1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 #include <linux/suspend.h>
63 #include <trace/events/kvm.h>
65 #include <asm/debugreg.h>
70 #include <linux/kernel_stat.h>
71 #include <asm/fpu/internal.h> /* Ugh! */
72 #include <asm/pvclock.h>
73 #include <asm/div64.h>
74 #include <asm/irq_remapping.h>
75 #include <asm/mshyperv.h>
76 #include <asm/hypervisor.h>
77 #include <asm/tlbflush.h>
78 #include <asm/intel_pt.h>
79 #include <asm/emulate_prefix.h>
81 #include <clocksource/hyperv_timer.h>
83 #define CREATE_TRACE_POINTS
86 #define MAX_IO_MSRS 256
87 #define KVM_MAX_MCE_BANKS 32
88 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
89 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
91 #define emul_to_vcpu(ctxt) \
92 ((struct kvm_vcpu *)(ctxt)->vcpu)
95 * - enable syscall per default because its emulated by KVM
96 * - enable LME and LMA per default on 64 bit KVM
100 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
102 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
105 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
107 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
109 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
110 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
112 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
113 static void process_nmi(struct kvm_vcpu *vcpu);
114 static void process_smi(struct kvm_vcpu *vcpu);
115 static void enter_smm(struct kvm_vcpu *vcpu);
116 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
117 static void store_regs(struct kvm_vcpu *vcpu);
118 static int sync_regs(struct kvm_vcpu *vcpu);
120 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
121 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
123 struct kvm_x86_ops kvm_x86_ops __read_mostly;
124 EXPORT_SYMBOL_GPL(kvm_x86_ops);
126 #define KVM_X86_OP(func) \
127 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
128 *(((struct kvm_x86_ops *)0)->func));
129 #define KVM_X86_OP_NULL KVM_X86_OP
130 #include <asm/kvm-x86-ops.h>
131 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
132 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
133 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
135 static bool __read_mostly ignore_msrs = 0;
136 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
138 bool __read_mostly report_ignored_msrs = true;
139 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
140 EXPORT_SYMBOL_GPL(report_ignored_msrs);
142 unsigned int min_timer_period_us = 200;
143 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
145 static bool __read_mostly kvmclock_periodic_sync = true;
146 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
148 bool __read_mostly kvm_has_tsc_control;
149 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
150 u32 __read_mostly kvm_max_guest_tsc_khz;
151 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
152 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
153 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
154 u64 __read_mostly kvm_max_tsc_scaling_ratio;
155 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
156 u64 __read_mostly kvm_default_tsc_scaling_ratio;
157 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
158 bool __read_mostly kvm_has_bus_lock_exit;
159 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
161 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
162 static u32 __read_mostly tsc_tolerance_ppm = 250;
163 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
166 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
167 * adaptive tuning starting from default advancement of 1000ns. '0' disables
168 * advancement entirely. Any other value is used as-is and disables adaptive
169 * tuning, i.e. allows privileged userspace to set an exact advancement time.
171 static int __read_mostly lapic_timer_advance_ns = -1;
172 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
174 static bool __read_mostly vector_hashing = true;
175 module_param(vector_hashing, bool, S_IRUGO);
177 bool __read_mostly enable_vmware_backdoor = false;
178 module_param(enable_vmware_backdoor, bool, S_IRUGO);
179 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
181 static bool __read_mostly force_emulation_prefix = false;
182 module_param(force_emulation_prefix, bool, S_IRUGO);
184 int __read_mostly pi_inject_timer = -1;
185 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
187 /* Enable/disable SMT_RSB bug mitigation */
188 bool __read_mostly mitigate_smt_rsb;
189 module_param(mitigate_smt_rsb, bool, 0444);
192 * Restoring the host value for MSRs that are only consumed when running in
193 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
194 * returns to userspace, i.e. the kernel can run with the guest's value.
196 #define KVM_MAX_NR_USER_RETURN_MSRS 16
198 struct kvm_user_return_msrs {
199 struct user_return_notifier urn;
201 struct kvm_user_return_msr_values {
204 } values[KVM_MAX_NR_USER_RETURN_MSRS];
207 u32 __read_mostly kvm_nr_uret_msrs;
208 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
209 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
210 static struct kvm_user_return_msrs __percpu *user_return_msrs;
212 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
213 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
214 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
215 | XFEATURE_MASK_PKRU)
217 u64 __read_mostly host_efer;
218 EXPORT_SYMBOL_GPL(host_efer);
220 bool __read_mostly allow_smaller_maxphyaddr = 0;
221 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
223 bool __read_mostly enable_apicv = true;
224 EXPORT_SYMBOL_GPL(enable_apicv);
226 u64 __read_mostly host_xss;
227 EXPORT_SYMBOL_GPL(host_xss);
228 u64 __read_mostly supported_xss;
229 EXPORT_SYMBOL_GPL(supported_xss);
231 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
232 KVM_GENERIC_VM_STATS(),
233 STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
234 STATS_DESC_COUNTER(VM, mmu_pte_write),
235 STATS_DESC_COUNTER(VM, mmu_pde_zapped),
236 STATS_DESC_COUNTER(VM, mmu_flooded),
237 STATS_DESC_COUNTER(VM, mmu_recycled),
238 STATS_DESC_COUNTER(VM, mmu_cache_miss),
239 STATS_DESC_ICOUNTER(VM, mmu_unsync),
240 STATS_DESC_ICOUNTER(VM, pages_4k),
241 STATS_DESC_ICOUNTER(VM, pages_2m),
242 STATS_DESC_ICOUNTER(VM, pages_1g),
243 STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
244 STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
245 STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
248 const struct kvm_stats_header kvm_vm_stats_header = {
249 .name_size = KVM_STATS_NAME_SIZE,
250 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
251 .id_offset = sizeof(struct kvm_stats_header),
252 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
253 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
254 sizeof(kvm_vm_stats_desc),
257 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
258 KVM_GENERIC_VCPU_STATS(),
259 STATS_DESC_COUNTER(VCPU, pf_fixed),
260 STATS_DESC_COUNTER(VCPU, pf_guest),
261 STATS_DESC_COUNTER(VCPU, tlb_flush),
262 STATS_DESC_COUNTER(VCPU, invlpg),
263 STATS_DESC_COUNTER(VCPU, exits),
264 STATS_DESC_COUNTER(VCPU, io_exits),
265 STATS_DESC_COUNTER(VCPU, mmio_exits),
266 STATS_DESC_COUNTER(VCPU, signal_exits),
267 STATS_DESC_COUNTER(VCPU, irq_window_exits),
268 STATS_DESC_COUNTER(VCPU, nmi_window_exits),
269 STATS_DESC_COUNTER(VCPU, l1d_flush),
270 STATS_DESC_COUNTER(VCPU, halt_exits),
271 STATS_DESC_COUNTER(VCPU, request_irq_exits),
272 STATS_DESC_COUNTER(VCPU, irq_exits),
273 STATS_DESC_COUNTER(VCPU, host_state_reload),
274 STATS_DESC_COUNTER(VCPU, fpu_reload),
275 STATS_DESC_COUNTER(VCPU, insn_emulation),
276 STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
277 STATS_DESC_COUNTER(VCPU, hypercalls),
278 STATS_DESC_COUNTER(VCPU, irq_injections),
279 STATS_DESC_COUNTER(VCPU, nmi_injections),
280 STATS_DESC_COUNTER(VCPU, req_event),
281 STATS_DESC_COUNTER(VCPU, nested_run),
282 STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
283 STATS_DESC_COUNTER(VCPU, directed_yield_successful),
284 STATS_DESC_COUNTER(VCPU, preemption_reported),
285 STATS_DESC_COUNTER(VCPU, preemption_other),
286 STATS_DESC_ICOUNTER(VCPU, guest_mode)
289 const struct kvm_stats_header kvm_vcpu_stats_header = {
290 .name_size = KVM_STATS_NAME_SIZE,
291 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
292 .id_offset = sizeof(struct kvm_stats_header),
293 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
294 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
295 sizeof(kvm_vcpu_stats_desc),
298 u64 __read_mostly host_xcr0;
299 u64 __read_mostly supported_xcr0;
300 EXPORT_SYMBOL_GPL(supported_xcr0);
302 static struct kmem_cache *x86_fpu_cache;
304 static struct kmem_cache *x86_emulator_cache;
307 * When called, it means the previous get/set msr reached an invalid msr.
308 * Return true if we want to ignore/silent this failed msr access.
310 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
312 const char *op = write ? "wrmsr" : "rdmsr";
315 if (report_ignored_msrs)
316 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
321 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
327 static struct kmem_cache *kvm_alloc_emulator_cache(void)
329 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
330 unsigned int size = sizeof(struct x86_emulate_ctxt);
332 return kmem_cache_create_usercopy("x86_emulator", size,
333 __alignof__(struct x86_emulate_ctxt),
334 SLAB_ACCOUNT, useroffset,
335 size - useroffset, NULL);
338 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
340 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
343 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
344 vcpu->arch.apf.gfns[i] = ~0;
347 static void kvm_on_user_return(struct user_return_notifier *urn)
350 struct kvm_user_return_msrs *msrs
351 = container_of(urn, struct kvm_user_return_msrs, urn);
352 struct kvm_user_return_msr_values *values;
356 * Disabling irqs at this point since the following code could be
357 * interrupted and executed through kvm_arch_hardware_disable()
359 local_irq_save(flags);
360 if (msrs->registered) {
361 msrs->registered = false;
362 user_return_notifier_unregister(urn);
364 local_irq_restore(flags);
365 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
366 values = &msrs->values[slot];
367 if (values->host != values->curr) {
368 wrmsrl(kvm_uret_msrs_list[slot], values->host);
369 values->curr = values->host;
374 static int kvm_probe_user_return_msr(u32 msr)
380 ret = rdmsrl_safe(msr, &val);
383 ret = wrmsrl_safe(msr, val);
389 int kvm_add_user_return_msr(u32 msr)
391 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
393 if (kvm_probe_user_return_msr(msr))
396 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
397 return kvm_nr_uret_msrs++;
399 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
401 int kvm_find_user_return_msr(u32 msr)
405 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
406 if (kvm_uret_msrs_list[i] == msr)
411 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
413 static void kvm_user_return_msr_cpu_online(void)
415 unsigned int cpu = smp_processor_id();
416 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
420 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
421 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
422 msrs->values[i].host = value;
423 msrs->values[i].curr = value;
427 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
429 unsigned int cpu = smp_processor_id();
430 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
433 value = (value & mask) | (msrs->values[slot].host & ~mask);
434 if (value == msrs->values[slot].curr)
436 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
440 msrs->values[slot].curr = value;
441 if (!msrs->registered) {
442 msrs->urn.on_user_return = kvm_on_user_return;
443 user_return_notifier_register(&msrs->urn);
444 msrs->registered = true;
448 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
450 static void drop_user_return_notifiers(void)
452 unsigned int cpu = smp_processor_id();
453 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
455 if (msrs->registered)
456 kvm_on_user_return(&msrs->urn);
459 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
461 return vcpu->arch.apic_base;
463 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
465 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
467 return kvm_apic_mode(kvm_get_apic_base(vcpu));
469 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
471 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
473 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
474 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
475 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
476 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
478 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
480 if (!msr_info->host_initiated) {
481 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
483 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
487 kvm_lapic_set_base(vcpu, msr_info->data);
488 kvm_recalculate_apic_map(vcpu->kvm);
491 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
494 * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
496 * Hardware virtualization extension instructions may fault if a reboot turns
497 * off virtualization while processes are running. Usually after catching the
498 * fault we just panic; during reboot instead the instruction is ignored.
500 noinstr void kvm_spurious_fault(void)
502 /* Fault while not rebooting. We want the trace. */
503 BUG_ON(!kvm_rebooting);
505 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
507 #define EXCPT_BENIGN 0
508 #define EXCPT_CONTRIBUTORY 1
511 static int exception_class(int vector)
521 return EXCPT_CONTRIBUTORY;
528 #define EXCPT_FAULT 0
530 #define EXCPT_ABORT 2
531 #define EXCPT_INTERRUPT 3
534 static int exception_type(int vector)
538 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
539 return EXCPT_INTERRUPT;
544 * #DBs can be trap-like or fault-like, the caller must check other CPU
545 * state, e.g. DR6, to determine whether a #DB is a trap or fault.
547 if (mask & (1 << DB_VECTOR))
550 if (mask & ((1 << BP_VECTOR) | (1 << OF_VECTOR)))
553 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
556 /* Reserved exceptions will result in fault */
560 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
562 unsigned nr = vcpu->arch.exception.nr;
563 bool has_payload = vcpu->arch.exception.has_payload;
564 unsigned long payload = vcpu->arch.exception.payload;
572 * "Certain debug exceptions may clear bit 0-3. The
573 * remaining contents of the DR6 register are never
574 * cleared by the processor".
576 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
578 * In order to reflect the #DB exception payload in guest
579 * dr6, three components need to be considered: active low
580 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
582 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
583 * In the target guest dr6:
584 * FIXED_1 bits should always be set.
585 * Active low bits should be cleared if 1-setting in payload.
586 * Active high bits should be set if 1-setting in payload.
588 * Note, the payload is compatible with the pending debug
589 * exceptions/exit qualification under VMX, that active_low bits
590 * are active high in payload.
591 * So they need to be flipped for DR6.
593 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
594 vcpu->arch.dr6 |= payload;
595 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
598 * The #DB payload is defined as compatible with the 'pending
599 * debug exceptions' field under VMX, not DR6. While bit 12 is
600 * defined in the 'pending debug exceptions' field (enabled
601 * breakpoint), it is reserved and must be zero in DR6.
603 vcpu->arch.dr6 &= ~BIT(12);
606 vcpu->arch.cr2 = payload;
610 vcpu->arch.exception.has_payload = false;
611 vcpu->arch.exception.payload = 0;
613 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
615 /* Forcibly leave the nested mode in cases like a vCPU reset */
616 static void kvm_leave_nested(struct kvm_vcpu *vcpu)
618 kvm_x86_ops.nested_ops->leave_nested(vcpu);
621 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
622 unsigned nr, bool has_error, u32 error_code,
623 bool has_payload, unsigned long payload, bool reinject)
628 kvm_make_request(KVM_REQ_EVENT, vcpu);
630 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
634 * On vmentry, vcpu->arch.exception.pending is only
635 * true if an event injection was blocked by
636 * nested_run_pending. In that case, however,
637 * vcpu_enter_guest requests an immediate exit,
638 * and the guest shouldn't proceed far enough to
641 WARN_ON_ONCE(vcpu->arch.exception.pending);
642 vcpu->arch.exception.injected = true;
643 if (WARN_ON_ONCE(has_payload)) {
645 * A reinjected event has already
646 * delivered its payload.
652 vcpu->arch.exception.pending = true;
653 vcpu->arch.exception.injected = false;
655 vcpu->arch.exception.has_error_code = has_error;
656 vcpu->arch.exception.nr = nr;
657 vcpu->arch.exception.error_code = error_code;
658 vcpu->arch.exception.has_payload = has_payload;
659 vcpu->arch.exception.payload = payload;
660 if (!is_guest_mode(vcpu))
661 kvm_deliver_exception_payload(vcpu);
665 /* to check exception */
666 prev_nr = vcpu->arch.exception.nr;
667 if (prev_nr == DF_VECTOR) {
668 /* triple fault -> shutdown */
669 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
672 class1 = exception_class(prev_nr);
673 class2 = exception_class(nr);
674 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
675 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
677 * Generate double fault per SDM Table 5-5. Set
678 * exception.pending = true so that the double fault
679 * can trigger a nested vmexit.
681 vcpu->arch.exception.pending = true;
682 vcpu->arch.exception.injected = false;
683 vcpu->arch.exception.has_error_code = true;
684 vcpu->arch.exception.nr = DF_VECTOR;
685 vcpu->arch.exception.error_code = 0;
686 vcpu->arch.exception.has_payload = false;
687 vcpu->arch.exception.payload = 0;
689 /* replace previous exception with a new one in a hope
690 that instruction re-execution will regenerate lost
695 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
697 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
699 EXPORT_SYMBOL_GPL(kvm_queue_exception);
701 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
703 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
705 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
707 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
708 unsigned long payload)
710 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
712 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
714 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
715 u32 error_code, unsigned long payload)
717 kvm_multiple_exception(vcpu, nr, true, error_code,
718 true, payload, false);
721 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
724 kvm_inject_gp(vcpu, 0);
726 return kvm_skip_emulated_instruction(vcpu);
730 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
732 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
734 ++vcpu->stat.pf_guest;
735 vcpu->arch.exception.nested_apf =
736 is_guest_mode(vcpu) && fault->async_page_fault;
737 if (vcpu->arch.exception.nested_apf) {
738 vcpu->arch.apf.nested_apf_token = fault->address;
739 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
741 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
745 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
747 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
748 struct x86_exception *fault)
750 struct kvm_mmu *fault_mmu;
751 WARN_ON_ONCE(fault->vector != PF_VECTOR);
753 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
757 * Invalidate the TLB entry for the faulting address, if it exists,
758 * else the access will fault indefinitely (and to emulate hardware).
760 if ((fault->error_code & PFERR_PRESENT_MASK) &&
761 !(fault->error_code & PFERR_RSVD_MASK))
762 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
763 fault_mmu->root_hpa);
765 fault_mmu->inject_page_fault(vcpu, fault);
766 return fault->nested_page_fault;
768 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
770 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
772 atomic_inc(&vcpu->arch.nmi_queued);
773 kvm_make_request(KVM_REQ_NMI, vcpu);
775 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
777 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
779 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
781 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
783 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
785 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
787 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
790 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
791 * a #GP and return false.
793 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
795 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
797 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
800 EXPORT_SYMBOL_GPL(kvm_require_cpl);
802 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
804 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
807 kvm_queue_exception(vcpu, UD_VECTOR);
810 EXPORT_SYMBOL_GPL(kvm_require_dr);
813 * This function will be used to read from the physical memory of the currently
814 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
815 * can read from guest physical or from the guest's guest physical memory.
817 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
818 gfn_t ngfn, void *data, int offset, int len,
821 struct x86_exception exception;
825 ngpa = gfn_to_gpa(ngfn);
826 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
827 if (real_gfn == UNMAPPED_GVA)
830 real_gfn = gpa_to_gfn(real_gfn);
832 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
834 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
836 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
838 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
842 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
844 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
846 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
847 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
850 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
852 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
853 offset * sizeof(u64), sizeof(pdpte),
854 PFERR_USER_MASK|PFERR_WRITE_MASK);
859 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
860 if ((pdpte[i] & PT_PRESENT_MASK) &&
861 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
868 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
869 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
870 kvm_make_request(KVM_REQ_LOAD_MMU_PGD, vcpu);
871 vcpu->arch.pdptrs_from_userspace = false;
877 EXPORT_SYMBOL_GPL(load_pdptrs);
879 static bool kvm_is_valid_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
882 if (cr0 & 0xffffffff00000000UL)
886 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
889 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
892 return static_call(kvm_x86_is_valid_cr0)(vcpu, cr0);
895 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
897 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
898 kvm_clear_async_pf_completion_queue(vcpu);
899 kvm_async_pf_hash_reset(vcpu);
902 if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
903 kvm_mmu_reset_context(vcpu);
905 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
906 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
907 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
908 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
910 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
912 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
914 unsigned long old_cr0 = kvm_read_cr0(vcpu);
915 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
917 if (!kvm_is_valid_cr0(vcpu, cr0))
922 /* Write to CR0 reserved bits are ignored, even on Intel. */
923 cr0 &= ~CR0_RESERVED_BITS;
926 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
927 (cr0 & X86_CR0_PG)) {
932 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
937 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
938 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
939 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
942 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
945 static_call(kvm_x86_set_cr0)(vcpu, cr0);
947 kvm_post_set_cr0(vcpu, old_cr0, cr0);
951 EXPORT_SYMBOL_GPL(kvm_set_cr0);
953 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
955 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
957 EXPORT_SYMBOL_GPL(kvm_lmsw);
959 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
961 if (vcpu->arch.guest_state_protected)
964 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
966 if (vcpu->arch.xcr0 != host_xcr0)
967 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
969 if (vcpu->arch.xsaves_enabled &&
970 vcpu->arch.ia32_xss != host_xss)
971 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
974 if (static_cpu_has(X86_FEATURE_PKU) &&
975 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
976 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
977 vcpu->arch.pkru != vcpu->arch.host_pkru)
978 write_pkru(vcpu->arch.pkru);
980 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
982 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
984 if (vcpu->arch.guest_state_protected)
987 if (static_cpu_has(X86_FEATURE_PKU) &&
988 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
989 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
990 vcpu->arch.pkru = rdpkru();
991 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
992 write_pkru(vcpu->arch.host_pkru);
995 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
997 if (vcpu->arch.xcr0 != host_xcr0)
998 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
1000 if (vcpu->arch.xsaves_enabled &&
1001 vcpu->arch.ia32_xss != host_xss)
1002 wrmsrl(MSR_IA32_XSS, host_xss);
1006 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
1008 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
1011 u64 old_xcr0 = vcpu->arch.xcr0;
1014 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
1015 if (index != XCR_XFEATURE_ENABLED_MASK)
1017 if (!(xcr0 & XFEATURE_MASK_FP))
1019 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
1023 * Do not allow the guest to set bits that we do not support
1024 * saving. However, xcr0 bit 0 is always set, even if the
1025 * emulated CPU does not support XSAVE (see fx_init).
1027 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
1028 if (xcr0 & ~valid_bits)
1031 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1032 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1035 if (xcr0 & XFEATURE_MASK_AVX512) {
1036 if (!(xcr0 & XFEATURE_MASK_YMM))
1038 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1041 vcpu->arch.xcr0 = xcr0;
1043 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1044 kvm_update_cpuid_runtime(vcpu);
1048 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1050 /* Note, #UD due to CR4.OSXSAVE=0 has priority over the intercept. */
1051 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1052 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1053 kvm_inject_gp(vcpu, 0);
1057 return kvm_skip_emulated_instruction(vcpu);
1059 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1061 bool __kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1063 if (cr4 & cr4_reserved_bits)
1066 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1071 EXPORT_SYMBOL_GPL(__kvm_is_valid_cr4);
1073 static bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1075 return __kvm_is_valid_cr4(vcpu, cr4) &&
1076 static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1079 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1081 if (((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS) ||
1082 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1083 kvm_mmu_reset_context(vcpu);
1085 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1087 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1089 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1090 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1093 if (!kvm_is_valid_cr4(vcpu, cr4))
1096 if (is_long_mode(vcpu)) {
1097 if (!(cr4 & X86_CR4_PAE))
1099 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1101 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1102 && ((cr4 ^ old_cr4) & pdptr_bits)
1103 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1104 kvm_read_cr3(vcpu)))
1107 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1108 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1111 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1112 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1116 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1118 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1122 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1124 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1126 struct kvm_mmu *mmu = vcpu->arch.mmu;
1127 unsigned long roots_to_free = 0;
1131 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1132 * this is reachable when running EPT=1 and unrestricted_guest=0, and
1133 * also via the emulator. KVM's TDP page tables are not in the scope of
1134 * the invalidation, but the guest's TLB entries need to be flushed as
1135 * the CPU may have cached entries in its TLB for the target PCID.
1137 if (unlikely(tdp_enabled)) {
1138 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1143 * If neither the current CR3 nor any of the prev_roots use the given
1144 * PCID, then nothing needs to be done here because a resync will
1145 * happen anyway before switching to any other CR3.
1147 if (kvm_get_active_pcid(vcpu) == pcid) {
1148 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1149 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1152 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1153 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1154 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1156 kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1159 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1161 bool skip_tlb_flush = false;
1162 unsigned long pcid = 0;
1163 #ifdef CONFIG_X86_64
1164 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1167 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1168 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1169 pcid = cr3 & X86_CR3_PCID_MASK;
1173 /* PDPTRs are always reloaded for PAE paging. */
1174 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1175 goto handle_tlb_flush;
1178 * Do not condition the GPA check on long mode, this helper is used to
1179 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1180 * the current vCPU mode is accurate.
1182 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1185 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1188 if (cr3 != kvm_read_cr3(vcpu))
1189 kvm_mmu_new_pgd(vcpu, cr3);
1191 vcpu->arch.cr3 = cr3;
1192 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1196 * A load of CR3 that flushes the TLB flushes only the current PCID,
1197 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1198 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1199 * and it's impossible to use a non-zero PCID when PCID is disabled,
1200 * i.e. only PCID=0 can be relevant.
1202 if (!skip_tlb_flush)
1203 kvm_invalidate_pcid(vcpu, pcid);
1207 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1209 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1211 if (cr8 & CR8_RESERVED_BITS)
1213 if (lapic_in_kernel(vcpu))
1214 kvm_lapic_set_tpr(vcpu, cr8);
1216 vcpu->arch.cr8 = cr8;
1219 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1221 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1223 if (lapic_in_kernel(vcpu))
1224 return kvm_lapic_get_cr8(vcpu);
1226 return vcpu->arch.cr8;
1228 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1230 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1234 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1235 for (i = 0; i < KVM_NR_DB_REGS; i++)
1236 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1240 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1244 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1245 dr7 = vcpu->arch.guest_debug_dr7;
1247 dr7 = vcpu->arch.dr7;
1248 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1249 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1250 if (dr7 & DR7_BP_EN_MASK)
1251 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1253 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1255 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1257 u64 fixed = DR6_FIXED_1;
1259 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1262 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1263 fixed |= DR6_BUS_LOCK;
1267 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1269 size_t size = ARRAY_SIZE(vcpu->arch.db);
1273 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1274 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1275 vcpu->arch.eff_db[dr] = val;
1279 if (!kvm_dr6_valid(val))
1281 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1285 if (!kvm_dr7_valid(val))
1287 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1288 kvm_update_dr7(vcpu);
1294 EXPORT_SYMBOL_GPL(kvm_set_dr);
1296 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1298 size_t size = ARRAY_SIZE(vcpu->arch.db);
1302 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1306 *val = vcpu->arch.dr6;
1310 *val = vcpu->arch.dr7;
1314 EXPORT_SYMBOL_GPL(kvm_get_dr);
1316 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1318 u32 ecx = kvm_rcx_read(vcpu);
1321 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1322 kvm_inject_gp(vcpu, 0);
1326 kvm_rax_write(vcpu, (u32)data);
1327 kvm_rdx_write(vcpu, data >> 32);
1328 return kvm_skip_emulated_instruction(vcpu);
1330 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1333 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1334 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1336 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1337 * extract the supported MSRs from the related const lists.
1338 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1339 * capabilities of the host cpu. This capabilities test skips MSRs that are
1340 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1341 * may depend on host virtualization features rather than host cpu features.
1344 static const u32 msrs_to_save_all[] = {
1345 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1347 #ifdef CONFIG_X86_64
1348 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1350 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1351 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1353 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1354 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1355 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1356 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1357 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1358 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1359 MSR_IA32_UMWAIT_CONTROL,
1361 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1362 MSR_ARCH_PERFMON_FIXED_CTR0 + 2,
1363 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1364 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1365 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1366 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1367 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1368 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1369 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1370 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1371 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1372 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1374 MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1375 MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1376 MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1377 MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1378 MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1379 MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1382 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1383 static unsigned num_msrs_to_save;
1385 static const u32 emulated_msrs_all[] = {
1386 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1387 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1388 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1389 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1390 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1391 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1392 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1394 HV_X64_MSR_VP_INDEX,
1395 HV_X64_MSR_VP_RUNTIME,
1396 HV_X64_MSR_SCONTROL,
1397 HV_X64_MSR_STIMER0_CONFIG,
1398 HV_X64_MSR_VP_ASSIST_PAGE,
1399 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1400 HV_X64_MSR_TSC_EMULATION_STATUS,
1401 HV_X64_MSR_SYNDBG_OPTIONS,
1402 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1403 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1404 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1406 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1407 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1409 MSR_IA32_TSC_ADJUST,
1410 MSR_IA32_TSC_DEADLINE,
1411 MSR_IA32_ARCH_CAPABILITIES,
1412 MSR_IA32_PERF_CAPABILITIES,
1413 MSR_IA32_MISC_ENABLE,
1414 MSR_IA32_MCG_STATUS,
1416 MSR_IA32_MCG_EXT_CTL,
1420 MSR_MISC_FEATURES_ENABLES,
1421 MSR_AMD64_VIRT_SPEC_CTRL,
1426 * The following list leaves out MSRs whose values are determined
1427 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1428 * We always support the "true" VMX control MSRs, even if the host
1429 * processor does not, so I am putting these registers here rather
1430 * than in msrs_to_save_all.
1433 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1434 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1435 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1436 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1438 MSR_IA32_VMX_CR0_FIXED0,
1439 MSR_IA32_VMX_CR4_FIXED0,
1440 MSR_IA32_VMX_VMCS_ENUM,
1441 MSR_IA32_VMX_PROCBASED_CTLS2,
1442 MSR_IA32_VMX_EPT_VPID_CAP,
1443 MSR_IA32_VMX_VMFUNC,
1446 MSR_KVM_POLL_CONTROL,
1449 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1450 static unsigned num_emulated_msrs;
1453 * List of msr numbers which are used to expose MSR-based features that
1454 * can be used by a hypervisor to validate requested CPU features.
