3758 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3764 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3769 case KVM_CAP_IRQCHIP:
3771 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3772 case KVM_CAP_SET_TSS_ADDR:
3773 case KVM_CAP_EXT_CPUID:
3774 case KVM_CAP_EXT_EMUL_CPUID:
3775 case KVM_CAP_CLOCKSOURCE:
3777 case KVM_CAP_NOP_IO_DELAY:
3778 case KVM_CAP_MP_STATE:
3779 case KVM_CAP_SYNC_MMU:
3780 case KVM_CAP_USER_NMI:
3781 case KVM_CAP_REINJECT_CONTROL:
3782 case KVM_CAP_IRQ_INJECT_STATUS:
3783 case KVM_CAP_IOEVENTFD:
3784 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3786 case KVM_CAP_PIT_STATE2:
3787 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3788 case KVM_CAP_VCPU_EVENTS:
3789 case KVM_CAP_HYPERV:
3790 case KVM_CAP_HYPERV_VAPIC:
3791 case KVM_CAP_HYPERV_SPIN:
3792 case KVM_CAP_HYPERV_SYNIC:
3793 case KVM_CAP_HYPERV_SYNIC2:
3794 case KVM_CAP_HYPERV_VP_INDEX:
3795 case KVM_CAP_HYPERV_EVENTFD:
3796 case KVM_CAP_HYPERV_TLBFLUSH:
3797 case KVM_CAP_HYPERV_SEND_IPI:
3798 case KVM_CAP_HYPERV_CPUID:
3799 case KVM_CAP_SYS_HYPERV_CPUID:
3800 case KVM_CAP_PCI_SEGMENT:
3801 case KVM_CAP_DEBUGREGS:
3802 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3804 case KVM_CAP_ASYNC_PF:
3805 case KVM_CAP_ASYNC_PF_INT:
3806 case KVM_CAP_GET_TSC_KHZ:
3807 case KVM_CAP_KVMCLOCK_CTRL:
3808 case KVM_CAP_READONLY_MEM:
3809 case KVM_CAP_HYPERV_TIME:
3810 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3811 case KVM_CAP_TSC_DEADLINE_TIMER:
3812 case KVM_CAP_DISABLE_QUIRKS:
3813 case KVM_CAP_SET_BOOT_CPU_ID:
3814 case KVM_CAP_SPLIT_IRQCHIP:
3815 case KVM_CAP_IMMEDIATE_EXIT:
3816 case KVM_CAP_PMU_EVENT_FILTER:
3817 case KVM_CAP_GET_MSR_FEATURES:
3818 case KVM_CAP_MSR_PLATFORM_INFO:
3819 case KVM_CAP_EXCEPTION_PAYLOAD:
3820 case KVM_CAP_SET_GUEST_DEBUG:
3821 case KVM_CAP_LAST_CPU:
3822 case KVM_CAP_X86_USER_SPACE_MSR:
3823 case KVM_CAP_X86_MSR_FILTER:
3824 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3825 #ifdef CONFIG_X86_SGX_KVM
3826 case KVM_CAP_SGX_ATTRIBUTE:
3828 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
3831 case KVM_CAP_SET_GUEST_DEBUG2:
3832 return KVM_GUESTDBG_VALID_MASK;
3833 #ifdef CONFIG_KVM_XEN
3834 case KVM_CAP_XEN_HVM:
3835 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
3836 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
3837 KVM_XEN_HVM_CONFIG_SHARED_INFO;
3838 if (sched_info_on())
3839 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
3842 case KVM_CAP_SYNC_REGS:
3843 r = KVM_SYNC_X86_VALID_FIELDS;
3845 case KVM_CAP_ADJUST_CLOCK:
3846 r = KVM_CLOCK_TSC_STABLE;
3848 case KVM_CAP_X86_DISABLE_EXITS:
3849 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3850 KVM_X86_DISABLE_EXITS_CSTATE;
3851 if(kvm_can_mwait_in_guest())
3852 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3854 case KVM_CAP_X86_SMM:
3855 /* SMBASE is usually relocated above 1M on modern chipsets,
3856 * and SMM handlers might indeed rely on 4G segment limits,
3857 * so do not report SMM to be available if real mode is
3858 * emulated via vm86 mode. Still, do not go to great lengths
3859 * to avoid userspace's usage of the feature, because it is a
3860 * fringe case that is not enabled except via specific settings
3861 * of the module parameters.
3863 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
3866 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
3868 case KVM_CAP_NR_VCPUS:
3869 r = KVM_SOFT_MAX_VCPUS;
3871 case KVM_CAP_MAX_VCPUS:
3874 case KVM_CAP_MAX_VCPU_ID:
3875 r = KVM_MAX_VCPU_ID;
3877 case KVM_CAP_PV_MMU: /* obsolete */
3881 r = KVM_MAX_MCE_BANKS;
3884 r = boot_cpu_has(X86_FEATURE_XSAVE);
3886 case KVM_CAP_TSC_CONTROL:
3887 r = kvm_has_tsc_control;
3889 case KVM_CAP_X2APIC_API:
3890 r = KVM_X2APIC_API_VALID_FLAGS;
3892 case KVM_CAP_NESTED_STATE:
3893 r = kvm_x86_ops.nested_ops->get_state ?
3894 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3896 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3897 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3899 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3900 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3902 case KVM_CAP_SMALLER_MAXPHYADDR:
3903 r = (int) allow_smaller_maxphyaddr;
3905 case KVM_CAP_STEAL_TIME:
3906 r = sched_info_on();
3908 case KVM_CAP_X86_BUS_LOCK_EXIT:
3909 if (kvm_has_bus_lock_exit)
3910 r = KVM_BUS_LOCK_DETECTION_OFF |
3911 KVM_BUS_LOCK_DETECTION_EXIT;
3922 long kvm_arch_dev_ioctl(struct file *filp,
3923 unsigned int ioctl, unsigned long arg)
3925 void __user *argp = (void __user *)arg;
3929 case KVM_GET_MSR_INDEX_LIST: {
3930 struct kvm_msr_list __user *user_msr_list = argp;
3931 struct kvm_msr_list msr_list;
3935 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3938 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3939 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3942 if (n < msr_list.nmsrs)
3945 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3946 num_msrs_to_save * sizeof(u32)))
3948 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3950 num_emulated_msrs * sizeof(u32)))
3955 case KVM_GET_SUPPORTED_CPUID:
3956 case KVM_GET_EMULATED_CPUID: {
3957 struct kvm_cpuid2 __user *cpuid_arg = argp;
3958 struct kvm_cpuid2 cpuid;
3961 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3964 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3970 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3975 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3977 if (copy_to_user(argp, &kvm_mce_cap_supported,
3978 sizeof(kvm_mce_cap_supported)))
3982 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3983 struct kvm_msr_list __user *user_msr_list = argp;
3984 struct kvm_msr_list msr_list;
3988 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3991 msr_list.nmsrs = num_msr_based_features;
3992 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3995 if (n < msr_list.nmsrs)
3998 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3999 num_msr_based_features * sizeof(u32)))
4005 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4007 case KVM_GET_SUPPORTED_HV_CPUID:
4008 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4018 static void wbinvd_ipi(void *garbage)
4023 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4025 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4028 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4030 /* Address WBINVD may be executed by guest */
4031 if (need_emulate_wbinvd(vcpu)) {
4032 if (static_call(kvm_x86_has_wbinvd_exit)())
4033 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4034 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4035 smp_call_function_single(vcpu->cpu,
4036 wbinvd_ipi, NULL, 1);
4039 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4041 /* Save host pkru register if supported */
4042 vcpu->arch.host_pkru = read_pkru();
4044 /* Apply any externally detected TSC adjustments (due to suspend) */
4045 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4046 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4047 vcpu->arch.tsc_offset_adjustment = 0;
4048 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4051 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4052 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4053 rdtsc() - vcpu->arch.last_host_tsc;
4055 mark_tsc_unstable("KVM discovered backwards TSC");
4057 if (kvm_check_tsc_unstable()) {
4058 u64 offset = kvm_compute_tsc_offset(vcpu,
4059 vcpu->arch.last_guest_tsc);
4060 kvm_vcpu_write_tsc_offset(vcpu, offset);
4061 vcpu->arch.tsc_catchup = 1;
4064 if (kvm_lapic_hv_timer_in_use(vcpu))
4065 kvm_lapic_restart_hv_timer(vcpu);
4068 * On a host with synchronized TSC, there is no need to update
4069 * kvmclock on vcpu->cpu migration
4071 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4072 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4073 if (vcpu->cpu != cpu)
4074 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4078 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4081 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4083 struct kvm_host_map map;
4084 struct kvm_steal_time *st;
4086 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4089 if (vcpu->arch.st.preempted)
4092 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4093 &vcpu->arch.st.cache, true))
4097 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4099 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4101 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4104 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4108 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4109 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4112 * Take the srcu lock as memslots will be accessed to check the gfn
4113 * cache generation against the memslots generation.
4115 idx = srcu_read_lock(&vcpu->kvm->srcu);
4116 if (kvm_xen_msr_enabled(vcpu->kvm))
4117 kvm_xen_runstate_set_preempted(vcpu);
4119 kvm_steal_time_set_preempted(vcpu);
4120 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4122 static_call(kvm_x86_vcpu_put)(vcpu);
4123 vcpu->arch.last_host_tsc = rdtsc();
4125 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4126 * on every vmexit, but if not, we might have a stale dr6 from the
4127 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4132 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4133 struct kvm_lapic_state *s)
4135 if (vcpu->arch.apicv_active)
4136 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4138 return kvm_apic_get_state(vcpu, s);
4141 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4142 struct kvm_lapic_state *s)
4146 r = kvm_apic_set_state(vcpu, s);
4149 update_cr8_intercept(vcpu);
4154 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4157 * We can accept userspace's request for interrupt injection
4158 * as long as we have a place to store the interrupt number.
4159 * The actual injection will happen when the CPU is able to
4160 * deliver the interrupt.
4162 if (kvm_cpu_has_extint(vcpu))
4165 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4166 return (!lapic_in_kernel(vcpu) ||
4167 kvm_apic_accept_pic_intr(vcpu));
4170 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4172 return kvm_arch_interrupt_allowed(vcpu) &&
4173 kvm_cpu_accept_dm_intr(vcpu);
4176 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4177 struct kvm_interrupt *irq)
4179 if (irq->irq >= KVM_NR_INTERRUPTS)
4182 if (!irqchip_in_kernel(vcpu->kvm)) {
4183 kvm_queue_interrupt(vcpu, irq->irq, false);
4184 kvm_make_request(KVM_REQ_EVENT, vcpu);
4189 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4190 * fail for in-kernel 8259.
4192 if (pic_in_kernel(vcpu->kvm))
4195 if (vcpu->arch.pending_external_vector != -1)
4198 vcpu->arch.pending_external_vector = irq->irq;
4199 kvm_make_request(KVM_REQ_EVENT, vcpu);
4203 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4205 kvm_inject_nmi(vcpu);
4210 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4212 kvm_make_request(KVM_REQ_SMI, vcpu);
4217 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4218 struct kvm_tpr_access_ctl *tac)
4222 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4226 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4230 unsigned bank_num = mcg_cap & 0xff, bank;
4233 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4235 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4238 vcpu->arch.mcg_cap = mcg_cap;
4239 /* Init IA32_MCG_CTL to all 1s */
4240 if (mcg_cap & MCG_CTL_P)
4241 vcpu->arch.mcg_ctl = ~(u64)0;
4242 /* Init IA32_MCi_CTL to all 1s */
4243 for (bank = 0; bank < bank_num; bank++)
4244 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4246 static_call(kvm_x86_setup_mce)(vcpu);
4251 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4252 struct kvm_x86_mce *mce)
4254 u64 mcg_cap = vcpu->arch.mcg_cap;
4255 unsigned bank_num = mcg_cap & 0xff;
4256 u64 *banks = vcpu->arch.mce_banks;
4258 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4261 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4262 * reporting is disabled
4264 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4265 vcpu->arch.mcg_ctl != ~(u64)0)
4267 banks += 4 * mce->bank;
4269 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4270 * reporting is disabled for the bank
4272 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4274 if (mce->status & MCI_STATUS_UC) {
4275 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4276 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4277 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4280 if (banks[1] & MCI_STATUS_VAL)
4281 mce->status |= MCI_STATUS_OVER;
4282 banks[2] = mce->addr;
4283 banks[3] = mce->misc;
4284 vcpu->arch.mcg_status = mce->mcg_status;
4285 banks[1] = mce->status;
4286 kvm_queue_exception(vcpu, MC_VECTOR);
4287 } else if (!(banks[1] & MCI_STATUS_VAL)
4288 || !(banks[1] & MCI_STATUS_UC)) {
4289 if (banks[1] & MCI_STATUS_VAL)
4290 mce->status |= MCI_STATUS_OVER;
4291 banks[2] = mce->addr;
4292 banks[3] = mce->misc;
4293 banks[1] = mce->status;
4295 banks[1] |= MCI_STATUS_OVER;
4299 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4300 struct kvm_vcpu_events *events)
4304 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4308 * In guest mode, payload delivery should be deferred,
4309 * so that the L1 hypervisor can intercept #PF before
4310 * CR2 is modified (or intercept #DB before DR6 is
4311 * modified under nVMX). Unless the per-VM capability,
4312 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4313 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4314 * opportunistically defer the exception payload, deliver it if the
4315 * capability hasn't been requested before processing a
4316 * KVM_GET_VCPU_EVENTS.
4318 if (!vcpu->kvm->arch.exception_payload_enabled &&
4319 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4320 kvm_deliver_exception_payload(vcpu);
4323 * The API doesn't provide the instruction length for software
4324 * exceptions, so don't report them. As long as the guest RIP
4325 * isn't advanced, we should expect to encounter the exception
4328 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4329 events->exception.injected = 0;
4330 events->exception.pending = 0;
4332 events->exception.injected = vcpu->arch.exception.injected;
4333 events->exception.pending = vcpu->arch.exception.pending;
4335 * For ABI compatibility, deliberately conflate
4336 * pending and injected exceptions when
4337 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4339 if (!vcpu->kvm->arch.exception_payload_enabled)
4340 events->exception.injected |=
4341 vcpu->arch.exception.pending;
4343 events->exception.nr = vcpu->arch.exception.nr;
4344 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4345 events->exception.error_code = vcpu->arch.exception.error_code;
4346 events->exception_has_payload = vcpu->arch.exception.has_payload;
4347 events->exception_payload = vcpu->arch.exception.payload;
4349 events->interrupt.injected =
4350 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4351 events->interrupt.nr = vcpu->arch.interrupt.nr;
4352 events->interrupt.soft = 0;
4353 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4355 events->nmi.injected = vcpu->arch.nmi_injected;
4356 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4357 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4358 events->nmi.pad = 0;
4360 events->sipi_vector = 0; /* never valid when reporting to user space */
4362 events->smi.smm = is_smm(vcpu);
4363 events->smi.pending = vcpu->arch.smi_pending;
4364 events->smi.smm_inside_nmi =
4365 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4366 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4368 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4369 | KVM_VCPUEVENT_VALID_SHADOW
4370 | KVM_VCPUEVENT_VALID_SMM);
4371 if (vcpu->kvm->arch.exception_payload_enabled)
4372 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4374 memset(&events->reserved, 0, sizeof(events->reserved));
4377 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4379 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4380 struct kvm_vcpu_events *events)
4382 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4383 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4384 | KVM_VCPUEVENT_VALID_SHADOW
4385 | KVM_VCPUEVENT_VALID_SMM
4386 | KVM_VCPUEVENT_VALID_PAYLOAD))
4389 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4390 if (!vcpu->kvm->arch.exception_payload_enabled)
4392 if (events->exception.pending)
4393 events->exception.injected = 0;
4395 events->exception_has_payload = 0;
4397 events->exception.pending = 0;
4398 events->exception_has_payload = 0;
4401 if ((events->exception.injected || events->exception.pending) &&
4402 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4405 /* INITs are latched while in SMM */
4406 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4407 (events->smi.smm || events->smi.pending) &&
4408 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4412 vcpu->arch.exception.injected = events->exception.injected;
4413 vcpu->arch.exception.pending = events->exception.pending;
4414 vcpu->arch.exception.nr = events->exception.nr;
4415 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4416 vcpu->arch.exception.error_code = events->exception.error_code;
4417 vcpu->arch.exception.has_payload = events->exception_has_payload;
4418 vcpu->arch.exception.payload = events->exception_payload;
4420 vcpu->arch.interrupt.injected = events->interrupt.injected;
4421 vcpu->arch.interrupt.nr = events->interrupt.nr;
4422 vcpu->arch.interrupt.soft = events->interrupt.soft;
4423 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4424 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4425 events->interrupt.shadow);
4427 vcpu->arch.nmi_injected = events->nmi.injected;
4428 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4429 vcpu->arch.nmi_pending = events->nmi.pending;
4430 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4432 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4433 lapic_in_kernel(vcpu))
4434 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4436 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4437 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4438 if (events->smi.smm)
4439 vcpu->arch.hflags |= HF_SMM_MASK;
4441 vcpu->arch.hflags &= ~HF_SMM_MASK;
4442 kvm_smm_changed(vcpu);
4445 vcpu->arch.smi_pending = events->smi.pending;
4447 if (events->smi.smm) {
4448 if (events->smi.smm_inside_nmi)
4449 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4451 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4454 if (lapic_in_kernel(vcpu)) {
4455 if (events->smi.latched_init)
4456 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4458 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4462 kvm_make_request(KVM_REQ_EVENT, vcpu);
4467 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4468 struct kvm_debugregs *dbgregs)
4472 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4473 kvm_get_dr(vcpu, 6, &val);
4475 dbgregs->dr7 = vcpu->arch.dr7;
4477 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4480 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4481 struct kvm_debugregs *dbgregs)
4486 if (!kvm_dr6_valid(dbgregs->dr6))
4488 if (!kvm_dr7_valid(dbgregs->dr7))
4491 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4492 kvm_update_dr0123(vcpu);
4493 vcpu->arch.dr6 = dbgregs->dr6;
4494 vcpu->arch.dr7 = dbgregs->dr7;
4495 kvm_update_dr7(vcpu);
4500 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4502 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4504 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4505 u64 xstate_bv = xsave->header.xfeatures;
4509 * Copy legacy XSAVE area, to avoid complications with CPUID
4510 * leaves 0 and 1 in the loop below.
4512 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4515 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4516 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4519 * Copy each region from the possibly compacted offset to the
4520 * non-compacted offset.
4522 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4524 u64 xfeature_mask = valid & -valid;
4525 int xfeature_nr = fls64(xfeature_mask) - 1;
4526 void *src = get_xsave_addr(xsave, xfeature_nr);
4529 u32 size, offset, ecx, edx;
4530 cpuid_count(XSTATE_CPUID, xfeature_nr,
4531 &size, &offset, &ecx, &edx);
4532 if (xfeature_nr == XFEATURE_PKRU)
4533 memcpy(dest + offset, &vcpu->arch.pkru,
4534 sizeof(vcpu->arch.pkru));
4536 memcpy(dest + offset, src, size);
4540 valid -= xfeature_mask;
4544 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4546 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4547 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4551 * Copy legacy XSAVE area, to avoid complications with CPUID
4552 * leaves 0 and 1 in the loop below.
4554 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4556 /* Set XSTATE_BV and possibly XCOMP_BV. */
4557 xsave->header.xfeatures = xstate_bv;
4558 if (boot_cpu_has(X86_FEATURE_XSAVES))
4559 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4562 * Copy each region from the non-compacted offset to the
4563 * possibly compacted offset.
