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>
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 #include <linux/kvm_host.h>
26 #include "kvm_cache_regs.h"
27 #include "kvm_emulate.h"
28 #include "mmu/page_track.h"
37 #include <linux/clocksource.h>
38 #include <linux/interrupt.h>
39 #include <linux/kvm.h>
41 #include <linux/vmalloc.h>
42 #include <linux/export.h>
43 #include <linux/moduleparam.h>
44 #include <linux/mman.h>
45 #include <linux/highmem.h>
46 #include <linux/iommu.h>
47 #include <linux/cpufreq.h>
48 #include <linux/user-return-notifier.h>
49 #include <linux/srcu.h>
50 #include <linux/slab.h>
51 #include <linux/perf_event.h>
52 #include <linux/uaccess.h>
53 #include <linux/hash.h>
54 #include <linux/pci.h>
55 #include <linux/timekeeper_internal.h>
56 #include <linux/pvclock_gtod.h>
57 #include <linux/kvm_irqfd.h>
58 #include <linux/irqbypass.h>
59 #include <linux/sched/stat.h>
60 #include <linux/sched/isolation.h>
61 #include <linux/mem_encrypt.h>
62 #include <linux/entry-kvm.h>
63 #include <linux/suspend.h>
64 #include <linux/smp.h>
66 #include <trace/events/ipi.h>
67 #include <trace/events/kvm.h>
69 #include <asm/debugreg.h>
74 #include <linux/kernel_stat.h>
75 #include <asm/fpu/api.h>
76 #include <asm/fpu/xcr.h>
77 #include <asm/fpu/xstate.h>
78 #include <asm/pvclock.h>
79 #include <asm/div64.h>
80 #include <asm/irq_remapping.h>
81 #include <asm/mshyperv.h>
82 #include <asm/hypervisor.h>
83 #include <asm/tlbflush.h>
84 #include <asm/intel_pt.h>
85 #include <asm/emulate_prefix.h>
87 #include <clocksource/hyperv_timer.h>
89 #define CREATE_TRACE_POINTS
92 #define MAX_IO_MSRS 256
93 #define KVM_MAX_MCE_BANKS 32
95 struct kvm_caps kvm_caps __read_mostly = {
96 .supported_mce_cap = MCG_CTL_P | MCG_SER_P,
98 EXPORT_SYMBOL_GPL(kvm_caps);
100 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
102 #define emul_to_vcpu(ctxt) \
103 ((struct kvm_vcpu *)(ctxt)->vcpu)
106 * - enable syscall per default because its emulated by KVM
107 * - enable LME and LMA per default on 64 bit KVM
111 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
113 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
116 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
118 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
120 #define KVM_CAP_PMU_VALID_MASK KVM_PMU_CAP_DISABLE
122 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
123 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
125 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
126 static void process_nmi(struct kvm_vcpu *vcpu);
127 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
128 static void store_regs(struct kvm_vcpu *vcpu);
129 static int sync_regs(struct kvm_vcpu *vcpu);
130 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu);
132 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
133 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
135 static DEFINE_MUTEX(vendor_module_lock);
136 struct kvm_x86_ops kvm_x86_ops __read_mostly;
138 #define KVM_X86_OP(func) \
139 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
140 *(((struct kvm_x86_ops *)0)->func));
141 #define KVM_X86_OP_OPTIONAL KVM_X86_OP
142 #define KVM_X86_OP_OPTIONAL_RET0 KVM_X86_OP
143 #include <asm/kvm-x86-ops.h>
144 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
145 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
147 static bool __read_mostly ignore_msrs = 0;
148 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
150 bool __read_mostly report_ignored_msrs = true;
151 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
152 EXPORT_SYMBOL_GPL(report_ignored_msrs);
154 unsigned int min_timer_period_us = 200;
155 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
157 static bool __read_mostly kvmclock_periodic_sync = true;
158 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
160 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
161 static u32 __read_mostly tsc_tolerance_ppm = 250;
162 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
165 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
166 * adaptive tuning starting from default advancement of 1000ns. '0' disables
167 * advancement entirely. Any other value is used as-is and disables adaptive
168 * tuning, i.e. allows privileged userspace to set an exact advancement time.
170 static int __read_mostly lapic_timer_advance_ns = -1;
171 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
173 static bool __read_mostly vector_hashing = true;
174 module_param(vector_hashing, bool, S_IRUGO);
176 bool __read_mostly enable_vmware_backdoor = false;
177 module_param(enable_vmware_backdoor, bool, S_IRUGO);
178 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
181 * Flags to manipulate forced emulation behavior (any non-zero value will
182 * enable forced emulation).
184 #define KVM_FEP_CLEAR_RFLAGS_RF BIT(1)
185 static int __read_mostly force_emulation_prefix;
186 module_param(force_emulation_prefix, int, 0644);
188 int __read_mostly pi_inject_timer = -1;
189 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
191 /* Enable/disable PMU virtualization */
192 bool __read_mostly enable_pmu = true;
193 EXPORT_SYMBOL_GPL(enable_pmu);
194 module_param(enable_pmu, bool, 0444);
196 bool __read_mostly eager_page_split = true;
197 module_param(eager_page_split, bool, 0644);
199 /* Enable/disable SMT_RSB bug mitigation */
200 static bool __read_mostly mitigate_smt_rsb;
201 module_param(mitigate_smt_rsb, bool, 0444);
204 * Restoring the host value for MSRs that are only consumed when running in
205 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
206 * returns to userspace, i.e. the kernel can run with the guest's value.
208 #define KVM_MAX_NR_USER_RETURN_MSRS 16
210 struct kvm_user_return_msrs {
211 struct user_return_notifier urn;
213 struct kvm_user_return_msr_values {
216 } values[KVM_MAX_NR_USER_RETURN_MSRS];
219 u32 __read_mostly kvm_nr_uret_msrs;
220 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
221 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
222 static struct kvm_user_return_msrs __percpu *user_return_msrs;
224 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
225 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
226 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
227 | XFEATURE_MASK_PKRU | XFEATURE_MASK_XTILE)
229 u64 __read_mostly host_efer;
230 EXPORT_SYMBOL_GPL(host_efer);
232 bool __read_mostly allow_smaller_maxphyaddr = 0;
233 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
235 bool __read_mostly enable_apicv = true;
236 EXPORT_SYMBOL_GPL(enable_apicv);
238 u64 __read_mostly host_xss;
239 EXPORT_SYMBOL_GPL(host_xss);
241 u64 __read_mostly host_arch_capabilities;
242 EXPORT_SYMBOL_GPL(host_arch_capabilities);
244 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
245 KVM_GENERIC_VM_STATS(),
246 STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
247 STATS_DESC_COUNTER(VM, mmu_pte_write),
248 STATS_DESC_COUNTER(VM, mmu_pde_zapped),
249 STATS_DESC_COUNTER(VM, mmu_flooded),
250 STATS_DESC_COUNTER(VM, mmu_recycled),
251 STATS_DESC_COUNTER(VM, mmu_cache_miss),
252 STATS_DESC_ICOUNTER(VM, mmu_unsync),
253 STATS_DESC_ICOUNTER(VM, pages_4k),
254 STATS_DESC_ICOUNTER(VM, pages_2m),
255 STATS_DESC_ICOUNTER(VM, pages_1g),
256 STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
257 STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
258 STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
261 const struct kvm_stats_header kvm_vm_stats_header = {
262 .name_size = KVM_STATS_NAME_SIZE,
263 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
264 .id_offset = sizeof(struct kvm_stats_header),
265 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
266 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
267 sizeof(kvm_vm_stats_desc),
270 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
271 KVM_GENERIC_VCPU_STATS(),
272 STATS_DESC_COUNTER(VCPU, pf_taken),
273 STATS_DESC_COUNTER(VCPU, pf_fixed),
274 STATS_DESC_COUNTER(VCPU, pf_emulate),
275 STATS_DESC_COUNTER(VCPU, pf_spurious),
276 STATS_DESC_COUNTER(VCPU, pf_fast),
277 STATS_DESC_COUNTER(VCPU, pf_mmio_spte_created),
278 STATS_DESC_COUNTER(VCPU, pf_guest),
279 STATS_DESC_COUNTER(VCPU, tlb_flush),
280 STATS_DESC_COUNTER(VCPU, invlpg),
281 STATS_DESC_COUNTER(VCPU, exits),
282 STATS_DESC_COUNTER(VCPU, io_exits),
283 STATS_DESC_COUNTER(VCPU, mmio_exits),
284 STATS_DESC_COUNTER(VCPU, signal_exits),
285 STATS_DESC_COUNTER(VCPU, irq_window_exits),
286 STATS_DESC_COUNTER(VCPU, nmi_window_exits),
287 STATS_DESC_COUNTER(VCPU, l1d_flush),
288 STATS_DESC_COUNTER(VCPU, halt_exits),
289 STATS_DESC_COUNTER(VCPU, request_irq_exits),
290 STATS_DESC_COUNTER(VCPU, irq_exits),
291 STATS_DESC_COUNTER(VCPU, host_state_reload),
292 STATS_DESC_COUNTER(VCPU, fpu_reload),
293 STATS_DESC_COUNTER(VCPU, insn_emulation),
294 STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
295 STATS_DESC_COUNTER(VCPU, hypercalls),
296 STATS_DESC_COUNTER(VCPU, irq_injections),
297 STATS_DESC_COUNTER(VCPU, nmi_injections),
298 STATS_DESC_COUNTER(VCPU, req_event),
299 STATS_DESC_COUNTER(VCPU, nested_run),
300 STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
301 STATS_DESC_COUNTER(VCPU, directed_yield_successful),
302 STATS_DESC_COUNTER(VCPU, preemption_reported),
303 STATS_DESC_COUNTER(VCPU, preemption_other),
304 STATS_DESC_IBOOLEAN(VCPU, guest_mode),
305 STATS_DESC_COUNTER(VCPU, notify_window_exits),
308 const struct kvm_stats_header kvm_vcpu_stats_header = {
309 .name_size = KVM_STATS_NAME_SIZE,
310 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
311 .id_offset = sizeof(struct kvm_stats_header),
312 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
313 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
314 sizeof(kvm_vcpu_stats_desc),
317 u64 __read_mostly host_xcr0;
319 static struct kmem_cache *x86_emulator_cache;
322 * When called, it means the previous get/set msr reached an invalid msr.
323 * Return true if we want to ignore/silent this failed msr access.
325 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
327 const char *op = write ? "wrmsr" : "rdmsr";
330 if (report_ignored_msrs)
331 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
336 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
342 static struct kmem_cache *kvm_alloc_emulator_cache(void)
344 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
345 unsigned int size = sizeof(struct x86_emulate_ctxt);
347 return kmem_cache_create_usercopy("x86_emulator", size,
348 __alignof__(struct x86_emulate_ctxt),
349 SLAB_ACCOUNT, useroffset,
350 size - useroffset, NULL);
353 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
355 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
358 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
359 vcpu->arch.apf.gfns[i] = ~0;
362 static void kvm_on_user_return(struct user_return_notifier *urn)
365 struct kvm_user_return_msrs *msrs
366 = container_of(urn, struct kvm_user_return_msrs, urn);
367 struct kvm_user_return_msr_values *values;
371 * Disabling irqs at this point since the following code could be
372 * interrupted and executed through kvm_arch_hardware_disable()
374 local_irq_save(flags);
375 if (msrs->registered) {
376 msrs->registered = false;
377 user_return_notifier_unregister(urn);
379 local_irq_restore(flags);
380 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
381 values = &msrs->values[slot];
382 if (values->host != values->curr) {
383 wrmsrl(kvm_uret_msrs_list[slot], values->host);
384 values->curr = values->host;
389 static int kvm_probe_user_return_msr(u32 msr)
395 ret = rdmsrl_safe(msr, &val);
398 ret = wrmsrl_safe(msr, val);
404 int kvm_add_user_return_msr(u32 msr)
406 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
408 if (kvm_probe_user_return_msr(msr))
411 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
412 return kvm_nr_uret_msrs++;
414 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
416 int kvm_find_user_return_msr(u32 msr)
420 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
421 if (kvm_uret_msrs_list[i] == msr)
426 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
428 static void kvm_user_return_msr_cpu_online(void)
430 unsigned int cpu = smp_processor_id();
431 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
435 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
436 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
437 msrs->values[i].host = value;
438 msrs->values[i].curr = value;
442 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
444 unsigned int cpu = smp_processor_id();
445 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
448 value = (value & mask) | (msrs->values[slot].host & ~mask);
449 if (value == msrs->values[slot].curr)
451 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
455 msrs->values[slot].curr = value;
456 if (!msrs->registered) {
457 msrs->urn.on_user_return = kvm_on_user_return;
458 user_return_notifier_register(&msrs->urn);
459 msrs->registered = true;
463 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
465 static void drop_user_return_notifiers(void)
467 unsigned int cpu = smp_processor_id();
468 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
470 if (msrs->registered)
471 kvm_on_user_return(&msrs->urn);
474 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
476 return vcpu->arch.apic_base;
479 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
481 return kvm_apic_mode(kvm_get_apic_base(vcpu));
483 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
485 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
487 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
488 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
489 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
490 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
492 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
494 if (!msr_info->host_initiated) {
495 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
497 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
501 kvm_lapic_set_base(vcpu, msr_info->data);
502 kvm_recalculate_apic_map(vcpu->kvm);
507 * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
509 * Hardware virtualization extension instructions may fault if a reboot turns
510 * off virtualization while processes are running. Usually after catching the
511 * fault we just panic; during reboot instead the instruction is ignored.
513 noinstr void kvm_spurious_fault(void)
515 /* Fault while not rebooting. We want the trace. */
516 BUG_ON(!kvm_rebooting);
518 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
520 #define EXCPT_BENIGN 0
521 #define EXCPT_CONTRIBUTORY 1
524 static int exception_class(int vector)
534 return EXCPT_CONTRIBUTORY;
541 #define EXCPT_FAULT 0
543 #define EXCPT_ABORT 2
544 #define EXCPT_INTERRUPT 3
547 static int exception_type(int vector)
551 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
552 return EXCPT_INTERRUPT;
557 * #DBs can be trap-like or fault-like, the caller must check other CPU
558 * state, e.g. DR6, to determine whether a #DB is a trap or fault.
560 if (mask & (1 << DB_VECTOR))
563 if (mask & ((1 << BP_VECTOR) | (1 << OF_VECTOR)))
566 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
569 /* Reserved exceptions will result in fault */
573 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu,
574 struct kvm_queued_exception *ex)
576 if (!ex->has_payload)
579 switch (ex->vector) {
582 * "Certain debug exceptions may clear bit 0-3. The
583 * remaining contents of the DR6 register are never
584 * cleared by the processor".
586 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
588 * In order to reflect the #DB exception payload in guest
589 * dr6, three components need to be considered: active low
590 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
592 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
593 * In the target guest dr6:
594 * FIXED_1 bits should always be set.
595 * Active low bits should be cleared if 1-setting in payload.
596 * Active high bits should be set if 1-setting in payload.
598 * Note, the payload is compatible with the pending debug
599 * exceptions/exit qualification under VMX, that active_low bits
600 * are active high in payload.
601 * So they need to be flipped for DR6.
603 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
604 vcpu->arch.dr6 |= ex->payload;
605 vcpu->arch.dr6 ^= ex->payload & DR6_ACTIVE_LOW;
608 * The #DB payload is defined as compatible with the 'pending
609 * debug exceptions' field under VMX, not DR6. While bit 12 is
610 * defined in the 'pending debug exceptions' field (enabled
611 * breakpoint), it is reserved and must be zero in DR6.
613 vcpu->arch.dr6 &= ~BIT(12);
616 vcpu->arch.cr2 = ex->payload;
620 ex->has_payload = false;
623 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
625 static void kvm_queue_exception_vmexit(struct kvm_vcpu *vcpu, unsigned int vector,
626 bool has_error_code, u32 error_code,
627 bool has_payload, unsigned long payload)
629 struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
632 ex->injected = false;
634 ex->has_error_code = has_error_code;
635 ex->error_code = error_code;
636 ex->has_payload = has_payload;
637 ex->payload = payload;
640 /* Forcibly leave the nested mode in cases like a vCPU reset */
641 static void kvm_leave_nested(struct kvm_vcpu *vcpu)
643 kvm_x86_ops.nested_ops->leave_nested(vcpu);
646 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
647 unsigned nr, bool has_error, u32 error_code,
648 bool has_payload, unsigned long payload, bool reinject)
653 kvm_make_request(KVM_REQ_EVENT, vcpu);
656 * If the exception is destined for L2 and isn't being reinjected,
657 * morph it to a VM-Exit if L1 wants to intercept the exception. A
658 * previously injected exception is not checked because it was checked
659 * when it was original queued, and re-checking is incorrect if _L1_
660 * injected the exception, in which case it's exempt from interception.
662 if (!reinject && is_guest_mode(vcpu) &&
663 kvm_x86_ops.nested_ops->is_exception_vmexit(vcpu, nr, error_code)) {
664 kvm_queue_exception_vmexit(vcpu, nr, has_error, error_code,
665 has_payload, payload);
669 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
673 * On VM-Entry, an exception can be pending if and only
674 * if event injection was blocked by nested_run_pending.
675 * In that case, however, vcpu_enter_guest() requests an
676 * immediate exit, and the guest shouldn't proceed far
677 * enough to need reinjection.
679 WARN_ON_ONCE(kvm_is_exception_pending(vcpu));
680 vcpu->arch.exception.injected = true;
681 if (WARN_ON_ONCE(has_payload)) {
683 * A reinjected event has already
684 * delivered its payload.
690 vcpu->arch.exception.pending = true;
691 vcpu->arch.exception.injected = false;
693 vcpu->arch.exception.has_error_code = has_error;
694 vcpu->arch.exception.vector = nr;
695 vcpu->arch.exception.error_code = error_code;
696 vcpu->arch.exception.has_payload = has_payload;
697 vcpu->arch.exception.payload = payload;
698 if (!is_guest_mode(vcpu))
699 kvm_deliver_exception_payload(vcpu,
700 &vcpu->arch.exception);
704 /* to check exception */
705 prev_nr = vcpu->arch.exception.vector;
706 if (prev_nr == DF_VECTOR) {
707 /* triple fault -> shutdown */
708 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
711 class1 = exception_class(prev_nr);
712 class2 = exception_class(nr);
713 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY) ||
714 (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
716 * Synthesize #DF. Clear the previously injected or pending
717 * exception so as not to incorrectly trigger shutdown.
719 vcpu->arch.exception.injected = false;
720 vcpu->arch.exception.pending = false;
722 kvm_queue_exception_e(vcpu, DF_VECTOR, 0);
724 /* replace previous exception with a new one in a hope
725 that instruction re-execution will regenerate lost
731 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
733 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
735 EXPORT_SYMBOL_GPL(kvm_queue_exception);
737 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
739 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
741 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
743 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
744 unsigned long payload)
746 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
748 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
750 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
751 u32 error_code, unsigned long payload)
753 kvm_multiple_exception(vcpu, nr, true, error_code,
754 true, payload, false);
757 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
760 kvm_inject_gp(vcpu, 0);
762 return kvm_skip_emulated_instruction(vcpu);
766 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
768 static int complete_emulated_insn_gp(struct kvm_vcpu *vcpu, int err)
771 kvm_inject_gp(vcpu, 0);
775 return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE | EMULTYPE_SKIP |
776 EMULTYPE_COMPLETE_USER_EXIT);
779 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
781 ++vcpu->stat.pf_guest;
784 * Async #PF in L2 is always forwarded to L1 as a VM-Exit regardless of
785 * whether or not L1 wants to intercept "regular" #PF.
787 if (is_guest_mode(vcpu) && fault->async_page_fault)
788 kvm_queue_exception_vmexit(vcpu, PF_VECTOR,
789 true, fault->error_code,
790 true, fault->address);
792 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
796 void kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
797 struct x86_exception *fault)
799 struct kvm_mmu *fault_mmu;
800 WARN_ON_ONCE(fault->vector != PF_VECTOR);
802 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
806 * Invalidate the TLB entry for the faulting address, if it exists,
807 * else the access will fault indefinitely (and to emulate hardware).
809 if ((fault->error_code & PFERR_PRESENT_MASK) &&
810 !(fault->error_code & PFERR_RSVD_MASK))
811 kvm_mmu_invalidate_addr(vcpu, fault_mmu, fault->address,
812 KVM_MMU_ROOT_CURRENT);
814 fault_mmu->inject_page_fault(vcpu, fault);
816 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
818 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
820 atomic_inc(&vcpu->arch.nmi_queued);
821 kvm_make_request(KVM_REQ_NMI, vcpu);
824 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
826 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
828 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
830 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
832 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
834 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
837 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
838 * a #GP and return false.
840 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
842 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
844 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
848 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
850 if ((dr != 4 && dr != 5) || !kvm_is_cr4_bit_set(vcpu, X86_CR4_DE))
853 kvm_queue_exception(vcpu, UD_VECTOR);
856 EXPORT_SYMBOL_GPL(kvm_require_dr);
858 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
860 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
864 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
866 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
868 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
869 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
873 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
876 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
879 real_gpa = kvm_translate_gpa(vcpu, mmu, gfn_to_gpa(pdpt_gfn),
880 PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
881 if (real_gpa == INVALID_GPA)
884 /* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
885 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
886 cr3 & GENMASK(11, 5), sizeof(pdpte));
890 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
891 if ((pdpte[i] & PT_PRESENT_MASK) &&
892 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
898 * Marking VCPU_EXREG_PDPTR dirty doesn't work for !tdp_enabled.
899 * Shadow page roots need to be reconstructed instead.
901 if (!tdp_enabled && memcmp(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)))
902 kvm_mmu_free_roots(vcpu->kvm, mmu, KVM_MMU_ROOT_CURRENT);
904 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
905 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
906 kvm_make_request(KVM_REQ_LOAD_MMU_PGD, vcpu);
907 vcpu->arch.pdptrs_from_userspace = false;
911 EXPORT_SYMBOL_GPL(load_pdptrs);
913 static bool kvm_is_valid_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
916 if (cr0 & 0xffffffff00000000UL)
920 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
923 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
926 return static_call(kvm_x86_is_valid_cr0)(vcpu, cr0);
929 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
932 * CR0.WP is incorporated into the MMU role, but only for non-nested,
933 * indirect shadow MMUs. If paging is disabled, no updates are needed
934 * as there are no permission bits to emulate. If TDP is enabled, the
935 * MMU's metadata needs to be updated, e.g. so that emulating guest
936 * translations does the right thing, but there's no need to unload the
937 * root as CR0.WP doesn't affect SPTEs.
939 if ((cr0 ^ old_cr0) == X86_CR0_WP) {
940 if (!(cr0 & X86_CR0_PG))
949 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
950 kvm_clear_async_pf_completion_queue(vcpu);
951 kvm_async_pf_hash_reset(vcpu);
954 * Clearing CR0.PG is defined to flush the TLB from the guest's
957 if (!(cr0 & X86_CR0_PG))
958 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
961 if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
962 kvm_mmu_reset_context(vcpu);
964 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
965 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
966 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
967 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
969 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
971 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
973 unsigned long old_cr0 = kvm_read_cr0(vcpu);
975 if (!kvm_is_valid_cr0(vcpu, cr0))
980 /* Write to CR0 reserved bits are ignored, even on Intel. */
981 cr0 &= ~CR0_RESERVED_BITS;
984 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
985 (cr0 & X86_CR0_PG)) {
990 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
995 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
996 is_pae(vcpu) && ((cr0 ^ old_cr0) & X86_CR0_PDPTR_BITS) &&
997 !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
1000 if (!(cr0 & X86_CR0_PG) &&
1001 (is_64_bit_mode(vcpu) || kvm_is_cr4_bit_set(vcpu, X86_CR4_PCIDE)))
1004 static_call(kvm_x86_set_cr0)(vcpu, cr0);
1006 kvm_post_set_cr0(vcpu, old_cr0, cr0);
1010 EXPORT_SYMBOL_GPL(kvm_set_cr0);
1012 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
1014 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
1016 EXPORT_SYMBOL_GPL(kvm_lmsw);
1018 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
1020 if (vcpu->arch.guest_state_protected)
1023 if (kvm_is_cr4_bit_set(vcpu, X86_CR4_OSXSAVE)) {
1025 if (vcpu->arch.xcr0 != host_xcr0)
1026 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
1028 if (guest_can_use(vcpu, X86_FEATURE_XSAVES) &&
1029 vcpu->arch.ia32_xss != host_xss)
1030 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
1033 if (cpu_feature_enabled(X86_FEATURE_PKU) &&
1034 vcpu->arch.pkru != vcpu->arch.host_pkru &&
1035 ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
1036 kvm_is_cr4_bit_set(vcpu, X86_CR4_PKE)))
1037 write_pkru(vcpu->arch.pkru);
1039 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
1041 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
1043 if (vcpu->arch.guest_state_protected)
1046 if (cpu_feature_enabled(X86_FEATURE_PKU) &&
1047 ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
1048 kvm_is_cr4_bit_set(vcpu, X86_CR4_PKE))) {
1049 vcpu->arch.pkru = rdpkru();
1050 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
1051 write_pkru(vcpu->arch.host_pkru);
1054 if (kvm_is_cr4_bit_set(vcpu, X86_CR4_OSXSAVE)) {
1056 if (vcpu->arch.xcr0 != host_xcr0)
1057 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
1059 if (guest_can_use(vcpu, X86_FEATURE_XSAVES) &&
1060 vcpu->arch.ia32_xss != host_xss)
1061 wrmsrl(MSR_IA32_XSS, host_xss);
1065 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
1067 #ifdef CONFIG_X86_64
1068 static inline u64 kvm_guest_supported_xfd(struct kvm_vcpu *vcpu)
1070 return vcpu->arch.guest_supported_xcr0 & XFEATURE_MASK_USER_DYNAMIC;
1074 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
1077 u64 old_xcr0 = vcpu->arch.xcr0;
1080 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
1081 if (index != XCR_XFEATURE_ENABLED_MASK)
1083 if (!(xcr0 & XFEATURE_MASK_FP))
1085 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
1089 * Do not allow the guest to set bits that we do not support
1090 * saving. However, xcr0 bit 0 is always set, even if the
1091 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
1093 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
1094 if (xcr0 & ~valid_bits)
1097 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1098 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1101 if (xcr0 & XFEATURE_MASK_AVX512) {
1102 if (!(xcr0 & XFEATURE_MASK_YMM))
1104 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1108 if ((xcr0 & XFEATURE_MASK_XTILE) &&
1109 ((xcr0 & XFEATURE_MASK_XTILE) != XFEATURE_MASK_XTILE))
1112 vcpu->arch.xcr0 = xcr0;
1114 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1115 kvm_update_cpuid_runtime(vcpu);
1119 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1121 /* Note, #UD due to CR4.OSXSAVE=0 has priority over the intercept. */
1122 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1123 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1124 kvm_inject_gp(vcpu, 0);
1128 return kvm_skip_emulated_instruction(vcpu);
1130 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1132 bool __kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1134 if (cr4 & cr4_reserved_bits)
1137 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1142 EXPORT_SYMBOL_GPL(__kvm_is_valid_cr4);
1144 static bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1146 return __kvm_is_valid_cr4(vcpu, cr4) &&
1147 static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1150 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1152 if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1153 kvm_mmu_reset_context(vcpu);
1156 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1157 * according to the SDM; however, stale prev_roots could be reused
1158 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1159 * free them all. This is *not* a superset of KVM_REQ_TLB_FLUSH_GUEST
1160 * or KVM_REQ_TLB_FLUSH_CURRENT, because the hardware TLB is not flushed,
1164 (cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE))
1165 kvm_mmu_unload(vcpu);
1168 * The TLB has to be flushed for all PCIDs if any of the following
1169 * (architecturally required) changes happen:
1170 * - CR4.PCIDE is changed from 1 to 0
1171 * - CR4.PGE is toggled
1173 * This is a superset of KVM_REQ_TLB_FLUSH_CURRENT.
1175 if (((cr4 ^ old_cr4) & X86_CR4_PGE) ||
1176 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1177 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1180 * The TLB has to be flushed for the current PCID if any of the
1181 * following (architecturally required) changes happen:
1182 * - CR4.SMEP is changed from 0 to 1
1183 * - CR4.PAE is toggled
1185 else if (((cr4 ^ old_cr4) & X86_CR4_PAE) ||
1186 ((cr4 & X86_CR4_SMEP) && !(old_cr4 & X86_CR4_SMEP)))
1187 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1190 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1192 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1194 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1196 if (!kvm_is_valid_cr4(vcpu, cr4))
1199 if (is_long_mode(vcpu)) {
1200 if (!(cr4 & X86_CR4_PAE))
1202 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1204 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1205 && ((cr4 ^ old_cr4) & X86_CR4_PDPTR_BITS)
1206 && !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
1209 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1210 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1211 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1215 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1217 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1221 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1223 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1225 struct kvm_mmu *mmu = vcpu->arch.mmu;
1226 unsigned long roots_to_free = 0;
1230 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1231 * this is reachable when running EPT=1 and unrestricted_guest=0, and
1232 * also via the emulator. KVM's TDP page tables are not in the scope of
1233 * the invalidation, but the guest's TLB entries need to be flushed as
1234 * the CPU may have cached entries in its TLB for the target PCID.
1236 if (unlikely(tdp_enabled)) {
1237 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1242 * If neither the current CR3 nor any of the prev_roots use the given
1243 * PCID, then nothing needs to be done here because a resync will
1244 * happen anyway before switching to any other CR3.
1246 if (kvm_get_active_pcid(vcpu) == pcid) {
1247 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1248 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1252 * If PCID is disabled, there is no need to free prev_roots even if the
1253 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1256 if (!kvm_is_cr4_bit_set(vcpu, X86_CR4_PCIDE))
1259 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1260 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1261 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1263 kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
1266 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1268 bool skip_tlb_flush = false;
1269 unsigned long pcid = 0;
1270 #ifdef CONFIG_X86_64
1271 if (kvm_is_cr4_bit_set(vcpu, X86_CR4_PCIDE)) {
1272 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1273 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1274 pcid = cr3 & X86_CR3_PCID_MASK;
1278 /* PDPTRs are always reloaded for PAE paging. */
1279 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1280 goto handle_tlb_flush;
1283 * Do not condition the GPA check on long mode, this helper is used to
1284 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1285 * the current vCPU mode is accurate.
1287 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1290 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, cr3))
1293 if (cr3 != kvm_read_cr3(vcpu))
1294 kvm_mmu_new_pgd(vcpu, cr3);
1296 vcpu->arch.cr3 = cr3;
1297 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1298 /* Do not call post_set_cr3, we do not get here for confidential guests. */
1302 * A load of CR3 that flushes the TLB flushes only the current PCID,
1303 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1304 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1305 * and it's impossible to use a non-zero PCID when PCID is disabled,
1306 * i.e. only PCID=0 can be relevant.
1308 if (!skip_tlb_flush)
1309 kvm_invalidate_pcid(vcpu, pcid);
1313 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1315 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1317 if (cr8 & CR8_RESERVED_BITS)
1319 if (lapic_in_kernel(vcpu))
1320 kvm_lapic_set_tpr(vcpu, cr8);
1322 vcpu->arch.cr8 = cr8;
1325 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1327 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1329 if (lapic_in_kernel(vcpu))
1330 return kvm_lapic_get_cr8(vcpu);
1332 return vcpu->arch.cr8;
1334 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1336 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1340 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1341 for (i = 0; i < KVM_NR_DB_REGS; i++)
1342 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1346 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1350 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1351 dr7 = vcpu->arch.guest_debug_dr7;
1353 dr7 = vcpu->arch.dr7;
1354 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1355 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1356 if (dr7 & DR7_BP_EN_MASK)
1357 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1359 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1361 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1363 u64 fixed = DR6_FIXED_1;
1365 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1368 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1369 fixed |= DR6_BUS_LOCK;
1373 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1375 size_t size = ARRAY_SIZE(vcpu->arch.db);
1379 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1380 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1381 vcpu->arch.eff_db[dr] = val;
1385 if (!kvm_dr6_valid(val))
1387 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1391 if (!kvm_dr7_valid(val))
1393 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1394 kvm_update_dr7(vcpu);
1400 EXPORT_SYMBOL_GPL(kvm_set_dr);
1402 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1404 size_t size = ARRAY_SIZE(vcpu->arch.db);
1408 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1412 *val = vcpu->arch.dr6;
1416 *val = vcpu->arch.dr7;
1420 EXPORT_SYMBOL_GPL(kvm_get_dr);
1422 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1424 u32 ecx = kvm_rcx_read(vcpu);
1427 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1428 kvm_inject_gp(vcpu, 0);
1432 kvm_rax_write(vcpu, (u32)data);
1433 kvm_rdx_write(vcpu, data >> 32);
1434 return kvm_skip_emulated_instruction(vcpu);
1436 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1439 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features) track
1440 * the set of MSRs that KVM exposes to userspace through KVM_GET_MSRS,
1441 * KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST. msrs_to_save holds MSRs that
1442 * require host support, i.e. should be probed via RDMSR. emulated_msrs holds
1443 * MSRs that KVM emulates without strictly requiring host support.
1444 * msr_based_features holds MSRs that enumerate features, i.e. are effectively
1445 * CPUID leafs. Note, msr_based_features isn't mutually exclusive with
1446 * msrs_to_save and emulated_msrs.
1449 static const u32 msrs_to_save_base[] = {
1450 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1452 #ifdef CONFIG_X86_64
1453 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1455 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1456 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1457 MSR_IA32_SPEC_CTRL, MSR_IA32_TSX_CTRL,
1458 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1459 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1460 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1461 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1462 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1463 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1464 MSR_IA32_UMWAIT_CONTROL,
1466 MSR_IA32_XFD, MSR_IA32_XFD_ERR,
1469 static const u32 msrs_to_save_pmu[] = {
1470 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1471 MSR_ARCH_PERFMON_FIXED_CTR0 + 2,
1472 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1473 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1474 MSR_IA32_PEBS_ENABLE, MSR_IA32_DS_AREA, MSR_PEBS_DATA_CFG,
1476 /* This part of MSRs should match KVM_INTEL_PMC_MAX_GENERIC. */
1477 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1478 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1479 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1480 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1481 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1482 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1483 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1484 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1486 MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1487 MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1489 /* This part of MSRs should match KVM_AMD_PMC_MAX_GENERIC. */
1490 MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1491 MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1492 MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1493 MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1495 MSR_AMD64_PERF_CNTR_GLOBAL_CTL,
1496 MSR_AMD64_PERF_CNTR_GLOBAL_STATUS,
1497 MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR,
1500 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_base) +
1501 ARRAY_SIZE(msrs_to_save_pmu)];
1502 static unsigned num_msrs_to_save;
1504 static const u32 emulated_msrs_all[] = {
1505 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1506 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1507 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1508 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1509 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1510 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1511 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1513 HV_X64_MSR_VP_INDEX,
1514 HV_X64_MSR_VP_RUNTIME,
1515 HV_X64_MSR_SCONTROL,
1516 HV_X64_MSR_STIMER0_CONFIG,
1517 HV_X64_MSR_VP_ASSIST_PAGE,
1518 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1519 HV_X64_MSR_TSC_EMULATION_STATUS, HV_X64_MSR_TSC_INVARIANT_CONTROL,
1520 HV_X64_MSR_SYNDBG_OPTIONS,
1521 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1522 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1523 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1525 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1526 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1528 MSR_IA32_TSC_ADJUST,
1529 MSR_IA32_TSC_DEADLINE,
1530 MSR_IA32_ARCH_CAPABILITIES,
1531 MSR_IA32_PERF_CAPABILITIES,
1532 MSR_IA32_MISC_ENABLE,
1533 MSR_IA32_MCG_STATUS,
1535 MSR_IA32_MCG_EXT_CTL,
1539 MSR_MISC_FEATURES_ENABLES,
1540 MSR_AMD64_VIRT_SPEC_CTRL,
1541 MSR_AMD64_TSC_RATIO,
1546 * KVM always supports the "true" VMX control MSRs, even if the host
1547 * does not. The VMX MSRs as a whole are considered "emulated" as KVM
1548 * doesn't strictly require them to exist in the host (ignoring that
1549 * KVM would refuse to load in the first place if the core set of MSRs
1550 * aren't supported).
1553 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1554 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1555 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1556 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1558 MSR_IA32_VMX_CR0_FIXED0,
1559 MSR_IA32_VMX_CR4_FIXED0,
1560 MSR_IA32_VMX_VMCS_ENUM,
1561 MSR_IA32_VMX_PROCBASED_CTLS2,
1562 MSR_IA32_VMX_EPT_VPID_CAP,
1563 MSR_IA32_VMX_VMFUNC,
1566 MSR_KVM_POLL_CONTROL,
1569 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1570 static unsigned num_emulated_msrs;
1573 * List of MSRs that control the existence of MSR-based features, i.e. MSRs
1574 * that are effectively CPUID leafs. VMX MSRs are also included in the set of
1575 * feature MSRs, but are handled separately to allow expedited lookups.
1577 static const u32 msr_based_features_all_except_vmx[] = {
1580 MSR_IA32_ARCH_CAPABILITIES,
1581 MSR_IA32_PERF_CAPABILITIES,
1584 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all_except_vmx) +
1585 (KVM_LAST_EMULATED_VMX_MSR - KVM_FIRST_EMULATED_VMX_MSR + 1)];
1586 static unsigned int num_msr_based_features;
1589 * All feature MSRs except uCode revID, which tracks the currently loaded uCode
1590 * patch, are immutable once the vCPU model is defined.
1592 static bool kvm_is_immutable_feature_msr(u32 msr)
1596 if (msr >= KVM_FIRST_EMULATED_VMX_MSR && msr <= KVM_LAST_EMULATED_VMX_MSR)
1599 for (i = 0; i < ARRAY_SIZE(msr_based_features_all_except_vmx); i++) {
1600 if (msr == msr_based_features_all_except_vmx[i])
1601 return msr != MSR_IA32_UCODE_REV;
1608 * Some IA32_ARCH_CAPABILITIES bits have dependencies on MSRs that KVM
1609 * does not yet virtualize. These include:
1610 * 10 - MISC_PACKAGE_CTRLS
1611 * 11 - ENERGY_FILTERING_CTL
1613 * 18 - FB_CLEAR_CTRL
1614 * 21 - XAPIC_DISABLE_STATUS
1615 * 23 - OVERCLOCKING_STATUS
1618 #define KVM_SUPPORTED_ARCH_CAP \
1619 (ARCH_CAP_RDCL_NO | ARCH_CAP_IBRS_ALL | ARCH_CAP_RSBA | \
1620 ARCH_CAP_SKIP_VMENTRY_L1DFLUSH | ARCH_CAP_SSB_NO | ARCH_CAP_MDS_NO | \
1621 ARCH_CAP_PSCHANGE_MC_NO | ARCH_CAP_TSX_CTRL_MSR | ARCH_CAP_TAA_NO | \
1622 ARCH_CAP_SBDR_SSDP_NO | ARCH_CAP_FBSDP_NO | ARCH_CAP_PSDP_NO | \
1623 ARCH_CAP_FB_CLEAR | ARCH_CAP_RRSBA | ARCH_CAP_PBRSB_NO | ARCH_CAP_GDS_NO)
1625 static u64 kvm_get_arch_capabilities(void)
1627 u64 data = host_arch_capabilities & KVM_SUPPORTED_ARCH_CAP;
1630 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1631 * the nested hypervisor runs with NX huge pages. If it is not,
1632 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1633 * L1 guests, so it need not worry about its own (L2) guests.
1635 data |= ARCH_CAP_PSCHANGE_MC_NO;
1638 * If we're doing cache flushes (either "always" or "cond")
1639 * we will do one whenever the guest does a vmlaunch/vmresume.
1640 * If an outer hypervisor is doing the cache flush for us
1641 * (ARCH_CAP_SKIP_VMENTRY_L1DFLUSH), we can safely pass that
1642 * capability to the guest too, and if EPT is disabled we're not
1643 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1644 * require a nested hypervisor to do a flush of its own.
1646 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1647 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1649 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1650 data |= ARCH_CAP_RDCL_NO;
1651 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1652 data |= ARCH_CAP_SSB_NO;
1653 if (!boot_cpu_has_bug(X86_BUG_MDS))
1654 data |= ARCH_CAP_MDS_NO;
1656 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1658 * If RTM=0 because the kernel has disabled TSX, the host might
1659 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1660 * and therefore knows that there cannot be TAA) but keep
1661 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1662 * and we want to allow migrating those guests to tsx=off hosts.
1664 data &= ~ARCH_CAP_TAA_NO;
1665 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1666 data |= ARCH_CAP_TAA_NO;
1669 * Nothing to do here; we emulate TSX_CTRL if present on the
1670 * host so the guest can choose between disabling TSX or
1671 * using VERW to clear CPU buffers.