1456 static const u32 msr_based_features_all[] = {
1458 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1459 MSR_IA32_VMX_PINBASED_CTLS,
1460 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1461 MSR_IA32_VMX_PROCBASED_CTLS,
1462 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1463 MSR_IA32_VMX_EXIT_CTLS,
1464 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1465 MSR_IA32_VMX_ENTRY_CTLS,
1467 MSR_IA32_VMX_CR0_FIXED0,
1468 MSR_IA32_VMX_CR0_FIXED1,
1469 MSR_IA32_VMX_CR4_FIXED0,
1470 MSR_IA32_VMX_CR4_FIXED1,
1471 MSR_IA32_VMX_VMCS_ENUM,
1472 MSR_IA32_VMX_PROCBASED_CTLS2,
1473 MSR_IA32_VMX_EPT_VPID_CAP,
1474 MSR_IA32_VMX_VMFUNC,
1478 MSR_IA32_ARCH_CAPABILITIES,
1479 MSR_IA32_PERF_CAPABILITIES,
1482 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1483 static unsigned int num_msr_based_features;
1486 * Some IA32_ARCH_CAPABILITIES bits have dependencies on MSRs that KVM
1487 * does not yet virtualize. These include:
1488 * 10 - MISC_PACKAGE_CTRLS
1489 * 11 - ENERGY_FILTERING_CTL
1491 * 18 - FB_CLEAR_CTRL
1492 * 21 - XAPIC_DISABLE_STATUS
1493 * 23 - OVERCLOCKING_STATUS
1496 #define KVM_SUPPORTED_ARCH_CAP \
1497 (ARCH_CAP_RDCL_NO | ARCH_CAP_IBRS_ALL | ARCH_CAP_RSBA | \
1498 ARCH_CAP_SKIP_VMENTRY_L1DFLUSH | ARCH_CAP_SSB_NO | ARCH_CAP_MDS_NO | \
1499 ARCH_CAP_PSCHANGE_MC_NO | ARCH_CAP_TSX_CTRL_MSR | ARCH_CAP_TAA_NO | \
1500 ARCH_CAP_SBDR_SSDP_NO | ARCH_CAP_FBSDP_NO | ARCH_CAP_PSDP_NO | \
1501 ARCH_CAP_FB_CLEAR | ARCH_CAP_RRSBA | ARCH_CAP_PBRSB_NO | ARCH_CAP_GDS_NO)
1503 static u64 kvm_get_arch_capabilities(void)
1507 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
1508 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1509 data &= KVM_SUPPORTED_ARCH_CAP;
1513 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1514 * the nested hypervisor runs with NX huge pages. If it is not,
1515 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1516 * L1 guests, so it need not worry about its own (L2) guests.
1518 data |= ARCH_CAP_PSCHANGE_MC_NO;
1521 * If we're doing cache flushes (either "always" or "cond")
1522 * we will do one whenever the guest does a vmlaunch/vmresume.
1523 * If an outer hypervisor is doing the cache flush for us
1524 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1525 * capability to the guest too, and if EPT is disabled we're not
1526 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1527 * require a nested hypervisor to do a flush of its own.
1529 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1530 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1532 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1533 data |= ARCH_CAP_RDCL_NO;
1534 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1535 data |= ARCH_CAP_SSB_NO;
1536 if (!boot_cpu_has_bug(X86_BUG_MDS))
1537 data |= ARCH_CAP_MDS_NO;
1539 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1541 * If RTM=0 because the kernel has disabled TSX, the host might
1542 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1543 * and therefore knows that there cannot be TAA) but keep
1544 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1545 * and we want to allow migrating those guests to tsx=off hosts.
1547 data &= ~ARCH_CAP_TAA_NO;
1548 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1549 data |= ARCH_CAP_TAA_NO;
1552 * Nothing to do here; we emulate TSX_CTRL if present on the
1553 * host so the guest can choose between disabling TSX or
1554 * using VERW to clear CPU buffers.
1558 if (!boot_cpu_has_bug(X86_BUG_GDS) || gds_ucode_mitigated())
1559 data |= ARCH_CAP_GDS_NO;
1564 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1566 switch (msr->index) {
1567 case MSR_IA32_ARCH_CAPABILITIES:
1568 msr->data = kvm_get_arch_capabilities();
1570 case MSR_IA32_UCODE_REV:
1571 rdmsrl_safe(msr->index, &msr->data);
1574 return static_call(kvm_x86_get_msr_feature)(msr);
1579 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1581 struct kvm_msr_entry msr;
1585 r = kvm_get_msr_feature(&msr);
1587 if (r == KVM_MSR_RET_INVALID) {
1588 /* Unconditionally clear the output for simplicity */
1590 if (kvm_msr_ignored_check(index, 0, false))
1602 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1604 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1607 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1610 if (efer & (EFER_LME | EFER_LMA) &&
1611 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1614 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1620 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1622 if (efer & efer_reserved_bits)
1625 return __kvm_valid_efer(vcpu, efer);
1627 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1629 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1631 u64 old_efer = vcpu->arch.efer;
1632 u64 efer = msr_info->data;
1635 if (efer & efer_reserved_bits)
1638 if (!msr_info->host_initiated) {
1639 if (!__kvm_valid_efer(vcpu, efer))
1642 if (is_paging(vcpu) &&
1643 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1648 efer |= vcpu->arch.efer & EFER_LMA;
1650 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1656 if ((efer ^ old_efer) & KVM_MMU_EFER_ROLE_BITS)
1657 kvm_mmu_reset_context(vcpu);
1662 void kvm_enable_efer_bits(u64 mask)
1664 efer_reserved_bits &= ~mask;
1666 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1668 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1670 struct kvm_x86_msr_filter *msr_filter;
1671 struct msr_bitmap_range *ranges;
1672 struct kvm *kvm = vcpu->kvm;
1677 /* x2APIC MSRs do not support filtering. */
1678 if (index >= 0x800 && index <= 0x8ff)
1681 idx = srcu_read_lock(&kvm->srcu);
1683 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1689 allowed = msr_filter->default_allow;
1690 ranges = msr_filter->ranges;
1692 for (i = 0; i < msr_filter->count; i++) {
1693 u32 start = ranges[i].base;
1694 u32 end = start + ranges[i].nmsrs;
1695 u32 flags = ranges[i].flags;
1696 unsigned long *bitmap = ranges[i].bitmap;
1698 if ((index >= start) && (index < end) && (flags & type)) {
1699 allowed = !!test_bit(index - start, bitmap);
1705 srcu_read_unlock(&kvm->srcu, idx);
1709 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1712 * Write @data into the MSR specified by @index. Select MSR specific fault
1713 * checks are bypassed if @host_initiated is %true.
1714 * Returns 0 on success, non-0 otherwise.
1715 * Assumes vcpu_load() was already called.
1717 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1718 bool host_initiated)
1720 struct msr_data msr;
1722 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1723 return KVM_MSR_RET_FILTERED;
1728 case MSR_KERNEL_GS_BASE:
1731 if (is_noncanonical_address(data, vcpu))
1734 case MSR_IA32_SYSENTER_EIP:
1735 case MSR_IA32_SYSENTER_ESP:
1737 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1738 * non-canonical address is written on Intel but not on
1739 * AMD (which ignores the top 32-bits, because it does
1740 * not implement 64-bit SYSENTER).
1742 * 64-bit code should hence be able to write a non-canonical
1743 * value on AMD. Making the address canonical ensures that
1744 * vmentry does not fail on Intel after writing a non-canonical
1745 * value, and that something deterministic happens if the guest
1746 * invokes 64-bit SYSENTER.
1748 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1751 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1754 if (!host_initiated &&
1755 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1756 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1760 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1761 * incomplete and conflicting architectural behavior. Current
1762 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1763 * reserved and always read as zeros. Enforce Intel's reserved
1764 * bits check if and only if the guest CPU is Intel, and clear
1765 * the bits in all other cases. This ensures cross-vendor
1766 * migration will provide consistent behavior for the guest.
1768 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1777 msr.host_initiated = host_initiated;
1779 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1782 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1783 u32 index, u64 data, bool host_initiated)
1785 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1787 if (ret == KVM_MSR_RET_INVALID)
1788 if (kvm_msr_ignored_check(index, data, true))
1795 * Read the MSR specified by @index into @data. Select MSR specific fault
1796 * checks are bypassed if @host_initiated is %true.
1797 * Returns 0 on success, non-0 otherwise.
1798 * Assumes vcpu_load() was already called.
1800 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1801 bool host_initiated)
1803 struct msr_data msr;
1806 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1807 return KVM_MSR_RET_FILTERED;
1811 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1814 if (!host_initiated &&
1815 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1816 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1822 msr.host_initiated = host_initiated;
1824 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1830 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1831 u32 index, u64 *data, bool host_initiated)
1833 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1835 if (ret == KVM_MSR_RET_INVALID) {
1836 /* Unconditionally clear *data for simplicity */
1838 if (kvm_msr_ignored_check(index, 0, false))
1845 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1847 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1849 EXPORT_SYMBOL_GPL(kvm_get_msr);
1851 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1853 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1855 EXPORT_SYMBOL_GPL(kvm_set_msr);
1857 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1859 int err = vcpu->run->msr.error;
1861 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1862 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1865 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1868 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1870 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1873 static u64 kvm_msr_reason(int r)
1876 case KVM_MSR_RET_INVALID:
1877 return KVM_MSR_EXIT_REASON_UNKNOWN;
1878 case KVM_MSR_RET_FILTERED:
1879 return KVM_MSR_EXIT_REASON_FILTER;
1881 return KVM_MSR_EXIT_REASON_INVAL;
1885 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1886 u32 exit_reason, u64 data,
1887 int (*completion)(struct kvm_vcpu *vcpu),
1890 u64 msr_reason = kvm_msr_reason(r);
1892 /* Check if the user wanted to know about this MSR fault */
1893 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1896 vcpu->run->exit_reason = exit_reason;
1897 vcpu->run->msr.error = 0;
1898 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1899 vcpu->run->msr.reason = msr_reason;
1900 vcpu->run->msr.index = index;
1901 vcpu->run->msr.data = data;
1902 vcpu->arch.complete_userspace_io = completion;
1907 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1909 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1910 complete_emulated_rdmsr, r);
1913 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1915 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1916 complete_emulated_wrmsr, r);
1919 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1921 u32 ecx = kvm_rcx_read(vcpu);
1925 r = kvm_get_msr(vcpu, ecx, &data);
1927 /* MSR read failed? See if we should ask user space */
1928 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1929 /* Bounce to user space */
1934 trace_kvm_msr_read(ecx, data);
1936 kvm_rax_write(vcpu, data & -1u);
1937 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1939 trace_kvm_msr_read_ex(ecx);
1942 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1944 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1946 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1948 u32 ecx = kvm_rcx_read(vcpu);
1949 u64 data = kvm_read_edx_eax(vcpu);
1952 r = kvm_set_msr(vcpu, ecx, data);
1954 /* MSR write failed? See if we should ask user space */
1955 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1956 /* Bounce to user space */
1959 /* Signal all other negative errors to userspace */
1964 trace_kvm_msr_write(ecx, data);
1966 trace_kvm_msr_write_ex(ecx, data);
1968 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1970 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1972 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1974 return kvm_skip_emulated_instruction(vcpu);
1976 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1978 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1980 /* Treat an INVD instruction as a NOP and just skip it. */
1981 return kvm_emulate_as_nop(vcpu);
1983 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1985 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1987 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1988 return kvm_emulate_as_nop(vcpu);
1990 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1992 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1994 kvm_queue_exception(vcpu, UD_VECTOR);
1997 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1999 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
2001 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
2002 return kvm_emulate_as_nop(vcpu);
2004 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
2006 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
2008 xfer_to_guest_mode_prepare();
2009 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
2010 xfer_to_guest_mode_work_pending();
2014 * The fast path for frequent and performance sensitive wrmsr emulation,
2015 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
2016 * the latency of virtual IPI by avoiding the expensive bits of transitioning
2017 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
2018 * other cases which must be called after interrupts are enabled on the host.
2020 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
2022 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
2025 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
2026 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
2027 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
2028 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
2031 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
2032 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
2033 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
2034 trace_kvm_apic_write(APIC_ICR, (u32)data);
2041 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2043 if (!kvm_can_use_hv_timer(vcpu))
2046 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2050 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2052 u32 msr = kvm_rcx_read(vcpu);
2054 fastpath_t ret = EXIT_FASTPATH_NONE;
2057 case APIC_BASE_MSR + (APIC_ICR >> 4):
2058 data = kvm_read_edx_eax(vcpu);
2059 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2060 kvm_skip_emulated_instruction(vcpu);
2061 ret = EXIT_FASTPATH_EXIT_HANDLED;
2064 case MSR_IA32_TSC_DEADLINE:
2065 data = kvm_read_edx_eax(vcpu);
2066 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2067 kvm_skip_emulated_instruction(vcpu);
2068 ret = EXIT_FASTPATH_REENTER_GUEST;
2075 if (ret != EXIT_FASTPATH_NONE)
2076 trace_kvm_msr_write(msr, data);
2080 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2083 * Adapt set_msr() to msr_io()'s calling convention
2085 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2087 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2090 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2092 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2095 #ifdef CONFIG_X86_64
2096 struct pvclock_clock {
2106 struct pvclock_gtod_data {
2109 struct pvclock_clock clock; /* extract of a clocksource struct */
2110 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2116 static struct pvclock_gtod_data pvclock_gtod_data;
2118 static void update_pvclock_gtod(struct timekeeper *tk)
2120 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2122 write_seqcount_begin(&vdata->seq);
2124 /* copy pvclock gtod data */
2125 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2126 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2127 vdata->clock.mask = tk->tkr_mono.mask;
2128 vdata->clock.mult = tk->tkr_mono.mult;
2129 vdata->clock.shift = tk->tkr_mono.shift;
2130 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2131 vdata->clock.offset = tk->tkr_mono.base;
2133 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2134 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2135 vdata->raw_clock.mask = tk->tkr_raw.mask;
2136 vdata->raw_clock.mult = tk->tkr_raw.mult;
2137 vdata->raw_clock.shift = tk->tkr_raw.shift;
2138 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2139 vdata->raw_clock.offset = tk->tkr_raw.base;
2141 vdata->wall_time_sec = tk->xtime_sec;
2143 vdata->offs_boot = tk->offs_boot;
2145 write_seqcount_end(&vdata->seq);
2148 static s64 get_kvmclock_base_ns(void)
2150 /* Count up from boot time, but with the frequency of the raw clock. */
2151 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2154 static s64 get_kvmclock_base_ns(void)
2156 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2157 return ktime_get_boottime_ns();
2161 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2165 struct pvclock_wall_clock wc;
2172 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2177 ++version; /* first time write, random junk */
2181 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2185 * The guest calculates current wall clock time by adding
2186 * system time (updated by kvm_guest_time_update below) to the
2187 * wall clock specified here. We do the reverse here.
2189 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2191 wc.nsec = do_div(wall_nsec, 1000000000);
2192 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2193 wc.version = version;
2195 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2198 wc_sec_hi = wall_nsec >> 32;
2199 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2200 &wc_sec_hi, sizeof(wc_sec_hi));
2204 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2207 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2208 bool old_msr, bool host_initiated)
2210 struct kvm_arch *ka = &vcpu->kvm->arch;
2212 if (vcpu->vcpu_id == 0 && !host_initiated) {
2213 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2214 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2216 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2219 vcpu->arch.time = system_time;
2220 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2222 /* we verify if the enable bit is set... */
2223 vcpu->arch.pv_time_enabled = false;
2224 if (!(system_time & 1))
2227 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2228 &vcpu->arch.pv_time, system_time & ~1ULL,
2229 sizeof(struct pvclock_vcpu_time_info)))
2230 vcpu->arch.pv_time_enabled = true;
2235 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2237 do_shl32_div32(dividend, divisor);
2241 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2242 s8 *pshift, u32 *pmultiplier)
2250 scaled64 = scaled_hz;
2251 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2256 tps32 = (uint32_t)tps64;
2257 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2258 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2266 *pmultiplier = div_frac(scaled64, tps32);
2269 #ifdef CONFIG_X86_64
2270 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2273 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2274 static unsigned long max_tsc_khz;
2276 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2278 u64 v = (u64)khz * (1000000 + ppm);
2283 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2285 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2289 /* Guest TSC same frequency as host TSC? */
2291 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2295 /* TSC scaling supported? */
2296 if (!kvm_has_tsc_control) {
2297 if (user_tsc_khz > tsc_khz) {
2298 vcpu->arch.tsc_catchup = 1;
2299 vcpu->arch.tsc_always_catchup = 1;
2302 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2307 /* TSC scaling required - calculate ratio */
2308 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2309 user_tsc_khz, tsc_khz);
2311 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2312 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2317 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2321 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2323 u32 thresh_lo, thresh_hi;
2324 int use_scaling = 0;
2326 /* tsc_khz can be zero if TSC calibration fails */
2327 if (user_tsc_khz == 0) {
2328 /* set tsc_scaling_ratio to a safe value */
2329 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2333 /* Compute a scale to convert nanoseconds in TSC cycles */
2334 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2335 &vcpu->arch.virtual_tsc_shift,
2336 &vcpu->arch.virtual_tsc_mult);
2337 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2340 * Compute the variation in TSC rate which is acceptable
2341 * within the range of tolerance and decide if the
2342 * rate being applied is within that bounds of the hardware
2343 * rate. If so, no scaling or compensation need be done.
2345 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2346 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2347 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2348 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2351 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2354 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2356 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2357 vcpu->arch.virtual_tsc_mult,
2358 vcpu->arch.virtual_tsc_shift);
2359 tsc += vcpu->arch.this_tsc_write;
2363 static inline int gtod_is_based_on_tsc(int mode)
2365 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2368 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2370 #ifdef CONFIG_X86_64
2372 struct kvm_arch *ka = &vcpu->kvm->arch;
2373 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2375 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2376 atomic_read(&vcpu->kvm->online_vcpus));
2379 * Once the masterclock is enabled, always perform request in
2380 * order to update it.
2382 * In order to enable masterclock, the host clocksource must be TSC
2383 * and the vcpus need to have matched TSCs. When that happens,
2384 * perform request to enable masterclock.
2386 if (ka->use_master_clock ||
2387 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2388 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2390 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2391 atomic_read(&vcpu->kvm->online_vcpus),
2392 ka->use_master_clock, gtod->clock.vclock_mode);
2397 * Multiply tsc by a fixed point number represented by ratio.
2399 * The most significant 64-N bits (mult) of ratio represent the
2400 * integral part of the fixed point number; the remaining N bits
2401 * (frac) represent the fractional part, ie. ratio represents a fixed
2402 * point number (mult + frac * 2^(-N)).
2404 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2406 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2408 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2411 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2415 if (ratio != kvm_default_tsc_scaling_ratio)
2416 _tsc = __scale_tsc(ratio, tsc);
2420 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2422 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2426 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2428 return target_tsc - tsc;
2431 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2433 return vcpu->arch.l1_tsc_offset +
2434 kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2436 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2438 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2442 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2443 nested_offset = l1_offset;
2445 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2446 kvm_tsc_scaling_ratio_frac_bits);
2448 nested_offset += l2_offset;
2449 return nested_offset;
2451 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2453 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2455 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2456 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2457 kvm_tsc_scaling_ratio_frac_bits);
2459 return l1_multiplier;
2461 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2463 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2465 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2466 vcpu->arch.l1_tsc_offset,
2469 vcpu->arch.l1_tsc_offset = l1_offset;
2472 * If we are here because L1 chose not to trap WRMSR to TSC then
2473 * according to the spec this should set L1's TSC (as opposed to
2474 * setting L1's offset for L2).
2476 if (is_guest_mode(vcpu))
2477 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2479 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2480 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2482 vcpu->arch.tsc_offset = l1_offset;
2484 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2487 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2489 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2491 /* Userspace is changing the multiplier while L2 is active */
2492 if (is_guest_mode(vcpu))
2493 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2495 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2497 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2499 if (kvm_has_tsc_control)
2500 static_call(kvm_x86_write_tsc_multiplier)(
2501 vcpu, vcpu->arch.tsc_scaling_ratio);
2504 static inline bool kvm_check_tsc_unstable(void)
2506 #ifdef CONFIG_X86_64
2508 * TSC is marked unstable when we're running on Hyper-V,
2509 * 'TSC page' clocksource is good.
2511 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2514 return check_tsc_unstable();
2517 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2519 struct kvm *kvm = vcpu->kvm;
2520 u64 offset, ns, elapsed;
2521 unsigned long flags;
2523 bool already_matched;
2524 bool synchronizing = false;
2526 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2527 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2528 ns = get_kvmclock_base_ns();
2529 elapsed = ns - kvm->arch.last_tsc_nsec;
2531 if (vcpu->arch.virtual_tsc_khz) {
2534 * detection of vcpu initialization -- need to sync
2535 * with other vCPUs. This particularly helps to keep
2536 * kvm_clock stable after CPU hotplug
2538 synchronizing = true;
2540 u64 tsc_exp = kvm->arch.last_tsc_write +
2541 nsec_to_cycles(vcpu, elapsed);
2542 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2544 * Special case: TSC write with a small delta (1 second)
2545 * of virtual cycle time against real time is
2546 * interpreted as an attempt to synchronize the CPU.
2548 synchronizing = data < tsc_exp + tsc_hz &&
2549 data + tsc_hz > tsc_exp;
2554 * For a reliable TSC, we can match TSC offsets, and for an unstable
2555 * TSC, we add elapsed time in this computation. We could let the
2556 * compensation code attempt to catch up if we fall behind, but
2557 * it's better to try to match offsets from the beginning.
2559 if (synchronizing &&
2560 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2561 if (!kvm_check_tsc_unstable()) {
2562 offset = kvm->arch.cur_tsc_offset;
2564 u64 delta = nsec_to_cycles(vcpu, elapsed);
2566 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2569 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2572 * We split periods of matched TSC writes into generations.
2573 * For each generation, we track the original measured
2574 * nanosecond time, offset, and write, so if TSCs are in
2575 * sync, we can match exact offset, and if not, we can match
2576 * exact software computation in compute_guest_tsc()
2578 * These values are tracked in kvm->arch.cur_xxx variables.
2580 kvm->arch.cur_tsc_generation++;
2581 kvm->arch.cur_tsc_nsec = ns;
2582 kvm->arch.cur_tsc_write = data;
2583 kvm->arch.cur_tsc_offset = offset;
2588 * We also track th most recent recorded KHZ, write and time to
2589 * allow the matching interval to be extended at each write.
2591 kvm->arch.last_tsc_nsec = ns;
2592 kvm->arch.last_tsc_write = data;
2593 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2595 vcpu->arch.last_guest_tsc = data;
2597 /* Keep track of which generation this VCPU has synchronized to */
2598 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2599 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2600 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2602 kvm_vcpu_write_tsc_offset(vcpu, offset);
2603 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2605 raw_spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2607 kvm->arch.nr_vcpus_matched_tsc = 0;
2608 } else if (!already_matched) {
2609 kvm->arch.nr_vcpus_matched_tsc++;
2612 kvm_track_tsc_matching(vcpu);
2613 raw_spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2616 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2619 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2620 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2623 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2625 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2626 WARN_ON(adjustment < 0);
2627 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2628 vcpu->arch.l1_tsc_scaling_ratio);
2629 adjust_tsc_offset_guest(vcpu, adjustment);
2632 #ifdef CONFIG_X86_64
2634 static u64 read_tsc(void)
2636 u64 ret = (u64)rdtsc_ordered();
2637 u64 last = pvclock_gtod_data.clock.cycle_last;
2639 if (likely(ret >= last))
2643 * GCC likes to generate cmov here, but this branch is extremely
2644 * predictable (it's just a function of time and the likely is
2645 * very likely) and there's a data dependence, so force GCC
2646 * to generate a branch instead. I don't barrier() because
2647 * we don't actually need a barrier, and if this function
2648 * ever gets inlined it will generate worse code.
2654 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2660 switch (clock->vclock_mode) {
2661 case VDSO_CLOCKMODE_HVCLOCK:
2662 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2664 if (tsc_pg_val != U64_MAX) {
2665 /* TSC page valid */
2666 *mode = VDSO_CLOCKMODE_HVCLOCK;
2667 v = (tsc_pg_val - clock->cycle_last) &
2670 /* TSC page invalid */
2671 *mode = VDSO_CLOCKMODE_NONE;
2674 case VDSO_CLOCKMODE_TSC:
2675 *mode = VDSO_CLOCKMODE_TSC;
2676 *tsc_timestamp = read_tsc();
2677 v = (*tsc_timestamp - clock->cycle_last) &
2681 *mode = VDSO_CLOCKMODE_NONE;
2684 if (*mode == VDSO_CLOCKMODE_NONE)
2685 *tsc_timestamp = v = 0;
2687 return v * clock->mult;
2690 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2692 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2698 seq = read_seqcount_begin(>od->seq);
2699 ns = gtod->raw_clock.base_cycles;
2700 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2701 ns >>= gtod->raw_clock.shift;
2702 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2703 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2709 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2711 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2717 seq = read_seqcount_begin(>od->seq);
2718 ts->tv_sec = gtod->wall_time_sec;
2719 ns = gtod->clock.base_cycles;
2720 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2721 ns >>= gtod->clock.shift;
2722 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2724 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2730 /* returns true if host is using TSC based clocksource */
2731 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2733 /* checked again under seqlock below */
2734 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2737 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2741 /* returns true if host is using TSC based clocksource */
2742 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2745 /* checked again under seqlock below */
2746 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2749 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2755 * Assuming a stable TSC across physical CPUS, and a stable TSC
2756 * across virtual CPUs, the following condition is possible.
2757 * Each numbered line represents an event visible to both
2758 * CPUs at the next numbered event.
2760 * "timespecX" represents host monotonic time. "tscX" represents
2763 * VCPU0 on CPU0 | VCPU1 on CPU1
2765 * 1. read timespec0,tsc0
2766 * 2. | timespec1 = timespec0 + N
2768 * 3. transition to guest | transition to guest
2769 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2770 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2771 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2773 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2776 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2778 * - 0 < N - M => M < N
2780 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2781 * always the case (the difference between two distinct xtime instances
2782 * might be smaller then the difference between corresponding TSC reads,
2783 * when updating guest vcpus pvclock areas).
2785 * To avoid that problem, do not allow visibility of distinct
2786 * system_timestamp/tsc_timestamp values simultaneously: use a master
2787 * copy of host monotonic time values. Update that master copy
2790 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2794 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2796 #ifdef CONFIG_X86_64
2797 struct kvm_arch *ka = &kvm->arch;
2799 bool host_tsc_clocksource, vcpus_matched;
2801 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2802 atomic_read(&kvm->online_vcpus));
2805 * If the host uses TSC clock, then passthrough TSC as stable
2808 host_tsc_clocksource = kvm_get_time_and_clockread(
2809 &ka->master_kernel_ns,
2810 &ka->master_cycle_now);
2812 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2813 && !ka->backwards_tsc_observed
2814 && !ka->boot_vcpu_runs_old_kvmclock;
2816 if (ka->use_master_clock)
2817 atomic_set(&kvm_guest_has_master_clock, 1);
2819 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2820 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2825 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2827 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2830 static void kvm_gen_update_masterclock(struct kvm *kvm)
2832 #ifdef CONFIG_X86_64
2834 struct kvm_vcpu *vcpu;
2835 struct kvm_arch *ka = &kvm->arch;
2836 unsigned long flags;
2838 kvm_hv_invalidate_tsc_page(kvm);
2840 kvm_make_mclock_inprogress_request(kvm);
2842 /* no guest entries from this point */
2843 raw_spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2844 pvclock_update_vm_gtod_copy(kvm);
2845 raw_spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2847 kvm_for_each_vcpu(i, vcpu, kvm)
2848 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2850 /* guest entries allowed */
2851 kvm_for_each_vcpu(i, vcpu, kvm)
2852 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2856 u64 get_kvmclock_ns(struct kvm *kvm)
2858 struct kvm_arch *ka = &kvm->arch;
2859 struct pvclock_vcpu_time_info hv_clock;
2860 unsigned long flags;
2863 raw_spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2864 if (!ka->use_master_clock) {
2865 raw_spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2866 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2869 hv_clock.tsc_timestamp = ka->master_cycle_now;
2870 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2871 raw_spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2873 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2876 if (__this_cpu_read(cpu_tsc_khz)) {
2877 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2878 &hv_clock.tsc_shift,
2879 &hv_clock.tsc_to_system_mul);
2880 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2882 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2889 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2890 struct gfn_to_hva_cache *cache,
2891 unsigned int offset)
2893 struct kvm_vcpu_arch *vcpu = &v->arch;
2894 struct pvclock_vcpu_time_info guest_hv_clock;
2896 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2897 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2900 /* This VCPU is paused, but it's legal for a guest to read another
2901 * VCPU's kvmclock, so we really have to follow the specification where
2902 * it says that version is odd if data is being modified, and even after
2905 * Version field updates must be kept separate. This is because
2906 * kvm_write_guest_cached might use a "rep movs" instruction, and
2907 * writes within a string instruction are weakly ordered. So there
2908 * are three writes overall.
2910 * As a small optimization, only write the version field in the first
2911 * and third write. The vcpu->pv_time cache is still valid, because the
2912 * version field is the first in the struct.
2914 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2916 if (guest_hv_clock.version & 1)
2917 ++guest_hv_clock.version; /* first time write, random junk */
2919 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2920 kvm_write_guest_offset_cached(v->kvm, cache,
2921 &vcpu->hv_clock, offset,
2922 sizeof(vcpu->hv_clock.version));
2926 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2927 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2929 if (vcpu->pvclock_set_guest_stopped_request) {
2930 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2931 vcpu->pvclock_set_guest_stopped_request = false;
2934 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2936 kvm_write_guest_offset_cached(v->kvm, cache,
2937 &vcpu->hv_clock, offset,
2938 sizeof(vcpu->hv_clock));
2942 vcpu->hv_clock.version++;
2943 kvm_write_guest_offset_cached(v->kvm, cache,
2944 &vcpu->hv_clock, offset,
2945 sizeof(vcpu->hv_clock.version));
2948 static int kvm_guest_time_update(struct kvm_vcpu *v)
2950 unsigned long flags, tgt_tsc_khz;
2951 struct kvm_vcpu_arch *vcpu = &v->arch;
2952 struct kvm_arch *ka = &v->kvm->arch;
2954 u64 tsc_timestamp, host_tsc;
2956 bool use_master_clock;
2962 * If the host uses TSC clock, then passthrough TSC as stable
2965 raw_spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2966 use_master_clock = ka->use_master_clock;
2967 if (use_master_clock) {
2968 host_tsc = ka->master_cycle_now;
2969 kernel_ns = ka->master_kernel_ns;
2971 raw_spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2973 /* Keep irq disabled to prevent changes to the clock */
2974 local_irq_save(flags);
2975 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2976 if (unlikely(tgt_tsc_khz == 0)) {
2977 local_irq_restore(flags);
2978 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2981 if (!use_master_clock) {
2983 kernel_ns = get_kvmclock_base_ns();
2986 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2989 * We may have to catch up the TSC to match elapsed wall clock
2990 * time for two reasons, even if kvmclock is used.