4565 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4567 u64 xfeature_mask = valid & -valid;
4568 int xfeature_nr = fls64(xfeature_mask) - 1;
4569 void *dest = get_xsave_addr(xsave, xfeature_nr);
4572 u32 size, offset, ecx, edx;
4573 cpuid_count(XSTATE_CPUID, xfeature_nr,
4574 &size, &offset, &ecx, &edx);
4575 if (xfeature_nr == XFEATURE_PKRU)
4576 memcpy(&vcpu->arch.pkru, src + offset,
4577 sizeof(vcpu->arch.pkru));
4579 memcpy(dest, src + offset, size);
4582 valid -= xfeature_mask;
4586 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4587 struct kvm_xsave *guest_xsave)
4589 if (!vcpu->arch.guest_fpu)
4592 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4593 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4594 fill_xsave((u8 *) guest_xsave->region, vcpu);
4596 memcpy(guest_xsave->region,
4597 &vcpu->arch.guest_fpu->state.fxsave,
4598 sizeof(struct fxregs_state));
4599 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4600 XFEATURE_MASK_FPSSE;
4604 #define XSAVE_MXCSR_OFFSET 24
4606 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4607 struct kvm_xsave *guest_xsave)
4612 if (!vcpu->arch.guest_fpu)
4615 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4616 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4618 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4620 * Here we allow setting states that are not present in
4621 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4622 * with old userspace.
4624 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4626 load_xsave(vcpu, (u8 *)guest_xsave->region);
4628 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4629 mxcsr & ~mxcsr_feature_mask)
4631 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4632 guest_xsave->region, sizeof(struct fxregs_state));
4637 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4638 struct kvm_xcrs *guest_xcrs)
4640 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4641 guest_xcrs->nr_xcrs = 0;
4645 guest_xcrs->nr_xcrs = 1;
4646 guest_xcrs->flags = 0;
4647 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4648 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4651 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4652 struct kvm_xcrs *guest_xcrs)
4656 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4659 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4662 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4663 /* Only support XCR0 currently */
4664 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4665 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4666 guest_xcrs->xcrs[i].value);
4675 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4676 * stopped by the hypervisor. This function will be called from the host only.
4677 * EINVAL is returned when the host attempts to set the flag for a guest that
4678 * does not support pv clocks.
4680 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4682 if (!vcpu->arch.pv_time_enabled)
4684 vcpu->arch.pvclock_set_guest_stopped_request = true;
4685 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4689 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4690 struct kvm_enable_cap *cap)
4693 uint16_t vmcs_version;
4694 void __user *user_ptr;
4700 case KVM_CAP_HYPERV_SYNIC2:
4705 case KVM_CAP_HYPERV_SYNIC:
4706 if (!irqchip_in_kernel(vcpu->kvm))
4708 return kvm_hv_activate_synic(vcpu, cap->cap ==
4709 KVM_CAP_HYPERV_SYNIC2);
4710 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4711 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4713 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4715 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4716 if (copy_to_user(user_ptr, &vmcs_version,
4717 sizeof(vmcs_version)))
4721 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4722 if (!kvm_x86_ops.enable_direct_tlbflush)
4725 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4727 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4728 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4729 if (vcpu->arch.pv_cpuid.enforce)
4730 kvm_update_pv_runtime(vcpu);
4738 long kvm_arch_vcpu_ioctl(struct file *filp,
4739 unsigned int ioctl, unsigned long arg)
4741 struct kvm_vcpu *vcpu = filp->private_data;
4742 void __user *argp = (void __user *)arg;
4745 struct kvm_lapic_state *lapic;
4746 struct kvm_xsave *xsave;
4747 struct kvm_xcrs *xcrs;
4755 case KVM_GET_LAPIC: {
4757 if (!lapic_in_kernel(vcpu))
4759 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4760 GFP_KERNEL_ACCOUNT);
4765 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4769 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4774 case KVM_SET_LAPIC: {
4776 if (!lapic_in_kernel(vcpu))
4778 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4779 if (IS_ERR(u.lapic)) {
4780 r = PTR_ERR(u.lapic);
4784 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4787 case KVM_INTERRUPT: {
4788 struct kvm_interrupt irq;
4791 if (copy_from_user(&irq, argp, sizeof(irq)))
4793 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4797 r = kvm_vcpu_ioctl_nmi(vcpu);
4801 r = kvm_vcpu_ioctl_smi(vcpu);
4804 case KVM_SET_CPUID: {
4805 struct kvm_cpuid __user *cpuid_arg = argp;
4806 struct kvm_cpuid cpuid;
4809 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4811 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4814 case KVM_SET_CPUID2: {
4815 struct kvm_cpuid2 __user *cpuid_arg = argp;
4816 struct kvm_cpuid2 cpuid;
4819 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4821 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4822 cpuid_arg->entries);
4825 case KVM_GET_CPUID2: {
4826 struct kvm_cpuid2 __user *cpuid_arg = argp;
4827 struct kvm_cpuid2 cpuid;
4830 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4832 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4833 cpuid_arg->entries);
4837 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4842 case KVM_GET_MSRS: {
4843 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4844 r = msr_io(vcpu, argp, do_get_msr, 1);
4845 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4848 case KVM_SET_MSRS: {
4849 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4850 r = msr_io(vcpu, argp, do_set_msr, 0);
4851 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4854 case KVM_TPR_ACCESS_REPORTING: {
4855 struct kvm_tpr_access_ctl tac;
4858 if (copy_from_user(&tac, argp, sizeof(tac)))
4860 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4864 if (copy_to_user(argp, &tac, sizeof(tac)))
4869 case KVM_SET_VAPIC_ADDR: {
4870 struct kvm_vapic_addr va;
4874 if (!lapic_in_kernel(vcpu))
4877 if (copy_from_user(&va, argp, sizeof(va)))
4879 idx = srcu_read_lock(&vcpu->kvm->srcu);
4880 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4881 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4884 case KVM_X86_SETUP_MCE: {
4888 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4890 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4893 case KVM_X86_SET_MCE: {
4894 struct kvm_x86_mce mce;
4897 if (copy_from_user(&mce, argp, sizeof(mce)))
4899 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4902 case KVM_GET_VCPU_EVENTS: {
4903 struct kvm_vcpu_events events;
4905 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4908 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4913 case KVM_SET_VCPU_EVENTS: {
4914 struct kvm_vcpu_events events;
4917 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4920 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4923 case KVM_GET_DEBUGREGS: {
4924 struct kvm_debugregs dbgregs;
4926 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4929 if (copy_to_user(argp, &dbgregs,
4930 sizeof(struct kvm_debugregs)))
4935 case KVM_SET_DEBUGREGS: {
4936 struct kvm_debugregs dbgregs;
4939 if (copy_from_user(&dbgregs, argp,
4940 sizeof(struct kvm_debugregs)))
4943 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4946 case KVM_GET_XSAVE: {
4947 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4952 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4955 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4960 case KVM_SET_XSAVE: {
4961 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4962 if (IS_ERR(u.xsave)) {
4963 r = PTR_ERR(u.xsave);
4967 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4970 case KVM_GET_XCRS: {
4971 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4976 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4979 if (copy_to_user(argp, u.xcrs,
4980 sizeof(struct kvm_xcrs)))
4985 case KVM_SET_XCRS: {
4986 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4987 if (IS_ERR(u.xcrs)) {
4988 r = PTR_ERR(u.xcrs);
4992 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4995 case KVM_SET_TSC_KHZ: {
4999 user_tsc_khz = (u32)arg;
5001 if (kvm_has_tsc_control &&
5002 user_tsc_khz >= kvm_max_guest_tsc_khz)
5005 if (user_tsc_khz == 0)
5006 user_tsc_khz = tsc_khz;
5008 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5013 case KVM_GET_TSC_KHZ: {
5014 r = vcpu->arch.virtual_tsc_khz;
5017 case KVM_KVMCLOCK_CTRL: {
5018 r = kvm_set_guest_paused(vcpu);
5021 case KVM_ENABLE_CAP: {
5022 struct kvm_enable_cap cap;
5025 if (copy_from_user(&cap, argp, sizeof(cap)))
5027 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5030 case KVM_GET_NESTED_STATE: {
5031 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5035 if (!kvm_x86_ops.nested_ops->get_state)
5038 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5040 if (get_user(user_data_size, &user_kvm_nested_state->size))
5043 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5048 if (r > user_data_size) {
5049 if (put_user(r, &user_kvm_nested_state->size))
5059 case KVM_SET_NESTED_STATE: {
5060 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5061 struct kvm_nested_state kvm_state;
5065 if (!kvm_x86_ops.nested_ops->set_state)
5069 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5073 if (kvm_state.size < sizeof(kvm_state))
5076 if (kvm_state.flags &
5077 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5078 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5079 | KVM_STATE_NESTED_GIF_SET))
5082 /* nested_run_pending implies guest_mode. */
5083 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5084 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5087 idx = srcu_read_lock(&vcpu->kvm->srcu);
5088 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5089 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5092 case KVM_GET_SUPPORTED_HV_CPUID:
5093 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5095 #ifdef CONFIG_KVM_XEN
5096 case KVM_XEN_VCPU_GET_ATTR: {
5097 struct kvm_xen_vcpu_attr xva;
5100 if (copy_from_user(&xva, argp, sizeof(xva)))
5102 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5103 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5107 case KVM_XEN_VCPU_SET_ATTR: {
5108 struct kvm_xen_vcpu_attr xva;
5111 if (copy_from_user(&xva, argp, sizeof(xva)))
5113 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5127 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5129 return VM_FAULT_SIGBUS;
5132 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5136 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5138 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5142 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5145 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5148 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5149 unsigned long kvm_nr_mmu_pages)
5151 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5154 mutex_lock(&kvm->slots_lock);
5156 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5157 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5159 mutex_unlock(&kvm->slots_lock);
5163 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5165 return kvm->arch.n_max_mmu_pages;
5168 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5170 struct kvm_pic *pic = kvm->arch.vpic;
5174 switch (chip->chip_id) {
5175 case KVM_IRQCHIP_PIC_MASTER:
5176 memcpy(&chip->chip.pic, &pic->pics[0],
5177 sizeof(struct kvm_pic_state));
5179 case KVM_IRQCHIP_PIC_SLAVE:
5180 memcpy(&chip->chip.pic, &pic->pics[1],
5181 sizeof(struct kvm_pic_state));
5183 case KVM_IRQCHIP_IOAPIC:
5184 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5193 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5195 struct kvm_pic *pic = kvm->arch.vpic;
5199 switch (chip->chip_id) {
5200 case KVM_IRQCHIP_PIC_MASTER:
5201 spin_lock(&pic->lock);
5202 memcpy(&pic->pics[0], &chip->chip.pic,
5203 sizeof(struct kvm_pic_state));
5204 spin_unlock(&pic->lock);
5206 case KVM_IRQCHIP_PIC_SLAVE:
5207 spin_lock(&pic->lock);
5208 memcpy(&pic->pics[1], &chip->chip.pic,
5209 sizeof(struct kvm_pic_state));
5210 spin_unlock(&pic->lock);
5212 case KVM_IRQCHIP_IOAPIC:
5213 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5219 kvm_pic_update_irq(pic);
5223 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5225 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5227 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5229 mutex_lock(&kps->lock);
5230 memcpy(ps, &kps->channels, sizeof(*ps));
5231 mutex_unlock(&kps->lock);
5235 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5238 struct kvm_pit *pit = kvm->arch.vpit;
5240 mutex_lock(&pit->pit_state.lock);
5241 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5242 for (i = 0; i < 3; i++)
5243 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5244 mutex_unlock(&pit->pit_state.lock);
5248 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5250 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5251 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5252 sizeof(ps->channels));
5253 ps->flags = kvm->arch.vpit->pit_state.flags;
5254 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5255 memset(&ps->reserved, 0, sizeof(ps->reserved));
5259 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5263 u32 prev_legacy, cur_legacy;
5264 struct kvm_pit *pit = kvm->arch.vpit;
5266 mutex_lock(&pit->pit_state.lock);
5267 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5268 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5269 if (!prev_legacy && cur_legacy)
5271 memcpy(&pit->pit_state.channels, &ps->channels,
5272 sizeof(pit->pit_state.channels));
5273 pit->pit_state.flags = ps->flags;
5274 for (i = 0; i < 3; i++)
5275 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5277 mutex_unlock(&pit->pit_state.lock);
5281 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5282 struct kvm_reinject_control *control)
5284 struct kvm_pit *pit = kvm->arch.vpit;
5286 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5287 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5288 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5290 mutex_lock(&pit->pit_state.lock);
5291 kvm_pit_set_reinject(pit, control->pit_reinject);
5292 mutex_unlock(&pit->pit_state.lock);
5297 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5301 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5302 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5303 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5306 struct kvm_vcpu *vcpu;
5309 kvm_for_each_vcpu(i, vcpu, kvm)
5310 kvm_vcpu_kick(vcpu);
5313 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5316 if (!irqchip_in_kernel(kvm))
5319 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5320 irq_event->irq, irq_event->level,
5325 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5326 struct kvm_enable_cap *cap)
5334 case KVM_CAP_DISABLE_QUIRKS:
5335 kvm->arch.disabled_quirks = cap->args[0];
5338 case KVM_CAP_SPLIT_IRQCHIP: {
5339 mutex_lock(&kvm->lock);
5341 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5342 goto split_irqchip_unlock;
5344 if (irqchip_in_kernel(kvm))
5345 goto split_irqchip_unlock;
5346 if (kvm->created_vcpus)
5347 goto split_irqchip_unlock;
5348 r = kvm_setup_empty_irq_routing(kvm);
5350 goto split_irqchip_unlock;
5351 /* Pairs with irqchip_in_kernel. */
5353 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5354 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5356 split_irqchip_unlock:
5357 mutex_unlock(&kvm->lock);
5360 case KVM_CAP_X2APIC_API:
5362 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5365 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5366 kvm->arch.x2apic_format = true;
5367 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5368 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5372 case KVM_CAP_X86_DISABLE_EXITS:
5374 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5377 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5378 kvm_can_mwait_in_guest())
5379 kvm->arch.mwait_in_guest = true;
5380 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5381 kvm->arch.hlt_in_guest = true;
5382 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5383 kvm->arch.pause_in_guest = true;
5384 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5385 kvm->arch.cstate_in_guest = true;
5388 case KVM_CAP_MSR_PLATFORM_INFO:
5389 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5392 case KVM_CAP_EXCEPTION_PAYLOAD:
5393 kvm->arch.exception_payload_enabled = cap->args[0];
5396 case KVM_CAP_X86_USER_SPACE_MSR:
5397 kvm->arch.user_space_msr_mask = cap->args[0];
5400 case KVM_CAP_X86_BUS_LOCK_EXIT:
5402 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5405 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5406 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5409 if (kvm_has_bus_lock_exit &&
5410 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5411 kvm->arch.bus_lock_detection_enabled = true;
5414 #ifdef CONFIG_X86_SGX_KVM
5415 case KVM_CAP_SGX_ATTRIBUTE: {
5416 unsigned long allowed_attributes = 0;
5418 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5422 /* KVM only supports the PROVISIONKEY privileged attribute. */
5423 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5424 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5425 kvm->arch.sgx_provisioning_allowed = true;
5431 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5433 if (kvm_x86_ops.vm_copy_enc_context_from)
5434 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5443 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5445 struct kvm_x86_msr_filter *msr_filter;
5447 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5451 msr_filter->default_allow = default_allow;
5455 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5462 for (i = 0; i < msr_filter->count; i++)
5463 kfree(msr_filter->ranges[i].bitmap);
5468 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5469 struct kvm_msr_filter_range *user_range)
5471 struct msr_bitmap_range range;
5472 unsigned long *bitmap = NULL;
5476 if (!user_range->nmsrs)
5479 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5480 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5483 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5485 return PTR_ERR(bitmap);
5487 range = (struct msr_bitmap_range) {
5488 .flags = user_range->flags,
5489 .base = user_range->base,
5490 .nmsrs = user_range->nmsrs,
5494 if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) {
5504 /* Everything ok, add this range identifier. */
5505 msr_filter->ranges[msr_filter->count] = range;
5506 msr_filter->count++;
5514 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5516 struct kvm_msr_filter __user *user_msr_filter = argp;
5517 struct kvm_x86_msr_filter *new_filter, *old_filter;
5518 struct kvm_msr_filter filter;
5524 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5527 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5528 empty &= !filter.ranges[i].nmsrs;
5530 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5531 if (empty && !default_allow)
5534 new_filter = kvm_alloc_msr_filter(default_allow);
5538 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5539 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5541 kvm_free_msr_filter(new_filter);
5546 mutex_lock(&kvm->lock);
5548 /* The per-VM filter is protected by kvm->lock... */
5549 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5551 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5552 synchronize_srcu(&kvm->srcu);
5554 kvm_free_msr_filter(old_filter);
5556 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5557 mutex_unlock(&kvm->lock);
5562 long kvm_arch_vm_ioctl(struct file *filp,
5563 unsigned int ioctl, unsigned long arg)
5565 struct kvm *kvm = filp->private_data;
5566 void __user *argp = (void __user *)arg;
5569 * This union makes it completely explicit to gcc-3.x
5570 * that these two variables' stack usage should be
5571 * combined, not added together.
5574 struct kvm_pit_state ps;
5575 struct kvm_pit_state2 ps2;
5576 struct kvm_pit_config pit_config;
5580 case KVM_SET_TSS_ADDR:
5581 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5583 case KVM_SET_IDENTITY_MAP_ADDR: {
5586 mutex_lock(&kvm->lock);
5588 if (kvm->created_vcpus)
5589 goto set_identity_unlock;
5591 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5592 goto set_identity_unlock;
5593 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5594 set_identity_unlock:
5595 mutex_unlock(&kvm->lock);
5598 case KVM_SET_NR_MMU_PAGES:
5599 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5601 case KVM_GET_NR_MMU_PAGES:
5602 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5604 case KVM_CREATE_IRQCHIP: {
5605 mutex_lock(&kvm->lock);
5608 if (irqchip_in_kernel(kvm))
5609 goto create_irqchip_unlock;
5612 if (kvm->created_vcpus)
5613 goto create_irqchip_unlock;
5615 r = kvm_pic_init(kvm);
5617 goto create_irqchip_unlock;
5619 r = kvm_ioapic_init(kvm);
5621 kvm_pic_destroy(kvm);
5622 goto create_irqchip_unlock;
5625 r = kvm_setup_default_irq_routing(kvm);
5627 kvm_ioapic_destroy(kvm);
5628 kvm_pic_destroy(kvm);
5629 goto create_irqchip_unlock;
5631 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5633 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5634 create_irqchip_unlock:
5635 mutex_unlock(&kvm->lock);
5638 case KVM_CREATE_PIT:
5639 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5641 case KVM_CREATE_PIT2:
5643 if (copy_from_user(&u.pit_config, argp,
5644 sizeof(struct kvm_pit_config)))
5647 mutex_lock(&kvm->lock);
5650 goto create_pit_unlock;
5652 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5656 mutex_unlock(&kvm->lock);
5658 case KVM_GET_IRQCHIP: {
5659 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5660 struct kvm_irqchip *chip;
5662 chip = memdup_user(argp, sizeof(*chip));
5669 if (!irqchip_kernel(kvm))
5670 goto get_irqchip_out;
5671 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5673 goto get_irqchip_out;
5675 if (copy_to_user(argp, chip, sizeof(*chip)))
5676 goto get_irqchip_out;
5682 case KVM_SET_IRQCHIP: {
5683 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5684 struct kvm_irqchip *chip;
5686 chip = memdup_user(argp, sizeof(*chip));
5693 if (!irqchip_kernel(kvm))
5694 goto set_irqchip_out;
5695 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5702 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5705 if (!kvm->arch.vpit)
5707 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5711 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5718 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5720 mutex_lock(&kvm->lock);
5722 if (!kvm->arch.vpit)
5724 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5726 mutex_unlock(&kvm->lock);
5729 case KVM_GET_PIT2: {
5731 if (!kvm->arch.vpit)
5733 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5737 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5742 case KVM_SET_PIT2: {
5744 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5746 mutex_lock(&kvm->lock);
5748 if (!kvm->arch.vpit)
5750 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5752 mutex_unlock(&kvm->lock);
5755 case KVM_REINJECT_CONTROL: {
5756 struct kvm_reinject_control control;
5758 if (copy_from_user(&control, argp, sizeof(control)))
5761 if (!kvm->arch.vpit)
5763 r = kvm_vm_ioctl_reinject(kvm, &control);
5766 case KVM_SET_BOOT_CPU_ID:
5768 mutex_lock(&kvm->lock);
5769 if (kvm->created_vcpus)
5772 kvm->arch.bsp_vcpu_id = arg;
5773 mutex_unlock(&kvm->lock);
5775 #ifdef CONFIG_KVM_XEN
5776 case KVM_XEN_HVM_CONFIG: {
5777 struct kvm_xen_hvm_config xhc;
5779 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5781 r = kvm_xen_hvm_config(kvm, &xhc);
5784 case KVM_XEN_HVM_GET_ATTR: {
5785 struct kvm_xen_hvm_attr xha;
5788 if (copy_from_user(&xha, argp, sizeof(xha)))
5790 r = kvm_xen_hvm_get_attr(kvm, &xha);
5791 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5795 case KVM_XEN_HVM_SET_ATTR: {
5796 struct kvm_xen_hvm_attr xha;
5799 if (copy_from_user(&xha, argp, sizeof(xha)))
5801 r = kvm_xen_hvm_set_attr(kvm, &xha);
5805 case KVM_SET_CLOCK: {
5806 struct kvm_arch *ka = &kvm->arch;
5807 struct kvm_clock_data user_ns;
5811 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5820 * TODO: userspace has to take care of races with VCPU_RUN, so
5821 * kvm_gen_update_masterclock() can be cut down to locked
5822 * pvclock_update_vm_gtod_copy().