1675 if (!boot_cpu_has_bug(X86_BUG_GDS) || gds_ucode_mitigated())
1676 data |= ARCH_CAP_GDS_NO;
1681 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1683 switch (msr->index) {
1684 case MSR_IA32_ARCH_CAPABILITIES:
1685 msr->data = kvm_get_arch_capabilities();
1687 case MSR_IA32_PERF_CAPABILITIES:
1688 msr->data = kvm_caps.supported_perf_cap;
1690 case MSR_IA32_UCODE_REV:
1691 rdmsrl_safe(msr->index, &msr->data);
1694 return static_call(kvm_x86_get_msr_feature)(msr);
1699 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1701 struct kvm_msr_entry msr;
1705 r = kvm_get_msr_feature(&msr);
1707 if (r == KVM_MSR_RET_INVALID) {
1708 /* Unconditionally clear the output for simplicity */
1710 if (kvm_msr_ignored_check(index, 0, false))
1722 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1724 if (efer & EFER_AUTOIBRS && !guest_cpuid_has(vcpu, X86_FEATURE_AUTOIBRS))
1727 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1730 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1733 if (efer & (EFER_LME | EFER_LMA) &&
1734 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1737 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1743 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1745 if (efer & efer_reserved_bits)
1748 return __kvm_valid_efer(vcpu, efer);
1750 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1752 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1754 u64 old_efer = vcpu->arch.efer;
1755 u64 efer = msr_info->data;
1758 if (efer & efer_reserved_bits)
1761 if (!msr_info->host_initiated) {
1762 if (!__kvm_valid_efer(vcpu, efer))
1765 if (is_paging(vcpu) &&
1766 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1771 efer |= vcpu->arch.efer & EFER_LMA;
1773 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1779 if ((efer ^ old_efer) & KVM_MMU_EFER_ROLE_BITS)
1780 kvm_mmu_reset_context(vcpu);
1785 void kvm_enable_efer_bits(u64 mask)
1787 efer_reserved_bits &= ~mask;
1789 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1791 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1793 struct kvm_x86_msr_filter *msr_filter;
1794 struct msr_bitmap_range *ranges;
1795 struct kvm *kvm = vcpu->kvm;
1800 /* x2APIC MSRs do not support filtering. */
1801 if (index >= 0x800 && index <= 0x8ff)
1804 idx = srcu_read_lock(&kvm->srcu);
1806 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1812 allowed = msr_filter->default_allow;
1813 ranges = msr_filter->ranges;
1815 for (i = 0; i < msr_filter->count; i++) {
1816 u32 start = ranges[i].base;
1817 u32 end = start + ranges[i].nmsrs;
1818 u32 flags = ranges[i].flags;
1819 unsigned long *bitmap = ranges[i].bitmap;
1821 if ((index >= start) && (index < end) && (flags & type)) {
1822 allowed = test_bit(index - start, bitmap);
1828 srcu_read_unlock(&kvm->srcu, idx);
1832 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1835 * Write @data into the MSR specified by @index. Select MSR specific fault
1836 * checks are bypassed if @host_initiated is %true.
1837 * Returns 0 on success, non-0 otherwise.
1838 * Assumes vcpu_load() was already called.
1840 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1841 bool host_initiated)
1843 struct msr_data msr;
1848 case MSR_KERNEL_GS_BASE:
1851 if (is_noncanonical_address(data, vcpu))
1854 case MSR_IA32_SYSENTER_EIP:
1855 case MSR_IA32_SYSENTER_ESP:
1857 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1858 * non-canonical address is written on Intel but not on
1859 * AMD (which ignores the top 32-bits, because it does
1860 * not implement 64-bit SYSENTER).
1862 * 64-bit code should hence be able to write a non-canonical
1863 * value on AMD. Making the address canonical ensures that
1864 * vmentry does not fail on Intel after writing a non-canonical
1865 * value, and that something deterministic happens if the guest
1866 * invokes 64-bit SYSENTER.
1868 data = __canonical_address(data, vcpu_virt_addr_bits(vcpu));
1871 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1874 if (!host_initiated &&
1875 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1876 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1880 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1881 * incomplete and conflicting architectural behavior. Current
1882 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1883 * reserved and always read as zeros. Enforce Intel's reserved
1884 * bits check if and only if the guest CPU is Intel, and clear
1885 * the bits in all other cases. This ensures cross-vendor
1886 * migration will provide consistent behavior for the guest.
1888 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1897 msr.host_initiated = host_initiated;
1899 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1902 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1903 u32 index, u64 data, bool host_initiated)
1905 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1907 if (ret == KVM_MSR_RET_INVALID)
1908 if (kvm_msr_ignored_check(index, data, true))
1915 * Read the MSR specified by @index into @data. Select MSR specific fault
1916 * checks are bypassed if @host_initiated is %true.
1917 * Returns 0 on success, non-0 otherwise.
1918 * Assumes vcpu_load() was already called.
1920 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1921 bool host_initiated)
1923 struct msr_data msr;
1928 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1931 if (!host_initiated &&
1932 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1933 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1939 msr.host_initiated = host_initiated;
1941 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1947 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1948 u32 index, u64 *data, bool host_initiated)
1950 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1952 if (ret == KVM_MSR_RET_INVALID) {
1953 /* Unconditionally clear *data for simplicity */
1955 if (kvm_msr_ignored_check(index, 0, false))
1962 static int kvm_get_msr_with_filter(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1964 if (!kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1965 return KVM_MSR_RET_FILTERED;
1966 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1969 static int kvm_set_msr_with_filter(struct kvm_vcpu *vcpu, u32 index, u64 data)
1971 if (!kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1972 return KVM_MSR_RET_FILTERED;
1973 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1976 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1978 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1980 EXPORT_SYMBOL_GPL(kvm_get_msr);
1982 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1984 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1986 EXPORT_SYMBOL_GPL(kvm_set_msr);
1988 static void complete_userspace_rdmsr(struct kvm_vcpu *vcpu)
1990 if (!vcpu->run->msr.error) {
1991 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1992 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1996 static int complete_emulated_msr_access(struct kvm_vcpu *vcpu)
1998 return complete_emulated_insn_gp(vcpu, vcpu->run->msr.error);
2001 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
2003 complete_userspace_rdmsr(vcpu);
2004 return complete_emulated_msr_access(vcpu);
2007 static int complete_fast_msr_access(struct kvm_vcpu *vcpu)
2009 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
2012 static int complete_fast_rdmsr(struct kvm_vcpu *vcpu)
2014 complete_userspace_rdmsr(vcpu);
2015 return complete_fast_msr_access(vcpu);
2018 static u64 kvm_msr_reason(int r)
2021 case KVM_MSR_RET_INVALID:
2022 return KVM_MSR_EXIT_REASON_UNKNOWN;
2023 case KVM_MSR_RET_FILTERED:
2024 return KVM_MSR_EXIT_REASON_FILTER;
2026 return KVM_MSR_EXIT_REASON_INVAL;
2030 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
2031 u32 exit_reason, u64 data,
2032 int (*completion)(struct kvm_vcpu *vcpu),
2035 u64 msr_reason = kvm_msr_reason(r);
2037 /* Check if the user wanted to know about this MSR fault */
2038 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
2041 vcpu->run->exit_reason = exit_reason;
2042 vcpu->run->msr.error = 0;
2043 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
2044 vcpu->run->msr.reason = msr_reason;
2045 vcpu->run->msr.index = index;
2046 vcpu->run->msr.data = data;
2047 vcpu->arch.complete_userspace_io = completion;
2052 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
2054 u32 ecx = kvm_rcx_read(vcpu);
2058 r = kvm_get_msr_with_filter(vcpu, ecx, &data);
2061 trace_kvm_msr_read(ecx, data);
2063 kvm_rax_write(vcpu, data & -1u);
2064 kvm_rdx_write(vcpu, (data >> 32) & -1u);
2066 /* MSR read failed? See if we should ask user space */
2067 if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_RDMSR, 0,
2068 complete_fast_rdmsr, r))
2070 trace_kvm_msr_read_ex(ecx);
2073 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
2075 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
2077 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
2079 u32 ecx = kvm_rcx_read(vcpu);
2080 u64 data = kvm_read_edx_eax(vcpu);
2083 r = kvm_set_msr_with_filter(vcpu, ecx, data);
2086 trace_kvm_msr_write(ecx, data);
2088 /* MSR write failed? See if we should ask user space */
2089 if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_WRMSR, data,
2090 complete_fast_msr_access, r))
2092 /* Signal all other negative errors to userspace */
2095 trace_kvm_msr_write_ex(ecx, data);
2098 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
2100 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
2102 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
2104 return kvm_skip_emulated_instruction(vcpu);
2107 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
2109 /* Treat an INVD instruction as a NOP and just skip it. */
2110 return kvm_emulate_as_nop(vcpu);
2112 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
2114 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
2116 kvm_queue_exception(vcpu, UD_VECTOR);
2119 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
2122 static int kvm_emulate_monitor_mwait(struct kvm_vcpu *vcpu, const char *insn)
2124 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MWAIT_NEVER_UD_FAULTS) &&
2125 !guest_cpuid_has(vcpu, X86_FEATURE_MWAIT))
2126 return kvm_handle_invalid_op(vcpu);
2128 pr_warn_once("%s instruction emulated as NOP!\n", insn);
2129 return kvm_emulate_as_nop(vcpu);
2131 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
2133 return kvm_emulate_monitor_mwait(vcpu, "MWAIT");
2135 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
2137 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
2139 return kvm_emulate_monitor_mwait(vcpu, "MONITOR");
2141 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
2143 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
2145 xfer_to_guest_mode_prepare();
2146 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
2147 xfer_to_guest_mode_work_pending();
2151 * The fast path for frequent and performance sensitive wrmsr emulation,
2152 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
2153 * the latency of virtual IPI by avoiding the expensive bits of transitioning
2154 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
2155 * other cases which must be called after interrupts are enabled on the host.
2157 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
2159 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
2162 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
2163 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
2164 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
2165 ((u32)(data >> 32) != X2APIC_BROADCAST))
2166 return kvm_x2apic_icr_write(vcpu->arch.apic, data);
2171 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2173 if (!kvm_can_use_hv_timer(vcpu))
2176 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2180 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2182 u32 msr = kvm_rcx_read(vcpu);
2184 fastpath_t ret = EXIT_FASTPATH_NONE;
2186 kvm_vcpu_srcu_read_lock(vcpu);
2189 case APIC_BASE_MSR + (APIC_ICR >> 4):
2190 data = kvm_read_edx_eax(vcpu);
2191 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2192 kvm_skip_emulated_instruction(vcpu);
2193 ret = EXIT_FASTPATH_EXIT_HANDLED;
2196 case MSR_IA32_TSC_DEADLINE:
2197 data = kvm_read_edx_eax(vcpu);
2198 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2199 kvm_skip_emulated_instruction(vcpu);
2200 ret = EXIT_FASTPATH_REENTER_GUEST;
2207 if (ret != EXIT_FASTPATH_NONE)
2208 trace_kvm_msr_write(msr, data);
2210 kvm_vcpu_srcu_read_unlock(vcpu);
2214 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2217 * Adapt set_msr() to msr_io()'s calling convention
2219 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2221 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2224 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2229 * Disallow writes to immutable feature MSRs after KVM_RUN. KVM does
2230 * not support modifying the guest vCPU model on the fly, e.g. changing
2231 * the nVMX capabilities while L2 is running is nonsensical. Ignore
2232 * writes of the same value, e.g. to allow userspace to blindly stuff
2233 * all MSRs when emulating RESET.
2235 if (kvm_vcpu_has_run(vcpu) && kvm_is_immutable_feature_msr(index)) {
2236 if (do_get_msr(vcpu, index, &val) || *data != val)
2242 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2245 #ifdef CONFIG_X86_64
2246 struct pvclock_clock {
2256 struct pvclock_gtod_data {
2259 struct pvclock_clock clock; /* extract of a clocksource struct */
2260 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2266 static struct pvclock_gtod_data pvclock_gtod_data;
2268 static void update_pvclock_gtod(struct timekeeper *tk)
2270 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2272 write_seqcount_begin(&vdata->seq);
2274 /* copy pvclock gtod data */
2275 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2276 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2277 vdata->clock.mask = tk->tkr_mono.mask;
2278 vdata->clock.mult = tk->tkr_mono.mult;
2279 vdata->clock.shift = tk->tkr_mono.shift;
2280 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2281 vdata->clock.offset = tk->tkr_mono.base;
2283 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2284 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2285 vdata->raw_clock.mask = tk->tkr_raw.mask;
2286 vdata->raw_clock.mult = tk->tkr_raw.mult;
2287 vdata->raw_clock.shift = tk->tkr_raw.shift;
2288 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2289 vdata->raw_clock.offset = tk->tkr_raw.base;
2291 vdata->wall_time_sec = tk->xtime_sec;
2293 vdata->offs_boot = tk->offs_boot;
2295 write_seqcount_end(&vdata->seq);
2298 static s64 get_kvmclock_base_ns(void)
2300 /* Count up from boot time, but with the frequency of the raw clock. */
2301 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2304 static s64 get_kvmclock_base_ns(void)
2306 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2307 return ktime_get_boottime_ns();
2311 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2315 struct pvclock_wall_clock wc;
2322 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2327 ++version; /* first time write, random junk */
2331 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2335 * The guest calculates current wall clock time by adding
2336 * system time (updated by kvm_guest_time_update below) to the
2337 * wall clock specified here. We do the reverse here.
2339 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2341 wc.nsec = do_div(wall_nsec, 1000000000);
2342 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2343 wc.version = version;
2345 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2348 wc_sec_hi = wall_nsec >> 32;
2349 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2350 &wc_sec_hi, sizeof(wc_sec_hi));
2354 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2357 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2358 bool old_msr, bool host_initiated)
2360 struct kvm_arch *ka = &vcpu->kvm->arch;
2362 if (vcpu->vcpu_id == 0 && !host_initiated) {
2363 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2364 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2366 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2369 vcpu->arch.time = system_time;
2370 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2372 /* we verify if the enable bit is set... */
2373 if (system_time & 1)
2374 kvm_gpc_activate(&vcpu->arch.pv_time, system_time & ~1ULL,
2375 sizeof(struct pvclock_vcpu_time_info));
2377 kvm_gpc_deactivate(&vcpu->arch.pv_time);
2382 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2384 do_shl32_div32(dividend, divisor);
2388 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2389 s8 *pshift, u32 *pmultiplier)
2397 scaled64 = scaled_hz;
2398 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2403 tps32 = (uint32_t)tps64;
2404 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2405 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2413 *pmultiplier = div_frac(scaled64, tps32);
2416 #ifdef CONFIG_X86_64
2417 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2420 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2421 static unsigned long max_tsc_khz;
2423 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2425 u64 v = (u64)khz * (1000000 + ppm);
2430 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2432 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2436 /* Guest TSC same frequency as host TSC? */
2438 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_caps.default_tsc_scaling_ratio);
2442 /* TSC scaling supported? */
2443 if (!kvm_caps.has_tsc_control) {
2444 if (user_tsc_khz > tsc_khz) {
2445 vcpu->arch.tsc_catchup = 1;
2446 vcpu->arch.tsc_always_catchup = 1;
2449 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2454 /* TSC scaling required - calculate ratio */
2455 ratio = mul_u64_u32_div(1ULL << kvm_caps.tsc_scaling_ratio_frac_bits,
2456 user_tsc_khz, tsc_khz);
2458 if (ratio == 0 || ratio >= kvm_caps.max_tsc_scaling_ratio) {
2459 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2464 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2468 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2470 u32 thresh_lo, thresh_hi;
2471 int use_scaling = 0;
2473 /* tsc_khz can be zero if TSC calibration fails */
2474 if (user_tsc_khz == 0) {
2475 /* set tsc_scaling_ratio to a safe value */
2476 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_caps.default_tsc_scaling_ratio);
2480 /* Compute a scale to convert nanoseconds in TSC cycles */
2481 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2482 &vcpu->arch.virtual_tsc_shift,
2483 &vcpu->arch.virtual_tsc_mult);
2484 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2487 * Compute the variation in TSC rate which is acceptable
2488 * within the range of tolerance and decide if the
2489 * rate being applied is within that bounds of the hardware
2490 * rate. If so, no scaling or compensation need be done.
2492 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2493 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2494 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2495 pr_debug("requested TSC rate %u falls outside tolerance [%u,%u]\n",
2496 user_tsc_khz, thresh_lo, thresh_hi);
2499 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2502 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2504 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2505 vcpu->arch.virtual_tsc_mult,
2506 vcpu->arch.virtual_tsc_shift);
2507 tsc += vcpu->arch.this_tsc_write;
2511 #ifdef CONFIG_X86_64
2512 static inline int gtod_is_based_on_tsc(int mode)
2514 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2518 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2520 #ifdef CONFIG_X86_64
2522 struct kvm_arch *ka = &vcpu->kvm->arch;
2523 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2525 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2526 atomic_read(&vcpu->kvm->online_vcpus));
2529 * Once the masterclock is enabled, always perform request in
2530 * order to update it.
2532 * In order to enable masterclock, the host clocksource must be TSC
2533 * and the vcpus need to have matched TSCs. When that happens,
2534 * perform request to enable masterclock.
2536 if (ka->use_master_clock ||
2537 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2538 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2540 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2541 atomic_read(&vcpu->kvm->online_vcpus),
2542 ka->use_master_clock, gtod->clock.vclock_mode);
2547 * Multiply tsc by a fixed point number represented by ratio.
2549 * The most significant 64-N bits (mult) of ratio represent the
2550 * integral part of the fixed point number; the remaining N bits
2551 * (frac) represent the fractional part, ie. ratio represents a fixed
2552 * point number (mult + frac * 2^(-N)).
2554 * N equals to kvm_caps.tsc_scaling_ratio_frac_bits.
2556 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2558 return mul_u64_u64_shr(tsc, ratio, kvm_caps.tsc_scaling_ratio_frac_bits);
2561 u64 kvm_scale_tsc(u64 tsc, u64 ratio)
2565 if (ratio != kvm_caps.default_tsc_scaling_ratio)
2566 _tsc = __scale_tsc(ratio, tsc);
2571 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2575 tsc = kvm_scale_tsc(rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2577 return target_tsc - tsc;
2580 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2582 return vcpu->arch.l1_tsc_offset +
2583 kvm_scale_tsc(host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2585 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2587 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2591 if (l2_multiplier == kvm_caps.default_tsc_scaling_ratio)
2592 nested_offset = l1_offset;
2594 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2595 kvm_caps.tsc_scaling_ratio_frac_bits);
2597 nested_offset += l2_offset;
2598 return nested_offset;
2600 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2602 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2604 if (l2_multiplier != kvm_caps.default_tsc_scaling_ratio)
2605 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2606 kvm_caps.tsc_scaling_ratio_frac_bits);
2608 return l1_multiplier;
2610 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2612 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2614 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2615 vcpu->arch.l1_tsc_offset,
2618 vcpu->arch.l1_tsc_offset = l1_offset;
2621 * If we are here because L1 chose not to trap WRMSR to TSC then
2622 * according to the spec this should set L1's TSC (as opposed to
2623 * setting L1's offset for L2).
2625 if (is_guest_mode(vcpu))
2626 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2628 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2629 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2631 vcpu->arch.tsc_offset = l1_offset;
2633 static_call(kvm_x86_write_tsc_offset)(vcpu);
2636 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2638 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2640 /* Userspace is changing the multiplier while L2 is active */
2641 if (is_guest_mode(vcpu))
2642 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2644 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2646 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2648 if (kvm_caps.has_tsc_control)
2649 static_call(kvm_x86_write_tsc_multiplier)(vcpu);
2652 static inline bool kvm_check_tsc_unstable(void)
2654 #ifdef CONFIG_X86_64
2656 * TSC is marked unstable when we're running on Hyper-V,
2657 * 'TSC page' clocksource is good.
2659 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2662 return check_tsc_unstable();
2666 * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2667 * offset for the vcpu and tracks the TSC matching generation that the vcpu
2670 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2671 u64 ns, bool matched)
2673 struct kvm *kvm = vcpu->kvm;
2675 lockdep_assert_held(&kvm->arch.tsc_write_lock);
2678 * We also track th most recent recorded KHZ, write and time to
2679 * allow the matching interval to be extended at each write.
2681 kvm->arch.last_tsc_nsec = ns;
2682 kvm->arch.last_tsc_write = tsc;
2683 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2684 kvm->arch.last_tsc_offset = offset;
2686 vcpu->arch.last_guest_tsc = tsc;
2688 kvm_vcpu_write_tsc_offset(vcpu, offset);
2692 * We split periods of matched TSC writes into generations.
2693 * For each generation, we track the original measured
2694 * nanosecond time, offset, and write, so if TSCs are in
2695 * sync, we can match exact offset, and if not, we can match
2696 * exact software computation in compute_guest_tsc()
2698 * These values are tracked in kvm->arch.cur_xxx variables.
2700 kvm->arch.cur_tsc_generation++;
2701 kvm->arch.cur_tsc_nsec = ns;
2702 kvm->arch.cur_tsc_write = tsc;
2703 kvm->arch.cur_tsc_offset = offset;
2704 kvm->arch.nr_vcpus_matched_tsc = 0;
2705 } else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2706 kvm->arch.nr_vcpus_matched_tsc++;
2709 /* Keep track of which generation this VCPU has synchronized to */
2710 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2711 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2712 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2714 kvm_track_tsc_matching(vcpu);
2717 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2719 struct kvm *kvm = vcpu->kvm;
2720 u64 offset, ns, elapsed;
2721 unsigned long flags;
2722 bool matched = false;
2723 bool synchronizing = false;
2725 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2726 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2727 ns = get_kvmclock_base_ns();
2728 elapsed = ns - kvm->arch.last_tsc_nsec;
2730 if (vcpu->arch.virtual_tsc_khz) {
2733 * detection of vcpu initialization -- need to sync
2734 * with other vCPUs. This particularly helps to keep
2735 * kvm_clock stable after CPU hotplug
2737 synchronizing = true;
2739 u64 tsc_exp = kvm->arch.last_tsc_write +
2740 nsec_to_cycles(vcpu, elapsed);
2741 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2743 * Special case: TSC write with a small delta (1 second)
2744 * of virtual cycle time against real time is
2745 * interpreted as an attempt to synchronize the CPU.
2747 synchronizing = data < tsc_exp + tsc_hz &&
2748 data + tsc_hz > tsc_exp;
2753 * For a reliable TSC, we can match TSC offsets, and for an unstable
2754 * TSC, we add elapsed time in this computation. We could let the
2755 * compensation code attempt to catch up if we fall behind, but
2756 * it's better to try to match offsets from the beginning.
2758 if (synchronizing &&
2759 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2760 if (!kvm_check_tsc_unstable()) {
2761 offset = kvm->arch.cur_tsc_offset;
2763 u64 delta = nsec_to_cycles(vcpu, elapsed);
2765 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2770 __kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2771 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2774 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2777 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2778 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2781 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2783 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_caps.default_tsc_scaling_ratio)
2784 WARN_ON(adjustment < 0);
2785 adjustment = kvm_scale_tsc((u64) adjustment,
2786 vcpu->arch.l1_tsc_scaling_ratio);
2787 adjust_tsc_offset_guest(vcpu, adjustment);
2790 #ifdef CONFIG_X86_64
2792 static u64 read_tsc(void)
2794 u64 ret = (u64)rdtsc_ordered();
2795 u64 last = pvclock_gtod_data.clock.cycle_last;
2797 if (likely(ret >= last))
2801 * GCC likes to generate cmov here, but this branch is extremely
2802 * predictable (it's just a function of time and the likely is
2803 * very likely) and there's a data dependence, so force GCC
2804 * to generate a branch instead. I don't barrier() because
2805 * we don't actually need a barrier, and if this function
2806 * ever gets inlined it will generate worse code.
2812 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2818 switch (clock->vclock_mode) {
2819 case VDSO_CLOCKMODE_HVCLOCK:
2820 if (hv_read_tsc_page_tsc(hv_get_tsc_page(),
2821 tsc_timestamp, &tsc_pg_val)) {
2822 /* TSC page valid */
2823 *mode = VDSO_CLOCKMODE_HVCLOCK;
2824 v = (tsc_pg_val - clock->cycle_last) &
2827 /* TSC page invalid */
2828 *mode = VDSO_CLOCKMODE_NONE;
2831 case VDSO_CLOCKMODE_TSC:
2832 *mode = VDSO_CLOCKMODE_TSC;
2833 *tsc_timestamp = read_tsc();
2834 v = (*tsc_timestamp - clock->cycle_last) &
2838 *mode = VDSO_CLOCKMODE_NONE;
2841 if (*mode == VDSO_CLOCKMODE_NONE)
2842 *tsc_timestamp = v = 0;
2844 return v * clock->mult;
2847 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2849 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2855 seq = read_seqcount_begin(>od->seq);
2856 ns = gtod->raw_clock.base_cycles;
2857 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2858 ns >>= gtod->raw_clock.shift;
2859 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2860 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2866 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2868 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2874 seq = read_seqcount_begin(>od->seq);
2875 ts->tv_sec = gtod->wall_time_sec;
2876 ns = gtod->clock.base_cycles;
2877 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2878 ns >>= gtod->clock.shift;
2879 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2881 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2887 /* returns true if host is using TSC based clocksource */
2888 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2890 /* checked again under seqlock below */
2891 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2894 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2898 /* returns true if host is using TSC based clocksource */
2899 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2902 /* checked again under seqlock below */
2903 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2906 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2912 * Assuming a stable TSC across physical CPUS, and a stable TSC
2913 * across virtual CPUs, the following condition is possible.
2914 * Each numbered line represents an event visible to both
2915 * CPUs at the next numbered event.
2917 * "timespecX" represents host monotonic time. "tscX" represents
2920 * VCPU0 on CPU0 | VCPU1 on CPU1
2922 * 1. read timespec0,tsc0
2923 * 2. | timespec1 = timespec0 + N
2925 * 3. transition to guest | transition to guest
2926 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2927 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2928 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2930 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2933 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2935 * - 0 < N - M => M < N
2937 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2938 * always the case (the difference between two distinct xtime instances
2939 * might be smaller then the difference between corresponding TSC reads,
2940 * when updating guest vcpus pvclock areas).
2942 * To avoid that problem, do not allow visibility of distinct
2943 * system_timestamp/tsc_timestamp values simultaneously: use a master
2944 * copy of host monotonic time values. Update that master copy
2947 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2951 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2953 #ifdef CONFIG_X86_64
2954 struct kvm_arch *ka = &kvm->arch;
2956 bool host_tsc_clocksource, vcpus_matched;
2958 lockdep_assert_held(&kvm->arch.tsc_write_lock);
2959 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2960 atomic_read(&kvm->online_vcpus));
2963 * If the host uses TSC clock, then passthrough TSC as stable
2966 host_tsc_clocksource = kvm_get_time_and_clockread(
2967 &ka->master_kernel_ns,
2968 &ka->master_cycle_now);
2970 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2971 && !ka->backwards_tsc_observed
2972 && !ka->boot_vcpu_runs_old_kvmclock;
2974 if (ka->use_master_clock)
2975 atomic_set(&kvm_guest_has_master_clock, 1);
2977 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2978 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2983 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2985 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2988 static void __kvm_start_pvclock_update(struct kvm *kvm)
2990 raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2991 write_seqcount_begin(&kvm->arch.pvclock_sc);
2994 static void kvm_start_pvclock_update(struct kvm *kvm)
2996 kvm_make_mclock_inprogress_request(kvm);
2998 /* no guest entries from this point */
2999 __kvm_start_pvclock_update(kvm);
3002 static void kvm_end_pvclock_update(struct kvm *kvm)
3004 struct kvm_arch *ka = &kvm->arch;
3005 struct kvm_vcpu *vcpu;
3008 write_seqcount_end(&ka->pvclock_sc);
3009 raw_spin_unlock_irq(&ka->tsc_write_lock);
3010 kvm_for_each_vcpu(i, vcpu, kvm)
3011 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3013 /* guest entries allowed */
3014 kvm_for_each_vcpu(i, vcpu, kvm)
3015 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
3018 static void kvm_update_masterclock(struct kvm *kvm)
3020 kvm_hv_request_tsc_page_update(kvm);
3021 kvm_start_pvclock_update(kvm);
3022 pvclock_update_vm_gtod_copy(kvm);
3023 kvm_end_pvclock_update(kvm);
3027 * Use the kernel's tsc_khz directly if the TSC is constant, otherwise use KVM's
3028 * per-CPU value (which may be zero if a CPU is going offline). Note, tsc_khz
3029 * can change during boot even if the TSC is constant, as it's possible for KVM
3030 * to be loaded before TSC calibration completes. Ideally, KVM would get a
3031 * notification when calibration completes, but practically speaking calibration
3032 * will complete before userspace is alive enough to create VMs.
3034 static unsigned long get_cpu_tsc_khz(void)
3036 if (static_cpu_has(X86_FEATURE_CONSTANT_TSC))
3039 return __this_cpu_read(cpu_tsc_khz);
3042 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc. */
3043 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
3045 struct kvm_arch *ka = &kvm->arch;
3046 struct pvclock_vcpu_time_info hv_clock;
3048 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
3052 if (ka->use_master_clock &&
3053 (static_cpu_has(X86_FEATURE_CONSTANT_TSC) || __this_cpu_read(cpu_tsc_khz))) {
3054 #ifdef CONFIG_X86_64
3055 struct timespec64 ts;
3057 if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
3058 data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
3059 data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
3062 data->host_tsc = rdtsc();
3064 data->flags |= KVM_CLOCK_TSC_STABLE;
3065 hv_clock.tsc_timestamp = ka->master_cycle_now;
3066 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
3067 kvm_get_time_scale(NSEC_PER_SEC, get_cpu_tsc_khz() * 1000LL,
3068 &hv_clock.tsc_shift,
3069 &hv_clock.tsc_to_system_mul);
3070 data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
3072 data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
3078 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
3080 struct kvm_arch *ka = &kvm->arch;
3084 seq = read_seqcount_begin(&ka->pvclock_sc);
3085 __get_kvmclock(kvm, data);
3086 } while (read_seqcount_retry(&ka->pvclock_sc, seq));
3089 u64 get_kvmclock_ns(struct kvm *kvm)
3091 struct kvm_clock_data data;
3093 get_kvmclock(kvm, &data);
3097 static void kvm_setup_guest_pvclock(struct kvm_vcpu *v,
3098 struct gfn_to_pfn_cache *gpc,
3099 unsigned int offset)
3101 struct kvm_vcpu_arch *vcpu = &v->arch;
3102 struct pvclock_vcpu_time_info *guest_hv_clock;
3103 unsigned long flags;
3105 read_lock_irqsave(&gpc->lock, flags);
3106 while (!kvm_gpc_check(gpc, offset + sizeof(*guest_hv_clock))) {
3107 read_unlock_irqrestore(&gpc->lock, flags);
3109 if (kvm_gpc_refresh(gpc, offset + sizeof(*guest_hv_clock)))
3112 read_lock_irqsave(&gpc->lock, flags);
3115 guest_hv_clock = (void *)(gpc->khva + offset);
3118 * This VCPU is paused, but it's legal for a guest to read another
3119 * VCPU's kvmclock, so we really have to follow the specification where
3120 * it says that version is odd if data is being modified, and even after
3124 guest_hv_clock->version = vcpu->hv_clock.version = (guest_hv_clock->version + 1) | 1;
3127 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
3128 vcpu->hv_clock.flags |= (guest_hv_clock->flags & PVCLOCK_GUEST_STOPPED);
3130 if (vcpu->pvclock_set_guest_stopped_request) {
3131 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
3132 vcpu->pvclock_set_guest_stopped_request = false;
3135 memcpy(guest_hv_clock, &vcpu->hv_clock, sizeof(*guest_hv_clock));
3138 guest_hv_clock->version = ++vcpu->hv_clock.version;
3140 mark_page_dirty_in_slot(v->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT);
3141 read_unlock_irqrestore(&gpc->lock, flags);
3143 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
3146 static int kvm_guest_time_update(struct kvm_vcpu *v)
3148 unsigned long flags, tgt_tsc_khz;
3150 struct kvm_vcpu_arch *vcpu = &v->arch;
3151 struct kvm_arch *ka = &v->kvm->arch;
3153 u64 tsc_timestamp, host_tsc;
3155 bool use_master_clock;
3161 * If the host uses TSC clock, then passthrough TSC as stable
3165 seq = read_seqcount_begin(&ka->pvclock_sc);
3166 use_master_clock = ka->use_master_clock;
3167 if (use_master_clock) {
3168 host_tsc = ka->master_cycle_now;
3169 kernel_ns = ka->master_kernel_ns;
3171 } while (read_seqcount_retry(&ka->pvclock_sc, seq));
3173 /* Keep irq disabled to prevent changes to the clock */
3174 local_irq_save(flags);
3175 tgt_tsc_khz = get_cpu_tsc_khz();
3176 if (unlikely(tgt_tsc_khz == 0)) {
3177 local_irq_restore(flags);
3178 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3181 if (!use_master_clock) {
3183 kernel_ns = get_kvmclock_base_ns();
3186 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
3189 * We may have to catch up the TSC to match elapsed wall clock
3190 * time for two reasons, even if kvmclock is used.
3191 * 1) CPU could have been running below the maximum TSC rate
3192 * 2) Broken TSC compensation resets the base at each VCPU
3193 * entry to avoid unknown leaps of TSC even when running
3194 * again on the same CPU. This may cause apparent elapsed
3195 * time to disappear, and the guest to stand still or run
3198 if (vcpu->tsc_catchup) {
3199 u64 tsc = compute_guest_tsc(v, kernel_ns);
3200 if (tsc > tsc_timestamp) {
3201 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3202 tsc_timestamp = tsc;
3206 local_irq_restore(flags);
3208 /* With all the info we got, fill in the values */
3210 if (kvm_caps.has_tsc_control)
3211 tgt_tsc_khz = kvm_scale_tsc(tgt_tsc_khz,
3212 v->arch.l1_tsc_scaling_ratio);
3214 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3215 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3216 &vcpu->hv_clock.tsc_shift,
3217 &vcpu->hv_clock.tsc_to_system_mul);
3218 vcpu->hw_tsc_khz = tgt_tsc_khz;
3219 kvm_xen_update_tsc_info(v);
3222 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3223 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3224 vcpu->last_guest_tsc = tsc_timestamp;
3226 /* If the host uses TSC clocksource, then it is stable */
3228 if (use_master_clock)
3229 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3231 vcpu->hv_clock.flags = pvclock_flags;
3233 if (vcpu->pv_time.active)
3234 kvm_setup_guest_pvclock(v, &vcpu->pv_time, 0);
3235 if (vcpu->xen.vcpu_info_cache.active)
3236 kvm_setup_guest_pvclock(v, &vcpu->xen.vcpu_info_cache,
3237 offsetof(struct compat_vcpu_info, time));
3238 if (vcpu->xen.vcpu_time_info_cache.active)
3239 kvm_setup_guest_pvclock(v, &vcpu->xen.vcpu_time_info_cache, 0);
3240 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3245 * kvmclock updates which are isolated to a given vcpu, such as
3246 * vcpu->cpu migration, should not allow system_timestamp from
3247 * the rest of the vcpus to remain static. Otherwise ntp frequency
3248 * correction applies to one vcpu's system_timestamp but not
3251 * So in those cases, request a kvmclock update for all vcpus.
3252 * We need to rate-limit these requests though, as they can
3253 * considerably slow guests that have a large number of vcpus.
3254 * The time for a remote vcpu to update its kvmclock is bound
3255 * by the delay we use to rate-limit the updates.
3258 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3260 static void kvmclock_update_fn(struct work_struct *work)
3263 struct delayed_work *dwork = to_delayed_work(work);
3264 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3265 kvmclock_update_work);
3266 struct kvm *kvm = container_of(ka, struct kvm, arch);
3267 struct kvm_vcpu *vcpu;
3269 kvm_for_each_vcpu(i, vcpu, kvm) {
3270 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3271 kvm_vcpu_kick(vcpu);
3275 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3277 struct kvm *kvm = v->kvm;
3279 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3280 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3281 KVMCLOCK_UPDATE_DELAY);
3284 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3286 static void kvmclock_sync_fn(struct work_struct *work)
3288 struct delayed_work *dwork = to_delayed_work(work);
3289 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3290 kvmclock_sync_work);
3291 struct kvm *kvm = container_of(ka, struct kvm, arch);
3293 if (!kvmclock_periodic_sync)
3296 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3297 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3298 KVMCLOCK_SYNC_PERIOD);
3301 /* These helpers are safe iff @msr is known to be an MCx bank MSR. */
3302 static bool is_mci_control_msr(u32 msr)
3304 return (msr & 3) == 0;
3306 static bool is_mci_status_msr(u32 msr)
3308 return (msr & 3) == 1;
3312 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3314 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3316 /* McStatusWrEn enabled? */
3317 if (guest_cpuid_is_amd_or_hygon(vcpu))
3318 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3323 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3325 u64 mcg_cap = vcpu->arch.mcg_cap;
3326 unsigned bank_num = mcg_cap & 0xff;
3327 u32 msr = msr_info->index;
3328 u64 data = msr_info->data;
3329 u32 offset, last_msr;
3332 case MSR_IA32_MCG_STATUS:
3333 vcpu->arch.mcg_status = data;
3335 case MSR_IA32_MCG_CTL:
3336 if (!(mcg_cap & MCG_CTL_P) &&
3337 (data || !msr_info->host_initiated))
3339 if (data != 0 && data != ~(u64)0)
3341 vcpu->arch.mcg_ctl = data;
3343 case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3344 last_msr = MSR_IA32_MCx_CTL2(bank_num) - 1;
3348 if (!(mcg_cap & MCG_CMCI_P) && (data || !msr_info->host_initiated))
3350 /* An attempt to write a 1 to a reserved bit raises #GP */
3351 if (data & ~(MCI_CTL2_CMCI_EN | MCI_CTL2_CMCI_THRESHOLD_MASK))
3353 offset = array_index_nospec(msr - MSR_IA32_MC0_CTL2,
3354 last_msr + 1 - MSR_IA32_MC0_CTL2);
3355 vcpu->arch.mci_ctl2_banks[offset] = data;
3357 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3358 last_msr = MSR_IA32_MCx_CTL(bank_num) - 1;
3363 * Only 0 or all 1s can be written to IA32_MCi_CTL, all other
3364 * values are architecturally undefined. But, some Linux
3365 * kernels clear bit 10 in bank 4 to workaround a BIOS/GART TLB
3366 * issue on AMD K8s, allow bit 10 to be clear when setting all
3367 * other bits in order to avoid an uncaught #GP in the guest.
3369 * UNIXWARE clears bit 0 of MC1_CTL to ignore correctable,
3370 * single-bit ECC data errors.
3372 if (is_mci_control_msr(msr) &&
3373 data != 0 && (data | (1 << 10) | 1) != ~(u64)0)
3377 * All CPUs allow writing 0 to MCi_STATUS MSRs to clear the MSR.
3378 * AMD-based CPUs allow non-zero values, but if and only if
3379 * HWCR[McStatusWrEn] is set.
3381 if (!msr_info->host_initiated && is_mci_status_msr(msr) &&
3382 data != 0 && !can_set_mci_status(vcpu))
3385 offset = array_index_nospec(msr - MSR_IA32_MC0_CTL,
3386 last_msr + 1 - MSR_IA32_MC0_CTL);
3387 vcpu->arch.mce_banks[offset] = data;
3395 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3397 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3399 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3402 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3404 gpa_t gpa = data & ~0x3f;
3406 /* Bits 4:5 are reserved, Should be zero */
3410 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3411 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3414 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3415 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3418 if (!lapic_in_kernel(vcpu))
3419 return data ? 1 : 0;
3421 vcpu->arch.apf.msr_en_val = data;
3423 if (!kvm_pv_async_pf_enabled(vcpu)) {
3424 kvm_clear_async_pf_completion_queue(vcpu);
3425 kvm_async_pf_hash_reset(vcpu);
3429 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3433 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3434 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3436 kvm_async_pf_wakeup_all(vcpu);
3441 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3443 /* Bits 8-63 are reserved */
3447 if (!lapic_in_kernel(vcpu))
3450 vcpu->arch.apf.msr_int_val = data;
3452 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3457 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3459 kvm_gpc_deactivate(&vcpu->arch.pv_time);
3460 vcpu->arch.time = 0;
3463 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3465 ++vcpu->stat.tlb_flush;
3466 static_call(kvm_x86_flush_tlb_all)(vcpu);
3468 /* Flushing all ASIDs flushes the current ASID... */
3469 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
3472 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3474 ++vcpu->stat.tlb_flush;
3478 * A TLB flush on behalf of the guest is equivalent to
3479 * INVPCID(all), toggling CR4.PGE, etc., which requires
3480 * a forced sync of the shadow page tables. Ensure all the
3481 * roots are synced and the guest TLB in hardware is clean.