2991 * 1) CPU could have been running below the maximum TSC rate
2992 * 2) Broken TSC compensation resets the base at each VCPU
2993 * entry to avoid unknown leaps of TSC even when running
2994 * again on the same CPU. This may cause apparent elapsed
2995 * time to disappear, and the guest to stand still or run
2998 if (vcpu->tsc_catchup) {
2999 u64 tsc = compute_guest_tsc(v, kernel_ns);
3000 if (tsc > tsc_timestamp) {
3001 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3002 tsc_timestamp = tsc;
3006 local_irq_restore(flags);
3008 /* With all the info we got, fill in the values */
3010 if (kvm_has_tsc_control)
3011 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
3012 v->arch.l1_tsc_scaling_ratio);
3014 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3015 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3016 &vcpu->hv_clock.tsc_shift,
3017 &vcpu->hv_clock.tsc_to_system_mul);
3018 vcpu->hw_tsc_khz = tgt_tsc_khz;
3021 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3022 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3023 vcpu->last_guest_tsc = tsc_timestamp;
3025 /* If the host uses TSC clocksource, then it is stable */
3027 if (use_master_clock)
3028 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3030 vcpu->hv_clock.flags = pvclock_flags;
3032 if (vcpu->pv_time_enabled)
3033 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
3034 if (vcpu->xen.vcpu_info_set)
3035 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
3036 offsetof(struct compat_vcpu_info, time));
3037 if (vcpu->xen.vcpu_time_info_set)
3038 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
3040 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3045 * kvmclock updates which are isolated to a given vcpu, such as
3046 * vcpu->cpu migration, should not allow system_timestamp from
3047 * the rest of the vcpus to remain static. Otherwise ntp frequency
3048 * correction applies to one vcpu's system_timestamp but not
3051 * So in those cases, request a kvmclock update for all vcpus.
3052 * We need to rate-limit these requests though, as they can
3053 * considerably slow guests that have a large number of vcpus.
3054 * The time for a remote vcpu to update its kvmclock is bound
3055 * by the delay we use to rate-limit the updates.
3058 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3060 static void kvmclock_update_fn(struct work_struct *work)
3063 struct delayed_work *dwork = to_delayed_work(work);
3064 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3065 kvmclock_update_work);
3066 struct kvm *kvm = container_of(ka, struct kvm, arch);
3067 struct kvm_vcpu *vcpu;
3069 kvm_for_each_vcpu(i, vcpu, kvm) {
3070 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3071 kvm_vcpu_kick(vcpu);
3075 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3077 struct kvm *kvm = v->kvm;
3079 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3080 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3081 KVMCLOCK_UPDATE_DELAY);
3084 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3086 static void kvmclock_sync_fn(struct work_struct *work)
3088 struct delayed_work *dwork = to_delayed_work(work);
3089 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3090 kvmclock_sync_work);
3091 struct kvm *kvm = container_of(ka, struct kvm, arch);
3093 if (!kvmclock_periodic_sync)
3096 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3097 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3098 KVMCLOCK_SYNC_PERIOD);
3102 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3104 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3106 /* McStatusWrEn enabled? */
3107 if (guest_cpuid_is_amd_or_hygon(vcpu))
3108 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3113 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3115 u64 mcg_cap = vcpu->arch.mcg_cap;
3116 unsigned bank_num = mcg_cap & 0xff;
3117 u32 msr = msr_info->index;
3118 u64 data = msr_info->data;
3121 case MSR_IA32_MCG_STATUS:
3122 vcpu->arch.mcg_status = data;
3124 case MSR_IA32_MCG_CTL:
3125 if (!(mcg_cap & MCG_CTL_P) &&
3126 (data || !msr_info->host_initiated))
3128 if (data != 0 && data != ~(u64)0)
3130 vcpu->arch.mcg_ctl = data;
3133 if (msr >= MSR_IA32_MC0_CTL &&
3134 msr < MSR_IA32_MCx_CTL(bank_num)) {
3135 u32 offset = array_index_nospec(
3136 msr - MSR_IA32_MC0_CTL,
3137 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3139 /* only 0 or all 1s can be written to IA32_MCi_CTL
3140 * some Linux kernels though clear bit 10 in bank 4 to
3141 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3142 * this to avoid an uncatched #GP in the guest.
3144 * UNIXWARE clears bit 0 of MC1_CTL to ignore
3145 * correctable, single-bit ECC data errors.
3147 if ((offset & 0x3) == 0 &&
3148 data != 0 && (data | (1 << 10) | 1) != ~(u64)0)
3152 if (!msr_info->host_initiated &&
3153 (offset & 0x3) == 1 && data != 0) {
3154 if (!can_set_mci_status(vcpu))
3158 vcpu->arch.mce_banks[offset] = data;
3166 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3168 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3170 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3173 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3175 gpa_t gpa = data & ~0x3f;
3177 /* Bits 4:5 are reserved, Should be zero */
3181 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3182 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3185 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3186 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3189 if (!lapic_in_kernel(vcpu))
3190 return data ? 1 : 0;
3192 vcpu->arch.apf.msr_en_val = data;
3194 if (!kvm_pv_async_pf_enabled(vcpu)) {
3195 kvm_clear_async_pf_completion_queue(vcpu);
3196 kvm_async_pf_hash_reset(vcpu);
3200 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3204 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3205 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3207 kvm_async_pf_wakeup_all(vcpu);
3212 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3214 /* Bits 8-63 are reserved */
3218 if (!lapic_in_kernel(vcpu))
3221 vcpu->arch.apf.msr_int_val = data;
3223 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3228 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3230 vcpu->arch.pv_time_enabled = false;
3231 vcpu->arch.time = 0;
3234 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3236 ++vcpu->stat.tlb_flush;
3237 static_call(kvm_x86_tlb_flush_all)(vcpu);
3240 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3242 ++vcpu->stat.tlb_flush;
3246 * A TLB flush on behalf of the guest is equivalent to
3247 * INVPCID(all), toggling CR4.PGE, etc., which requires
3248 * a forced sync of the shadow page tables. Unload the
3249 * entire MMU here and the subsequent load will sync the
3250 * shadow page tables, and also flush the TLB.
3252 kvm_mmu_unload(vcpu);
3256 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3260 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
3262 ++vcpu->stat.tlb_flush;
3263 static_call(kvm_x86_tlb_flush_current)(vcpu);
3267 * Service "local" TLB flush requests, which are specific to the current MMU
3268 * context. In addition to the generic event handling in vcpu_enter_guest(),
3269 * TLB flushes that are targeted at an MMU context also need to be serviced
3270 * prior before nested VM-Enter/VM-Exit.
3272 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu)
3274 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3275 kvm_vcpu_flush_tlb_current(vcpu);
3277 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
3278 kvm_vcpu_flush_tlb_guest(vcpu);
3280 EXPORT_SYMBOL_GPL(kvm_service_local_tlb_flush_requests);
3282 static void record_steal_time(struct kvm_vcpu *vcpu)
3284 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3285 struct kvm_steal_time __user *st;
3286 struct kvm_memslots *slots;
3287 gpa_t gpa = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3291 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3292 kvm_xen_runstate_set_running(vcpu);
3296 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3299 if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3302 slots = kvm_memslots(vcpu->kvm);
3304 if (unlikely(slots->generation != ghc->generation ||
3306 kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3307 /* We rely on the fact that it fits in a single page. */
3308 BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3310 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gpa, sizeof(*st)) ||
3311 kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3315 st = (struct kvm_steal_time __user *)ghc->hva;
3317 * Doing a TLB flush here, on the guest's behalf, can avoid
3320 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3321 u8 st_preempted = 0;
3324 if (!user_access_begin(st, sizeof(*st)))
3327 asm volatile("1: xchgb %0, %2\n"
3330 _ASM_EXTABLE_UA(1b, 2b)
3331 : "+q" (st_preempted),
3333 "+m" (st->preempted));
3339 vcpu->arch.st.preempted = 0;
3341 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3342 st_preempted & KVM_VCPU_FLUSH_TLB);
3343 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3344 kvm_vcpu_flush_tlb_guest(vcpu);
3346 if (!user_access_begin(st, sizeof(*st)))
3349 if (!user_access_begin(st, sizeof(*st)))
3352 unsafe_put_user(0, &st->preempted, out);
3353 vcpu->arch.st.preempted = 0;
3356 unsafe_get_user(version, &st->version, out);
3358 version += 1; /* first time write, random junk */
3361 unsafe_put_user(version, &st->version, out);
3365 unsafe_get_user(steal, &st->steal, out);
3366 steal += current->sched_info.run_delay -
3367 vcpu->arch.st.last_steal;
3368 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3369 unsafe_put_user(steal, &st->steal, out);
3372 unsafe_put_user(version, &st->version, out);
3377 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3380 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3383 u32 msr = msr_info->index;
3384 u64 data = msr_info->data;
3386 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3387 return kvm_xen_write_hypercall_page(vcpu, data);
3390 case MSR_AMD64_NB_CFG:
3391 case MSR_IA32_UCODE_WRITE:
3392 case MSR_VM_HSAVE_PA:
3393 case MSR_AMD64_PATCH_LOADER:
3394 case MSR_AMD64_BU_CFG2:
3395 case MSR_AMD64_DC_CFG:
3396 case MSR_F15H_EX_CFG:
3399 case MSR_IA32_UCODE_REV:
3400 if (msr_info->host_initiated)
3401 vcpu->arch.microcode_version = data;
3403 case MSR_IA32_ARCH_CAPABILITIES:
3404 if (!msr_info->host_initiated)
3406 vcpu->arch.arch_capabilities = data;
3408 case MSR_IA32_PERF_CAPABILITIES: {
3409 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3411 if (!msr_info->host_initiated)
3413 if (kvm_get_msr_feature(&msr_ent))
3415 if (data & ~msr_ent.data)
3418 vcpu->arch.perf_capabilities = data;
3423 return set_efer(vcpu, msr_info);
3425 data &= ~(u64)0x40; /* ignore flush filter disable */
3426 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3427 data &= ~(u64)0x8; /* ignore TLB cache disable */
3429 /* Handle McStatusWrEn */
3430 if (data == BIT_ULL(18)) {
3431 vcpu->arch.msr_hwcr = data;
3432 } else if (data != 0) {
3433 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3438 case MSR_FAM10H_MMIO_CONF_BASE:
3440 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3445 case 0x200 ... 0x2ff:
3446 return kvm_mtrr_set_msr(vcpu, msr, data);
3447 case MSR_IA32_APICBASE:
3448 return kvm_set_apic_base(vcpu, msr_info);
3449 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3450 return kvm_x2apic_msr_write(vcpu, msr, data);
3451 case MSR_IA32_TSC_DEADLINE:
3452 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3454 case MSR_IA32_TSC_ADJUST:
3455 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3456 if (!msr_info->host_initiated) {
3457 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3458 adjust_tsc_offset_guest(vcpu, adj);
3459 /* Before back to guest, tsc_timestamp must be adjusted
3460 * as well, otherwise guest's percpu pvclock time could jump.
3462 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3464 vcpu->arch.ia32_tsc_adjust_msr = data;
3467 case MSR_IA32_MISC_ENABLE:
3468 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3469 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3470 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3472 vcpu->arch.ia32_misc_enable_msr = data;
3473 kvm_update_cpuid_runtime(vcpu);
3475 vcpu->arch.ia32_misc_enable_msr = data;
3478 case MSR_IA32_SMBASE:
3479 if (!msr_info->host_initiated)
3481 vcpu->arch.smbase = data;
3483 case MSR_IA32_POWER_CTL:
3484 vcpu->arch.msr_ia32_power_ctl = data;
3487 if (msr_info->host_initiated) {
3488 kvm_synchronize_tsc(vcpu, data);
3490 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3491 adjust_tsc_offset_guest(vcpu, adj);
3492 vcpu->arch.ia32_tsc_adjust_msr += adj;
3496 if (!msr_info->host_initiated &&
3497 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3500 * KVM supports exposing PT to the guest, but does not support
3501 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3502 * XSAVES/XRSTORS to save/restore PT MSRs.
3504 if (data & ~supported_xss)
3506 vcpu->arch.ia32_xss = data;
3507 kvm_update_cpuid_runtime(vcpu);
3510 if (!msr_info->host_initiated)
3512 vcpu->arch.smi_count = data;
3514 case MSR_KVM_WALL_CLOCK_NEW:
3515 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3518 vcpu->kvm->arch.wall_clock = data;
3519 kvm_write_wall_clock(vcpu->kvm, data, 0);
3521 case MSR_KVM_WALL_CLOCK:
3522 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3525 vcpu->kvm->arch.wall_clock = data;
3526 kvm_write_wall_clock(vcpu->kvm, data, 0);
3528 case MSR_KVM_SYSTEM_TIME_NEW:
3529 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3532 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3534 case MSR_KVM_SYSTEM_TIME:
3535 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3538 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3540 case MSR_KVM_ASYNC_PF_EN:
3541 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3544 if (kvm_pv_enable_async_pf(vcpu, data))
3547 case MSR_KVM_ASYNC_PF_INT:
3548 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3551 if (kvm_pv_enable_async_pf_int(vcpu, data))
3554 case MSR_KVM_ASYNC_PF_ACK:
3555 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3558 vcpu->arch.apf.pageready_pending = false;
3559 kvm_check_async_pf_completion(vcpu);
3562 case MSR_KVM_STEAL_TIME:
3563 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3566 if (unlikely(!sched_info_on()))
3569 if (data & KVM_STEAL_RESERVED_MASK)
3572 vcpu->arch.st.msr_val = data;
3574 if (!(data & KVM_MSR_ENABLED))
3577 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3580 case MSR_KVM_PV_EOI_EN:
3581 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3584 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3588 case MSR_KVM_POLL_CONTROL:
3589 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3592 /* only enable bit supported */
3593 if (data & (-1ULL << 1))
3596 vcpu->arch.msr_kvm_poll_control = data;
3599 case MSR_IA32_MCG_CTL:
3600 case MSR_IA32_MCG_STATUS:
3601 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3602 return set_msr_mce(vcpu, msr_info);
3604 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3605 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3608 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3609 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3610 if (kvm_pmu_is_valid_msr(vcpu, msr))
3611 return kvm_pmu_set_msr(vcpu, msr_info);
3613 if (pr || data != 0)
3614 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3615 "0x%x data 0x%llx\n", msr, data);
3617 case MSR_K7_CLK_CTL:
3619 * Ignore all writes to this no longer documented MSR.
3620 * Writes are only relevant for old K7 processors,
3621 * all pre-dating SVM, but a recommended workaround from
3622 * AMD for these chips. It is possible to specify the
3623 * affected processor models on the command line, hence
3624 * the need to ignore the workaround.
3627 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3628 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3629 case HV_X64_MSR_SYNDBG_OPTIONS:
3630 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3631 case HV_X64_MSR_CRASH_CTL:
3632 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3633 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3634 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3635 case HV_X64_MSR_TSC_EMULATION_STATUS:
3636 return kvm_hv_set_msr_common(vcpu, msr, data,
3637 msr_info->host_initiated);
3638 case MSR_IA32_BBL_CR_CTL3:
3639 /* Drop writes to this legacy MSR -- see rdmsr
3640 * counterpart for further detail.
3642 if (report_ignored_msrs)
3643 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3646 case MSR_AMD64_OSVW_ID_LENGTH:
3647 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3649 vcpu->arch.osvw.length = data;
3651 case MSR_AMD64_OSVW_STATUS:
3652 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3654 vcpu->arch.osvw.status = data;
3656 case MSR_PLATFORM_INFO:
3657 if (!msr_info->host_initiated ||
3658 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3659 cpuid_fault_enabled(vcpu)))
3661 vcpu->arch.msr_platform_info = data;
3663 case MSR_MISC_FEATURES_ENABLES:
3664 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3665 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3666 !supports_cpuid_fault(vcpu)))
3668 vcpu->arch.msr_misc_features_enables = data;
3671 if (kvm_pmu_is_valid_msr(vcpu, msr))
3672 return kvm_pmu_set_msr(vcpu, msr_info);
3673 return KVM_MSR_RET_INVALID;
3677 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3679 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3682 u64 mcg_cap = vcpu->arch.mcg_cap;
3683 unsigned bank_num = mcg_cap & 0xff;
3686 case MSR_IA32_P5_MC_ADDR:
3687 case MSR_IA32_P5_MC_TYPE:
3690 case MSR_IA32_MCG_CAP:
3691 data = vcpu->arch.mcg_cap;
3693 case MSR_IA32_MCG_CTL:
3694 if (!(mcg_cap & MCG_CTL_P) && !host)
3696 data = vcpu->arch.mcg_ctl;
3698 case MSR_IA32_MCG_STATUS:
3699 data = vcpu->arch.mcg_status;
3702 if (msr >= MSR_IA32_MC0_CTL &&
3703 msr < MSR_IA32_MCx_CTL(bank_num)) {
3704 u32 offset = array_index_nospec(
3705 msr - MSR_IA32_MC0_CTL,
3706 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3708 data = vcpu->arch.mce_banks[offset];
3717 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3719 switch (msr_info->index) {
3720 case MSR_IA32_PLATFORM_ID:
3721 case MSR_IA32_EBL_CR_POWERON:
3722 case MSR_IA32_LASTBRANCHFROMIP:
3723 case MSR_IA32_LASTBRANCHTOIP:
3724 case MSR_IA32_LASTINTFROMIP:
3725 case MSR_IA32_LASTINTTOIP:
3726 case MSR_AMD64_SYSCFG:
3727 case MSR_K8_TSEG_ADDR:
3728 case MSR_K8_TSEG_MASK:
3729 case MSR_VM_HSAVE_PA:
3730 case MSR_K8_INT_PENDING_MSG:
3731 case MSR_AMD64_NB_CFG:
3732 case MSR_FAM10H_MMIO_CONF_BASE:
3733 case MSR_AMD64_BU_CFG2:
3734 case MSR_IA32_PERF_CTL:
3735 case MSR_AMD64_DC_CFG:
3736 case MSR_F15H_EX_CFG:
3738 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3739 * limit) MSRs. Just return 0, as we do not want to expose the host
3740 * data here. Do not conditionalize this on CPUID, as KVM does not do
3741 * so for existing CPU-specific MSRs.
3743 case MSR_RAPL_POWER_UNIT:
3744 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3745 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3746 case MSR_PKG_ENERGY_STATUS: /* Total package */
3747 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3750 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3751 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3752 return kvm_pmu_get_msr(vcpu, msr_info);
3753 if (!msr_info->host_initiated)
3757 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3758 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3759 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3760 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3761 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3762 return kvm_pmu_get_msr(vcpu, msr_info);
3765 case MSR_IA32_UCODE_REV:
3766 msr_info->data = vcpu->arch.microcode_version;
3768 case MSR_IA32_ARCH_CAPABILITIES:
3769 if (!msr_info->host_initiated &&
3770 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3772 msr_info->data = vcpu->arch.arch_capabilities;
3774 case MSR_IA32_PERF_CAPABILITIES:
3775 if (!msr_info->host_initiated &&
3776 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3778 msr_info->data = vcpu->arch.perf_capabilities;
3780 case MSR_IA32_POWER_CTL:
3781 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3783 case MSR_IA32_TSC: {
3785 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3786 * even when not intercepted. AMD manual doesn't explicitly
3787 * state this but appears to behave the same.
3789 * On userspace reads and writes, however, we unconditionally
3790 * return L1's TSC value to ensure backwards-compatible
3791 * behavior for migration.
3795 if (msr_info->host_initiated) {
3796 offset = vcpu->arch.l1_tsc_offset;
3797 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3799 offset = vcpu->arch.tsc_offset;
3800 ratio = vcpu->arch.tsc_scaling_ratio;
3803 msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3807 case 0x200 ... 0x2ff:
3808 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3809 case 0xcd: /* fsb frequency */
3813 * MSR_EBC_FREQUENCY_ID
3814 * Conservative value valid for even the basic CPU models.
3815 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3816 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3817 * and 266MHz for model 3, or 4. Set Core Clock
3818 * Frequency to System Bus Frequency Ratio to 1 (bits
3819 * 31:24) even though these are only valid for CPU
3820 * models > 2, however guests may end up dividing or
3821 * multiplying by zero otherwise.
3823 case MSR_EBC_FREQUENCY_ID:
3824 msr_info->data = 1 << 24;
3826 case MSR_IA32_APICBASE:
3827 msr_info->data = kvm_get_apic_base(vcpu);
3829 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3830 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3831 case MSR_IA32_TSC_DEADLINE:
3832 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3834 case MSR_IA32_TSC_ADJUST:
3835 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3837 case MSR_IA32_MISC_ENABLE:
3838 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3840 case MSR_IA32_SMBASE:
3841 if (!msr_info->host_initiated)
3843 msr_info->data = vcpu->arch.smbase;
3846 msr_info->data = vcpu->arch.smi_count;
3848 case MSR_IA32_PERF_STATUS:
3849 /* TSC increment by tick */
3850 msr_info->data = 1000ULL;
3851 /* CPU multiplier */
3852 msr_info->data |= (((uint64_t)4ULL) << 40);
3855 msr_info->data = vcpu->arch.efer;
3857 case MSR_KVM_WALL_CLOCK:
3858 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3861 msr_info->data = vcpu->kvm->arch.wall_clock;
3863 case MSR_KVM_WALL_CLOCK_NEW:
3864 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3867 msr_info->data = vcpu->kvm->arch.wall_clock;
3869 case MSR_KVM_SYSTEM_TIME:
3870 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3873 msr_info->data = vcpu->arch.time;
3875 case MSR_KVM_SYSTEM_TIME_NEW:
3876 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3879 msr_info->data = vcpu->arch.time;
3881 case MSR_KVM_ASYNC_PF_EN:
3882 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3885 msr_info->data = vcpu->arch.apf.msr_en_val;
3887 case MSR_KVM_ASYNC_PF_INT:
3888 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3891 msr_info->data = vcpu->arch.apf.msr_int_val;
3893 case MSR_KVM_ASYNC_PF_ACK:
3894 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3899 case MSR_KVM_STEAL_TIME:
3900 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3903 msr_info->data = vcpu->arch.st.msr_val;
3905 case MSR_KVM_PV_EOI_EN:
3906 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3909 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3911 case MSR_KVM_POLL_CONTROL:
3912 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3915 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3917 case MSR_IA32_P5_MC_ADDR:
3918 case MSR_IA32_P5_MC_TYPE:
3919 case MSR_IA32_MCG_CAP:
3920 case MSR_IA32_MCG_CTL:
3921 case MSR_IA32_MCG_STATUS:
3922 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3923 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3924 msr_info->host_initiated);
3926 if (!msr_info->host_initiated &&
3927 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3929 msr_info->data = vcpu->arch.ia32_xss;
3931 case MSR_K7_CLK_CTL:
3933 * Provide expected ramp-up count for K7. All other
3934 * are set to zero, indicating minimum divisors for
3937 * This prevents guest kernels on AMD host with CPU
3938 * type 6, model 8 and higher from exploding due to
3939 * the rdmsr failing.
3941 msr_info->data = 0x20000000;
3943 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3944 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3945 case HV_X64_MSR_SYNDBG_OPTIONS:
3946 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3947 case HV_X64_MSR_CRASH_CTL:
3948 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3949 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3950 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3951 case HV_X64_MSR_TSC_EMULATION_STATUS:
3952 return kvm_hv_get_msr_common(vcpu,
3953 msr_info->index, &msr_info->data,
3954 msr_info->host_initiated);
3955 case MSR_IA32_BBL_CR_CTL3:
3956 /* This legacy MSR exists but isn't fully documented in current
3957 * silicon. It is however accessed by winxp in very narrow
3958 * scenarios where it sets bit #19, itself documented as
3959 * a "reserved" bit. Best effort attempt to source coherent
3960 * read data here should the balance of the register be
3961 * interpreted by the guest:
3963 * L2 cache control register 3: 64GB range, 256KB size,
3964 * enabled, latency 0x1, configured
3966 msr_info->data = 0xbe702111;
3968 case MSR_AMD64_OSVW_ID_LENGTH:
3969 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3971 msr_info->data = vcpu->arch.osvw.length;
3973 case MSR_AMD64_OSVW_STATUS:
3974 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3976 msr_info->data = vcpu->arch.osvw.status;
3978 case MSR_PLATFORM_INFO:
3979 if (!msr_info->host_initiated &&
3980 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3982 msr_info->data = vcpu->arch.msr_platform_info;
3984 case MSR_MISC_FEATURES_ENABLES:
3985 msr_info->data = vcpu->arch.msr_misc_features_enables;
3988 msr_info->data = vcpu->arch.msr_hwcr;
3991 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3992 return kvm_pmu_get_msr(vcpu, msr_info);
3993 return KVM_MSR_RET_INVALID;
3997 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
4000 * Read or write a bunch of msrs. All parameters are kernel addresses.
4002 * @return number of msrs set successfully.
4004 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
4005 struct kvm_msr_entry *entries,
4006 int (*do_msr)(struct kvm_vcpu *vcpu,
4007 unsigned index, u64 *data))
4011 for (i = 0; i < msrs->nmsrs; ++i)
4012 if (do_msr(vcpu, entries[i].index, &entries[i].data))
4019 * Read or write a bunch of msrs. Parameters are user addresses.
4021 * @return number of msrs set successfully.
4023 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
4024 int (*do_msr)(struct kvm_vcpu *vcpu,
4025 unsigned index, u64 *data),
4028 struct kvm_msrs msrs;
4029 struct kvm_msr_entry *entries;
4034 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4038 if (msrs.nmsrs >= MAX_IO_MSRS)
4041 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4042 entries = memdup_user(user_msrs->entries, size);
4043 if (IS_ERR(entries)) {
4044 r = PTR_ERR(entries);
4048 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
4053 if (writeback && copy_to_user(user_msrs->entries, entries, size))
4064 static inline bool kvm_can_mwait_in_guest(void)
4066 return boot_cpu_has(X86_FEATURE_MWAIT) &&
4067 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
4068 boot_cpu_has(X86_FEATURE_ARAT);
4071 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4072 struct kvm_cpuid2 __user *cpuid_arg)
4074 struct kvm_cpuid2 cpuid;
4078 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4081 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4086 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4092 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4097 case KVM_CAP_IRQCHIP:
4099 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4100 case KVM_CAP_SET_TSS_ADDR:
4101 case KVM_CAP_EXT_CPUID:
4102 case KVM_CAP_EXT_EMUL_CPUID:
4103 case KVM_CAP_CLOCKSOURCE:
4105 case KVM_CAP_NOP_IO_DELAY:
4106 case KVM_CAP_MP_STATE:
4107 case KVM_CAP_SYNC_MMU:
4108 case KVM_CAP_USER_NMI:
4109 case KVM_CAP_REINJECT_CONTROL:
4110 case KVM_CAP_IRQ_INJECT_STATUS:
4111 case KVM_CAP_IOEVENTFD:
4112 case KVM_CAP_IOEVENTFD_NO_LENGTH:
4114 case KVM_CAP_PIT_STATE2:
4115 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4116 case KVM_CAP_VCPU_EVENTS:
4117 case KVM_CAP_HYPERV:
4118 case KVM_CAP_HYPERV_VAPIC:
4119 case KVM_CAP_HYPERV_SPIN:
4120 case KVM_CAP_HYPERV_SYNIC:
4121 case KVM_CAP_HYPERV_SYNIC2:
4122 case KVM_CAP_HYPERV_VP_INDEX:
4123 case KVM_CAP_HYPERV_EVENTFD:
4124 case KVM_CAP_HYPERV_TLBFLUSH:
4125 case KVM_CAP_HYPERV_SEND_IPI:
4126 case KVM_CAP_HYPERV_CPUID:
4127 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4128 case KVM_CAP_SYS_HYPERV_CPUID:
4129 case KVM_CAP_PCI_SEGMENT:
4130 case KVM_CAP_DEBUGREGS:
4131 case KVM_CAP_X86_ROBUST_SINGLESTEP:
4133 case KVM_CAP_ASYNC_PF:
4134 case KVM_CAP_ASYNC_PF_INT:
4135 case KVM_CAP_GET_TSC_KHZ:
4136 case KVM_CAP_KVMCLOCK_CTRL:
4137 case KVM_CAP_READONLY_MEM:
4138 case KVM_CAP_HYPERV_TIME:
4139 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4140 case KVM_CAP_TSC_DEADLINE_TIMER:
4141 case KVM_CAP_DISABLE_QUIRKS:
4142 case KVM_CAP_SET_BOOT_CPU_ID:
4143 case KVM_CAP_SPLIT_IRQCHIP:
4144 case KVM_CAP_IMMEDIATE_EXIT:
4145 case KVM_CAP_PMU_EVENT_FILTER:
4146 case KVM_CAP_GET_MSR_FEATURES:
4147 case KVM_CAP_MSR_PLATFORM_INFO:
4148 case KVM_CAP_EXCEPTION_PAYLOAD:
4149 case KVM_CAP_SET_GUEST_DEBUG:
4150 case KVM_CAP_LAST_CPU:
4151 case KVM_CAP_X86_USER_SPACE_MSR:
4152 case KVM_CAP_X86_MSR_FILTER:
4153 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4154 #ifdef CONFIG_X86_SGX_KVM
4155 case KVM_CAP_SGX_ATTRIBUTE:
4157 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4158 case KVM_CAP_SREGS2:
4159 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4162 case KVM_CAP_EXIT_HYPERCALL:
4163 r = KVM_EXIT_HYPERCALL_VALID_MASK;
4165 case KVM_CAP_SET_GUEST_DEBUG2:
4166 return KVM_GUESTDBG_VALID_MASK;
4167 #ifdef CONFIG_KVM_XEN
4168 case KVM_CAP_XEN_HVM:
4169 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4170 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4171 KVM_XEN_HVM_CONFIG_SHARED_INFO;
4172 if (sched_info_on())
4173 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4176 case KVM_CAP_SYNC_REGS:
4177 r = KVM_SYNC_X86_VALID_FIELDS;
4179 case KVM_CAP_ADJUST_CLOCK:
4180 r = KVM_CLOCK_TSC_STABLE;
4182 case KVM_CAP_X86_DISABLE_EXITS:
4183 r = KVM_X86_DISABLE_EXITS_PAUSE;
4185 if (!mitigate_smt_rsb) {
4186 r |= KVM_X86_DISABLE_EXITS_HLT |
4187 KVM_X86_DISABLE_EXITS_CSTATE;
4189 if (kvm_can_mwait_in_guest())
4190 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4193 case KVM_CAP_X86_SMM:
4194 /* SMBASE is usually relocated above 1M on modern chipsets,
4195 * and SMM handlers might indeed rely on 4G segment limits,
4196 * so do not report SMM to be available if real mode is
4197 * emulated via vm86 mode. Still, do not go to great lengths
4198 * to avoid userspace's usage of the feature, because it is a
4199 * fringe case that is not enabled except via specific settings
4200 * of the module parameters.