5824 kvm_gen_update_masterclock(kvm);
5827 * This pairs with kvm_guest_time_update(): when masterclock is
5828 * in use, we use master_kernel_ns + kvmclock_offset to set
5829 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
5830 * is slightly ahead) here we risk going negative on unsigned
5831 * 'system_time' when 'user_ns.clock' is very small.
5833 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
5834 if (kvm->arch.use_master_clock)
5835 now_ns = ka->master_kernel_ns;
5837 now_ns = get_kvmclock_base_ns();
5838 ka->kvmclock_offset = user_ns.clock - now_ns;
5839 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
5841 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5844 case KVM_GET_CLOCK: {
5845 struct kvm_clock_data user_ns;
5848 now_ns = get_kvmclock_ns(kvm);
5849 user_ns.clock = now_ns;
5850 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5851 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5854 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5859 case KVM_MEMORY_ENCRYPT_OP: {
5861 if (kvm_x86_ops.mem_enc_op)
5862 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
5865 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5866 struct kvm_enc_region region;
5869 if (copy_from_user(®ion, argp, sizeof(region)))
5873 if (kvm_x86_ops.mem_enc_reg_region)
5874 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
5877 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5878 struct kvm_enc_region region;
5881 if (copy_from_user(®ion, argp, sizeof(region)))
5885 if (kvm_x86_ops.mem_enc_unreg_region)
5886 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
5889 case KVM_HYPERV_EVENTFD: {
5890 struct kvm_hyperv_eventfd hvevfd;
5893 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5895 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5898 case KVM_SET_PMU_EVENT_FILTER:
5899 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5901 case KVM_X86_SET_MSR_FILTER:
5902 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5911 static void kvm_init_msr_list(void)
5913 struct x86_pmu_capability x86_pmu;
5917 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5918 "Please update the fixed PMCs in msrs_to_saved_all[]");
5920 perf_get_x86_pmu_capability(&x86_pmu);
5922 num_msrs_to_save = 0;
5923 num_emulated_msrs = 0;
5924 num_msr_based_features = 0;
5926 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5927 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5931 * Even MSRs that are valid in the host may not be exposed
5932 * to the guests in some cases.
5934 switch (msrs_to_save_all[i]) {
5935 case MSR_IA32_BNDCFGS:
5936 if (!kvm_mpx_supported())
5940 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5943 case MSR_IA32_UMWAIT_CONTROL:
5944 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5947 case MSR_IA32_RTIT_CTL:
5948 case MSR_IA32_RTIT_STATUS:
5949 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5952 case MSR_IA32_RTIT_CR3_MATCH:
5953 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5954 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5957 case MSR_IA32_RTIT_OUTPUT_BASE:
5958 case MSR_IA32_RTIT_OUTPUT_MASK:
5959 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5960 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5961 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5964 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5965 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5966 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5967 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5970 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5971 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5972 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5975 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5976 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5977 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5984 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5987 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5988 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
5991 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5994 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5995 struct kvm_msr_entry msr;
5997 msr.index = msr_based_features_all[i];
5998 if (kvm_get_msr_feature(&msr))
6001 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6005 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6013 if (!(lapic_in_kernel(vcpu) &&
6014 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6015 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6026 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6033 if (!(lapic_in_kernel(vcpu) &&
6034 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6036 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6038 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6048 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6049 struct kvm_segment *var, int seg)
6051 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6054 void kvm_get_segment(struct kvm_vcpu *vcpu,
6055 struct kvm_segment *var, int seg)
6057 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6060 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6061 struct x86_exception *exception)
6065 BUG_ON(!mmu_is_nested(vcpu));
6067 /* NPT walks are always user-walks */
6068 access |= PFERR_USER_MASK;
6069 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6074 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6075 struct x86_exception *exception)
6077 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6078 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6080 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6082 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6083 struct x86_exception *exception)
6085 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6086 access |= PFERR_FETCH_MASK;
6087 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6090 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6091 struct x86_exception *exception)
6093 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6094 access |= PFERR_WRITE_MASK;
6095 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6097 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6099 /* uses this to access any guest's mapped memory without checking CPL */
6100 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6101 struct x86_exception *exception)
6103 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6106 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6107 struct kvm_vcpu *vcpu, u32 access,
6108 struct x86_exception *exception)
6111 int r = X86EMUL_CONTINUE;
6114 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6116 unsigned offset = addr & (PAGE_SIZE-1);
6117 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6120 if (gpa == UNMAPPED_GVA)
6121 return X86EMUL_PROPAGATE_FAULT;
6122 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6125 r = X86EMUL_IO_NEEDED;
6137 /* used for instruction fetching */
6138 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6139 gva_t addr, void *val, unsigned int bytes,
6140 struct x86_exception *exception)
6142 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6143 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6147 /* Inline kvm_read_guest_virt_helper for speed. */
6148 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6150 if (unlikely(gpa == UNMAPPED_GVA))
6151 return X86EMUL_PROPAGATE_FAULT;
6153 offset = addr & (PAGE_SIZE-1);
6154 if (WARN_ON(offset + bytes > PAGE_SIZE))
6155 bytes = (unsigned)PAGE_SIZE - offset;
6156 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6158 if (unlikely(ret < 0))
6159 return X86EMUL_IO_NEEDED;
6161 return X86EMUL_CONTINUE;
6164 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6165 gva_t addr, void *val, unsigned int bytes,
6166 struct x86_exception *exception)
6168 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6171 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6172 * is returned, but our callers are not ready for that and they blindly
6173 * call kvm_inject_page_fault. Ensure that they at least do not leak
6174 * uninitialized kernel stack memory into cr2 and error code.
6176 memset(exception, 0, sizeof(*exception));
6177 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6180 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6182 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6183 gva_t addr, void *val, unsigned int bytes,
6184 struct x86_exception *exception, bool system)
6186 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6189 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6190 access |= PFERR_USER_MASK;
6192 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6195 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6196 unsigned long addr, void *val, unsigned int bytes)
6198 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6199 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6201 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6204 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6205 struct kvm_vcpu *vcpu, u32 access,
6206 struct x86_exception *exception)
6209 int r = X86EMUL_CONTINUE;
6212 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6215 unsigned offset = addr & (PAGE_SIZE-1);
6216 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6219 if (gpa == UNMAPPED_GVA)
6220 return X86EMUL_PROPAGATE_FAULT;
6221 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6223 r = X86EMUL_IO_NEEDED;
6235 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6236 unsigned int bytes, struct x86_exception *exception,
6239 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6240 u32 access = PFERR_WRITE_MASK;
6242 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6243 access |= PFERR_USER_MASK;
6245 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6249 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6250 unsigned int bytes, struct x86_exception *exception)
6252 /* kvm_write_guest_virt_system can pull in tons of pages. */
6253 vcpu->arch.l1tf_flush_l1d = true;
6255 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6256 PFERR_WRITE_MASK, exception);
6258 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6260 int handle_ud(struct kvm_vcpu *vcpu)
6262 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6263 int emul_type = EMULTYPE_TRAP_UD;
6264 char sig[5]; /* ud2; .ascii "kvm" */
6265 struct x86_exception e;
6267 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6270 if (force_emulation_prefix &&
6271 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6272 sig, sizeof(sig), &e) == 0 &&
6273 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6274 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6275 emul_type = EMULTYPE_TRAP_UD_FORCED;
6278 return kvm_emulate_instruction(vcpu, emul_type);
6280 EXPORT_SYMBOL_GPL(handle_ud);
6282 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6283 gpa_t gpa, bool write)
6285 /* For APIC access vmexit */
6286 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6289 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6290 trace_vcpu_match_mmio(gva, gpa, write, true);
6297 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6298 gpa_t *gpa, struct x86_exception *exception,
6301 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6302 | (write ? PFERR_WRITE_MASK : 0);
6305 * currently PKRU is only applied to ept enabled guest so
6306 * there is no pkey in EPT page table for L1 guest or EPT
6307 * shadow page table for L2 guest.
6309 if (vcpu_match_mmio_gva(vcpu, gva)
6310 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6311 vcpu->arch.mmio_access, 0, access)) {
6312 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6313 (gva & (PAGE_SIZE - 1));
6314 trace_vcpu_match_mmio(gva, *gpa, write, false);
6318 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6320 if (*gpa == UNMAPPED_GVA)
6323 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6326 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6327 const void *val, int bytes)
6331 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6334 kvm_page_track_write(vcpu, gpa, val, bytes);
6338 struct read_write_emulator_ops {
6339 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6341 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6342 void *val, int bytes);
6343 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6344 int bytes, void *val);
6345 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6346 void *val, int bytes);
6350 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6352 if (vcpu->mmio_read_completed) {
6353 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6354 vcpu->mmio_fragments[0].gpa, val);
6355 vcpu->mmio_read_completed = 0;
6362 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6363 void *val, int bytes)
6365 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6368 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6369 void *val, int bytes)
6371 return emulator_write_phys(vcpu, gpa, val, bytes);
6374 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6376 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6377 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6380 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6381 void *val, int bytes)
6383 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6384 return X86EMUL_IO_NEEDED;
6387 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6388 void *val, int bytes)
6390 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6392 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6393 return X86EMUL_CONTINUE;
6396 static const struct read_write_emulator_ops read_emultor = {
6397 .read_write_prepare = read_prepare,
6398 .read_write_emulate = read_emulate,
6399 .read_write_mmio = vcpu_mmio_read,
6400 .read_write_exit_mmio = read_exit_mmio,
6403 static const struct read_write_emulator_ops write_emultor = {
6404 .read_write_emulate = write_emulate,
6405 .read_write_mmio = write_mmio,
6406 .read_write_exit_mmio = write_exit_mmio,
6410 static int emulator_read_write_onepage(unsigned long addr, void *val,
6412 struct x86_exception *exception,
6413 struct kvm_vcpu *vcpu,
6414 const struct read_write_emulator_ops *ops)
6418 bool write = ops->write;
6419 struct kvm_mmio_fragment *frag;
6420 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6423 * If the exit was due to a NPF we may already have a GPA.
6424 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6425 * Note, this cannot be used on string operations since string
6426 * operation using rep will only have the initial GPA from the NPF
6429 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6430 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6431 gpa = ctxt->gpa_val;
6432 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6434 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6436 return X86EMUL_PROPAGATE_FAULT;
6439 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6440 return X86EMUL_CONTINUE;
6443 * Is this MMIO handled locally?
6445 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6446 if (handled == bytes)
6447 return X86EMUL_CONTINUE;
6453 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6454 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6458 return X86EMUL_CONTINUE;
6461 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6463 void *val, unsigned int bytes,
6464 struct x86_exception *exception,
6465 const struct read_write_emulator_ops *ops)
6467 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6471 if (ops->read_write_prepare &&
6472 ops->read_write_prepare(vcpu, val, bytes))
6473 return X86EMUL_CONTINUE;
6475 vcpu->mmio_nr_fragments = 0;
6477 /* Crossing a page boundary? */
6478 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6481 now = -addr & ~PAGE_MASK;
6482 rc = emulator_read_write_onepage(addr, val, now, exception,
6485 if (rc != X86EMUL_CONTINUE)
6488 if (ctxt->mode != X86EMUL_MODE_PROT64)
6494 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6496 if (rc != X86EMUL_CONTINUE)
6499 if (!vcpu->mmio_nr_fragments)
6502 gpa = vcpu->mmio_fragments[0].gpa;
6504 vcpu->mmio_needed = 1;
6505 vcpu->mmio_cur_fragment = 0;
6507 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6508 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6509 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6510 vcpu->run->mmio.phys_addr = gpa;
6512 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6515 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6519 struct x86_exception *exception)
6521 return emulator_read_write(ctxt, addr, val, bytes,
6522 exception, &read_emultor);
6525 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6529 struct x86_exception *exception)
6531 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6532 exception, &write_emultor);
6535 #define CMPXCHG_TYPE(t, ptr, old, new) \
6536 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6538 #ifdef CONFIG_X86_64
6539 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6541 # define CMPXCHG64(ptr, old, new) \
6542 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6545 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6550 struct x86_exception *exception)
6552 struct kvm_host_map map;
6553 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6559 /* guests cmpxchg8b have to be emulated atomically */
6560 if (bytes > 8 || (bytes & (bytes - 1)))
6563 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6565 if (gpa == UNMAPPED_GVA ||
6566 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6570 * Emulate the atomic as a straight write to avoid #AC if SLD is
6571 * enabled in the host and the access splits a cache line.
6573 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6574 page_line_mask = ~(cache_line_size() - 1);
6576 page_line_mask = PAGE_MASK;
6578 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6581 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6584 kaddr = map.hva + offset_in_page(gpa);
6588 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6591 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6594 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6597 exchanged = CMPXCHG64(kaddr, old, new);
6603 kvm_vcpu_unmap(vcpu, &map, true);
6606 return X86EMUL_CMPXCHG_FAILED;
6608 kvm_page_track_write(vcpu, gpa, new, bytes);
6610 return X86EMUL_CONTINUE;
6613 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6615 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6618 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6622 for (i = 0; i < vcpu->arch.pio.count; i++) {
6623 if (vcpu->arch.pio.in)
6624 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6625 vcpu->arch.pio.size, pd);
6627 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6628 vcpu->arch.pio.port, vcpu->arch.pio.size,
6632 pd += vcpu->arch.pio.size;
6637 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6638 unsigned short port, void *val,
6639 unsigned int count, bool in)
6641 vcpu->arch.pio.port = port;
6642 vcpu->arch.pio.in = in;
6643 vcpu->arch.pio.count = count;
6644 vcpu->arch.pio.size = size;
6646 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6647 vcpu->arch.pio.count = 0;
6651 vcpu->run->exit_reason = KVM_EXIT_IO;
6652 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6653 vcpu->run->io.size = size;
6654 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6655 vcpu->run->io.count = count;
6656 vcpu->run->io.port = port;
6661 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6662 unsigned short port, void *val, unsigned int count)
6666 if (vcpu->arch.pio.count)
6669 memset(vcpu->arch.pio_data, 0, size * count);
6671 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6674 memcpy(val, vcpu->arch.pio_data, size * count);
6675 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6676 vcpu->arch.pio.count = 0;
6683 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6684 int size, unsigned short port, void *val,
6687 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6691 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6692 unsigned short port, const void *val,
6695 memcpy(vcpu->arch.pio_data, val, size * count);
6696 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6697 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6700 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6701 int size, unsigned short port,
6702 const void *val, unsigned int count)
6704 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6707 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6709 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6712 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6714 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6717 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6719 if (!need_emulate_wbinvd(vcpu))
6720 return X86EMUL_CONTINUE;
6722 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6723 int cpu = get_cpu();
6725 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6726 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6727 wbinvd_ipi, NULL, 1);
6729 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6732 return X86EMUL_CONTINUE;
6735 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6737 kvm_emulate_wbinvd_noskip(vcpu);
6738 return kvm_skip_emulated_instruction(vcpu);
6740 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6744 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6746 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6749 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6750 unsigned long *dest)
6752 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6755 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6756 unsigned long value)
6759 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6762 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6764 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6767 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6769 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6770 unsigned long value;
6774 value = kvm_read_cr0(vcpu);
6777 value = vcpu->arch.cr2;
6780 value = kvm_read_cr3(vcpu);
6783 value = kvm_read_cr4(vcpu);
6786 value = kvm_get_cr8(vcpu);
6789 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6796 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6798 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6803 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6806 vcpu->arch.cr2 = val;
6809 res = kvm_set_cr3(vcpu, val);
6812 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6815 res = kvm_set_cr8(vcpu, val);
6818 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6825 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6827 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
6830 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6832 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
6835 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6837 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
6840 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6842 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
6845 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6847 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
6850 static unsigned long emulator_get_cached_segment_base(
6851 struct x86_emulate_ctxt *ctxt, int seg)
6853 return get_segment_base(emul_to_vcpu(ctxt), seg);
6856 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6857 struct desc_struct *desc, u32 *base3,
6860 struct kvm_segment var;
6862 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6863 *selector = var.selector;
6866 memset(desc, 0, sizeof(*desc));
6874 set_desc_limit(desc, var.limit);
6875 set_desc_base(desc, (unsigned long)var.base);
6876 #ifdef CONFIG_X86_64
6878 *base3 = var.base >> 32;
6880 desc->type = var.type;
6882 desc->dpl = var.dpl;
6883 desc->p = var.present;
6884 desc->avl = var.avl;
6892 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6893 struct desc_struct *desc, u32 base3,
6896 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6897 struct kvm_segment var;
6899 var.selector = selector;
6900 var.base = get_desc_base(desc);
6901 #ifdef CONFIG_X86_64
6902 var.base |= ((u64)base3) << 32;
6904 var.limit = get_desc_limit(desc);
6906 var.limit = (var.limit << 12) | 0xfff;
6907 var.type = desc->type;
6908 var.dpl = desc->dpl;
6913 var.avl = desc->avl;
6914 var.present = desc->p;
6915 var.unusable = !var.present;
6918 kvm_set_segment(vcpu, &var, seg);
6922 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6923 u32 msr_index, u64 *pdata)
6925 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6928 r = kvm_get_msr(vcpu, msr_index, pdata);
6930 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6931 /* Bounce to user space */
6932 return X86EMUL_IO_NEEDED;
6938 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6939 u32 msr_index, u64 data)
6941 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6944 r = kvm_set_msr(vcpu, msr_index, data);
6946 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6947 /* Bounce to user space */
6948 return X86EMUL_IO_NEEDED;
6954 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6956 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6958 return vcpu->arch.smbase;
6961 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6963 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6965 vcpu->arch.smbase = smbase;
6968 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6971 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6974 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6975 u32 pmc, u64 *pdata)
6977 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6980 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6982 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6985 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6986 struct x86_instruction_info *info,
6987 enum x86_intercept_stage stage)
6989 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
6993 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6994 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6997 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7000 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7002 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7005 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7007 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7010 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7012 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7015 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7017 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7020 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7022 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7025 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7027 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7030 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7032 return emul_to_vcpu(ctxt)->arch.hflags;
7035 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
7037 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
7040 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
7041 const char *smstate)
7043 return static_call(kvm_x86_pre_leave_smm)(emul_to_vcpu(ctxt), smstate);
7046 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
7048 kvm_smm_changed(emul_to_vcpu(ctxt));
7051 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7053 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7056 static const struct x86_emulate_ops emulate_ops = {
7057 .read_gpr = emulator_read_gpr,
7058 .write_gpr = emulator_write_gpr,
7059 .read_std = emulator_read_std,
7060 .write_std = emulator_write_std,
7061 .read_phys = kvm_read_guest_phys_system,
7062 .fetch = kvm_fetch_guest_virt,
7063 .read_emulated = emulator_read_emulated,
7064 .write_emulated = emulator_write_emulated,
7065 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7066 .invlpg = emulator_invlpg,
7067 .pio_in_emulated = emulator_pio_in_emulated,
7068 .pio_out_emulated = emulator_pio_out_emulated,
7069 .get_segment = emulator_get_segment,
7070 .set_segment = emulator_set_segment,
7071 .get_cached_segment_base = emulator_get_cached_segment_base,
7072 .get_gdt = emulator_get_gdt,
7073 .get_idt = emulator_get_idt,
7074 .set_gdt = emulator_set_gdt,
7075 .set_idt = emulator_set_idt,
7076 .get_cr = emulator_get_cr,
7077 .set_cr = emulator_set_cr,
7078 .cpl = emulator_get_cpl,
7079 .get_dr = emulator_get_dr,
7080 .set_dr = emulator_set_dr,
7081 .get_smbase = emulator_get_smbase,
7082 .set_smbase = emulator_set_smbase,
7083 .set_msr = emulator_set_msr,
7084 .get_msr = emulator_get_msr,
7085 .check_pmc = emulator_check_pmc,
7086 .read_pmc = emulator_read_pmc,
7087 .halt = emulator_halt,
7088 .wbinvd = emulator_wbinvd,
7089 .fix_hypercall = emulator_fix_hypercall,
7090 .intercept = emulator_intercept,
7091 .get_cpuid = emulator_get_cpuid,
7092 .guest_has_long_mode = emulator_guest_has_long_mode,
7093 .guest_has_movbe = emulator_guest_has_movbe,
7094 .guest_has_fxsr = emulator_guest_has_fxsr,
7095 .set_nmi_mask = emulator_set_nmi_mask,
7096 .get_hflags = emulator_get_hflags,
7097 .set_hflags = emulator_set_hflags,
7098 .pre_leave_smm = emulator_pre_leave_smm,
7099 .post_leave_smm = emulator_post_leave_smm,
7100 .set_xcr = emulator_set_xcr,
7103 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7105 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7107 * an sti; sti; sequence only disable interrupts for the first
7108 * instruction. So, if the last instruction, be it emulated or
7109 * not, left the system with the INT_STI flag enabled, it
7110 * means that the last instruction is an sti. We should not
7111 * leave the flag on in this case. The same goes for mov ss
7113 if (int_shadow & mask)
7115 if (unlikely(int_shadow || mask)) {
7116 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7118 kvm_make_request(KVM_REQ_EVENT, vcpu);
7122 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7124 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7125 if (ctxt->exception.vector == PF_VECTOR)
7126 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7128 if (ctxt->exception.error_code_valid)
7129 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7130 ctxt->exception.error_code);
7132 kvm_queue_exception(vcpu, ctxt->exception.vector);
7136 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7138 struct x86_emulate_ctxt *ctxt;
7140 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7142 pr_err("kvm: failed to allocate vcpu's emulator\n");
7147 ctxt->ops = &emulate_ops;
7148 vcpu->arch.emulate_ctxt = ctxt;
7153 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7155 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7158 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7160 ctxt->gpa_available = false;
7161 ctxt->eflags = kvm_get_rflags(vcpu);
7162 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7164 ctxt->eip = kvm_rip_read(vcpu);
7165 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7166 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7167 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7168 cs_db ? X86EMUL_MODE_PROT32 :
7169 X86EMUL_MODE_PROT16;
7170 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7171 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7172 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7174 init_decode_cache(ctxt);
7175 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7178 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7180 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7183 init_emulate_ctxt(vcpu);
7187 ctxt->_eip = ctxt->eip + inc_eip;
7188 ret = emulate_int_real(ctxt, irq);
7190 if (ret != X86EMUL_CONTINUE) {
7191 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7193 ctxt->eip = ctxt->_eip;
7194 kvm_rip_write(vcpu, ctxt->eip);
7195 kvm_set_rflags(vcpu, ctxt->eflags);
7198 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7200 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7202 ++vcpu->stat.insn_emulation_fail;
7203 trace_kvm_emulate_insn_failed(vcpu);
7205 if (emulation_type & EMULTYPE_VMWARE_GP) {
7206 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7210 if (emulation_type & EMULTYPE_SKIP) {
7211 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7212 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7213 vcpu->run->internal.ndata = 0;
7217 kvm_queue_exception(vcpu, UD_VECTOR);
7219 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7220 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7221 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7222 vcpu->run->internal.ndata = 0;
7229 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7230 bool write_fault_to_shadow_pgtable,
7233 gpa_t gpa = cr2_or_gpa;
7236 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7239 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7240 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7243 if (!vcpu->arch.mmu->direct_map) {
7245 * Write permission should be allowed since only
7246 * write access need to be emulated.