3483 kvm_mmu_sync_roots(vcpu);
3484 kvm_mmu_sync_prev_roots(vcpu);
3487 static_call(kvm_x86_flush_tlb_guest)(vcpu);
3490 * Flushing all "guest" TLB is always a superset of Hyper-V's fine
3493 kvm_hv_vcpu_purge_flush_tlb(vcpu);
3497 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
3499 ++vcpu->stat.tlb_flush;
3500 static_call(kvm_x86_flush_tlb_current)(vcpu);
3504 * Service "local" TLB flush requests, which are specific to the current MMU
3505 * context. In addition to the generic event handling in vcpu_enter_guest(),
3506 * TLB flushes that are targeted at an MMU context also need to be serviced
3507 * prior before nested VM-Enter/VM-Exit.
3509 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu)
3511 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3512 kvm_vcpu_flush_tlb_current(vcpu);
3514 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
3515 kvm_vcpu_flush_tlb_guest(vcpu);
3517 EXPORT_SYMBOL_GPL(kvm_service_local_tlb_flush_requests);
3519 static void record_steal_time(struct kvm_vcpu *vcpu)
3521 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3522 struct kvm_steal_time __user *st;
3523 struct kvm_memslots *slots;
3524 gpa_t gpa = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3528 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3529 kvm_xen_runstate_set_running(vcpu);
3533 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3536 if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3539 slots = kvm_memslots(vcpu->kvm);
3541 if (unlikely(slots->generation != ghc->generation ||
3543 kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3544 /* We rely on the fact that it fits in a single page. */
3545 BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3547 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gpa, sizeof(*st)) ||
3548 kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3552 st = (struct kvm_steal_time __user *)ghc->hva;
3554 * Doing a TLB flush here, on the guest's behalf, can avoid
3557 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3558 u8 st_preempted = 0;
3561 if (!user_access_begin(st, sizeof(*st)))
3564 asm volatile("1: xchgb %0, %2\n"
3567 _ASM_EXTABLE_UA(1b, 2b)
3568 : "+q" (st_preempted),
3570 "+m" (st->preempted));
3576 vcpu->arch.st.preempted = 0;
3578 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3579 st_preempted & KVM_VCPU_FLUSH_TLB);
3580 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3581 kvm_vcpu_flush_tlb_guest(vcpu);
3583 if (!user_access_begin(st, sizeof(*st)))
3586 if (!user_access_begin(st, sizeof(*st)))
3589 unsafe_put_user(0, &st->preempted, out);
3590 vcpu->arch.st.preempted = 0;
3593 unsafe_get_user(version, &st->version, out);
3595 version += 1; /* first time write, random junk */
3598 unsafe_put_user(version, &st->version, out);
3602 unsafe_get_user(steal, &st->steal, out);
3603 steal += current->sched_info.run_delay -
3604 vcpu->arch.st.last_steal;
3605 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3606 unsafe_put_user(steal, &st->steal, out);
3609 unsafe_put_user(version, &st->version, out);
3614 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3617 static bool kvm_is_msr_to_save(u32 msr_index)
3621 for (i = 0; i < num_msrs_to_save; i++) {
3622 if (msrs_to_save[i] == msr_index)
3629 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3631 u32 msr = msr_info->index;
3632 u64 data = msr_info->data;
3634 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3635 return kvm_xen_write_hypercall_page(vcpu, data);
3638 case MSR_AMD64_NB_CFG:
3639 case MSR_IA32_UCODE_WRITE:
3640 case MSR_VM_HSAVE_PA:
3641 case MSR_AMD64_PATCH_LOADER:
3642 case MSR_AMD64_BU_CFG2:
3643 case MSR_AMD64_DC_CFG:
3644 case MSR_F15H_EX_CFG:
3647 case MSR_IA32_UCODE_REV:
3648 if (msr_info->host_initiated)
3649 vcpu->arch.microcode_version = data;
3651 case MSR_IA32_ARCH_CAPABILITIES:
3652 if (!msr_info->host_initiated)
3654 vcpu->arch.arch_capabilities = data;
3656 case MSR_IA32_PERF_CAPABILITIES:
3657 if (!msr_info->host_initiated)
3659 if (data & ~kvm_caps.supported_perf_cap)
3663 * Note, this is not just a performance optimization! KVM
3664 * disallows changing feature MSRs after the vCPU has run; PMU
3665 * refresh will bug the VM if called after the vCPU has run.
3667 if (vcpu->arch.perf_capabilities == data)
3670 vcpu->arch.perf_capabilities = data;
3671 kvm_pmu_refresh(vcpu);
3673 case MSR_IA32_PRED_CMD:
3674 if (!msr_info->host_initiated && !guest_has_pred_cmd_msr(vcpu))
3677 if (!boot_cpu_has(X86_FEATURE_IBPB) || (data & ~PRED_CMD_IBPB))
3682 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
3684 case MSR_IA32_FLUSH_CMD:
3685 if (!msr_info->host_initiated &&
3686 !guest_cpuid_has(vcpu, X86_FEATURE_FLUSH_L1D))
3689 if (!boot_cpu_has(X86_FEATURE_FLUSH_L1D) || (data & ~L1D_FLUSH))
3694 wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
3697 return set_efer(vcpu, msr_info);
3699 data &= ~(u64)0x40; /* ignore flush filter disable */
3700 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3701 data &= ~(u64)0x8; /* ignore TLB cache disable */
3703 /* Handle McStatusWrEn */
3704 if (data == BIT_ULL(18)) {
3705 vcpu->arch.msr_hwcr = data;
3706 } else if (data != 0) {
3707 kvm_pr_unimpl_wrmsr(vcpu, msr, data);
3711 case MSR_FAM10H_MMIO_CONF_BASE:
3713 kvm_pr_unimpl_wrmsr(vcpu, msr, data);
3717 case MSR_IA32_CR_PAT:
3718 if (!kvm_pat_valid(data))
3721 vcpu->arch.pat = data;
3723 case MTRRphysBase_MSR(0) ... MSR_MTRRfix4K_F8000:
3724 case MSR_MTRRdefType:
3725 return kvm_mtrr_set_msr(vcpu, msr, data);
3726 case MSR_IA32_APICBASE:
3727 return kvm_set_apic_base(vcpu, msr_info);
3728 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3729 return kvm_x2apic_msr_write(vcpu, msr, data);
3730 case MSR_IA32_TSC_DEADLINE:
3731 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3733 case MSR_IA32_TSC_ADJUST:
3734 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3735 if (!msr_info->host_initiated) {
3736 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3737 adjust_tsc_offset_guest(vcpu, adj);
3738 /* Before back to guest, tsc_timestamp must be adjusted
3739 * as well, otherwise guest's percpu pvclock time could jump.
3741 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3743 vcpu->arch.ia32_tsc_adjust_msr = data;
3746 case MSR_IA32_MISC_ENABLE: {
3747 u64 old_val = vcpu->arch.ia32_misc_enable_msr;
3749 if (!msr_info->host_initiated) {
3751 if ((old_val ^ data) & MSR_IA32_MISC_ENABLE_PMU_RO_MASK)
3754 /* R bits, i.e. writes are ignored, but don't fault. */
3755 data = data & ~MSR_IA32_MISC_ENABLE_EMON;
3756 data |= old_val & MSR_IA32_MISC_ENABLE_EMON;
3759 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3760 ((old_val ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3761 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3763 vcpu->arch.ia32_misc_enable_msr = data;
3764 kvm_update_cpuid_runtime(vcpu);
3766 vcpu->arch.ia32_misc_enable_msr = data;
3770 case MSR_IA32_SMBASE:
3771 if (!IS_ENABLED(CONFIG_KVM_SMM) || !msr_info->host_initiated)
3773 vcpu->arch.smbase = data;
3775 case MSR_IA32_POWER_CTL:
3776 vcpu->arch.msr_ia32_power_ctl = data;
3779 if (msr_info->host_initiated) {
3780 kvm_synchronize_tsc(vcpu, data);
3782 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3783 adjust_tsc_offset_guest(vcpu, adj);
3784 vcpu->arch.ia32_tsc_adjust_msr += adj;
3788 if (!msr_info->host_initiated &&
3789 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3792 * KVM supports exposing PT to the guest, but does not support
3793 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3794 * XSAVES/XRSTORS to save/restore PT MSRs.
3796 if (data & ~kvm_caps.supported_xss)
3798 vcpu->arch.ia32_xss = data;
3799 kvm_update_cpuid_runtime(vcpu);
3802 if (!msr_info->host_initiated)
3804 vcpu->arch.smi_count = data;
3806 case MSR_KVM_WALL_CLOCK_NEW:
3807 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3810 vcpu->kvm->arch.wall_clock = data;
3811 kvm_write_wall_clock(vcpu->kvm, data, 0);
3813 case MSR_KVM_WALL_CLOCK:
3814 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3817 vcpu->kvm->arch.wall_clock = data;
3818 kvm_write_wall_clock(vcpu->kvm, data, 0);
3820 case MSR_KVM_SYSTEM_TIME_NEW:
3821 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3824 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3826 case MSR_KVM_SYSTEM_TIME:
3827 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3830 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3832 case MSR_KVM_ASYNC_PF_EN:
3833 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3836 if (kvm_pv_enable_async_pf(vcpu, data))
3839 case MSR_KVM_ASYNC_PF_INT:
3840 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3843 if (kvm_pv_enable_async_pf_int(vcpu, data))
3846 case MSR_KVM_ASYNC_PF_ACK:
3847 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3850 vcpu->arch.apf.pageready_pending = false;
3851 kvm_check_async_pf_completion(vcpu);
3854 case MSR_KVM_STEAL_TIME:
3855 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3858 if (unlikely(!sched_info_on()))
3861 if (data & KVM_STEAL_RESERVED_MASK)
3864 vcpu->arch.st.msr_val = data;
3866 if (!(data & KVM_MSR_ENABLED))
3869 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3872 case MSR_KVM_PV_EOI_EN:
3873 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3876 if (kvm_lapic_set_pv_eoi(vcpu, data, sizeof(u8)))
3880 case MSR_KVM_POLL_CONTROL:
3881 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3884 /* only enable bit supported */
3885 if (data & (-1ULL << 1))
3888 vcpu->arch.msr_kvm_poll_control = data;
3891 case MSR_IA32_MCG_CTL:
3892 case MSR_IA32_MCG_STATUS:
3893 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3894 case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3895 return set_msr_mce(vcpu, msr_info);
3897 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3898 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3899 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3900 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3901 if (kvm_pmu_is_valid_msr(vcpu, msr))
3902 return kvm_pmu_set_msr(vcpu, msr_info);
3905 kvm_pr_unimpl_wrmsr(vcpu, msr, data);
3907 case MSR_K7_CLK_CTL:
3909 * Ignore all writes to this no longer documented MSR.
3910 * Writes are only relevant for old K7 processors,
3911 * all pre-dating SVM, but a recommended workaround from
3912 * AMD for these chips. It is possible to specify the
3913 * affected processor models on the command line, hence
3914 * the need to ignore the workaround.
3917 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3918 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3919 case HV_X64_MSR_SYNDBG_OPTIONS:
3920 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3921 case HV_X64_MSR_CRASH_CTL:
3922 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3923 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3924 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3925 case HV_X64_MSR_TSC_EMULATION_STATUS:
3926 case HV_X64_MSR_TSC_INVARIANT_CONTROL:
3927 return kvm_hv_set_msr_common(vcpu, msr, data,
3928 msr_info->host_initiated);
3929 case MSR_IA32_BBL_CR_CTL3:
3930 /* Drop writes to this legacy MSR -- see rdmsr
3931 * counterpart for further detail.
3933 kvm_pr_unimpl_wrmsr(vcpu, msr, data);
3935 case MSR_AMD64_OSVW_ID_LENGTH:
3936 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3938 vcpu->arch.osvw.length = data;
3940 case MSR_AMD64_OSVW_STATUS:
3941 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3943 vcpu->arch.osvw.status = data;
3945 case MSR_PLATFORM_INFO:
3946 if (!msr_info->host_initiated ||
3947 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3948 cpuid_fault_enabled(vcpu)))
3950 vcpu->arch.msr_platform_info = data;
3952 case MSR_MISC_FEATURES_ENABLES:
3953 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3954 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3955 !supports_cpuid_fault(vcpu)))
3957 vcpu->arch.msr_misc_features_enables = data;
3959 #ifdef CONFIG_X86_64
3961 if (!msr_info->host_initiated &&
3962 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3965 if (data & ~kvm_guest_supported_xfd(vcpu))
3968 fpu_update_guest_xfd(&vcpu->arch.guest_fpu, data);
3970 case MSR_IA32_XFD_ERR:
3971 if (!msr_info->host_initiated &&
3972 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3975 if (data & ~kvm_guest_supported_xfd(vcpu))
3978 vcpu->arch.guest_fpu.xfd_err = data;
3982 if (kvm_pmu_is_valid_msr(vcpu, msr))
3983 return kvm_pmu_set_msr(vcpu, msr_info);
3986 * Userspace is allowed to write '0' to MSRs that KVM reports
3987 * as to-be-saved, even if an MSRs isn't fully supported.
3989 if (msr_info->host_initiated && !data &&
3990 kvm_is_msr_to_save(msr))
3993 return KVM_MSR_RET_INVALID;
3997 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3999 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
4002 u64 mcg_cap = vcpu->arch.mcg_cap;
4003 unsigned bank_num = mcg_cap & 0xff;
4004 u32 offset, last_msr;
4007 case MSR_IA32_P5_MC_ADDR:
4008 case MSR_IA32_P5_MC_TYPE:
4011 case MSR_IA32_MCG_CAP:
4012 data = vcpu->arch.mcg_cap;
4014 case MSR_IA32_MCG_CTL:
4015 if (!(mcg_cap & MCG_CTL_P) && !host)
4017 data = vcpu->arch.mcg_ctl;
4019 case MSR_IA32_MCG_STATUS:
4020 data = vcpu->arch.mcg_status;
4022 case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
4023 last_msr = MSR_IA32_MCx_CTL2(bank_num) - 1;
4027 if (!(mcg_cap & MCG_CMCI_P) && !host)
4029 offset = array_index_nospec(msr - MSR_IA32_MC0_CTL2,
4030 last_msr + 1 - MSR_IA32_MC0_CTL2);
4031 data = vcpu->arch.mci_ctl2_banks[offset];
4033 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
4034 last_msr = MSR_IA32_MCx_CTL(bank_num) - 1;
4038 offset = array_index_nospec(msr - MSR_IA32_MC0_CTL,
4039 last_msr + 1 - MSR_IA32_MC0_CTL);
4040 data = vcpu->arch.mce_banks[offset];
4049 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
4051 switch (msr_info->index) {
4052 case MSR_IA32_PLATFORM_ID:
4053 case MSR_IA32_EBL_CR_POWERON:
4054 case MSR_IA32_LASTBRANCHFROMIP:
4055 case MSR_IA32_LASTBRANCHTOIP:
4056 case MSR_IA32_LASTINTFROMIP:
4057 case MSR_IA32_LASTINTTOIP:
4058 case MSR_AMD64_SYSCFG:
4059 case MSR_K8_TSEG_ADDR:
4060 case MSR_K8_TSEG_MASK:
4061 case MSR_VM_HSAVE_PA:
4062 case MSR_K8_INT_PENDING_MSG:
4063 case MSR_AMD64_NB_CFG:
4064 case MSR_FAM10H_MMIO_CONF_BASE:
4065 case MSR_AMD64_BU_CFG2:
4066 case MSR_IA32_PERF_CTL:
4067 case MSR_AMD64_DC_CFG:
4068 case MSR_F15H_EX_CFG:
4070 * Intel Sandy Bridge CPUs must support the RAPL (running average power
4071 * limit) MSRs. Just return 0, as we do not want to expose the host
4072 * data here. Do not conditionalize this on CPUID, as KVM does not do
4073 * so for existing CPU-specific MSRs.
4075 case MSR_RAPL_POWER_UNIT:
4076 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
4077 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
4078 case MSR_PKG_ENERGY_STATUS: /* Total package */
4079 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
4082 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
4083 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
4084 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
4085 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
4086 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
4087 return kvm_pmu_get_msr(vcpu, msr_info);
4090 case MSR_IA32_UCODE_REV:
4091 msr_info->data = vcpu->arch.microcode_version;
4093 case MSR_IA32_ARCH_CAPABILITIES:
4094 if (!msr_info->host_initiated &&
4095 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
4097 msr_info->data = vcpu->arch.arch_capabilities;
4099 case MSR_IA32_PERF_CAPABILITIES:
4100 if (!msr_info->host_initiated &&
4101 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
4103 msr_info->data = vcpu->arch.perf_capabilities;
4105 case MSR_IA32_POWER_CTL:
4106 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
4108 case MSR_IA32_TSC: {
4110 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
4111 * even when not intercepted. AMD manual doesn't explicitly
4112 * state this but appears to behave the same.
4114 * On userspace reads and writes, however, we unconditionally
4115 * return L1's TSC value to ensure backwards-compatible
4116 * behavior for migration.
4120 if (msr_info->host_initiated) {
4121 offset = vcpu->arch.l1_tsc_offset;
4122 ratio = vcpu->arch.l1_tsc_scaling_ratio;
4124 offset = vcpu->arch.tsc_offset;
4125 ratio = vcpu->arch.tsc_scaling_ratio;
4128 msr_info->data = kvm_scale_tsc(rdtsc(), ratio) + offset;
4131 case MSR_IA32_CR_PAT:
4132 msr_info->data = vcpu->arch.pat;
4135 case MTRRphysBase_MSR(0) ... MSR_MTRRfix4K_F8000:
4136 case MSR_MTRRdefType:
4137 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
4138 case 0xcd: /* fsb frequency */
4142 * MSR_EBC_FREQUENCY_ID
4143 * Conservative value valid for even the basic CPU models.
4144 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
4145 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
4146 * and 266MHz for model 3, or 4. Set Core Clock
4147 * Frequency to System Bus Frequency Ratio to 1 (bits
4148 * 31:24) even though these are only valid for CPU
4149 * models > 2, however guests may end up dividing or
4150 * multiplying by zero otherwise.
4152 case MSR_EBC_FREQUENCY_ID:
4153 msr_info->data = 1 << 24;
4155 case MSR_IA32_APICBASE:
4156 msr_info->data = kvm_get_apic_base(vcpu);
4158 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
4159 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
4160 case MSR_IA32_TSC_DEADLINE:
4161 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
4163 case MSR_IA32_TSC_ADJUST:
4164 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
4166 case MSR_IA32_MISC_ENABLE:
4167 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
4169 case MSR_IA32_SMBASE:
4170 if (!IS_ENABLED(CONFIG_KVM_SMM) || !msr_info->host_initiated)
4172 msr_info->data = vcpu->arch.smbase;
4175 msr_info->data = vcpu->arch.smi_count;
4177 case MSR_IA32_PERF_STATUS:
4178 /* TSC increment by tick */
4179 msr_info->data = 1000ULL;
4180 /* CPU multiplier */
4181 msr_info->data |= (((uint64_t)4ULL) << 40);
4184 msr_info->data = vcpu->arch.efer;
4186 case MSR_KVM_WALL_CLOCK:
4187 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
4190 msr_info->data = vcpu->kvm->arch.wall_clock;
4192 case MSR_KVM_WALL_CLOCK_NEW:
4193 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
4196 msr_info->data = vcpu->kvm->arch.wall_clock;
4198 case MSR_KVM_SYSTEM_TIME:
4199 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
4202 msr_info->data = vcpu->arch.time;
4204 case MSR_KVM_SYSTEM_TIME_NEW:
4205 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
4208 msr_info->data = vcpu->arch.time;
4210 case MSR_KVM_ASYNC_PF_EN:
4211 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
4214 msr_info->data = vcpu->arch.apf.msr_en_val;
4216 case MSR_KVM_ASYNC_PF_INT:
4217 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
4220 msr_info->data = vcpu->arch.apf.msr_int_val;
4222 case MSR_KVM_ASYNC_PF_ACK:
4223 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
4228 case MSR_KVM_STEAL_TIME:
4229 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
4232 msr_info->data = vcpu->arch.st.msr_val;
4234 case MSR_KVM_PV_EOI_EN:
4235 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
4238 msr_info->data = vcpu->arch.pv_eoi.msr_val;
4240 case MSR_KVM_POLL_CONTROL:
4241 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
4244 msr_info->data = vcpu->arch.msr_kvm_poll_control;
4246 case MSR_IA32_P5_MC_ADDR:
4247 case MSR_IA32_P5_MC_TYPE:
4248 case MSR_IA32_MCG_CAP:
4249 case MSR_IA32_MCG_CTL:
4250 case MSR_IA32_MCG_STATUS:
4251 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
4252 case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
4253 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
4254 msr_info->host_initiated);
4256 if (!msr_info->host_initiated &&
4257 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
4259 msr_info->data = vcpu->arch.ia32_xss;
4261 case MSR_K7_CLK_CTL:
4263 * Provide expected ramp-up count for K7. All other
4264 * are set to zero, indicating minimum divisors for
4267 * This prevents guest kernels on AMD host with CPU
4268 * type 6, model 8 and higher from exploding due to
4269 * the rdmsr failing.
4271 msr_info->data = 0x20000000;
4273 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
4274 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
4275 case HV_X64_MSR_SYNDBG_OPTIONS:
4276 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
4277 case HV_X64_MSR_CRASH_CTL:
4278 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
4279 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
4280 case HV_X64_MSR_TSC_EMULATION_CONTROL:
4281 case HV_X64_MSR_TSC_EMULATION_STATUS:
4282 case HV_X64_MSR_TSC_INVARIANT_CONTROL:
4283 return kvm_hv_get_msr_common(vcpu,
4284 msr_info->index, &msr_info->data,
4285 msr_info->host_initiated);
4286 case MSR_IA32_BBL_CR_CTL3:
4287 /* This legacy MSR exists but isn't fully documented in current
4288 * silicon. It is however accessed by winxp in very narrow
4289 * scenarios where it sets bit #19, itself documented as
4290 * a "reserved" bit. Best effort attempt to source coherent
4291 * read data here should the balance of the register be
4292 * interpreted by the guest:
4294 * L2 cache control register 3: 64GB range, 256KB size,
4295 * enabled, latency 0x1, configured
4297 msr_info->data = 0xbe702111;
4299 case MSR_AMD64_OSVW_ID_LENGTH:
4300 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4302 msr_info->data = vcpu->arch.osvw.length;
4304 case MSR_AMD64_OSVW_STATUS:
4305 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4307 msr_info->data = vcpu->arch.osvw.status;
4309 case MSR_PLATFORM_INFO:
4310 if (!msr_info->host_initiated &&
4311 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
4313 msr_info->data = vcpu->arch.msr_platform_info;
4315 case MSR_MISC_FEATURES_ENABLES:
4316 msr_info->data = vcpu->arch.msr_misc_features_enables;
4319 msr_info->data = vcpu->arch.msr_hwcr;
4321 #ifdef CONFIG_X86_64
4323 if (!msr_info->host_initiated &&
4324 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4327 msr_info->data = vcpu->arch.guest_fpu.fpstate->xfd;
4329 case MSR_IA32_XFD_ERR:
4330 if (!msr_info->host_initiated &&
4331 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4334 msr_info->data = vcpu->arch.guest_fpu.xfd_err;
4338 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
4339 return kvm_pmu_get_msr(vcpu, msr_info);
4342 * Userspace is allowed to read MSRs that KVM reports as
4343 * to-be-saved, even if an MSR isn't fully supported.
4345 if (msr_info->host_initiated &&
4346 kvm_is_msr_to_save(msr_info->index)) {
4351 return KVM_MSR_RET_INVALID;
4355 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
4358 * Read or write a bunch of msrs. All parameters are kernel addresses.
4360 * @return number of msrs set successfully.
4362 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
4363 struct kvm_msr_entry *entries,
4364 int (*do_msr)(struct kvm_vcpu *vcpu,
4365 unsigned index, u64 *data))
4369 for (i = 0; i < msrs->nmsrs; ++i)
4370 if (do_msr(vcpu, entries[i].index, &entries[i].data))
4377 * Read or write a bunch of msrs. Parameters are user addresses.
4379 * @return number of msrs set successfully.
4381 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
4382 int (*do_msr)(struct kvm_vcpu *vcpu,
4383 unsigned index, u64 *data),
4386 struct kvm_msrs msrs;
4387 struct kvm_msr_entry *entries;
4392 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4396 if (msrs.nmsrs >= MAX_IO_MSRS)
4399 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4400 entries = memdup_user(user_msrs->entries, size);
4401 if (IS_ERR(entries)) {
4402 r = PTR_ERR(entries);
4406 r = __msr_io(vcpu, &msrs, entries, do_msr);
4408 if (writeback && copy_to_user(user_msrs->entries, entries, size))
4416 static inline bool kvm_can_mwait_in_guest(void)
4418 return boot_cpu_has(X86_FEATURE_MWAIT) &&
4419 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
4420 boot_cpu_has(X86_FEATURE_ARAT);
4423 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4424 struct kvm_cpuid2 __user *cpuid_arg)
4426 struct kvm_cpuid2 cpuid;
4430 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4433 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4438 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4444 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4449 case KVM_CAP_IRQCHIP:
4451 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4452 case KVM_CAP_SET_TSS_ADDR:
4453 case KVM_CAP_EXT_CPUID:
4454 case KVM_CAP_EXT_EMUL_CPUID:
4455 case KVM_CAP_CLOCKSOURCE:
4457 case KVM_CAP_NOP_IO_DELAY:
4458 case KVM_CAP_MP_STATE:
4459 case KVM_CAP_SYNC_MMU:
4460 case KVM_CAP_USER_NMI:
4461 case KVM_CAP_REINJECT_CONTROL:
4462 case KVM_CAP_IRQ_INJECT_STATUS:
4463 case KVM_CAP_IOEVENTFD:
4464 case KVM_CAP_IOEVENTFD_NO_LENGTH:
4466 case KVM_CAP_PIT_STATE2:
4467 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4468 case KVM_CAP_VCPU_EVENTS:
4469 case KVM_CAP_HYPERV:
4470 case KVM_CAP_HYPERV_VAPIC:
4471 case KVM_CAP_HYPERV_SPIN:
4472 case KVM_CAP_HYPERV_SYNIC:
4473 case KVM_CAP_HYPERV_SYNIC2:
4474 case KVM_CAP_HYPERV_VP_INDEX:
4475 case KVM_CAP_HYPERV_EVENTFD:
4476 case KVM_CAP_HYPERV_TLBFLUSH:
4477 case KVM_CAP_HYPERV_SEND_IPI:
4478 case KVM_CAP_HYPERV_CPUID:
4479 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4480 case KVM_CAP_SYS_HYPERV_CPUID:
4481 case KVM_CAP_PCI_SEGMENT:
4482 case KVM_CAP_DEBUGREGS:
4483 case KVM_CAP_X86_ROBUST_SINGLESTEP:
4485 case KVM_CAP_ASYNC_PF:
4486 case KVM_CAP_ASYNC_PF_INT:
4487 case KVM_CAP_GET_TSC_KHZ:
4488 case KVM_CAP_KVMCLOCK_CTRL:
4489 case KVM_CAP_READONLY_MEM:
4490 case KVM_CAP_HYPERV_TIME:
4491 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4492 case KVM_CAP_TSC_DEADLINE_TIMER:
4493 case KVM_CAP_DISABLE_QUIRKS:
4494 case KVM_CAP_SET_BOOT_CPU_ID:
4495 case KVM_CAP_SPLIT_IRQCHIP:
4496 case KVM_CAP_IMMEDIATE_EXIT:
4497 case KVM_CAP_PMU_EVENT_FILTER:
4498 case KVM_CAP_PMU_EVENT_MASKED_EVENTS:
4499 case KVM_CAP_GET_MSR_FEATURES:
4500 case KVM_CAP_MSR_PLATFORM_INFO:
4501 case KVM_CAP_EXCEPTION_PAYLOAD:
4502 case KVM_CAP_X86_TRIPLE_FAULT_EVENT:
4503 case KVM_CAP_SET_GUEST_DEBUG:
4504 case KVM_CAP_LAST_CPU:
4505 case KVM_CAP_X86_USER_SPACE_MSR:
4506 case KVM_CAP_X86_MSR_FILTER:
4507 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4508 #ifdef CONFIG_X86_SGX_KVM
4509 case KVM_CAP_SGX_ATTRIBUTE:
4511 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4512 case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
4513 case KVM_CAP_SREGS2:
4514 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4515 case KVM_CAP_VCPU_ATTRIBUTES:
4516 case KVM_CAP_SYS_ATTRIBUTES:
4518 case KVM_CAP_ENABLE_CAP:
4519 case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
4520 case KVM_CAP_IRQFD_RESAMPLE:
4523 case KVM_CAP_EXIT_HYPERCALL:
4524 r = KVM_EXIT_HYPERCALL_VALID_MASK;
4526 case KVM_CAP_SET_GUEST_DEBUG2:
4527 return KVM_GUESTDBG_VALID_MASK;
4528 #ifdef CONFIG_KVM_XEN
4529 case KVM_CAP_XEN_HVM:
4530 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4531 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4532 KVM_XEN_HVM_CONFIG_SHARED_INFO |
4533 KVM_XEN_HVM_CONFIG_EVTCHN_2LEVEL |
4534 KVM_XEN_HVM_CONFIG_EVTCHN_SEND;
4535 if (sched_info_on())
4536 r |= KVM_XEN_HVM_CONFIG_RUNSTATE |
4537 KVM_XEN_HVM_CONFIG_RUNSTATE_UPDATE_FLAG;
4540 case KVM_CAP_SYNC_REGS:
4541 r = KVM_SYNC_X86_VALID_FIELDS;
4543 case KVM_CAP_ADJUST_CLOCK:
4544 r = KVM_CLOCK_VALID_FLAGS;
4546 case KVM_CAP_X86_DISABLE_EXITS:
4547 r = KVM_X86_DISABLE_EXITS_PAUSE;
4549 if (!mitigate_smt_rsb) {
4550 r |= KVM_X86_DISABLE_EXITS_HLT |
4551 KVM_X86_DISABLE_EXITS_CSTATE;
4553 if (kvm_can_mwait_in_guest())
4554 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4557 case KVM_CAP_X86_SMM:
4558 if (!IS_ENABLED(CONFIG_KVM_SMM))
4561 /* SMBASE is usually relocated above 1M on modern chipsets,
4562 * and SMM handlers might indeed rely on 4G segment limits,
4563 * so do not report SMM to be available if real mode is
4564 * emulated via vm86 mode. Still, do not go to great lengths
4565 * to avoid userspace's usage of the feature, because it is a
4566 * fringe case that is not enabled except via specific settings
4567 * of the module parameters.
4569 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4571 case KVM_CAP_NR_VCPUS:
4572 r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
4574 case KVM_CAP_MAX_VCPUS:
4577 case KVM_CAP_MAX_VCPU_ID:
4578 r = KVM_MAX_VCPU_IDS;
4580 case KVM_CAP_PV_MMU: /* obsolete */
4584 r = KVM_MAX_MCE_BANKS;
4587 r = boot_cpu_has(X86_FEATURE_XSAVE);
4589 case KVM_CAP_TSC_CONTROL:
4590 case KVM_CAP_VM_TSC_CONTROL:
4591 r = kvm_caps.has_tsc_control;
4593 case KVM_CAP_X2APIC_API:
4594 r = KVM_X2APIC_API_VALID_FLAGS;
4596 case KVM_CAP_NESTED_STATE:
4597 r = kvm_x86_ops.nested_ops->get_state ?
4598 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4600 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4601 r = kvm_x86_ops.enable_l2_tlb_flush != NULL;
4603 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4604 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4606 case KVM_CAP_SMALLER_MAXPHYADDR:
4607 r = (int) allow_smaller_maxphyaddr;
4609 case KVM_CAP_STEAL_TIME:
4610 r = sched_info_on();
4612 case KVM_CAP_X86_BUS_LOCK_EXIT:
4613 if (kvm_caps.has_bus_lock_exit)
4614 r = KVM_BUS_LOCK_DETECTION_OFF |
4615 KVM_BUS_LOCK_DETECTION_EXIT;
4619 case KVM_CAP_XSAVE2: {
4620 r = xstate_required_size(kvm_get_filtered_xcr0(), false);
4621 if (r < sizeof(struct kvm_xsave))
4622 r = sizeof(struct kvm_xsave);
4625 case KVM_CAP_PMU_CAPABILITY:
4626 r = enable_pmu ? KVM_CAP_PMU_VALID_MASK : 0;
4628 case KVM_CAP_DISABLE_QUIRKS2:
4629 r = KVM_X86_VALID_QUIRKS;
4631 case KVM_CAP_X86_NOTIFY_VMEXIT:
4632 r = kvm_caps.has_notify_vmexit;
4640 static inline void __user *kvm_get_attr_addr(struct kvm_device_attr *attr)
4642 void __user *uaddr = (void __user*)(unsigned long)attr->addr;
4644 if ((u64)(unsigned long)uaddr != attr->addr)
4645 return ERR_PTR_USR(-EFAULT);
4649 static int kvm_x86_dev_get_attr(struct kvm_device_attr *attr)
4651 u64 __user *uaddr = kvm_get_attr_addr(attr);
4657 return PTR_ERR(uaddr);
4659 switch (attr->attr) {
4660 case KVM_X86_XCOMP_GUEST_SUPP:
4661 if (put_user(kvm_caps.supported_xcr0, uaddr))
4669 static int kvm_x86_dev_has_attr(struct kvm_device_attr *attr)
4674 switch (attr->attr) {
4675 case KVM_X86_XCOMP_GUEST_SUPP:
4682 long kvm_arch_dev_ioctl(struct file *filp,
4683 unsigned int ioctl, unsigned long arg)
4685 void __user *argp = (void __user *)arg;
4689 case KVM_GET_MSR_INDEX_LIST: {
4690 struct kvm_msr_list __user *user_msr_list = argp;
4691 struct kvm_msr_list msr_list;
4695 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4698 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4699 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4702 if (n < msr_list.nmsrs)
4705 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4706 num_msrs_to_save * sizeof(u32)))
4708 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4710 num_emulated_msrs * sizeof(u32)))
4715 case KVM_GET_SUPPORTED_CPUID:
4716 case KVM_GET_EMULATED_CPUID: {
4717 struct kvm_cpuid2 __user *cpuid_arg = argp;
4718 struct kvm_cpuid2 cpuid;
4721 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4724 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4730 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4735 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4737 if (copy_to_user(argp, &kvm_caps.supported_mce_cap,
4738 sizeof(kvm_caps.supported_mce_cap)))
4742 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4743 struct kvm_msr_list __user *user_msr_list = argp;
4744 struct kvm_msr_list msr_list;
4748 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4751 msr_list.nmsrs = num_msr_based_features;
4752 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4755 if (n < msr_list.nmsrs)
4758 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4759 num_msr_based_features * sizeof(u32)))
4765 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4767 case KVM_GET_SUPPORTED_HV_CPUID:
4768 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4770 case KVM_GET_DEVICE_ATTR: {
4771 struct kvm_device_attr attr;
4773 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4775 r = kvm_x86_dev_get_attr(&attr);
4778 case KVM_HAS_DEVICE_ATTR: {
4779 struct kvm_device_attr attr;
4781 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4783 r = kvm_x86_dev_has_attr(&attr);
4794 static void wbinvd_ipi(void *garbage)
4799 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4801 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4804 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4806 /* Address WBINVD may be executed by guest */
4807 if (need_emulate_wbinvd(vcpu)) {
4808 if (static_call(kvm_x86_has_wbinvd_exit)())
4809 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4810 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4811 smp_call_function_single(vcpu->cpu,
4812 wbinvd_ipi, NULL, 1);
4815 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4817 /* Save host pkru register if supported */
4818 vcpu->arch.host_pkru = read_pkru();
4820 /* Apply any externally detected TSC adjustments (due to suspend) */
4821 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4822 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4823 vcpu->arch.tsc_offset_adjustment = 0;
4824 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4827 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4828 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4829 rdtsc() - vcpu->arch.last_host_tsc;
4831 mark_tsc_unstable("KVM discovered backwards TSC");
4833 if (kvm_check_tsc_unstable()) {
4834 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4835 vcpu->arch.last_guest_tsc);
4836 kvm_vcpu_write_tsc_offset(vcpu, offset);
4837 vcpu->arch.tsc_catchup = 1;
4840 if (kvm_lapic_hv_timer_in_use(vcpu))
4841 kvm_lapic_restart_hv_timer(vcpu);
4844 * On a host with synchronized TSC, there is no need to update
4845 * kvmclock on vcpu->cpu migration
4847 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4848 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4849 if (vcpu->cpu != cpu)
4850 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4854 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4857 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4859 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4860 struct kvm_steal_time __user *st;
4861 struct kvm_memslots *slots;
4862 static const u8 preempted = KVM_VCPU_PREEMPTED;
4863 gpa_t gpa = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
4866 * The vCPU can be marked preempted if and only if the VM-Exit was on
4867 * an instruction boundary and will not trigger guest emulation of any
4868 * kind (see vcpu_run). Vendor specific code controls (conservatively)
4869 * when this is true, for example allowing the vCPU to be marked
4870 * preempted if and only if the VM-Exit was due to a host interrupt.
4872 if (!vcpu->arch.at_instruction_boundary) {
4873 vcpu->stat.preemption_other++;
4877 vcpu->stat.preemption_reported++;
4878 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4881 if (vcpu->arch.st.preempted)
4884 /* This happens on process exit */
4885 if (unlikely(current->mm != vcpu->kvm->mm))
4888 slots = kvm_memslots(vcpu->kvm);
4890 if (unlikely(slots->generation != ghc->generation ||
4892 kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4895 st = (struct kvm_steal_time __user *)ghc->hva;
4896 BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4898 if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4899 vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4901 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4904 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4908 if (vcpu->preempted) {
4909 if (!vcpu->arch.guest_state_protected)
4910 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4913 * Take the srcu lock as memslots will be accessed to check the gfn
4914 * cache generation against the memslots generation.
4916 idx = srcu_read_lock(&vcpu->kvm->srcu);
4917 if (kvm_xen_msr_enabled(vcpu->kvm))
4918 kvm_xen_runstate_set_preempted(vcpu);
4920 kvm_steal_time_set_preempted(vcpu);
4921 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4924 static_call(kvm_x86_vcpu_put)(vcpu);
4925 vcpu->arch.last_host_tsc = rdtsc();
4928 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4929 struct kvm_lapic_state *s)
4931 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
4933 return kvm_apic_get_state(vcpu, s);
4936 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4937 struct kvm_lapic_state *s)
4941 r = kvm_apic_set_state(vcpu, s);
4944 update_cr8_intercept(vcpu);
4949 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4952 * We can accept userspace's request for interrupt injection
4953 * as long as we have a place to store the interrupt number.
4954 * The actual injection will happen when the CPU is able to
4955 * deliver the interrupt.
4957 if (kvm_cpu_has_extint(vcpu))
4960 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4961 return (!lapic_in_kernel(vcpu) ||
4962 kvm_apic_accept_pic_intr(vcpu));
4965 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4968 * Do not cause an interrupt window exit if an exception
4969 * is pending or an event needs reinjection; userspace
4970 * might want to inject the interrupt manually using KVM_SET_REGS
4971 * or KVM_SET_SREGS. For that to work, we must be at an
4972 * instruction boundary and with no events half-injected.
4974 return (kvm_arch_interrupt_allowed(vcpu) &&
4975 kvm_cpu_accept_dm_intr(vcpu) &&
4976 !kvm_event_needs_reinjection(vcpu) &&
4977 !kvm_is_exception_pending(vcpu));
4980 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4981 struct kvm_interrupt *irq)
4983 if (irq->irq >= KVM_NR_INTERRUPTS)
4986 if (!irqchip_in_kernel(vcpu->kvm)) {
4987 kvm_queue_interrupt(vcpu, irq->irq, false);
4988 kvm_make_request(KVM_REQ_EVENT, vcpu);
4993 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4994 * fail for in-kernel 8259.