4202 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4205 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4207 case KVM_CAP_NR_VCPUS:
4208 r = KVM_SOFT_MAX_VCPUS;
4210 case KVM_CAP_MAX_VCPUS:
4213 case KVM_CAP_MAX_VCPU_ID:
4214 r = KVM_MAX_VCPU_ID;
4216 case KVM_CAP_PV_MMU: /* obsolete */
4220 r = KVM_MAX_MCE_BANKS;
4223 r = boot_cpu_has(X86_FEATURE_XSAVE);
4225 case KVM_CAP_TSC_CONTROL:
4226 r = kvm_has_tsc_control;
4228 case KVM_CAP_X2APIC_API:
4229 r = KVM_X2APIC_API_VALID_FLAGS;
4231 case KVM_CAP_NESTED_STATE:
4232 r = kvm_x86_ops.nested_ops->get_state ?
4233 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4235 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4236 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4238 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4239 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4241 case KVM_CAP_SMALLER_MAXPHYADDR:
4242 r = (int) allow_smaller_maxphyaddr;
4244 case KVM_CAP_STEAL_TIME:
4245 r = sched_info_on();
4247 case KVM_CAP_X86_BUS_LOCK_EXIT:
4248 if (kvm_has_bus_lock_exit)
4249 r = KVM_BUS_LOCK_DETECTION_OFF |
4250 KVM_BUS_LOCK_DETECTION_EXIT;
4261 long kvm_arch_dev_ioctl(struct file *filp,
4262 unsigned int ioctl, unsigned long arg)
4264 void __user *argp = (void __user *)arg;
4268 case KVM_GET_MSR_INDEX_LIST: {
4269 struct kvm_msr_list __user *user_msr_list = argp;
4270 struct kvm_msr_list msr_list;
4274 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4277 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4278 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4281 if (n < msr_list.nmsrs)
4284 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4285 num_msrs_to_save * sizeof(u32)))
4287 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4289 num_emulated_msrs * sizeof(u32)))
4294 case KVM_GET_SUPPORTED_CPUID:
4295 case KVM_GET_EMULATED_CPUID: {
4296 struct kvm_cpuid2 __user *cpuid_arg = argp;
4297 struct kvm_cpuid2 cpuid;
4300 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4303 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4309 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4314 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4316 if (copy_to_user(argp, &kvm_mce_cap_supported,
4317 sizeof(kvm_mce_cap_supported)))
4321 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4322 struct kvm_msr_list __user *user_msr_list = argp;
4323 struct kvm_msr_list msr_list;
4327 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4330 msr_list.nmsrs = num_msr_based_features;
4331 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4334 if (n < msr_list.nmsrs)
4337 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4338 num_msr_based_features * sizeof(u32)))
4344 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4346 case KVM_GET_SUPPORTED_HV_CPUID:
4347 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4357 static void wbinvd_ipi(void *garbage)
4362 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4364 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4367 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4369 /* Address WBINVD may be executed by guest */
4370 if (need_emulate_wbinvd(vcpu)) {
4371 if (static_call(kvm_x86_has_wbinvd_exit)())
4372 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4373 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4374 smp_call_function_single(vcpu->cpu,
4375 wbinvd_ipi, NULL, 1);
4378 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4380 /* Save host pkru register if supported */
4381 vcpu->arch.host_pkru = read_pkru();
4383 /* Apply any externally detected TSC adjustments (due to suspend) */
4384 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4385 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4386 vcpu->arch.tsc_offset_adjustment = 0;
4387 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4390 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4391 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4392 rdtsc() - vcpu->arch.last_host_tsc;
4394 mark_tsc_unstable("KVM discovered backwards TSC");
4396 if (kvm_check_tsc_unstable()) {
4397 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4398 vcpu->arch.last_guest_tsc);
4399 kvm_vcpu_write_tsc_offset(vcpu, offset);
4400 vcpu->arch.tsc_catchup = 1;
4403 if (kvm_lapic_hv_timer_in_use(vcpu))
4404 kvm_lapic_restart_hv_timer(vcpu);
4407 * On a host with synchronized TSC, there is no need to update
4408 * kvmclock on vcpu->cpu migration
4410 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4411 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4412 if (vcpu->cpu != cpu)
4413 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4417 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4420 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4422 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4423 struct kvm_steal_time __user *st;
4424 struct kvm_memslots *slots;
4425 static const u8 preempted = KVM_VCPU_PREEMPTED;
4426 gpa_t gpa = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
4429 * The vCPU can be marked preempted if and only if the VM-Exit was on
4430 * an instruction boundary and will not trigger guest emulation of any
4431 * kind (see vcpu_run). Vendor specific code controls (conservatively)
4432 * when this is true, for example allowing the vCPU to be marked
4433 * preempted if and only if the VM-Exit was due to a host interrupt.
4435 if (!vcpu->arch.at_instruction_boundary) {
4436 vcpu->stat.preemption_other++;
4440 vcpu->stat.preemption_reported++;
4441 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4444 if (vcpu->arch.st.preempted)
4447 /* This happens on process exit */
4448 if (unlikely(current->mm != vcpu->kvm->mm))
4451 slots = kvm_memslots(vcpu->kvm);
4453 if (unlikely(slots->generation != ghc->generation ||
4455 kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4458 st = (struct kvm_steal_time __user *)ghc->hva;
4459 BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4461 if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4462 vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4464 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4467 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4471 if (vcpu->preempted) {
4472 if (!vcpu->arch.guest_state_protected)
4473 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4476 * Take the srcu lock as memslots will be accessed to check the gfn
4477 * cache generation against the memslots generation.
4479 idx = srcu_read_lock(&vcpu->kvm->srcu);
4480 if (kvm_xen_msr_enabled(vcpu->kvm))
4481 kvm_xen_runstate_set_preempted(vcpu);
4483 kvm_steal_time_set_preempted(vcpu);
4484 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4487 static_call(kvm_x86_vcpu_put)(vcpu);
4488 vcpu->arch.last_host_tsc = rdtsc();
4491 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4492 struct kvm_lapic_state *s)
4494 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
4496 return kvm_apic_get_state(vcpu, s);
4499 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4500 struct kvm_lapic_state *s)
4504 r = kvm_apic_set_state(vcpu, s);
4507 update_cr8_intercept(vcpu);
4512 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4515 * We can accept userspace's request for interrupt injection
4516 * as long as we have a place to store the interrupt number.
4517 * The actual injection will happen when the CPU is able to
4518 * deliver the interrupt.
4520 if (kvm_cpu_has_extint(vcpu))
4523 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4524 return (!lapic_in_kernel(vcpu) ||
4525 kvm_apic_accept_pic_intr(vcpu));
4528 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4531 * Do not cause an interrupt window exit if an exception
4532 * is pending or an event needs reinjection; userspace
4533 * might want to inject the interrupt manually using KVM_SET_REGS
4534 * or KVM_SET_SREGS. For that to work, we must be at an
4535 * instruction boundary and with no events half-injected.
4537 return (kvm_arch_interrupt_allowed(vcpu) &&
4538 kvm_cpu_accept_dm_intr(vcpu) &&
4539 !kvm_event_needs_reinjection(vcpu) &&
4540 !vcpu->arch.exception.pending);
4543 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4544 struct kvm_interrupt *irq)
4546 if (irq->irq >= KVM_NR_INTERRUPTS)
4549 if (!irqchip_in_kernel(vcpu->kvm)) {
4550 kvm_queue_interrupt(vcpu, irq->irq, false);
4551 kvm_make_request(KVM_REQ_EVENT, vcpu);
4556 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4557 * fail for in-kernel 8259.
4559 if (pic_in_kernel(vcpu->kvm))
4562 if (vcpu->arch.pending_external_vector != -1)
4565 vcpu->arch.pending_external_vector = irq->irq;
4566 kvm_make_request(KVM_REQ_EVENT, vcpu);
4570 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4572 kvm_inject_nmi(vcpu);
4577 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4579 kvm_make_request(KVM_REQ_SMI, vcpu);
4584 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4585 struct kvm_tpr_access_ctl *tac)
4589 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4593 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4597 unsigned bank_num = mcg_cap & 0xff, bank;
4600 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4602 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4605 vcpu->arch.mcg_cap = mcg_cap;
4606 /* Init IA32_MCG_CTL to all 1s */
4607 if (mcg_cap & MCG_CTL_P)
4608 vcpu->arch.mcg_ctl = ~(u64)0;
4609 /* Init IA32_MCi_CTL to all 1s */
4610 for (bank = 0; bank < bank_num; bank++)
4611 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4613 static_call(kvm_x86_setup_mce)(vcpu);
4618 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4619 struct kvm_x86_mce *mce)
4621 u64 mcg_cap = vcpu->arch.mcg_cap;
4622 unsigned bank_num = mcg_cap & 0xff;
4623 u64 *banks = vcpu->arch.mce_banks;
4625 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4628 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4629 * reporting is disabled
4631 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4632 vcpu->arch.mcg_ctl != ~(u64)0)
4634 banks += 4 * mce->bank;
4636 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4637 * reporting is disabled for the bank
4639 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4641 if (mce->status & MCI_STATUS_UC) {
4642 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4643 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4644 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4647 if (banks[1] & MCI_STATUS_VAL)
4648 mce->status |= MCI_STATUS_OVER;
4649 banks[2] = mce->addr;
4650 banks[3] = mce->misc;
4651 vcpu->arch.mcg_status = mce->mcg_status;
4652 banks[1] = mce->status;
4653 kvm_queue_exception(vcpu, MC_VECTOR);
4654 } else if (!(banks[1] & MCI_STATUS_VAL)
4655 || !(banks[1] & MCI_STATUS_UC)) {
4656 if (banks[1] & MCI_STATUS_VAL)
4657 mce->status |= MCI_STATUS_OVER;
4658 banks[2] = mce->addr;
4659 banks[3] = mce->misc;
4660 banks[1] = mce->status;
4662 banks[1] |= MCI_STATUS_OVER;
4666 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4667 struct kvm_vcpu_events *events)
4671 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4675 * In guest mode, payload delivery should be deferred,
4676 * so that the L1 hypervisor can intercept #PF before
4677 * CR2 is modified (or intercept #DB before DR6 is
4678 * modified under nVMX). Unless the per-VM capability,
4679 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4680 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4681 * opportunistically defer the exception payload, deliver it if the
4682 * capability hasn't been requested before processing a
4683 * KVM_GET_VCPU_EVENTS.
4685 if (!vcpu->kvm->arch.exception_payload_enabled &&
4686 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4687 kvm_deliver_exception_payload(vcpu);
4690 * The API doesn't provide the instruction length for software
4691 * exceptions, so don't report them. As long as the guest RIP
4692 * isn't advanced, we should expect to encounter the exception
4695 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4696 events->exception.injected = 0;
4697 events->exception.pending = 0;
4699 events->exception.injected = vcpu->arch.exception.injected;
4700 events->exception.pending = vcpu->arch.exception.pending;
4702 * For ABI compatibility, deliberately conflate
4703 * pending and injected exceptions when
4704 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4706 if (!vcpu->kvm->arch.exception_payload_enabled)
4707 events->exception.injected |=
4708 vcpu->arch.exception.pending;
4710 events->exception.nr = vcpu->arch.exception.nr;
4711 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4712 events->exception.error_code = vcpu->arch.exception.error_code;
4713 events->exception_has_payload = vcpu->arch.exception.has_payload;
4714 events->exception_payload = vcpu->arch.exception.payload;
4716 events->interrupt.injected =
4717 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4718 events->interrupt.nr = vcpu->arch.interrupt.nr;
4719 events->interrupt.soft = 0;
4720 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4722 events->nmi.injected = vcpu->arch.nmi_injected;
4723 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4724 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4725 events->nmi.pad = 0;
4727 events->sipi_vector = 0; /* never valid when reporting to user space */
4729 events->smi.smm = is_smm(vcpu);
4730 events->smi.pending = vcpu->arch.smi_pending;
4731 events->smi.smm_inside_nmi =
4732 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4733 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4735 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4736 | KVM_VCPUEVENT_VALID_SHADOW
4737 | KVM_VCPUEVENT_VALID_SMM);
4738 if (vcpu->kvm->arch.exception_payload_enabled)
4739 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4741 memset(&events->reserved, 0, sizeof(events->reserved));
4744 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4746 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4747 struct kvm_vcpu_events *events)
4749 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4750 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4751 | KVM_VCPUEVENT_VALID_SHADOW
4752 | KVM_VCPUEVENT_VALID_SMM
4753 | KVM_VCPUEVENT_VALID_PAYLOAD))
4756 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4757 if (!vcpu->kvm->arch.exception_payload_enabled)
4759 if (events->exception.pending)
4760 events->exception.injected = 0;
4762 events->exception_has_payload = 0;
4764 events->exception.pending = 0;
4765 events->exception_has_payload = 0;
4768 if ((events->exception.injected || events->exception.pending) &&
4769 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4772 /* INITs are latched while in SMM */
4773 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4774 (events->smi.smm || events->smi.pending) &&
4775 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4779 vcpu->arch.exception.injected = events->exception.injected;
4780 vcpu->arch.exception.pending = events->exception.pending;
4781 vcpu->arch.exception.nr = events->exception.nr;
4782 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4783 vcpu->arch.exception.error_code = events->exception.error_code;
4784 vcpu->arch.exception.has_payload = events->exception_has_payload;
4785 vcpu->arch.exception.payload = events->exception_payload;
4787 vcpu->arch.interrupt.injected = events->interrupt.injected;
4788 vcpu->arch.interrupt.nr = events->interrupt.nr;
4789 vcpu->arch.interrupt.soft = events->interrupt.soft;
4790 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4791 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4792 events->interrupt.shadow);
4794 vcpu->arch.nmi_injected = events->nmi.injected;
4795 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4796 vcpu->arch.nmi_pending = events->nmi.pending;
4797 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4799 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4800 lapic_in_kernel(vcpu))
4801 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4803 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4804 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4805 kvm_leave_nested(vcpu);
4806 kvm_smm_changed(vcpu, events->smi.smm);
4809 vcpu->arch.smi_pending = events->smi.pending;
4811 if (events->smi.smm) {
4812 if (events->smi.smm_inside_nmi)
4813 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4815 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4818 if (lapic_in_kernel(vcpu)) {
4819 if (events->smi.latched_init)
4820 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4822 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4826 kvm_make_request(KVM_REQ_EVENT, vcpu);
4831 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4832 struct kvm_debugregs *dbgregs)
4836 memset(dbgregs, 0, sizeof(*dbgregs));
4837 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4838 kvm_get_dr(vcpu, 6, &val);
4840 dbgregs->dr7 = vcpu->arch.dr7;
4843 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4844 struct kvm_debugregs *dbgregs)
4849 if (!kvm_dr6_valid(dbgregs->dr6))
4851 if (!kvm_dr7_valid(dbgregs->dr7))
4854 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4855 kvm_update_dr0123(vcpu);
4856 vcpu->arch.dr6 = dbgregs->dr6;
4857 vcpu->arch.dr7 = dbgregs->dr7;
4858 kvm_update_dr7(vcpu);
4863 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4865 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4867 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4868 u64 xstate_bv = xsave->header.xfeatures;
4872 * Copy legacy XSAVE area, to avoid complications with CPUID
4873 * leaves 0 and 1 in the loop below.
4875 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4878 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4879 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4882 * Copy each region from the possibly compacted offset to the
4883 * non-compacted offset.
4885 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4887 u32 size, offset, ecx, edx;
4888 u64 xfeature_mask = valid & -valid;
4889 int xfeature_nr = fls64(xfeature_mask) - 1;
4892 cpuid_count(XSTATE_CPUID, xfeature_nr,
4893 &size, &offset, &ecx, &edx);
4895 if (xfeature_nr == XFEATURE_PKRU) {
4896 memcpy(dest + offset, &vcpu->arch.pkru,
4897 sizeof(vcpu->arch.pkru));
4899 src = get_xsave_addr(xsave, xfeature_nr);
4901 memcpy(dest + offset, src, size);
4904 valid -= xfeature_mask;
4908 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4910 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4911 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4915 * Copy legacy XSAVE area, to avoid complications with CPUID
4916 * leaves 0 and 1 in the loop below.
4918 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4920 /* Set XSTATE_BV and possibly XCOMP_BV. */
4921 xsave->header.xfeatures = xstate_bv;
4922 if (boot_cpu_has(X86_FEATURE_XSAVES))
4923 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4926 * Copy each region from the non-compacted offset to the
4927 * possibly compacted offset.
4929 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4931 u32 size, offset, ecx, edx;
4932 u64 xfeature_mask = valid & -valid;
4933 int xfeature_nr = fls64(xfeature_mask) - 1;
4935 cpuid_count(XSTATE_CPUID, xfeature_nr,
4936 &size, &offset, &ecx, &edx);
4938 if (xfeature_nr == XFEATURE_PKRU) {
4939 memcpy(&vcpu->arch.pkru, src + offset,
4940 sizeof(vcpu->arch.pkru));
4942 void *dest = get_xsave_addr(xsave, xfeature_nr);
4945 memcpy(dest, src + offset, size);
4948 valid -= xfeature_mask;
4952 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4953 struct kvm_xsave *guest_xsave)
4955 if (!vcpu->arch.guest_fpu)
4958 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4959 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4960 fill_xsave((u8 *) guest_xsave->region, vcpu);
4962 memcpy(guest_xsave->region,
4963 &vcpu->arch.guest_fpu->state.fxsave,
4964 sizeof(struct fxregs_state));
4965 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4966 XFEATURE_MASK_FPSSE;
4970 #define XSAVE_MXCSR_OFFSET 24
4972 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4973 struct kvm_xsave *guest_xsave)
4978 if (!vcpu->arch.guest_fpu)
4981 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4982 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4984 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4986 * Here we allow setting states that are not present in
4987 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4988 * with old userspace.
4990 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4992 load_xsave(vcpu, (u8 *)guest_xsave->region);
4994 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4995 mxcsr & ~mxcsr_feature_mask)
4997 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4998 guest_xsave->region, sizeof(struct fxregs_state));
5003 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
5004 struct kvm_xcrs *guest_xcrs)
5006 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
5007 guest_xcrs->nr_xcrs = 0;
5011 guest_xcrs->nr_xcrs = 1;
5012 guest_xcrs->flags = 0;
5013 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
5014 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
5017 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
5018 struct kvm_xcrs *guest_xcrs)
5022 if (!boot_cpu_has(X86_FEATURE_XSAVE))
5025 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
5028 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
5029 /* Only support XCR0 currently */
5030 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
5031 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
5032 guest_xcrs->xcrs[i].value);
5041 * kvm_set_guest_paused() indicates to the guest kernel that it has been
5042 * stopped by the hypervisor. This function will be called from the host only.
5043 * EINVAL is returned when the host attempts to set the flag for a guest that
5044 * does not support pv clocks.
5046 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
5048 if (!vcpu->arch.pv_time_enabled)
5050 vcpu->arch.pvclock_set_guest_stopped_request = true;
5051 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5055 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5056 struct kvm_enable_cap *cap)
5059 uint16_t vmcs_version;
5060 void __user *user_ptr;
5066 case KVM_CAP_HYPERV_SYNIC2:
5071 case KVM_CAP_HYPERV_SYNIC:
5072 if (!irqchip_in_kernel(vcpu->kvm))
5074 return kvm_hv_activate_synic(vcpu, cap->cap ==
5075 KVM_CAP_HYPERV_SYNIC2);
5076 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5077 if (!kvm_x86_ops.nested_ops->enable_evmcs)
5079 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5081 user_ptr = (void __user *)(uintptr_t)cap->args[0];
5082 if (copy_to_user(user_ptr, &vmcs_version,
5083 sizeof(vmcs_version)))
5087 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5088 if (!kvm_x86_ops.enable_direct_tlbflush)
5091 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
5093 case KVM_CAP_HYPERV_ENFORCE_CPUID:
5094 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5096 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5097 vcpu->arch.pv_cpuid.enforce = cap->args[0];
5098 if (vcpu->arch.pv_cpuid.enforce)
5099 kvm_update_pv_runtime(vcpu);
5107 long kvm_arch_vcpu_ioctl(struct file *filp,
5108 unsigned int ioctl, unsigned long arg)
5110 struct kvm_vcpu *vcpu = filp->private_data;
5111 void __user *argp = (void __user *)arg;
5114 struct kvm_sregs2 *sregs2;
5115 struct kvm_lapic_state *lapic;
5116 struct kvm_xsave *xsave;
5117 struct kvm_xcrs *xcrs;
5125 case KVM_GET_LAPIC: {
5127 if (!lapic_in_kernel(vcpu))
5129 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5130 GFP_KERNEL_ACCOUNT);
5135 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5139 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5144 case KVM_SET_LAPIC: {
5146 if (!lapic_in_kernel(vcpu))
5148 u.lapic = memdup_user(argp, sizeof(*u.lapic));
5149 if (IS_ERR(u.lapic)) {
5150 r = PTR_ERR(u.lapic);
5154 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5157 case KVM_INTERRUPT: {
5158 struct kvm_interrupt irq;
5161 if (copy_from_user(&irq, argp, sizeof(irq)))
5163 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5167 r = kvm_vcpu_ioctl_nmi(vcpu);
5171 r = kvm_vcpu_ioctl_smi(vcpu);
5174 case KVM_SET_CPUID: {
5175 struct kvm_cpuid __user *cpuid_arg = argp;
5176 struct kvm_cpuid cpuid;
5179 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5181 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5184 case KVM_SET_CPUID2: {
5185 struct kvm_cpuid2 __user *cpuid_arg = argp;
5186 struct kvm_cpuid2 cpuid;
5189 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5191 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5192 cpuid_arg->entries);
5195 case KVM_GET_CPUID2: {
5196 struct kvm_cpuid2 __user *cpuid_arg = argp;
5197 struct kvm_cpuid2 cpuid;
5200 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5202 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5203 cpuid_arg->entries);
5207 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5212 case KVM_GET_MSRS: {
5213 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5214 r = msr_io(vcpu, argp, do_get_msr, 1);
5215 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5218 case KVM_SET_MSRS: {
5219 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5220 r = msr_io(vcpu, argp, do_set_msr, 0);
5221 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5224 case KVM_TPR_ACCESS_REPORTING: {
5225 struct kvm_tpr_access_ctl tac;
5228 if (copy_from_user(&tac, argp, sizeof(tac)))
5230 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5234 if (copy_to_user(argp, &tac, sizeof(tac)))
5239 case KVM_SET_VAPIC_ADDR: {
5240 struct kvm_vapic_addr va;
5244 if (!lapic_in_kernel(vcpu))
5247 if (copy_from_user(&va, argp, sizeof(va)))
5249 idx = srcu_read_lock(&vcpu->kvm->srcu);
5250 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5251 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5254 case KVM_X86_SETUP_MCE: {
5258 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5260 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5263 case KVM_X86_SET_MCE: {
5264 struct kvm_x86_mce mce;
5267 if (copy_from_user(&mce, argp, sizeof(mce)))
5269 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5272 case KVM_GET_VCPU_EVENTS: {
5273 struct kvm_vcpu_events events;
5275 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5278 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5283 case KVM_SET_VCPU_EVENTS: {
5284 struct kvm_vcpu_events events;
5287 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5290 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5293 case KVM_GET_DEBUGREGS: {
5294 struct kvm_debugregs dbgregs;
5296 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5299 if (copy_to_user(argp, &dbgregs,
5300 sizeof(struct kvm_debugregs)))
5305 case KVM_SET_DEBUGREGS: {
5306 struct kvm_debugregs dbgregs;
5309 if (copy_from_user(&dbgregs, argp,
5310 sizeof(struct kvm_debugregs)))
5313 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5316 case KVM_GET_XSAVE: {
5317 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5322 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5325 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5330 case KVM_SET_XSAVE: {
5331 u.xsave = memdup_user(argp, sizeof(*u.xsave));
5332 if (IS_ERR(u.xsave)) {
5333 r = PTR_ERR(u.xsave);
5337 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5340 case KVM_GET_XCRS: {
5341 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5346 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5349 if (copy_to_user(argp, u.xcrs,
5350 sizeof(struct kvm_xcrs)))
5355 case KVM_SET_XCRS: {
5356 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5357 if (IS_ERR(u.xcrs)) {
5358 r = PTR_ERR(u.xcrs);
5362 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5365 case KVM_SET_TSC_KHZ: {
5369 user_tsc_khz = (u32)arg;
5371 if (kvm_has_tsc_control &&
5372 user_tsc_khz >= kvm_max_guest_tsc_khz)
5375 if (user_tsc_khz == 0)
5376 user_tsc_khz = tsc_khz;
5378 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5383 case KVM_GET_TSC_KHZ: {
5384 r = vcpu->arch.virtual_tsc_khz;
5387 case KVM_KVMCLOCK_CTRL: {
5388 r = kvm_set_guest_paused(vcpu);
5391 case KVM_ENABLE_CAP: {
5392 struct kvm_enable_cap cap;
5395 if (copy_from_user(&cap, argp, sizeof(cap)))
5397 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5400 case KVM_GET_NESTED_STATE: {
5401 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5405 if (!kvm_x86_ops.nested_ops->get_state)
5408 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5410 if (get_user(user_data_size, &user_kvm_nested_state->size))
5413 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5418 if (r > user_data_size) {
5419 if (put_user(r, &user_kvm_nested_state->size))
5429 case KVM_SET_NESTED_STATE: {
5430 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5431 struct kvm_nested_state kvm_state;
5435 if (!kvm_x86_ops.nested_ops->set_state)
5439 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5443 if (kvm_state.size < sizeof(kvm_state))
5446 if (kvm_state.flags &
5447 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5448 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5449 | KVM_STATE_NESTED_GIF_SET))
5452 /* nested_run_pending implies guest_mode. */
5453 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5454 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5457 idx = srcu_read_lock(&vcpu->kvm->srcu);
5458 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5459 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5462 case KVM_GET_SUPPORTED_HV_CPUID:
5463 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5465 #ifdef CONFIG_KVM_XEN
5466 case KVM_XEN_VCPU_GET_ATTR: {
5467 struct kvm_xen_vcpu_attr xva;
5470 if (copy_from_user(&xva, argp, sizeof(xva)))
5472 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5473 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5477 case KVM_XEN_VCPU_SET_ATTR: {
5478 struct kvm_xen_vcpu_attr xva;
5481 if (copy_from_user(&xva, argp, sizeof(xva)))
5483 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5487 case KVM_GET_SREGS2: {
5488 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5492 __get_sregs2(vcpu, u.sregs2);
5494 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5499 case KVM_SET_SREGS2: {
5500 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5501 if (IS_ERR(u.sregs2)) {
5502 r = PTR_ERR(u.sregs2);
5506 r = __set_sregs2(vcpu, u.sregs2);
5519 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5521 return VM_FAULT_SIGBUS;
5524 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5528 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5530 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5534 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5537 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5540 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5541 unsigned long kvm_nr_mmu_pages)
5543 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5546 mutex_lock(&kvm->slots_lock);
5548 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5549 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5551 mutex_unlock(&kvm->slots_lock);
5555 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5557 return kvm->arch.n_max_mmu_pages;
5560 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5562 struct kvm_pic *pic = kvm->arch.vpic;
5566 switch (chip->chip_id) {
5567 case KVM_IRQCHIP_PIC_MASTER:
5568 memcpy(&chip->chip.pic, &pic->pics[0],
5569 sizeof(struct kvm_pic_state));
5571 case KVM_IRQCHIP_PIC_SLAVE:
5572 memcpy(&chip->chip.pic, &pic->pics[1],
5573 sizeof(struct kvm_pic_state));
5575 case KVM_IRQCHIP_IOAPIC:
5576 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5585 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5587 struct kvm_pic *pic = kvm->arch.vpic;
5591 switch (chip->chip_id) {
5592 case KVM_IRQCHIP_PIC_MASTER:
5593 spin_lock(&pic->lock);
5594 memcpy(&pic->pics[0], &chip->chip.pic,
5595 sizeof(struct kvm_pic_state));
5596 spin_unlock(&pic->lock);
5598 case KVM_IRQCHIP_PIC_SLAVE:
5599 spin_lock(&pic->lock);
5600 memcpy(&pic->pics[1], &chip->chip.pic,
5601 sizeof(struct kvm_pic_state));
5602 spin_unlock(&pic->lock);
5604 case KVM_IRQCHIP_IOAPIC:
5605 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5611 kvm_pic_update_irq(pic);
5615 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5617 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5619 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5621 mutex_lock(&kps->lock);
5622 memcpy(ps, &kps->channels, sizeof(*ps));
5623 mutex_unlock(&kps->lock);
5627 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5630 struct kvm_pit *pit = kvm->arch.vpit;
5632 mutex_lock(&pit->pit_state.lock);
5633 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5634 for (i = 0; i < 3; i++)
5635 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5636 mutex_unlock(&pit->pit_state.lock);
5640 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5642 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5643 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5644 sizeof(ps->channels));
5645 ps->flags = kvm->arch.vpit->pit_state.flags;
5646 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5647 memset(&ps->reserved, 0, sizeof(ps->reserved));
5651 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5655 u32 prev_legacy, cur_legacy;
5656 struct kvm_pit *pit = kvm->arch.vpit;
5658 mutex_lock(&pit->pit_state.lock);
5659 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5660 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5661 if (!prev_legacy && cur_legacy)
5663 memcpy(&pit->pit_state.channels, &ps->channels,
5664 sizeof(pit->pit_state.channels));
5665 pit->pit_state.flags = ps->flags;
5666 for (i = 0; i < 3; i++)
5667 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5669 mutex_unlock(&pit->pit_state.lock);
5673 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5674 struct kvm_reinject_control *control)
5676 struct kvm_pit *pit = kvm->arch.vpit;
5678 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5679 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5680 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5682 mutex_lock(&pit->pit_state.lock);
5683 kvm_pit_set_reinject(pit, control->pit_reinject);
5684 mutex_unlock(&pit->pit_state.lock);
5689 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5693 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5694 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5695 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5698 struct kvm_vcpu *vcpu;
5701 kvm_for_each_vcpu(i, vcpu, kvm)
5702 kvm_vcpu_kick(vcpu);
5705 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5708 if (!irqchip_in_kernel(kvm))
5711 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5712 irq_event->irq, irq_event->level,
5717 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5718 struct kvm_enable_cap *cap)
5726 case KVM_CAP_DISABLE_QUIRKS:
5727 kvm->arch.disabled_quirks = cap->args[0];
5730 case KVM_CAP_SPLIT_IRQCHIP: {
5731 mutex_lock(&kvm->lock);
5733 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5734 goto split_irqchip_unlock;
5736 if (irqchip_in_kernel(kvm))
5737 goto split_irqchip_unlock;
5738 if (kvm->created_vcpus)
5739 goto split_irqchip_unlock;
5740 r = kvm_setup_empty_irq_routing(kvm);
5742 goto split_irqchip_unlock;
5743 /* Pairs with irqchip_in_kernel. */
5745 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5746 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5748 split_irqchip_unlock:
5749 mutex_unlock(&kvm->lock);
5752 case KVM_CAP_X2APIC_API:
5754 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5757 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5758 kvm->arch.x2apic_format = true;
5759 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5760 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5764 case KVM_CAP_X86_DISABLE_EXITS:
5766 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5769 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5770 kvm->arch.pause_in_guest = true;
5772 #define SMT_RSB_MSG "This processor is affected by the Cross-Thread Return Predictions vulnerability. " \
5773 "KVM_CAP_X86_DISABLE_EXITS should only be used with SMT disabled or trusted guests."