7248 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7251 * If the mapping is invalid in guest, let cpu retry
7252 * it to generate fault.
7254 if (gpa == UNMAPPED_GVA)
7259 * Do not retry the unhandleable instruction if it faults on the
7260 * readonly host memory, otherwise it will goto a infinite loop:
7261 * retry instruction -> write #PF -> emulation fail -> retry
7262 * instruction -> ...
7264 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7267 * If the instruction failed on the error pfn, it can not be fixed,
7268 * report the error to userspace.
7270 if (is_error_noslot_pfn(pfn))
7273 kvm_release_pfn_clean(pfn);
7275 /* The instructions are well-emulated on direct mmu. */
7276 if (vcpu->arch.mmu->direct_map) {
7277 unsigned int indirect_shadow_pages;
7279 write_lock(&vcpu->kvm->mmu_lock);
7280 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7281 write_unlock(&vcpu->kvm->mmu_lock);
7283 if (indirect_shadow_pages)
7284 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7290 * if emulation was due to access to shadowed page table
7291 * and it failed try to unshadow page and re-enter the
7292 * guest to let CPU execute the instruction.
7294 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7297 * If the access faults on its page table, it can not
7298 * be fixed by unprotecting shadow page and it should
7299 * be reported to userspace.
7301 return !write_fault_to_shadow_pgtable;
7304 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7305 gpa_t cr2_or_gpa, int emulation_type)
7307 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7308 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7310 last_retry_eip = vcpu->arch.last_retry_eip;
7311 last_retry_addr = vcpu->arch.last_retry_addr;
7314 * If the emulation is caused by #PF and it is non-page_table
7315 * writing instruction, it means the VM-EXIT is caused by shadow
7316 * page protected, we can zap the shadow page and retry this
7317 * instruction directly.
7319 * Note: if the guest uses a non-page-table modifying instruction
7320 * on the PDE that points to the instruction, then we will unmap
7321 * the instruction and go to an infinite loop. So, we cache the
7322 * last retried eip and the last fault address, if we meet the eip
7323 * and the address again, we can break out of the potential infinite
7326 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7328 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7331 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7332 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7335 if (x86_page_table_writing_insn(ctxt))
7338 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7341 vcpu->arch.last_retry_eip = ctxt->eip;
7342 vcpu->arch.last_retry_addr = cr2_or_gpa;
7344 if (!vcpu->arch.mmu->direct_map)
7345 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7347 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7352 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7353 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7355 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7357 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7358 /* This is a good place to trace that we are exiting SMM. */
7359 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7361 /* Process a latched INIT or SMI, if any. */
7362 kvm_make_request(KVM_REQ_EVENT, vcpu);
7365 kvm_mmu_reset_context(vcpu);
7368 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7377 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7378 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7383 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7385 struct kvm_run *kvm_run = vcpu->run;
7387 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7388 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7389 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7390 kvm_run->debug.arch.exception = DB_VECTOR;
7391 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7394 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7398 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7400 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7403 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7408 * rflags is the old, "raw" value of the flags. The new value has
7409 * not been saved yet.
7411 * This is correct even for TF set by the guest, because "the
7412 * processor will not generate this exception after the instruction
7413 * that sets the TF flag".
7415 if (unlikely(rflags & X86_EFLAGS_TF))
7416 r = kvm_vcpu_do_singlestep(vcpu);
7419 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7421 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7423 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7424 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7425 struct kvm_run *kvm_run = vcpu->run;
7426 unsigned long eip = kvm_get_linear_rip(vcpu);
7427 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7428 vcpu->arch.guest_debug_dr7,
7432 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7433 kvm_run->debug.arch.pc = eip;
7434 kvm_run->debug.arch.exception = DB_VECTOR;
7435 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7441 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7442 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7443 unsigned long eip = kvm_get_linear_rip(vcpu);
7444 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7449 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7458 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7460 switch (ctxt->opcode_len) {
7467 case 0xe6: /* OUT */
7471 case 0x6c: /* INS */
7473 case 0x6e: /* OUTS */
7480 case 0x33: /* RDPMC */
7490 * Decode to be emulated instruction. Return EMULATION_OK if success.
7492 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7493 void *insn, int insn_len)
7495 int r = EMULATION_OK;
7496 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7498 init_emulate_ctxt(vcpu);
7501 * We will reenter on the same instruction since we do not set
7502 * complete_userspace_io. This does not handle watchpoints yet,
7503 * those would be handled in the emulate_ops.
7505 if (!(emulation_type & EMULTYPE_SKIP) &&
7506 kvm_vcpu_check_breakpoint(vcpu, &r))
7509 ctxt->interruptibility = 0;
7510 ctxt->have_exception = false;
7511 ctxt->exception.vector = -1;
7512 ctxt->perm_ok = false;
7514 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7516 r = x86_decode_insn(ctxt, insn, insn_len);
7518 trace_kvm_emulate_insn_start(vcpu);
7519 ++vcpu->stat.insn_emulation;
7523 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7525 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7526 int emulation_type, void *insn, int insn_len)
7529 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7530 bool writeback = true;
7531 bool write_fault_to_spt;
7533 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7536 vcpu->arch.l1tf_flush_l1d = true;
7539 * Clear write_fault_to_shadow_pgtable here to ensure it is
7542 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7543 vcpu->arch.write_fault_to_shadow_pgtable = false;
7545 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7546 kvm_clear_exception_queue(vcpu);
7548 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7550 if (r != EMULATION_OK) {
7551 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7552 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7553 kvm_queue_exception(vcpu, UD_VECTOR);
7556 if (reexecute_instruction(vcpu, cr2_or_gpa,
7560 if (ctxt->have_exception) {
7562 * #UD should result in just EMULATION_FAILED, and trap-like
7563 * exception should not be encountered during decode.
7565 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7566 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7567 inject_emulated_exception(vcpu);
7570 return handle_emulation_failure(vcpu, emulation_type);
7574 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7575 !is_vmware_backdoor_opcode(ctxt)) {
7576 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7581 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7582 * for kvm_skip_emulated_instruction(). The caller is responsible for
7583 * updating interruptibility state and injecting single-step #DBs.
7585 if (emulation_type & EMULTYPE_SKIP) {
7586 kvm_rip_write(vcpu, ctxt->_eip);
7587 if (ctxt->eflags & X86_EFLAGS_RF)
7588 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7592 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7595 /* this is needed for vmware backdoor interface to work since it
7596 changes registers values during IO operation */
7597 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7598 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7599 emulator_invalidate_register_cache(ctxt);
7603 if (emulation_type & EMULTYPE_PF) {
7604 /* Save the faulting GPA (cr2) in the address field */
7605 ctxt->exception.address = cr2_or_gpa;
7607 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7608 if (vcpu->arch.mmu->direct_map) {
7609 ctxt->gpa_available = true;
7610 ctxt->gpa_val = cr2_or_gpa;
7613 /* Sanitize the address out of an abundance of paranoia. */
7614 ctxt->exception.address = 0;
7617 r = x86_emulate_insn(ctxt);
7619 if (r == EMULATION_INTERCEPTED)
7622 if (r == EMULATION_FAILED) {
7623 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7627 return handle_emulation_failure(vcpu, emulation_type);
7630 if (ctxt->have_exception) {
7632 if (inject_emulated_exception(vcpu))
7634 } else if (vcpu->arch.pio.count) {
7635 if (!vcpu->arch.pio.in) {
7636 /* FIXME: return into emulator if single-stepping. */
7637 vcpu->arch.pio.count = 0;
7640 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7643 } else if (vcpu->mmio_needed) {
7644 ++vcpu->stat.mmio_exits;
7646 if (!vcpu->mmio_is_write)
7649 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7650 } else if (r == EMULATION_RESTART)
7656 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7657 toggle_interruptibility(vcpu, ctxt->interruptibility);
7658 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7659 if (!ctxt->have_exception ||
7660 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7661 kvm_rip_write(vcpu, ctxt->eip);
7662 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7663 r = kvm_vcpu_do_singlestep(vcpu);
7664 if (kvm_x86_ops.update_emulated_instruction)
7665 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7666 __kvm_set_rflags(vcpu, ctxt->eflags);
7670 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7671 * do nothing, and it will be requested again as soon as
7672 * the shadow expires. But we still need to check here,
7673 * because POPF has no interrupt shadow.
7675 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7676 kvm_make_request(KVM_REQ_EVENT, vcpu);
7678 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7683 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7685 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7687 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7689 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7690 void *insn, int insn_len)
7692 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7694 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7696 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7698 vcpu->arch.pio.count = 0;
7702 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7704 vcpu->arch.pio.count = 0;
7706 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7709 return kvm_skip_emulated_instruction(vcpu);
7712 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7713 unsigned short port)
7715 unsigned long val = kvm_rax_read(vcpu);
7716 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7722 * Workaround userspace that relies on old KVM behavior of %rip being
7723 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7726 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7727 vcpu->arch.complete_userspace_io =
7728 complete_fast_pio_out_port_0x7e;
7729 kvm_skip_emulated_instruction(vcpu);
7731 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7732 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7737 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7741 /* We should only ever be called with arch.pio.count equal to 1 */
7742 BUG_ON(vcpu->arch.pio.count != 1);
7744 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7745 vcpu->arch.pio.count = 0;
7749 /* For size less than 4 we merge, else we zero extend */
7750 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7753 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7754 * the copy and tracing
7756 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7757 kvm_rax_write(vcpu, val);
7759 return kvm_skip_emulated_instruction(vcpu);
7762 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7763 unsigned short port)
7768 /* For size less than 4 we merge, else we zero extend */
7769 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7771 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7773 kvm_rax_write(vcpu, val);
7777 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7778 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7783 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7788 ret = kvm_fast_pio_in(vcpu, size, port);
7790 ret = kvm_fast_pio_out(vcpu, size, port);
7791 return ret && kvm_skip_emulated_instruction(vcpu);
7793 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7795 static int kvmclock_cpu_down_prep(unsigned int cpu)
7797 __this_cpu_write(cpu_tsc_khz, 0);
7801 static void tsc_khz_changed(void *data)
7803 struct cpufreq_freqs *freq = data;
7804 unsigned long khz = 0;
7808 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7809 khz = cpufreq_quick_get(raw_smp_processor_id());
7812 __this_cpu_write(cpu_tsc_khz, khz);
7815 #ifdef CONFIG_X86_64
7816 static void kvm_hyperv_tsc_notifier(void)
7819 struct kvm_vcpu *vcpu;
7821 unsigned long flags;
7823 mutex_lock(&kvm_lock);
7824 list_for_each_entry(kvm, &vm_list, vm_list)
7825 kvm_make_mclock_inprogress_request(kvm);
7827 hyperv_stop_tsc_emulation();
7829 /* TSC frequency always matches when on Hyper-V */
7830 for_each_present_cpu(cpu)
7831 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7832 kvm_max_guest_tsc_khz = tsc_khz;
7834 list_for_each_entry(kvm, &vm_list, vm_list) {
7835 struct kvm_arch *ka = &kvm->arch;
7837 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
7838 pvclock_update_vm_gtod_copy(kvm);
7839 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
7841 kvm_for_each_vcpu(cpu, vcpu, kvm)
7842 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7844 kvm_for_each_vcpu(cpu, vcpu, kvm)
7845 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7847 mutex_unlock(&kvm_lock);
7851 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7854 struct kvm_vcpu *vcpu;
7855 int i, send_ipi = 0;
7858 * We allow guests to temporarily run on slowing clocks,
7859 * provided we notify them after, or to run on accelerating
7860 * clocks, provided we notify them before. Thus time never
7863 * However, we have a problem. We can't atomically update
7864 * the frequency of a given CPU from this function; it is
7865 * merely a notifier, which can be called from any CPU.
7866 * Changing the TSC frequency at arbitrary points in time
7867 * requires a recomputation of local variables related to
7868 * the TSC for each VCPU. We must flag these local variables
7869 * to be updated and be sure the update takes place with the
7870 * new frequency before any guests proceed.
7872 * Unfortunately, the combination of hotplug CPU and frequency
7873 * change creates an intractable locking scenario; the order
7874 * of when these callouts happen is undefined with respect to
7875 * CPU hotplug, and they can race with each other. As such,
7876 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7877 * undefined; you can actually have a CPU frequency change take
7878 * place in between the computation of X and the setting of the
7879 * variable. To protect against this problem, all updates of
7880 * the per_cpu tsc_khz variable are done in an interrupt
7881 * protected IPI, and all callers wishing to update the value
7882 * must wait for a synchronous IPI to complete (which is trivial
7883 * if the caller is on the CPU already). This establishes the
7884 * necessary total order on variable updates.
7886 * Note that because a guest time update may take place
7887 * anytime after the setting of the VCPU's request bit, the
7888 * correct TSC value must be set before the request. However,
7889 * to ensure the update actually makes it to any guest which
7890 * starts running in hardware virtualization between the set
7891 * and the acquisition of the spinlock, we must also ping the
7892 * CPU after setting the request bit.
7896 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7898 mutex_lock(&kvm_lock);
7899 list_for_each_entry(kvm, &vm_list, vm_list) {
7900 kvm_for_each_vcpu(i, vcpu, kvm) {
7901 if (vcpu->cpu != cpu)
7903 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7904 if (vcpu->cpu != raw_smp_processor_id())
7908 mutex_unlock(&kvm_lock);
7910 if (freq->old < freq->new && send_ipi) {
7912 * We upscale the frequency. Must make the guest
7913 * doesn't see old kvmclock values while running with
7914 * the new frequency, otherwise we risk the guest sees
7915 * time go backwards.
7917 * In case we update the frequency for another cpu
7918 * (which might be in guest context) send an interrupt
7919 * to kick the cpu out of guest context. Next time
7920 * guest context is entered kvmclock will be updated,
7921 * so the guest will not see stale values.
7923 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7927 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7930 struct cpufreq_freqs *freq = data;
7933 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7935 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7938 for_each_cpu(cpu, freq->policy->cpus)
7939 __kvmclock_cpufreq_notifier(freq, cpu);
7944 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7945 .notifier_call = kvmclock_cpufreq_notifier
7948 static int kvmclock_cpu_online(unsigned int cpu)
7950 tsc_khz_changed(NULL);
7954 static void kvm_timer_init(void)
7956 max_tsc_khz = tsc_khz;
7958 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7959 #ifdef CONFIG_CPU_FREQ
7960 struct cpufreq_policy *policy;
7964 policy = cpufreq_cpu_get(cpu);
7966 if (policy->cpuinfo.max_freq)
7967 max_tsc_khz = policy->cpuinfo.max_freq;
7968 cpufreq_cpu_put(policy);
7972 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7973 CPUFREQ_TRANSITION_NOTIFIER);
7976 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7977 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7980 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7981 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7983 int kvm_is_in_guest(void)
7985 return __this_cpu_read(current_vcpu) != NULL;
7988 static int kvm_is_user_mode(void)
7992 if (__this_cpu_read(current_vcpu))
7993 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
7995 return user_mode != 0;
7998 static unsigned long kvm_get_guest_ip(void)
8000 unsigned long ip = 0;
8002 if (__this_cpu_read(current_vcpu))
8003 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8008 static void kvm_handle_intel_pt_intr(void)
8010 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8012 kvm_make_request(KVM_REQ_PMI, vcpu);
8013 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8014 (unsigned long *)&vcpu->arch.pmu.global_status);
8017 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8018 .is_in_guest = kvm_is_in_guest,
8019 .is_user_mode = kvm_is_user_mode,
8020 .get_guest_ip = kvm_get_guest_ip,
8021 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8024 #ifdef CONFIG_X86_64
8025 static void pvclock_gtod_update_fn(struct work_struct *work)
8029 struct kvm_vcpu *vcpu;
8032 mutex_lock(&kvm_lock);
8033 list_for_each_entry(kvm, &vm_list, vm_list)
8034 kvm_for_each_vcpu(i, vcpu, kvm)
8035 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8036 atomic_set(&kvm_guest_has_master_clock, 0);
8037 mutex_unlock(&kvm_lock);
8040 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8043 * Notification about pvclock gtod data update.