4996 if (pic_in_kernel(vcpu->kvm))
4999 if (vcpu->arch.pending_external_vector != -1)
5002 vcpu->arch.pending_external_vector = irq->irq;
5003 kvm_make_request(KVM_REQ_EVENT, vcpu);
5007 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
5009 kvm_inject_nmi(vcpu);
5014 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
5015 struct kvm_tpr_access_ctl *tac)
5019 vcpu->arch.tpr_access_reporting = !!tac->enabled;
5023 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
5027 unsigned bank_num = mcg_cap & 0xff, bank;
5030 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
5032 if (mcg_cap & ~(kvm_caps.supported_mce_cap | 0xff | 0xff0000))
5035 vcpu->arch.mcg_cap = mcg_cap;
5036 /* Init IA32_MCG_CTL to all 1s */
5037 if (mcg_cap & MCG_CTL_P)
5038 vcpu->arch.mcg_ctl = ~(u64)0;
5039 /* Init IA32_MCi_CTL to all 1s, IA32_MCi_CTL2 to all 0s */
5040 for (bank = 0; bank < bank_num; bank++) {
5041 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
5042 if (mcg_cap & MCG_CMCI_P)
5043 vcpu->arch.mci_ctl2_banks[bank] = 0;
5046 kvm_apic_after_set_mcg_cap(vcpu);
5048 static_call(kvm_x86_setup_mce)(vcpu);
5054 * Validate this is an UCNA (uncorrectable no action) error by checking the
5055 * MCG_STATUS and MCi_STATUS registers:
5056 * - none of the bits for Machine Check Exceptions are set
5057 * - both the VAL (valid) and UC (uncorrectable) bits are set
5058 * MCI_STATUS_PCC - Processor Context Corrupted
5059 * MCI_STATUS_S - Signaled as a Machine Check Exception
5060 * MCI_STATUS_AR - Software recoverable Action Required
5062 static bool is_ucna(struct kvm_x86_mce *mce)
5064 return !mce->mcg_status &&
5065 !(mce->status & (MCI_STATUS_PCC | MCI_STATUS_S | MCI_STATUS_AR)) &&
5066 (mce->status & MCI_STATUS_VAL) &&
5067 (mce->status & MCI_STATUS_UC);
5070 static int kvm_vcpu_x86_set_ucna(struct kvm_vcpu *vcpu, struct kvm_x86_mce *mce, u64* banks)
5072 u64 mcg_cap = vcpu->arch.mcg_cap;
5074 banks[1] = mce->status;
5075 banks[2] = mce->addr;
5076 banks[3] = mce->misc;
5077 vcpu->arch.mcg_status = mce->mcg_status;
5079 if (!(mcg_cap & MCG_CMCI_P) ||
5080 !(vcpu->arch.mci_ctl2_banks[mce->bank] & MCI_CTL2_CMCI_EN))
5083 if (lapic_in_kernel(vcpu))
5084 kvm_apic_local_deliver(vcpu->arch.apic, APIC_LVTCMCI);
5089 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
5090 struct kvm_x86_mce *mce)
5092 u64 mcg_cap = vcpu->arch.mcg_cap;
5093 unsigned bank_num = mcg_cap & 0xff;
5094 u64 *banks = vcpu->arch.mce_banks;
5096 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
5099 banks += array_index_nospec(4 * mce->bank, 4 * bank_num);
5102 return kvm_vcpu_x86_set_ucna(vcpu, mce, banks);
5105 * if IA32_MCG_CTL is not all 1s, the uncorrected error
5106 * reporting is disabled
5108 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
5109 vcpu->arch.mcg_ctl != ~(u64)0)
5112 * if IA32_MCi_CTL is not all 1s, the uncorrected error
5113 * reporting is disabled for the bank
5115 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
5117 if (mce->status & MCI_STATUS_UC) {
5118 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
5119 !kvm_is_cr4_bit_set(vcpu, X86_CR4_MCE)) {
5120 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5123 if (banks[1] & MCI_STATUS_VAL)
5124 mce->status |= MCI_STATUS_OVER;
5125 banks[2] = mce->addr;
5126 banks[3] = mce->misc;
5127 vcpu->arch.mcg_status = mce->mcg_status;
5128 banks[1] = mce->status;
5129 kvm_queue_exception(vcpu, MC_VECTOR);
5130 } else if (!(banks[1] & MCI_STATUS_VAL)
5131 || !(banks[1] & MCI_STATUS_UC)) {
5132 if (banks[1] & MCI_STATUS_VAL)
5133 mce->status |= MCI_STATUS_OVER;
5134 banks[2] = mce->addr;
5135 banks[3] = mce->misc;
5136 banks[1] = mce->status;
5138 banks[1] |= MCI_STATUS_OVER;
5142 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
5143 struct kvm_vcpu_events *events)
5145 struct kvm_queued_exception *ex;
5149 #ifdef CONFIG_KVM_SMM
5150 if (kvm_check_request(KVM_REQ_SMI, vcpu))
5155 * KVM's ABI only allows for one exception to be migrated. Luckily,
5156 * the only time there can be two queued exceptions is if there's a
5157 * non-exiting _injected_ exception, and a pending exiting exception.
5158 * In that case, ignore the VM-Exiting exception as it's an extension
5159 * of the injected exception.
5161 if (vcpu->arch.exception_vmexit.pending &&
5162 !vcpu->arch.exception.pending &&
5163 !vcpu->arch.exception.injected)
5164 ex = &vcpu->arch.exception_vmexit;
5166 ex = &vcpu->arch.exception;
5169 * In guest mode, payload delivery should be deferred if the exception
5170 * will be intercepted by L1, e.g. KVM should not modifying CR2 if L1
5171 * intercepts #PF, ditto for DR6 and #DBs. If the per-VM capability,
5172 * KVM_CAP_EXCEPTION_PAYLOAD, is not set, userspace may or may not
5173 * propagate the payload and so it cannot be safely deferred. Deliver
5174 * the payload if the capability hasn't been requested.
5176 if (!vcpu->kvm->arch.exception_payload_enabled &&
5177 ex->pending && ex->has_payload)
5178 kvm_deliver_exception_payload(vcpu, ex);
5180 memset(events, 0, sizeof(*events));
5183 * The API doesn't provide the instruction length for software
5184 * exceptions, so don't report them. As long as the guest RIP
5185 * isn't advanced, we should expect to encounter the exception
5188 if (!kvm_exception_is_soft(ex->vector)) {
5189 events->exception.injected = ex->injected;
5190 events->exception.pending = ex->pending;
5192 * For ABI compatibility, deliberately conflate
5193 * pending and injected exceptions when
5194 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
5196 if (!vcpu->kvm->arch.exception_payload_enabled)
5197 events->exception.injected |= ex->pending;
5199 events->exception.nr = ex->vector;
5200 events->exception.has_error_code = ex->has_error_code;
5201 events->exception.error_code = ex->error_code;
5202 events->exception_has_payload = ex->has_payload;
5203 events->exception_payload = ex->payload;
5205 events->interrupt.injected =
5206 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
5207 events->interrupt.nr = vcpu->arch.interrupt.nr;
5208 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
5210 events->nmi.injected = vcpu->arch.nmi_injected;
5211 events->nmi.pending = kvm_get_nr_pending_nmis(vcpu);
5212 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
5214 /* events->sipi_vector is never valid when reporting to user space */
5216 #ifdef CONFIG_KVM_SMM
5217 events->smi.smm = is_smm(vcpu);
5218 events->smi.pending = vcpu->arch.smi_pending;
5219 events->smi.smm_inside_nmi =
5220 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
5222 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
5224 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
5225 | KVM_VCPUEVENT_VALID_SHADOW
5226 | KVM_VCPUEVENT_VALID_SMM);
5227 if (vcpu->kvm->arch.exception_payload_enabled)
5228 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
5229 if (vcpu->kvm->arch.triple_fault_event) {
5230 events->triple_fault.pending = kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5231 events->flags |= KVM_VCPUEVENT_VALID_TRIPLE_FAULT;
5235 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
5236 struct kvm_vcpu_events *events)
5238 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
5239 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
5240 | KVM_VCPUEVENT_VALID_SHADOW
5241 | KVM_VCPUEVENT_VALID_SMM
5242 | KVM_VCPUEVENT_VALID_PAYLOAD
5243 | KVM_VCPUEVENT_VALID_TRIPLE_FAULT))
5246 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
5247 if (!vcpu->kvm->arch.exception_payload_enabled)
5249 if (events->exception.pending)
5250 events->exception.injected = 0;
5252 events->exception_has_payload = 0;
5254 events->exception.pending = 0;
5255 events->exception_has_payload = 0;
5258 if ((events->exception.injected || events->exception.pending) &&
5259 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
5262 /* INITs are latched while in SMM */
5263 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
5264 (events->smi.smm || events->smi.pending) &&
5265 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
5271 * Flag that userspace is stuffing an exception, the next KVM_RUN will
5272 * morph the exception to a VM-Exit if appropriate. Do this only for
5273 * pending exceptions, already-injected exceptions are not subject to
5274 * intercpetion. Note, userspace that conflates pending and injected
5275 * is hosed, and will incorrectly convert an injected exception into a
5276 * pending exception, which in turn may cause a spurious VM-Exit.
5278 vcpu->arch.exception_from_userspace = events->exception.pending;
5280 vcpu->arch.exception_vmexit.pending = false;
5282 vcpu->arch.exception.injected = events->exception.injected;
5283 vcpu->arch.exception.pending = events->exception.pending;
5284 vcpu->arch.exception.vector = events->exception.nr;
5285 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
5286 vcpu->arch.exception.error_code = events->exception.error_code;
5287 vcpu->arch.exception.has_payload = events->exception_has_payload;
5288 vcpu->arch.exception.payload = events->exception_payload;
5290 vcpu->arch.interrupt.injected = events->interrupt.injected;
5291 vcpu->arch.interrupt.nr = events->interrupt.nr;
5292 vcpu->arch.interrupt.soft = events->interrupt.soft;
5293 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
5294 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
5295 events->interrupt.shadow);
5297 vcpu->arch.nmi_injected = events->nmi.injected;
5298 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING) {
5299 vcpu->arch.nmi_pending = 0;
5300 atomic_set(&vcpu->arch.nmi_queued, events->nmi.pending);
5301 kvm_make_request(KVM_REQ_NMI, vcpu);
5303 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
5305 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
5306 lapic_in_kernel(vcpu))
5307 vcpu->arch.apic->sipi_vector = events->sipi_vector;
5309 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
5310 #ifdef CONFIG_KVM_SMM
5311 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
5312 kvm_leave_nested(vcpu);
5313 kvm_smm_changed(vcpu, events->smi.smm);
5316 vcpu->arch.smi_pending = events->smi.pending;
5318 if (events->smi.smm) {
5319 if (events->smi.smm_inside_nmi)
5320 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
5322 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
5326 if (events->smi.smm || events->smi.pending ||
5327 events->smi.smm_inside_nmi)
5331 if (lapic_in_kernel(vcpu)) {
5332 if (events->smi.latched_init)
5333 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
5335 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
5339 if (events->flags & KVM_VCPUEVENT_VALID_TRIPLE_FAULT) {
5340 if (!vcpu->kvm->arch.triple_fault_event)
5342 if (events->triple_fault.pending)
5343 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5345 kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5348 kvm_make_request(KVM_REQ_EVENT, vcpu);
5353 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
5354 struct kvm_debugregs *dbgregs)
5358 memset(dbgregs, 0, sizeof(*dbgregs));
5359 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
5360 kvm_get_dr(vcpu, 6, &val);
5362 dbgregs->dr7 = vcpu->arch.dr7;
5365 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
5366 struct kvm_debugregs *dbgregs)
5371 if (!kvm_dr6_valid(dbgregs->dr6))
5373 if (!kvm_dr7_valid(dbgregs->dr7))
5376 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
5377 kvm_update_dr0123(vcpu);
5378 vcpu->arch.dr6 = dbgregs->dr6;
5379 vcpu->arch.dr7 = dbgregs->dr7;
5380 kvm_update_dr7(vcpu);
5385 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
5386 struct kvm_xsave *guest_xsave)
5388 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5391 fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
5392 guest_xsave->region,
5393 sizeof(guest_xsave->region),
5397 static void kvm_vcpu_ioctl_x86_get_xsave2(struct kvm_vcpu *vcpu,
5398 u8 *state, unsigned int size)
5400 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5403 fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
5404 state, size, vcpu->arch.pkru);
5407 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
5408 struct kvm_xsave *guest_xsave)
5410 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5413 return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
5414 guest_xsave->region,
5415 kvm_caps.supported_xcr0,
5419 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
5420 struct kvm_xcrs *guest_xcrs)
5422 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
5423 guest_xcrs->nr_xcrs = 0;
5427 guest_xcrs->nr_xcrs = 1;
5428 guest_xcrs->flags = 0;
5429 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
5430 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
5433 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
5434 struct kvm_xcrs *guest_xcrs)
5438 if (!boot_cpu_has(X86_FEATURE_XSAVE))
5441 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
5444 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
5445 /* Only support XCR0 currently */
5446 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
5447 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
5448 guest_xcrs->xcrs[i].value);
5457 * kvm_set_guest_paused() indicates to the guest kernel that it has been
5458 * stopped by the hypervisor. This function will be called from the host only.
5459 * EINVAL is returned when the host attempts to set the flag for a guest that
5460 * does not support pv clocks.
5462 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
5464 if (!vcpu->arch.pv_time.active)
5466 vcpu->arch.pvclock_set_guest_stopped_request = true;
5467 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5471 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
5472 struct kvm_device_attr *attr)
5476 switch (attr->attr) {
5477 case KVM_VCPU_TSC_OFFSET:
5487 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
5488 struct kvm_device_attr *attr)
5490 u64 __user *uaddr = kvm_get_attr_addr(attr);
5494 return PTR_ERR(uaddr);
5496 switch (attr->attr) {
5497 case KVM_VCPU_TSC_OFFSET:
5499 if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
5510 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
5511 struct kvm_device_attr *attr)
5513 u64 __user *uaddr = kvm_get_attr_addr(attr);
5514 struct kvm *kvm = vcpu->kvm;
5518 return PTR_ERR(uaddr);
5520 switch (attr->attr) {
5521 case KVM_VCPU_TSC_OFFSET: {
5522 u64 offset, tsc, ns;
5523 unsigned long flags;
5527 if (get_user(offset, uaddr))
5530 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
5532 matched = (vcpu->arch.virtual_tsc_khz &&
5533 kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
5534 kvm->arch.last_tsc_offset == offset);
5536 tsc = kvm_scale_tsc(rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
5537 ns = get_kvmclock_base_ns();
5539 __kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
5540 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
5552 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
5556 struct kvm_device_attr attr;
5559 if (copy_from_user(&attr, argp, sizeof(attr)))
5562 if (attr.group != KVM_VCPU_TSC_CTRL)
5566 case KVM_HAS_DEVICE_ATTR:
5567 r = kvm_arch_tsc_has_attr(vcpu, &attr);
5569 case KVM_GET_DEVICE_ATTR:
5570 r = kvm_arch_tsc_get_attr(vcpu, &attr);
5572 case KVM_SET_DEVICE_ATTR:
5573 r = kvm_arch_tsc_set_attr(vcpu, &attr);
5580 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5581 struct kvm_enable_cap *cap)
5584 uint16_t vmcs_version;
5585 void __user *user_ptr;
5591 case KVM_CAP_HYPERV_SYNIC2:
5596 case KVM_CAP_HYPERV_SYNIC:
5597 if (!irqchip_in_kernel(vcpu->kvm))
5599 return kvm_hv_activate_synic(vcpu, cap->cap ==
5600 KVM_CAP_HYPERV_SYNIC2);
5601 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5602 if (!kvm_x86_ops.nested_ops->enable_evmcs)
5604 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5606 user_ptr = (void __user *)(uintptr_t)cap->args[0];
5607 if (copy_to_user(user_ptr, &vmcs_version,
5608 sizeof(vmcs_version)))
5612 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5613 if (!kvm_x86_ops.enable_l2_tlb_flush)
5616 return static_call(kvm_x86_enable_l2_tlb_flush)(vcpu);
5618 case KVM_CAP_HYPERV_ENFORCE_CPUID:
5619 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5621 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5622 vcpu->arch.pv_cpuid.enforce = cap->args[0];
5623 if (vcpu->arch.pv_cpuid.enforce)
5624 kvm_update_pv_runtime(vcpu);
5632 long kvm_arch_vcpu_ioctl(struct file *filp,
5633 unsigned int ioctl, unsigned long arg)
5635 struct kvm_vcpu *vcpu = filp->private_data;
5636 void __user *argp = (void __user *)arg;
5639 struct kvm_sregs2 *sregs2;
5640 struct kvm_lapic_state *lapic;
5641 struct kvm_xsave *xsave;
5642 struct kvm_xcrs *xcrs;
5650 case KVM_GET_LAPIC: {
5652 if (!lapic_in_kernel(vcpu))
5654 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5655 GFP_KERNEL_ACCOUNT);
5660 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5664 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5669 case KVM_SET_LAPIC: {
5671 if (!lapic_in_kernel(vcpu))
5673 u.lapic = memdup_user(argp, sizeof(*u.lapic));
5674 if (IS_ERR(u.lapic)) {
5675 r = PTR_ERR(u.lapic);
5679 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5682 case KVM_INTERRUPT: {
5683 struct kvm_interrupt irq;
5686 if (copy_from_user(&irq, argp, sizeof(irq)))
5688 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5692 r = kvm_vcpu_ioctl_nmi(vcpu);
5696 r = kvm_inject_smi(vcpu);
5699 case KVM_SET_CPUID: {
5700 struct kvm_cpuid __user *cpuid_arg = argp;
5701 struct kvm_cpuid cpuid;
5704 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5706 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5709 case KVM_SET_CPUID2: {
5710 struct kvm_cpuid2 __user *cpuid_arg = argp;
5711 struct kvm_cpuid2 cpuid;
5714 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5716 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5717 cpuid_arg->entries);
5720 case KVM_GET_CPUID2: {
5721 struct kvm_cpuid2 __user *cpuid_arg = argp;
5722 struct kvm_cpuid2 cpuid;
5725 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5727 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5728 cpuid_arg->entries);
5732 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5737 case KVM_GET_MSRS: {
5738 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5739 r = msr_io(vcpu, argp, do_get_msr, 1);
5740 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5743 case KVM_SET_MSRS: {
5744 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5745 r = msr_io(vcpu, argp, do_set_msr, 0);
5746 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5749 case KVM_TPR_ACCESS_REPORTING: {
5750 struct kvm_tpr_access_ctl tac;
5753 if (copy_from_user(&tac, argp, sizeof(tac)))
5755 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5759 if (copy_to_user(argp, &tac, sizeof(tac)))
5764 case KVM_SET_VAPIC_ADDR: {
5765 struct kvm_vapic_addr va;
5769 if (!lapic_in_kernel(vcpu))
5772 if (copy_from_user(&va, argp, sizeof(va)))
5774 idx = srcu_read_lock(&vcpu->kvm->srcu);
5775 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5776 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5779 case KVM_X86_SETUP_MCE: {
5783 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5785 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5788 case KVM_X86_SET_MCE: {
5789 struct kvm_x86_mce mce;
5792 if (copy_from_user(&mce, argp, sizeof(mce)))
5794 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5797 case KVM_GET_VCPU_EVENTS: {
5798 struct kvm_vcpu_events events;
5800 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5803 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5808 case KVM_SET_VCPU_EVENTS: {
5809 struct kvm_vcpu_events events;
5812 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5815 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5818 case KVM_GET_DEBUGREGS: {
5819 struct kvm_debugregs dbgregs;
5821 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5824 if (copy_to_user(argp, &dbgregs,
5825 sizeof(struct kvm_debugregs)))
5830 case KVM_SET_DEBUGREGS: {
5831 struct kvm_debugregs dbgregs;
5834 if (copy_from_user(&dbgregs, argp,
5835 sizeof(struct kvm_debugregs)))
5838 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5841 case KVM_GET_XSAVE: {
5843 if (vcpu->arch.guest_fpu.uabi_size > sizeof(struct kvm_xsave))
5846 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5851 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5854 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5859 case KVM_SET_XSAVE: {
5860 int size = vcpu->arch.guest_fpu.uabi_size;
5862 u.xsave = memdup_user(argp, size);
5863 if (IS_ERR(u.xsave)) {
5864 r = PTR_ERR(u.xsave);
5868 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5872 case KVM_GET_XSAVE2: {
5873 int size = vcpu->arch.guest_fpu.uabi_size;
5875 u.xsave = kzalloc(size, GFP_KERNEL_ACCOUNT);
5880 kvm_vcpu_ioctl_x86_get_xsave2(vcpu, u.buffer, size);
5883 if (copy_to_user(argp, u.xsave, size))
5890 case KVM_GET_XCRS: {
5891 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5896 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5899 if (copy_to_user(argp, u.xcrs,
5900 sizeof(struct kvm_xcrs)))
5905 case KVM_SET_XCRS: {
5906 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5907 if (IS_ERR(u.xcrs)) {
5908 r = PTR_ERR(u.xcrs);
5912 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5915 case KVM_SET_TSC_KHZ: {
5919 user_tsc_khz = (u32)arg;
5921 if (kvm_caps.has_tsc_control &&
5922 user_tsc_khz >= kvm_caps.max_guest_tsc_khz)
5925 if (user_tsc_khz == 0)
5926 user_tsc_khz = tsc_khz;
5928 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5933 case KVM_GET_TSC_KHZ: {
5934 r = vcpu->arch.virtual_tsc_khz;
5937 case KVM_KVMCLOCK_CTRL: {
5938 r = kvm_set_guest_paused(vcpu);
5941 case KVM_ENABLE_CAP: {
5942 struct kvm_enable_cap cap;
5945 if (copy_from_user(&cap, argp, sizeof(cap)))
5947 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5950 case KVM_GET_NESTED_STATE: {
5951 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5955 if (!kvm_x86_ops.nested_ops->get_state)
5958 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5960 if (get_user(user_data_size, &user_kvm_nested_state->size))
5963 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5968 if (r > user_data_size) {
5969 if (put_user(r, &user_kvm_nested_state->size))
5979 case KVM_SET_NESTED_STATE: {
5980 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5981 struct kvm_nested_state kvm_state;
5985 if (!kvm_x86_ops.nested_ops->set_state)
5989 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5993 if (kvm_state.size < sizeof(kvm_state))
5996 if (kvm_state.flags &
5997 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5998 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5999 | KVM_STATE_NESTED_GIF_SET))
6002 /* nested_run_pending implies guest_mode. */
6003 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
6004 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
6007 idx = srcu_read_lock(&vcpu->kvm->srcu);
6008 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
6009 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6012 case KVM_GET_SUPPORTED_HV_CPUID:
6013 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
6015 #ifdef CONFIG_KVM_XEN
6016 case KVM_XEN_VCPU_GET_ATTR: {
6017 struct kvm_xen_vcpu_attr xva;
6020 if (copy_from_user(&xva, argp, sizeof(xva)))
6022 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
6023 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
6027 case KVM_XEN_VCPU_SET_ATTR: {
6028 struct kvm_xen_vcpu_attr xva;
6031 if (copy_from_user(&xva, argp, sizeof(xva)))
6033 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
6037 case KVM_GET_SREGS2: {
6038 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
6042 __get_sregs2(vcpu, u.sregs2);
6044 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
6049 case KVM_SET_SREGS2: {
6050 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
6051 if (IS_ERR(u.sregs2)) {
6052 r = PTR_ERR(u.sregs2);
6056 r = __set_sregs2(vcpu, u.sregs2);
6059 case KVM_HAS_DEVICE_ATTR:
6060 case KVM_GET_DEVICE_ATTR:
6061 case KVM_SET_DEVICE_ATTR:
6062 r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
6074 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
6076 return VM_FAULT_SIGBUS;
6079 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
6083 if (addr > (unsigned int)(-3 * PAGE_SIZE))
6085 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
6089 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
6092 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
6095 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
6096 unsigned long kvm_nr_mmu_pages)
6098 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
6101 mutex_lock(&kvm->slots_lock);
6103 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
6104 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
6106 mutex_unlock(&kvm->slots_lock);
6110 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
6112 struct kvm_pic *pic = kvm->arch.vpic;
6116 switch (chip->chip_id) {
6117 case KVM_IRQCHIP_PIC_MASTER:
6118 memcpy(&chip->chip.pic, &pic->pics[0],
6119 sizeof(struct kvm_pic_state));
6121 case KVM_IRQCHIP_PIC_SLAVE:
6122 memcpy(&chip->chip.pic, &pic->pics[1],
6123 sizeof(struct kvm_pic_state));
6125 case KVM_IRQCHIP_IOAPIC:
6126 kvm_get_ioapic(kvm, &chip->chip.ioapic);
6135 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
6137 struct kvm_pic *pic = kvm->arch.vpic;
6141 switch (chip->chip_id) {
6142 case KVM_IRQCHIP_PIC_MASTER:
6143 spin_lock(&pic->lock);
6144 memcpy(&pic->pics[0], &chip->chip.pic,
6145 sizeof(struct kvm_pic_state));
6146 spin_unlock(&pic->lock);
6148 case KVM_IRQCHIP_PIC_SLAVE:
6149 spin_lock(&pic->lock);
6150 memcpy(&pic->pics[1], &chip->chip.pic,
6151 sizeof(struct kvm_pic_state));
6152 spin_unlock(&pic->lock);
6154 case KVM_IRQCHIP_IOAPIC:
6155 kvm_set_ioapic(kvm, &chip->chip.ioapic);
6161 kvm_pic_update_irq(pic);
6165 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
6167 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
6169 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
6171 mutex_lock(&kps->lock);
6172 memcpy(ps, &kps->channels, sizeof(*ps));
6173 mutex_unlock(&kps->lock);
6177 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
6180 struct kvm_pit *pit = kvm->arch.vpit;
6182 mutex_lock(&pit->pit_state.lock);
6183 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
6184 for (i = 0; i < 3; i++)
6185 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
6186 mutex_unlock(&pit->pit_state.lock);
6190 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
6192 mutex_lock(&kvm->arch.vpit->pit_state.lock);
6193 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
6194 sizeof(ps->channels));
6195 ps->flags = kvm->arch.vpit->pit_state.flags;
6196 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
6197 memset(&ps->reserved, 0, sizeof(ps->reserved));
6201 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
6205 u32 prev_legacy, cur_legacy;
6206 struct kvm_pit *pit = kvm->arch.vpit;
6208 mutex_lock(&pit->pit_state.lock);
6209 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
6210 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
6211 if (!prev_legacy && cur_legacy)
6213 memcpy(&pit->pit_state.channels, &ps->channels,
6214 sizeof(pit->pit_state.channels));
6215 pit->pit_state.flags = ps->flags;
6216 for (i = 0; i < 3; i++)
6217 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
6219 mutex_unlock(&pit->pit_state.lock);
6223 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
6224 struct kvm_reinject_control *control)
6226 struct kvm_pit *pit = kvm->arch.vpit;
6228 /* pit->pit_state.lock was overloaded to prevent userspace from getting
6229 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
6230 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
6232 mutex_lock(&pit->pit_state.lock);
6233 kvm_pit_set_reinject(pit, control->pit_reinject);
6234 mutex_unlock(&pit->pit_state.lock);
6239 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
6243 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
6244 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
6245 * on all VM-Exits, thus we only need to kick running vCPUs to force a
6248 struct kvm_vcpu *vcpu;
6251 kvm_for_each_vcpu(i, vcpu, kvm)
6252 kvm_vcpu_kick(vcpu);
6255 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
6258 if (!irqchip_in_kernel(kvm))
6261 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
6262 irq_event->irq, irq_event->level,
6267 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
6268 struct kvm_enable_cap *cap)
6276 case KVM_CAP_DISABLE_QUIRKS2:
6278 if (cap->args[0] & ~KVM_X86_VALID_QUIRKS)
6281 case KVM_CAP_DISABLE_QUIRKS:
6282 kvm->arch.disabled_quirks = cap->args[0];
6285 case KVM_CAP_SPLIT_IRQCHIP: {
6286 mutex_lock(&kvm->lock);
6288 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
6289 goto split_irqchip_unlock;
6291 if (irqchip_in_kernel(kvm))
6292 goto split_irqchip_unlock;
6293 if (kvm->created_vcpus)
6294 goto split_irqchip_unlock;
6295 r = kvm_setup_empty_irq_routing(kvm);
6297 goto split_irqchip_unlock;
6298 /* Pairs with irqchip_in_kernel. */
6300 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
6301 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
6302 kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_ABSENT);
6304 split_irqchip_unlock:
6305 mutex_unlock(&kvm->lock);
6308 case KVM_CAP_X2APIC_API:
6310 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
6313 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
6314 kvm->arch.x2apic_format = true;
6315 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
6316 kvm->arch.x2apic_broadcast_quirk_disabled = true;
6320 case KVM_CAP_X86_DISABLE_EXITS:
6322 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
6325 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
6326 kvm->arch.pause_in_guest = true;
6328 #define SMT_RSB_MSG "This processor is affected by the Cross-Thread Return Predictions vulnerability. " \
6329 "KVM_CAP_X86_DISABLE_EXITS should only be used with SMT disabled or trusted guests."
6331 if (!mitigate_smt_rsb) {
6332 if (boot_cpu_has_bug(X86_BUG_SMT_RSB) && cpu_smt_possible() &&
6333 (cap->args[0] & ~KVM_X86_DISABLE_EXITS_PAUSE))
6334 pr_warn_once(SMT_RSB_MSG);
6336 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
6337 kvm_can_mwait_in_guest())
6338 kvm->arch.mwait_in_guest = true;
6339 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
6340 kvm->arch.hlt_in_guest = true;
6341 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
6342 kvm->arch.cstate_in_guest = true;
6347 case KVM_CAP_MSR_PLATFORM_INFO:
6348 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
6351 case KVM_CAP_EXCEPTION_PAYLOAD:
6352 kvm->arch.exception_payload_enabled = cap->args[0];
6355 case KVM_CAP_X86_TRIPLE_FAULT_EVENT:
6356 kvm->arch.triple_fault_event = cap->args[0];
6359 case KVM_CAP_X86_USER_SPACE_MSR:
6361 if (cap->args[0] & ~KVM_MSR_EXIT_REASON_VALID_MASK)
6363 kvm->arch.user_space_msr_mask = cap->args[0];
6366 case KVM_CAP_X86_BUS_LOCK_EXIT:
6368 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
6371 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
6372 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
6375 if (kvm_caps.has_bus_lock_exit &&
6376 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
6377 kvm->arch.bus_lock_detection_enabled = true;
6380 #ifdef CONFIG_X86_SGX_KVM
6381 case KVM_CAP_SGX_ATTRIBUTE: {
6382 unsigned long allowed_attributes = 0;
6384 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
6388 /* KVM only supports the PROVISIONKEY privileged attribute. */
6389 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
6390 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
6391 kvm->arch.sgx_provisioning_allowed = true;
6397 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
6399 if (!kvm_x86_ops.vm_copy_enc_context_from)
6402 r = static_call(kvm_x86_vm_copy_enc_context_from)(kvm, cap->args[0]);
6404 case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
6406 if (!kvm_x86_ops.vm_move_enc_context_from)
6409 r = static_call(kvm_x86_vm_move_enc_context_from)(kvm, cap->args[0]);
6411 case KVM_CAP_EXIT_HYPERCALL:
6412 if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
6416 kvm->arch.hypercall_exit_enabled = cap->args[0];
6419 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
6421 if (cap->args[0] & ~1)
6423 kvm->arch.exit_on_emulation_error = cap->args[0];
6426 case KVM_CAP_PMU_CAPABILITY:
6428 if (!enable_pmu || (cap->args[0] & ~KVM_CAP_PMU_VALID_MASK))
6431 mutex_lock(&kvm->lock);
6432 if (!kvm->created_vcpus) {
6433 kvm->arch.enable_pmu = !(cap->args[0] & KVM_PMU_CAP_DISABLE);
6436 mutex_unlock(&kvm->lock);
6438 case KVM_CAP_MAX_VCPU_ID:
6440 if (cap->args[0] > KVM_MAX_VCPU_IDS)
6443 mutex_lock(&kvm->lock);
6444 if (kvm->arch.max_vcpu_ids == cap->args[0]) {
6446 } else if (!kvm->arch.max_vcpu_ids) {
6447 kvm->arch.max_vcpu_ids = cap->args[0];
6450 mutex_unlock(&kvm->lock);
6452 case KVM_CAP_X86_NOTIFY_VMEXIT:
6454 if ((u32)cap->args[0] & ~KVM_X86_NOTIFY_VMEXIT_VALID_BITS)
6456 if (!kvm_caps.has_notify_vmexit)
6458 if (!((u32)cap->args[0] & KVM_X86_NOTIFY_VMEXIT_ENABLED))
6460 mutex_lock(&kvm->lock);
6461 if (!kvm->created_vcpus) {
6462 kvm->arch.notify_window = cap->args[0] >> 32;
6463 kvm->arch.notify_vmexit_flags = (u32)cap->args[0];
6466 mutex_unlock(&kvm->lock);
6468 case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
6472 * Since the risk of disabling NX hugepages is a guest crashing
6473 * the system, ensure the userspace process has permission to
6474 * reboot the system.
6476 * Note that unlike the reboot() syscall, the process must have
6477 * this capability in the root namespace because exposing
6478 * /dev/kvm into a container does not limit the scope of the
6479 * iTLB multihit bug to that container. In other words,
6480 * this must use capable(), not ns_capable().
6482 if (!capable(CAP_SYS_BOOT)) {
6490 mutex_lock(&kvm->lock);
6491 if (!kvm->created_vcpus) {
6492 kvm->arch.disable_nx_huge_pages = true;
6495 mutex_unlock(&kvm->lock);
6504 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
6506 struct kvm_x86_msr_filter *msr_filter;
6508 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
6512 msr_filter->default_allow = default_allow;
6516 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
6523 for (i = 0; i < msr_filter->count; i++)
6524 kfree(msr_filter->ranges[i].bitmap);
6529 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
6530 struct kvm_msr_filter_range *user_range)
6532 unsigned long *bitmap;
6535 if (!user_range->nmsrs)
6538 if (user_range->flags & ~KVM_MSR_FILTER_RANGE_VALID_MASK)
6541 if (!user_range->flags)
6544 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
6545 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
6548 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
6550 return PTR_ERR(bitmap);
6552 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
6553 .flags = user_range->flags,
6554 .base = user_range->base,
6555 .nmsrs = user_range->nmsrs,
6559 msr_filter->count++;
6563 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm,
6564 struct kvm_msr_filter *filter)
6566 struct kvm_x86_msr_filter *new_filter, *old_filter;
6572 if (filter->flags & ~KVM_MSR_FILTER_VALID_MASK)
6575 for (i = 0; i < ARRAY_SIZE(filter->ranges); i++)
6576 empty &= !filter->ranges[i].nmsrs;
6578 default_allow = !(filter->flags & KVM_MSR_FILTER_DEFAULT_DENY);
6579 if (empty && !default_allow)
6582 new_filter = kvm_alloc_msr_filter(default_allow);
6586 for (i = 0; i < ARRAY_SIZE(filter->ranges); i++) {
6587 r = kvm_add_msr_filter(new_filter, &filter->ranges[i]);
6589 kvm_free_msr_filter(new_filter);
6594 mutex_lock(&kvm->lock);
6595 old_filter = rcu_replace_pointer(kvm->arch.msr_filter, new_filter,
6596 mutex_is_locked(&kvm->lock));
6597 mutex_unlock(&kvm->lock);
6598 synchronize_srcu(&kvm->srcu);
6600 kvm_free_msr_filter(old_filter);
6602 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
6607 #ifdef CONFIG_KVM_COMPAT
6608 /* for KVM_X86_SET_MSR_FILTER */
6609 struct kvm_msr_filter_range_compat {
6616 struct kvm_msr_filter_compat {
6618 struct kvm_msr_filter_range_compat ranges[KVM_MSR_FILTER_MAX_RANGES];
6621 #define KVM_X86_SET_MSR_FILTER_COMPAT _IOW(KVMIO, 0xc6, struct kvm_msr_filter_compat)
6623 long kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
6626 void __user *argp = (void __user *)arg;
6627 struct kvm *kvm = filp->private_data;
6631 case KVM_X86_SET_MSR_FILTER_COMPAT: {
6632 struct kvm_msr_filter __user *user_msr_filter = argp;
6633 struct kvm_msr_filter_compat filter_compat;
6634 struct kvm_msr_filter filter;
6637 if (copy_from_user(&filter_compat, user_msr_filter,
6638 sizeof(filter_compat)))
6641 filter.flags = filter_compat.flags;
6642 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
6643 struct kvm_msr_filter_range_compat *cr;
6645 cr = &filter_compat.ranges[i];
6646 filter.ranges[i] = (struct kvm_msr_filter_range) {
6650 .bitmap = (__u8 *)(ulong)cr->bitmap,
6654 r = kvm_vm_ioctl_set_msr_filter(kvm, &filter);
6663 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
6664 static int kvm_arch_suspend_notifier(struct kvm *kvm)
6666 struct kvm_vcpu *vcpu;
6670 mutex_lock(&kvm->lock);
6671 kvm_for_each_vcpu(i, vcpu, kvm) {
6672 if (!vcpu->arch.pv_time.active)
6675 ret = kvm_set_guest_paused(vcpu);
6677 kvm_err("Failed to pause guest VCPU%d: %d\n",
6678 vcpu->vcpu_id, ret);
6682 mutex_unlock(&kvm->lock);
6684 return ret ? NOTIFY_BAD : NOTIFY_DONE;
6687 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
6690 case PM_HIBERNATION_PREPARE:
6691 case PM_SUSPEND_PREPARE:
6692 return kvm_arch_suspend_notifier(kvm);
6697 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
6699 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
6701 struct kvm_clock_data data = { 0 };
6703 get_kvmclock(kvm, &data);
6704 if (copy_to_user(argp, &data, sizeof(data)))
6710 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
6712 struct kvm_arch *ka = &kvm->arch;
6713 struct kvm_clock_data data;
6716 if (copy_from_user(&data, argp, sizeof(data)))
6720 * Only KVM_CLOCK_REALTIME is used, but allow passing the
6721 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
6723 if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
6726 kvm_hv_request_tsc_page_update(kvm);
6727 kvm_start_pvclock_update(kvm);
6728 pvclock_update_vm_gtod_copy(kvm);
6731 * This pairs with kvm_guest_time_update(): when masterclock is
6732 * in use, we use master_kernel_ns + kvmclock_offset to set
6733 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6734 * is slightly ahead) here we risk going negative on unsigned
6735 * 'system_time' when 'data.clock' is very small.
6737 if (data.flags & KVM_CLOCK_REALTIME) {
6738 u64 now_real_ns = ktime_get_real_ns();
6741 * Avoid stepping the kvmclock backwards.
6743 if (now_real_ns > data.realtime)
6744 data.clock += now_real_ns - data.realtime;
6747 if (ka->use_master_clock)
6748 now_raw_ns = ka->master_kernel_ns;
6750 now_raw_ns = get_kvmclock_base_ns();
6751 ka->kvmclock_offset = data.clock - now_raw_ns;
6752 kvm_end_pvclock_update(kvm);
6756 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
6758 struct kvm *kvm = filp->private_data;
6759 void __user *argp = (void __user *)arg;
6762 * This union makes it completely explicit to gcc-3.x
6763 * that these two variables' stack usage should be
6764 * combined, not added together.