5775 if (!mitigate_smt_rsb) {
5776 if (boot_cpu_has_bug(X86_BUG_SMT_RSB) && cpu_smt_possible() &&
5777 (cap->args[0] & ~KVM_X86_DISABLE_EXITS_PAUSE))
5778 pr_warn_once(SMT_RSB_MSG);
5780 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5781 kvm_can_mwait_in_guest())
5782 kvm->arch.mwait_in_guest = true;
5783 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5784 kvm->arch.hlt_in_guest = true;
5785 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5786 kvm->arch.cstate_in_guest = true;
5791 case KVM_CAP_MSR_PLATFORM_INFO:
5792 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5795 case KVM_CAP_EXCEPTION_PAYLOAD:
5796 kvm->arch.exception_payload_enabled = cap->args[0];
5799 case KVM_CAP_X86_USER_SPACE_MSR:
5801 if (cap->args[0] & ~(KVM_MSR_EXIT_REASON_INVAL |
5802 KVM_MSR_EXIT_REASON_UNKNOWN |
5803 KVM_MSR_EXIT_REASON_FILTER))
5805 kvm->arch.user_space_msr_mask = cap->args[0];
5808 case KVM_CAP_X86_BUS_LOCK_EXIT:
5810 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5813 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5814 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5817 if (kvm_has_bus_lock_exit &&
5818 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5819 kvm->arch.bus_lock_detection_enabled = true;
5822 #ifdef CONFIG_X86_SGX_KVM
5823 case KVM_CAP_SGX_ATTRIBUTE: {
5824 unsigned long allowed_attributes = 0;
5826 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5830 /* KVM only supports the PROVISIONKEY privileged attribute. */
5831 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5832 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5833 kvm->arch.sgx_provisioning_allowed = true;
5839 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5841 if (kvm_x86_ops.vm_copy_enc_context_from)
5842 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5844 case KVM_CAP_EXIT_HYPERCALL:
5845 if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5849 kvm->arch.hypercall_exit_enabled = cap->args[0];
5852 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5854 if (cap->args[0] & ~1)
5856 kvm->arch.exit_on_emulation_error = cap->args[0];
5866 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5868 struct kvm_x86_msr_filter *msr_filter;
5870 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5874 msr_filter->default_allow = default_allow;
5878 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5885 for (i = 0; i < msr_filter->count; i++)
5886 kfree(msr_filter->ranges[i].bitmap);
5891 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5892 struct kvm_msr_filter_range *user_range)
5894 unsigned long *bitmap = NULL;
5897 if (!user_range->nmsrs)
5900 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5903 if (!user_range->flags)
5906 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5907 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5910 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5912 return PTR_ERR(bitmap);
5914 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5915 .flags = user_range->flags,
5916 .base = user_range->base,
5917 .nmsrs = user_range->nmsrs,
5921 msr_filter->count++;
5925 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm,
5926 struct kvm_msr_filter *filter)
5928 struct kvm_x86_msr_filter *new_filter, *old_filter;
5934 if (filter->flags & ~KVM_MSR_FILTER_DEFAULT_DENY)
5937 for (i = 0; i < ARRAY_SIZE(filter->ranges); i++)
5938 empty &= !filter->ranges[i].nmsrs;
5940 default_allow = !(filter->flags & KVM_MSR_FILTER_DEFAULT_DENY);
5941 if (empty && !default_allow)
5944 new_filter = kvm_alloc_msr_filter(default_allow);
5948 for (i = 0; i < ARRAY_SIZE(filter->ranges); i++) {
5949 r = kvm_add_msr_filter(new_filter, &filter->ranges[i]);
5951 kvm_free_msr_filter(new_filter);
5956 mutex_lock(&kvm->lock);
5958 /* The per-VM filter is protected by kvm->lock... */
5959 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5961 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5962 synchronize_srcu(&kvm->srcu);
5964 kvm_free_msr_filter(old_filter);
5966 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5967 mutex_unlock(&kvm->lock);
5972 #ifdef CONFIG_KVM_COMPAT
5973 /* for KVM_X86_SET_MSR_FILTER */
5974 struct kvm_msr_filter_range_compat {
5981 struct kvm_msr_filter_compat {
5983 struct kvm_msr_filter_range_compat ranges[KVM_MSR_FILTER_MAX_RANGES];
5986 #define KVM_X86_SET_MSR_FILTER_COMPAT _IOW(KVMIO, 0xc6, struct kvm_msr_filter_compat)
5988 long kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
5991 void __user *argp = (void __user *)arg;
5992 struct kvm *kvm = filp->private_data;
5996 case KVM_X86_SET_MSR_FILTER_COMPAT: {
5997 struct kvm_msr_filter __user *user_msr_filter = argp;
5998 struct kvm_msr_filter_compat filter_compat;
5999 struct kvm_msr_filter filter;
6002 if (copy_from_user(&filter_compat, user_msr_filter,
6003 sizeof(filter_compat)))
6006 filter.flags = filter_compat.flags;
6007 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
6008 struct kvm_msr_filter_range_compat *cr;
6010 cr = &filter_compat.ranges[i];
6011 filter.ranges[i] = (struct kvm_msr_filter_range) {
6015 .bitmap = (__u8 *)(ulong)cr->bitmap,
6019 r = kvm_vm_ioctl_set_msr_filter(kvm, &filter);
6028 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
6029 static int kvm_arch_suspend_notifier(struct kvm *kvm)
6031 struct kvm_vcpu *vcpu;
6034 mutex_lock(&kvm->lock);
6035 kvm_for_each_vcpu(i, vcpu, kvm) {
6036 if (!vcpu->arch.pv_time_enabled)
6039 ret = kvm_set_guest_paused(vcpu);
6041 kvm_err("Failed to pause guest VCPU%d: %d\n",
6042 vcpu->vcpu_id, ret);
6046 mutex_unlock(&kvm->lock);
6048 return ret ? NOTIFY_BAD : NOTIFY_DONE;
6051 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
6054 case PM_HIBERNATION_PREPARE:
6055 case PM_SUSPEND_PREPARE:
6056 return kvm_arch_suspend_notifier(kvm);
6061 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
6063 long kvm_arch_vm_ioctl(struct file *filp,
6064 unsigned int ioctl, unsigned long arg)
6066 struct kvm *kvm = filp->private_data;
6067 void __user *argp = (void __user *)arg;
6070 * This union makes it completely explicit to gcc-3.x
6071 * that these two variables' stack usage should be
6072 * combined, not added together.
6075 struct kvm_pit_state ps;
6076 struct kvm_pit_state2 ps2;
6077 struct kvm_pit_config pit_config;
6081 case KVM_SET_TSS_ADDR:
6082 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6084 case KVM_SET_IDENTITY_MAP_ADDR: {
6087 mutex_lock(&kvm->lock);
6089 if (kvm->created_vcpus)
6090 goto set_identity_unlock;
6092 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6093 goto set_identity_unlock;
6094 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6095 set_identity_unlock:
6096 mutex_unlock(&kvm->lock);
6099 case KVM_SET_NR_MMU_PAGES:
6100 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6102 case KVM_GET_NR_MMU_PAGES:
6103 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
6105 case KVM_CREATE_IRQCHIP: {
6106 mutex_lock(&kvm->lock);
6109 if (irqchip_in_kernel(kvm))
6110 goto create_irqchip_unlock;
6113 if (kvm->created_vcpus)
6114 goto create_irqchip_unlock;
6116 r = kvm_pic_init(kvm);
6118 goto create_irqchip_unlock;
6120 r = kvm_ioapic_init(kvm);
6122 kvm_pic_destroy(kvm);
6123 goto create_irqchip_unlock;
6126 r = kvm_setup_default_irq_routing(kvm);
6128 kvm_ioapic_destroy(kvm);
6129 kvm_pic_destroy(kvm);
6130 goto create_irqchip_unlock;
6132 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6134 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6135 create_irqchip_unlock:
6136 mutex_unlock(&kvm->lock);
6139 case KVM_CREATE_PIT:
6140 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6142 case KVM_CREATE_PIT2:
6144 if (copy_from_user(&u.pit_config, argp,
6145 sizeof(struct kvm_pit_config)))
6148 mutex_lock(&kvm->lock);
6151 goto create_pit_unlock;
6153 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6157 mutex_unlock(&kvm->lock);
6159 case KVM_GET_IRQCHIP: {
6160 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6161 struct kvm_irqchip *chip;
6163 chip = memdup_user(argp, sizeof(*chip));
6170 if (!irqchip_kernel(kvm))
6171 goto get_irqchip_out;
6172 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6174 goto get_irqchip_out;
6176 if (copy_to_user(argp, chip, sizeof(*chip)))
6177 goto get_irqchip_out;
6183 case KVM_SET_IRQCHIP: {
6184 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6185 struct kvm_irqchip *chip;
6187 chip = memdup_user(argp, sizeof(*chip));
6194 if (!irqchip_kernel(kvm))
6195 goto set_irqchip_out;
6196 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6203 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6206 if (!kvm->arch.vpit)
6208 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6212 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6219 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6221 mutex_lock(&kvm->lock);
6223 if (!kvm->arch.vpit)
6225 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6227 mutex_unlock(&kvm->lock);
6230 case KVM_GET_PIT2: {
6232 if (!kvm->arch.vpit)
6234 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6238 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6243 case KVM_SET_PIT2: {
6245 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6247 mutex_lock(&kvm->lock);
6249 if (!kvm->arch.vpit)
6251 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6253 mutex_unlock(&kvm->lock);
6256 case KVM_REINJECT_CONTROL: {
6257 struct kvm_reinject_control control;
6259 if (copy_from_user(&control, argp, sizeof(control)))
6262 if (!kvm->arch.vpit)
6264 r = kvm_vm_ioctl_reinject(kvm, &control);
6267 case KVM_SET_BOOT_CPU_ID:
6269 mutex_lock(&kvm->lock);
6270 if (kvm->created_vcpus)
6273 kvm->arch.bsp_vcpu_id = arg;
6274 mutex_unlock(&kvm->lock);
6276 #ifdef CONFIG_KVM_XEN
6277 case KVM_XEN_HVM_CONFIG: {
6278 struct kvm_xen_hvm_config xhc;
6280 if (copy_from_user(&xhc, argp, sizeof(xhc)))
6282 r = kvm_xen_hvm_config(kvm, &xhc);
6285 case KVM_XEN_HVM_GET_ATTR: {
6286 struct kvm_xen_hvm_attr xha;
6289 if (copy_from_user(&xha, argp, sizeof(xha)))
6291 r = kvm_xen_hvm_get_attr(kvm, &xha);
6292 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6296 case KVM_XEN_HVM_SET_ATTR: {
6297 struct kvm_xen_hvm_attr xha;
6300 if (copy_from_user(&xha, argp, sizeof(xha)))
6302 r = kvm_xen_hvm_set_attr(kvm, &xha);
6306 case KVM_SET_CLOCK: {
6307 struct kvm_arch *ka = &kvm->arch;
6308 struct kvm_clock_data user_ns;
6312 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
6321 * TODO: userspace has to take care of races with VCPU_RUN, so
6322 * kvm_gen_update_masterclock() can be cut down to locked
6323 * pvclock_update_vm_gtod_copy().
6325 kvm_gen_update_masterclock(kvm);
6328 * This pairs with kvm_guest_time_update(): when masterclock is
6329 * in use, we use master_kernel_ns + kvmclock_offset to set
6330 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6331 * is slightly ahead) here we risk going negative on unsigned
6332 * 'system_time' when 'user_ns.clock' is very small.
6334 raw_spin_lock_irq(&ka->pvclock_gtod_sync_lock);
6335 if (kvm->arch.use_master_clock)
6336 now_ns = ka->master_kernel_ns;
6338 now_ns = get_kvmclock_base_ns();
6339 ka->kvmclock_offset = user_ns.clock - now_ns;
6340 raw_spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
6342 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
6345 case KVM_GET_CLOCK: {
6346 struct kvm_clock_data user_ns;
6349 now_ns = get_kvmclock_ns(kvm);
6350 user_ns.clock = now_ns;
6351 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
6352 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
6355 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
6360 case KVM_MEMORY_ENCRYPT_OP: {
6362 if (kvm_x86_ops.mem_enc_op)
6363 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6366 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6367 struct kvm_enc_region region;
6370 if (copy_from_user(®ion, argp, sizeof(region)))
6374 if (kvm_x86_ops.mem_enc_reg_region)
6375 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
6378 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6379 struct kvm_enc_region region;
6382 if (copy_from_user(®ion, argp, sizeof(region)))
6386 if (kvm_x86_ops.mem_enc_unreg_region)
6387 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
6390 case KVM_HYPERV_EVENTFD: {
6391 struct kvm_hyperv_eventfd hvevfd;
6394 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6396 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6399 case KVM_SET_PMU_EVENT_FILTER:
6400 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6402 case KVM_X86_SET_MSR_FILTER: {
6403 struct kvm_msr_filter __user *user_msr_filter = argp;
6404 struct kvm_msr_filter filter;
6406 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
6409 r = kvm_vm_ioctl_set_msr_filter(kvm, &filter);
6419 static void kvm_init_msr_list(void)
6421 struct x86_pmu_capability x86_pmu;
6425 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6426 "Please update the fixed PMCs in msrs_to_saved_all[]");
6428 perf_get_x86_pmu_capability(&x86_pmu);
6430 num_msrs_to_save = 0;
6431 num_emulated_msrs = 0;
6432 num_msr_based_features = 0;
6434 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6435 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6439 * Even MSRs that are valid in the host may not be exposed
6440 * to the guests in some cases.
6442 switch (msrs_to_save_all[i]) {
6443 case MSR_IA32_BNDCFGS:
6444 if (!kvm_mpx_supported())
6448 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6449 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6452 case MSR_IA32_UMWAIT_CONTROL:
6453 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6456 case MSR_IA32_RTIT_CTL:
6457 case MSR_IA32_RTIT_STATUS:
6458 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6461 case MSR_IA32_RTIT_CR3_MATCH:
6462 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6463 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6466 case MSR_IA32_RTIT_OUTPUT_BASE:
6467 case MSR_IA32_RTIT_OUTPUT_MASK:
6468 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6469 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6470 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6473 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6474 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6475 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6476 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6479 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 7:
6480 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6481 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6484 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 7:
6485 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6486 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6493 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6496 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6497 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6500 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6503 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6504 struct kvm_msr_entry msr;
6506 msr.index = msr_based_features_all[i];
6507 if (kvm_get_msr_feature(&msr))
6510 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6514 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6522 if (!(lapic_in_kernel(vcpu) &&
6523 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6524 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6535 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6542 if (!(lapic_in_kernel(vcpu) &&
6543 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6545 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6547 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6557 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6558 struct kvm_segment *var, int seg)
6560 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6563 void kvm_get_segment(struct kvm_vcpu *vcpu,
6564 struct kvm_segment *var, int seg)
6566 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6569 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6570 struct x86_exception *exception)
6574 BUG_ON(!mmu_is_nested(vcpu));
6576 /* NPT walks are always user-walks */
6577 access |= PFERR_USER_MASK;
6578 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6583 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6584 struct x86_exception *exception)
6586 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6587 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6589 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6591 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6592 struct x86_exception *exception)
6594 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6595 access |= PFERR_FETCH_MASK;
6596 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6599 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6600 struct x86_exception *exception)
6602 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6603 access |= PFERR_WRITE_MASK;
6604 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6606 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6608 /* uses this to access any guest's mapped memory without checking CPL */
6609 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6610 struct x86_exception *exception)
6612 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6615 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6616 struct kvm_vcpu *vcpu, u32 access,
6617 struct x86_exception *exception)
6620 int r = X86EMUL_CONTINUE;
6623 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6625 unsigned offset = addr & (PAGE_SIZE-1);
6626 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6629 if (gpa == UNMAPPED_GVA)
6630 return X86EMUL_PROPAGATE_FAULT;
6631 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6634 r = X86EMUL_IO_NEEDED;
6646 /* used for instruction fetching */
6647 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6648 gva_t addr, void *val, unsigned int bytes,
6649 struct x86_exception *exception)
6651 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6652 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6656 /* Inline kvm_read_guest_virt_helper for speed. */
6657 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6659 if (unlikely(gpa == UNMAPPED_GVA))
6660 return X86EMUL_PROPAGATE_FAULT;
6662 offset = addr & (PAGE_SIZE-1);
6663 if (WARN_ON(offset + bytes > PAGE_SIZE))
6664 bytes = (unsigned)PAGE_SIZE - offset;
6665 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6667 if (unlikely(ret < 0))
6668 return X86EMUL_IO_NEEDED;
6670 return X86EMUL_CONTINUE;
6673 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6674 gva_t addr, void *val, unsigned int bytes,
6675 struct x86_exception *exception)
6677 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6680 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6681 * is returned, but our callers are not ready for that and they blindly
6682 * call kvm_inject_page_fault. Ensure that they at least do not leak
6683 * uninitialized kernel stack memory into cr2 and error code.
6685 memset(exception, 0, sizeof(*exception));
6686 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6689 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6691 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6692 gva_t addr, void *val, unsigned int bytes,
6693 struct x86_exception *exception, bool system)
6695 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6698 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6699 access |= PFERR_USER_MASK;
6701 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6704 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6705 unsigned long addr, void *val, unsigned int bytes)
6707 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6708 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6710 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6713 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6714 struct kvm_vcpu *vcpu, u32 access,
6715 struct x86_exception *exception)
6718 int r = X86EMUL_CONTINUE;
6721 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6724 unsigned offset = addr & (PAGE_SIZE-1);
6725 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6728 if (gpa == UNMAPPED_GVA)
6729 return X86EMUL_PROPAGATE_FAULT;
6730 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6732 r = X86EMUL_IO_NEEDED;
6744 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6745 unsigned int bytes, struct x86_exception *exception,
6748 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6749 u32 access = PFERR_WRITE_MASK;
6751 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6752 access |= PFERR_USER_MASK;
6754 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6758 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6759 unsigned int bytes, struct x86_exception *exception)
6761 /* kvm_write_guest_virt_system can pull in tons of pages. */
6762 vcpu->arch.l1tf_flush_l1d = true;
6764 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6765 PFERR_WRITE_MASK, exception);
6767 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6769 int handle_ud(struct kvm_vcpu *vcpu)
6771 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6772 int emul_type = EMULTYPE_TRAP_UD;
6773 char sig[5]; /* ud2; .ascii "kvm" */
6774 struct x86_exception e;
6776 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6779 if (force_emulation_prefix &&
6780 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6781 sig, sizeof(sig), &e) == 0 &&
6782 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6783 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6784 emul_type = EMULTYPE_TRAP_UD_FORCED;
6787 return kvm_emulate_instruction(vcpu, emul_type);
6789 EXPORT_SYMBOL_GPL(handle_ud);
6791 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6792 gpa_t gpa, bool write)
6794 /* For APIC access vmexit */
6795 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6798 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6799 trace_vcpu_match_mmio(gva, gpa, write, true);
6806 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6807 gpa_t *gpa, struct x86_exception *exception,
6810 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6811 | (write ? PFERR_WRITE_MASK : 0);
6814 * currently PKRU is only applied to ept enabled guest so
6815 * there is no pkey in EPT page table for L1 guest or EPT
6816 * shadow page table for L2 guest.
6818 if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6819 !permission_fault(vcpu, vcpu->arch.walk_mmu,
6820 vcpu->arch.mmio_access, 0, access))) {
6821 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6822 (gva & (PAGE_SIZE - 1));
6823 trace_vcpu_match_mmio(gva, *gpa, write, false);
6827 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6829 if (*gpa == UNMAPPED_GVA)
6832 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6835 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6836 const void *val, int bytes)
6840 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6843 kvm_page_track_write(vcpu, gpa, val, bytes);
6847 struct read_write_emulator_ops {
6848 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6850 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6851 void *val, int bytes);
6852 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6853 int bytes, void *val);
6854 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6855 void *val, int bytes);
6859 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6861 if (vcpu->mmio_read_completed) {
6862 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6863 vcpu->mmio_fragments[0].gpa, val);
6864 vcpu->mmio_read_completed = 0;
6871 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6872 void *val, int bytes)
6874 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6877 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6878 void *val, int bytes)
6880 return emulator_write_phys(vcpu, gpa, val, bytes);
6883 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6885 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6886 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6889 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6890 void *val, int bytes)
6892 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6893 return X86EMUL_IO_NEEDED;
6896 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6897 void *val, int bytes)
6899 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6901 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6902 return X86EMUL_CONTINUE;
6905 static const struct read_write_emulator_ops read_emultor = {
6906 .read_write_prepare = read_prepare,
6907 .read_write_emulate = read_emulate,
6908 .read_write_mmio = vcpu_mmio_read,
6909 .read_write_exit_mmio = read_exit_mmio,
6912 static const struct read_write_emulator_ops write_emultor = {
6913 .read_write_emulate = write_emulate,
6914 .read_write_mmio = write_mmio,
6915 .read_write_exit_mmio = write_exit_mmio,
6919 static int emulator_read_write_onepage(unsigned long addr, void *val,
6921 struct x86_exception *exception,
6922 struct kvm_vcpu *vcpu,
6923 const struct read_write_emulator_ops *ops)
6927 bool write = ops->write;
6928 struct kvm_mmio_fragment *frag;
6929 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6932 * If the exit was due to a NPF we may already have a GPA.
6933 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6934 * Note, this cannot be used on string operations since string
6935 * operation using rep will only have the initial GPA from the NPF
6938 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6939 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6940 gpa = ctxt->gpa_val;
6941 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6943 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6945 return X86EMUL_PROPAGATE_FAULT;
6948 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6949 return X86EMUL_CONTINUE;
6952 * Is this MMIO handled locally?
6954 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6955 if (handled == bytes)
6956 return X86EMUL_CONTINUE;
6962 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6963 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6967 return X86EMUL_CONTINUE;
6970 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6972 void *val, unsigned int bytes,
6973 struct x86_exception *exception,
6974 const struct read_write_emulator_ops *ops)
6976 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6980 if (ops->read_write_prepare &&
6981 ops->read_write_prepare(vcpu, val, bytes))
6982 return X86EMUL_CONTINUE;
6984 vcpu->mmio_nr_fragments = 0;
6986 /* Crossing a page boundary? */
6987 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6990 now = -addr & ~PAGE_MASK;
6991 rc = emulator_read_write_onepage(addr, val, now, exception,
6994 if (rc != X86EMUL_CONTINUE)
6997 if (ctxt->mode != X86EMUL_MODE_PROT64)
7003 rc = emulator_read_write_onepage(addr, val, bytes, exception,
7005 if (rc != X86EMUL_CONTINUE)
7008 if (!vcpu->mmio_nr_fragments)
7011 gpa = vcpu->mmio_fragments[0].gpa;
7013 vcpu->mmio_needed = 1;
7014 vcpu->mmio_cur_fragment = 0;
7016 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
7017 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
7018 vcpu->run->exit_reason = KVM_EXIT_MMIO;
7019 vcpu->run->mmio.phys_addr = gpa;
7021 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
7024 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
7028 struct x86_exception *exception)
7030 return emulator_read_write(ctxt, addr, val, bytes,
7031 exception, &read_emultor);
7034 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
7038 struct x86_exception *exception)
7040 return emulator_read_write(ctxt, addr, (void *)val, bytes,
7041 exception, &write_emultor);
7044 #define emulator_try_cmpxchg_user(t, ptr, old, new) \
7045 (__try_cmpxchg_user((t __user *)(ptr), (t *)(old), *(t *)(new), efault ## t))
7047 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
7052 struct x86_exception *exception)
7054 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7060 /* guests cmpxchg8b have to be emulated atomically */
7061 if (bytes > 8 || (bytes & (bytes - 1)))
7064 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
7066 if (gpa == UNMAPPED_GVA ||
7067 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7071 * Emulate the atomic as a straight write to avoid #AC if SLD is
7072 * enabled in the host and the access splits a cache line.
7074 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
7075 page_line_mask = ~(cache_line_size() - 1);
7077 page_line_mask = PAGE_MASK;
7079 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
7082 hva = kvm_vcpu_gfn_to_hva(vcpu, gpa_to_gfn(gpa));
7083 if (kvm_is_error_hva(hva))
7086 hva += offset_in_page(gpa);
7090 r = emulator_try_cmpxchg_user(u8, hva, old, new);
7093 r = emulator_try_cmpxchg_user(u16, hva, old, new);
7096 r = emulator_try_cmpxchg_user(u32, hva, old, new);
7099 r = emulator_try_cmpxchg_user(u64, hva, old, new);
7108 return X86EMUL_CMPXCHG_FAILED;
7110 kvm_page_track_write(vcpu, gpa, new, bytes);
7112 return X86EMUL_CONTINUE;
7115 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
7117 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7120 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
7124 for (i = 0; i < vcpu->arch.pio.count; i++) {
7125 if (vcpu->arch.pio.in)
7126 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
7127 vcpu->arch.pio.size, pd);
7129 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
7130 vcpu->arch.pio.port, vcpu->arch.pio.size,
7134 pd += vcpu->arch.pio.size;
7139 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7140 unsigned short port,
7141 unsigned int count, bool in)
7143 vcpu->arch.pio.port = port;
7144 vcpu->arch.pio.in = in;
7145 vcpu->arch.pio.count = count;
7146 vcpu->arch.pio.size = size;
7148 if (!kernel_pio(vcpu, vcpu->arch.pio_data))
7151 vcpu->run->exit_reason = KVM_EXIT_IO;
7152 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7153 vcpu->run->io.size = size;
7154 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7155 vcpu->run->io.count = count;
7156 vcpu->run->io.port = port;
7161 static int __emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7162 unsigned short port, unsigned int count)
7164 WARN_ON(vcpu->arch.pio.count);
7165 memset(vcpu->arch.pio_data, 0, size * count);
7166 return emulator_pio_in_out(vcpu, size, port, count, true);
7169 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7171 int size = vcpu->arch.pio.size;
7172 unsigned count = vcpu->arch.pio.count;
7173 memcpy(val, vcpu->arch.pio_data, size * count);
7174 trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7175 vcpu->arch.pio.count = 0;
7178 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7179 unsigned short port, void *val, unsigned int count)
7181 if (vcpu->arch.pio.count) {
7183 * Complete a previous iteration that required userspace I/O.
7184 * Note, @count isn't guaranteed to match pio.count as userspace
7185 * can modify ECX before rerunning the vCPU. Ignore any such
7186 * shenanigans as KVM doesn't support modifying the rep count,
7187 * and the emulator ensures @count doesn't overflow the buffer.