8045 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8048 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8049 struct timekeeper *tk = priv;
8051 update_pvclock_gtod(tk);
8053 /* disable master clock if host does not trust, or does not
8054 * use, TSC based clocksource.
8056 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8057 atomic_read(&kvm_guest_has_master_clock) != 0)
8058 queue_work(system_long_wq, &pvclock_gtod_work);
8063 static struct notifier_block pvclock_gtod_notifier = {
8064 .notifier_call = pvclock_gtod_notify,
8068 int kvm_arch_init(void *opaque)
8070 struct kvm_x86_init_ops *ops = opaque;
8073 if (kvm_x86_ops.hardware_enable) {
8074 printk(KERN_ERR "kvm: already loaded the other module\n");
8079 if (!ops->cpu_has_kvm_support()) {
8080 pr_err_ratelimited("kvm: no hardware support\n");
8084 if (ops->disabled_by_bios()) {
8085 pr_err_ratelimited("kvm: disabled by bios\n");
8091 * KVM explicitly assumes that the guest has an FPU and
8092 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8093 * vCPU's FPU state as a fxregs_state struct.
8095 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8096 printk(KERN_ERR "kvm: inadequate fpu\n");
8102 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8103 __alignof__(struct fpu), SLAB_ACCOUNT,
8105 if (!x86_fpu_cache) {
8106 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8110 x86_emulator_cache = kvm_alloc_emulator_cache();
8111 if (!x86_emulator_cache) {
8112 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8113 goto out_free_x86_fpu_cache;
8116 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8117 if (!user_return_msrs) {
8118 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8119 goto out_free_x86_emulator_cache;
8122 r = kvm_mmu_module_init();
8124 goto out_free_percpu;
8128 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8130 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8131 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8132 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8135 if (pi_inject_timer == -1)
8136 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8137 #ifdef CONFIG_X86_64
8138 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8140 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8141 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8147 free_percpu(user_return_msrs);
8148 out_free_x86_emulator_cache:
8149 kmem_cache_destroy(x86_emulator_cache);
8150 out_free_x86_fpu_cache:
8151 kmem_cache_destroy(x86_fpu_cache);
8156 void kvm_arch_exit(void)
8158 #ifdef CONFIG_X86_64
8159 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8160 clear_hv_tscchange_cb();
8163 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8165 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8166 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8167 CPUFREQ_TRANSITION_NOTIFIER);
8168 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8169 #ifdef CONFIG_X86_64
8170 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8172 kvm_x86_ops.hardware_enable = NULL;
8173 kvm_mmu_module_exit();
8174 free_percpu(user_return_msrs);
8175 kmem_cache_destroy(x86_fpu_cache);
8176 #ifdef CONFIG_KVM_XEN
8177 static_key_deferred_flush(&kvm_xen_enabled);
8178 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8182 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8184 ++vcpu->stat.halt_exits;
8185 if (lapic_in_kernel(vcpu)) {
8186 vcpu->arch.mp_state = state;
8189 vcpu->run->exit_reason = reason;
8194 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8196 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8198 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8200 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8202 int ret = kvm_skip_emulated_instruction(vcpu);
8204 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8205 * KVM_EXIT_DEBUG here.
8207 return kvm_vcpu_halt(vcpu) && ret;
8209 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8211 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8213 int ret = kvm_skip_emulated_instruction(vcpu);
8215 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8217 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8219 #ifdef CONFIG_X86_64
8220 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8221 unsigned long clock_type)
8223 struct kvm_clock_pairing clock_pairing;
8224 struct timespec64 ts;
8228 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8229 return -KVM_EOPNOTSUPP;
8231 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8232 return -KVM_EOPNOTSUPP;
8234 clock_pairing.sec = ts.tv_sec;
8235 clock_pairing.nsec = ts.tv_nsec;
8236 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8237 clock_pairing.flags = 0;
8238 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8241 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8242 sizeof(struct kvm_clock_pairing)))
8250 * kvm_pv_kick_cpu_op: Kick a vcpu.
8252 * @apicid - apicid of vcpu to be kicked.
8254 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8256 struct kvm_lapic_irq lapic_irq;
8258 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8259 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8260 lapic_irq.level = 0;
8261 lapic_irq.dest_id = apicid;
8262 lapic_irq.msi_redir_hint = false;
8264 lapic_irq.delivery_mode = APIC_DM_REMRD;
8265 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8268 bool kvm_apicv_activated(struct kvm *kvm)
8270 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8272 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8274 void kvm_apicv_init(struct kvm *kvm, bool enable)
8277 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8278 &kvm->arch.apicv_inhibit_reasons);
8280 set_bit(APICV_INHIBIT_REASON_DISABLE,
8281 &kvm->arch.apicv_inhibit_reasons);
8283 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8285 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8287 struct kvm_vcpu *target = NULL;
8288 struct kvm_apic_map *map;
8290 vcpu->stat.directed_yield_attempted++;
8293 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8295 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8296 target = map->phys_map[dest_id]->vcpu;
8300 if (!target || !READ_ONCE(target->ready))
8303 /* Ignore requests to yield to self */
8307 if (kvm_vcpu_yield_to(target) <= 0)
8310 vcpu->stat.directed_yield_successful++;
8316 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8318 unsigned long nr, a0, a1, a2, a3, ret;
8321 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8322 return kvm_xen_hypercall(vcpu);
8324 if (kvm_hv_hypercall_enabled(vcpu))
8325 return kvm_hv_hypercall(vcpu);
8327 nr = kvm_rax_read(vcpu);
8328 a0 = kvm_rbx_read(vcpu);
8329 a1 = kvm_rcx_read(vcpu);
8330 a2 = kvm_rdx_read(vcpu);
8331 a3 = kvm_rsi_read(vcpu);
8333 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8335 op_64_bit = is_64_bit_mode(vcpu);
8344 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8352 case KVM_HC_VAPIC_POLL_IRQ:
8355 case KVM_HC_KICK_CPU:
8356 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8359 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8360 kvm_sched_yield(vcpu, a1);
8363 #ifdef CONFIG_X86_64
8364 case KVM_HC_CLOCK_PAIRING:
8365 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8368 case KVM_HC_SEND_IPI:
8369 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8372 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8374 case KVM_HC_SCHED_YIELD:
8375 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8378 kvm_sched_yield(vcpu, a0);
8388 kvm_rax_write(vcpu, ret);
8390 ++vcpu->stat.hypercalls;
8391 return kvm_skip_emulated_instruction(vcpu);
8393 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8395 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8397 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8398 char instruction[3];
8399 unsigned long rip = kvm_rip_read(vcpu);
8401 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8403 return emulator_write_emulated(ctxt, rip, instruction, 3,
8407 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8409 return vcpu->run->request_interrupt_window &&
8410 likely(!pic_in_kernel(vcpu->kvm));
8413 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8415 struct kvm_run *kvm_run = vcpu->run;
8418 * if_flag is obsolete and useless, so do not bother
8419 * setting it for SEV-ES guests. Userspace can just
8420 * use kvm_run->ready_for_interrupt_injection.
8422 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8423 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8425 kvm_run->cr8 = kvm_get_cr8(vcpu);
8426 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8427 kvm_run->ready_for_interrupt_injection =
8428 pic_in_kernel(vcpu->kvm) ||
8429 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8432 kvm_run->flags |= KVM_RUN_X86_SMM;
8435 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8439 if (!kvm_x86_ops.update_cr8_intercept)
8442 if (!lapic_in_kernel(vcpu))
8445 if (vcpu->arch.apicv_active)
8448 if (!vcpu->arch.apic->vapic_addr)
8449 max_irr = kvm_lapic_find_highest_irr(vcpu);
8456 tpr = kvm_lapic_get_cr8(vcpu);
8458 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8462 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8464 if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
8467 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8468 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8472 return kvm_x86_ops.nested_ops->check_events(vcpu);
8475 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8477 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8478 vcpu->arch.exception.error_code = false;
8479 static_call(kvm_x86_queue_exception)(vcpu);
8482 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8485 bool can_inject = true;
8487 /* try to reinject previous events if any */
8489 if (vcpu->arch.exception.injected) {
8490 kvm_inject_exception(vcpu);
8494 * Do not inject an NMI or interrupt if there is a pending
8495 * exception. Exceptions and interrupts are recognized at
8496 * instruction boundaries, i.e. the start of an instruction.
8497 * Trap-like exceptions, e.g. #DB, have higher priority than
8498 * NMIs and interrupts, i.e. traps are recognized before an
8499 * NMI/interrupt that's pending on the same instruction.
8500 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8501 * priority, but are only generated (pended) during instruction
8502 * execution, i.e. a pending fault-like exception means the
8503 * fault occurred on the *previous* instruction and must be
8504 * serviced prior to recognizing any new events in order to
8505 * fully complete the previous instruction.
8507 else if (!vcpu->arch.exception.pending) {
8508 if (vcpu->arch.nmi_injected) {
8509 static_call(kvm_x86_set_nmi)(vcpu);
8511 } else if (vcpu->arch.interrupt.injected) {
8512 static_call(kvm_x86_set_irq)(vcpu);
8517 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8518 vcpu->arch.exception.pending);
8521 * Call check_nested_events() even if we reinjected a previous event
8522 * in order for caller to determine if it should require immediate-exit
8523 * from L2 to L1 due to pending L1 events which require exit
8526 if (is_guest_mode(vcpu)) {
8527 r = kvm_check_nested_events(vcpu);
8532 /* try to inject new event if pending */
8533 if (vcpu->arch.exception.pending) {
8534 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8535 vcpu->arch.exception.has_error_code,
8536 vcpu->arch.exception.error_code);
8538 vcpu->arch.exception.pending = false;
8539 vcpu->arch.exception.injected = true;
8541 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8542 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8545 if (vcpu->arch.exception.nr == DB_VECTOR) {
8546 kvm_deliver_exception_payload(vcpu);
8547 if (vcpu->arch.dr7 & DR7_GD) {
8548 vcpu->arch.dr7 &= ~DR7_GD;
8549 kvm_update_dr7(vcpu);
8553 kvm_inject_exception(vcpu);
8558 * Finally, inject interrupt events. If an event cannot be injected
8559 * due to architectural conditions (e.g. IF=0) a window-open exit
8560 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8561 * and can architecturally be injected, but we cannot do it right now:
8562 * an interrupt could have arrived just now and we have to inject it
8563 * as a vmexit, or there could already an event in the queue, which is
8564 * indicated by can_inject. In that case we request an immediate exit
8565 * in order to make progress and get back here for another iteration.
8566 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8568 if (vcpu->arch.smi_pending) {
8569 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8573 vcpu->arch.smi_pending = false;
8574 ++vcpu->arch.smi_count;
8578 static_call(kvm_x86_enable_smi_window)(vcpu);
8581 if (vcpu->arch.nmi_pending) {
8582 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8586 --vcpu->arch.nmi_pending;
8587 vcpu->arch.nmi_injected = true;
8588 static_call(kvm_x86_set_nmi)(vcpu);
8590 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8592 if (vcpu->arch.nmi_pending)
8593 static_call(kvm_x86_enable_nmi_window)(vcpu);
8596 if (kvm_cpu_has_injectable_intr(vcpu)) {
8597 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8601 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8602 static_call(kvm_x86_set_irq)(vcpu);
8603 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8605 if (kvm_cpu_has_injectable_intr(vcpu))
8606 static_call(kvm_x86_enable_irq_window)(vcpu);
8609 if (is_guest_mode(vcpu) &&
8610 kvm_x86_ops.nested_ops->hv_timer_pending &&
8611 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8612 *req_immediate_exit = true;
8614 WARN_ON(vcpu->arch.exception.pending);
8618 *req_immediate_exit = true;
8622 static void process_nmi(struct kvm_vcpu *vcpu)
8627 * x86 is limited to one NMI running, and one NMI pending after it.
8628 * If an NMI is already in progress, limit further NMIs to just one.
8629 * Otherwise, allow two (and we'll inject the first one immediately).
8631 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8634 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8635 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8636 kvm_make_request(KVM_REQ_EVENT, vcpu);
8639 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8642 flags |= seg->g << 23;
8643 flags |= seg->db << 22;
8644 flags |= seg->l << 21;
8645 flags |= seg->avl << 20;
8646 flags |= seg->present << 15;
8647 flags |= seg->dpl << 13;
8648 flags |= seg->s << 12;
8649 flags |= seg->type << 8;
8653 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8655 struct kvm_segment seg;
8658 kvm_get_segment(vcpu, &seg, n);
8659 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8662 offset = 0x7f84 + n * 12;
8664 offset = 0x7f2c + (n - 3) * 12;
8666 put_smstate(u32, buf, offset + 8, seg.base);
8667 put_smstate(u32, buf, offset + 4, seg.limit);
8668 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8671 #ifdef CONFIG_X86_64
8672 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8674 struct kvm_segment seg;
8678 kvm_get_segment(vcpu, &seg, n);
8679 offset = 0x7e00 + n * 16;
8681 flags = enter_smm_get_segment_flags(&seg) >> 8;
8682 put_smstate(u16, buf, offset, seg.selector);
8683 put_smstate(u16, buf, offset + 2, flags);
8684 put_smstate(u32, buf, offset + 4, seg.limit);
8685 put_smstate(u64, buf, offset + 8, seg.base);
8689 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8692 struct kvm_segment seg;
8696 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8697 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8698 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8699 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8701 for (i = 0; i < 8; i++)
8702 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
8704 kvm_get_dr(vcpu, 6, &val);
8705 put_smstate(u32, buf, 0x7fcc, (u32)val);
8706 kvm_get_dr(vcpu, 7, &val);
8707 put_smstate(u32, buf, 0x7fc8, (u32)val);
8709 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8710 put_smstate(u32, buf, 0x7fc4, seg.selector);
8711 put_smstate(u32, buf, 0x7f64, seg.base);
8712 put_smstate(u32, buf, 0x7f60, seg.limit);
8713 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8715 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8716 put_smstate(u32, buf, 0x7fc0, seg.selector);
8717 put_smstate(u32, buf, 0x7f80, seg.base);
8718 put_smstate(u32, buf, 0x7f7c, seg.limit);
8719 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8721 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8722 put_smstate(u32, buf, 0x7f74, dt.address);
8723 put_smstate(u32, buf, 0x7f70, dt.size);
8725 static_call(kvm_x86_get_idt)(vcpu, &dt);
8726 put_smstate(u32, buf, 0x7f58, dt.address);
8727 put_smstate(u32, buf, 0x7f54, dt.size);
8729 for (i = 0; i < 6; i++)
8730 enter_smm_save_seg_32(vcpu, buf, i);
8732 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8735 put_smstate(u32, buf, 0x7efc, 0x00020000);
8736 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8739 #ifdef CONFIG_X86_64
8740 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8743 struct kvm_segment seg;
8747 for (i = 0; i < 16; i++)
8748 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
8750 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8751 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8753 kvm_get_dr(vcpu, 6, &val);
8754 put_smstate(u64, buf, 0x7f68, val);
8755 kvm_get_dr(vcpu, 7, &val);
8756 put_smstate(u64, buf, 0x7f60, val);
8758 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8759 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8760 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8762 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8765 put_smstate(u32, buf, 0x7efc, 0x00020064);
8767 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8769 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8770 put_smstate(u16, buf, 0x7e90, seg.selector);
8771 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8772 put_smstate(u32, buf, 0x7e94, seg.limit);
8773 put_smstate(u64, buf, 0x7e98, seg.base);
8775 static_call(kvm_x86_get_idt)(vcpu, &dt);
8776 put_smstate(u32, buf, 0x7e84, dt.size);
8777 put_smstate(u64, buf, 0x7e88, dt.address);
8779 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8780 put_smstate(u16, buf, 0x7e70, seg.selector);
8781 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8782 put_smstate(u32, buf, 0x7e74, seg.limit);
8783 put_smstate(u64, buf, 0x7e78, seg.base);
8785 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8786 put_smstate(u32, buf, 0x7e64, dt.size);
8787 put_smstate(u64, buf, 0x7e68, dt.address);
8789 for (i = 0; i < 6; i++)
8790 enter_smm_save_seg_64(vcpu, buf, i);
8794 static void enter_smm(struct kvm_vcpu *vcpu)
8796 struct kvm_segment cs, ds;
8801 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8802 memset(buf, 0, 512);
8803 #ifdef CONFIG_X86_64
8804 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8805 enter_smm_save_state_64(vcpu, buf);
8808 enter_smm_save_state_32(vcpu, buf);
8811 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8812 * vCPU state (e.g. leave guest mode) after we've saved the state into
8813 * the SMM state-save area.
8815 static_call(kvm_x86_pre_enter_smm)(vcpu, buf);
8817 vcpu->arch.hflags |= HF_SMM_MASK;
8818 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8820 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
8821 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8823 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
8825 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8826 kvm_rip_write(vcpu, 0x8000);
8828 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8829 static_call(kvm_x86_set_cr0)(vcpu, cr0);
8830 vcpu->arch.cr0 = cr0;
8832 static_call(kvm_x86_set_cr4)(vcpu, 0);
8834 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8835 dt.address = dt.size = 0;
8836 static_call(kvm_x86_set_idt)(vcpu, &dt);
8838 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8840 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8841 cs.base = vcpu->arch.smbase;
8846 cs.limit = ds.limit = 0xffffffff;
8847 cs.type = ds.type = 0x3;
8848 cs.dpl = ds.dpl = 0;
8853 cs.avl = ds.avl = 0;
8854 cs.present = ds.present = 1;
8855 cs.unusable = ds.unusable = 0;
8856 cs.padding = ds.padding = 0;
8858 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8859 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8860 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8861 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8862 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8863 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8865 #ifdef CONFIG_X86_64
8866 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8867 static_call(kvm_x86_set_efer)(vcpu, 0);
8870 kvm_update_cpuid_runtime(vcpu);
8871 kvm_mmu_reset_context(vcpu);
8874 static void process_smi(struct kvm_vcpu *vcpu)
8876 vcpu->arch.smi_pending = true;
8877 kvm_make_request(KVM_REQ_EVENT, vcpu);
8880 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8881 unsigned long *vcpu_bitmap)
8885 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8887 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8888 NULL, vcpu_bitmap, cpus);
8890 free_cpumask_var(cpus);
8893 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8895 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8898 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8900 if (!lapic_in_kernel(vcpu))
8903 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8904 kvm_apic_update_apicv(vcpu);
8905 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
8907 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8910 * NOTE: Do not hold any lock prior to calling this.
8912 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8913 * locked, because it calls __x86_set_memory_region() which does
8914 * synchronize_srcu(&kvm->srcu).
8916 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8918 struct kvm_vcpu *except;
8919 unsigned long old, new, expected;
8921 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8922 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
8925 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8927 expected = new = old;
8929 __clear_bit(bit, &new);
8931 __set_bit(bit, &new);
8934 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8935 } while (old != expected);
8940 trace_kvm_apicv_update_request(activate, bit);
8941 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8942 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
8945 * Sending request to update APICV for all other vcpus,
8946 * while update the calling vcpu immediately instead of
8947 * waiting for another #VMEXIT to handle the request.
8949 except = kvm_get_running_vcpu();
8950 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8953 kvm_vcpu_update_apicv(except);
8955 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8957 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8959 if (!kvm_apic_present(vcpu))
8962 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8964 if (irqchip_split(vcpu->kvm))
8965 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8967 if (vcpu->arch.apicv_active)
8968 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
8969 if (ioapic_in_kernel(vcpu->kvm))
8970 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8973 if (is_guest_mode(vcpu))
8974 vcpu->arch.load_eoi_exitmap_pending = true;
8976 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8979 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8981 u64 eoi_exit_bitmap[4];
8983 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8986 if (to_hv_vcpu(vcpu))
8987 bitmap_or((ulong *)eoi_exit_bitmap,
8988 vcpu->arch.ioapic_handled_vectors,
8989 to_hv_synic(vcpu)->vec_bitmap, 256);
8991 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
8994 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8995 unsigned long start, unsigned long end)
8997 unsigned long apic_address;
9000 * The physical address of apic access page is stored in the VMCS.