6767 struct kvm_pit_state ps;
6768 struct kvm_pit_state2 ps2;
6769 struct kvm_pit_config pit_config;
6773 case KVM_SET_TSS_ADDR:
6774 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6776 case KVM_SET_IDENTITY_MAP_ADDR: {
6779 mutex_lock(&kvm->lock);
6781 if (kvm->created_vcpus)
6782 goto set_identity_unlock;
6784 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6785 goto set_identity_unlock;
6786 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6787 set_identity_unlock:
6788 mutex_unlock(&kvm->lock);
6791 case KVM_SET_NR_MMU_PAGES:
6792 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6794 case KVM_CREATE_IRQCHIP: {
6795 mutex_lock(&kvm->lock);
6798 if (irqchip_in_kernel(kvm))
6799 goto create_irqchip_unlock;
6802 if (kvm->created_vcpus)
6803 goto create_irqchip_unlock;
6805 r = kvm_pic_init(kvm);
6807 goto create_irqchip_unlock;
6809 r = kvm_ioapic_init(kvm);
6811 kvm_pic_destroy(kvm);
6812 goto create_irqchip_unlock;
6815 r = kvm_setup_default_irq_routing(kvm);
6817 kvm_ioapic_destroy(kvm);
6818 kvm_pic_destroy(kvm);
6819 goto create_irqchip_unlock;
6821 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6823 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6824 kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_ABSENT);
6825 create_irqchip_unlock:
6826 mutex_unlock(&kvm->lock);
6829 case KVM_CREATE_PIT:
6830 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6832 case KVM_CREATE_PIT2:
6834 if (copy_from_user(&u.pit_config, argp,
6835 sizeof(struct kvm_pit_config)))
6838 mutex_lock(&kvm->lock);
6841 goto create_pit_unlock;
6843 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6847 mutex_unlock(&kvm->lock);
6849 case KVM_GET_IRQCHIP: {
6850 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6851 struct kvm_irqchip *chip;
6853 chip = memdup_user(argp, sizeof(*chip));
6860 if (!irqchip_kernel(kvm))
6861 goto get_irqchip_out;
6862 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6864 goto get_irqchip_out;
6866 if (copy_to_user(argp, chip, sizeof(*chip)))
6867 goto get_irqchip_out;
6873 case KVM_SET_IRQCHIP: {
6874 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6875 struct kvm_irqchip *chip;
6877 chip = memdup_user(argp, sizeof(*chip));
6884 if (!irqchip_kernel(kvm))
6885 goto set_irqchip_out;
6886 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6893 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6896 if (!kvm->arch.vpit)
6898 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6902 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6909 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6911 mutex_lock(&kvm->lock);
6913 if (!kvm->arch.vpit)
6915 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6917 mutex_unlock(&kvm->lock);
6920 case KVM_GET_PIT2: {
6922 if (!kvm->arch.vpit)
6924 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6928 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6933 case KVM_SET_PIT2: {
6935 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6937 mutex_lock(&kvm->lock);
6939 if (!kvm->arch.vpit)
6941 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6943 mutex_unlock(&kvm->lock);
6946 case KVM_REINJECT_CONTROL: {
6947 struct kvm_reinject_control control;
6949 if (copy_from_user(&control, argp, sizeof(control)))
6952 if (!kvm->arch.vpit)
6954 r = kvm_vm_ioctl_reinject(kvm, &control);
6957 case KVM_SET_BOOT_CPU_ID:
6959 mutex_lock(&kvm->lock);
6960 if (kvm->created_vcpus)
6963 kvm->arch.bsp_vcpu_id = arg;
6964 mutex_unlock(&kvm->lock);
6966 #ifdef CONFIG_KVM_XEN
6967 case KVM_XEN_HVM_CONFIG: {
6968 struct kvm_xen_hvm_config xhc;
6970 if (copy_from_user(&xhc, argp, sizeof(xhc)))
6972 r = kvm_xen_hvm_config(kvm, &xhc);
6975 case KVM_XEN_HVM_GET_ATTR: {
6976 struct kvm_xen_hvm_attr xha;
6979 if (copy_from_user(&xha, argp, sizeof(xha)))
6981 r = kvm_xen_hvm_get_attr(kvm, &xha);
6982 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6986 case KVM_XEN_HVM_SET_ATTR: {
6987 struct kvm_xen_hvm_attr xha;
6990 if (copy_from_user(&xha, argp, sizeof(xha)))
6992 r = kvm_xen_hvm_set_attr(kvm, &xha);
6995 case KVM_XEN_HVM_EVTCHN_SEND: {
6996 struct kvm_irq_routing_xen_evtchn uxe;
6999 if (copy_from_user(&uxe, argp, sizeof(uxe)))
7001 r = kvm_xen_hvm_evtchn_send(kvm, &uxe);
7006 r = kvm_vm_ioctl_set_clock(kvm, argp);
7009 r = kvm_vm_ioctl_get_clock(kvm, argp);
7011 case KVM_SET_TSC_KHZ: {
7015 user_tsc_khz = (u32)arg;
7017 if (kvm_caps.has_tsc_control &&
7018 user_tsc_khz >= kvm_caps.max_guest_tsc_khz)
7021 if (user_tsc_khz == 0)
7022 user_tsc_khz = tsc_khz;
7024 WRITE_ONCE(kvm->arch.default_tsc_khz, user_tsc_khz);
7029 case KVM_GET_TSC_KHZ: {
7030 r = READ_ONCE(kvm->arch.default_tsc_khz);
7033 case KVM_MEMORY_ENCRYPT_OP: {
7035 if (!kvm_x86_ops.mem_enc_ioctl)
7038 r = static_call(kvm_x86_mem_enc_ioctl)(kvm, argp);
7041 case KVM_MEMORY_ENCRYPT_REG_REGION: {
7042 struct kvm_enc_region region;
7045 if (copy_from_user(®ion, argp, sizeof(region)))
7049 if (!kvm_x86_ops.mem_enc_register_region)
7052 r = static_call(kvm_x86_mem_enc_register_region)(kvm, ®ion);
7055 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
7056 struct kvm_enc_region region;
7059 if (copy_from_user(®ion, argp, sizeof(region)))
7063 if (!kvm_x86_ops.mem_enc_unregister_region)
7066 r = static_call(kvm_x86_mem_enc_unregister_region)(kvm, ®ion);
7069 case KVM_HYPERV_EVENTFD: {
7070 struct kvm_hyperv_eventfd hvevfd;
7073 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
7075 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
7078 case KVM_SET_PMU_EVENT_FILTER:
7079 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
7081 case KVM_X86_SET_MSR_FILTER: {
7082 struct kvm_msr_filter __user *user_msr_filter = argp;
7083 struct kvm_msr_filter filter;
7085 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
7088 r = kvm_vm_ioctl_set_msr_filter(kvm, &filter);
7098 static void kvm_probe_feature_msr(u32 msr_index)
7100 struct kvm_msr_entry msr = {
7104 if (kvm_get_msr_feature(&msr))
7107 msr_based_features[num_msr_based_features++] = msr_index;
7110 static void kvm_probe_msr_to_save(u32 msr_index)
7114 if (rdmsr_safe(msr_index, &dummy[0], &dummy[1]))
7118 * Even MSRs that are valid in the host may not be exposed to guests in
7121 switch (msr_index) {
7122 case MSR_IA32_BNDCFGS:
7123 if (!kvm_mpx_supported())
7127 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
7128 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
7131 case MSR_IA32_UMWAIT_CONTROL:
7132 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
7135 case MSR_IA32_RTIT_CTL:
7136 case MSR_IA32_RTIT_STATUS:
7137 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
7140 case MSR_IA32_RTIT_CR3_MATCH:
7141 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
7142 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
7145 case MSR_IA32_RTIT_OUTPUT_BASE:
7146 case MSR_IA32_RTIT_OUTPUT_MASK:
7147 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
7148 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
7149 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
7152 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
7153 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
7154 (msr_index - MSR_IA32_RTIT_ADDR0_A >=
7155 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2))
7158 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR_MAX:
7159 if (msr_index - MSR_ARCH_PERFMON_PERFCTR0 >=
7160 kvm_pmu_cap.num_counters_gp)
7163 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL_MAX:
7164 if (msr_index - MSR_ARCH_PERFMON_EVENTSEL0 >=
7165 kvm_pmu_cap.num_counters_gp)
7168 case MSR_ARCH_PERFMON_FIXED_CTR0 ... MSR_ARCH_PERFMON_FIXED_CTR_MAX:
7169 if (msr_index - MSR_ARCH_PERFMON_FIXED_CTR0 >=
7170 kvm_pmu_cap.num_counters_fixed)
7173 case MSR_AMD64_PERF_CNTR_GLOBAL_CTL:
7174 case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS:
7175 case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR:
7176 if (!kvm_cpu_cap_has(X86_FEATURE_PERFMON_V2))
7180 case MSR_IA32_XFD_ERR:
7181 if (!kvm_cpu_cap_has(X86_FEATURE_XFD))
7184 case MSR_IA32_TSX_CTRL:
7185 if (!(kvm_get_arch_capabilities() & ARCH_CAP_TSX_CTRL_MSR))
7192 msrs_to_save[num_msrs_to_save++] = msr_index;
7195 static void kvm_init_msr_lists(void)
7199 BUILD_BUG_ON_MSG(KVM_PMC_MAX_FIXED != 3,
7200 "Please update the fixed PMCs in msrs_to_save_pmu[]");
7202 num_msrs_to_save = 0;
7203 num_emulated_msrs = 0;
7204 num_msr_based_features = 0;
7206 for (i = 0; i < ARRAY_SIZE(msrs_to_save_base); i++)
7207 kvm_probe_msr_to_save(msrs_to_save_base[i]);
7210 for (i = 0; i < ARRAY_SIZE(msrs_to_save_pmu); i++)
7211 kvm_probe_msr_to_save(msrs_to_save_pmu[i]);
7214 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
7215 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
7218 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
7221 for (i = KVM_FIRST_EMULATED_VMX_MSR; i <= KVM_LAST_EMULATED_VMX_MSR; i++)
7222 kvm_probe_feature_msr(i);
7224 for (i = 0; i < ARRAY_SIZE(msr_based_features_all_except_vmx); i++)
7225 kvm_probe_feature_msr(msr_based_features_all_except_vmx[i]);
7228 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
7236 if (!(lapic_in_kernel(vcpu) &&
7237 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
7238 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
7249 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
7256 if (!(lapic_in_kernel(vcpu) &&
7257 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
7259 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
7261 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
7271 void kvm_set_segment(struct kvm_vcpu *vcpu,
7272 struct kvm_segment *var, int seg)
7274 static_call(kvm_x86_set_segment)(vcpu, var, seg);
7277 void kvm_get_segment(struct kvm_vcpu *vcpu,
7278 struct kvm_segment *var, int seg)
7280 static_call(kvm_x86_get_segment)(vcpu, var, seg);
7283 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u64 access,
7284 struct x86_exception *exception)
7286 struct kvm_mmu *mmu = vcpu->arch.mmu;
7289 BUG_ON(!mmu_is_nested(vcpu));
7291 /* NPT walks are always user-walks */
7292 access |= PFERR_USER_MASK;
7293 t_gpa = mmu->gva_to_gpa(vcpu, mmu, gpa, access, exception);
7298 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
7299 struct x86_exception *exception)
7301 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7303 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7304 return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7306 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
7308 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
7309 struct x86_exception *exception)
7311 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7313 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7314 access |= PFERR_WRITE_MASK;
7315 return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7317 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
7319 /* uses this to access any guest's mapped memory without checking CPL */
7320 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
7321 struct x86_exception *exception)
7323 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7325 return mmu->gva_to_gpa(vcpu, mmu, gva, 0, exception);
7328 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
7329 struct kvm_vcpu *vcpu, u64 access,
7330 struct x86_exception *exception)
7332 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7334 int r = X86EMUL_CONTINUE;
7337 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
7338 unsigned offset = addr & (PAGE_SIZE-1);
7339 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
7342 if (gpa == INVALID_GPA)
7343 return X86EMUL_PROPAGATE_FAULT;
7344 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
7347 r = X86EMUL_IO_NEEDED;
7359 /* used for instruction fetching */
7360 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
7361 gva_t addr, void *val, unsigned int bytes,
7362 struct x86_exception *exception)
7364 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7365 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7366 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7370 /* Inline kvm_read_guest_virt_helper for speed. */
7371 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access|PFERR_FETCH_MASK,
7373 if (unlikely(gpa == INVALID_GPA))
7374 return X86EMUL_PROPAGATE_FAULT;
7376 offset = addr & (PAGE_SIZE-1);
7377 if (WARN_ON(offset + bytes > PAGE_SIZE))
7378 bytes = (unsigned)PAGE_SIZE - offset;
7379 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
7381 if (unlikely(ret < 0))
7382 return X86EMUL_IO_NEEDED;
7384 return X86EMUL_CONTINUE;
7387 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
7388 gva_t addr, void *val, unsigned int bytes,
7389 struct x86_exception *exception)
7391 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7394 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
7395 * is returned, but our callers are not ready for that and they blindly
7396 * call kvm_inject_page_fault. Ensure that they at least do not leak
7397 * uninitialized kernel stack memory into cr2 and error code.
7399 memset(exception, 0, sizeof(*exception));
7400 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
7403 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
7405 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
7406 gva_t addr, void *val, unsigned int bytes,
7407 struct x86_exception *exception, bool system)
7409 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7413 access |= PFERR_IMPLICIT_ACCESS;
7414 else if (static_call(kvm_x86_get_cpl)(vcpu) == 3)
7415 access |= PFERR_USER_MASK;
7417 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
7420 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
7421 struct kvm_vcpu *vcpu, u64 access,
7422 struct x86_exception *exception)
7424 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7426 int r = X86EMUL_CONTINUE;
7429 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
7430 unsigned offset = addr & (PAGE_SIZE-1);
7431 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
7434 if (gpa == INVALID_GPA)
7435 return X86EMUL_PROPAGATE_FAULT;
7436 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
7438 r = X86EMUL_IO_NEEDED;
7450 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
7451 unsigned int bytes, struct x86_exception *exception,
7454 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7455 u64 access = PFERR_WRITE_MASK;
7458 access |= PFERR_IMPLICIT_ACCESS;
7459 else if (static_call(kvm_x86_get_cpl)(vcpu) == 3)
7460 access |= PFERR_USER_MASK;
7462 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
7466 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
7467 unsigned int bytes, struct x86_exception *exception)
7469 /* kvm_write_guest_virt_system can pull in tons of pages. */
7470 vcpu->arch.l1tf_flush_l1d = true;
7472 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
7473 PFERR_WRITE_MASK, exception);
7475 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
7477 static int kvm_can_emulate_insn(struct kvm_vcpu *vcpu, int emul_type,
7478 void *insn, int insn_len)
7480 return static_call(kvm_x86_can_emulate_instruction)(vcpu, emul_type,
7484 int handle_ud(struct kvm_vcpu *vcpu)
7486 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
7487 int fep_flags = READ_ONCE(force_emulation_prefix);
7488 int emul_type = EMULTYPE_TRAP_UD;
7489 char sig[5]; /* ud2; .ascii "kvm" */
7490 struct x86_exception e;
7492 if (unlikely(!kvm_can_emulate_insn(vcpu, emul_type, NULL, 0)))
7496 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
7497 sig, sizeof(sig), &e) == 0 &&
7498 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
7499 if (fep_flags & KVM_FEP_CLEAR_RFLAGS_RF)
7500 kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) & ~X86_EFLAGS_RF);
7501 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
7502 emul_type = EMULTYPE_TRAP_UD_FORCED;
7505 return kvm_emulate_instruction(vcpu, emul_type);
7507 EXPORT_SYMBOL_GPL(handle_ud);
7509 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
7510 gpa_t gpa, bool write)
7512 /* For APIC access vmexit */
7513 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7516 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
7517 trace_vcpu_match_mmio(gva, gpa, write, true);
7524 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
7525 gpa_t *gpa, struct x86_exception *exception,
7528 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7529 u64 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
7530 | (write ? PFERR_WRITE_MASK : 0);
7533 * currently PKRU is only applied to ept enabled guest so
7534 * there is no pkey in EPT page table for L1 guest or EPT
7535 * shadow page table for L2 guest.
7537 if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
7538 !permission_fault(vcpu, vcpu->arch.walk_mmu,
7539 vcpu->arch.mmio_access, 0, access))) {
7540 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
7541 (gva & (PAGE_SIZE - 1));
7542 trace_vcpu_match_mmio(gva, *gpa, write, false);
7546 *gpa = mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7548 if (*gpa == INVALID_GPA)
7551 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
7554 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
7555 const void *val, int bytes)
7559 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
7562 kvm_page_track_write(vcpu, gpa, val, bytes);
7566 struct read_write_emulator_ops {
7567 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
7569 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
7570 void *val, int bytes);
7571 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
7572 int bytes, void *val);
7573 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
7574 void *val, int bytes);
7578 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
7580 if (vcpu->mmio_read_completed) {
7581 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
7582 vcpu->mmio_fragments[0].gpa, val);
7583 vcpu->mmio_read_completed = 0;
7590 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
7591 void *val, int bytes)
7593 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
7596 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
7597 void *val, int bytes)
7599 return emulator_write_phys(vcpu, gpa, val, bytes);
7602 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
7604 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
7605 return vcpu_mmio_write(vcpu, gpa, bytes, val);
7608 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
7609 void *val, int bytes)
7611 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
7612 return X86EMUL_IO_NEEDED;
7615 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
7616 void *val, int bytes)
7618 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
7620 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
7621 return X86EMUL_CONTINUE;
7624 static const struct read_write_emulator_ops read_emultor = {
7625 .read_write_prepare = read_prepare,
7626 .read_write_emulate = read_emulate,
7627 .read_write_mmio = vcpu_mmio_read,
7628 .read_write_exit_mmio = read_exit_mmio,
7631 static const struct read_write_emulator_ops write_emultor = {
7632 .read_write_emulate = write_emulate,
7633 .read_write_mmio = write_mmio,
7634 .read_write_exit_mmio = write_exit_mmio,
7638 static int emulator_read_write_onepage(unsigned long addr, void *val,
7640 struct x86_exception *exception,
7641 struct kvm_vcpu *vcpu,
7642 const struct read_write_emulator_ops *ops)
7646 bool write = ops->write;
7647 struct kvm_mmio_fragment *frag;
7648 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7651 * If the exit was due to a NPF we may already have a GPA.
7652 * If the GPA is present, use it to avoid the GVA to GPA table walk.
7653 * Note, this cannot be used on string operations since string
7654 * operation using rep will only have the initial GPA from the NPF
7657 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
7658 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
7659 gpa = ctxt->gpa_val;
7660 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
7662 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
7664 return X86EMUL_PROPAGATE_FAULT;
7667 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
7668 return X86EMUL_CONTINUE;
7671 * Is this MMIO handled locally?
7673 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
7674 if (handled == bytes)
7675 return X86EMUL_CONTINUE;
7681 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
7682 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
7686 return X86EMUL_CONTINUE;
7689 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
7691 void *val, unsigned int bytes,
7692 struct x86_exception *exception,
7693 const struct read_write_emulator_ops *ops)
7695 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7699 if (ops->read_write_prepare &&
7700 ops->read_write_prepare(vcpu, val, bytes))
7701 return X86EMUL_CONTINUE;
7703 vcpu->mmio_nr_fragments = 0;
7705 /* Crossing a page boundary? */
7706 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
7709 now = -addr & ~PAGE_MASK;
7710 rc = emulator_read_write_onepage(addr, val, now, exception,
7713 if (rc != X86EMUL_CONTINUE)
7716 if (ctxt->mode != X86EMUL_MODE_PROT64)
7722 rc = emulator_read_write_onepage(addr, val, bytes, exception,
7724 if (rc != X86EMUL_CONTINUE)
7727 if (!vcpu->mmio_nr_fragments)
7730 gpa = vcpu->mmio_fragments[0].gpa;
7732 vcpu->mmio_needed = 1;
7733 vcpu->mmio_cur_fragment = 0;
7735 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
7736 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
7737 vcpu->run->exit_reason = KVM_EXIT_MMIO;
7738 vcpu->run->mmio.phys_addr = gpa;
7740 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
7743 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
7747 struct x86_exception *exception)
7749 return emulator_read_write(ctxt, addr, val, bytes,
7750 exception, &read_emultor);
7753 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
7757 struct x86_exception *exception)
7759 return emulator_read_write(ctxt, addr, (void *)val, bytes,
7760 exception, &write_emultor);
7763 #define emulator_try_cmpxchg_user(t, ptr, old, new) \
7764 (__try_cmpxchg_user((t __user *)(ptr), (t *)(old), *(t *)(new), efault ## t))
7766 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
7771 struct x86_exception *exception)
7773 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7779 /* guests cmpxchg8b have to be emulated atomically */
7780 if (bytes > 8 || (bytes & (bytes - 1)))
7783 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
7785 if (gpa == INVALID_GPA ||
7786 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7790 * Emulate the atomic as a straight write to avoid #AC if SLD is
7791 * enabled in the host and the access splits a cache line.
7793 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
7794 page_line_mask = ~(cache_line_size() - 1);
7796 page_line_mask = PAGE_MASK;
7798 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
7801 hva = kvm_vcpu_gfn_to_hva(vcpu, gpa_to_gfn(gpa));
7802 if (kvm_is_error_hva(hva))
7805 hva += offset_in_page(gpa);
7809 r = emulator_try_cmpxchg_user(u8, hva, old, new);
7812 r = emulator_try_cmpxchg_user(u16, hva, old, new);
7815 r = emulator_try_cmpxchg_user(u32, hva, old, new);
7818 r = emulator_try_cmpxchg_user(u64, hva, old, new);
7825 return X86EMUL_UNHANDLEABLE;
7827 return X86EMUL_CMPXCHG_FAILED;
7829 kvm_page_track_write(vcpu, gpa, new, bytes);
7831 return X86EMUL_CONTINUE;
7834 pr_warn_once("emulating exchange as write\n");
7836 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7839 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7840 unsigned short port, void *data,
7841 unsigned int count, bool in)
7846 WARN_ON_ONCE(vcpu->arch.pio.count);
7847 for (i = 0; i < count; i++) {
7849 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, port, size, data);
7851 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS, port, size, data);
7858 * Userspace must have unregistered the device while PIO
7859 * was running. Drop writes / read as 0.
7862 memset(data, 0, size * (count - i));
7871 vcpu->arch.pio.port = port;
7872 vcpu->arch.pio.in = in;
7873 vcpu->arch.pio.count = count;
7874 vcpu->arch.pio.size = size;
7877 memset(vcpu->arch.pio_data, 0, size * count);
7879 memcpy(vcpu->arch.pio_data, data, size * count);
7881 vcpu->run->exit_reason = KVM_EXIT_IO;
7882 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7883 vcpu->run->io.size = size;
7884 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7885 vcpu->run->io.count = count;
7886 vcpu->run->io.port = port;
7890 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7891 unsigned short port, void *val, unsigned int count)
7893 int r = emulator_pio_in_out(vcpu, size, port, val, count, true);
7895 trace_kvm_pio(KVM_PIO_IN, port, size, count, val);
7900 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7902 int size = vcpu->arch.pio.size;
7903 unsigned int count = vcpu->arch.pio.count;
7904 memcpy(val, vcpu->arch.pio_data, size * count);
7905 trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7906 vcpu->arch.pio.count = 0;
7909 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7910 int size, unsigned short port, void *val,
7913 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7914 if (vcpu->arch.pio.count) {
7916 * Complete a previous iteration that required userspace I/O.
7917 * Note, @count isn't guaranteed to match pio.count as userspace
7918 * can modify ECX before rerunning the vCPU. Ignore any such
7919 * shenanigans as KVM doesn't support modifying the rep count,
7920 * and the emulator ensures @count doesn't overflow the buffer.
7922 complete_emulator_pio_in(vcpu, val);
7926 return emulator_pio_in(vcpu, size, port, val, count);
7929 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7930 unsigned short port, const void *val,
7933 trace_kvm_pio(KVM_PIO_OUT, port, size, count, val);
7934 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
7937 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7938 int size, unsigned short port,
7939 const void *val, unsigned int count)
7941 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7944 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7946 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7949 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7951 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7954 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7956 if (!need_emulate_wbinvd(vcpu))
7957 return X86EMUL_CONTINUE;
7959 if (static_call(kvm_x86_has_wbinvd_exit)()) {
7960 int cpu = get_cpu();
7962 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7963 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7964 wbinvd_ipi, NULL, 1);
7966 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7969 return X86EMUL_CONTINUE;
7972 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7974 kvm_emulate_wbinvd_noskip(vcpu);
7975 return kvm_skip_emulated_instruction(vcpu);
7977 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7981 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7983 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7986 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7987 unsigned long *dest)
7989 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7992 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7993 unsigned long value)
7996 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7999 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
8001 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
8004 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
8006 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8007 unsigned long value;
8011 value = kvm_read_cr0(vcpu);
8014 value = vcpu->arch.cr2;
8017 value = kvm_read_cr3(vcpu);
8020 value = kvm_read_cr4(vcpu);
8023 value = kvm_get_cr8(vcpu);
8026 kvm_err("%s: unexpected cr %u\n", __func__, cr);
8033 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
8035 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8040 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
8043 vcpu->arch.cr2 = val;
8046 res = kvm_set_cr3(vcpu, val);
8049 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
8052 res = kvm_set_cr8(vcpu, val);
8055 kvm_err("%s: unexpected cr %u\n", __func__, cr);
8062 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
8064 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
8067 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
8069 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
8072 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
8074 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
8077 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
8079 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
8082 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
8084 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
8087 static unsigned long emulator_get_cached_segment_base(
8088 struct x86_emulate_ctxt *ctxt, int seg)
8090 return get_segment_base(emul_to_vcpu(ctxt), seg);
8093 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
8094 struct desc_struct *desc, u32 *base3,
8097 struct kvm_segment var;
8099 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
8100 *selector = var.selector;
8103 memset(desc, 0, sizeof(*desc));
8111 set_desc_limit(desc, var.limit);
8112 set_desc_base(desc, (unsigned long)var.base);
8113 #ifdef CONFIG_X86_64
8115 *base3 = var.base >> 32;
8117 desc->type = var.type;
8119 desc->dpl = var.dpl;
8120 desc->p = var.present;
8121 desc->avl = var.avl;
8129 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
8130 struct desc_struct *desc, u32 base3,
8133 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8134 struct kvm_segment var;
8136 var.selector = selector;
8137 var.base = get_desc_base(desc);
8138 #ifdef CONFIG_X86_64
8139 var.base |= ((u64)base3) << 32;
8141 var.limit = get_desc_limit(desc);
8143 var.limit = (var.limit << 12) | 0xfff;
8144 var.type = desc->type;
8145 var.dpl = desc->dpl;
8150 var.avl = desc->avl;
8151 var.present = desc->p;
8152 var.unusable = !var.present;
8155 kvm_set_segment(vcpu, &var, seg);
8159 static int emulator_get_msr_with_filter(struct x86_emulate_ctxt *ctxt,
8160 u32 msr_index, u64 *pdata)
8162 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8165 r = kvm_get_msr_with_filter(vcpu, msr_index, pdata);
8167 return X86EMUL_UNHANDLEABLE;
8170 if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
8171 complete_emulated_rdmsr, r))
8172 return X86EMUL_IO_NEEDED;
8174 trace_kvm_msr_read_ex(msr_index);
8175 return X86EMUL_PROPAGATE_FAULT;
8178 trace_kvm_msr_read(msr_index, *pdata);
8179 return X86EMUL_CONTINUE;
8182 static int emulator_set_msr_with_filter(struct x86_emulate_ctxt *ctxt,
8183 u32 msr_index, u64 data)
8185 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8188 r = kvm_set_msr_with_filter(vcpu, msr_index, data);
8190 return X86EMUL_UNHANDLEABLE;
8193 if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
8194 complete_emulated_msr_access, r))
8195 return X86EMUL_IO_NEEDED;
8197 trace_kvm_msr_write_ex(msr_index, data);
8198 return X86EMUL_PROPAGATE_FAULT;
8201 trace_kvm_msr_write(msr_index, data);
8202 return X86EMUL_CONTINUE;
8205 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
8206 u32 msr_index, u64 *pdata)
8208 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
8211 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
8214 if (kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc))
8219 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
8220 u32 pmc, u64 *pdata)
8222 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
8225 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
8227 emul_to_vcpu(ctxt)->arch.halt_request = 1;
8230 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8231 struct x86_instruction_info *info,
8232 enum x86_intercept_stage stage)
8234 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
8238 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
8239 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
8242 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
8245 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
8247 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
8250 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
8252 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
8255 static bool emulator_guest_has_rdpid(struct x86_emulate_ctxt *ctxt)
8257 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_RDPID);
8260 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
8262 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
8265 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
8267 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
8270 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
8272 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
8275 static bool emulator_is_smm(struct x86_emulate_ctxt *ctxt)
8277 return is_smm(emul_to_vcpu(ctxt));
8280 static bool emulator_is_guest_mode(struct x86_emulate_ctxt *ctxt)
8282 return is_guest_mode(emul_to_vcpu(ctxt));
8285 #ifndef CONFIG_KVM_SMM
8286 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt)
8289 return X86EMUL_UNHANDLEABLE;
8293 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
8295 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
8298 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
8300 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
8303 static void emulator_vm_bugged(struct x86_emulate_ctxt *ctxt)
8305 struct kvm *kvm = emul_to_vcpu(ctxt)->kvm;
8307 if (!kvm->vm_bugged)
8311 static const struct x86_emulate_ops emulate_ops = {
8312 .vm_bugged = emulator_vm_bugged,
8313 .read_gpr = emulator_read_gpr,
8314 .write_gpr = emulator_write_gpr,
8315 .read_std = emulator_read_std,
8316 .write_std = emulator_write_std,
8317 .fetch = kvm_fetch_guest_virt,
8318 .read_emulated = emulator_read_emulated,
8319 .write_emulated = emulator_write_emulated,
8320 .cmpxchg_emulated = emulator_cmpxchg_emulated,
8321 .invlpg = emulator_invlpg,
8322 .pio_in_emulated = emulator_pio_in_emulated,
8323 .pio_out_emulated = emulator_pio_out_emulated,
8324 .get_segment = emulator_get_segment,
8325 .set_segment = emulator_set_segment,
8326 .get_cached_segment_base = emulator_get_cached_segment_base,
8327 .get_gdt = emulator_get_gdt,
8328 .get_idt = emulator_get_idt,
8329 .set_gdt = emulator_set_gdt,
8330 .set_idt = emulator_set_idt,
8331 .get_cr = emulator_get_cr,
8332 .set_cr = emulator_set_cr,
8333 .cpl = emulator_get_cpl,
8334 .get_dr = emulator_get_dr,
8335 .set_dr = emulator_set_dr,
8336 .set_msr_with_filter = emulator_set_msr_with_filter,
8337 .get_msr_with_filter = emulator_get_msr_with_filter,
8338 .get_msr = emulator_get_msr,
8339 .check_pmc = emulator_check_pmc,
8340 .read_pmc = emulator_read_pmc,
8341 .halt = emulator_halt,
8342 .wbinvd = emulator_wbinvd,
8343 .fix_hypercall = emulator_fix_hypercall,
8344 .intercept = emulator_intercept,
8345 .get_cpuid = emulator_get_cpuid,
8346 .guest_has_movbe = emulator_guest_has_movbe,
8347 .guest_has_fxsr = emulator_guest_has_fxsr,
8348 .guest_has_rdpid = emulator_guest_has_rdpid,
8349 .set_nmi_mask = emulator_set_nmi_mask,
8350 .is_smm = emulator_is_smm,
8351 .is_guest_mode = emulator_is_guest_mode,
8352 .leave_smm = emulator_leave_smm,
8353 .triple_fault = emulator_triple_fault,
8354 .set_xcr = emulator_set_xcr,
8357 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
8359 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
8361 * an sti; sti; sequence only disable interrupts for the first
8362 * instruction. So, if the last instruction, be it emulated or
8363 * not, left the system with the INT_STI flag enabled, it
8364 * means that the last instruction is an sti. We should not
8365 * leave the flag on in this case. The same goes for mov ss
8367 if (int_shadow & mask)
8369 if (unlikely(int_shadow || mask)) {
8370 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
8372 kvm_make_request(KVM_REQ_EVENT, vcpu);
8376 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
8378 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8380 if (ctxt->exception.vector == PF_VECTOR)
8381 kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
8382 else if (ctxt->exception.error_code_valid)
8383 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
8384 ctxt->exception.error_code);
8386 kvm_queue_exception(vcpu, ctxt->exception.vector);
8389 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
8391 struct x86_emulate_ctxt *ctxt;
8393 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
8395 pr_err("failed to allocate vcpu's emulator\n");
8400 ctxt->ops = &emulate_ops;
8401 vcpu->arch.emulate_ctxt = ctxt;
8406 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
8408 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8411 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
8413 ctxt->gpa_available = false;
8414 ctxt->eflags = kvm_get_rflags(vcpu);
8415 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
8417 ctxt->eip = kvm_rip_read(vcpu);
8418 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
8419 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
8420 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
8421 cs_db ? X86EMUL_MODE_PROT32 :
8422 X86EMUL_MODE_PROT16;
8423 ctxt->interruptibility = 0;
8424 ctxt->have_exception = false;
8425 ctxt->exception.vector = -1;
8426 ctxt->perm_ok = false;
8428 init_decode_cache(ctxt);
8429 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8432 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
8434 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8437 init_emulate_ctxt(vcpu);
8441 ctxt->_eip = ctxt->eip + inc_eip;
8442 ret = emulate_int_real(ctxt, irq);
8444 if (ret != X86EMUL_CONTINUE) {
8445 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
8447 ctxt->eip = ctxt->_eip;
8448 kvm_rip_write(vcpu, ctxt->eip);
8449 kvm_set_rflags(vcpu, ctxt->eflags);
8452 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
8454 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
8455 u8 ndata, u8 *insn_bytes, u8 insn_size)
8457 struct kvm_run *run = vcpu->run;
8462 * Zero the whole array used to retrieve the exit info, as casting to
8463 * u32 for select entries will leave some chunks uninitialized.
8465 memset(&info, 0, sizeof(info));
8467 static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
8468 &info[2], (u32 *)&info[3],
8471 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8472 run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
8475 * There's currently space for 13 entries, but 5 are used for the exit
8476 * reason and info. Restrict to 4 to reduce the maintenance burden
8477 * when expanding kvm_run.emulation_failure in the future.
8479 if (WARN_ON_ONCE(ndata > 4))
8482 /* Always include the flags as a 'data' entry. */
8484 run->emulation_failure.flags = 0;
8487 BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
8488 sizeof(run->emulation_failure.insn_bytes) != 16));
8490 run->emulation_failure.flags |=
8491 KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
8492 run->emulation_failure.insn_size = insn_size;
8493 memset(run->emulation_failure.insn_bytes, 0x90,
8494 sizeof(run->emulation_failure.insn_bytes));
8495 memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
8498 memcpy(&run->internal.data[info_start], info, sizeof(info));
8499 memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
8500 ndata * sizeof(data[0]));
8502 run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
8505 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
8507 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8509 prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
8510 ctxt->fetch.end - ctxt->fetch.data);
8513 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
8516 prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
8518 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
8520 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
8522 __kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
8524 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
8526 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
8528 struct kvm *kvm = vcpu->kvm;
8530 ++vcpu->stat.insn_emulation_fail;
8531 trace_kvm_emulate_insn_failed(vcpu);
8533 if (emulation_type & EMULTYPE_VMWARE_GP) {
8534 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8538 if (kvm->arch.exit_on_emulation_error ||
8539 (emulation_type & EMULTYPE_SKIP)) {
8540 prepare_emulation_ctxt_failure_exit(vcpu);
8544 kvm_queue_exception(vcpu, UD_VECTOR);
8546 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
8547 prepare_emulation_ctxt_failure_exit(vcpu);
8554 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8557 gpa_t gpa = cr2_or_gpa;
8560 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8563 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8564 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8567 if (!vcpu->arch.mmu->root_role.direct) {
8569 * Write permission should be allowed since only
8570 * write access need to be emulated.
8572 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8575 * If the mapping is invalid in guest, let cpu retry
8576 * it to generate fault.
8578 if (gpa == INVALID_GPA)
8583 * Do not retry the unhandleable instruction if it faults on the
8584 * readonly host memory, otherwise it will goto a infinite loop:
8585 * retry instruction -> write #PF -> emulation fail -> retry
8586 * instruction -> ...
8588 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
8591 * If the instruction failed on the error pfn, it can not be fixed,
8592 * report the error to userspace.
8594 if (is_error_noslot_pfn(pfn))
8597 kvm_release_pfn_clean(pfn);
8599 /* The instructions are well-emulated on direct mmu. */
8600 if (vcpu->arch.mmu->root_role.direct) {
8601 unsigned int indirect_shadow_pages;
8603 write_lock(&vcpu->kvm->mmu_lock);
8604 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
8605 write_unlock(&vcpu->kvm->mmu_lock);
8607 if (indirect_shadow_pages)
8608 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8614 * if emulation was due to access to shadowed page table
8615 * and it failed try to unshadow page and re-enter the
8616 * guest to let CPU execute the instruction.
8618 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8621 * If the access faults on its page table, it can not
8622 * be fixed by unprotecting shadow page and it should
8623 * be reported to userspace.
8625 return !(emulation_type & EMULTYPE_WRITE_PF_TO_SP);
8628 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
8629 gpa_t cr2_or_gpa, int emulation_type)
8631 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8632 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
8634 last_retry_eip = vcpu->arch.last_retry_eip;
8635 last_retry_addr = vcpu->arch.last_retry_addr;
8638 * If the emulation is caused by #PF and it is non-page_table
8639 * writing instruction, it means the VM-EXIT is caused by shadow
8640 * page protected, we can zap the shadow page and retry this
8641 * instruction directly.
8643 * Note: if the guest uses a non-page-table modifying instruction
8644 * on the PDE that points to the instruction, then we will unmap
8645 * the instruction and go to an infinite loop. So, we cache the
8646 * last retried eip and the last fault address, if we meet the eip
8647 * and the address again, we can break out of the potential infinite
8650 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
8652 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8655 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8656 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8659 if (x86_page_table_writing_insn(ctxt))
8662 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
8665 vcpu->arch.last_retry_eip = ctxt->eip;
8666 vcpu->arch.last_retry_addr = cr2_or_gpa;
8668 if (!vcpu->arch.mmu->root_role.direct)
8669 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8671 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8676 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
8677 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
8679 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
8688 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
8689 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
8694 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
8696 struct kvm_run *kvm_run = vcpu->run;
8698 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
8699 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
8700 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
8701 kvm_run->debug.arch.exception = DB_VECTOR;
8702 kvm_run->exit_reason = KVM_EXIT_DEBUG;
8705 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
8709 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
8711 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8714 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
8718 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8721 * rflags is the old, "raw" value of the flags. The new value has
8722 * not been saved yet.
8724 * This is correct even for TF set by the guest, because "the
8725 * processor will not generate this exception after the instruction
8726 * that sets the TF flag".
8728 if (unlikely(rflags & X86_EFLAGS_TF))
8729 r = kvm_vcpu_do_singlestep(vcpu);
8732 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
8734 static bool kvm_is_code_breakpoint_inhibited(struct kvm_vcpu *vcpu)
8738 if (kvm_get_rflags(vcpu) & X86_EFLAGS_RF)
8742 * Intel CPUs inhibit code #DBs when MOV/POP SS blocking is active,
8743 * but AMD CPUs do not. MOV/POP SS blocking is rare, check that first
8744 * to avoid the relatively expensive CPUID lookup.
8746 shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
8747 return (shadow & KVM_X86_SHADOW_INT_MOV_SS) &&
8748 guest_cpuid_is_intel(vcpu);
8751 static bool kvm_vcpu_check_code_breakpoint(struct kvm_vcpu *vcpu,
8752 int emulation_type, int *r)
8754 WARN_ON_ONCE(emulation_type & EMULTYPE_NO_DECODE);
8757 * Do not check for code breakpoints if hardware has already done the
8758 * checks, as inferred from the emulation type. On NO_DECODE and SKIP,
8759 * the instruction has passed all exception checks, and all intercepted
8760 * exceptions that trigger emulation have lower priority than code
8761 * breakpoints, i.e. the fact that the intercepted exception occurred
8762 * means any code breakpoints have already been serviced.
8764 * Note, KVM needs to check for code #DBs on EMULTYPE_TRAP_UD_FORCED as
8765 * hardware has checked the RIP of the magic prefix, but not the RIP of
8766 * the instruction being emulated. The intent of forced emulation is
8767 * to behave as if KVM intercepted the instruction without an exception
8768 * and without a prefix.
8770 if (emulation_type & (EMULTYPE_NO_DECODE | EMULTYPE_SKIP |
8771 EMULTYPE_TRAP_UD | EMULTYPE_VMWARE_GP | EMULTYPE_PF))
8774 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
8775 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
8776 struct kvm_run *kvm_run = vcpu->run;
8777 unsigned long eip = kvm_get_linear_rip(vcpu);
8778 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8779 vcpu->arch.guest_debug_dr7,
8783 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
8784 kvm_run->debug.arch.pc = eip;
8785 kvm_run->debug.arch.exception = DB_VECTOR;
8786 kvm_run->exit_reason = KVM_EXIT_DEBUG;
8792 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
8793 !kvm_is_code_breakpoint_inhibited(vcpu)) {
8794 unsigned long eip = kvm_get_linear_rip(vcpu);
8795 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8800 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
8809 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
8811 switch (ctxt->opcode_len) {
8818 case 0xe6: /* OUT */
8822 case 0x6c: /* INS */
8824 case 0x6e: /* OUTS */
8831 case 0x33: /* RDPMC */
8841 * Decode an instruction for emulation. The caller is responsible for handling
8842 * code breakpoints. Note, manually detecting code breakpoints is unnecessary
8843 * (and wrong) when emulating on an intercepted fault-like exception[*], as
8844 * code breakpoints have higher priority and thus have already been done by
8847 * [*] Except #MC, which is higher priority, but KVM should never emulate in
8848 * response to a machine check.
8850 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8851 void *insn, int insn_len)
8853 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8856 init_emulate_ctxt(vcpu);
8858 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8860 trace_kvm_emulate_insn_start(vcpu);
8861 ++vcpu->stat.insn_emulation;
8865 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8867 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8868 int emulation_type, void *insn, int insn_len)
8871 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8872 bool writeback = true;
8874 if (unlikely(!kvm_can_emulate_insn(vcpu, emulation_type, insn, insn_len)))
8877 vcpu->arch.l1tf_flush_l1d = true;
8879 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8880 kvm_clear_exception_queue(vcpu);
8883 * Return immediately if RIP hits a code breakpoint, such #DBs
8884 * are fault-like and are higher priority than any faults on
8885 * the code fetch itself.
8887 if (kvm_vcpu_check_code_breakpoint(vcpu, emulation_type, &r))
8890 r = x86_decode_emulated_instruction(vcpu, emulation_type,
8892 if (r != EMULATION_OK) {
8893 if ((emulation_type & EMULTYPE_TRAP_UD) ||
8894 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8895 kvm_queue_exception(vcpu, UD_VECTOR);
8898 if (reexecute_instruction(vcpu, cr2_or_gpa,
8902 if (ctxt->have_exception &&
8903 !(emulation_type & EMULTYPE_SKIP)) {
8905 * #UD should result in just EMULATION_FAILED, and trap-like
8906 * exception should not be encountered during decode.