7190 int r = __emulator_pio_in(vcpu, size, port, count);
7194 /* Results already available, fall through. */
7197 complete_emulator_pio_in(vcpu, val);
7201 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7202 int size, unsigned short port, void *val,
7205 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
7209 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7210 unsigned short port, const void *val,
7215 memcpy(vcpu->arch.pio_data, val, size * count);
7216 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
7217 ret = emulator_pio_in_out(vcpu, size, port, count, false);
7219 vcpu->arch.pio.count = 0;
7224 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7225 int size, unsigned short port,
7226 const void *val, unsigned int count)
7228 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7231 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7233 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7236 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7238 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7241 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7243 if (!need_emulate_wbinvd(vcpu))
7244 return X86EMUL_CONTINUE;
7246 if (static_call(kvm_x86_has_wbinvd_exit)()) {
7247 int cpu = get_cpu();
7249 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7250 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7251 wbinvd_ipi, NULL, 1);
7253 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7256 return X86EMUL_CONTINUE;
7259 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7261 kvm_emulate_wbinvd_noskip(vcpu);
7262 return kvm_skip_emulated_instruction(vcpu);
7264 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7268 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7270 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7273 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7274 unsigned long *dest)
7276 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7279 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7280 unsigned long value)
7283 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7286 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7288 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7291 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7293 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7294 unsigned long value;
7298 value = kvm_read_cr0(vcpu);
7301 value = vcpu->arch.cr2;
7304 value = kvm_read_cr3(vcpu);
7307 value = kvm_read_cr4(vcpu);
7310 value = kvm_get_cr8(vcpu);
7313 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7320 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7322 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7327 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7330 vcpu->arch.cr2 = val;
7333 res = kvm_set_cr3(vcpu, val);
7336 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7339 res = kvm_set_cr8(vcpu, val);
7342 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7349 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7351 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7354 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7356 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7359 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7361 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7364 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7366 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7369 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7371 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7374 static unsigned long emulator_get_cached_segment_base(
7375 struct x86_emulate_ctxt *ctxt, int seg)
7377 return get_segment_base(emul_to_vcpu(ctxt), seg);
7380 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7381 struct desc_struct *desc, u32 *base3,
7384 struct kvm_segment var;
7386 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7387 *selector = var.selector;
7390 memset(desc, 0, sizeof(*desc));
7398 set_desc_limit(desc, var.limit);
7399 set_desc_base(desc, (unsigned long)var.base);
7400 #ifdef CONFIG_X86_64
7402 *base3 = var.base >> 32;
7404 desc->type = var.type;
7406 desc->dpl = var.dpl;
7407 desc->p = var.present;
7408 desc->avl = var.avl;
7416 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7417 struct desc_struct *desc, u32 base3,
7420 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7421 struct kvm_segment var;
7423 var.selector = selector;
7424 var.base = get_desc_base(desc);
7425 #ifdef CONFIG_X86_64
7426 var.base |= ((u64)base3) << 32;
7428 var.limit = get_desc_limit(desc);
7430 var.limit = (var.limit << 12) | 0xfff;
7431 var.type = desc->type;
7432 var.dpl = desc->dpl;
7437 var.avl = desc->avl;
7438 var.present = desc->p;
7439 var.unusable = !var.present;
7442 kvm_set_segment(vcpu, &var, seg);
7446 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7447 u32 msr_index, u64 *pdata)
7449 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7452 r = kvm_get_msr(vcpu, msr_index, pdata);
7454 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7455 /* Bounce to user space */
7456 return X86EMUL_IO_NEEDED;
7462 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7463 u32 msr_index, u64 data)
7465 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7468 r = kvm_set_msr(vcpu, msr_index, data);
7470 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7471 /* Bounce to user space */
7472 return X86EMUL_IO_NEEDED;
7478 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7480 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7482 return vcpu->arch.smbase;
7485 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7487 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7489 vcpu->arch.smbase = smbase;
7492 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7495 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7498 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7499 u32 pmc, u64 *pdata)
7501 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7504 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7506 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7509 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7510 struct x86_instruction_info *info,
7511 enum x86_intercept_stage stage)
7513 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7517 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7518 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7521 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7524 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7526 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7529 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7531 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7534 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7536 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7539 static bool emulator_guest_has_rdpid(struct x86_emulate_ctxt *ctxt)
7541 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_RDPID);
7544 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7546 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7549 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7551 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7554 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7556 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7559 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7561 return emul_to_vcpu(ctxt)->arch.hflags;
7564 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7566 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7568 kvm_smm_changed(vcpu, false);
7571 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7572 const char *smstate)
7574 return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7577 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7579 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7582 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7584 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7587 static const struct x86_emulate_ops emulate_ops = {
7588 .read_gpr = emulator_read_gpr,
7589 .write_gpr = emulator_write_gpr,
7590 .read_std = emulator_read_std,
7591 .write_std = emulator_write_std,
7592 .read_phys = kvm_read_guest_phys_system,
7593 .fetch = kvm_fetch_guest_virt,
7594 .read_emulated = emulator_read_emulated,
7595 .write_emulated = emulator_write_emulated,
7596 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7597 .invlpg = emulator_invlpg,
7598 .pio_in_emulated = emulator_pio_in_emulated,
7599 .pio_out_emulated = emulator_pio_out_emulated,
7600 .get_segment = emulator_get_segment,
7601 .set_segment = emulator_set_segment,
7602 .get_cached_segment_base = emulator_get_cached_segment_base,
7603 .get_gdt = emulator_get_gdt,
7604 .get_idt = emulator_get_idt,
7605 .set_gdt = emulator_set_gdt,
7606 .set_idt = emulator_set_idt,
7607 .get_cr = emulator_get_cr,
7608 .set_cr = emulator_set_cr,
7609 .cpl = emulator_get_cpl,
7610 .get_dr = emulator_get_dr,
7611 .set_dr = emulator_set_dr,
7612 .get_smbase = emulator_get_smbase,
7613 .set_smbase = emulator_set_smbase,
7614 .set_msr = emulator_set_msr,
7615 .get_msr = emulator_get_msr,
7616 .check_pmc = emulator_check_pmc,
7617 .read_pmc = emulator_read_pmc,
7618 .halt = emulator_halt,
7619 .wbinvd = emulator_wbinvd,
7620 .fix_hypercall = emulator_fix_hypercall,
7621 .intercept = emulator_intercept,
7622 .get_cpuid = emulator_get_cpuid,
7623 .guest_has_long_mode = emulator_guest_has_long_mode,
7624 .guest_has_movbe = emulator_guest_has_movbe,
7625 .guest_has_fxsr = emulator_guest_has_fxsr,
7626 .guest_has_rdpid = emulator_guest_has_rdpid,
7627 .set_nmi_mask = emulator_set_nmi_mask,
7628 .get_hflags = emulator_get_hflags,
7629 .exiting_smm = emulator_exiting_smm,
7630 .leave_smm = emulator_leave_smm,
7631 .triple_fault = emulator_triple_fault,
7632 .set_xcr = emulator_set_xcr,
7635 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7637 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7639 * an sti; sti; sequence only disable interrupts for the first
7640 * instruction. So, if the last instruction, be it emulated or
7641 * not, left the system with the INT_STI flag enabled, it
7642 * means that the last instruction is an sti. We should not
7643 * leave the flag on in this case. The same goes for mov ss
7645 if (int_shadow & mask)
7647 if (unlikely(int_shadow || mask)) {
7648 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7650 kvm_make_request(KVM_REQ_EVENT, vcpu);
7654 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7656 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7657 if (ctxt->exception.vector == PF_VECTOR)
7658 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7660 if (ctxt->exception.error_code_valid)
7661 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7662 ctxt->exception.error_code);
7664 kvm_queue_exception(vcpu, ctxt->exception.vector);
7668 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7670 struct x86_emulate_ctxt *ctxt;
7672 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7674 pr_err("kvm: failed to allocate vcpu's emulator\n");
7679 ctxt->ops = &emulate_ops;
7680 vcpu->arch.emulate_ctxt = ctxt;
7685 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7687 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7690 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7692 ctxt->gpa_available = false;
7693 ctxt->eflags = kvm_get_rflags(vcpu);
7694 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7696 ctxt->eip = kvm_rip_read(vcpu);
7697 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7698 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7699 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7700 cs_db ? X86EMUL_MODE_PROT32 :
7701 X86EMUL_MODE_PROT16;
7702 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7703 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7704 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7706 ctxt->interruptibility = 0;
7707 ctxt->have_exception = false;
7708 ctxt->exception.vector = -1;
7709 ctxt->perm_ok = false;
7711 init_decode_cache(ctxt);
7712 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7715 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7717 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7720 init_emulate_ctxt(vcpu);
7724 ctxt->_eip = ctxt->eip + inc_eip;
7725 ret = emulate_int_real(ctxt, irq);
7727 if (ret != X86EMUL_CONTINUE) {
7728 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7730 ctxt->eip = ctxt->_eip;
7731 kvm_rip_write(vcpu, ctxt->eip);
7732 kvm_set_rflags(vcpu, ctxt->eflags);
7735 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7737 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7739 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7740 u32 insn_size = ctxt->fetch.end - ctxt->fetch.data;
7741 struct kvm_run *run = vcpu->run;
7743 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7744 run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7745 run->emulation_failure.ndata = 0;
7746 run->emulation_failure.flags = 0;
7749 run->emulation_failure.ndata = 3;
7750 run->emulation_failure.flags |=
7751 KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7752 run->emulation_failure.insn_size = insn_size;
7753 memset(run->emulation_failure.insn_bytes, 0x90,
7754 sizeof(run->emulation_failure.insn_bytes));
7755 memcpy(run->emulation_failure.insn_bytes,
7756 ctxt->fetch.data, insn_size);
7760 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7762 struct kvm *kvm = vcpu->kvm;
7764 ++vcpu->stat.insn_emulation_fail;
7765 trace_kvm_emulate_insn_failed(vcpu);
7767 if (emulation_type & EMULTYPE_VMWARE_GP) {
7768 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7772 if (kvm->arch.exit_on_emulation_error ||
7773 (emulation_type & EMULTYPE_SKIP)) {
7774 prepare_emulation_failure_exit(vcpu);
7778 kvm_queue_exception(vcpu, UD_VECTOR);
7780 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7781 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7782 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7783 vcpu->run->internal.ndata = 0;
7790 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7791 bool write_fault_to_shadow_pgtable,
7794 gpa_t gpa = cr2_or_gpa;
7797 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7800 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7801 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7804 if (!vcpu->arch.mmu->direct_map) {
7806 * Write permission should be allowed since only
7807 * write access need to be emulated.
7809 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7812 * If the mapping is invalid in guest, let cpu retry
7813 * it to generate fault.
7815 if (gpa == UNMAPPED_GVA)
7820 * Do not retry the unhandleable instruction if it faults on the
7821 * readonly host memory, otherwise it will goto a infinite loop:
7822 * retry instruction -> write #PF -> emulation fail -> retry
7823 * instruction -> ...
7825 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7828 * If the instruction failed on the error pfn, it can not be fixed,
7829 * report the error to userspace.
7831 if (is_error_noslot_pfn(pfn))
7834 kvm_release_pfn_clean(pfn);
7836 /* The instructions are well-emulated on direct mmu. */
7837 if (vcpu->arch.mmu->direct_map) {
7838 unsigned int indirect_shadow_pages;
7840 write_lock(&vcpu->kvm->mmu_lock);
7841 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7842 write_unlock(&vcpu->kvm->mmu_lock);
7844 if (indirect_shadow_pages)
7845 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7851 * if emulation was due to access to shadowed page table
7852 * and it failed try to unshadow page and re-enter the
7853 * guest to let CPU execute the instruction.
7855 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7858 * If the access faults on its page table, it can not
7859 * be fixed by unprotecting shadow page and it should
7860 * be reported to userspace.
7862 return !write_fault_to_shadow_pgtable;
7865 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7866 gpa_t cr2_or_gpa, int emulation_type)
7868 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7869 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7871 last_retry_eip = vcpu->arch.last_retry_eip;
7872 last_retry_addr = vcpu->arch.last_retry_addr;
7875 * If the emulation is caused by #PF and it is non-page_table
7876 * writing instruction, it means the VM-EXIT is caused by shadow
7877 * page protected, we can zap the shadow page and retry this
7878 * instruction directly.
7880 * Note: if the guest uses a non-page-table modifying instruction
7881 * on the PDE that points to the instruction, then we will unmap
7882 * the instruction and go to an infinite loop. So, we cache the
7883 * last retried eip and the last fault address, if we meet the eip
7884 * and the address again, we can break out of the potential infinite
7887 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7889 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7892 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7893 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7896 if (x86_page_table_writing_insn(ctxt))
7899 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7902 vcpu->arch.last_retry_eip = ctxt->eip;
7903 vcpu->arch.last_retry_addr = cr2_or_gpa;
7905 if (!vcpu->arch.mmu->direct_map)
7906 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7908 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7913 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7914 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7916 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7918 trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7921 vcpu->arch.hflags |= HF_SMM_MASK;
7923 vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7925 /* Process a latched INIT or SMI, if any. */
7926 kvm_make_request(KVM_REQ_EVENT, vcpu);
7929 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
7930 * on SMM exit we still need to reload them from
7933 vcpu->arch.pdptrs_from_userspace = false;
7936 kvm_mmu_reset_context(vcpu);
7939 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7948 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7949 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7954 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7956 struct kvm_run *kvm_run = vcpu->run;
7958 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7959 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7960 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7961 kvm_run->debug.arch.exception = DB_VECTOR;
7962 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7965 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7969 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7971 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7974 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7979 * rflags is the old, "raw" value of the flags. The new value has
7980 * not been saved yet.
7982 * This is correct even for TF set by the guest, because "the
7983 * processor will not generate this exception after the instruction
7984 * that sets the TF flag".
7986 if (unlikely(rflags & X86_EFLAGS_TF))
7987 r = kvm_vcpu_do_singlestep(vcpu);
7990 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7992 static bool kvm_vcpu_check_code_breakpoint(struct kvm_vcpu *vcpu, int *r)
7994 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7995 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7996 struct kvm_run *kvm_run = vcpu->run;
7997 unsigned long eip = kvm_get_linear_rip(vcpu);
7998 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7999 vcpu->arch.guest_debug_dr7,
8003 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
8004 kvm_run->debug.arch.pc = eip;
8005 kvm_run->debug.arch.exception = DB_VECTOR;
8006 kvm_run->exit_reason = KVM_EXIT_DEBUG;
8012 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
8013 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
8014 unsigned long eip = kvm_get_linear_rip(vcpu);
8015 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8020 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
8029 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
8031 switch (ctxt->opcode_len) {
8038 case 0xe6: /* OUT */
8042 case 0x6c: /* INS */
8044 case 0x6e: /* OUTS */
8051 case 0x33: /* RDPMC */
8061 * Decode an instruction for emulation. The caller is responsible for handling
8062 * code breakpoints. Note, manually detecting code breakpoints is unnecessary
8063 * (and wrong) when emulating on an intercepted fault-like exception[*], as
8064 * code breakpoints have higher priority and thus have already been done by
8067 * [*] Except #MC, which is higher priority, but KVM should never emulate in
8068 * response to a machine check.
8070 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8071 void *insn, int insn_len)
8073 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8076 init_emulate_ctxt(vcpu);
8078 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8080 trace_kvm_emulate_insn_start(vcpu);
8081 ++vcpu->stat.insn_emulation;
8085 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8087 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8088 int emulation_type, void *insn, int insn_len)
8091 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8092 bool writeback = true;
8093 bool write_fault_to_spt;
8095 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
8098 vcpu->arch.l1tf_flush_l1d = true;
8101 * Clear write_fault_to_shadow_pgtable here to ensure it is
8104 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
8105 vcpu->arch.write_fault_to_shadow_pgtable = false;
8107 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8108 kvm_clear_exception_queue(vcpu);
8111 * Return immediately if RIP hits a code breakpoint, such #DBs
8112 * are fault-like and are higher priority than any faults on
8113 * the code fetch itself.
8115 if (!(emulation_type & EMULTYPE_SKIP) &&
8116 kvm_vcpu_check_code_breakpoint(vcpu, &r))
8119 r = x86_decode_emulated_instruction(vcpu, emulation_type,
8121 if (r != EMULATION_OK) {
8122 if ((emulation_type & EMULTYPE_TRAP_UD) ||
8123 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8124 kvm_queue_exception(vcpu, UD_VECTOR);
8127 if (reexecute_instruction(vcpu, cr2_or_gpa,
8132 if (ctxt->have_exception &&
8133 !(emulation_type & EMULTYPE_SKIP)) {
8135 * #UD should result in just EMULATION_FAILED, and trap-like
8136 * exception should not be encountered during decode.
8138 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8139 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8140 inject_emulated_exception(vcpu);
8143 return handle_emulation_failure(vcpu, emulation_type);
8147 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8148 !is_vmware_backdoor_opcode(ctxt)) {
8149 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8154 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
8155 * for kvm_skip_emulated_instruction(). The caller is responsible for
8156 * updating interruptibility state and injecting single-step #DBs.
8158 if (emulation_type & EMULTYPE_SKIP) {
8159 if (ctxt->mode != X86EMUL_MODE_PROT64)
8160 ctxt->eip = (u32)ctxt->_eip;
8162 ctxt->eip = ctxt->_eip;
8164 kvm_rip_write(vcpu, ctxt->eip);
8165 if (ctxt->eflags & X86_EFLAGS_RF)
8166 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8170 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8173 /* this is needed for vmware backdoor interface to work since it
8174 changes registers values during IO operation */
8175 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8176 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8177 emulator_invalidate_register_cache(ctxt);
8181 if (emulation_type & EMULTYPE_PF) {
8182 /* Save the faulting GPA (cr2) in the address field */
8183 ctxt->exception.address = cr2_or_gpa;
8185 /* With shadow page tables, cr2 contains a GVA or nGPA. */
8186 if (vcpu->arch.mmu->direct_map) {
8187 ctxt->gpa_available = true;
8188 ctxt->gpa_val = cr2_or_gpa;
8191 /* Sanitize the address out of an abundance of paranoia. */
8192 ctxt->exception.address = 0;
8195 r = x86_emulate_insn(ctxt);
8197 if (r == EMULATION_INTERCEPTED)
8200 if (r == EMULATION_FAILED) {
8201 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8205 return handle_emulation_failure(vcpu, emulation_type);
8208 if (ctxt->have_exception) {
8210 if (inject_emulated_exception(vcpu))
8212 } else if (vcpu->arch.pio.count) {
8213 if (!vcpu->arch.pio.in) {
8214 /* FIXME: return into emulator if single-stepping. */
8215 vcpu->arch.pio.count = 0;
8218 vcpu->arch.complete_userspace_io = complete_emulated_pio;
8221 } else if (vcpu->mmio_needed) {
8222 ++vcpu->stat.mmio_exits;
8224 if (!vcpu->mmio_is_write)
8227 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8228 } else if (vcpu->arch.complete_userspace_io) {
8231 } else if (r == EMULATION_RESTART)
8237 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8238 toggle_interruptibility(vcpu, ctxt->interruptibility);
8239 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8242 * Note, EXCPT_DB is assumed to be fault-like as the emulator
8243 * only supports code breakpoints and general detect #DB, both
8244 * of which are fault-like.
8246 if (!ctxt->have_exception ||
8247 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8248 kvm_rip_write(vcpu, ctxt->eip);
8249 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8250 r = kvm_vcpu_do_singlestep(vcpu);
8251 if (kvm_x86_ops.update_emulated_instruction)
8252 static_call(kvm_x86_update_emulated_instruction)(vcpu);
8253 __kvm_set_rflags(vcpu, ctxt->eflags);
8257 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8258 * do nothing, and it will be requested again as soon as
8259 * the shadow expires. But we still need to check here,
8260 * because POPF has no interrupt shadow.
8262 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8263 kvm_make_request(KVM_REQ_EVENT, vcpu);
8265 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8270 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8272 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8274 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8276 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8277 void *insn, int insn_len)
8279 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8281 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8283 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8285 vcpu->arch.pio.count = 0;
8289 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8291 vcpu->arch.pio.count = 0;
8293 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8296 return kvm_skip_emulated_instruction(vcpu);
8299 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8300 unsigned short port)
8302 unsigned long val = kvm_rax_read(vcpu);
8303 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8309 * Workaround userspace that relies on old KVM behavior of %rip being
8310 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8313 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8314 vcpu->arch.complete_userspace_io =
8315 complete_fast_pio_out_port_0x7e;
8316 kvm_skip_emulated_instruction(vcpu);
8318 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8319 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8324 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8328 /* We should only ever be called with arch.pio.count equal to 1 */
8329 BUG_ON(vcpu->arch.pio.count != 1);
8331 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8332 vcpu->arch.pio.count = 0;
8336 /* For size less than 4 we merge, else we zero extend */
8337 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8340 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8341 * the copy and tracing
8343 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8344 kvm_rax_write(vcpu, val);
8346 return kvm_skip_emulated_instruction(vcpu);
8349 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8350 unsigned short port)
8355 /* For size less than 4 we merge, else we zero extend */
8356 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8358 ret = emulator_pio_in(vcpu, size, port, &val, 1);
8360 kvm_rax_write(vcpu, val);
8364 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8365 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8370 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8375 ret = kvm_fast_pio_in(vcpu, size, port);
8377 ret = kvm_fast_pio_out(vcpu, size, port);
8378 return ret && kvm_skip_emulated_instruction(vcpu);
8380 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8382 static int kvmclock_cpu_down_prep(unsigned int cpu)
8384 __this_cpu_write(cpu_tsc_khz, 0);
8388 static void tsc_khz_changed(void *data)
8390 struct cpufreq_freqs *freq = data;
8391 unsigned long khz = 0;
8395 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8396 khz = cpufreq_quick_get(raw_smp_processor_id());
8399 __this_cpu_write(cpu_tsc_khz, khz);
8402 #ifdef CONFIG_X86_64
8403 static void kvm_hyperv_tsc_notifier(void)
8406 struct kvm_vcpu *vcpu;
8408 unsigned long flags;
8410 mutex_lock(&kvm_lock);
8411 list_for_each_entry(kvm, &vm_list, vm_list)
8412 kvm_make_mclock_inprogress_request(kvm);
8414 hyperv_stop_tsc_emulation();
8416 /* TSC frequency always matches when on Hyper-V */
8417 for_each_present_cpu(cpu)
8418 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8419 kvm_max_guest_tsc_khz = tsc_khz;
8421 list_for_each_entry(kvm, &vm_list, vm_list) {
8422 struct kvm_arch *ka = &kvm->arch;
8424 raw_spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
8425 pvclock_update_vm_gtod_copy(kvm);
8426 raw_spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
8428 kvm_for_each_vcpu(cpu, vcpu, kvm)
8429 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8431 kvm_for_each_vcpu(cpu, vcpu, kvm)
8432 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
8434 mutex_unlock(&kvm_lock);
8438 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8441 struct kvm_vcpu *vcpu;
8442 int i, send_ipi = 0;
8445 * We allow guests to temporarily run on slowing clocks,
8446 * provided we notify them after, or to run on accelerating
8447 * clocks, provided we notify them before. Thus time never
8450 * However, we have a problem. We can't atomically update
8451 * the frequency of a given CPU from this function; it is
8452 * merely a notifier, which can be called from any CPU.
8453 * Changing the TSC frequency at arbitrary points in time
8454 * requires a recomputation of local variables related to
8455 * the TSC for each VCPU. We must flag these local variables
8456 * to be updated and be sure the update takes place with the
8457 * new frequency before any guests proceed.
8459 * Unfortunately, the combination of hotplug CPU and frequency
8460 * change creates an intractable locking scenario; the order
8461 * of when these callouts happen is undefined with respect to
8462 * CPU hotplug, and they can race with each other. As such,
8463 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8464 * undefined; you can actually have a CPU frequency change take
8465 * place in between the computation of X and the setting of the
8466 * variable. To protect against this problem, all updates of
8467 * the per_cpu tsc_khz variable are done in an interrupt
8468 * protected IPI, and all callers wishing to update the value
8469 * must wait for a synchronous IPI to complete (which is trivial
8470 * if the caller is on the CPU already). This establishes the
8471 * necessary total order on variable updates.
8473 * Note that because a guest time update may take place
8474 * anytime after the setting of the VCPU's request bit, the
8475 * correct TSC value must be set before the request. However,
8476 * to ensure the update actually makes it to any guest which
8477 * starts running in hardware virtualization between the set
8478 * and the acquisition of the spinlock, we must also ping the
8479 * CPU after setting the request bit.
8483 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8485 mutex_lock(&kvm_lock);
8486 list_for_each_entry(kvm, &vm_list, vm_list) {
8487 kvm_for_each_vcpu(i, vcpu, kvm) {
8488 if (vcpu->cpu != cpu)
8490 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8491 if (vcpu->cpu != raw_smp_processor_id())
8495 mutex_unlock(&kvm_lock);
8497 if (freq->old < freq->new && send_ipi) {
8499 * We upscale the frequency. Must make the guest
8500 * doesn't see old kvmclock values while running with
8501 * the new frequency, otherwise we risk the guest sees
8502 * time go backwards.
8504 * In case we update the frequency for another cpu
8505 * (which might be in guest context) send an interrupt
8506 * to kick the cpu out of guest context. Next time
8507 * guest context is entered kvmclock will be updated,
8508 * so the guest will not see stale values.
8510 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8514 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8517 struct cpufreq_freqs *freq = data;
8520 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8522 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8525 for_each_cpu(cpu, freq->policy->cpus)
8526 __kvmclock_cpufreq_notifier(freq, cpu);
8531 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8532 .notifier_call = kvmclock_cpufreq_notifier
8535 static int kvmclock_cpu_online(unsigned int cpu)
8537 tsc_khz_changed(NULL);
8541 static void kvm_timer_init(void)
8543 max_tsc_khz = tsc_khz;
8545 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8546 #ifdef CONFIG_CPU_FREQ
8547 struct cpufreq_policy *policy;
8551 policy = cpufreq_cpu_get(cpu);
8553 if (policy->cpuinfo.max_freq)
8554 max_tsc_khz = policy->cpuinfo.max_freq;
8555 cpufreq_cpu_put(policy);
8559 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8560 CPUFREQ_TRANSITION_NOTIFIER);
8563 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8564 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8567 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8568 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8570 int kvm_is_in_guest(void)
8572 return __this_cpu_read(current_vcpu) != NULL;
8575 static int kvm_is_user_mode(void)
8579 if (__this_cpu_read(current_vcpu))
8580 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8582 return user_mode != 0;
8585 static unsigned long kvm_get_guest_ip(void)
8587 unsigned long ip = 0;
8589 if (__this_cpu_read(current_vcpu))
8590 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8595 static void kvm_handle_intel_pt_intr(void)
8597 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8599 kvm_make_request(KVM_REQ_PMI, vcpu);
8600 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8601 (unsigned long *)&vcpu->arch.pmu.global_status);
8604 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8605 .is_in_guest = kvm_is_in_guest,
8606 .is_user_mode = kvm_is_user_mode,
8607 .get_guest_ip = kvm_get_guest_ip,
8608 .handle_intel_pt_intr = NULL,
8611 #ifdef CONFIG_X86_64
8612 static void pvclock_gtod_update_fn(struct work_struct *work)
8616 struct kvm_vcpu *vcpu;
8619 mutex_lock(&kvm_lock);
8620 list_for_each_entry(kvm, &vm_list, vm_list)
8621 kvm_for_each_vcpu(i, vcpu, kvm)
8622 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8623 atomic_set(&kvm_guest_has_master_clock, 0);
8624 mutex_unlock(&kvm_lock);
8627 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8630 * Indirection to move queue_work() out of the tk_core.seq write held
8631 * region to prevent possible deadlocks against time accessors which
8632 * are invoked with work related locks held.
8634 static void pvclock_irq_work_fn(struct irq_work *w)
8636 queue_work(system_long_wq, &pvclock_gtod_work);
8639 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8642 * Notification about pvclock gtod data update.
8644 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8647 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8648 struct timekeeper *tk = priv;
8650 update_pvclock_gtod(tk);
8653 * Disable master clock if host does not trust, or does not use,
8654 * TSC based clocksource. Delegate queue_work() to irq_work as
8655 * this is invoked with tk_core.seq write held.
8657 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8658 atomic_read(&kvm_guest_has_master_clock) != 0)
8659 irq_work_queue(&pvclock_irq_work);
8663 static struct notifier_block pvclock_gtod_notifier = {
8664 .notifier_call = pvclock_gtod_notify,
8668 int kvm_arch_init(void *opaque)
8670 struct kvm_x86_init_ops *ops = opaque;
8673 if (kvm_x86_ops.hardware_enable) {
8674 printk(KERN_ERR "kvm: already loaded the other module\n");
8679 if (!ops->cpu_has_kvm_support()) {
8680 pr_err_ratelimited("kvm: no hardware support\n");
8684 if (ops->disabled_by_bios()) {
8685 pr_err_ratelimited("kvm: disabled by bios\n");
8691 * KVM explicitly assumes that the guest has an FPU and
8692 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8693 * vCPU's FPU state as a fxregs_state struct.
8695 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8696 printk(KERN_ERR "kvm: inadequate fpu\n");
8702 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8703 __alignof__(struct fpu), SLAB_ACCOUNT,
8705 if (!x86_fpu_cache) {
8706 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8710 x86_emulator_cache = kvm_alloc_emulator_cache();
8711 if (!x86_emulator_cache) {
8712 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8713 goto out_free_x86_fpu_cache;
8716 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8717 if (!user_return_msrs) {
8718 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8719 goto out_free_x86_emulator_cache;
8721 kvm_nr_uret_msrs = 0;
8723 r = kvm_mmu_vendor_module_init();
8725 goto out_free_percpu;
8729 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8730 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8731 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8734 if (pi_inject_timer == -1)
8735 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8736 #ifdef CONFIG_X86_64
8737 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8739 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8740 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8746 free_percpu(user_return_msrs);
8747 out_free_x86_emulator_cache:
8748 kmem_cache_destroy(x86_emulator_cache);
8749 out_free_x86_fpu_cache:
8750 kmem_cache_destroy(x86_fpu_cache);
8755 void kvm_arch_exit(void)
8757 #ifdef CONFIG_X86_64
8758 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8759 clear_hv_tscchange_cb();
8763 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8764 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8765 CPUFREQ_TRANSITION_NOTIFIER);
8766 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8767 #ifdef CONFIG_X86_64
8768 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8769 irq_work_sync(&pvclock_irq_work);
8770 cancel_work_sync(&pvclock_gtod_work);
8772 kvm_x86_ops.hardware_enable = NULL;
8773 kvm_mmu_vendor_module_exit();
8774 free_percpu(user_return_msrs);
8775 kmem_cache_destroy(x86_emulator_cache);
8776 kmem_cache_destroy(x86_fpu_cache);
8777 #ifdef CONFIG_KVM_XEN
8778 static_key_deferred_flush(&kvm_xen_enabled);
8779 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8783 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8785 ++vcpu->stat.halt_exits;
8786 if (lapic_in_kernel(vcpu)) {
8787 vcpu->arch.mp_state = state;
8790 vcpu->run->exit_reason = reason;
8795 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8797 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8799 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8801 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8803 int ret = kvm_skip_emulated_instruction(vcpu);
8805 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8806 * KVM_EXIT_DEBUG here.
8808 return kvm_vcpu_halt(vcpu) && ret;
8810 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8812 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8814 int ret = kvm_skip_emulated_instruction(vcpu);
8816 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8818 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8820 #ifdef CONFIG_X86_64
8821 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8822 unsigned long clock_type)
8824 struct kvm_clock_pairing clock_pairing;
8825 struct timespec64 ts;
8829 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8830 return -KVM_EOPNOTSUPP;
8833 * When tsc is in permanent catchup mode guests won't be able to use
8834 * pvclock_read_retry loop to get consistent view of pvclock
8836 if (vcpu->arch.tsc_always_catchup)
8837 return -KVM_EOPNOTSUPP;
8839 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8840 return -KVM_EOPNOTSUPP;
8842 clock_pairing.sec = ts.tv_sec;
8843 clock_pairing.nsec = ts.tv_nsec;
8844 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8845 clock_pairing.flags = 0;
8846 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8849 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8850 sizeof(struct kvm_clock_pairing)))
8858 * kvm_pv_kick_cpu_op: Kick a vcpu.
8860 * @apicid - apicid of vcpu to be kicked.
8862 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8865 * All other fields are unused for APIC_DM_REMRD, but may be consumed by
8866 * common code, e.g. for tracing. Defer initialization to the compiler.