9001 * Update it when it becomes invalid.
9003 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9004 if (start <= apic_address && apic_address < end)
9005 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9008 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9010 if (!lapic_in_kernel(vcpu))
9013 if (!kvm_x86_ops.set_apic_access_page_addr)
9016 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9019 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9021 smp_send_reschedule(vcpu->cpu);
9023 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9026 * Returns 1 to let vcpu_run() continue the guest execution loop without
9027 * exiting to the userspace. Otherwise, the value will be returned to the
9030 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9034 dm_request_for_irq_injection(vcpu) &&
9035 kvm_cpu_accept_dm_intr(vcpu);
9036 fastpath_t exit_fastpath;
9038 bool req_immediate_exit = false;
9040 /* Forbid vmenter if vcpu dirty ring is soft-full */
9041 if (unlikely(vcpu->kvm->dirty_ring_size &&
9042 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9043 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9044 trace_kvm_dirty_ring_exit(vcpu);
9049 if (kvm_request_pending(vcpu)) {
9050 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9051 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9056 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9057 kvm_mmu_unload(vcpu);
9058 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9059 __kvm_migrate_timers(vcpu);
9060 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9061 kvm_gen_update_masterclock(vcpu->kvm);
9062 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9063 kvm_gen_kvmclock_update(vcpu);
9064 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9065 r = kvm_guest_time_update(vcpu);
9069 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9070 kvm_mmu_sync_roots(vcpu);
9071 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9072 kvm_mmu_load_pgd(vcpu);
9073 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9074 kvm_vcpu_flush_tlb_all(vcpu);
9076 /* Flushing all ASIDs flushes the current ASID... */
9077 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9079 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9080 kvm_vcpu_flush_tlb_current(vcpu);
9081 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
9082 kvm_vcpu_flush_tlb_guest(vcpu);
9084 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9085 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9089 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9090 if (is_guest_mode(vcpu)) {
9091 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9093 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9094 vcpu->mmio_needed = 0;
9099 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9100 /* Page is swapped out. Do synthetic halt */
9101 vcpu->arch.apf.halted = true;
9105 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9106 record_steal_time(vcpu);
9107 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9109 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9111 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9112 kvm_pmu_handle_event(vcpu);
9113 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9114 kvm_pmu_deliver_pmi(vcpu);
9115 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9116 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9117 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9118 vcpu->arch.ioapic_handled_vectors)) {
9119 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9120 vcpu->run->eoi.vector =
9121 vcpu->arch.pending_ioapic_eoi;
9126 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9127 vcpu_scan_ioapic(vcpu);
9128 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9129 vcpu_load_eoi_exitmap(vcpu);
9130 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9131 kvm_vcpu_reload_apic_access_page(vcpu);
9132 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9133 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9134 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9138 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9139 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9140 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9144 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9145 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9147 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9148 vcpu->run->hyperv = hv_vcpu->exit;
9154 * KVM_REQ_HV_STIMER has to be processed after
9155 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9156 * depend on the guest clock being up-to-date
9158 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9159 kvm_hv_process_stimers(vcpu);
9160 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9161 kvm_vcpu_update_apicv(vcpu);
9162 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9163 kvm_check_async_pf_completion(vcpu);
9164 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9165 static_call(kvm_x86_msr_filter_changed)(vcpu);
9167 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9168 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9171 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9172 kvm_xen_has_interrupt(vcpu)) {
9173 ++vcpu->stat.req_event;
9174 kvm_apic_accept_events(vcpu);
9175 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9180 inject_pending_event(vcpu, &req_immediate_exit);
9182 static_call(kvm_x86_enable_irq_window)(vcpu);
9184 if (kvm_lapic_enabled(vcpu)) {
9185 update_cr8_intercept(vcpu);
9186 kvm_lapic_sync_to_vapic(vcpu);
9190 r = kvm_mmu_reload(vcpu);
9192 goto cancel_injection;
9197 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9200 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9201 * IPI are then delayed after guest entry, which ensures that they
9202 * result in virtual interrupt delivery.
9204 local_irq_disable();
9205 vcpu->mode = IN_GUEST_MODE;
9207 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9210 * 1) We should set ->mode before checking ->requests. Please see
9211 * the comment in kvm_vcpu_exiting_guest_mode().
9213 * 2) For APICv, we should set ->mode before checking PID.ON. This
9214 * pairs with the memory barrier implicit in pi_test_and_set_on
9215 * (see vmx_deliver_posted_interrupt).
9217 * 3) This also orders the write to mode from any reads to the page
9218 * tables done while the VCPU is running. Please see the comment
9219 * in kvm_flush_remote_tlbs.
9221 smp_mb__after_srcu_read_unlock();
9224 * This handles the case where a posted interrupt was
9225 * notified with kvm_vcpu_kick.
9227 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9228 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9230 if (kvm_vcpu_exit_request(vcpu)) {
9231 vcpu->mode = OUTSIDE_GUEST_MODE;
9235 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9237 goto cancel_injection;
9240 if (req_immediate_exit) {
9241 kvm_make_request(KVM_REQ_EVENT, vcpu);
9242 static_call(kvm_x86_request_immediate_exit)(vcpu);
9245 fpregs_assert_state_consistent();
9246 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9247 switch_fpu_return();
9249 if (unlikely(vcpu->arch.switch_db_regs)) {
9251 set_debugreg(vcpu->arch.eff_db[0], 0);
9252 set_debugreg(vcpu->arch.eff_db[1], 1);
9253 set_debugreg(vcpu->arch.eff_db[2], 2);
9254 set_debugreg(vcpu->arch.eff_db[3], 3);
9255 set_debugreg(vcpu->arch.dr6, 6);
9256 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9260 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9261 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9264 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9265 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9269 if (vcpu->arch.apicv_active)
9270 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9274 * Do this here before restoring debug registers on the host. And
9275 * since we do this before handling the vmexit, a DR access vmexit
9276 * can (a) read the correct value of the debug registers, (b) set
9277 * KVM_DEBUGREG_WONT_EXIT again.
9279 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9280 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9281 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9282 kvm_update_dr0123(vcpu);
9283 kvm_update_dr7(vcpu);
9284 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9288 * If the guest has used debug registers, at least dr7
9289 * will be disabled while returning to the host.
9290 * If we don't have active breakpoints in the host, we don't
9291 * care about the messed up debug address registers. But if
9292 * we have some of them active, restore the old state.
9294 if (hw_breakpoint_active())
9295 hw_breakpoint_restore();
9297 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9298 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9300 vcpu->mode = OUTSIDE_GUEST_MODE;
9303 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9306 * Consume any pending interrupts, including the possible source of
9307 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9308 * An instruction is required after local_irq_enable() to fully unblock
9309 * interrupts on processors that implement an interrupt shadow, the
9310 * stat.exits increment will do nicely.
9312 kvm_before_interrupt(vcpu);
9315 local_irq_disable();
9316 kvm_after_interrupt(vcpu);
9318 if (lapic_in_kernel(vcpu)) {
9319 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9320 if (delta != S64_MIN) {
9321 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9322 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9329 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9332 * Profile KVM exit RIPs:
9334 if (unlikely(prof_on == KVM_PROFILING)) {
9335 unsigned long rip = kvm_rip_read(vcpu);
9336 profile_hit(KVM_PROFILING, (void *)rip);
9339 if (unlikely(vcpu->arch.tsc_always_catchup))
9340 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9342 if (vcpu->arch.apic_attention)
9343 kvm_lapic_sync_from_vapic(vcpu);
9345 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9349 if (req_immediate_exit)
9350 kvm_make_request(KVM_REQ_EVENT, vcpu);
9351 static_call(kvm_x86_cancel_injection)(vcpu);
9352 if (unlikely(vcpu->arch.apic_attention))
9353 kvm_lapic_sync_from_vapic(vcpu);
9358 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9360 if (!kvm_arch_vcpu_runnable(vcpu) &&
9361 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9362 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9363 kvm_vcpu_block(vcpu);
9364 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9366 if (kvm_x86_ops.post_block)
9367 static_call(kvm_x86_post_block)(vcpu);
9369 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9373 kvm_apic_accept_events(vcpu);
9374 switch(vcpu->arch.mp_state) {
9375 case KVM_MP_STATE_HALTED:
9376 case KVM_MP_STATE_AP_RESET_HOLD:
9377 vcpu->arch.pv.pv_unhalted = false;
9378 vcpu->arch.mp_state =
9379 KVM_MP_STATE_RUNNABLE;
9381 case KVM_MP_STATE_RUNNABLE:
9382 vcpu->arch.apf.halted = false;
9384 case KVM_MP_STATE_INIT_RECEIVED:
9392 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9394 if (is_guest_mode(vcpu))
9395 kvm_check_nested_events(vcpu);
9397 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9398 !vcpu->arch.apf.halted);
9401 static int vcpu_run(struct kvm_vcpu *vcpu)
9404 struct kvm *kvm = vcpu->kvm;
9406 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9407 vcpu->arch.l1tf_flush_l1d = true;
9410 if (kvm_vcpu_running(vcpu)) {
9411 r = vcpu_enter_guest(vcpu);
9413 r = vcpu_block(kvm, vcpu);
9419 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9420 if (kvm_cpu_has_pending_timer(vcpu))
9421 kvm_inject_pending_timer_irqs(vcpu);
9423 if (dm_request_for_irq_injection(vcpu) &&
9424 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9426 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9427 ++vcpu->stat.request_irq_exits;
9431 if (__xfer_to_guest_mode_work_pending()) {
9432 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9433 r = xfer_to_guest_mode_handle_work(vcpu);
9436 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9440 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9445 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9449 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9450 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9451 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9455 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9457 BUG_ON(!vcpu->arch.pio.count);
9459 return complete_emulated_io(vcpu);
9463 * Implements the following, as a state machine:
9467 * for each mmio piece in the fragment
9475 * for each mmio piece in the fragment
9480 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9482 struct kvm_run *run = vcpu->run;
9483 struct kvm_mmio_fragment *frag;
9486 BUG_ON(!vcpu->mmio_needed);
9488 /* Complete previous fragment */
9489 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9490 len = min(8u, frag->len);
9491 if (!vcpu->mmio_is_write)
9492 memcpy(frag->data, run->mmio.data, len);
9494 if (frag->len <= 8) {
9495 /* Switch to the next fragment. */
9497 vcpu->mmio_cur_fragment++;
9499 /* Go forward to the next mmio piece. */
9505 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9506 vcpu->mmio_needed = 0;
9508 /* FIXME: return into emulator if single-stepping. */
9509 if (vcpu->mmio_is_write)
9511 vcpu->mmio_read_completed = 1;
9512 return complete_emulated_io(vcpu);
9515 run->exit_reason = KVM_EXIT_MMIO;
9516 run->mmio.phys_addr = frag->gpa;
9517 if (vcpu->mmio_is_write)
9518 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9519 run->mmio.len = min(8u, frag->len);
9520 run->mmio.is_write = vcpu->mmio_is_write;
9521 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9525 static void kvm_save_current_fpu(struct fpu *fpu)
9528 * If the target FPU state is not resident in the CPU registers, just
9529 * memcpy() from current, else save CPU state directly to the target.
9531 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9532 memcpy(&fpu->state, ¤t->thread.fpu.state,
9533 fpu_kernel_xstate_size);
9535 copy_fpregs_to_fpstate(fpu);
9538 /* Swap (qemu) user FPU context for the guest FPU context. */
9539 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9543 kvm_save_current_fpu(vcpu->arch.user_fpu);
9546 * Guests with protected state can't have it set by the hypervisor,
9547 * so skip trying to set it.
9549 if (vcpu->arch.guest_fpu)
9550 /* PKRU is separately restored in kvm_x86_ops.run. */
9551 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9552 ~XFEATURE_MASK_PKRU);
9554 fpregs_mark_activate();
9560 /* When vcpu_run ends, restore user space FPU context. */
9561 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9566 * Guests with protected state can't have it read by the hypervisor,
9567 * so skip trying to save it.
9569 if (vcpu->arch.guest_fpu)
9570 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9572 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9574 fpregs_mark_activate();
9577 ++vcpu->stat.fpu_reload;
9581 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9583 struct kvm_run *kvm_run = vcpu->run;
9587 kvm_sigset_activate(vcpu);
9589 kvm_load_guest_fpu(vcpu);
9591 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9592 if (kvm_run->immediate_exit) {
9596 kvm_vcpu_block(vcpu);
9597 kvm_apic_accept_events(vcpu);
9598 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9600 if (signal_pending(current)) {
9602 kvm_run->exit_reason = KVM_EXIT_INTR;
9603 ++vcpu->stat.signal_exits;
9608 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9613 if (kvm_run->kvm_dirty_regs) {
9614 r = sync_regs(vcpu);
9619 /* re-sync apic's tpr */
9620 if (!lapic_in_kernel(vcpu)) {
9621 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9627 if (unlikely(vcpu->arch.complete_userspace_io)) {
9628 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9629 vcpu->arch.complete_userspace_io = NULL;
9634 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9636 if (kvm_run->immediate_exit)
9642 kvm_put_guest_fpu(vcpu);
9643 if (kvm_run->kvm_valid_regs)
9645 post_kvm_run_save(vcpu);
9646 kvm_sigset_deactivate(vcpu);
9652 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9654 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9656 * We are here if userspace calls get_regs() in the middle of
9657 * instruction emulation. Registers state needs to be copied
9658 * back from emulation context to vcpu. Userspace shouldn't do
9659 * that usually, but some bad designed PV devices (vmware
9660 * backdoor interface) need this to work
9662 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9663 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9665 regs->rax = kvm_rax_read(vcpu);
9666 regs->rbx = kvm_rbx_read(vcpu);
9667 regs->rcx = kvm_rcx_read(vcpu);
9668 regs->rdx = kvm_rdx_read(vcpu);
9669 regs->rsi = kvm_rsi_read(vcpu);
9670 regs->rdi = kvm_rdi_read(vcpu);
9671 regs->rsp = kvm_rsp_read(vcpu);
9672 regs->rbp = kvm_rbp_read(vcpu);
9673 #ifdef CONFIG_X86_64
9674 regs->r8 = kvm_r8_read(vcpu);
9675 regs->r9 = kvm_r9_read(vcpu);
9676 regs->r10 = kvm_r10_read(vcpu);
9677 regs->r11 = kvm_r11_read(vcpu);
9678 regs->r12 = kvm_r12_read(vcpu);
9679 regs->r13 = kvm_r13_read(vcpu);
9680 regs->r14 = kvm_r14_read(vcpu);
9681 regs->r15 = kvm_r15_read(vcpu);
9684 regs->rip = kvm_rip_read(vcpu);
9685 regs->rflags = kvm_get_rflags(vcpu);
9688 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9691 __get_regs(vcpu, regs);
9696 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9698 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9699 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9701 kvm_rax_write(vcpu, regs->rax);
9702 kvm_rbx_write(vcpu, regs->rbx);
9703 kvm_rcx_write(vcpu, regs->rcx);
9704 kvm_rdx_write(vcpu, regs->rdx);
9705 kvm_rsi_write(vcpu, regs->rsi);
9706 kvm_rdi_write(vcpu, regs->rdi);
9707 kvm_rsp_write(vcpu, regs->rsp);
9708 kvm_rbp_write(vcpu, regs->rbp);
9709 #ifdef CONFIG_X86_64
9710 kvm_r8_write(vcpu, regs->r8);
9711 kvm_r9_write(vcpu, regs->r9);
9712 kvm_r10_write(vcpu, regs->r10);
9713 kvm_r11_write(vcpu, regs->r11);
9714 kvm_r12_write(vcpu, regs->r12);
9715 kvm_r13_write(vcpu, regs->r13);
9716 kvm_r14_write(vcpu, regs->r14);
9717 kvm_r15_write(vcpu, regs->r15);
9720 kvm_rip_write(vcpu, regs->rip);
9721 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9723 vcpu->arch.exception.pending = false;
9725 kvm_make_request(KVM_REQ_EVENT, vcpu);
9728 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9731 __set_regs(vcpu, regs);
9736 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9738 struct kvm_segment cs;
9740 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9744 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9746 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9750 if (vcpu->arch.guest_state_protected)
9751 goto skip_protected_regs;
9753 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9754 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9755 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9756 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9757 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9758 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9760 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9761 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9763 static_call(kvm_x86_get_idt)(vcpu, &dt);
9764 sregs->idt.limit = dt.size;
9765 sregs->idt.base = dt.address;
9766 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9767 sregs->gdt.limit = dt.size;
9768 sregs->gdt.base = dt.address;
9770 sregs->cr2 = vcpu->arch.cr2;
9771 sregs->cr3 = kvm_read_cr3(vcpu);
9773 skip_protected_regs:
9774 sregs->cr0 = kvm_read_cr0(vcpu);
9775 sregs->cr4 = kvm_read_cr4(vcpu);
9776 sregs->cr8 = kvm_get_cr8(vcpu);
9777 sregs->efer = vcpu->arch.efer;
9778 sregs->apic_base = kvm_get_apic_base(vcpu);
9780 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9782 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9783 set_bit(vcpu->arch.interrupt.nr,
9784 (unsigned long *)sregs->interrupt_bitmap);
9787 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9788 struct kvm_sregs *sregs)
9791 __get_sregs(vcpu, sregs);
9796 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9797 struct kvm_mp_state *mp_state)
9800 if (kvm_mpx_supported())
9801 kvm_load_guest_fpu(vcpu);
9803 kvm_apic_accept_events(vcpu);
9804 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9805 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9806 vcpu->arch.pv.pv_unhalted)
9807 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9809 mp_state->mp_state = vcpu->arch.mp_state;
9811 if (kvm_mpx_supported())
9812 kvm_put_guest_fpu(vcpu);
9817 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9818 struct kvm_mp_state *mp_state)
9824 if (!lapic_in_kernel(vcpu) &&
9825 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9829 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9830 * INIT state; latched init should be reported using
9831 * KVM_SET_VCPU_EVENTS, so reject it here.
9833 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9834 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9835 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9838 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9839 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9840 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9842 vcpu->arch.mp_state = mp_state->mp_state;
9843 kvm_make_request(KVM_REQ_EVENT, vcpu);
9851 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9852 int reason, bool has_error_code, u32 error_code)
9854 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9857 init_emulate_ctxt(vcpu);
9859 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9860 has_error_code, error_code);
9862 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9863 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9864 vcpu->run->internal.ndata = 0;
9868 kvm_rip_write(vcpu, ctxt->eip);
9869 kvm_set_rflags(vcpu, ctxt->eflags);
9872 EXPORT_SYMBOL_GPL(kvm_task_switch);
9874 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9876 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9878 * When EFER.LME and CR0.PG are set, the processor is in
9879 * 64-bit mode (though maybe in a 32-bit code segment).
9880 * CR4.PAE and EFER.LMA must be set.
9882 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9884 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
9888 * Not in 64-bit mode: EFER.LMA is clear and the code
9889 * segment cannot be 64-bit.