8908 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8909 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8910 inject_emulated_exception(vcpu);
8913 return handle_emulation_failure(vcpu, emulation_type);
8917 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8918 !is_vmware_backdoor_opcode(ctxt)) {
8919 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8924 * EMULTYPE_SKIP without EMULTYPE_COMPLETE_USER_EXIT is intended for
8925 * use *only* by vendor callbacks for kvm_skip_emulated_instruction().
8926 * The caller is responsible for updating interruptibility state and
8927 * injecting single-step #DBs.
8929 if (emulation_type & EMULTYPE_SKIP) {
8930 if (ctxt->mode != X86EMUL_MODE_PROT64)
8931 ctxt->eip = (u32)ctxt->_eip;
8933 ctxt->eip = ctxt->_eip;
8935 if (emulation_type & EMULTYPE_COMPLETE_USER_EXIT) {
8940 kvm_rip_write(vcpu, ctxt->eip);
8941 if (ctxt->eflags & X86_EFLAGS_RF)
8942 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8946 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8949 /* this is needed for vmware backdoor interface to work since it
8950 changes registers values during IO operation */
8951 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8952 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8953 emulator_invalidate_register_cache(ctxt);
8957 if (emulation_type & EMULTYPE_PF) {
8958 /* Save the faulting GPA (cr2) in the address field */
8959 ctxt->exception.address = cr2_or_gpa;
8961 /* With shadow page tables, cr2 contains a GVA or nGPA. */
8962 if (vcpu->arch.mmu->root_role.direct) {
8963 ctxt->gpa_available = true;
8964 ctxt->gpa_val = cr2_or_gpa;
8967 /* Sanitize the address out of an abundance of paranoia. */
8968 ctxt->exception.address = 0;
8971 r = x86_emulate_insn(ctxt);
8973 if (r == EMULATION_INTERCEPTED)
8976 if (r == EMULATION_FAILED) {
8977 if (reexecute_instruction(vcpu, cr2_or_gpa, emulation_type))
8980 return handle_emulation_failure(vcpu, emulation_type);
8983 if (ctxt->have_exception) {
8984 WARN_ON_ONCE(vcpu->mmio_needed && !vcpu->mmio_is_write);
8985 vcpu->mmio_needed = false;
8987 inject_emulated_exception(vcpu);
8988 } else if (vcpu->arch.pio.count) {
8989 if (!vcpu->arch.pio.in) {
8990 /* FIXME: return into emulator if single-stepping. */
8991 vcpu->arch.pio.count = 0;
8994 vcpu->arch.complete_userspace_io = complete_emulated_pio;
8997 } else if (vcpu->mmio_needed) {
8998 ++vcpu->stat.mmio_exits;
9000 if (!vcpu->mmio_is_write)
9003 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9004 } else if (vcpu->arch.complete_userspace_io) {
9007 } else if (r == EMULATION_RESTART)
9014 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
9015 toggle_interruptibility(vcpu, ctxt->interruptibility);
9016 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9019 * Note, EXCPT_DB is assumed to be fault-like as the emulator
9020 * only supports code breakpoints and general detect #DB, both
9021 * of which are fault-like.
9023 if (!ctxt->have_exception ||
9024 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
9025 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
9026 if (ctxt->is_branch)
9027 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
9028 kvm_rip_write(vcpu, ctxt->eip);
9029 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
9030 r = kvm_vcpu_do_singlestep(vcpu);
9031 static_call_cond(kvm_x86_update_emulated_instruction)(vcpu);
9032 __kvm_set_rflags(vcpu, ctxt->eflags);
9036 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
9037 * do nothing, and it will be requested again as soon as
9038 * the shadow expires. But we still need to check here,
9039 * because POPF has no interrupt shadow.
9041 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
9042 kvm_make_request(KVM_REQ_EVENT, vcpu);
9044 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
9049 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
9051 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
9053 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
9055 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
9056 void *insn, int insn_len)
9058 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
9060 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
9062 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
9064 vcpu->arch.pio.count = 0;
9068 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
9070 vcpu->arch.pio.count = 0;
9072 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
9075 return kvm_skip_emulated_instruction(vcpu);
9078 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
9079 unsigned short port)
9081 unsigned long val = kvm_rax_read(vcpu);
9082 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
9088 * Workaround userspace that relies on old KVM behavior of %rip being
9089 * incremented prior to exiting to userspace to handle "OUT 0x7e".
9092 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
9093 vcpu->arch.complete_userspace_io =
9094 complete_fast_pio_out_port_0x7e;
9095 kvm_skip_emulated_instruction(vcpu);
9097 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
9098 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
9103 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
9107 /* We should only ever be called with arch.pio.count equal to 1 */
9108 BUG_ON(vcpu->arch.pio.count != 1);
9110 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
9111 vcpu->arch.pio.count = 0;
9115 /* For size less than 4 we merge, else we zero extend */
9116 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
9118 complete_emulator_pio_in(vcpu, &val);
9119 kvm_rax_write(vcpu, val);
9121 return kvm_skip_emulated_instruction(vcpu);
9124 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
9125 unsigned short port)
9130 /* For size less than 4 we merge, else we zero extend */
9131 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
9133 ret = emulator_pio_in(vcpu, size, port, &val, 1);
9135 kvm_rax_write(vcpu, val);
9139 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
9140 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
9145 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
9150 ret = kvm_fast_pio_in(vcpu, size, port);
9152 ret = kvm_fast_pio_out(vcpu, size, port);
9153 return ret && kvm_skip_emulated_instruction(vcpu);
9155 EXPORT_SYMBOL_GPL(kvm_fast_pio);
9157 static int kvmclock_cpu_down_prep(unsigned int cpu)
9159 __this_cpu_write(cpu_tsc_khz, 0);
9163 static void tsc_khz_changed(void *data)
9165 struct cpufreq_freqs *freq = data;
9168 WARN_ON_ONCE(boot_cpu_has(X86_FEATURE_CONSTANT_TSC));
9173 khz = cpufreq_quick_get(raw_smp_processor_id());
9176 __this_cpu_write(cpu_tsc_khz, khz);
9179 #ifdef CONFIG_X86_64
9180 static void kvm_hyperv_tsc_notifier(void)
9185 mutex_lock(&kvm_lock);
9186 list_for_each_entry(kvm, &vm_list, vm_list)
9187 kvm_make_mclock_inprogress_request(kvm);
9189 /* no guest entries from this point */
9190 hyperv_stop_tsc_emulation();
9192 /* TSC frequency always matches when on Hyper-V */
9193 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9194 for_each_present_cpu(cpu)
9195 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
9197 kvm_caps.max_guest_tsc_khz = tsc_khz;
9199 list_for_each_entry(kvm, &vm_list, vm_list) {
9200 __kvm_start_pvclock_update(kvm);
9201 pvclock_update_vm_gtod_copy(kvm);
9202 kvm_end_pvclock_update(kvm);
9205 mutex_unlock(&kvm_lock);
9209 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
9212 struct kvm_vcpu *vcpu;
9217 * We allow guests to temporarily run on slowing clocks,
9218 * provided we notify them after, or to run on accelerating
9219 * clocks, provided we notify them before. Thus time never
9222 * However, we have a problem. We can't atomically update
9223 * the frequency of a given CPU from this function; it is
9224 * merely a notifier, which can be called from any CPU.
9225 * Changing the TSC frequency at arbitrary points in time
9226 * requires a recomputation of local variables related to
9227 * the TSC for each VCPU. We must flag these local variables
9228 * to be updated and be sure the update takes place with the
9229 * new frequency before any guests proceed.
9231 * Unfortunately, the combination of hotplug CPU and frequency
9232 * change creates an intractable locking scenario; the order
9233 * of when these callouts happen is undefined with respect to
9234 * CPU hotplug, and they can race with each other. As such,
9235 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
9236 * undefined; you can actually have a CPU frequency change take
9237 * place in between the computation of X and the setting of the
9238 * variable. To protect against this problem, all updates of
9239 * the per_cpu tsc_khz variable are done in an interrupt
9240 * protected IPI, and all callers wishing to update the value
9241 * must wait for a synchronous IPI to complete (which is trivial
9242 * if the caller is on the CPU already). This establishes the
9243 * necessary total order on variable updates.
9245 * Note that because a guest time update may take place
9246 * anytime after the setting of the VCPU's request bit, the
9247 * correct TSC value must be set before the request. However,
9248 * to ensure the update actually makes it to any guest which
9249 * starts running in hardware virtualization between the set
9250 * and the acquisition of the spinlock, we must also ping the
9251 * CPU after setting the request bit.
9255 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
9257 mutex_lock(&kvm_lock);
9258 list_for_each_entry(kvm, &vm_list, vm_list) {
9259 kvm_for_each_vcpu(i, vcpu, kvm) {
9260 if (vcpu->cpu != cpu)
9262 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9263 if (vcpu->cpu != raw_smp_processor_id())
9267 mutex_unlock(&kvm_lock);
9269 if (freq->old < freq->new && send_ipi) {
9271 * We upscale the frequency. Must make the guest
9272 * doesn't see old kvmclock values while running with
9273 * the new frequency, otherwise we risk the guest sees
9274 * time go backwards.
9276 * In case we update the frequency for another cpu
9277 * (which might be in guest context) send an interrupt
9278 * to kick the cpu out of guest context. Next time
9279 * guest context is entered kvmclock will be updated,
9280 * so the guest will not see stale values.
9282 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
9286 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
9289 struct cpufreq_freqs *freq = data;
9292 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
9294 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
9297 for_each_cpu(cpu, freq->policy->cpus)
9298 __kvmclock_cpufreq_notifier(freq, cpu);
9303 static struct notifier_block kvmclock_cpufreq_notifier_block = {
9304 .notifier_call = kvmclock_cpufreq_notifier
9307 static int kvmclock_cpu_online(unsigned int cpu)
9309 tsc_khz_changed(NULL);
9313 static void kvm_timer_init(void)
9315 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9316 max_tsc_khz = tsc_khz;
9318 if (IS_ENABLED(CONFIG_CPU_FREQ)) {
9319 struct cpufreq_policy *policy;
9323 policy = cpufreq_cpu_get(cpu);
9325 if (policy->cpuinfo.max_freq)
9326 max_tsc_khz = policy->cpuinfo.max_freq;
9327 cpufreq_cpu_put(policy);
9331 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
9332 CPUFREQ_TRANSITION_NOTIFIER);
9334 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
9335 kvmclock_cpu_online, kvmclock_cpu_down_prep);
9339 #ifdef CONFIG_X86_64
9340 static void pvclock_gtod_update_fn(struct work_struct *work)
9343 struct kvm_vcpu *vcpu;
9346 mutex_lock(&kvm_lock);
9347 list_for_each_entry(kvm, &vm_list, vm_list)
9348 kvm_for_each_vcpu(i, vcpu, kvm)
9349 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9350 atomic_set(&kvm_guest_has_master_clock, 0);
9351 mutex_unlock(&kvm_lock);
9354 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
9357 * Indirection to move queue_work() out of the tk_core.seq write held
9358 * region to prevent possible deadlocks against time accessors which
9359 * are invoked with work related locks held.
9361 static void pvclock_irq_work_fn(struct irq_work *w)
9363 queue_work(system_long_wq, &pvclock_gtod_work);
9366 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
9369 * Notification about pvclock gtod data update.
9371 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
9374 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
9375 struct timekeeper *tk = priv;
9377 update_pvclock_gtod(tk);
9380 * Disable master clock if host does not trust, or does not use,
9381 * TSC based clocksource. Delegate queue_work() to irq_work as
9382 * this is invoked with tk_core.seq write held.
9384 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
9385 atomic_read(&kvm_guest_has_master_clock) != 0)
9386 irq_work_queue(&pvclock_irq_work);
9390 static struct notifier_block pvclock_gtod_notifier = {
9391 .notifier_call = pvclock_gtod_notify,
9395 static inline void kvm_ops_update(struct kvm_x86_init_ops *ops)
9397 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9399 #define __KVM_X86_OP(func) \
9400 static_call_update(kvm_x86_##func, kvm_x86_ops.func);
9401 #define KVM_X86_OP(func) \
9402 WARN_ON(!kvm_x86_ops.func); __KVM_X86_OP(func)
9403 #define KVM_X86_OP_OPTIONAL __KVM_X86_OP
9404 #define KVM_X86_OP_OPTIONAL_RET0(func) \
9405 static_call_update(kvm_x86_##func, (void *)kvm_x86_ops.func ? : \
9406 (void *)__static_call_return0);
9407 #include <asm/kvm-x86-ops.h>
9410 kvm_pmu_ops_update(ops->pmu_ops);
9413 static int kvm_x86_check_processor_compatibility(void)
9415 int cpu = smp_processor_id();
9416 struct cpuinfo_x86 *c = &cpu_data(cpu);
9419 * Compatibility checks are done when loading KVM and when enabling
9420 * hardware, e.g. during CPU hotplug, to ensure all online CPUs are
9421 * compatible, i.e. KVM should never perform a compatibility check on
9424 WARN_ON(!cpu_online(cpu));
9426 if (__cr4_reserved_bits(cpu_has, c) !=
9427 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
9430 return static_call(kvm_x86_check_processor_compatibility)();
9433 static void kvm_x86_check_cpu_compat(void *ret)
9435 *(int *)ret = kvm_x86_check_processor_compatibility();
9438 static int __kvm_x86_vendor_init(struct kvm_x86_init_ops *ops)
9443 if (kvm_x86_ops.hardware_enable) {
9444 pr_err("already loaded vendor module '%s'\n", kvm_x86_ops.name);
9449 * KVM explicitly assumes that the guest has an FPU and
9450 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
9451 * vCPU's FPU state as a fxregs_state struct.
9453 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
9454 pr_err("inadequate fpu\n");
9458 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9459 pr_err("RT requires X86_FEATURE_CONSTANT_TSC\n");
9464 * KVM assumes that PAT entry '0' encodes WB memtype and simply zeroes
9465 * the PAT bits in SPTEs. Bail if PAT[0] is programmed to something
9466 * other than WB. Note, EPT doesn't utilize the PAT, but don't bother
9467 * with an exception. PAT[0] is set to WB on RESET and also by the
9468 * kernel, i.e. failure indicates a kernel bug or broken firmware.
9470 if (rdmsrl_safe(MSR_IA32_CR_PAT, &host_pat) ||
9471 (host_pat & GENMASK(2, 0)) != 6) {
9472 pr_err("host PAT[0] is not WB\n");
9476 x86_emulator_cache = kvm_alloc_emulator_cache();
9477 if (!x86_emulator_cache) {
9478 pr_err("failed to allocate cache for x86 emulator\n");
9482 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
9483 if (!user_return_msrs) {
9484 pr_err("failed to allocate percpu kvm_user_return_msrs\n");
9486 goto out_free_x86_emulator_cache;
9488 kvm_nr_uret_msrs = 0;
9490 r = kvm_mmu_vendor_module_init();
9492 goto out_free_percpu;
9494 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
9495 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
9496 kvm_caps.supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
9499 rdmsrl_safe(MSR_EFER, &host_efer);
9501 if (boot_cpu_has(X86_FEATURE_XSAVES))
9502 rdmsrl(MSR_IA32_XSS, host_xss);
9504 kvm_init_pmu_capability(ops->pmu_ops);
9506 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
9507 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, host_arch_capabilities);
9509 r = ops->hardware_setup();
9513 kvm_ops_update(ops);
9515 for_each_online_cpu(cpu) {
9516 smp_call_function_single(cpu, kvm_x86_check_cpu_compat, &r, 1);
9518 goto out_unwind_ops;
9522 * Point of no return! DO NOT add error paths below this point unless
9523 * absolutely necessary, as most operations from this point forward
9524 * require unwinding.
9528 if (pi_inject_timer == -1)
9529 pi_inject_timer = housekeeping_enabled(HK_TYPE_TIMER);
9530 #ifdef CONFIG_X86_64
9531 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
9533 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
9534 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
9537 kvm_register_perf_callbacks(ops->handle_intel_pt_intr);
9539 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9540 kvm_caps.supported_xss = 0;
9542 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
9543 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
9544 #undef __kvm_cpu_cap_has
9546 if (kvm_caps.has_tsc_control) {
9548 * Make sure the user can only configure tsc_khz values that
9549 * fit into a signed integer.
9550 * A min value is not calculated because it will always
9551 * be 1 on all machines.
9553 u64 max = min(0x7fffffffULL,
9554 __scale_tsc(kvm_caps.max_tsc_scaling_ratio, tsc_khz));
9555 kvm_caps.max_guest_tsc_khz = max;
9557 kvm_caps.default_tsc_scaling_ratio = 1ULL << kvm_caps.tsc_scaling_ratio_frac_bits;
9558 kvm_init_msr_lists();
9562 kvm_x86_ops.hardware_enable = NULL;
9563 static_call(kvm_x86_hardware_unsetup)();
9565 kvm_mmu_vendor_module_exit();
9567 free_percpu(user_return_msrs);
9568 out_free_x86_emulator_cache:
9569 kmem_cache_destroy(x86_emulator_cache);
9573 int kvm_x86_vendor_init(struct kvm_x86_init_ops *ops)
9577 mutex_lock(&vendor_module_lock);
9578 r = __kvm_x86_vendor_init(ops);
9579 mutex_unlock(&vendor_module_lock);
9583 EXPORT_SYMBOL_GPL(kvm_x86_vendor_init);
9585 void kvm_x86_vendor_exit(void)
9587 kvm_unregister_perf_callbacks();
9589 #ifdef CONFIG_X86_64
9590 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
9591 clear_hv_tscchange_cb();
9595 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9596 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
9597 CPUFREQ_TRANSITION_NOTIFIER);
9598 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
9600 #ifdef CONFIG_X86_64
9601 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
9602 irq_work_sync(&pvclock_irq_work);
9603 cancel_work_sync(&pvclock_gtod_work);
9605 static_call(kvm_x86_hardware_unsetup)();
9606 kvm_mmu_vendor_module_exit();
9607 free_percpu(user_return_msrs);
9608 kmem_cache_destroy(x86_emulator_cache);
9609 #ifdef CONFIG_KVM_XEN
9610 static_key_deferred_flush(&kvm_xen_enabled);
9611 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
9613 mutex_lock(&vendor_module_lock);
9614 kvm_x86_ops.hardware_enable = NULL;
9615 mutex_unlock(&vendor_module_lock);
9617 EXPORT_SYMBOL_GPL(kvm_x86_vendor_exit);
9619 static int __kvm_emulate_halt(struct kvm_vcpu *vcpu, int state, int reason)
9622 * The vCPU has halted, e.g. executed HLT. Update the run state if the
9623 * local APIC is in-kernel, the run loop will detect the non-runnable
9624 * state and halt the vCPU. Exit to userspace if the local APIC is
9625 * managed by userspace, in which case userspace is responsible for
9626 * handling wake events.
9628 ++vcpu->stat.halt_exits;
9629 if (lapic_in_kernel(vcpu)) {
9630 vcpu->arch.mp_state = state;
9633 vcpu->run->exit_reason = reason;
9638 int kvm_emulate_halt_noskip(struct kvm_vcpu *vcpu)
9640 return __kvm_emulate_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
9642 EXPORT_SYMBOL_GPL(kvm_emulate_halt_noskip);
9644 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
9646 int ret = kvm_skip_emulated_instruction(vcpu);
9648 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
9649 * KVM_EXIT_DEBUG here.
9651 return kvm_emulate_halt_noskip(vcpu) && ret;
9653 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
9655 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
9657 int ret = kvm_skip_emulated_instruction(vcpu);
9659 return __kvm_emulate_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD,
9660 KVM_EXIT_AP_RESET_HOLD) && ret;
9662 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
9664 #ifdef CONFIG_X86_64
9665 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
9666 unsigned long clock_type)
9668 struct kvm_clock_pairing clock_pairing;
9669 struct timespec64 ts;
9673 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
9674 return -KVM_EOPNOTSUPP;
9677 * When tsc is in permanent catchup mode guests won't be able to use
9678 * pvclock_read_retry loop to get consistent view of pvclock
9680 if (vcpu->arch.tsc_always_catchup)
9681 return -KVM_EOPNOTSUPP;
9683 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
9684 return -KVM_EOPNOTSUPP;
9686 clock_pairing.sec = ts.tv_sec;
9687 clock_pairing.nsec = ts.tv_nsec;
9688 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
9689 clock_pairing.flags = 0;
9690 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
9693 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
9694 sizeof(struct kvm_clock_pairing)))
9702 * kvm_pv_kick_cpu_op: Kick a vcpu.
9704 * @apicid - apicid of vcpu to be kicked.
9706 static void kvm_pv_kick_cpu_op(struct kvm *kvm, int apicid)
9709 * All other fields are unused for APIC_DM_REMRD, but may be consumed by
9710 * common code, e.g. for tracing. Defer initialization to the compiler.
9712 struct kvm_lapic_irq lapic_irq = {
9713 .delivery_mode = APIC_DM_REMRD,
9714 .dest_mode = APIC_DEST_PHYSICAL,
9715 .shorthand = APIC_DEST_NOSHORT,
9719 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
9722 bool kvm_apicv_activated(struct kvm *kvm)
9724 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
9726 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
9728 bool kvm_vcpu_apicv_activated(struct kvm_vcpu *vcpu)
9730 ulong vm_reasons = READ_ONCE(vcpu->kvm->arch.apicv_inhibit_reasons);
9731 ulong vcpu_reasons = static_call(kvm_x86_vcpu_get_apicv_inhibit_reasons)(vcpu);
9733 return (vm_reasons | vcpu_reasons) == 0;
9735 EXPORT_SYMBOL_GPL(kvm_vcpu_apicv_activated);
9737 static void set_or_clear_apicv_inhibit(unsigned long *inhibits,
9738 enum kvm_apicv_inhibit reason, bool set)
9741 __set_bit(reason, inhibits);
9743 __clear_bit(reason, inhibits);
9745 trace_kvm_apicv_inhibit_changed(reason, set, *inhibits);
9748 static void kvm_apicv_init(struct kvm *kvm)
9750 unsigned long *inhibits = &kvm->arch.apicv_inhibit_reasons;
9752 init_rwsem(&kvm->arch.apicv_update_lock);
9754 set_or_clear_apicv_inhibit(inhibits, APICV_INHIBIT_REASON_ABSENT, true);
9757 set_or_clear_apicv_inhibit(inhibits,
9758 APICV_INHIBIT_REASON_DISABLE, true);
9761 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
9763 struct kvm_vcpu *target = NULL;
9764 struct kvm_apic_map *map;
9766 vcpu->stat.directed_yield_attempted++;
9768 if (single_task_running())
9772 map = rcu_dereference(vcpu->kvm->arch.apic_map);
9774 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
9775 target = map->phys_map[dest_id]->vcpu;
9779 if (!target || !READ_ONCE(target->ready))
9782 /* Ignore requests to yield to self */
9786 if (kvm_vcpu_yield_to(target) <= 0)
9789 vcpu->stat.directed_yield_successful++;
9795 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
9797 u64 ret = vcpu->run->hypercall.ret;
9799 if (!is_64_bit_mode(vcpu))
9801 kvm_rax_write(vcpu, ret);
9802 ++vcpu->stat.hypercalls;
9803 return kvm_skip_emulated_instruction(vcpu);
9806 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
9808 unsigned long nr, a0, a1, a2, a3, ret;
9811 if (kvm_xen_hypercall_enabled(vcpu->kvm))
9812 return kvm_xen_hypercall(vcpu);
9814 if (kvm_hv_hypercall_enabled(vcpu))
9815 return kvm_hv_hypercall(vcpu);
9817 nr = kvm_rax_read(vcpu);
9818 a0 = kvm_rbx_read(vcpu);
9819 a1 = kvm_rcx_read(vcpu);
9820 a2 = kvm_rdx_read(vcpu);
9821 a3 = kvm_rsi_read(vcpu);
9823 trace_kvm_hypercall(nr, a0, a1, a2, a3);
9825 op_64_bit = is_64_bit_hypercall(vcpu);
9834 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
9842 case KVM_HC_VAPIC_POLL_IRQ:
9845 case KVM_HC_KICK_CPU:
9846 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
9849 kvm_pv_kick_cpu_op(vcpu->kvm, a1);
9850 kvm_sched_yield(vcpu, a1);
9853 #ifdef CONFIG_X86_64
9854 case KVM_HC_CLOCK_PAIRING:
9855 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
9858 case KVM_HC_SEND_IPI:
9859 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
9862 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
9864 case KVM_HC_SCHED_YIELD:
9865 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
9868 kvm_sched_yield(vcpu, a0);
9871 case KVM_HC_MAP_GPA_RANGE: {
9872 u64 gpa = a0, npages = a1, attrs = a2;
9875 if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
9878 if (!PAGE_ALIGNED(gpa) || !npages ||
9879 gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
9884 vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
9885 vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
9886 vcpu->run->hypercall.args[0] = gpa;
9887 vcpu->run->hypercall.args[1] = npages;
9888 vcpu->run->hypercall.args[2] = attrs;
9889 vcpu->run->hypercall.flags = 0;
9891 vcpu->run->hypercall.flags |= KVM_EXIT_HYPERCALL_LONG_MODE;
9893 WARN_ON_ONCE(vcpu->run->hypercall.flags & KVM_EXIT_HYPERCALL_MBZ);
9894 vcpu->arch.complete_userspace_io = complete_hypercall_exit;
9904 kvm_rax_write(vcpu, ret);
9906 ++vcpu->stat.hypercalls;
9907 return kvm_skip_emulated_instruction(vcpu);
9909 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
9911 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
9913 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
9914 char instruction[3];
9915 unsigned long rip = kvm_rip_read(vcpu);
9918 * If the quirk is disabled, synthesize a #UD and let the guest pick up
9921 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_FIX_HYPERCALL_INSN)) {
9922 ctxt->exception.error_code_valid = false;
9923 ctxt->exception.vector = UD_VECTOR;
9924 ctxt->have_exception = true;
9925 return X86EMUL_PROPAGATE_FAULT;
9928 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
9930 return emulator_write_emulated(ctxt, rip, instruction, 3,
9934 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
9936 return vcpu->run->request_interrupt_window &&
9937 likely(!pic_in_kernel(vcpu->kvm));
9940 /* Called within kvm->srcu read side. */
9941 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
9943 struct kvm_run *kvm_run = vcpu->run;
9945 kvm_run->if_flag = static_call(kvm_x86_get_if_flag)(vcpu);
9946 kvm_run->cr8 = kvm_get_cr8(vcpu);
9947 kvm_run->apic_base = kvm_get_apic_base(vcpu);
9949 kvm_run->ready_for_interrupt_injection =
9950 pic_in_kernel(vcpu->kvm) ||
9951 kvm_vcpu_ready_for_interrupt_injection(vcpu);
9954 kvm_run->flags |= KVM_RUN_X86_SMM;
9957 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
9961 if (!kvm_x86_ops.update_cr8_intercept)
9964 if (!lapic_in_kernel(vcpu))
9967 if (vcpu->arch.apic->apicv_active)
9970 if (!vcpu->arch.apic->vapic_addr)
9971 max_irr = kvm_lapic_find_highest_irr(vcpu);
9978 tpr = kvm_lapic_get_cr8(vcpu);
9980 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9984 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9986 if (kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9987 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9991 return kvm_x86_ops.nested_ops->check_events(vcpu);
9994 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9997 * Suppress the error code if the vCPU is in Real Mode, as Real Mode
9998 * exceptions don't report error codes. The presence of an error code
9999 * is carried with the exception and only stripped when the exception
10000 * is injected as intercepted #PF VM-Exits for AMD's Paged Real Mode do
10001 * report an error code despite the CPU being in Real Mode.
10003 vcpu->arch.exception.has_error_code &= is_protmode(vcpu);
10005 trace_kvm_inj_exception(vcpu->arch.exception.vector,
10006 vcpu->arch.exception.has_error_code,
10007 vcpu->arch.exception.error_code,
10008 vcpu->arch.exception.injected);
10010 static_call(kvm_x86_inject_exception)(vcpu);
10014 * Check for any event (interrupt or exception) that is ready to be injected,
10015 * and if there is at least one event, inject the event with the highest
10016 * priority. This handles both "pending" events, i.e. events that have never
10017 * been injected into the guest, and "injected" events, i.e. events that were
10018 * injected as part of a previous VM-Enter, but weren't successfully delivered
10019 * and need to be re-injected.
10021 * Note, this is not guaranteed to be invoked on a guest instruction boundary,
10022 * i.e. doesn't guarantee that there's an event window in the guest. KVM must
10023 * be able to inject exceptions in the "middle" of an instruction, and so must
10024 * also be able to re-inject NMIs and IRQs in the middle of an instruction.
10025 * I.e. for exceptions and re-injected events, NOT invoking this on instruction
10026 * boundaries is necessary and correct.
10028 * For simplicity, KVM uses a single path to inject all events (except events
10029 * that are injected directly from L1 to L2) and doesn't explicitly track
10030 * instruction boundaries for asynchronous events. However, because VM-Exits
10031 * that can occur during instruction execution typically result in KVM skipping
10032 * the instruction or injecting an exception, e.g. instruction and exception
10033 * intercepts, and because pending exceptions have higher priority than pending
10034 * interrupts, KVM still honors instruction boundaries in most scenarios.
10036 * But, if a VM-Exit occurs during instruction execution, and KVM does NOT skip
10037 * the instruction or inject an exception, then KVM can incorrecty inject a new
10038 * asynchrounous event if the event became pending after the CPU fetched the
10039 * instruction (in the guest). E.g. if a page fault (#PF, #NPF, EPT violation)
10040 * occurs and is resolved by KVM, a coincident NMI, SMI, IRQ, etc... can be
10041 * injected on the restarted instruction instead of being deferred until the
10042 * instruction completes.
10044 * In practice, this virtualization hole is unlikely to be observed by the
10045 * guest, and even less likely to cause functional problems. To detect the
10046 * hole, the guest would have to trigger an event on a side effect of an early
10047 * phase of instruction execution, e.g. on the instruction fetch from memory.
10048 * And for it to be a functional problem, the guest would need to depend on the
10049 * ordering between that side effect, the instruction completing, _and_ the
10050 * delivery of the asynchronous event.
10052 static int kvm_check_and_inject_events(struct kvm_vcpu *vcpu,
10053 bool *req_immediate_exit)
10059 * Process nested events first, as nested VM-Exit supercedes event
10060 * re-injection. If there's an event queued for re-injection, it will
10061 * be saved into the appropriate vmc{b,s}12 fields on nested VM-Exit.
10063 if (is_guest_mode(vcpu))
10064 r = kvm_check_nested_events(vcpu);
10069 * Re-inject exceptions and events *especially* if immediate entry+exit
10070 * to/from L2 is needed, as any event that has already been injected
10071 * into L2 needs to complete its lifecycle before injecting a new event.
10073 * Don't re-inject an NMI or interrupt if there is a pending exception.
10074 * This collision arises if an exception occurred while vectoring the
10075 * injected event, KVM intercepted said exception, and KVM ultimately
10076 * determined the fault belongs to the guest and queues the exception
10077 * for injection back into the guest.
10079 * "Injected" interrupts can also collide with pending exceptions if
10080 * userspace ignores the "ready for injection" flag and blindly queues
10081 * an interrupt. In that case, prioritizing the exception is correct,
10082 * as the exception "occurred" before the exit to userspace. Trap-like
10083 * exceptions, e.g. most #DBs, have higher priority than interrupts.
10084 * And while fault-like exceptions, e.g. #GP and #PF, are the lowest
10085 * priority, they're only generated (pended) during instruction
10086 * execution, and interrupts are recognized at instruction boundaries.
10087 * Thus a pending fault-like exception means the fault occurred on the
10088 * *previous* instruction and must be serviced prior to recognizing any
10089 * new events in order to fully complete the previous instruction.
10091 if (vcpu->arch.exception.injected)
10092 kvm_inject_exception(vcpu);
10093 else if (kvm_is_exception_pending(vcpu))
10095 else if (vcpu->arch.nmi_injected)
10096 static_call(kvm_x86_inject_nmi)(vcpu);
10097 else if (vcpu->arch.interrupt.injected)
10098 static_call(kvm_x86_inject_irq)(vcpu, true);
10101 * Exceptions that morph to VM-Exits are handled above, and pending
10102 * exceptions on top of injected exceptions that do not VM-Exit should
10103 * either morph to #DF or, sadly, override the injected exception.
10105 WARN_ON_ONCE(vcpu->arch.exception.injected &&
10106 vcpu->arch.exception.pending);
10109 * Bail if immediate entry+exit to/from the guest is needed to complete
10110 * nested VM-Enter or event re-injection so that a different pending
10111 * event can be serviced (or if KVM needs to exit to userspace).
10113 * Otherwise, continue processing events even if VM-Exit occurred. The
10114 * VM-Exit will have cleared exceptions that were meant for L2, but
10115 * there may now be events that can be injected into L1.
10121 * A pending exception VM-Exit should either result in nested VM-Exit
10122 * or force an immediate re-entry and exit to/from L2, and exception
10123 * VM-Exits cannot be injected (flag should _never_ be set).
10125 WARN_ON_ONCE(vcpu->arch.exception_vmexit.injected ||
10126 vcpu->arch.exception_vmexit.pending);
10129 * New events, other than exceptions, cannot be injected if KVM needs
10130 * to re-inject a previous event. See above comments on re-injecting
10131 * for why pending exceptions get priority.
10133 can_inject = !kvm_event_needs_reinjection(vcpu);
10135 if (vcpu->arch.exception.pending) {
10137 * Fault-class exceptions, except #DBs, set RF=1 in the RFLAGS
10138 * value pushed on the stack. Trap-like exception and all #DBs
10139 * leave RF as-is (KVM follows Intel's behavior in this regard;
10140 * AMD states that code breakpoint #DBs excplitly clear RF=0).
10142 * Note, most versions of Intel's SDM and AMD's APM incorrectly
10143 * describe the behavior of General Detect #DBs, which are
10144 * fault-like. They do _not_ set RF, a la code breakpoints.
10146 if (exception_type(vcpu->arch.exception.vector) == EXCPT_FAULT)
10147 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
10150 if (vcpu->arch.exception.vector == DB_VECTOR) {
10151 kvm_deliver_exception_payload(vcpu, &vcpu->arch.exception);
10152 if (vcpu->arch.dr7 & DR7_GD) {
10153 vcpu->arch.dr7 &= ~DR7_GD;
10154 kvm_update_dr7(vcpu);
10158 kvm_inject_exception(vcpu);
10160 vcpu->arch.exception.pending = false;
10161 vcpu->arch.exception.injected = true;
10163 can_inject = false;
10166 /* Don't inject interrupts if the user asked to avoid doing so */
10167 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
10171 * Finally, inject interrupt events. If an event cannot be injected
10172 * due to architectural conditions (e.g. IF=0) a window-open exit
10173 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
10174 * and can architecturally be injected, but we cannot do it right now:
10175 * an interrupt could have arrived just now and we have to inject it
10176 * as a vmexit, or there could already an event in the queue, which is
10177 * indicated by can_inject. In that case we request an immediate exit
10178 * in order to make progress and get back here for another iteration.
10179 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
10181 #ifdef CONFIG_KVM_SMM
10182 if (vcpu->arch.smi_pending) {
10183 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
10187 vcpu->arch.smi_pending = false;
10188 ++vcpu->arch.smi_count;
10190 can_inject = false;
10192 static_call(kvm_x86_enable_smi_window)(vcpu);
10196 if (vcpu->arch.nmi_pending) {
10197 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
10201 --vcpu->arch.nmi_pending;
10202 vcpu->arch.nmi_injected = true;
10203 static_call(kvm_x86_inject_nmi)(vcpu);
10204 can_inject = false;
10205 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
10207 if (vcpu->arch.nmi_pending)
10208 static_call(kvm_x86_enable_nmi_window)(vcpu);
10211 if (kvm_cpu_has_injectable_intr(vcpu)) {
10212 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
10216 int irq = kvm_cpu_get_interrupt(vcpu);
10218 if (!WARN_ON_ONCE(irq == -1)) {
10219 kvm_queue_interrupt(vcpu, irq, false);
10220 static_call(kvm_x86_inject_irq)(vcpu, false);
10221 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
10224 if (kvm_cpu_has_injectable_intr(vcpu))
10225 static_call(kvm_x86_enable_irq_window)(vcpu);
10228 if (is_guest_mode(vcpu) &&
10229 kvm_x86_ops.nested_ops->has_events &&
10230 kvm_x86_ops.nested_ops->has_events(vcpu))
10231 *req_immediate_exit = true;
10234 * KVM must never queue a new exception while injecting an event; KVM
10235 * is done emulating and should only propagate the to-be-injected event
10236 * to the VMCS/VMCB. Queueing a new exception can put the vCPU into an
10237 * infinite loop as KVM will bail from VM-Enter to inject the pending
10238 * exception and start the cycle all over.
10240 * Exempt triple faults as they have special handling and won't put the
10241 * vCPU into an infinite loop. Triple fault can be queued when running
10242 * VMX without unrestricted guest, as that requires KVM to emulate Real
10243 * Mode events (see kvm_inject_realmode_interrupt()).
10245 WARN_ON_ONCE(vcpu->arch.exception.pending ||
10246 vcpu->arch.exception_vmexit.pending);
10251 *req_immediate_exit = true;
10257 static void process_nmi(struct kvm_vcpu *vcpu)
10259 unsigned int limit;
10262 * x86 is limited to one NMI pending, but because KVM can't react to
10263 * incoming NMIs as quickly as bare metal, e.g. if the vCPU is
10264 * scheduled out, KVM needs to play nice with two queued NMIs showing
10265 * up at the same time. To handle this scenario, allow two NMIs to be
10266 * (temporarily) pending so long as NMIs are not blocked and KVM is not
10267 * waiting for a previous NMI injection to complete (which effectively
10268 * blocks NMIs). KVM will immediately inject one of the two NMIs, and
10269 * will request an NMI window to handle the second NMI.
10271 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
10277 * Adjust the limit to account for pending virtual NMIs, which aren't
10278 * tracked in vcpu->arch.nmi_pending.
10280 if (static_call(kvm_x86_is_vnmi_pending)(vcpu))
10283 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
10284 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
10286 if (vcpu->arch.nmi_pending &&
10287 (static_call(kvm_x86_set_vnmi_pending)(vcpu)))
10288 vcpu->arch.nmi_pending--;
10290 if (vcpu->arch.nmi_pending)
10291 kvm_make_request(KVM_REQ_EVENT, vcpu);
10294 /* Return total number of NMIs pending injection to the VM */
10295 int kvm_get_nr_pending_nmis(struct kvm_vcpu *vcpu)
10297 return vcpu->arch.nmi_pending +
10298 static_call(kvm_x86_is_vnmi_pending)(vcpu);
10301 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
10302 unsigned long *vcpu_bitmap)
10304 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
10307 void kvm_make_scan_ioapic_request(struct kvm *kvm)
10309 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
10312 void __kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
10314 struct kvm_lapic *apic = vcpu->arch.apic;
10317 if (!lapic_in_kernel(vcpu))
10320 down_read(&vcpu->kvm->arch.apicv_update_lock);
10323 /* Do not activate APICV when APIC is disabled */
10324 activate = kvm_vcpu_apicv_activated(vcpu) &&
10325 (kvm_get_apic_mode(vcpu) != LAPIC_MODE_DISABLED);
10327 if (apic->apicv_active == activate)
10330 apic->apicv_active = activate;
10331 kvm_apic_update_apicv(vcpu);
10332 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
10335 * When APICv gets disabled, we may still have injected interrupts
10336 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
10337 * still active when the interrupt got accepted. Make sure
10338 * kvm_check_and_inject_events() is called to check for that.
10340 if (!apic->apicv_active)
10341 kvm_make_request(KVM_REQ_EVENT, vcpu);
10345 up_read(&vcpu->kvm->arch.apicv_update_lock);
10347 EXPORT_SYMBOL_GPL(__kvm_vcpu_update_apicv);
10349 static void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
10351 if (!lapic_in_kernel(vcpu))
10355 * Due to sharing page tables across vCPUs, the xAPIC memslot must be
10356 * deleted if any vCPU has xAPIC virtualization and x2APIC enabled, but
10357 * and hardware doesn't support x2APIC virtualization. E.g. some AMD
10358 * CPUs support AVIC but not x2APIC. KVM still allows enabling AVIC in
10359 * this case so that KVM can the AVIC doorbell to inject interrupts to
10360 * running vCPUs, but KVM must not create SPTEs for the APIC base as
10361 * the vCPU would incorrectly be able to access the vAPIC page via MMIO
10362 * despite being in x2APIC mode. For simplicity, inhibiting the APIC
10363 * access page is sticky.