8868 struct kvm_lapic_irq lapic_irq = {
8869 .delivery_mode = APIC_DM_REMRD,
8870 .dest_mode = APIC_DEST_PHYSICAL,
8871 .shorthand = APIC_DEST_NOSHORT,
8875 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8878 bool kvm_apicv_activated(struct kvm *kvm)
8880 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8882 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8884 static void kvm_apicv_init(struct kvm *kvm)
8886 mutex_init(&kvm->arch.apicv_update_lock);
8889 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8890 &kvm->arch.apicv_inhibit_reasons);
8892 set_bit(APICV_INHIBIT_REASON_DISABLE,
8893 &kvm->arch.apicv_inhibit_reasons);
8896 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8898 struct kvm_vcpu *target = NULL;
8899 struct kvm_apic_map *map;
8901 vcpu->stat.directed_yield_attempted++;
8903 if (single_task_running())
8907 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8909 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8910 target = map->phys_map[dest_id]->vcpu;
8914 if (!target || !READ_ONCE(target->ready))
8917 /* Ignore requests to yield to self */
8921 if (kvm_vcpu_yield_to(target) <= 0)
8924 vcpu->stat.directed_yield_successful++;
8930 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8932 u64 ret = vcpu->run->hypercall.ret;
8934 if (!is_64_bit_mode(vcpu))
8936 kvm_rax_write(vcpu, ret);
8937 ++vcpu->stat.hypercalls;
8938 return kvm_skip_emulated_instruction(vcpu);
8941 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8943 unsigned long nr, a0, a1, a2, a3, ret;
8946 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8947 return kvm_xen_hypercall(vcpu);
8949 if (kvm_hv_hypercall_enabled(vcpu))
8950 return kvm_hv_hypercall(vcpu);
8952 nr = kvm_rax_read(vcpu);
8953 a0 = kvm_rbx_read(vcpu);
8954 a1 = kvm_rcx_read(vcpu);
8955 a2 = kvm_rdx_read(vcpu);
8956 a3 = kvm_rsi_read(vcpu);
8958 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8960 op_64_bit = is_64_bit_hypercall(vcpu);
8969 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8977 case KVM_HC_VAPIC_POLL_IRQ:
8980 case KVM_HC_KICK_CPU:
8981 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8984 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8985 kvm_sched_yield(vcpu, a1);
8988 #ifdef CONFIG_X86_64
8989 case KVM_HC_CLOCK_PAIRING:
8990 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8993 case KVM_HC_SEND_IPI:
8994 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8997 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8999 case KVM_HC_SCHED_YIELD:
9000 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
9003 kvm_sched_yield(vcpu, a0);
9006 case KVM_HC_MAP_GPA_RANGE: {
9007 u64 gpa = a0, npages = a1, attrs = a2;
9010 if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
9013 if (!PAGE_ALIGNED(gpa) || !npages ||
9014 gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
9019 vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
9020 vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
9021 vcpu->run->hypercall.args[0] = gpa;
9022 vcpu->run->hypercall.args[1] = npages;
9023 vcpu->run->hypercall.args[2] = attrs;
9024 vcpu->run->hypercall.longmode = op_64_bit;
9025 vcpu->arch.complete_userspace_io = complete_hypercall_exit;
9035 kvm_rax_write(vcpu, ret);
9037 ++vcpu->stat.hypercalls;
9038 return kvm_skip_emulated_instruction(vcpu);
9040 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
9042 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
9044 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
9045 char instruction[3];
9046 unsigned long rip = kvm_rip_read(vcpu);
9048 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
9050 return emulator_write_emulated(ctxt, rip, instruction, 3,
9054 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
9056 return vcpu->run->request_interrupt_window &&
9057 likely(!pic_in_kernel(vcpu->kvm));
9060 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
9062 struct kvm_run *kvm_run = vcpu->run;
9064 kvm_run->if_flag = static_call(kvm_x86_get_if_flag)(vcpu);
9065 kvm_run->cr8 = kvm_get_cr8(vcpu);
9066 kvm_run->apic_base = kvm_get_apic_base(vcpu);
9069 * The call to kvm_ready_for_interrupt_injection() may end up in
9070 * kvm_xen_has_interrupt() which may require the srcu lock to be
9071 * held, to protect against changes in the vcpu_info address.
9073 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9074 kvm_run->ready_for_interrupt_injection =
9075 pic_in_kernel(vcpu->kvm) ||
9076 kvm_vcpu_ready_for_interrupt_injection(vcpu);
9077 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9080 kvm_run->flags |= KVM_RUN_X86_SMM;
9083 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
9087 if (!kvm_x86_ops.update_cr8_intercept)
9090 if (!lapic_in_kernel(vcpu))
9093 if (vcpu->arch.apicv_active)
9096 if (!vcpu->arch.apic->vapic_addr)
9097 max_irr = kvm_lapic_find_highest_irr(vcpu);
9104 tpr = kvm_lapic_get_cr8(vcpu);
9106 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9110 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9112 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9113 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9117 return kvm_x86_ops.nested_ops->check_events(vcpu);
9120 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9122 trace_kvm_inj_exception(vcpu->arch.exception.nr,
9123 vcpu->arch.exception.has_error_code,
9124 vcpu->arch.exception.error_code,
9125 vcpu->arch.exception.injected);
9127 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
9128 vcpu->arch.exception.error_code = false;
9129 static_call(kvm_x86_queue_exception)(vcpu);
9132 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
9135 bool can_inject = true;
9137 /* try to reinject previous events if any */
9139 if (vcpu->arch.exception.injected) {
9140 kvm_inject_exception(vcpu);
9144 * Do not inject an NMI or interrupt if there is a pending
9145 * exception. Exceptions and interrupts are recognized at
9146 * instruction boundaries, i.e. the start of an instruction.
9147 * Trap-like exceptions, e.g. #DB, have higher priority than
9148 * NMIs and interrupts, i.e. traps are recognized before an
9149 * NMI/interrupt that's pending on the same instruction.
9150 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
9151 * priority, but are only generated (pended) during instruction
9152 * execution, i.e. a pending fault-like exception means the
9153 * fault occurred on the *previous* instruction and must be
9154 * serviced prior to recognizing any new events in order to
9155 * fully complete the previous instruction.
9157 else if (!vcpu->arch.exception.pending) {
9158 if (vcpu->arch.nmi_injected) {
9159 static_call(kvm_x86_set_nmi)(vcpu);
9161 } else if (vcpu->arch.interrupt.injected) {
9162 static_call(kvm_x86_set_irq)(vcpu);
9167 WARN_ON_ONCE(vcpu->arch.exception.injected &&
9168 vcpu->arch.exception.pending);
9171 * Call check_nested_events() even if we reinjected a previous event
9172 * in order for caller to determine if it should require immediate-exit
9173 * from L2 to L1 due to pending L1 events which require exit
9176 if (is_guest_mode(vcpu)) {
9177 r = kvm_check_nested_events(vcpu);
9182 /* try to inject new event if pending */
9183 if (vcpu->arch.exception.pending) {
9185 * Fault-class exceptions, except #DBs, set RF=1 in the RFLAGS
9186 * value pushed on the stack. Trap-like exception and all #DBs
9187 * leave RF as-is (KVM follows Intel's behavior in this regard;
9188 * AMD states that code breakpoint #DBs excplitly clear RF=0).
9190 * Note, most versions of Intel's SDM and AMD's APM incorrectly
9191 * describe the behavior of General Detect #DBs, which are
9192 * fault-like. They do _not_ set RF, a la code breakpoints.
9194 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
9195 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
9198 if (vcpu->arch.exception.nr == DB_VECTOR) {
9199 kvm_deliver_exception_payload(vcpu);
9200 if (vcpu->arch.dr7 & DR7_GD) {
9201 vcpu->arch.dr7 &= ~DR7_GD;
9202 kvm_update_dr7(vcpu);
9206 kvm_inject_exception(vcpu);
9208 vcpu->arch.exception.pending = false;
9209 vcpu->arch.exception.injected = true;
9214 /* Don't inject interrupts if the user asked to avoid doing so */
9215 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
9219 * Finally, inject interrupt events. If an event cannot be injected
9220 * due to architectural conditions (e.g. IF=0) a window-open exit
9221 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
9222 * and can architecturally be injected, but we cannot do it right now:
9223 * an interrupt could have arrived just now and we have to inject it
9224 * as a vmexit, or there could already an event in the queue, which is
9225 * indicated by can_inject. In that case we request an immediate exit
9226 * in order to make progress and get back here for another iteration.
9227 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
9229 if (vcpu->arch.smi_pending) {
9230 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
9234 vcpu->arch.smi_pending = false;
9235 ++vcpu->arch.smi_count;
9239 static_call(kvm_x86_enable_smi_window)(vcpu);
9242 if (vcpu->arch.nmi_pending) {
9243 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
9247 --vcpu->arch.nmi_pending;
9248 vcpu->arch.nmi_injected = true;
9249 static_call(kvm_x86_set_nmi)(vcpu);
9251 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
9253 if (vcpu->arch.nmi_pending)
9254 static_call(kvm_x86_enable_nmi_window)(vcpu);
9257 if (kvm_cpu_has_injectable_intr(vcpu)) {
9258 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
9262 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
9263 static_call(kvm_x86_set_irq)(vcpu);
9264 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
9266 if (kvm_cpu_has_injectable_intr(vcpu))
9267 static_call(kvm_x86_enable_irq_window)(vcpu);
9270 if (is_guest_mode(vcpu) &&
9271 kvm_x86_ops.nested_ops->hv_timer_pending &&
9272 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
9273 *req_immediate_exit = true;
9275 WARN_ON(vcpu->arch.exception.pending);
9280 *req_immediate_exit = true;
9286 static void process_nmi(struct kvm_vcpu *vcpu)
9291 * x86 is limited to one NMI running, and one NMI pending after it.
9292 * If an NMI is already in progress, limit further NMIs to just one.
9293 * Otherwise, allow two (and we'll inject the first one immediately).
9295 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
9298 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
9299 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
9300 kvm_make_request(KVM_REQ_EVENT, vcpu);
9303 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
9306 flags |= seg->g << 23;
9307 flags |= seg->db << 22;
9308 flags |= seg->l << 21;
9309 flags |= seg->avl << 20;
9310 flags |= seg->present << 15;
9311 flags |= seg->dpl << 13;
9312 flags |= seg->s << 12;
9313 flags |= seg->type << 8;
9317 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
9319 struct kvm_segment seg;
9322 kvm_get_segment(vcpu, &seg, n);
9323 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9326 offset = 0x7f84 + n * 12;
9328 offset = 0x7f2c + (n - 3) * 12;
9330 put_smstate(u32, buf, offset + 8, seg.base);
9331 put_smstate(u32, buf, offset + 4, seg.limit);
9332 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9335 #ifdef CONFIG_X86_64
9336 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9338 struct kvm_segment seg;
9342 kvm_get_segment(vcpu, &seg, n);
9343 offset = 0x7e00 + n * 16;
9345 flags = enter_smm_get_segment_flags(&seg) >> 8;
9346 put_smstate(u16, buf, offset, seg.selector);
9347 put_smstate(u16, buf, offset + 2, flags);
9348 put_smstate(u32, buf, offset + 4, seg.limit);
9349 put_smstate(u64, buf, offset + 8, seg.base);
9353 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9356 struct kvm_segment seg;
9360 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9361 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9362 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9363 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9365 for (i = 0; i < 8; i++)
9366 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9368 kvm_get_dr(vcpu, 6, &val);
9369 put_smstate(u32, buf, 0x7fcc, (u32)val);
9370 kvm_get_dr(vcpu, 7, &val);
9371 put_smstate(u32, buf, 0x7fc8, (u32)val);
9373 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9374 put_smstate(u32, buf, 0x7fc4, seg.selector);
9375 put_smstate(u32, buf, 0x7f64, seg.base);
9376 put_smstate(u32, buf, 0x7f60, seg.limit);
9377 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9379 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9380 put_smstate(u32, buf, 0x7fc0, seg.selector);
9381 put_smstate(u32, buf, 0x7f80, seg.base);
9382 put_smstate(u32, buf, 0x7f7c, seg.limit);
9383 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9385 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9386 put_smstate(u32, buf, 0x7f74, dt.address);
9387 put_smstate(u32, buf, 0x7f70, dt.size);
9389 static_call(kvm_x86_get_idt)(vcpu, &dt);
9390 put_smstate(u32, buf, 0x7f58, dt.address);
9391 put_smstate(u32, buf, 0x7f54, dt.size);
9393 for (i = 0; i < 6; i++)
9394 enter_smm_save_seg_32(vcpu, buf, i);
9396 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9399 put_smstate(u32, buf, 0x7efc, 0x00020000);
9400 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9403 #ifdef CONFIG_X86_64
9404 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9407 struct kvm_segment seg;
9411 for (i = 0; i < 16; i++)
9412 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9414 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9415 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9417 kvm_get_dr(vcpu, 6, &val);
9418 put_smstate(u64, buf, 0x7f68, val);
9419 kvm_get_dr(vcpu, 7, &val);
9420 put_smstate(u64, buf, 0x7f60, val);
9422 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9423 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9424 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9426 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9429 put_smstate(u32, buf, 0x7efc, 0x00020064);
9431 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9433 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9434 put_smstate(u16, buf, 0x7e90, seg.selector);
9435 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9436 put_smstate(u32, buf, 0x7e94, seg.limit);
9437 put_smstate(u64, buf, 0x7e98, seg.base);
9439 static_call(kvm_x86_get_idt)(vcpu, &dt);
9440 put_smstate(u32, buf, 0x7e84, dt.size);
9441 put_smstate(u64, buf, 0x7e88, dt.address);
9443 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9444 put_smstate(u16, buf, 0x7e70, seg.selector);
9445 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9446 put_smstate(u32, buf, 0x7e74, seg.limit);
9447 put_smstate(u64, buf, 0x7e78, seg.base);
9449 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9450 put_smstate(u32, buf, 0x7e64, dt.size);
9451 put_smstate(u64, buf, 0x7e68, dt.address);
9453 for (i = 0; i < 6; i++)
9454 enter_smm_save_seg_64(vcpu, buf, i);
9458 static void enter_smm(struct kvm_vcpu *vcpu)
9460 struct kvm_segment cs, ds;
9465 memset(buf, 0, 512);
9466 #ifdef CONFIG_X86_64
9467 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9468 enter_smm_save_state_64(vcpu, buf);
9471 enter_smm_save_state_32(vcpu, buf);
9474 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9475 * state (e.g. leave guest mode) after we've saved the state into the
9476 * SMM state-save area.
9478 static_call(kvm_x86_enter_smm)(vcpu, buf);
9480 kvm_smm_changed(vcpu, true);
9481 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9483 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9484 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9486 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9488 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9489 kvm_rip_write(vcpu, 0x8000);
9491 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9492 static_call(kvm_x86_set_cr0)(vcpu, cr0);
9493 vcpu->arch.cr0 = cr0;
9495 static_call(kvm_x86_set_cr4)(vcpu, 0);
9497 /* Undocumented: IDT limit is set to zero on entry to SMM. */
9498 dt.address = dt.size = 0;
9499 static_call(kvm_x86_set_idt)(vcpu, &dt);
9501 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9503 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9504 cs.base = vcpu->arch.smbase;
9509 cs.limit = ds.limit = 0xffffffff;
9510 cs.type = ds.type = 0x3;
9511 cs.dpl = ds.dpl = 0;
9516 cs.avl = ds.avl = 0;
9517 cs.present = ds.present = 1;
9518 cs.unusable = ds.unusable = 0;
9519 cs.padding = ds.padding = 0;
9521 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9522 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9523 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9524 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9525 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9526 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9528 #ifdef CONFIG_X86_64
9529 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9530 static_call(kvm_x86_set_efer)(vcpu, 0);
9533 kvm_update_cpuid_runtime(vcpu);
9534 kvm_mmu_reset_context(vcpu);
9537 static void process_smi(struct kvm_vcpu *vcpu)
9539 vcpu->arch.smi_pending = true;
9540 kvm_make_request(KVM_REQ_EVENT, vcpu);
9543 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9544 unsigned long *vcpu_bitmap)
9548 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
9550 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
9551 NULL, vcpu_bitmap, cpus);
9553 free_cpumask_var(cpus);
9556 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9558 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9561 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9565 if (!lapic_in_kernel(vcpu))
9568 mutex_lock(&vcpu->kvm->arch.apicv_update_lock);
9570 activate = kvm_apicv_activated(vcpu->kvm);
9571 if (vcpu->arch.apicv_active == activate)
9574 vcpu->arch.apicv_active = activate;
9575 kvm_apic_update_apicv(vcpu);
9576 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9579 * When APICv gets disabled, we may still have injected interrupts
9580 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9581 * still active when the interrupt got accepted. Make sure
9582 * inject_pending_event() is called to check for that.
9584 if (!vcpu->arch.apicv_active)
9585 kvm_make_request(KVM_REQ_EVENT, vcpu);
9588 mutex_unlock(&vcpu->kvm->arch.apicv_update_lock);
9590 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9592 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9594 unsigned long old, new;
9596 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9597 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9600 old = new = kvm->arch.apicv_inhibit_reasons;
9603 __clear_bit(bit, &new);
9605 __set_bit(bit, &new);
9607 if (!!old != !!new) {
9608 trace_kvm_apicv_update_request(activate, bit);
9609 kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9610 kvm->arch.apicv_inhibit_reasons = new;
9612 unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9613 kvm_zap_gfn_range(kvm, gfn, gfn+1);
9616 kvm->arch.apicv_inhibit_reasons = new;
9618 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9620 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9622 mutex_lock(&kvm->arch.apicv_update_lock);
9623 __kvm_request_apicv_update(kvm, activate, bit);
9624 mutex_unlock(&kvm->arch.apicv_update_lock);
9626 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9628 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9630 if (!kvm_apic_present(vcpu))
9633 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9635 if (irqchip_split(vcpu->kvm))
9636 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9638 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9639 if (ioapic_in_kernel(vcpu->kvm))
9640 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9643 if (is_guest_mode(vcpu))
9644 vcpu->arch.load_eoi_exitmap_pending = true;
9646 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9649 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9651 u64 eoi_exit_bitmap[4];
9653 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9656 if (to_hv_vcpu(vcpu)) {
9657 bitmap_or((ulong *)eoi_exit_bitmap,
9658 vcpu->arch.ioapic_handled_vectors,
9659 to_hv_synic(vcpu)->vec_bitmap, 256);
9660 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9664 static_call(kvm_x86_load_eoi_exitmap)(
9665 vcpu, (u64 *)vcpu->arch.ioapic_handled_vectors);
9668 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9669 unsigned long start, unsigned long end)
9671 unsigned long apic_address;
9674 * The physical address of apic access page is stored in the VMCS.
9675 * Update it when it becomes invalid.
9677 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9678 if (start <= apic_address && apic_address < end)
9679 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9682 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
9684 static_call_cond(kvm_x86_guest_memory_reclaimed)(kvm);
9687 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9689 if (!lapic_in_kernel(vcpu))
9692 if (!kvm_x86_ops.set_apic_access_page_addr)
9695 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9698 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9700 smp_send_reschedule(vcpu->cpu);
9702 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9705 * Returns 1 to let vcpu_run() continue the guest execution loop without
9706 * exiting to the userspace. Otherwise, the value will be returned to the
9709 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9713 dm_request_for_irq_injection(vcpu) &&
9714 kvm_cpu_accept_dm_intr(vcpu);
9715 fastpath_t exit_fastpath;
9717 bool req_immediate_exit = false;
9719 /* Forbid vmenter if vcpu dirty ring is soft-full */
9720 if (unlikely(vcpu->kvm->dirty_ring_size &&
9721 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9722 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9723 trace_kvm_dirty_ring_exit(vcpu);
9728 if (kvm_request_pending(vcpu)) {
9729 if (kvm_check_request(KVM_REQ_VM_BUGGED, vcpu)) {
9733 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9734 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9739 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9740 kvm_mmu_unload(vcpu);
9741 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9742 __kvm_migrate_timers(vcpu);
9743 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9744 kvm_gen_update_masterclock(vcpu->kvm);
9745 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9746 kvm_gen_kvmclock_update(vcpu);
9747 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9748 r = kvm_guest_time_update(vcpu);
9752 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9753 kvm_mmu_sync_roots(vcpu);
9754 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9755 kvm_mmu_load_pgd(vcpu);
9756 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9757 kvm_vcpu_flush_tlb_all(vcpu);
9759 /* Flushing all ASIDs flushes the current ASID... */
9760 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9762 kvm_service_local_tlb_flush_requests(vcpu);
9764 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9765 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9769 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9770 if (is_guest_mode(vcpu)) {
9771 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9773 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9774 vcpu->mmio_needed = 0;
9779 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9780 /* Page is swapped out. Do synthetic halt */
9781 vcpu->arch.apf.halted = true;
9785 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9786 record_steal_time(vcpu);
9787 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9789 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9791 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9792 kvm_pmu_handle_event(vcpu);
9793 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9794 kvm_pmu_deliver_pmi(vcpu);
9795 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9796 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9797 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9798 vcpu->arch.ioapic_handled_vectors)) {
9799 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9800 vcpu->run->eoi.vector =
9801 vcpu->arch.pending_ioapic_eoi;
9806 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9807 vcpu_scan_ioapic(vcpu);
9808 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9809 vcpu_load_eoi_exitmap(vcpu);
9810 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9811 kvm_vcpu_reload_apic_access_page(vcpu);
9812 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9813 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9814 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9818 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9819 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9820 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9824 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9825 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9827 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9828 vcpu->run->hyperv = hv_vcpu->exit;
9834 * KVM_REQ_HV_STIMER has to be processed after
9835 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9836 * depend on the guest clock being up-to-date
9838 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9839 kvm_hv_process_stimers(vcpu);
9840 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9841 kvm_vcpu_update_apicv(vcpu);
9842 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9843 kvm_check_async_pf_completion(vcpu);
9844 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9845 static_call(kvm_x86_msr_filter_changed)(vcpu);
9847 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9848 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9851 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9852 kvm_xen_has_interrupt(vcpu)) {
9853 ++vcpu->stat.req_event;
9854 r = kvm_apic_accept_events(vcpu);
9859 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9864 r = inject_pending_event(vcpu, &req_immediate_exit);
9870 static_call(kvm_x86_enable_irq_window)(vcpu);
9872 if (kvm_lapic_enabled(vcpu)) {
9873 update_cr8_intercept(vcpu);
9874 kvm_lapic_sync_to_vapic(vcpu);
9878 r = kvm_mmu_reload(vcpu);
9880 goto cancel_injection;
9885 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9888 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9889 * IPI are then delayed after guest entry, which ensures that they
9890 * result in virtual interrupt delivery.
9892 local_irq_disable();
9893 vcpu->mode = IN_GUEST_MODE;
9895 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9898 * 1) We should set ->mode before checking ->requests. Please see
9899 * the comment in kvm_vcpu_exiting_guest_mode().
9901 * 2) For APICv, we should set ->mode before checking PID.ON. This
9902 * pairs with the memory barrier implicit in pi_test_and_set_on
9903 * (see vmx_deliver_posted_interrupt).
9905 * 3) This also orders the write to mode from any reads to the page
9906 * tables done while the VCPU is running. Please see the comment
9907 * in kvm_flush_remote_tlbs.
9909 smp_mb__after_srcu_read_unlock();
9912 * This handles the case where a posted interrupt was
9913 * notified with kvm_vcpu_kick. Assigned devices can
9914 * use the POSTED_INTR_VECTOR even if APICv is disabled,
9915 * so do it even if APICv is disabled on this vCPU.
9917 if (kvm_lapic_enabled(vcpu))
9918 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9920 if (kvm_vcpu_exit_request(vcpu)) {
9921 vcpu->mode = OUTSIDE_GUEST_MODE;
9925 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9927 goto cancel_injection;
9930 if (req_immediate_exit) {
9931 kvm_make_request(KVM_REQ_EVENT, vcpu);
9932 static_call(kvm_x86_request_immediate_exit)(vcpu);
9935 fpregs_assert_state_consistent();
9936 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9937 switch_fpu_return();
9939 if (unlikely(vcpu->arch.switch_db_regs)) {
9941 set_debugreg(vcpu->arch.eff_db[0], 0);
9942 set_debugreg(vcpu->arch.eff_db[1], 1);
9943 set_debugreg(vcpu->arch.eff_db[2], 2);
9944 set_debugreg(vcpu->arch.eff_db[3], 3);
9945 } else if (unlikely(hw_breakpoint_active())) {
9950 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9951 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9954 if (kvm_lapic_enabled(vcpu))
9955 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9957 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9958 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9962 /* Note, VM-Exits that go down the "slow" path are accounted below. */
9967 * Do this here before restoring debug registers on the host. And
9968 * since we do this before handling the vmexit, a DR access vmexit
9969 * can (a) read the correct value of the debug registers, (b) set
9970 * KVM_DEBUGREG_WONT_EXIT again.
9972 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9973 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9974 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9975 kvm_update_dr0123(vcpu);
9976 kvm_update_dr7(vcpu);
9980 * If the guest has used debug registers, at least dr7
9981 * will be disabled while returning to the host.
9982 * If we don't have active breakpoints in the host, we don't
9983 * care about the messed up debug address registers. But if
9984 * we have some of them active, restore the old state.
9986 if (hw_breakpoint_active())
9987 hw_breakpoint_restore();
9989 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9990 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9992 vcpu->mode = OUTSIDE_GUEST_MODE;
9995 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9998 * Consume any pending interrupts, including the possible source of
9999 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
10000 * An instruction is required after local_irq_enable() to fully unblock
10001 * interrupts on processors that implement an interrupt shadow, the
10002 * stat.exits increment will do nicely.
10004 kvm_before_interrupt(vcpu);
10005 local_irq_enable();
10006 ++vcpu->stat.exits;
10007 local_irq_disable();
10008 kvm_after_interrupt(vcpu);
10011 * Wait until after servicing IRQs to account guest time so that any
10012 * ticks that occurred while running the guest are properly accounted
10013 * to the guest. Waiting until IRQs are enabled degrades the accuracy
10014 * of accounting via context tracking, but the loss of accuracy is
10015 * acceptable for all known use cases.
10017 vtime_account_guest_exit();
10019 if (lapic_in_kernel(vcpu)) {
10020 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
10021 if (delta != S64_MIN) {
10022 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
10023 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
10027 local_irq_enable();
10030 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
10033 * Profile KVM exit RIPs:
10035 if (unlikely(prof_on == KVM_PROFILING)) {
10036 unsigned long rip = kvm_rip_read(vcpu);
10037 profile_hit(KVM_PROFILING, (void *)rip);
10040 if (unlikely(vcpu->arch.tsc_always_catchup))
10041 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10043 if (vcpu->arch.apic_attention)
10044 kvm_lapic_sync_from_vapic(vcpu);
10046 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
10050 if (req_immediate_exit)
10051 kvm_make_request(KVM_REQ_EVENT, vcpu);
10052 static_call(kvm_x86_cancel_injection)(vcpu);
10053 if (unlikely(vcpu->arch.apic_attention))
10054 kvm_lapic_sync_from_vapic(vcpu);
10059 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
10063 if (!kvm_arch_vcpu_runnable(vcpu) &&
10064 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
10066 * Switch to the software timer before halt-polling/blocking as
10067 * the guest's timer may be a break event for the vCPU, and the
10068 * hypervisor timer runs only when the CPU is in guest mode.
10069 * Switch before halt-polling so that KVM recognizes an expired
10070 * timer before blocking.
10072 hv_timer = kvm_lapic_hv_timer_in_use(vcpu);
10074 kvm_lapic_switch_to_sw_timer(vcpu);
10076 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10077 kvm_vcpu_block(vcpu);
10078 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10081 kvm_lapic_switch_to_hv_timer(vcpu);
10083 if (kvm_x86_ops.post_block)
10084 static_call(kvm_x86_post_block)(vcpu);
10086 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
10090 if (kvm_apic_accept_events(vcpu) < 0)
10092 switch(vcpu->arch.mp_state) {
10093 case KVM_MP_STATE_HALTED:
10094 case KVM_MP_STATE_AP_RESET_HOLD:
10095 vcpu->arch.pv.pv_unhalted = false;
10096 vcpu->arch.mp_state =
10097 KVM_MP_STATE_RUNNABLE;
10099 case KVM_MP_STATE_RUNNABLE:
10100 vcpu->arch.apf.halted = false;
10102 case KVM_MP_STATE_INIT_RECEIVED:
10110 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
10112 if (is_guest_mode(vcpu))
10113 kvm_check_nested_events(vcpu);
10115 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
10116 !vcpu->arch.apf.halted);
10119 static int vcpu_run(struct kvm_vcpu *vcpu)
10122 struct kvm *kvm = vcpu->kvm;
10124 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10125 vcpu->arch.l1tf_flush_l1d = true;
10129 * If another guest vCPU requests a PV TLB flush in the middle
10130 * of instruction emulation, the rest of the emulation could
10131 * use a stale page translation. Assume that any code after
10132 * this point can start executing an instruction.
10134 vcpu->arch.at_instruction_boundary = false;
10135 if (kvm_vcpu_running(vcpu)) {
10136 r = vcpu_enter_guest(vcpu);
10138 r = vcpu_block(kvm, vcpu);
10144 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10145 if (kvm_cpu_has_pending_timer(vcpu))
10146 kvm_inject_pending_timer_irqs(vcpu);
10148 if (dm_request_for_irq_injection(vcpu) &&
10149 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10151 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10152 ++vcpu->stat.request_irq_exits;
10156 if (__xfer_to_guest_mode_work_pending()) {
10157 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10158 r = xfer_to_guest_mode_handle_work(vcpu);
10161 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10165 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10170 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
10174 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
10175 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
10176 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
10180 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
10182 BUG_ON(!vcpu->arch.pio.count);
10184 return complete_emulated_io(vcpu);
10188 * Implements the following, as a state machine:
10191 * for each fragment
10192 * for each mmio piece in the fragment
10199 * for each fragment
10200 * for each mmio piece in the fragment
10205 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10207 struct kvm_run *run = vcpu->run;
10208 struct kvm_mmio_fragment *frag;
10211 BUG_ON(!vcpu->mmio_needed);
10213 /* Complete previous fragment */
10214 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10215 len = min(8u, frag->len);
10216 if (!vcpu->mmio_is_write)
10217 memcpy(frag->data, run->mmio.data, len);
10219 if (frag->len <= 8) {
10220 /* Switch to the next fragment. */
10222 vcpu->mmio_cur_fragment++;
10224 /* Go forward to the next mmio piece. */
10230 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10231 vcpu->mmio_needed = 0;
10233 /* FIXME: return into emulator if single-stepping. */
10234 if (vcpu->mmio_is_write)
10236 vcpu->mmio_read_completed = 1;
10237 return complete_emulated_io(vcpu);
10240 run->exit_reason = KVM_EXIT_MMIO;
10241 run->mmio.phys_addr = frag->gpa;
10242 if (vcpu->mmio_is_write)
10243 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10244 run->mmio.len = min(8u, frag->len);
10245 run->mmio.is_write = vcpu->mmio_is_write;
10246 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10250 static void kvm_save_current_fpu(struct fpu *fpu)
10253 * If the target FPU state is not resident in the CPU registers, just
10254 * memcpy() from current, else save CPU state directly to the target.