9891 if (sregs->efer & EFER_LMA || sregs->cs.l)
9895 return kvm_is_valid_cr4(vcpu, sregs->cr4);
9898 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9900 struct msr_data apic_base_msr;
9901 int mmu_reset_needed = 0;
9902 int pending_vec, max_bits, idx;
9906 if (!kvm_is_valid_sregs(vcpu, sregs))
9909 apic_base_msr.data = sregs->apic_base;
9910 apic_base_msr.host_initiated = true;
9911 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9914 if (vcpu->arch.guest_state_protected)
9915 goto skip_protected_regs;
9917 dt.size = sregs->idt.limit;
9918 dt.address = sregs->idt.base;
9919 static_call(kvm_x86_set_idt)(vcpu, &dt);
9920 dt.size = sregs->gdt.limit;
9921 dt.address = sregs->gdt.base;
9922 static_call(kvm_x86_set_gdt)(vcpu, &dt);
9924 vcpu->arch.cr2 = sregs->cr2;
9925 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9926 vcpu->arch.cr3 = sregs->cr3;
9927 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9929 kvm_set_cr8(vcpu, sregs->cr8);
9931 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9932 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
9934 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9935 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
9936 vcpu->arch.cr0 = sregs->cr0;
9938 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9939 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
9941 idx = srcu_read_lock(&vcpu->kvm->srcu);
9942 if (is_pae_paging(vcpu)) {
9943 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9944 mmu_reset_needed = 1;
9946 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9948 if (mmu_reset_needed)
9949 kvm_mmu_reset_context(vcpu);
9951 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9952 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9953 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9954 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9955 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9956 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9958 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9959 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9961 update_cr8_intercept(vcpu);
9963 /* Older userspace won't unhalt the vcpu on reset. */
9964 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9965 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9967 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9969 skip_protected_regs:
9970 max_bits = KVM_NR_INTERRUPTS;
9971 pending_vec = find_first_bit(
9972 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9973 if (pending_vec < max_bits) {
9974 kvm_queue_interrupt(vcpu, pending_vec, false);
9975 pr_debug("Set back pending irq %d\n", pending_vec);
9978 kvm_make_request(KVM_REQ_EVENT, vcpu);
9985 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9986 struct kvm_sregs *sregs)
9991 ret = __set_sregs(vcpu, sregs);
9996 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9997 struct kvm_guest_debug *dbg)
9999 unsigned long rflags;
10002 if (vcpu->arch.guest_state_protected)
10007 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10009 if (vcpu->arch.exception.pending)
10011 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10012 kvm_queue_exception(vcpu, DB_VECTOR);
10014 kvm_queue_exception(vcpu, BP_VECTOR);
10018 * Read rflags as long as potentially injected trace flags are still
10021 rflags = kvm_get_rflags(vcpu);
10023 vcpu->guest_debug = dbg->control;
10024 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10025 vcpu->guest_debug = 0;
10027 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10028 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10029 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10030 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10032 for (i = 0; i < KVM_NR_DB_REGS; i++)
10033 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10035 kvm_update_dr7(vcpu);
10037 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10038 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
10039 get_segment_base(vcpu, VCPU_SREG_CS);
10042 * Trigger an rflags update that will inject or remove the trace
10045 kvm_set_rflags(vcpu, rflags);
10047 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10057 * Translate a guest virtual address to a guest physical address.
10059 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10060 struct kvm_translation *tr)
10062 unsigned long vaddr = tr->linear_address;
10068 idx = srcu_read_lock(&vcpu->kvm->srcu);
10069 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10070 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10071 tr->physical_address = gpa;
10072 tr->valid = gpa != UNMAPPED_GVA;
10080 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10082 struct fxregs_state *fxsave;
10084 if (!vcpu->arch.guest_fpu)
10089 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10090 memcpy(fpu->fpr, fxsave->st_space, 128);
10091 fpu->fcw = fxsave->cwd;
10092 fpu->fsw = fxsave->swd;
10093 fpu->ftwx = fxsave->twd;
10094 fpu->last_opcode = fxsave->fop;
10095 fpu->last_ip = fxsave->rip;
10096 fpu->last_dp = fxsave->rdp;
10097 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10103 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10105 struct fxregs_state *fxsave;
10107 if (!vcpu->arch.guest_fpu)
10112 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10114 memcpy(fxsave->st_space, fpu->fpr, 128);
10115 fxsave->cwd = fpu->fcw;
10116 fxsave->swd = fpu->fsw;
10117 fxsave->twd = fpu->ftwx;
10118 fxsave->fop = fpu->last_opcode;
10119 fxsave->rip = fpu->last_ip;
10120 fxsave->rdp = fpu->last_dp;
10121 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10127 static void store_regs(struct kvm_vcpu *vcpu)
10129 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10131 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10132 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10134 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10135 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10137 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10138 kvm_vcpu_ioctl_x86_get_vcpu_events(
10139 vcpu, &vcpu->run->s.regs.events);
10142 static int sync_regs(struct kvm_vcpu *vcpu)
10144 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10147 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10148 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10149 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10151 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10152 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10154 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10156 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10157 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10158 vcpu, &vcpu->run->s.regs.events))
10160 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10166 static void fx_init(struct kvm_vcpu *vcpu)
10168 if (!vcpu->arch.guest_fpu)
10171 fpstate_init(&vcpu->arch.guest_fpu->state);
10172 if (boot_cpu_has(X86_FEATURE_XSAVES))
10173 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10174 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10177 * Ensure guest xcr0 is valid for loading
10179 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10181 vcpu->arch.cr0 |= X86_CR0_ET;
10184 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10186 if (vcpu->arch.guest_fpu) {
10187 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10188 vcpu->arch.guest_fpu = NULL;
10191 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10193 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10195 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10196 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10197 "guest TSC will not be reliable\n");
10202 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10207 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10208 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10210 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10212 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10214 r = kvm_mmu_create(vcpu);
10218 if (irqchip_in_kernel(vcpu->kvm)) {
10219 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10221 goto fail_mmu_destroy;
10222 if (kvm_apicv_activated(vcpu->kvm))
10223 vcpu->arch.apicv_active = true;
10225 static_branch_inc(&kvm_has_noapic_vcpu);
10229 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10231 goto fail_free_lapic;
10232 vcpu->arch.pio_data = page_address(page);
10234 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10235 GFP_KERNEL_ACCOUNT);
10236 if (!vcpu->arch.mce_banks)
10237 goto fail_free_pio_data;
10238 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10240 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10241 GFP_KERNEL_ACCOUNT))
10242 goto fail_free_mce_banks;
10244 if (!alloc_emulate_ctxt(vcpu))
10245 goto free_wbinvd_dirty_mask;
10247 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10248 GFP_KERNEL_ACCOUNT);
10249 if (!vcpu->arch.user_fpu) {
10250 pr_err("kvm: failed to allocate userspace's fpu\n");
10251 goto free_emulate_ctxt;
10254 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10255 GFP_KERNEL_ACCOUNT);
10256 if (!vcpu->arch.guest_fpu) {
10257 pr_err("kvm: failed to allocate vcpu's fpu\n");
10258 goto free_user_fpu;
10262 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10263 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10265 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10267 kvm_async_pf_hash_reset(vcpu);
10268 kvm_pmu_init(vcpu);
10270 vcpu->arch.pending_external_vector = -1;
10271 vcpu->arch.preempted_in_kernel = false;
10273 r = static_call(kvm_x86_vcpu_create)(vcpu);
10275 goto free_guest_fpu;
10277 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10278 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10279 kvm_vcpu_mtrr_init(vcpu);
10281 kvm_vcpu_reset(vcpu, false);
10282 kvm_init_mmu(vcpu, false);
10287 kvm_free_guest_fpu(vcpu);
10289 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10291 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10292 free_wbinvd_dirty_mask:
10293 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10294 fail_free_mce_banks:
10295 kfree(vcpu->arch.mce_banks);
10296 fail_free_pio_data:
10297 free_page((unsigned long)vcpu->arch.pio_data);
10299 kvm_free_lapic(vcpu);
10301 kvm_mmu_destroy(vcpu);
10305 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10307 struct kvm *kvm = vcpu->kvm;
10309 if (mutex_lock_killable(&vcpu->mutex))
10312 kvm_synchronize_tsc(vcpu, 0);
10315 /* poll control enabled by default */
10316 vcpu->arch.msr_kvm_poll_control = 1;
10318 mutex_unlock(&vcpu->mutex);
10320 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10321 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10322 KVMCLOCK_SYNC_PERIOD);
10325 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10327 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10330 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10332 kvmclock_reset(vcpu);
10334 static_call(kvm_x86_vcpu_free)(vcpu);
10336 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10337 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10338 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10339 kvm_free_guest_fpu(vcpu);
10341 kvm_hv_vcpu_uninit(vcpu);
10342 kvm_pmu_destroy(vcpu);
10343 kfree(vcpu->arch.mce_banks);
10344 kvm_free_lapic(vcpu);
10345 idx = srcu_read_lock(&vcpu->kvm->srcu);
10346 kvm_mmu_destroy(vcpu);
10347 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10348 free_page((unsigned long)vcpu->arch.pio_data);
10349 kvfree(vcpu->arch.cpuid_entries);
10350 if (!lapic_in_kernel(vcpu))
10351 static_branch_dec(&kvm_has_noapic_vcpu);
10354 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10356 kvm_lapic_reset(vcpu, init_event);
10358 vcpu->arch.hflags = 0;
10360 vcpu->arch.smi_pending = 0;
10361 vcpu->arch.smi_count = 0;
10362 atomic_set(&vcpu->arch.nmi_queued, 0);
10363 vcpu->arch.nmi_pending = 0;
10364 vcpu->arch.nmi_injected = false;
10365 kvm_clear_interrupt_queue(vcpu);
10366 kvm_clear_exception_queue(vcpu);
10368 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10369 kvm_update_dr0123(vcpu);
10370 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10371 vcpu->arch.dr7 = DR7_FIXED_1;
10372 kvm_update_dr7(vcpu);
10374 vcpu->arch.cr2 = 0;
10376 kvm_make_request(KVM_REQ_EVENT, vcpu);
10377 vcpu->arch.apf.msr_en_val = 0;
10378 vcpu->arch.apf.msr_int_val = 0;
10379 vcpu->arch.st.msr_val = 0;
10381 kvmclock_reset(vcpu);
10383 kvm_clear_async_pf_completion_queue(vcpu);
10384 kvm_async_pf_hash_reset(vcpu);
10385 vcpu->arch.apf.halted = false;
10387 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10388 void *mpx_state_buffer;
10391 * To avoid have the INIT path from kvm_apic_has_events() that be
10392 * called with loaded FPU and does not let userspace fix the state.
10395 kvm_put_guest_fpu(vcpu);
10396 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10398 if (mpx_state_buffer)
10399 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
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_bndcsr));
10405 kvm_load_guest_fpu(vcpu);
10409 kvm_pmu_reset(vcpu);
10410 vcpu->arch.smbase = 0x30000;
10412 vcpu->arch.msr_misc_features_enables = 0;
10414 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10417 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10418 vcpu->arch.regs_avail = ~0;
10419 vcpu->arch.regs_dirty = ~0;
10421 vcpu->arch.ia32_xss = 0;
10423 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10426 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10428 struct kvm_segment cs;
10430 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10431 cs.selector = vector << 8;
10432 cs.base = vector << 12;
10433 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10434 kvm_rip_write(vcpu, 0);
10436 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10438 int kvm_arch_hardware_enable(void)
10441 struct kvm_vcpu *vcpu;
10446 bool stable, backwards_tsc = false;
10448 kvm_user_return_msr_cpu_online();
10449 ret = static_call(kvm_x86_hardware_enable)();
10453 local_tsc = rdtsc();
10454 stable = !kvm_check_tsc_unstable();
10455 list_for_each_entry(kvm, &vm_list, vm_list) {
10456 kvm_for_each_vcpu(i, vcpu, kvm) {
10457 if (!stable && vcpu->cpu == smp_processor_id())
10458 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10459 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10460 backwards_tsc = true;
10461 if (vcpu->arch.last_host_tsc > max_tsc)
10462 max_tsc = vcpu->arch.last_host_tsc;
10468 * Sometimes, even reliable TSCs go backwards. This happens on
10469 * platforms that reset TSC during suspend or hibernate actions, but
10470 * maintain synchronization. We must compensate. Fortunately, we can
10471 * detect that condition here, which happens early in CPU bringup,
10472 * before any KVM threads can be running. Unfortunately, we can't
10473 * bring the TSCs fully up to date with real time, as we aren't yet far
10474 * enough into CPU bringup that we know how much real time has actually
10475 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10476 * variables that haven't been updated yet.
10478 * So we simply find the maximum observed TSC above, then record the
10479 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10480 * the adjustment will be applied. Note that we accumulate
10481 * adjustments, in case multiple suspend cycles happen before some VCPU
10482 * gets a chance to run again. In the event that no KVM threads get a
10483 * chance to run, we will miss the entire elapsed period, as we'll have
10484 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10485 * loose cycle time. This isn't too big a deal, since the loss will be
10486 * uniform across all VCPUs (not to mention the scenario is extremely
10487 * unlikely). It is possible that a second hibernate recovery happens
10488 * much faster than a first, causing the observed TSC here to be
10489 * smaller; this would require additional padding adjustment, which is
10490 * why we set last_host_tsc to the local tsc observed here.
10492 * N.B. - this code below runs only on platforms with reliable TSC,
10493 * as that is the only way backwards_tsc is set above. Also note
10494 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10495 * have the same delta_cyc adjustment applied if backwards_tsc
10496 * is detected. Note further, this adjustment is only done once,
10497 * as we reset last_host_tsc on all VCPUs to stop this from being
10498 * called multiple times (one for each physical CPU bringup).
10500 * Platforms with unreliable TSCs don't have to deal with this, they
10501 * will be compensated by the logic in vcpu_load, which sets the TSC to
10502 * catchup mode. This will catchup all VCPUs to real time, but cannot
10503 * guarantee that they stay in perfect synchronization.
10505 if (backwards_tsc) {
10506 u64 delta_cyc = max_tsc - local_tsc;
10507 list_for_each_entry(kvm, &vm_list, vm_list) {
10508 kvm->arch.backwards_tsc_observed = true;
10509 kvm_for_each_vcpu(i, vcpu, kvm) {
10510 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10511 vcpu->arch.last_host_tsc = local_tsc;
10512 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10516 * We have to disable TSC offset matching.. if you were
10517 * booting a VM while issuing an S4 host suspend....
10518 * you may have some problem. Solving this issue is
10519 * left as an exercise to the reader.
10521 kvm->arch.last_tsc_nsec = 0;
10522 kvm->arch.last_tsc_write = 0;
10529 void kvm_arch_hardware_disable(void)
10531 static_call(kvm_x86_hardware_disable)();
10532 drop_user_return_notifiers();
10535 int kvm_arch_hardware_setup(void *opaque)
10537 struct kvm_x86_init_ops *ops = opaque;
10540 rdmsrl_safe(MSR_EFER, &host_efer);
10542 if (boot_cpu_has(X86_FEATURE_XSAVES))
10543 rdmsrl(MSR_IA32_XSS, host_xss);
10545 r = ops->hardware_setup();
10549 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10550 kvm_ops_static_call_update();
10552 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10555 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10556 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10557 #undef __kvm_cpu_cap_has
10559 if (kvm_has_tsc_control) {
10561 * Make sure the user can only configure tsc_khz values that
10562 * fit into a signed integer.
10563 * A min value is not calculated because it will always
10564 * be 1 on all machines.
10566 u64 max = min(0x7fffffffULL,
10567 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10568 kvm_max_guest_tsc_khz = max;
10570 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10573 kvm_init_msr_list();
10577 void kvm_arch_hardware_unsetup(void)
10579 static_call(kvm_x86_hardware_unsetup)();
10582 int kvm_arch_check_processor_compat(void *opaque)
10584 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10585 struct kvm_x86_init_ops *ops = opaque;
10587 WARN_ON(!irqs_disabled());
10589 if (__cr4_reserved_bits(cpu_has, c) !=
10590 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10593 return ops->check_processor_compatibility();
10596 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10598 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10600 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10602 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10604 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10607 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10608 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10610 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10612 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10614 vcpu->arch.l1tf_flush_l1d = true;
10615 if (pmu->version && unlikely(pmu->event_count)) {
10616 pmu->need_cleanup = true;
10617 kvm_make_request(KVM_REQ_PMU, vcpu);
10619 static_call(kvm_x86_sched_in)(vcpu, cpu);
10622 void kvm_arch_free_vm(struct kvm *kvm)
10624 kfree(to_kvm_hv(kvm)->hv_pa_pg);
10629 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10634 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10635 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10636 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10637 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10638 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10639 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10641 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10642 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10643 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10644 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10645 &kvm->arch.irq_sources_bitmap);
10647 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10648 mutex_init(&kvm->arch.apic_map_lock);
10649 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10651 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10652 pvclock_update_vm_gtod_copy(kvm);
10654 kvm->arch.guest_can_read_msr_platform_info = true;
10656 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10657 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10659 kvm_hv_init_vm(kvm);
10660 kvm_page_track_init(kvm);
10661 kvm_mmu_init_vm(kvm);
10663 return static_call(kvm_x86_vm_init)(kvm);
10666 int kvm_arch_post_init_vm(struct kvm *kvm)
10668 return kvm_mmu_post_init_vm(kvm);
10671 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10674 kvm_mmu_unload(vcpu);
10678 static void kvm_free_vcpus(struct kvm *kvm)
10681 struct kvm_vcpu *vcpu;
10684 * Unpin any mmu pages first.
10686 kvm_for_each_vcpu(i, vcpu, kvm) {
10687 kvm_clear_async_pf_completion_queue(vcpu);
10688 kvm_unload_vcpu_mmu(vcpu);
10690 kvm_for_each_vcpu(i, vcpu, kvm)
10691 kvm_vcpu_destroy(vcpu);
10693 mutex_lock(&kvm->lock);
10694 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10695 kvm->vcpus[i] = NULL;
10697 atomic_set(&kvm->online_vcpus, 0);
10698 mutex_unlock(&kvm->lock);
10701 void kvm_arch_sync_events(struct kvm *kvm)
10703 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10704 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10708 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
10711 * __x86_set_memory_region: Setup KVM internal memory slot
10713 * @kvm: the kvm pointer to the VM.
10714 * @id: the slot ID to setup.
10715 * @gpa: the GPA to install the slot (unused when @size == 0).
10716 * @size: the size of the slot. Set to zero to uninstall a slot.
10718 * This function helps to setup a KVM internal memory slot. Specify
10719 * @size > 0 to install a new slot, while @size == 0 to uninstall a
10720 * slot. The return code can be one of the following:
10722 * HVA: on success (uninstall will return a bogus HVA)
10725 * The caller should always use IS_ERR() to check the return value
10726 * before use. Note, the KVM internal memory slots are guaranteed to
10727 * remain valid and unchanged until the VM is destroyed, i.e., the
10728 * GPA->HVA translation will not change. However, the HVA is a user
10729 * address, i.e. its accessibility is not guaranteed, and must be
10730 * accessed via __copy_{to,from}_user().
10732 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10736 unsigned long hva, old_npages;
10737 struct kvm_memslots *slots = kvm_memslots(kvm);
10738 struct kvm_memory_slot *slot;
10740 /* Called with kvm->slots_lock held. */
10741 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10742 return ERR_PTR_USR(-EINVAL);
10744 slot = id_to_memslot(slots, id);
10746 if (slot && slot->npages)
10747 return ERR_PTR_USR(-EEXIST);
10750 * MAP_SHARED to prevent internal slot pages from being moved
10753 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10754 MAP_SHARED | MAP_ANONYMOUS, 0);
10755 if (IS_ERR((void *)hva))
10756 return (void __user *)hva;
10758 if (!slot || !slot->npages)
10761 old_npages = slot->npages;
10762 hva = slot->userspace_addr;
10765 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10766 struct kvm_userspace_memory_region m;
10768 m.slot = id | (i << 16);
10770 m.guest_phys_addr = gpa;
10771 m.userspace_addr = hva;
10772 m.memory_size = size;
10773 r = __kvm_set_memory_region(kvm, &m);
10775 return ERR_PTR_USR(r);
10779 vm_munmap(hva, old_npages * PAGE_SIZE);
10781 return (void __user *)hva;
10783 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10785 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10787 kvm_mmu_pre_destroy_vm(kvm);
10790 void kvm_arch_destroy_vm(struct kvm *kvm)
10792 if (current->mm == kvm->mm) {
10794 * Free memory regions allocated on behalf of userspace,
10795 * unless the the memory map has changed due to process exit
10798 mutex_lock(&kvm->slots_lock);
10799 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10801 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10803 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10804 mutex_unlock(&kvm->slots_lock);
10806 static_call_cond(kvm_x86_vm_destroy)(kvm);
10807 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
10808 kvm_pic_destroy(kvm);
10809 kvm_ioapic_destroy(kvm);
10810 kvm_free_vcpus(kvm);
10811 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10812 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10813 kvm_mmu_uninit_vm(kvm);
10814 kvm_page_track_cleanup(kvm);
10815 kvm_xen_destroy_vm(kvm);
10816 kvm_hv_destroy_vm(kvm);
10819 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10823 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10824 kvfree(slot->arch.rmap[i]);
10825 slot->arch.rmap[i] = NULL;
10830 kvfree(slot->arch.lpage_info[i - 1]);
10831 slot->arch.lpage_info[i - 1] = NULL;
10834 kvm_page_track_free_memslot(slot);
10837 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10838 unsigned long npages)
10843 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10844 * old arrays will be freed by __kvm_set_memory_region() if installing
10845 * the new memslot is successful.