10365 if (apic_x2apic_mode(vcpu->arch.apic) &&
10366 kvm_x86_ops.allow_apicv_in_x2apic_without_x2apic_virtualization)
10367 kvm_inhibit_apic_access_page(vcpu);
10369 __kvm_vcpu_update_apicv(vcpu);
10372 void __kvm_set_or_clear_apicv_inhibit(struct kvm *kvm,
10373 enum kvm_apicv_inhibit reason, bool set)
10375 unsigned long old, new;
10377 lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
10379 if (!(kvm_x86_ops.required_apicv_inhibits & BIT(reason)))
10382 old = new = kvm->arch.apicv_inhibit_reasons;
10384 set_or_clear_apicv_inhibit(&new, reason, set);
10386 if (!!old != !!new) {
10388 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
10389 * false positives in the sanity check WARN in svm_vcpu_run().
10390 * This task will wait for all vCPUs to ack the kick IRQ before
10391 * updating apicv_inhibit_reasons, and all other vCPUs will
10392 * block on acquiring apicv_update_lock so that vCPUs can't
10393 * redo svm_vcpu_run() without seeing the new inhibit state.
10395 * Note, holding apicv_update_lock and taking it in the read
10396 * side (handling the request) also prevents other vCPUs from
10397 * servicing the request with a stale apicv_inhibit_reasons.
10399 kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
10400 kvm->arch.apicv_inhibit_reasons = new;
10402 unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
10403 int idx = srcu_read_lock(&kvm->srcu);
10405 kvm_zap_gfn_range(kvm, gfn, gfn+1);
10406 srcu_read_unlock(&kvm->srcu, idx);
10409 kvm->arch.apicv_inhibit_reasons = new;
10413 void kvm_set_or_clear_apicv_inhibit(struct kvm *kvm,
10414 enum kvm_apicv_inhibit reason, bool set)
10419 down_write(&kvm->arch.apicv_update_lock);
10420 __kvm_set_or_clear_apicv_inhibit(kvm, reason, set);
10421 up_write(&kvm->arch.apicv_update_lock);
10423 EXPORT_SYMBOL_GPL(kvm_set_or_clear_apicv_inhibit);
10425 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
10427 if (!kvm_apic_present(vcpu))
10430 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
10432 if (irqchip_split(vcpu->kvm))
10433 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
10435 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10436 if (ioapic_in_kernel(vcpu->kvm))
10437 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
10440 if (is_guest_mode(vcpu))
10441 vcpu->arch.load_eoi_exitmap_pending = true;
10443 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
10446 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
10448 u64 eoi_exit_bitmap[4];
10450 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
10453 if (to_hv_vcpu(vcpu)) {
10454 bitmap_or((ulong *)eoi_exit_bitmap,
10455 vcpu->arch.ioapic_handled_vectors,
10456 to_hv_synic(vcpu)->vec_bitmap, 256);
10457 static_call_cond(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
10461 static_call_cond(kvm_x86_load_eoi_exitmap)(
10462 vcpu, (u64 *)vcpu->arch.ioapic_handled_vectors);
10465 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
10467 static_call_cond(kvm_x86_guest_memory_reclaimed)(kvm);
10470 static void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
10472 if (!lapic_in_kernel(vcpu))
10475 static_call_cond(kvm_x86_set_apic_access_page_addr)(vcpu);
10478 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
10480 smp_send_reschedule(vcpu->cpu);
10482 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
10485 * Called within kvm->srcu read side.
10486 * Returns 1 to let vcpu_run() continue the guest execution loop without
10487 * exiting to the userspace. Otherwise, the value will be returned to the
10490 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
10494 dm_request_for_irq_injection(vcpu) &&
10495 kvm_cpu_accept_dm_intr(vcpu);
10496 fastpath_t exit_fastpath;
10498 bool req_immediate_exit = false;
10500 if (kvm_request_pending(vcpu)) {
10501 if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu)) {
10506 if (kvm_dirty_ring_check_request(vcpu)) {
10511 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
10512 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
10517 if (kvm_check_request(KVM_REQ_MMU_FREE_OBSOLETE_ROOTS, vcpu))
10518 kvm_mmu_free_obsolete_roots(vcpu);
10519 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
10520 __kvm_migrate_timers(vcpu);
10521 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
10522 kvm_update_masterclock(vcpu->kvm);
10523 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
10524 kvm_gen_kvmclock_update(vcpu);
10525 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
10526 r = kvm_guest_time_update(vcpu);
10530 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
10531 kvm_mmu_sync_roots(vcpu);
10532 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
10533 kvm_mmu_load_pgd(vcpu);
10536 * Note, the order matters here, as flushing "all" TLB entries
10537 * also flushes the "current" TLB entries, i.e. servicing the
10538 * flush "all" will clear any request to flush "current".
10540 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
10541 kvm_vcpu_flush_tlb_all(vcpu);
10543 kvm_service_local_tlb_flush_requests(vcpu);
10546 * Fall back to a "full" guest flush if Hyper-V's precise
10547 * flushing fails. Note, Hyper-V's flushing is per-vCPU, but
10548 * the flushes are considered "remote" and not "local" because
10549 * the requests can be initiated from other vCPUs.
10551 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu) &&
10552 kvm_hv_vcpu_flush_tlb(vcpu))
10553 kvm_vcpu_flush_tlb_guest(vcpu);
10555 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
10556 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
10560 if (kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
10561 if (is_guest_mode(vcpu))
10562 kvm_x86_ops.nested_ops->triple_fault(vcpu);
10564 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
10565 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
10566 vcpu->mmio_needed = 0;
10571 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
10572 /* Page is swapped out. Do synthetic halt */
10573 vcpu->arch.apf.halted = true;
10577 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
10578 record_steal_time(vcpu);
10579 #ifdef CONFIG_KVM_SMM
10580 if (kvm_check_request(KVM_REQ_SMI, vcpu))
10583 if (kvm_check_request(KVM_REQ_NMI, vcpu))
10585 if (kvm_check_request(KVM_REQ_PMU, vcpu))
10586 kvm_pmu_handle_event(vcpu);
10587 if (kvm_check_request(KVM_REQ_PMI, vcpu))
10588 kvm_pmu_deliver_pmi(vcpu);
10589 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
10590 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
10591 if (test_bit(vcpu->arch.pending_ioapic_eoi,
10592 vcpu->arch.ioapic_handled_vectors)) {
10593 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
10594 vcpu->run->eoi.vector =
10595 vcpu->arch.pending_ioapic_eoi;
10600 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
10601 vcpu_scan_ioapic(vcpu);
10602 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
10603 vcpu_load_eoi_exitmap(vcpu);
10604 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
10605 kvm_vcpu_reload_apic_access_page(vcpu);
10606 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
10607 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
10608 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
10609 vcpu->run->system_event.ndata = 0;
10613 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
10614 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
10615 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
10616 vcpu->run->system_event.ndata = 0;
10620 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
10621 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
10623 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
10624 vcpu->run->hyperv = hv_vcpu->exit;
10630 * KVM_REQ_HV_STIMER has to be processed after
10631 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
10632 * depend on the guest clock being up-to-date
10634 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
10635 kvm_hv_process_stimers(vcpu);
10636 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
10637 kvm_vcpu_update_apicv(vcpu);
10638 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
10639 kvm_check_async_pf_completion(vcpu);
10640 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
10641 static_call(kvm_x86_msr_filter_changed)(vcpu);
10643 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
10644 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
10647 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
10648 kvm_xen_has_interrupt(vcpu)) {
10649 ++vcpu->stat.req_event;
10650 r = kvm_apic_accept_events(vcpu);
10655 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
10660 r = kvm_check_and_inject_events(vcpu, &req_immediate_exit);
10666 static_call(kvm_x86_enable_irq_window)(vcpu);
10668 if (kvm_lapic_enabled(vcpu)) {
10669 update_cr8_intercept(vcpu);
10670 kvm_lapic_sync_to_vapic(vcpu);
10674 r = kvm_mmu_reload(vcpu);
10676 goto cancel_injection;
10681 static_call(kvm_x86_prepare_switch_to_guest)(vcpu);
10684 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
10685 * IPI are then delayed after guest entry, which ensures that they
10686 * result in virtual interrupt delivery.
10688 local_irq_disable();
10690 /* Store vcpu->apicv_active before vcpu->mode. */
10691 smp_store_release(&vcpu->mode, IN_GUEST_MODE);
10693 kvm_vcpu_srcu_read_unlock(vcpu);
10696 * 1) We should set ->mode before checking ->requests. Please see
10697 * the comment in kvm_vcpu_exiting_guest_mode().
10699 * 2) For APICv, we should set ->mode before checking PID.ON. This
10700 * pairs with the memory barrier implicit in pi_test_and_set_on
10701 * (see vmx_deliver_posted_interrupt).
10703 * 3) This also orders the write to mode from any reads to the page
10704 * tables done while the VCPU is running. Please see the comment
10705 * in kvm_flush_remote_tlbs.
10707 smp_mb__after_srcu_read_unlock();
10710 * Process pending posted interrupts to handle the case where the
10711 * notification IRQ arrived in the host, or was never sent (because the
10712 * target vCPU wasn't running). Do this regardless of the vCPU's APICv
10713 * status, KVM doesn't update assigned devices when APICv is inhibited,
10714 * i.e. they can post interrupts even if APICv is temporarily disabled.
10716 if (kvm_lapic_enabled(vcpu))
10717 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10719 if (kvm_vcpu_exit_request(vcpu)) {
10720 vcpu->mode = OUTSIDE_GUEST_MODE;
10722 local_irq_enable();
10724 kvm_vcpu_srcu_read_lock(vcpu);
10726 goto cancel_injection;
10729 if (req_immediate_exit) {
10730 kvm_make_request(KVM_REQ_EVENT, vcpu);
10731 static_call(kvm_x86_request_immediate_exit)(vcpu);
10734 fpregs_assert_state_consistent();
10735 if (test_thread_flag(TIF_NEED_FPU_LOAD))
10736 switch_fpu_return();
10738 if (vcpu->arch.guest_fpu.xfd_err)
10739 wrmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
10741 if (unlikely(vcpu->arch.switch_db_regs)) {
10742 set_debugreg(0, 7);
10743 set_debugreg(vcpu->arch.eff_db[0], 0);
10744 set_debugreg(vcpu->arch.eff_db[1], 1);
10745 set_debugreg(vcpu->arch.eff_db[2], 2);
10746 set_debugreg(vcpu->arch.eff_db[3], 3);
10747 } else if (unlikely(hw_breakpoint_active())) {
10748 set_debugreg(0, 7);
10751 guest_timing_enter_irqoff();
10755 * Assert that vCPU vs. VM APICv state is consistent. An APICv
10756 * update must kick and wait for all vCPUs before toggling the
10757 * per-VM state, and responsing vCPUs must wait for the update
10758 * to complete before servicing KVM_REQ_APICV_UPDATE.
10760 WARN_ON_ONCE((kvm_vcpu_apicv_activated(vcpu) != kvm_vcpu_apicv_active(vcpu)) &&
10761 (kvm_get_apic_mode(vcpu) != LAPIC_MODE_DISABLED));
10763 exit_fastpath = static_call(kvm_x86_vcpu_run)(vcpu);
10764 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
10767 if (kvm_lapic_enabled(vcpu))
10768 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10770 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
10771 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
10775 /* Note, VM-Exits that go down the "slow" path are accounted below. */
10776 ++vcpu->stat.exits;
10780 * Do this here before restoring debug registers on the host. And
10781 * since we do this before handling the vmexit, a DR access vmexit
10782 * can (a) read the correct value of the debug registers, (b) set
10783 * KVM_DEBUGREG_WONT_EXIT again.
10785 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
10786 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
10787 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
10788 kvm_update_dr0123(vcpu);
10789 kvm_update_dr7(vcpu);
10793 * If the guest has used debug registers, at least dr7
10794 * will be disabled while returning to the host.
10795 * If we don't have active breakpoints in the host, we don't
10796 * care about the messed up debug address registers. But if
10797 * we have some of them active, restore the old state.
10799 if (hw_breakpoint_active())
10800 hw_breakpoint_restore();
10802 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
10803 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
10805 vcpu->mode = OUTSIDE_GUEST_MODE;
10809 * Sync xfd before calling handle_exit_irqoff() which may
10810 * rely on the fact that guest_fpu::xfd is up-to-date (e.g.
10811 * in #NM irqoff handler).
10813 if (vcpu->arch.xfd_no_write_intercept)
10814 fpu_sync_guest_vmexit_xfd_state();
10816 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
10818 if (vcpu->arch.guest_fpu.xfd_err)
10819 wrmsrl(MSR_IA32_XFD_ERR, 0);
10822 * Consume any pending interrupts, including the possible source of
10823 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
10824 * An instruction is required after local_irq_enable() to fully unblock
10825 * interrupts on processors that implement an interrupt shadow, the
10826 * stat.exits increment will do nicely.
10828 kvm_before_interrupt(vcpu, KVM_HANDLING_IRQ);
10829 local_irq_enable();
10830 ++vcpu->stat.exits;
10831 local_irq_disable();
10832 kvm_after_interrupt(vcpu);
10835 * Wait until after servicing IRQs to account guest time so that any
10836 * ticks that occurred while running the guest are properly accounted
10837 * to the guest. Waiting until IRQs are enabled degrades the accuracy
10838 * of accounting via context tracking, but the loss of accuracy is
10839 * acceptable for all known use cases.
10841 guest_timing_exit_irqoff();
10843 local_irq_enable();
10846 kvm_vcpu_srcu_read_lock(vcpu);
10849 * Profile KVM exit RIPs:
10851 if (unlikely(prof_on == KVM_PROFILING)) {
10852 unsigned long rip = kvm_rip_read(vcpu);
10853 profile_hit(KVM_PROFILING, (void *)rip);
10856 if (unlikely(vcpu->arch.tsc_always_catchup))
10857 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10859 if (vcpu->arch.apic_attention)
10860 kvm_lapic_sync_from_vapic(vcpu);
10862 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
10866 if (req_immediate_exit)
10867 kvm_make_request(KVM_REQ_EVENT, vcpu);
10868 static_call(kvm_x86_cancel_injection)(vcpu);
10869 if (unlikely(vcpu->arch.apic_attention))
10870 kvm_lapic_sync_from_vapic(vcpu);
10875 /* Called within kvm->srcu read side. */
10876 static inline int vcpu_block(struct kvm_vcpu *vcpu)
10880 if (!kvm_arch_vcpu_runnable(vcpu)) {
10882 * Switch to the software timer before halt-polling/blocking as
10883 * the guest's timer may be a break event for the vCPU, and the
10884 * hypervisor timer runs only when the CPU is in guest mode.
10885 * Switch before halt-polling so that KVM recognizes an expired
10886 * timer before blocking.
10888 hv_timer = kvm_lapic_hv_timer_in_use(vcpu);
10890 kvm_lapic_switch_to_sw_timer(vcpu);
10892 kvm_vcpu_srcu_read_unlock(vcpu);
10893 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10894 kvm_vcpu_halt(vcpu);
10896 kvm_vcpu_block(vcpu);
10897 kvm_vcpu_srcu_read_lock(vcpu);
10900 kvm_lapic_switch_to_hv_timer(vcpu);
10903 * If the vCPU is not runnable, a signal or another host event
10904 * of some kind is pending; service it without changing the
10905 * vCPU's activity state.
10907 if (!kvm_arch_vcpu_runnable(vcpu))
10912 * Evaluate nested events before exiting the halted state. This allows
10913 * the halt state to be recorded properly in the VMCS12's activity
10914 * state field (AMD does not have a similar field and a VM-Exit always
10915 * causes a spurious wakeup from HLT).
10917 if (is_guest_mode(vcpu)) {
10918 if (kvm_check_nested_events(vcpu) < 0)
10922 if (kvm_apic_accept_events(vcpu) < 0)
10924 switch(vcpu->arch.mp_state) {
10925 case KVM_MP_STATE_HALTED:
10926 case KVM_MP_STATE_AP_RESET_HOLD:
10927 vcpu->arch.pv.pv_unhalted = false;
10928 vcpu->arch.mp_state =
10929 KVM_MP_STATE_RUNNABLE;
10931 case KVM_MP_STATE_RUNNABLE:
10932 vcpu->arch.apf.halted = false;
10934 case KVM_MP_STATE_INIT_RECEIVED:
10943 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
10945 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
10946 !vcpu->arch.apf.halted);
10949 /* Called within kvm->srcu read side. */
10950 static int vcpu_run(struct kvm_vcpu *vcpu)
10954 vcpu->arch.l1tf_flush_l1d = true;
10958 * If another guest vCPU requests a PV TLB flush in the middle
10959 * of instruction emulation, the rest of the emulation could
10960 * use a stale page translation. Assume that any code after
10961 * this point can start executing an instruction.
10963 vcpu->arch.at_instruction_boundary = false;
10964 if (kvm_vcpu_running(vcpu)) {
10965 r = vcpu_enter_guest(vcpu);
10967 r = vcpu_block(vcpu);
10973 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10974 if (kvm_xen_has_pending_events(vcpu))
10975 kvm_xen_inject_pending_events(vcpu);
10977 if (kvm_cpu_has_pending_timer(vcpu))
10978 kvm_inject_pending_timer_irqs(vcpu);
10980 if (dm_request_for_irq_injection(vcpu) &&
10981 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10983 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10984 ++vcpu->stat.request_irq_exits;
10988 if (__xfer_to_guest_mode_work_pending()) {
10989 kvm_vcpu_srcu_read_unlock(vcpu);
10990 r = xfer_to_guest_mode_handle_work(vcpu);
10991 kvm_vcpu_srcu_read_lock(vcpu);
11000 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
11002 return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
11005 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
11007 BUG_ON(!vcpu->arch.pio.count);
11009 return complete_emulated_io(vcpu);
11013 * Implements the following, as a state machine:
11016 * for each fragment
11017 * for each mmio piece in the fragment
11024 * for each fragment
11025 * for each mmio piece in the fragment
11030 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
11032 struct kvm_run *run = vcpu->run;
11033 struct kvm_mmio_fragment *frag;
11036 BUG_ON(!vcpu->mmio_needed);
11038 /* Complete previous fragment */
11039 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11040 len = min(8u, frag->len);
11041 if (!vcpu->mmio_is_write)
11042 memcpy(frag->data, run->mmio.data, len);
11044 if (frag->len <= 8) {
11045 /* Switch to the next fragment. */
11047 vcpu->mmio_cur_fragment++;
11049 /* Go forward to the next mmio piece. */
11055 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11056 vcpu->mmio_needed = 0;
11058 /* FIXME: return into emulator if single-stepping. */
11059 if (vcpu->mmio_is_write)
11061 vcpu->mmio_read_completed = 1;
11062 return complete_emulated_io(vcpu);
11065 run->exit_reason = KVM_EXIT_MMIO;
11066 run->mmio.phys_addr = frag->gpa;
11067 if (vcpu->mmio_is_write)
11068 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11069 run->mmio.len = min(8u, frag->len);
11070 run->mmio.is_write = vcpu->mmio_is_write;
11071 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
11075 /* Swap (qemu) user FPU context for the guest FPU context. */
11076 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
11078 /* Exclude PKRU, it's restored separately immediately after VM-Exit. */
11079 fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
11083 /* When vcpu_run ends, restore user space FPU context. */
11084 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
11086 fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
11087 ++vcpu->stat.fpu_reload;
11091 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
11093 struct kvm_queued_exception *ex = &vcpu->arch.exception;
11094 struct kvm_run *kvm_run = vcpu->run;
11098 kvm_sigset_activate(vcpu);
11099 kvm_run->flags = 0;
11100 kvm_load_guest_fpu(vcpu);
11102 kvm_vcpu_srcu_read_lock(vcpu);
11103 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
11104 if (kvm_run->immediate_exit) {
11110 * Don't bother switching APIC timer emulation from the
11111 * hypervisor timer to the software timer, the only way for the
11112 * APIC timer to be active is if userspace stuffed vCPU state,
11113 * i.e. put the vCPU into a nonsensical state. Only an INIT
11114 * will transition the vCPU out of UNINITIALIZED (without more
11115 * state stuffing from userspace), which will reset the local
11116 * APIC and thus cancel the timer or drop the IRQ (if the timer
11117 * already expired).
11119 kvm_vcpu_srcu_read_unlock(vcpu);
11120 kvm_vcpu_block(vcpu);
11121 kvm_vcpu_srcu_read_lock(vcpu);
11123 if (kvm_apic_accept_events(vcpu) < 0) {
11128 if (signal_pending(current)) {
11130 kvm_run->exit_reason = KVM_EXIT_INTR;
11131 ++vcpu->stat.signal_exits;
11136 if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
11137 (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
11142 if (kvm_run->kvm_dirty_regs) {
11143 r = sync_regs(vcpu);
11148 /* re-sync apic's tpr */
11149 if (!lapic_in_kernel(vcpu)) {
11150 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
11157 * If userspace set a pending exception and L2 is active, convert it to
11158 * a pending VM-Exit if L1 wants to intercept the exception.
11160 if (vcpu->arch.exception_from_userspace && is_guest_mode(vcpu) &&
11161 kvm_x86_ops.nested_ops->is_exception_vmexit(vcpu, ex->vector,
11163 kvm_queue_exception_vmexit(vcpu, ex->vector,
11164 ex->has_error_code, ex->error_code,
11165 ex->has_payload, ex->payload);
11166 ex->injected = false;
11167 ex->pending = false;
11169 vcpu->arch.exception_from_userspace = false;
11171 if (unlikely(vcpu->arch.complete_userspace_io)) {
11172 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
11173 vcpu->arch.complete_userspace_io = NULL;
11178 WARN_ON_ONCE(vcpu->arch.pio.count);
11179 WARN_ON_ONCE(vcpu->mmio_needed);
11182 if (kvm_run->immediate_exit) {
11187 r = static_call(kvm_x86_vcpu_pre_run)(vcpu);
11191 r = vcpu_run(vcpu);
11194 kvm_put_guest_fpu(vcpu);
11195 if (kvm_run->kvm_valid_regs)
11197 post_kvm_run_save(vcpu);
11198 kvm_vcpu_srcu_read_unlock(vcpu);
11200 kvm_sigset_deactivate(vcpu);
11205 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11207 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
11209 * We are here if userspace calls get_regs() in the middle of
11210 * instruction emulation. Registers state needs to be copied
11211 * back from emulation context to vcpu. Userspace shouldn't do
11212 * that usually, but some bad designed PV devices (vmware
11213 * backdoor interface) need this to work
11215 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
11216 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
11218 regs->rax = kvm_rax_read(vcpu);
11219 regs->rbx = kvm_rbx_read(vcpu);
11220 regs->rcx = kvm_rcx_read(vcpu);
11221 regs->rdx = kvm_rdx_read(vcpu);
11222 regs->rsi = kvm_rsi_read(vcpu);
11223 regs->rdi = kvm_rdi_read(vcpu);
11224 regs->rsp = kvm_rsp_read(vcpu);
11225 regs->rbp = kvm_rbp_read(vcpu);
11226 #ifdef CONFIG_X86_64
11227 regs->r8 = kvm_r8_read(vcpu);
11228 regs->r9 = kvm_r9_read(vcpu);
11229 regs->r10 = kvm_r10_read(vcpu);
11230 regs->r11 = kvm_r11_read(vcpu);
11231 regs->r12 = kvm_r12_read(vcpu);
11232 regs->r13 = kvm_r13_read(vcpu);
11233 regs->r14 = kvm_r14_read(vcpu);
11234 regs->r15 = kvm_r15_read(vcpu);
11237 regs->rip = kvm_rip_read(vcpu);
11238 regs->rflags = kvm_get_rflags(vcpu);
11241 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11244 __get_regs(vcpu, regs);
11249 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11251 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
11252 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
11254 kvm_rax_write(vcpu, regs->rax);
11255 kvm_rbx_write(vcpu, regs->rbx);
11256 kvm_rcx_write(vcpu, regs->rcx);
11257 kvm_rdx_write(vcpu, regs->rdx);
11258 kvm_rsi_write(vcpu, regs->rsi);
11259 kvm_rdi_write(vcpu, regs->rdi);
11260 kvm_rsp_write(vcpu, regs->rsp);
11261 kvm_rbp_write(vcpu, regs->rbp);
11262 #ifdef CONFIG_X86_64
11263 kvm_r8_write(vcpu, regs->r8);
11264 kvm_r9_write(vcpu, regs->r9);
11265 kvm_r10_write(vcpu, regs->r10);
11266 kvm_r11_write(vcpu, regs->r11);
11267 kvm_r12_write(vcpu, regs->r12);
11268 kvm_r13_write(vcpu, regs->r13);
11269 kvm_r14_write(vcpu, regs->r14);
11270 kvm_r15_write(vcpu, regs->r15);
11273 kvm_rip_write(vcpu, regs->rip);
11274 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
11276 vcpu->arch.exception.pending = false;
11277 vcpu->arch.exception_vmexit.pending = false;
11279 kvm_make_request(KVM_REQ_EVENT, vcpu);
11282 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11285 __set_regs(vcpu, regs);
11290 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11292 struct desc_ptr dt;
11294 if (vcpu->arch.guest_state_protected)
11295 goto skip_protected_regs;
11297 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
11298 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
11299 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
11300 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
11301 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
11302 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
11304 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
11305 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
11307 static_call(kvm_x86_get_idt)(vcpu, &dt);
11308 sregs->idt.limit = dt.size;
11309 sregs->idt.base = dt.address;
11310 static_call(kvm_x86_get_gdt)(vcpu, &dt);
11311 sregs->gdt.limit = dt.size;
11312 sregs->gdt.base = dt.address;
11314 sregs->cr2 = vcpu->arch.cr2;
11315 sregs->cr3 = kvm_read_cr3(vcpu);
11317 skip_protected_regs:
11318 sregs->cr0 = kvm_read_cr0(vcpu);
11319 sregs->cr4 = kvm_read_cr4(vcpu);
11320 sregs->cr8 = kvm_get_cr8(vcpu);
11321 sregs->efer = vcpu->arch.efer;
11322 sregs->apic_base = kvm_get_apic_base(vcpu);
11325 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11327 __get_sregs_common(vcpu, sregs);
11329 if (vcpu->arch.guest_state_protected)
11332 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
11333 set_bit(vcpu->arch.interrupt.nr,
11334 (unsigned long *)sregs->interrupt_bitmap);
11337 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
11341 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
11343 if (vcpu->arch.guest_state_protected)
11346 if (is_pae_paging(vcpu)) {
11347 for (i = 0 ; i < 4 ; i++)
11348 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
11349 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
11353 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
11354 struct kvm_sregs *sregs)
11357 __get_sregs(vcpu, sregs);
11362 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
11363 struct kvm_mp_state *mp_state)
11368 if (kvm_mpx_supported())
11369 kvm_load_guest_fpu(vcpu);
11371 r = kvm_apic_accept_events(vcpu);
11376 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
11377 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
11378 vcpu->arch.pv.pv_unhalted)
11379 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
11381 mp_state->mp_state = vcpu->arch.mp_state;
11384 if (kvm_mpx_supported())
11385 kvm_put_guest_fpu(vcpu);
11390 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
11391 struct kvm_mp_state *mp_state)
11397 switch (mp_state->mp_state) {
11398 case KVM_MP_STATE_UNINITIALIZED:
11399 case KVM_MP_STATE_HALTED:
11400 case KVM_MP_STATE_AP_RESET_HOLD:
11401 case KVM_MP_STATE_INIT_RECEIVED:
11402 case KVM_MP_STATE_SIPI_RECEIVED:
11403 if (!lapic_in_kernel(vcpu))
11407 case KVM_MP_STATE_RUNNABLE:
11415 * Pending INITs are reported using KVM_SET_VCPU_EVENTS, disallow
11416 * forcing the guest into INIT/SIPI if those events are supposed to be
11417 * blocked. KVM prioritizes SMI over INIT, so reject INIT/SIPI state
11418 * if an SMI is pending as well.
11420 if ((!kvm_apic_init_sipi_allowed(vcpu) || vcpu->arch.smi_pending) &&
11421 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
11422 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
11425 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
11426 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
11427 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
11429 vcpu->arch.mp_state = mp_state->mp_state;
11430 kvm_make_request(KVM_REQ_EVENT, vcpu);
11438 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
11439 int reason, bool has_error_code, u32 error_code)
11441 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
11444 init_emulate_ctxt(vcpu);
11446 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
11447 has_error_code, error_code);
11449 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11450 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11451 vcpu->run->internal.ndata = 0;
11455 kvm_rip_write(vcpu, ctxt->eip);
11456 kvm_set_rflags(vcpu, ctxt->eflags);
11459 EXPORT_SYMBOL_GPL(kvm_task_switch);
11461 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11463 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
11465 * When EFER.LME and CR0.PG are set, the processor is in
11466 * 64-bit mode (though maybe in a 32-bit code segment).
11467 * CR4.PAE and EFER.LMA must be set.
11469 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
11471 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
11475 * Not in 64-bit mode: EFER.LMA is clear and the code
11476 * segment cannot be 64-bit.
11478 if (sregs->efer & EFER_LMA || sregs->cs.l)
11482 return kvm_is_valid_cr4(vcpu, sregs->cr4) &&
11483 kvm_is_valid_cr0(vcpu, sregs->cr0);
11486 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
11487 int *mmu_reset_needed, bool update_pdptrs)
11489 struct msr_data apic_base_msr;
11491 struct desc_ptr dt;
11493 if (!kvm_is_valid_sregs(vcpu, sregs))
11496 apic_base_msr.data = sregs->apic_base;
11497 apic_base_msr.host_initiated = true;
11498 if (kvm_set_apic_base(vcpu, &apic_base_msr))
11501 if (vcpu->arch.guest_state_protected)
11504 dt.size = sregs->idt.limit;
11505 dt.address = sregs->idt.base;
11506 static_call(kvm_x86_set_idt)(vcpu, &dt);
11507 dt.size = sregs->gdt.limit;
11508 dt.address = sregs->gdt.base;
11509 static_call(kvm_x86_set_gdt)(vcpu, &dt);
11511 vcpu->arch.cr2 = sregs->cr2;
11512 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
11513 vcpu->arch.cr3 = sregs->cr3;
11514 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11515 static_call_cond(kvm_x86_post_set_cr3)(vcpu, sregs->cr3);
11517 kvm_set_cr8(vcpu, sregs->cr8);
11519 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
11520 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
11522 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
11523 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
11524 vcpu->arch.cr0 = sregs->cr0;
11526 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
11527 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
11529 if (update_pdptrs) {
11530 idx = srcu_read_lock(&vcpu->kvm->srcu);
11531 if (is_pae_paging(vcpu)) {
11532 load_pdptrs(vcpu, kvm_read_cr3(vcpu));
11533 *mmu_reset_needed = 1;
11535 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11538 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
11539 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
11540 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
11541 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
11542 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
11543 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
11545 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
11546 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
11548 update_cr8_intercept(vcpu);
11550 /* Older userspace won't unhalt the vcpu on reset. */
11551 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
11552 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
11553 !is_protmode(vcpu))
11554 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11559 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11561 int pending_vec, max_bits;
11562 int mmu_reset_needed = 0;
11563 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
11568 if (mmu_reset_needed)
11569 kvm_mmu_reset_context(vcpu);
11571 max_bits = KVM_NR_INTERRUPTS;
11572 pending_vec = find_first_bit(
11573 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
11575 if (pending_vec < max_bits) {
11576 kvm_queue_interrupt(vcpu, pending_vec, false);
11577 pr_debug("Set back pending irq %d\n", pending_vec);
11578 kvm_make_request(KVM_REQ_EVENT, vcpu);
11583 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
11585 int mmu_reset_needed = 0;
11586 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
11587 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
11588 !(sregs2->efer & EFER_LMA);
11591 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
11594 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
11597 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
11598 &mmu_reset_needed, !valid_pdptrs);
11602 if (valid_pdptrs) {
11603 for (i = 0; i < 4 ; i++)
11604 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
11606 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
11607 mmu_reset_needed = 1;
11608 vcpu->arch.pdptrs_from_userspace = true;
11610 if (mmu_reset_needed)
11611 kvm_mmu_reset_context(vcpu);
11615 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
11616 struct kvm_sregs *sregs)
11621 ret = __set_sregs(vcpu, sregs);
11626 static void kvm_arch_vcpu_guestdbg_update_apicv_inhibit(struct kvm *kvm)
11629 struct kvm_vcpu *vcpu;
11635 down_write(&kvm->arch.apicv_update_lock);
11637 kvm_for_each_vcpu(i, vcpu, kvm) {
11638 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ) {
11643 __kvm_set_or_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_BLOCKIRQ, set);
11644 up_write(&kvm->arch.apicv_update_lock);
11647 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
11648 struct kvm_guest_debug *dbg)
11650 unsigned long rflags;
11653 if (vcpu->arch.guest_state_protected)
11658 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
11660 if (kvm_is_exception_pending(vcpu))
11662 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
11663 kvm_queue_exception(vcpu, DB_VECTOR);
11665 kvm_queue_exception(vcpu, BP_VECTOR);
11669 * Read rflags as long as potentially injected trace flags are still
11672 rflags = kvm_get_rflags(vcpu);
11674 vcpu->guest_debug = dbg->control;
11675 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
11676 vcpu->guest_debug = 0;
11678 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
11679 for (i = 0; i < KVM_NR_DB_REGS; ++i)
11680 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
11681 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
11683 for (i = 0; i < KVM_NR_DB_REGS; i++)
11684 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
11686 kvm_update_dr7(vcpu);
11688 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11689 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
11692 * Trigger an rflags update that will inject or remove the trace
11695 kvm_set_rflags(vcpu, rflags);
11697 static_call(kvm_x86_update_exception_bitmap)(vcpu);
11699 kvm_arch_vcpu_guestdbg_update_apicv_inhibit(vcpu->kvm);
11709 * Translate a guest virtual address to a guest physical address.
11711 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
11712 struct kvm_translation *tr)
11714 unsigned long vaddr = tr->linear_address;
11720 idx = srcu_read_lock(&vcpu->kvm->srcu);
11721 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
11722 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11723 tr->physical_address = gpa;
11724 tr->valid = gpa != INVALID_GPA;
11732 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
11734 struct fxregs_state *fxsave;
11736 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
11741 fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
11742 memcpy(fpu->fpr, fxsave->st_space, 128);
11743 fpu->fcw = fxsave->cwd;
11744 fpu->fsw = fxsave->swd;
11745 fpu->ftwx = fxsave->twd;
11746 fpu->last_opcode = fxsave->fop;
11747 fpu->last_ip = fxsave->rip;
11748 fpu->last_dp = fxsave->rdp;
11749 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
11755 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
11757 struct fxregs_state *fxsave;
11759 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
11764 fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
11766 memcpy(fxsave->st_space, fpu->fpr, 128);
11767 fxsave->cwd = fpu->fcw;
11768 fxsave->swd = fpu->fsw;
11769 fxsave->twd = fpu->ftwx;
11770 fxsave->fop = fpu->last_opcode;
11771 fxsave->rip = fpu->last_ip;
11772 fxsave->rdp = fpu->last_dp;
11773 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
11779 static void store_regs(struct kvm_vcpu *vcpu)
11781 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
11783 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
11784 __get_regs(vcpu, &vcpu->run->s.regs.regs);
11786 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
11787 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
11789 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
11790 kvm_vcpu_ioctl_x86_get_vcpu_events(
11791 vcpu, &vcpu->run->s.regs.events);
11794 static int sync_regs(struct kvm_vcpu *vcpu)
11796 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
11797 __set_regs(vcpu, &vcpu->run->s.regs.regs);
11798 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
11801 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
11802 struct kvm_sregs sregs = vcpu->run->s.regs.sregs;
11804 if (__set_sregs(vcpu, &sregs))
11807 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
11810 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
11811 struct kvm_vcpu_events events = vcpu->run->s.regs.events;
11813 if (kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events))
11816 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
11822 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
11824 if (kvm_check_tsc_unstable() && kvm->created_vcpus)
11825 pr_warn_once("SMP vm created on host with unstable TSC; "
11826 "guest TSC will not be reliable\n");
11828 if (!kvm->arch.max_vcpu_ids)
11829 kvm->arch.max_vcpu_ids = KVM_MAX_VCPU_IDS;
11831 if (id >= kvm->arch.max_vcpu_ids)
11834 return static_call(kvm_x86_vcpu_precreate)(kvm);
11837 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
11842 vcpu->arch.last_vmentry_cpu = -1;
11843 vcpu->arch.regs_avail = ~0;
11844 vcpu->arch.regs_dirty = ~0;
11846 kvm_gpc_init(&vcpu->arch.pv_time, vcpu->kvm, vcpu, KVM_HOST_USES_PFN);
11848 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
11849 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11851 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
11853 r = kvm_mmu_create(vcpu);
11857 if (irqchip_in_kernel(vcpu->kvm)) {
11858 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
11860 goto fail_mmu_destroy;
11863 * Defer evaluating inhibits until the vCPU is first run, as
11864 * this vCPU will not get notified of any changes until this
11865 * vCPU is visible to other vCPUs (marked online and added to
11866 * the set of vCPUs). Opportunistically mark APICv active as
11867 * VMX in particularly is highly unlikely to have inhibits.
11868 * Ignore the current per-VM APICv state so that vCPU creation
11869 * is guaranteed to run with a deterministic value, the request
11870 * will ensure the vCPU gets the correct state before VM-Entry.
11872 if (enable_apicv) {
11873 vcpu->arch.apic->apicv_active = true;
11874 kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
11877 static_branch_inc(&kvm_has_noapic_vcpu);
11881 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
11883 goto fail_free_lapic;
11884 vcpu->arch.pio_data = page_address(page);
11886 vcpu->arch.mce_banks = kcalloc(KVM_MAX_MCE_BANKS * 4, sizeof(u64),
11887 GFP_KERNEL_ACCOUNT);
11888 vcpu->arch.mci_ctl2_banks = kcalloc(KVM_MAX_MCE_BANKS, sizeof(u64),
11889 GFP_KERNEL_ACCOUNT);
11890 if (!vcpu->arch.mce_banks || !vcpu->arch.mci_ctl2_banks)
11891 goto fail_free_mce_banks;
11892 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
11894 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
11895 GFP_KERNEL_ACCOUNT))
11896 goto fail_free_mce_banks;
11898 if (!alloc_emulate_ctxt(vcpu))
11899 goto free_wbinvd_dirty_mask;
11901 if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
11902 pr_err("failed to allocate vcpu's fpu\n");
11903 goto free_emulate_ctxt;
11906 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
11907 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
11909 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
11911 kvm_async_pf_hash_reset(vcpu);
11913 vcpu->arch.perf_capabilities = kvm_caps.supported_perf_cap;
11914 kvm_pmu_init(vcpu);
11916 vcpu->arch.pending_external_vector = -1;
11917 vcpu->arch.preempted_in_kernel = false;
11919 #if IS_ENABLED(CONFIG_HYPERV)
11920 vcpu->arch.hv_root_tdp = INVALID_PAGE;
11923 r = static_call(kvm_x86_vcpu_create)(vcpu);
11925 goto free_guest_fpu;
11927 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
11928 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
11929 kvm_xen_init_vcpu(vcpu);
11930 kvm_vcpu_mtrr_init(vcpu);
11932 kvm_set_tsc_khz(vcpu, vcpu->kvm->arch.default_tsc_khz);
11933 kvm_vcpu_reset(vcpu, false);
11934 kvm_init_mmu(vcpu);
11939 fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11941 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11942 free_wbinvd_dirty_mask:
11943 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11944 fail_free_mce_banks:
11945 kfree(vcpu->arch.mce_banks);
11946 kfree(vcpu->arch.mci_ctl2_banks);
11947 free_page((unsigned long)vcpu->arch.pio_data);
11949 kvm_free_lapic(vcpu);
11951 kvm_mmu_destroy(vcpu);
11955 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
11957 struct kvm *kvm = vcpu->kvm;
11959 if (mutex_lock_killable(&vcpu->mutex))
11962 kvm_synchronize_tsc(vcpu, 0);
11965 /* poll control enabled by default */
11966 vcpu->arch.msr_kvm_poll_control = 1;
11968 mutex_unlock(&vcpu->mutex);
11970 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
11971 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
11972 KVMCLOCK_SYNC_PERIOD);
11975 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
11979 kvmclock_reset(vcpu);
11981 static_call(kvm_x86_vcpu_free)(vcpu);
11983 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11984 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11985 fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11987 kvm_xen_destroy_vcpu(vcpu);
11988 kvm_hv_vcpu_uninit(vcpu);
11989 kvm_pmu_destroy(vcpu);
11990 kfree(vcpu->arch.mce_banks);
11991 kfree(vcpu->arch.mci_ctl2_banks);
11992 kvm_free_lapic(vcpu);
11993 idx = srcu_read_lock(&vcpu->kvm->srcu);
11994 kvm_mmu_destroy(vcpu);
11995 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11996 free_page((unsigned long)vcpu->arch.pio_data);
11997 kvfree(vcpu->arch.cpuid_entries);
11998 if (!lapic_in_kernel(vcpu))
11999 static_branch_dec(&kvm_has_noapic_vcpu);
12002 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
12004 struct kvm_cpuid_entry2 *cpuid_0x1;
12005 unsigned long old_cr0 = kvm_read_cr0(vcpu);
12006 unsigned long new_cr0;
12009 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
12010 * to handle side effects. RESET emulation hits those flows and relies
12011 * on emulated/virtualized registers, including those that are loaded
12012 * into hardware, to be zeroed at vCPU creation. Use CRs as a sentinel
12013 * to detect improper or missing initialization.