10847 memset(&slot->arch, 0, sizeof(slot->arch));
10849 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10850 struct kvm_lpage_info *linfo;
10851 unsigned long ugfn;
10855 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10856 slot->base_gfn, level) + 1;
10858 slot->arch.rmap[i] =
10859 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10860 GFP_KERNEL_ACCOUNT);
10861 if (!slot->arch.rmap[i])
10866 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10870 slot->arch.lpage_info[i - 1] = linfo;
10872 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10873 linfo[0].disallow_lpage = 1;
10874 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10875 linfo[lpages - 1].disallow_lpage = 1;
10876 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10878 * If the gfn and userspace address are not aligned wrt each
10879 * other, disable large page support for this slot.
10881 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10884 for (j = 0; j < lpages; ++j)
10885 linfo[j].disallow_lpage = 1;
10889 if (kvm_page_track_create_memslot(slot, npages))
10895 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10896 kvfree(slot->arch.rmap[i]);
10897 slot->arch.rmap[i] = NULL;
10901 kvfree(slot->arch.lpage_info[i - 1]);
10902 slot->arch.lpage_info[i - 1] = NULL;
10907 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10909 struct kvm_vcpu *vcpu;
10913 * memslots->generation has been incremented.
10914 * mmio generation may have reached its maximum value.
10916 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10918 /* Force re-initialization of steal_time cache */
10919 kvm_for_each_vcpu(i, vcpu, kvm)
10920 kvm_vcpu_kick(vcpu);
10923 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10924 struct kvm_memory_slot *memslot,
10925 const struct kvm_userspace_memory_region *mem,
10926 enum kvm_mr_change change)
10928 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10929 return kvm_alloc_memslot_metadata(memslot,
10930 mem->memory_size >> PAGE_SHIFT);
10935 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
10937 struct kvm_arch *ka = &kvm->arch;
10939 if (!kvm_x86_ops.cpu_dirty_log_size)
10942 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
10943 (!enable && --ka->cpu_dirty_logging_count == 0))
10944 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
10946 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
10949 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10950 struct kvm_memory_slot *old,
10951 struct kvm_memory_slot *new,
10952 enum kvm_mr_change change)
10954 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
10957 * Update CPU dirty logging if dirty logging is being toggled. This
10958 * applies to all operations.
10960 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
10961 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
10964 * Nothing more to do for RO slots (which can't be dirtied and can't be
10965 * made writable) or CREATE/MOVE/DELETE of a slot.
10967 * For a memslot with dirty logging disabled:
10968 * CREATE: No dirty mappings will already exist.
10969 * MOVE/DELETE: The old mappings will already have been cleaned up by
10970 * kvm_arch_flush_shadow_memslot()
10972 * For a memslot with dirty logging enabled:
10973 * CREATE: No shadow pages exist, thus nothing to write-protect
10974 * and no dirty bits to clear.
10975 * MOVE/DELETE: The old mappings will already have been cleaned up by
10976 * kvm_arch_flush_shadow_memslot().
10978 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10982 * READONLY and non-flags changes were filtered out above, and the only
10983 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
10984 * logging isn't being toggled on or off.
10986 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
10989 if (!log_dirty_pages) {
10991 * Dirty logging tracks sptes in 4k granularity, meaning that
10992 * large sptes have to be split. If live migration succeeds,
10993 * the guest in the source machine will be destroyed and large
10994 * sptes will be created in the destination. However, if the
10995 * guest continues to run in the source machine (for example if
10996 * live migration fails), small sptes will remain around and
10997 * cause bad performance.
10999 * Scan sptes if dirty logging has been stopped, dropping those
11000 * which can be collapsed into a single large-page spte. Later
11001 * page faults will create the large-page sptes.
11003 kvm_mmu_zap_collapsible_sptes(kvm, new);
11005 /* By default, write-protect everything to log writes. */
11006 int level = PG_LEVEL_4K;
11008 if (kvm_x86_ops.cpu_dirty_log_size) {
11010 * Clear all dirty bits, unless pages are treated as
11011 * dirty from the get-go.
11013 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
11014 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11017 * Write-protect large pages on write so that dirty
11018 * logging happens at 4k granularity. No need to
11019 * write-protect small SPTEs since write accesses are
11020 * logged by the CPU via dirty bits.
11022 level = PG_LEVEL_2M;
11023 } else if (kvm_dirty_log_manual_protect_and_init_set(kvm)) {
11025 * If we're with initial-all-set, we don't need
11026 * to write protect any small page because
11027 * they're reported as dirty already. However
11028 * we still need to write-protect huge pages
11029 * so that the page split can happen lazily on
11030 * the first write to the huge page.
11032 level = PG_LEVEL_2M;
11034 kvm_mmu_slot_remove_write_access(kvm, new, level);
11038 void kvm_arch_commit_memory_region(struct kvm *kvm,
11039 const struct kvm_userspace_memory_region *mem,
11040 struct kvm_memory_slot *old,
11041 const struct kvm_memory_slot *new,
11042 enum kvm_mr_change change)
11044 if (!kvm->arch.n_requested_mmu_pages)
11045 kvm_mmu_change_mmu_pages(kvm,
11046 kvm_mmu_calculate_default_mmu_pages(kvm));
11049 * FIXME: const-ify all uses of struct kvm_memory_slot.
11051 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11053 /* Free the arrays associated with the old memslot. */
11054 if (change == KVM_MR_MOVE)
11055 kvm_arch_free_memslot(kvm, old);
11058 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11060 kvm_mmu_zap_all(kvm);
11063 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11064 struct kvm_memory_slot *slot)
11066 kvm_page_track_flush_slot(kvm, slot);
11069 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11071 return (is_guest_mode(vcpu) &&
11072 kvm_x86_ops.guest_apic_has_interrupt &&
11073 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11076 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11078 if (!list_empty_careful(&vcpu->async_pf.done))
11081 if (kvm_apic_has_events(vcpu))
11084 if (vcpu->arch.pv.pv_unhalted)
11087 if (vcpu->arch.exception.pending)
11090 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11091 (vcpu->arch.nmi_pending &&
11092 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11095 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11096 (vcpu->arch.smi_pending &&
11097 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11100 if (kvm_arch_interrupt_allowed(vcpu) &&
11101 (kvm_cpu_has_interrupt(vcpu) ||
11102 kvm_guest_apic_has_interrupt(vcpu)))
11105 if (kvm_hv_has_stimer_pending(vcpu))
11108 if (is_guest_mode(vcpu) &&
11109 kvm_x86_ops.nested_ops->hv_timer_pending &&
11110 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11116 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11118 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11121 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11123 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11129 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11131 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11134 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11135 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11136 kvm_test_request(KVM_REQ_EVENT, vcpu))
11139 return kvm_arch_dy_has_pending_interrupt(vcpu);
11142 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11144 if (vcpu->arch.guest_state_protected)
11147 return vcpu->arch.preempted_in_kernel;
11150 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11152 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11155 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11157 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11160 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11162 /* Can't read the RIP when guest state is protected, just return 0 */
11163 if (vcpu->arch.guest_state_protected)
11166 if (is_64_bit_mode(vcpu))
11167 return kvm_rip_read(vcpu);
11168 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11169 kvm_rip_read(vcpu));
11171 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11173 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11175 return kvm_get_linear_rip(vcpu) == linear_rip;
11177 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11179 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11181 unsigned long rflags;
11183 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11184 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11185 rflags &= ~X86_EFLAGS_TF;
11188 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11190 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11192 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11193 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11194 rflags |= X86_EFLAGS_TF;
11195 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11198 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11200 __kvm_set_rflags(vcpu, rflags);
11201 kvm_make_request(KVM_REQ_EVENT, vcpu);
11203 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11205 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11209 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11213 r = kvm_mmu_reload(vcpu);
11217 if (!vcpu->arch.mmu->direct_map &&
11218 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11221 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11224 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11226 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11228 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11231 static inline u32 kvm_async_pf_next_probe(u32 key)
11233 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11236 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11238 u32 key = kvm_async_pf_hash_fn(gfn);
11240 while (vcpu->arch.apf.gfns[key] != ~0)
11241 key = kvm_async_pf_next_probe(key);
11243 vcpu->arch.apf.gfns[key] = gfn;
11246 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11249 u32 key = kvm_async_pf_hash_fn(gfn);
11251 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11252 (vcpu->arch.apf.gfns[key] != gfn &&
11253 vcpu->arch.apf.gfns[key] != ~0); i++)
11254 key = kvm_async_pf_next_probe(key);
11259 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11261 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11264 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11268 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11270 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11274 vcpu->arch.apf.gfns[i] = ~0;
11276 j = kvm_async_pf_next_probe(j);
11277 if (vcpu->arch.apf.gfns[j] == ~0)
11279 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11281 * k lies cyclically in ]i,j]
11283 * |....j i.k.| or |.k..j i...|
11285 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11286 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11291 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11293 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11295 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11299 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11301 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11303 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11304 &token, offset, sizeof(token));
11307 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11309 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11312 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11313 &val, offset, sizeof(val)))
11319 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11321 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11324 if (!kvm_pv_async_pf_enabled(vcpu) ||
11325 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11331 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11333 if (unlikely(!lapic_in_kernel(vcpu) ||
11334 kvm_event_needs_reinjection(vcpu) ||
11335 vcpu->arch.exception.pending))
11338 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11342 * If interrupts are off we cannot even use an artificial
11345 return kvm_arch_interrupt_allowed(vcpu);
11348 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11349 struct kvm_async_pf *work)
11351 struct x86_exception fault;
11353 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11354 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11356 if (kvm_can_deliver_async_pf(vcpu) &&
11357 !apf_put_user_notpresent(vcpu)) {
11358 fault.vector = PF_VECTOR;
11359 fault.error_code_valid = true;
11360 fault.error_code = 0;
11361 fault.nested_page_fault = false;
11362 fault.address = work->arch.token;
11363 fault.async_page_fault = true;
11364 kvm_inject_page_fault(vcpu, &fault);
11368 * It is not possible to deliver a paravirtualized asynchronous
11369 * page fault, but putting the guest in an artificial halt state
11370 * can be beneficial nevertheless: if an interrupt arrives, we
11371 * can deliver it timely and perhaps the guest will schedule
11372 * another process. When the instruction that triggered a page
11373 * fault is retried, hopefully the page will be ready in the host.
11375 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11380 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11381 struct kvm_async_pf *work)
11383 struct kvm_lapic_irq irq = {
11384 .delivery_mode = APIC_DM_FIXED,
11385 .vector = vcpu->arch.apf.vec
11388 if (work->wakeup_all)
11389 work->arch.token = ~0; /* broadcast wakeup */
11391 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11392 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11394 if ((work->wakeup_all || work->notpresent_injected) &&
11395 kvm_pv_async_pf_enabled(vcpu) &&
11396 !apf_put_user_ready(vcpu, work->arch.token)) {
11397 vcpu->arch.apf.pageready_pending = true;
11398 kvm_apic_set_irq(vcpu, &irq, NULL);
11401 vcpu->arch.apf.halted = false;
11402 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11405 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11407 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11408 if (!vcpu->arch.apf.pageready_pending)
11409 kvm_vcpu_kick(vcpu);
11412 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11414 if (!kvm_pv_async_pf_enabled(vcpu))
11417 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11420 void kvm_arch_start_assignment(struct kvm *kvm)
11422 atomic_inc(&kvm->arch.assigned_device_count);
11424 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11426 void kvm_arch_end_assignment(struct kvm *kvm)
11428 atomic_dec(&kvm->arch.assigned_device_count);
11430 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11432 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11434 return atomic_read(&kvm->arch.assigned_device_count);
11436 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11438 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11440 atomic_inc(&kvm->arch.noncoherent_dma_count);
11442 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11444 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11446 atomic_dec(&kvm->arch.noncoherent_dma_count);
11448 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11450 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11452 return atomic_read(&kvm->arch.noncoherent_dma_count);
11454 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11456 bool kvm_arch_has_irq_bypass(void)
11461 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11462 struct irq_bypass_producer *prod)
11464 struct kvm_kernel_irqfd *irqfd =
11465 container_of(cons, struct kvm_kernel_irqfd, consumer);
11468 irqfd->producer = prod;
11469 kvm_arch_start_assignment(irqfd->kvm);
11470 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11471 prod->irq, irqfd->gsi, 1);
11474 kvm_arch_end_assignment(irqfd->kvm);
11479 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11480 struct irq_bypass_producer *prod)
11483 struct kvm_kernel_irqfd *irqfd =
11484 container_of(cons, struct kvm_kernel_irqfd, consumer);
11486 WARN_ON(irqfd->producer != prod);
11487 irqfd->producer = NULL;
11490 * When producer of consumer is unregistered, we change back to
11491 * remapped mode, so we can re-use the current implementation
11492 * when the irq is masked/disabled or the consumer side (KVM
11493 * int this case doesn't want to receive the interrupts.
11495 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11497 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11498 " fails: %d\n", irqfd->consumer.token, ret);
11500 kvm_arch_end_assignment(irqfd->kvm);
11503 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11504 uint32_t guest_irq, bool set)
11506 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11509 bool kvm_vector_hashing_enabled(void)
11511 return vector_hashing;
11514 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11516 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11518 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11521 int kvm_spec_ctrl_test_value(u64 value)
11524 * test that setting IA32_SPEC_CTRL to given value
11525 * is allowed by the host processor
11529 unsigned long flags;
11532 local_irq_save(flags);
11534 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11536 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11539 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11541 local_irq_restore(flags);
11545 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11547 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11549 struct x86_exception fault;
11550 u32 access = error_code &
11551 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11553 if (!(error_code & PFERR_PRESENT_MASK) ||
11554 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11556 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11557 * tables probably do not match the TLB. Just proceed
11558 * with the error code that the processor gave.
11560 fault.vector = PF_VECTOR;
11561 fault.error_code_valid = true;
11562 fault.error_code = error_code;
11563 fault.nested_page_fault = false;
11564 fault.address = gva;
11566 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11568 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11571 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11572 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11573 * indicates whether exit to userspace is needed.
11575 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11576 struct x86_exception *e)
11578 if (r == X86EMUL_PROPAGATE_FAULT) {
11579 kvm_inject_emulated_page_fault(vcpu, e);
11584 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11585 * while handling a VMX instruction KVM could've handled the request
11586 * correctly by exiting to userspace and performing I/O but there
11587 * doesn't seem to be a real use-case behind such requests, just return
11588 * KVM_EXIT_INTERNAL_ERROR for now.
11590 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11591 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11592 vcpu->run->internal.ndata = 0;
11596 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11598 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11601 struct x86_exception e;
11603 unsigned long roots_to_free = 0;
11610 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11611 if (r != X86EMUL_CONTINUE)
11612 return kvm_handle_memory_failure(vcpu, r, &e);
11614 if (operand.pcid >> 12 != 0) {
11615 kvm_inject_gp(vcpu, 0);
11619 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11622 case INVPCID_TYPE_INDIV_ADDR:
11623 if ((!pcid_enabled && (operand.pcid != 0)) ||
11624 is_noncanonical_address(operand.gla, vcpu)) {
11625 kvm_inject_gp(vcpu, 0);
11628 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11629 return kvm_skip_emulated_instruction(vcpu);
11631 case INVPCID_TYPE_SINGLE_CTXT:
11632 if (!pcid_enabled && (operand.pcid != 0)) {
11633 kvm_inject_gp(vcpu, 0);
11637 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11638 kvm_mmu_sync_roots(vcpu);
11639 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11642 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11643 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11645 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11647 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11649 * If neither the current cr3 nor any of the prev_roots use the
11650 * given PCID, then nothing needs to be done here because a
11651 * resync will happen anyway before switching to any other CR3.
11654 return kvm_skip_emulated_instruction(vcpu);
11656 case INVPCID_TYPE_ALL_NON_GLOBAL:
11658 * Currently, KVM doesn't mark global entries in the shadow
11659 * page tables, so a non-global flush just degenerates to a
11660 * global flush. If needed, we could optimize this later by
11661 * keeping track of global entries in shadow page tables.
11665 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11666 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
11667 return kvm_skip_emulated_instruction(vcpu);
11670 BUG(); /* We have already checked above that type <= 3 */
11673 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11675 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11677 struct kvm_run *run = vcpu->run;
11678 struct kvm_mmio_fragment *frag;
11681 BUG_ON(!vcpu->mmio_needed);
11683 /* Complete previous fragment */
11684 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11685 len = min(8u, frag->len);
11686 if (!vcpu->mmio_is_write)
11687 memcpy(frag->data, run->mmio.data, len);
11689 if (frag->len <= 8) {
11690 /* Switch to the next fragment. */
11692 vcpu->mmio_cur_fragment++;
11694 /* Go forward to the next mmio piece. */
11700 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11701 vcpu->mmio_needed = 0;
11703 // VMG change, at this point, we're always done
11704 // RIP has already been advanced
11708 // More MMIO is needed
11709 run->mmio.phys_addr = frag->gpa;
11710 run->mmio.len = min(8u, frag->len);
11711 run->mmio.is_write = vcpu->mmio_is_write;
11712 if (run->mmio.is_write)
11713 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11714 run->exit_reason = KVM_EXIT_MMIO;
11716 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11721 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11725 struct kvm_mmio_fragment *frag;
11730 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11731 if (handled == bytes)
11738 /*TODO: Check if need to increment number of frags */
11739 frag = vcpu->mmio_fragments;
11740 vcpu->mmio_nr_fragments = 1;
11745 vcpu->mmio_needed = 1;
11746 vcpu->mmio_cur_fragment = 0;
11748 vcpu->run->mmio.phys_addr = gpa;
11749 vcpu->run->mmio.len = min(8u, frag->len);
11750 vcpu->run->mmio.is_write = 1;
11751 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11752 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11754 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11758 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11760 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11764 struct kvm_mmio_fragment *frag;
11769 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11770 if (handled == bytes)
11777 /*TODO: Check if need to increment number of frags */
11778 frag = vcpu->mmio_fragments;
11779 vcpu->mmio_nr_fragments = 1;
11784 vcpu->mmio_needed = 1;
11785 vcpu->mmio_cur_fragment = 0;
11787 vcpu->run->mmio.phys_addr = gpa;
11788 vcpu->run->mmio.len = min(8u, frag->len);
11789 vcpu->run->mmio.is_write = 0;
11790 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11792 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11796 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11798 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11800 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11801 vcpu->arch.pio.count * vcpu->arch.pio.size);
11802 vcpu->arch.pio.count = 0;
11807 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11808 unsigned int port, void *data, unsigned int count)
11812 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11817 vcpu->arch.pio.count = 0;
11822 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11823 unsigned int port, void *data, unsigned int count)
11827 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11830 vcpu->arch.pio.count = 0;
11832 vcpu->arch.guest_ins_data = data;
11833 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11839 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11840 unsigned int port, void *data, unsigned int count,
11843 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11844 : kvm_sev_es_outs(vcpu, size, port, data, count);
11846 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11848 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11849 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11850 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11851 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11852 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11853 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11854 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11855 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11856 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11857 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11858 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11859 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11860 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11861 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11862 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11863 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11864 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11865 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11866 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11867 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11868 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11869 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11870 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11871 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11872 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11873 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11874 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);