12015 WARN_ON_ONCE(!init_event &&
12016 (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
12019 * SVM doesn't unconditionally VM-Exit on INIT and SHUTDOWN, thus it's
12020 * possible to INIT the vCPU while L2 is active. Force the vCPU back
12021 * into L1 as EFER.SVME is cleared on INIT (along with all other EFER
12022 * bits), i.e. virtualization is disabled.
12024 if (is_guest_mode(vcpu))
12025 kvm_leave_nested(vcpu);
12027 kvm_lapic_reset(vcpu, init_event);
12029 WARN_ON_ONCE(is_guest_mode(vcpu) || is_smm(vcpu));
12030 vcpu->arch.hflags = 0;
12032 vcpu->arch.smi_pending = 0;
12033 vcpu->arch.smi_count = 0;
12034 atomic_set(&vcpu->arch.nmi_queued, 0);
12035 vcpu->arch.nmi_pending = 0;
12036 vcpu->arch.nmi_injected = false;
12037 kvm_clear_interrupt_queue(vcpu);
12038 kvm_clear_exception_queue(vcpu);
12040 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
12041 kvm_update_dr0123(vcpu);
12042 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
12043 vcpu->arch.dr7 = DR7_FIXED_1;
12044 kvm_update_dr7(vcpu);
12046 vcpu->arch.cr2 = 0;
12048 kvm_make_request(KVM_REQ_EVENT, vcpu);
12049 vcpu->arch.apf.msr_en_val = 0;
12050 vcpu->arch.apf.msr_int_val = 0;
12051 vcpu->arch.st.msr_val = 0;
12053 kvmclock_reset(vcpu);
12055 kvm_clear_async_pf_completion_queue(vcpu);
12056 kvm_async_pf_hash_reset(vcpu);
12057 vcpu->arch.apf.halted = false;
12059 if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
12060 struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
12063 * All paths that lead to INIT are required to load the guest's
12064 * FPU state (because most paths are buried in KVM_RUN).
12067 kvm_put_guest_fpu(vcpu);
12069 fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
12070 fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
12073 kvm_load_guest_fpu(vcpu);
12077 kvm_pmu_reset(vcpu);
12078 vcpu->arch.smbase = 0x30000;
12080 vcpu->arch.msr_misc_features_enables = 0;
12081 vcpu->arch.ia32_misc_enable_msr = MSR_IA32_MISC_ENABLE_PEBS_UNAVAIL |
12082 MSR_IA32_MISC_ENABLE_BTS_UNAVAIL;
12084 __kvm_set_xcr(vcpu, 0, XFEATURE_MASK_FP);
12085 __kvm_set_msr(vcpu, MSR_IA32_XSS, 0, true);
12088 /* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
12089 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
12090 kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
12093 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
12094 * if no CPUID match is found. Note, it's impossible to get a match at
12095 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
12096 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
12097 * on RESET. But, go through the motions in case that's ever remedied.
12099 cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1);
12100 kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
12102 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
12104 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
12105 kvm_rip_write(vcpu, 0xfff0);
12107 vcpu->arch.cr3 = 0;
12108 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
12111 * CR0.CD/NW are set on RESET, preserved on INIT. Note, some versions
12112 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
12113 * (or qualify) that with a footnote stating that CD/NW are preserved.
12115 new_cr0 = X86_CR0_ET;
12117 new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
12119 new_cr0 |= X86_CR0_NW | X86_CR0_CD;
12121 static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
12122 static_call(kvm_x86_set_cr4)(vcpu, 0);
12123 static_call(kvm_x86_set_efer)(vcpu, 0);
12124 static_call(kvm_x86_update_exception_bitmap)(vcpu);
12127 * On the standard CR0/CR4/EFER modification paths, there are several
12128 * complex conditions determining whether the MMU has to be reset and/or
12129 * which PCIDs have to be flushed. However, CR0.WP and the paging-related
12130 * bits in CR4 and EFER are irrelevant if CR0.PG was '0'; and a reset+flush
12131 * is needed anyway if CR0.PG was '1' (which can only happen for INIT, as
12132 * CR0 will be '0' prior to RESET). So we only need to check CR0.PG here.
12134 if (old_cr0 & X86_CR0_PG) {
12135 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12136 kvm_mmu_reset_context(vcpu);
12140 * Intel's SDM states that all TLB entries are flushed on INIT. AMD's
12141 * APM states the TLBs are untouched by INIT, but it also states that
12142 * the TLBs are flushed on "External initialization of the processor."
12143 * Flush the guest TLB regardless of vendor, there is no meaningful
12144 * benefit in relying on the guest to flush the TLB immediately after
12145 * INIT. A spurious TLB flush is benign and likely negligible from a
12146 * performance perspective.
12149 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12151 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
12153 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
12155 struct kvm_segment cs;
12157 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
12158 cs.selector = vector << 8;
12159 cs.base = vector << 12;
12160 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
12161 kvm_rip_write(vcpu, 0);
12163 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
12165 int kvm_arch_hardware_enable(void)
12168 struct kvm_vcpu *vcpu;
12173 bool stable, backwards_tsc = false;
12175 kvm_user_return_msr_cpu_online();
12177 ret = kvm_x86_check_processor_compatibility();
12181 ret = static_call(kvm_x86_hardware_enable)();
12185 local_tsc = rdtsc();
12186 stable = !kvm_check_tsc_unstable();
12187 list_for_each_entry(kvm, &vm_list, vm_list) {
12188 kvm_for_each_vcpu(i, vcpu, kvm) {
12189 if (!stable && vcpu->cpu == smp_processor_id())
12190 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
12191 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
12192 backwards_tsc = true;
12193 if (vcpu->arch.last_host_tsc > max_tsc)
12194 max_tsc = vcpu->arch.last_host_tsc;
12200 * Sometimes, even reliable TSCs go backwards. This happens on
12201 * platforms that reset TSC during suspend or hibernate actions, but
12202 * maintain synchronization. We must compensate. Fortunately, we can
12203 * detect that condition here, which happens early in CPU bringup,
12204 * before any KVM threads can be running. Unfortunately, we can't
12205 * bring the TSCs fully up to date with real time, as we aren't yet far
12206 * enough into CPU bringup that we know how much real time has actually
12207 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
12208 * variables that haven't been updated yet.
12210 * So we simply find the maximum observed TSC above, then record the
12211 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
12212 * the adjustment will be applied. Note that we accumulate
12213 * adjustments, in case multiple suspend cycles happen before some VCPU
12214 * gets a chance to run again. In the event that no KVM threads get a
12215 * chance to run, we will miss the entire elapsed period, as we'll have
12216 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
12217 * loose cycle time. This isn't too big a deal, since the loss will be
12218 * uniform across all VCPUs (not to mention the scenario is extremely
12219 * unlikely). It is possible that a second hibernate recovery happens
12220 * much faster than a first, causing the observed TSC here to be
12221 * smaller; this would require additional padding adjustment, which is
12222 * why we set last_host_tsc to the local tsc observed here.
12224 * N.B. - this code below runs only on platforms with reliable TSC,
12225 * as that is the only way backwards_tsc is set above. Also note
12226 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
12227 * have the same delta_cyc adjustment applied if backwards_tsc
12228 * is detected. Note further, this adjustment is only done once,
12229 * as we reset last_host_tsc on all VCPUs to stop this from being
12230 * called multiple times (one for each physical CPU bringup).
12232 * Platforms with unreliable TSCs don't have to deal with this, they
12233 * will be compensated by the logic in vcpu_load, which sets the TSC to
12234 * catchup mode. This will catchup all VCPUs to real time, but cannot
12235 * guarantee that they stay in perfect synchronization.
12237 if (backwards_tsc) {
12238 u64 delta_cyc = max_tsc - local_tsc;
12239 list_for_each_entry(kvm, &vm_list, vm_list) {
12240 kvm->arch.backwards_tsc_observed = true;
12241 kvm_for_each_vcpu(i, vcpu, kvm) {
12242 vcpu->arch.tsc_offset_adjustment += delta_cyc;
12243 vcpu->arch.last_host_tsc = local_tsc;
12244 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
12248 * We have to disable TSC offset matching.. if you were
12249 * booting a VM while issuing an S4 host suspend....
12250 * you may have some problem. Solving this issue is
12251 * left as an exercise to the reader.
12253 kvm->arch.last_tsc_nsec = 0;
12254 kvm->arch.last_tsc_write = 0;
12261 void kvm_arch_hardware_disable(void)
12263 static_call(kvm_x86_hardware_disable)();
12264 drop_user_return_notifiers();
12267 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
12269 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
12272 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
12274 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
12277 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
12278 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
12280 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
12282 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
12284 vcpu->arch.l1tf_flush_l1d = true;
12285 if (pmu->version && unlikely(pmu->event_count)) {
12286 pmu->need_cleanup = true;
12287 kvm_make_request(KVM_REQ_PMU, vcpu);
12289 static_call(kvm_x86_sched_in)(vcpu, cpu);
12292 void kvm_arch_free_vm(struct kvm *kvm)
12294 kfree(to_kvm_hv(kvm)->hv_pa_pg);
12295 __kvm_arch_free_vm(kvm);
12299 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
12302 unsigned long flags;
12307 ret = kvm_page_track_init(kvm);
12311 ret = kvm_mmu_init_vm(kvm);
12313 goto out_page_track;
12315 ret = static_call(kvm_x86_vm_init)(kvm);
12317 goto out_uninit_mmu;
12319 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
12320 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
12321 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
12323 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
12324 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
12325 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
12326 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
12327 &kvm->arch.irq_sources_bitmap);
12329 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
12330 mutex_init(&kvm->arch.apic_map_lock);
12331 seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
12332 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
12334 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
12335 pvclock_update_vm_gtod_copy(kvm);
12336 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
12338 kvm->arch.default_tsc_khz = max_tsc_khz ? : tsc_khz;
12339 kvm->arch.guest_can_read_msr_platform_info = true;
12340 kvm->arch.enable_pmu = enable_pmu;
12342 #if IS_ENABLED(CONFIG_HYPERV)
12343 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
12344 kvm->arch.hv_root_tdp = INVALID_PAGE;
12347 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
12348 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
12350 kvm_apicv_init(kvm);
12351 kvm_hv_init_vm(kvm);
12352 kvm_xen_init_vm(kvm);
12357 kvm_mmu_uninit_vm(kvm);
12359 kvm_page_track_cleanup(kvm);
12364 int kvm_arch_post_init_vm(struct kvm *kvm)
12366 return kvm_mmu_post_init_vm(kvm);
12369 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
12372 kvm_mmu_unload(vcpu);
12376 static void kvm_unload_vcpu_mmus(struct kvm *kvm)
12379 struct kvm_vcpu *vcpu;
12381 kvm_for_each_vcpu(i, vcpu, kvm) {
12382 kvm_clear_async_pf_completion_queue(vcpu);
12383 kvm_unload_vcpu_mmu(vcpu);
12387 void kvm_arch_sync_events(struct kvm *kvm)
12389 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
12390 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
12395 * __x86_set_memory_region: Setup KVM internal memory slot
12397 * @kvm: the kvm pointer to the VM.
12398 * @id: the slot ID to setup.
12399 * @gpa: the GPA to install the slot (unused when @size == 0).
12400 * @size: the size of the slot. Set to zero to uninstall a slot.
12402 * This function helps to setup a KVM internal memory slot. Specify
12403 * @size > 0 to install a new slot, while @size == 0 to uninstall a
12404 * slot. The return code can be one of the following:
12406 * HVA: on success (uninstall will return a bogus HVA)
12409 * The caller should always use IS_ERR() to check the return value
12410 * before use. Note, the KVM internal memory slots are guaranteed to
12411 * remain valid and unchanged until the VM is destroyed, i.e., the
12412 * GPA->HVA translation will not change. However, the HVA is a user
12413 * address, i.e. its accessibility is not guaranteed, and must be
12414 * accessed via __copy_{to,from}_user().
12416 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
12420 unsigned long hva, old_npages;
12421 struct kvm_memslots *slots = kvm_memslots(kvm);
12422 struct kvm_memory_slot *slot;
12424 /* Called with kvm->slots_lock held. */
12425 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
12426 return ERR_PTR_USR(-EINVAL);
12428 slot = id_to_memslot(slots, id);
12430 if (slot && slot->npages)
12431 return ERR_PTR_USR(-EEXIST);
12434 * MAP_SHARED to prevent internal slot pages from being moved
12437 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
12438 MAP_SHARED | MAP_ANONYMOUS, 0);
12439 if (IS_ERR_VALUE(hva))
12440 return (void __user *)hva;
12442 if (!slot || !slot->npages)
12445 old_npages = slot->npages;
12446 hva = slot->userspace_addr;
12449 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
12450 struct kvm_userspace_memory_region m;
12452 m.slot = id | (i << 16);
12454 m.guest_phys_addr = gpa;
12455 m.userspace_addr = hva;
12456 m.memory_size = size;
12457 r = __kvm_set_memory_region(kvm, &m);
12459 return ERR_PTR_USR(r);
12463 vm_munmap(hva, old_npages * PAGE_SIZE);
12465 return (void __user *)hva;
12467 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
12469 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
12471 kvm_mmu_pre_destroy_vm(kvm);
12474 void kvm_arch_destroy_vm(struct kvm *kvm)
12476 if (current->mm == kvm->mm) {
12478 * Free memory regions allocated on behalf of userspace,
12479 * unless the memory map has changed due to process exit
12482 mutex_lock(&kvm->slots_lock);
12483 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
12485 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
12487 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
12488 mutex_unlock(&kvm->slots_lock);
12490 kvm_unload_vcpu_mmus(kvm);
12491 static_call_cond(kvm_x86_vm_destroy)(kvm);
12492 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
12493 kvm_pic_destroy(kvm);
12494 kvm_ioapic_destroy(kvm);
12495 kvm_destroy_vcpus(kvm);
12496 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
12497 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
12498 kvm_mmu_uninit_vm(kvm);
12499 kvm_page_track_cleanup(kvm);
12500 kvm_xen_destroy_vm(kvm);
12501 kvm_hv_destroy_vm(kvm);
12504 static void memslot_rmap_free(struct kvm_memory_slot *slot)
12508 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
12509 kvfree(slot->arch.rmap[i]);
12510 slot->arch.rmap[i] = NULL;
12514 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
12518 memslot_rmap_free(slot);
12520 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12521 kvfree(slot->arch.lpage_info[i - 1]);
12522 slot->arch.lpage_info[i - 1] = NULL;
12525 kvm_page_track_free_memslot(slot);
12528 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
12530 const int sz = sizeof(*slot->arch.rmap[0]);
12533 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
12535 int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
12537 if (slot->arch.rmap[i])
12540 slot->arch.rmap[i] = __vcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
12541 if (!slot->arch.rmap[i]) {
12542 memslot_rmap_free(slot);
12550 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
12551 struct kvm_memory_slot *slot)
12553 unsigned long npages = slot->npages;
12557 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
12558 * old arrays will be freed by __kvm_set_memory_region() if installing
12559 * the new memslot is successful.
12561 memset(&slot->arch, 0, sizeof(slot->arch));
12563 if (kvm_memslots_have_rmaps(kvm)) {
12564 r = memslot_rmap_alloc(slot, npages);
12569 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12570 struct kvm_lpage_info *linfo;
12571 unsigned long ugfn;
12575 lpages = __kvm_mmu_slot_lpages(slot, npages, level);
12577 linfo = __vcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
12581 slot->arch.lpage_info[i - 1] = linfo;
12583 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
12584 linfo[0].disallow_lpage = 1;
12585 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
12586 linfo[lpages - 1].disallow_lpage = 1;
12587 ugfn = slot->userspace_addr >> PAGE_SHIFT;
12589 * If the gfn and userspace address are not aligned wrt each
12590 * other, disable large page support for this slot.
12592 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
12595 for (j = 0; j < lpages; ++j)
12596 linfo[j].disallow_lpage = 1;
12600 if (kvm_page_track_create_memslot(kvm, slot, npages))
12606 memslot_rmap_free(slot);
12608 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12609 kvfree(slot->arch.lpage_info[i - 1]);
12610 slot->arch.lpage_info[i - 1] = NULL;
12615 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
12617 struct kvm_vcpu *vcpu;
12621 * memslots->generation has been incremented.
12622 * mmio generation may have reached its maximum value.
12624 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
12626 /* Force re-initialization of steal_time cache */
12627 kvm_for_each_vcpu(i, vcpu, kvm)
12628 kvm_vcpu_kick(vcpu);
12631 int kvm_arch_prepare_memory_region(struct kvm *kvm,
12632 const struct kvm_memory_slot *old,
12633 struct kvm_memory_slot *new,
12634 enum kvm_mr_change change)
12637 * KVM doesn't support moving memslots when there are external page
12638 * trackers attached to the VM, i.e. if KVMGT is in use.
12640 if (change == KVM_MR_MOVE && kvm_page_track_has_external_user(kvm))
12643 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE) {
12644 if ((new->base_gfn + new->npages - 1) > kvm_mmu_max_gfn())
12647 return kvm_alloc_memslot_metadata(kvm, new);
12650 if (change == KVM_MR_FLAGS_ONLY)
12651 memcpy(&new->arch, &old->arch, sizeof(old->arch));
12652 else if (WARN_ON_ONCE(change != KVM_MR_DELETE))
12659 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
12663 if (!kvm_x86_ops.cpu_dirty_log_size)
12666 nr_slots = atomic_read(&kvm->nr_memslots_dirty_logging);
12667 if ((enable && nr_slots == 1) || !nr_slots)
12668 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
12671 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
12672 struct kvm_memory_slot *old,
12673 const struct kvm_memory_slot *new,
12674 enum kvm_mr_change change)
12676 u32 old_flags = old ? old->flags : 0;
12677 u32 new_flags = new ? new->flags : 0;
12678 bool log_dirty_pages = new_flags & KVM_MEM_LOG_DIRTY_PAGES;
12681 * Update CPU dirty logging if dirty logging is being toggled. This
12682 * applies to all operations.
12684 if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)
12685 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
12688 * Nothing more to do for RO slots (which can't be dirtied and can't be
12689 * made writable) or CREATE/MOVE/DELETE of a slot.
12691 * For a memslot with dirty logging disabled:
12692 * CREATE: No dirty mappings will already exist.
12693 * MOVE/DELETE: The old mappings will already have been cleaned up by
12694 * kvm_arch_flush_shadow_memslot()
12696 * For a memslot with dirty logging enabled:
12697 * CREATE: No shadow pages exist, thus nothing to write-protect
12698 * and no dirty bits to clear.
12699 * MOVE/DELETE: The old mappings will already have been cleaned up by
12700 * kvm_arch_flush_shadow_memslot().
12702 if ((change != KVM_MR_FLAGS_ONLY) || (new_flags & KVM_MEM_READONLY))
12706 * READONLY and non-flags changes were filtered out above, and the only
12707 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
12708 * logging isn't being toggled on or off.
12710 if (WARN_ON_ONCE(!((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)))
12713 if (!log_dirty_pages) {
12715 * Dirty logging tracks sptes in 4k granularity, meaning that
12716 * large sptes have to be split. If live migration succeeds,
12717 * the guest in the source machine will be destroyed and large
12718 * sptes will be created in the destination. However, if the
12719 * guest continues to run in the source machine (for example if
12720 * live migration fails), small sptes will remain around and
12721 * cause bad performance.
12723 * Scan sptes if dirty logging has been stopped, dropping those
12724 * which can be collapsed into a single large-page spte. Later
12725 * page faults will create the large-page sptes.
12727 kvm_mmu_zap_collapsible_sptes(kvm, new);
12730 * Initially-all-set does not require write protecting any page,
12731 * because they're all assumed to be dirty.
12733 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
12736 if (READ_ONCE(eager_page_split))
12737 kvm_mmu_slot_try_split_huge_pages(kvm, new, PG_LEVEL_4K);
12739 if (kvm_x86_ops.cpu_dirty_log_size) {
12740 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
12741 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
12743 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
12747 * Unconditionally flush the TLBs after enabling dirty logging.
12748 * A flush is almost always going to be necessary (see below),
12749 * and unconditionally flushing allows the helpers to omit
12750 * the subtly complex checks when removing write access.
12752 * Do the flush outside of mmu_lock to reduce the amount of
12753 * time mmu_lock is held. Flushing after dropping mmu_lock is
12754 * safe as KVM only needs to guarantee the slot is fully
12755 * write-protected before returning to userspace, i.e. before
12756 * userspace can consume the dirty status.
12758 * Flushing outside of mmu_lock requires KVM to be careful when
12759 * making decisions based on writable status of an SPTE, e.g. a
12760 * !writable SPTE doesn't guarantee a CPU can't perform writes.
12762 * Specifically, KVM also write-protects guest page tables to
12763 * monitor changes when using shadow paging, and must guarantee
12764 * no CPUs can write to those page before mmu_lock is dropped.
12765 * Because CPUs may have stale TLB entries at this point, a
12766 * !writable SPTE doesn't guarantee CPUs can't perform writes.
12768 * KVM also allows making SPTES writable outside of mmu_lock,
12769 * e.g. to allow dirty logging without taking mmu_lock.
12771 * To handle these scenarios, KVM uses a separate software-only
12772 * bit (MMU-writable) to track if a SPTE is !writable due to
12773 * a guest page table being write-protected (KVM clears the
12774 * MMU-writable flag when write-protecting for shadow paging).
12776 * The use of MMU-writable is also the primary motivation for
12777 * the unconditional flush. Because KVM must guarantee that a
12778 * CPU doesn't contain stale, writable TLB entries for a
12779 * !MMU-writable SPTE, KVM must flush if it encounters any
12780 * MMU-writable SPTE regardless of whether the actual hardware
12781 * writable bit was set. I.e. KVM is almost guaranteed to need
12782 * to flush, while unconditionally flushing allows the "remove
12783 * write access" helpers to ignore MMU-writable entirely.
12785 * See is_writable_pte() for more details (the case involving
12786 * access-tracked SPTEs is particularly relevant).
12788 kvm_flush_remote_tlbs_memslot(kvm, new);
12792 void kvm_arch_commit_memory_region(struct kvm *kvm,
12793 struct kvm_memory_slot *old,
12794 const struct kvm_memory_slot *new,
12795 enum kvm_mr_change change)
12797 if (change == KVM_MR_DELETE)
12798 kvm_page_track_delete_slot(kvm, old);
12800 if (!kvm->arch.n_requested_mmu_pages &&
12801 (change == KVM_MR_CREATE || change == KVM_MR_DELETE)) {
12802 unsigned long nr_mmu_pages;
12804 nr_mmu_pages = kvm->nr_memslot_pages / KVM_MEMSLOT_PAGES_TO_MMU_PAGES_RATIO;
12805 nr_mmu_pages = max(nr_mmu_pages, KVM_MIN_ALLOC_MMU_PAGES);
12806 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
12809 kvm_mmu_slot_apply_flags(kvm, old, new, change);
12811 /* Free the arrays associated with the old memslot. */
12812 if (change == KVM_MR_MOVE)
12813 kvm_arch_free_memslot(kvm, old);
12816 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
12818 return (is_guest_mode(vcpu) &&
12819 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
12822 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
12824 if (!list_empty_careful(&vcpu->async_pf.done))
12827 if (kvm_apic_has_pending_init_or_sipi(vcpu) &&
12828 kvm_apic_init_sipi_allowed(vcpu))
12831 if (vcpu->arch.pv.pv_unhalted)
12834 if (kvm_is_exception_pending(vcpu))
12837 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12838 (vcpu->arch.nmi_pending &&
12839 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
12842 #ifdef CONFIG_KVM_SMM
12843 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
12844 (vcpu->arch.smi_pending &&
12845 static_call(kvm_x86_smi_allowed)(vcpu, false)))
12849 if (kvm_arch_interrupt_allowed(vcpu) &&
12850 (kvm_cpu_has_interrupt(vcpu) ||
12851 kvm_guest_apic_has_interrupt(vcpu)))
12854 if (kvm_hv_has_stimer_pending(vcpu))
12857 if (is_guest_mode(vcpu) &&
12858 kvm_x86_ops.nested_ops->has_events &&
12859 kvm_x86_ops.nested_ops->has_events(vcpu))
12862 if (kvm_xen_has_pending_events(vcpu))
12868 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
12870 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
12873 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
12875 if (kvm_vcpu_apicv_active(vcpu) &&
12876 static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
12882 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
12884 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
12887 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12888 #ifdef CONFIG_KVM_SMM
12889 kvm_test_request(KVM_REQ_SMI, vcpu) ||
12891 kvm_test_request(KVM_REQ_EVENT, vcpu))
12894 return kvm_arch_dy_has_pending_interrupt(vcpu);
12897 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
12899 if (vcpu->arch.guest_state_protected)
12902 return vcpu->arch.preempted_in_kernel;
12905 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
12907 return kvm_rip_read(vcpu);
12910 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
12912 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
12915 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
12917 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
12920 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
12922 /* Can't read the RIP when guest state is protected, just return 0 */
12923 if (vcpu->arch.guest_state_protected)
12926 if (is_64_bit_mode(vcpu))
12927 return kvm_rip_read(vcpu);
12928 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
12929 kvm_rip_read(vcpu));
12931 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
12933 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
12935 return kvm_get_linear_rip(vcpu) == linear_rip;
12937 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
12939 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
12941 unsigned long rflags;
12943 rflags = static_call(kvm_x86_get_rflags)(vcpu);
12944 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
12945 rflags &= ~X86_EFLAGS_TF;
12948 EXPORT_SYMBOL_GPL(kvm_get_rflags);
12950 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12952 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
12953 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
12954 rflags |= X86_EFLAGS_TF;
12955 static_call(kvm_x86_set_rflags)(vcpu, rflags);
12958 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12960 __kvm_set_rflags(vcpu, rflags);
12961 kvm_make_request(KVM_REQ_EVENT, vcpu);
12963 EXPORT_SYMBOL_GPL(kvm_set_rflags);
12965 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
12967 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
12969 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
12972 static inline u32 kvm_async_pf_next_probe(u32 key)
12974 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
12977 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12979 u32 key = kvm_async_pf_hash_fn(gfn);
12981 while (vcpu->arch.apf.gfns[key] != ~0)
12982 key = kvm_async_pf_next_probe(key);
12984 vcpu->arch.apf.gfns[key] = gfn;
12987 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
12990 u32 key = kvm_async_pf_hash_fn(gfn);
12992 for (i = 0; i < ASYNC_PF_PER_VCPU &&
12993 (vcpu->arch.apf.gfns[key] != gfn &&
12994 vcpu->arch.apf.gfns[key] != ~0); i++)
12995 key = kvm_async_pf_next_probe(key);
13000 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
13002 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
13005 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
13009 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
13011 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
13015 vcpu->arch.apf.gfns[i] = ~0;
13017 j = kvm_async_pf_next_probe(j);
13018 if (vcpu->arch.apf.gfns[j] == ~0)
13020 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
13022 * k lies cyclically in ]i,j]
13024 * |....j i.k.| or |.k..j i...|
13026 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
13027 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
13032 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
13034 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
13036 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
13040 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
13042 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
13044 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
13045 &token, offset, sizeof(token));
13048 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
13050 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
13053 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
13054 &val, offset, sizeof(val)))
13060 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
13063 if (!kvm_pv_async_pf_enabled(vcpu))
13066 if (vcpu->arch.apf.send_user_only &&
13067 static_call(kvm_x86_get_cpl)(vcpu) == 0)
13070 if (is_guest_mode(vcpu)) {
13072 * L1 needs to opt into the special #PF vmexits that are
13073 * used to deliver async page faults.
13075 return vcpu->arch.apf.delivery_as_pf_vmexit;
13078 * Play it safe in case the guest temporarily disables paging.
13079 * The real mode IDT in particular is unlikely to have a #PF
13082 return is_paging(vcpu);
13086 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
13088 if (unlikely(!lapic_in_kernel(vcpu) ||
13089 kvm_event_needs_reinjection(vcpu) ||
13090 kvm_is_exception_pending(vcpu)))
13093 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
13097 * If interrupts are off we cannot even use an artificial
13100 return kvm_arch_interrupt_allowed(vcpu);
13103 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
13104 struct kvm_async_pf *work)
13106 struct x86_exception fault;
13108 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
13109 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
13111 if (kvm_can_deliver_async_pf(vcpu) &&
13112 !apf_put_user_notpresent(vcpu)) {
13113 fault.vector = PF_VECTOR;
13114 fault.error_code_valid = true;
13115 fault.error_code = 0;
13116 fault.nested_page_fault = false;
13117 fault.address = work->arch.token;
13118 fault.async_page_fault = true;
13119 kvm_inject_page_fault(vcpu, &fault);
13123 * It is not possible to deliver a paravirtualized asynchronous
13124 * page fault, but putting the guest in an artificial halt state
13125 * can be beneficial nevertheless: if an interrupt arrives, we
13126 * can deliver it timely and perhaps the guest will schedule
13127 * another process. When the instruction that triggered a page
13128 * fault is retried, hopefully the page will be ready in the host.
13130 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
13135 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
13136 struct kvm_async_pf *work)
13138 struct kvm_lapic_irq irq = {
13139 .delivery_mode = APIC_DM_FIXED,
13140 .vector = vcpu->arch.apf.vec
13143 if (work->wakeup_all)
13144 work->arch.token = ~0; /* broadcast wakeup */
13146 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
13147 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
13149 if ((work->wakeup_all || work->notpresent_injected) &&
13150 kvm_pv_async_pf_enabled(vcpu) &&
13151 !apf_put_user_ready(vcpu, work->arch.token)) {
13152 vcpu->arch.apf.pageready_pending = true;
13153 kvm_apic_set_irq(vcpu, &irq, NULL);
13156 vcpu->arch.apf.halted = false;
13157 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
13160 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
13162 kvm_make_request(KVM_REQ_APF_READY, vcpu);
13163 if (!vcpu->arch.apf.pageready_pending)
13164 kvm_vcpu_kick(vcpu);
13167 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
13169 if (!kvm_pv_async_pf_enabled(vcpu))
13172 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
13175 void kvm_arch_start_assignment(struct kvm *kvm)
13177 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
13178 static_call_cond(kvm_x86_pi_start_assignment)(kvm);
13180 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
13182 void kvm_arch_end_assignment(struct kvm *kvm)
13184 atomic_dec(&kvm->arch.assigned_device_count);
13186 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
13188 bool noinstr kvm_arch_has_assigned_device(struct kvm *kvm)
13190 return raw_atomic_read(&kvm->arch.assigned_device_count);
13192 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
13194 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
13196 atomic_inc(&kvm->arch.noncoherent_dma_count);
13198 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
13200 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
13202 atomic_dec(&kvm->arch.noncoherent_dma_count);
13204 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
13206 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
13208 return atomic_read(&kvm->arch.noncoherent_dma_count);
13210 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
13212 bool kvm_arch_has_irq_bypass(void)
13214 return enable_apicv && irq_remapping_cap(IRQ_POSTING_CAP);
13217 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
13218 struct irq_bypass_producer *prod)
13220 struct kvm_kernel_irqfd *irqfd =
13221 container_of(cons, struct kvm_kernel_irqfd, consumer);
13224 irqfd->producer = prod;
13225 kvm_arch_start_assignment(irqfd->kvm);
13226 ret = static_call(kvm_x86_pi_update_irte)(irqfd->kvm,
13227 prod->irq, irqfd->gsi, 1);
13230 kvm_arch_end_assignment(irqfd->kvm);
13235 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
13236 struct irq_bypass_producer *prod)
13239 struct kvm_kernel_irqfd *irqfd =
13240 container_of(cons, struct kvm_kernel_irqfd, consumer);
13242 WARN_ON(irqfd->producer != prod);
13243 irqfd->producer = NULL;
13246 * When producer of consumer is unregistered, we change back to
13247 * remapped mode, so we can re-use the current implementation
13248 * when the irq is masked/disabled or the consumer side (KVM
13249 * int this case doesn't want to receive the interrupts.
13251 ret = static_call(kvm_x86_pi_update_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
13253 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
13254 " fails: %d\n", irqfd->consumer.token, ret);
13256 kvm_arch_end_assignment(irqfd->kvm);
13259 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
13260 uint32_t guest_irq, bool set)
13262 return static_call(kvm_x86_pi_update_irte)(kvm, host_irq, guest_irq, set);
13265 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
13266 struct kvm_kernel_irq_routing_entry *new)
13268 if (new->type != KVM_IRQ_ROUTING_MSI)
13271 return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
13274 bool kvm_vector_hashing_enabled(void)
13276 return vector_hashing;
13279 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
13281 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
13283 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
13286 int kvm_spec_ctrl_test_value(u64 value)
13289 * test that setting IA32_SPEC_CTRL to given value
13290 * is allowed by the host processor
13294 unsigned long flags;
13297 local_irq_save(flags);
13299 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
13301 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
13304 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
13306 local_irq_restore(flags);
13310 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
13312 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
13314 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
13315 struct x86_exception fault;
13316 u64 access = error_code &
13317 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
13319 if (!(error_code & PFERR_PRESENT_MASK) ||
13320 mmu->gva_to_gpa(vcpu, mmu, gva, access, &fault) != INVALID_GPA) {
13322 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
13323 * tables probably do not match the TLB. Just proceed
13324 * with the error code that the processor gave.
13326 fault.vector = PF_VECTOR;
13327 fault.error_code_valid = true;
13328 fault.error_code = error_code;
13329 fault.nested_page_fault = false;
13330 fault.address = gva;
13331 fault.async_page_fault = false;
13333 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
13335 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
13338 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
13339 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
13340 * indicates whether exit to userspace is needed.
13342 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
13343 struct x86_exception *e)
13345 if (r == X86EMUL_PROPAGATE_FAULT) {
13346 if (KVM_BUG_ON(!e, vcpu->kvm))
13349 kvm_inject_emulated_page_fault(vcpu, e);
13354 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
13355 * while handling a VMX instruction KVM could've handled the request
13356 * correctly by exiting to userspace and performing I/O but there
13357 * doesn't seem to be a real use-case behind such requests, just return
13358 * KVM_EXIT_INTERNAL_ERROR for now.
13360 kvm_prepare_emulation_failure_exit(vcpu);
13364 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
13366 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
13369 struct x86_exception e;
13376 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
13377 if (r != X86EMUL_CONTINUE)
13378 return kvm_handle_memory_failure(vcpu, r, &e);
13380 if (operand.pcid >> 12 != 0) {
13381 kvm_inject_gp(vcpu, 0);
13385 pcid_enabled = kvm_is_cr4_bit_set(vcpu, X86_CR4_PCIDE);
13388 case INVPCID_TYPE_INDIV_ADDR:
13389 if ((!pcid_enabled && (operand.pcid != 0)) ||
13390 is_noncanonical_address(operand.gla, vcpu)) {
13391 kvm_inject_gp(vcpu, 0);
13394 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
13395 return kvm_skip_emulated_instruction(vcpu);
13397 case INVPCID_TYPE_SINGLE_CTXT:
13398 if (!pcid_enabled && (operand.pcid != 0)) {
13399 kvm_inject_gp(vcpu, 0);
13403 kvm_invalidate_pcid(vcpu, operand.pcid);
13404 return kvm_skip_emulated_instruction(vcpu);
13406 case INVPCID_TYPE_ALL_NON_GLOBAL:
13408 * Currently, KVM doesn't mark global entries in the shadow
13409 * page tables, so a non-global flush just degenerates to a
13410 * global flush. If needed, we could optimize this later by
13411 * keeping track of global entries in shadow page tables.
13415 case INVPCID_TYPE_ALL_INCL_GLOBAL:
13416 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
13417 return kvm_skip_emulated_instruction(vcpu);
13420 kvm_inject_gp(vcpu, 0);
13424 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
13426 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
13428 struct kvm_run *run = vcpu->run;
13429 struct kvm_mmio_fragment *frag;
13432 BUG_ON(!vcpu->mmio_needed);
13434 /* Complete previous fragment */
13435 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
13436 len = min(8u, frag->len);
13437 if (!vcpu->mmio_is_write)
13438 memcpy(frag->data, run->mmio.data, len);
13440 if (frag->len <= 8) {
13441 /* Switch to the next fragment. */
13443 vcpu->mmio_cur_fragment++;
13445 /* Go forward to the next mmio piece. */
13451 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
13452 vcpu->mmio_needed = 0;
13454 // VMG change, at this point, we're always done
13455 // RIP has already been advanced
13459 // More MMIO is needed
13460 run->mmio.phys_addr = frag->gpa;
13461 run->mmio.len = min(8u, frag->len);
13462 run->mmio.is_write = vcpu->mmio_is_write;
13463 if (run->mmio.is_write)
13464 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
13465 run->exit_reason = KVM_EXIT_MMIO;
13467 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13472 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
13476 struct kvm_mmio_fragment *frag;
13481 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
13482 if (handled == bytes)
13489 /*TODO: Check if need to increment number of frags */
13490 frag = vcpu->mmio_fragments;
13491 vcpu->mmio_nr_fragments = 1;
13496 vcpu->mmio_needed = 1;
13497 vcpu->mmio_cur_fragment = 0;
13499 vcpu->run->mmio.phys_addr = gpa;
13500 vcpu->run->mmio.len = min(8u, frag->len);
13501 vcpu->run->mmio.is_write = 1;
13502 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
13503 vcpu->run->exit_reason = KVM_EXIT_MMIO;
13505 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13509 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
13511 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
13515 struct kvm_mmio_fragment *frag;
13520 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
13521 if (handled == bytes)
13528 /*TODO: Check if need to increment number of frags */
13529 frag = vcpu->mmio_fragments;
13530 vcpu->mmio_nr_fragments = 1;
13535 vcpu->mmio_needed = 1;
13536 vcpu->mmio_cur_fragment = 0;
13538 vcpu->run->mmio.phys_addr = gpa;
13539 vcpu->run->mmio.len = min(8u, frag->len);
13540 vcpu->run->mmio.is_write = 0;
13541 vcpu->run->exit_reason = KVM_EXIT_MMIO;
13543 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13547 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
13549 static void advance_sev_es_emulated_pio(struct kvm_vcpu *vcpu, unsigned count, int size)
13551 vcpu->arch.sev_pio_count -= count;
13552 vcpu->arch.sev_pio_data += count * size;
13555 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
13556 unsigned int port);
13558 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
13560 int size = vcpu->arch.pio.size;
13561 int port = vcpu->arch.pio.port;
13563 vcpu->arch.pio.count = 0;
13564 if (vcpu->arch.sev_pio_count)
13565 return kvm_sev_es_outs(vcpu, size, port);
13569 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
13573 unsigned int count =
13574 min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
13575 int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
13577 /* memcpy done already by emulator_pio_out. */
13578 advance_sev_es_emulated_pio(vcpu, count, size);
13582 /* Emulation done by the kernel. */
13583 if (!vcpu->arch.sev_pio_count)
13587 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
13591 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
13592 unsigned int port);
13594 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
13596 unsigned count = vcpu->arch.pio.count;
13597 int size = vcpu->arch.pio.size;
13598 int port = vcpu->arch.pio.port;
13600 complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
13601 advance_sev_es_emulated_pio(vcpu, count, size);
13602 if (vcpu->arch.sev_pio_count)
13603 return kvm_sev_es_ins(vcpu, size, port);
13607 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
13611 unsigned int count =
13612 min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
13613 if (!emulator_pio_in(vcpu, size, port, vcpu->arch.sev_pio_data, count))
13616 /* Emulation done by the kernel. */
13617 advance_sev_es_emulated_pio(vcpu, count, size);
13618 if (!vcpu->arch.sev_pio_count)
13622 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
13626 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
13627 unsigned int port, void *data, unsigned int count,
13630 vcpu->arch.sev_pio_data = data;
13631 vcpu->arch.sev_pio_count = count;
13632 return in ? kvm_sev_es_ins(vcpu, size, port)
13633 : kvm_sev_es_outs(vcpu, size, port);
13635 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
13637 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
13638 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
13639 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
13640 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
13641 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
13642 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
13643 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
13644 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter);
13645 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
13646 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
13647 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
13648 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
13649 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
13650 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
13651 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
13652 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
13653 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
13654 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
13655 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
13656 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
13657 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
13658 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
13659 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_kick_vcpu_slowpath);
13660 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_doorbell);
13661 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_accept_irq);
13662 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
13663 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
13664 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
13665 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
13667 static int __init kvm_x86_init(void)
13669 kvm_mmu_x86_module_init();
13670 mitigate_smt_rsb &= boot_cpu_has_bug(X86_BUG_SMT_RSB) && cpu_smt_possible();
13673 module_init(kvm_x86_init);
13675 static void __exit kvm_x86_exit(void)
13678 * If module_init() is implemented, module_exit() must also be
13679 * implemented to allow module unload.
13682 module_exit(kvm_x86_exit);