10256 if (test_thread_flag(TIF_NEED_FPU_LOAD))
10257 memcpy(&fpu->state, ¤t->thread.fpu.state,
10258 fpu_kernel_xstate_size);
10260 save_fpregs_to_fpstate(fpu);
10263 /* Swap (qemu) user FPU context for the guest FPU context. */
10264 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10268 kvm_save_current_fpu(vcpu->arch.user_fpu);
10271 * Guests with protected state can't have it set by the hypervisor,
10272 * so skip trying to set it.
10274 if (vcpu->arch.guest_fpu)
10275 /* PKRU is separately restored in kvm_x86_ops.run. */
10276 __restore_fpregs_from_fpstate(&vcpu->arch.guest_fpu->state,
10277 ~XFEATURE_MASK_PKRU);
10279 fpregs_mark_activate();
10285 /* When vcpu_run ends, restore user space FPU context. */
10286 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10291 * Guests with protected state can't have it read by the hypervisor,
10292 * so skip trying to save it.
10294 if (vcpu->arch.guest_fpu)
10295 kvm_save_current_fpu(vcpu->arch.guest_fpu);
10297 restore_fpregs_from_fpstate(&vcpu->arch.user_fpu->state);
10299 fpregs_mark_activate();
10302 ++vcpu->stat.fpu_reload;
10306 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10308 struct kvm_run *kvm_run = vcpu->run;
10312 kvm_sigset_activate(vcpu);
10313 kvm_run->flags = 0;
10314 kvm_load_guest_fpu(vcpu);
10316 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10317 if (kvm_run->immediate_exit) {
10322 * It should be impossible for the hypervisor timer to be in
10323 * use before KVM has ever run the vCPU.
10325 WARN_ON_ONCE(kvm_lapic_hv_timer_in_use(vcpu));
10326 kvm_vcpu_block(vcpu);
10327 if (kvm_apic_accept_events(vcpu) < 0) {
10331 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
10333 if (signal_pending(current)) {
10335 kvm_run->exit_reason = KVM_EXIT_INTR;
10336 ++vcpu->stat.signal_exits;
10341 if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10342 (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10347 if (kvm_run->kvm_dirty_regs) {
10348 r = sync_regs(vcpu);
10353 /* re-sync apic's tpr */
10354 if (!lapic_in_kernel(vcpu)) {
10355 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10361 if (unlikely(vcpu->arch.complete_userspace_io)) {
10362 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10363 vcpu->arch.complete_userspace_io = NULL;
10368 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10370 if (kvm_run->immediate_exit)
10373 r = vcpu_run(vcpu);
10376 kvm_put_guest_fpu(vcpu);
10377 if (kvm_run->kvm_valid_regs)
10379 post_kvm_run_save(vcpu);
10380 kvm_sigset_deactivate(vcpu);
10386 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10388 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10390 * We are here if userspace calls get_regs() in the middle of
10391 * instruction emulation. Registers state needs to be copied
10392 * back from emulation context to vcpu. Userspace shouldn't do
10393 * that usually, but some bad designed PV devices (vmware
10394 * backdoor interface) need this to work
10396 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10397 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10399 regs->rax = kvm_rax_read(vcpu);
10400 regs->rbx = kvm_rbx_read(vcpu);
10401 regs->rcx = kvm_rcx_read(vcpu);
10402 regs->rdx = kvm_rdx_read(vcpu);
10403 regs->rsi = kvm_rsi_read(vcpu);
10404 regs->rdi = kvm_rdi_read(vcpu);
10405 regs->rsp = kvm_rsp_read(vcpu);
10406 regs->rbp = kvm_rbp_read(vcpu);
10407 #ifdef CONFIG_X86_64
10408 regs->r8 = kvm_r8_read(vcpu);
10409 regs->r9 = kvm_r9_read(vcpu);
10410 regs->r10 = kvm_r10_read(vcpu);
10411 regs->r11 = kvm_r11_read(vcpu);
10412 regs->r12 = kvm_r12_read(vcpu);
10413 regs->r13 = kvm_r13_read(vcpu);
10414 regs->r14 = kvm_r14_read(vcpu);
10415 regs->r15 = kvm_r15_read(vcpu);
10418 regs->rip = kvm_rip_read(vcpu);
10419 regs->rflags = kvm_get_rflags(vcpu);
10422 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10425 __get_regs(vcpu, regs);
10430 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10432 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10433 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10435 kvm_rax_write(vcpu, regs->rax);
10436 kvm_rbx_write(vcpu, regs->rbx);
10437 kvm_rcx_write(vcpu, regs->rcx);
10438 kvm_rdx_write(vcpu, regs->rdx);
10439 kvm_rsi_write(vcpu, regs->rsi);
10440 kvm_rdi_write(vcpu, regs->rdi);
10441 kvm_rsp_write(vcpu, regs->rsp);
10442 kvm_rbp_write(vcpu, regs->rbp);
10443 #ifdef CONFIG_X86_64
10444 kvm_r8_write(vcpu, regs->r8);
10445 kvm_r9_write(vcpu, regs->r9);
10446 kvm_r10_write(vcpu, regs->r10);
10447 kvm_r11_write(vcpu, regs->r11);
10448 kvm_r12_write(vcpu, regs->r12);
10449 kvm_r13_write(vcpu, regs->r13);
10450 kvm_r14_write(vcpu, regs->r14);
10451 kvm_r15_write(vcpu, regs->r15);
10454 kvm_rip_write(vcpu, regs->rip);
10455 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10457 vcpu->arch.exception.pending = false;
10459 kvm_make_request(KVM_REQ_EVENT, vcpu);
10462 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10465 __set_regs(vcpu, regs);
10470 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10472 struct kvm_segment cs;
10474 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10478 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10480 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10482 struct desc_ptr dt;
10484 if (vcpu->arch.guest_state_protected)
10485 goto skip_protected_regs;
10487 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10488 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10489 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10490 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10491 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10492 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10494 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10495 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10497 static_call(kvm_x86_get_idt)(vcpu, &dt);
10498 sregs->idt.limit = dt.size;
10499 sregs->idt.base = dt.address;
10500 static_call(kvm_x86_get_gdt)(vcpu, &dt);
10501 sregs->gdt.limit = dt.size;
10502 sregs->gdt.base = dt.address;
10504 sregs->cr2 = vcpu->arch.cr2;
10505 sregs->cr3 = kvm_read_cr3(vcpu);
10507 skip_protected_regs:
10508 sregs->cr0 = kvm_read_cr0(vcpu);
10509 sregs->cr4 = kvm_read_cr4(vcpu);
10510 sregs->cr8 = kvm_get_cr8(vcpu);
10511 sregs->efer = vcpu->arch.efer;
10512 sregs->apic_base = kvm_get_apic_base(vcpu);
10515 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10517 __get_sregs_common(vcpu, sregs);
10519 if (vcpu->arch.guest_state_protected)
10522 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10523 set_bit(vcpu->arch.interrupt.nr,
10524 (unsigned long *)sregs->interrupt_bitmap);
10527 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10531 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10533 if (vcpu->arch.guest_state_protected)
10536 if (is_pae_paging(vcpu)) {
10537 for (i = 0 ; i < 4 ; i++)
10538 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10539 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10543 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10544 struct kvm_sregs *sregs)
10547 __get_sregs(vcpu, sregs);
10552 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10553 struct kvm_mp_state *mp_state)
10558 if (kvm_mpx_supported())
10559 kvm_load_guest_fpu(vcpu);
10561 r = kvm_apic_accept_events(vcpu);
10566 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10567 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10568 vcpu->arch.pv.pv_unhalted)
10569 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10571 mp_state->mp_state = vcpu->arch.mp_state;
10574 if (kvm_mpx_supported())
10575 kvm_put_guest_fpu(vcpu);
10580 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10581 struct kvm_mp_state *mp_state)
10587 if (!lapic_in_kernel(vcpu) &&
10588 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10592 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10593 * INIT state; latched init should be reported using
10594 * KVM_SET_VCPU_EVENTS, so reject it here.
10596 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10597 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10598 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10601 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10602 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10603 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10605 vcpu->arch.mp_state = mp_state->mp_state;
10606 kvm_make_request(KVM_REQ_EVENT, vcpu);
10614 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10615 int reason, bool has_error_code, u32 error_code)
10617 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10620 init_emulate_ctxt(vcpu);
10622 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10623 has_error_code, error_code);
10625 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10626 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10627 vcpu->run->internal.ndata = 0;
10631 kvm_rip_write(vcpu, ctxt->eip);
10632 kvm_set_rflags(vcpu, ctxt->eflags);
10635 EXPORT_SYMBOL_GPL(kvm_task_switch);
10637 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10639 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10641 * When EFER.LME and CR0.PG are set, the processor is in
10642 * 64-bit mode (though maybe in a 32-bit code segment).
10643 * CR4.PAE and EFER.LMA must be set.
10645 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10647 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10651 * Not in 64-bit mode: EFER.LMA is clear and the code
10652 * segment cannot be 64-bit.
10654 if (sregs->efer & EFER_LMA || sregs->cs.l)
10658 return kvm_is_valid_cr4(vcpu, sregs->cr4) &&
10659 kvm_is_valid_cr0(vcpu, sregs->cr0);
10662 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10663 int *mmu_reset_needed, bool update_pdptrs)
10665 struct msr_data apic_base_msr;
10667 struct desc_ptr dt;
10669 if (!kvm_is_valid_sregs(vcpu, sregs))
10672 apic_base_msr.data = sregs->apic_base;
10673 apic_base_msr.host_initiated = true;
10674 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10677 if (vcpu->arch.guest_state_protected)
10680 dt.size = sregs->idt.limit;
10681 dt.address = sregs->idt.base;
10682 static_call(kvm_x86_set_idt)(vcpu, &dt);
10683 dt.size = sregs->gdt.limit;
10684 dt.address = sregs->gdt.base;
10685 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10687 vcpu->arch.cr2 = sregs->cr2;
10688 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10689 vcpu->arch.cr3 = sregs->cr3;
10690 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10692 kvm_set_cr8(vcpu, sregs->cr8);
10694 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10695 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10697 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10698 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10699 vcpu->arch.cr0 = sregs->cr0;
10701 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10702 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10704 if (update_pdptrs) {
10705 idx = srcu_read_lock(&vcpu->kvm->srcu);
10706 if (is_pae_paging(vcpu)) {
10707 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10708 *mmu_reset_needed = 1;
10710 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10713 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10714 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10715 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10716 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10717 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10718 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10720 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10721 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10723 update_cr8_intercept(vcpu);
10725 /* Older userspace won't unhalt the vcpu on reset. */
10726 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10727 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10728 !is_protmode(vcpu))
10729 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10734 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10736 int pending_vec, max_bits;
10737 int mmu_reset_needed = 0;
10738 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10743 if (mmu_reset_needed)
10744 kvm_mmu_reset_context(vcpu);
10746 max_bits = KVM_NR_INTERRUPTS;
10747 pending_vec = find_first_bit(
10748 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10750 if (pending_vec < max_bits) {
10751 kvm_queue_interrupt(vcpu, pending_vec, false);
10752 pr_debug("Set back pending irq %d\n", pending_vec);
10753 kvm_make_request(KVM_REQ_EVENT, vcpu);
10758 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10760 int mmu_reset_needed = 0;
10761 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10762 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10763 !(sregs2->efer & EFER_LMA);
10766 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10769 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10772 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10773 &mmu_reset_needed, !valid_pdptrs);
10777 if (valid_pdptrs) {
10778 for (i = 0; i < 4 ; i++)
10779 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10781 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10782 mmu_reset_needed = 1;
10783 vcpu->arch.pdptrs_from_userspace = true;
10785 if (mmu_reset_needed)
10786 kvm_mmu_reset_context(vcpu);
10790 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10791 struct kvm_sregs *sregs)
10796 ret = __set_sregs(vcpu, sregs);
10801 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10802 struct kvm_guest_debug *dbg)
10804 unsigned long rflags;
10807 if (vcpu->arch.guest_state_protected)
10812 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10814 if (vcpu->arch.exception.pending)
10816 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10817 kvm_queue_exception(vcpu, DB_VECTOR);
10819 kvm_queue_exception(vcpu, BP_VECTOR);
10823 * Read rflags as long as potentially injected trace flags are still
10826 rflags = kvm_get_rflags(vcpu);
10828 vcpu->guest_debug = dbg->control;
10829 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10830 vcpu->guest_debug = 0;
10832 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10833 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10834 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10835 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10837 for (i = 0; i < KVM_NR_DB_REGS; i++)
10838 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10840 kvm_update_dr7(vcpu);
10842 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10843 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10846 * Trigger an rflags update that will inject or remove the trace
10849 kvm_set_rflags(vcpu, rflags);
10851 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10861 * Translate a guest virtual address to a guest physical address.
10863 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10864 struct kvm_translation *tr)
10866 unsigned long vaddr = tr->linear_address;
10872 idx = srcu_read_lock(&vcpu->kvm->srcu);
10873 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10874 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10875 tr->physical_address = gpa;
10876 tr->valid = gpa != UNMAPPED_GVA;
10884 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10886 struct fxregs_state *fxsave;
10888 if (!vcpu->arch.guest_fpu)
10893 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10894 memcpy(fpu->fpr, fxsave->st_space, 128);
10895 fpu->fcw = fxsave->cwd;
10896 fpu->fsw = fxsave->swd;
10897 fpu->ftwx = fxsave->twd;
10898 fpu->last_opcode = fxsave->fop;
10899 fpu->last_ip = fxsave->rip;
10900 fpu->last_dp = fxsave->rdp;
10901 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10907 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10909 struct fxregs_state *fxsave;
10911 if (!vcpu->arch.guest_fpu)
10916 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10918 memcpy(fxsave->st_space, fpu->fpr, 128);
10919 fxsave->cwd = fpu->fcw;
10920 fxsave->swd = fpu->fsw;
10921 fxsave->twd = fpu->ftwx;
10922 fxsave->fop = fpu->last_opcode;
10923 fxsave->rip = fpu->last_ip;
10924 fxsave->rdp = fpu->last_dp;
10925 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10931 static void store_regs(struct kvm_vcpu *vcpu)
10933 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10935 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10936 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10938 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10939 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10941 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10942 kvm_vcpu_ioctl_x86_get_vcpu_events(
10943 vcpu, &vcpu->run->s.regs.events);
10946 static int sync_regs(struct kvm_vcpu *vcpu)
10948 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10949 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10950 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10952 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10953 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10955 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10957 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10958 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10959 vcpu, &vcpu->run->s.regs.events))
10961 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10967 static void fx_init(struct kvm_vcpu *vcpu)
10969 if (!vcpu->arch.guest_fpu)
10972 fpstate_init(&vcpu->arch.guest_fpu->state);
10973 if (boot_cpu_has(X86_FEATURE_XSAVES))
10974 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10975 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10978 * Ensure guest xcr0 is valid for loading
10980 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10982 vcpu->arch.cr0 |= X86_CR0_ET;
10985 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10987 if (vcpu->arch.guest_fpu) {
10988 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10989 vcpu->arch.guest_fpu = NULL;
10992 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10994 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10996 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10997 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10998 "guest TSC will not be reliable\n");
11003 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
11008 vcpu->arch.last_vmentry_cpu = -1;
11009 vcpu->arch.regs_avail = ~0;
11010 vcpu->arch.regs_dirty = ~0;
11012 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
11013 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11015 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
11017 r = kvm_mmu_create(vcpu);
11021 if (irqchip_in_kernel(vcpu->kvm)) {
11022 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
11024 goto fail_mmu_destroy;
11027 * Defer evaluating inhibits until the vCPU is first run, as
11028 * this vCPU will not get notified of any changes until this
11029 * vCPU is visible to other vCPUs (marked online and added to
11030 * the set of vCPUs). Opportunistically mark APICv active as
11031 * VMX in particularly is highly unlikely to have inhibits.
11032 * Ignore the current per-VM APICv state so that vCPU creation
11033 * is guaranteed to run with a deterministic value, the request
11034 * will ensure the vCPU gets the correct state before VM-Entry.
11036 if (enable_apicv) {
11037 vcpu->arch.apicv_active = true;
11038 kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
11041 static_branch_inc(&kvm_has_noapic_vcpu);
11045 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
11047 goto fail_free_lapic;
11048 vcpu->arch.pio_data = page_address(page);
11050 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
11051 GFP_KERNEL_ACCOUNT);
11052 if (!vcpu->arch.mce_banks)
11053 goto fail_free_pio_data;
11054 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
11056 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
11057 GFP_KERNEL_ACCOUNT))
11058 goto fail_free_mce_banks;
11060 if (!alloc_emulate_ctxt(vcpu))
11061 goto free_wbinvd_dirty_mask;
11063 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
11064 GFP_KERNEL_ACCOUNT);
11065 if (!vcpu->arch.user_fpu) {
11066 pr_err("kvm: failed to allocate userspace's fpu\n");
11067 goto free_emulate_ctxt;
11070 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
11071 GFP_KERNEL_ACCOUNT);
11072 if (!vcpu->arch.guest_fpu) {
11073 pr_err("kvm: failed to allocate vcpu's fpu\n");
11074 goto free_user_fpu;
11078 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
11079 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
11081 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
11083 kvm_async_pf_hash_reset(vcpu);
11084 kvm_pmu_init(vcpu);
11086 vcpu->arch.pending_external_vector = -1;
11087 vcpu->arch.preempted_in_kernel = false;
11089 #if IS_ENABLED(CONFIG_HYPERV)
11090 vcpu->arch.hv_root_tdp = INVALID_PAGE;
11093 r = static_call(kvm_x86_vcpu_create)(vcpu);
11095 goto free_guest_fpu;
11097 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
11098 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
11099 kvm_vcpu_mtrr_init(vcpu);
11101 kvm_set_tsc_khz(vcpu, max_tsc_khz);
11102 kvm_vcpu_reset(vcpu, false);
11103 kvm_init_mmu(vcpu);
11108 kvm_free_guest_fpu(vcpu);
11110 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
11112 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11113 free_wbinvd_dirty_mask:
11114 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11115 fail_free_mce_banks:
11116 kfree(vcpu->arch.mce_banks);
11117 fail_free_pio_data:
11118 free_page((unsigned long)vcpu->arch.pio_data);
11120 kvm_free_lapic(vcpu);
11122 kvm_mmu_destroy(vcpu);
11126 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
11128 struct kvm *kvm = vcpu->kvm;
11130 if (mutex_lock_killable(&vcpu->mutex))
11133 kvm_synchronize_tsc(vcpu, 0);
11136 /* poll control enabled by default */
11137 vcpu->arch.msr_kvm_poll_control = 1;
11139 mutex_unlock(&vcpu->mutex);
11141 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
11142 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
11143 KVMCLOCK_SYNC_PERIOD);
11146 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
11150 kvmclock_reset(vcpu);
11152 static_call(kvm_x86_vcpu_free)(vcpu);
11154 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11155 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11156 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
11157 kvm_free_guest_fpu(vcpu);
11159 kvm_hv_vcpu_uninit(vcpu);
11160 kvm_pmu_destroy(vcpu);
11161 kfree(vcpu->arch.mce_banks);
11162 kvm_free_lapic(vcpu);
11163 idx = srcu_read_lock(&vcpu->kvm->srcu);
11164 kvm_mmu_destroy(vcpu);
11165 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11166 free_page((unsigned long)vcpu->arch.pio_data);
11167 kvfree(vcpu->arch.cpuid_entries);
11168 if (!lapic_in_kernel(vcpu))
11169 static_branch_dec(&kvm_has_noapic_vcpu);
11172 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
11174 unsigned long old_cr0 = kvm_read_cr0(vcpu);
11175 unsigned long new_cr0;
11179 * SVM doesn't unconditionally VM-Exit on INIT and SHUTDOWN, thus it's
11180 * possible to INIT the vCPU while L2 is active. Force the vCPU back
11181 * into L1 as EFER.SVME is cleared on INIT (along with all other EFER
11182 * bits), i.e. virtualization is disabled.
11184 if (is_guest_mode(vcpu))
11185 kvm_leave_nested(vcpu);
11187 kvm_lapic_reset(vcpu, init_event);
11189 WARN_ON_ONCE(is_guest_mode(vcpu) || is_smm(vcpu));
11190 vcpu->arch.hflags = 0;
11192 vcpu->arch.smi_pending = 0;
11193 vcpu->arch.smi_count = 0;
11194 atomic_set(&vcpu->arch.nmi_queued, 0);
11195 vcpu->arch.nmi_pending = 0;
11196 vcpu->arch.nmi_injected = false;
11197 kvm_clear_interrupt_queue(vcpu);
11198 kvm_clear_exception_queue(vcpu);
11200 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
11201 kvm_update_dr0123(vcpu);
11202 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
11203 vcpu->arch.dr7 = DR7_FIXED_1;
11204 kvm_update_dr7(vcpu);
11206 vcpu->arch.cr2 = 0;
11208 kvm_make_request(KVM_REQ_EVENT, vcpu);
11209 vcpu->arch.apf.msr_en_val = 0;
11210 vcpu->arch.apf.msr_int_val = 0;
11211 vcpu->arch.st.msr_val = 0;
11213 kvmclock_reset(vcpu);
11215 kvm_clear_async_pf_completion_queue(vcpu);
11216 kvm_async_pf_hash_reset(vcpu);
11217 vcpu->arch.apf.halted = false;
11219 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
11220 void *mpx_state_buffer;
11223 * To avoid have the INIT path from kvm_apic_has_events() that be
11224 * called with loaded FPU and does not let userspace fix the state.
11227 kvm_put_guest_fpu(vcpu);
11228 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
11230 if (mpx_state_buffer)
11231 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
11232 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
11234 if (mpx_state_buffer)
11235 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
11237 kvm_load_guest_fpu(vcpu);
11241 kvm_pmu_reset(vcpu);
11242 vcpu->arch.smbase = 0x30000;
11244 vcpu->arch.msr_misc_features_enables = 0;
11246 __kvm_set_xcr(vcpu, 0, XFEATURE_MASK_FP);
11247 __kvm_set_msr(vcpu, MSR_IA32_XSS, 0, true);
11250 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
11251 vcpu->arch.regs_avail = ~0;
11252 vcpu->arch.regs_dirty = ~0;
11255 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
11256 * if no CPUID match is found. Note, it's impossible to get a match at
11257 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
11258 * i.e. it'simpossible for kvm_cpuid() to find a valid entry on RESET.
11259 * But, go through the motions in case that's ever remedied.
11262 if (!kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true))
11264 kvm_rdx_write(vcpu, eax);
11266 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
11268 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
11269 kvm_rip_write(vcpu, 0xfff0);
11271 vcpu->arch.cr3 = 0;
11272 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11275 * CR0.CD/NW are set on RESET, preserved on INIT. Note, some versions
11276 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
11277 * (or qualify) that with a footnote stating that CD/NW are preserved.
11279 new_cr0 = X86_CR0_ET;
11281 new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
11283 new_cr0 |= X86_CR0_NW | X86_CR0_CD;
11285 static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
11286 static_call(kvm_x86_set_cr4)(vcpu, 0);
11287 static_call(kvm_x86_set_efer)(vcpu, 0);
11288 static_call(kvm_x86_update_exception_bitmap)(vcpu);
11291 * Reset the MMU context if paging was enabled prior to INIT (which is
11292 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET). Unlike the
11293 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
11294 * checked because it is unconditionally cleared on INIT and all other
11295 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
11296 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
11298 if (old_cr0 & X86_CR0_PG)
11299 kvm_mmu_reset_context(vcpu);
11302 * Intel's SDM states that all TLB entries are flushed on INIT. AMD's
11303 * APM states the TLBs are untouched by INIT, but it also states that
11304 * the TLBs are flushed on "External initialization of the processor."
11305 * Flush the guest TLB regardless of vendor, there is no meaningful
11306 * benefit in relying on the guest to flush the TLB immediately after
11307 * INIT. A spurious TLB flush is benign and likely negligible from a
11308 * performance perspective.
11311 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11313 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11315 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11317 struct kvm_segment cs;
11319 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
11320 cs.selector = vector << 8;
11321 cs.base = vector << 12;
11322 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
11323 kvm_rip_write(vcpu, 0);
11325 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
11327 int kvm_arch_hardware_enable(void)
11330 struct kvm_vcpu *vcpu;
11335 bool stable, backwards_tsc = false;
11337 kvm_user_return_msr_cpu_online();
11338 ret = static_call(kvm_x86_hardware_enable)();
11342 local_tsc = rdtsc();
11343 stable = !kvm_check_tsc_unstable();
11344 list_for_each_entry(kvm, &vm_list, vm_list) {
11345 kvm_for_each_vcpu(i, vcpu, kvm) {
11346 if (!stable && vcpu->cpu == smp_processor_id())
11347 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
11348 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
11349 backwards_tsc = true;
11350 if (vcpu->arch.last_host_tsc > max_tsc)
11351 max_tsc = vcpu->arch.last_host_tsc;
11357 * Sometimes, even reliable TSCs go backwards. This happens on
11358 * platforms that reset TSC during suspend or hibernate actions, but
11359 * maintain synchronization. We must compensate. Fortunately, we can
11360 * detect that condition here, which happens early in CPU bringup,
11361 * before any KVM threads can be running. Unfortunately, we can't
11362 * bring the TSCs fully up to date with real time, as we aren't yet far
11363 * enough into CPU bringup that we know how much real time has actually
11364 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
11365 * variables that haven't been updated yet.
11367 * So we simply find the maximum observed TSC above, then record the
11368 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
11369 * the adjustment will be applied. Note that we accumulate
11370 * adjustments, in case multiple suspend cycles happen before some VCPU
11371 * gets a chance to run again. In the event that no KVM threads get a
11372 * chance to run, we will miss the entire elapsed period, as we'll have
11373 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
11374 * loose cycle time. This isn't too big a deal, since the loss will be
11375 * uniform across all VCPUs (not to mention the scenario is extremely
11376 * unlikely). It is possible that a second hibernate recovery happens
11377 * much faster than a first, causing the observed TSC here to be
11378 * smaller; this would require additional padding adjustment, which is
11379 * why we set last_host_tsc to the local tsc observed here.
11381 * N.B. - this code below runs only on platforms with reliable TSC,
11382 * as that is the only way backwards_tsc is set above. Also note
11383 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
11384 * have the same delta_cyc adjustment applied if backwards_tsc
11385 * is detected. Note further, this adjustment is only done once,
11386 * as we reset last_host_tsc on all VCPUs to stop this from being
11387 * called multiple times (one for each physical CPU bringup).
11389 * Platforms with unreliable TSCs don't have to deal with this, they
11390 * will be compensated by the logic in vcpu_load, which sets the TSC to
11391 * catchup mode. This will catchup all VCPUs to real time, but cannot
11392 * guarantee that they stay in perfect synchronization.
11394 if (backwards_tsc) {
11395 u64 delta_cyc = max_tsc - local_tsc;
11396 list_for_each_entry(kvm, &vm_list, vm_list) {
11397 kvm->arch.backwards_tsc_observed = true;
11398 kvm_for_each_vcpu(i, vcpu, kvm) {
11399 vcpu->arch.tsc_offset_adjustment += delta_cyc;
11400 vcpu->arch.last_host_tsc = local_tsc;
11401 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11405 * We have to disable TSC offset matching.. if you were
11406 * booting a VM while issuing an S4 host suspend....
11407 * you may have some problem. Solving this issue is
11408 * left as an exercise to the reader.
11410 kvm->arch.last_tsc_nsec = 0;
11411 kvm->arch.last_tsc_write = 0;
11418 void kvm_arch_hardware_disable(void)
11420 static_call(kvm_x86_hardware_disable)();
11421 drop_user_return_notifiers();
11424 int kvm_arch_hardware_setup(void *opaque)
11426 struct kvm_x86_init_ops *ops = opaque;
11429 rdmsrl_safe(MSR_EFER, &host_efer);
11431 if (boot_cpu_has(X86_FEATURE_XSAVES))
11432 rdmsrl(MSR_IA32_XSS, host_xss);
11434 r = ops->hardware_setup();
11438 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11439 kvm_ops_static_call_update();
11441 if (ops->intel_pt_intr_in_guest && ops->intel_pt_intr_in_guest())
11442 kvm_guest_cbs.handle_intel_pt_intr = kvm_handle_intel_pt_intr;
11443 perf_register_guest_info_callbacks(&kvm_guest_cbs);
11445 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11448 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11449 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11450 #undef __kvm_cpu_cap_has
11452 if (kvm_has_tsc_control) {
11454 * Make sure the user can only configure tsc_khz values that
11455 * fit into a signed integer.
11456 * A min value is not calculated because it will always
11457 * be 1 on all machines.
11459 u64 max = min(0x7fffffffULL,
11460 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11461 kvm_max_guest_tsc_khz = max;
11463 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11466 kvm_init_msr_list();
11470 void kvm_arch_hardware_unsetup(void)
11472 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
11473 kvm_guest_cbs.handle_intel_pt_intr = NULL;
11475 static_call(kvm_x86_hardware_unsetup)();
11478 int kvm_arch_check_processor_compat(void *opaque)
11480 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11481 struct kvm_x86_init_ops *ops = opaque;
11483 WARN_ON(!irqs_disabled());
11485 if (__cr4_reserved_bits(cpu_has, c) !=
11486 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11489 return ops->check_processor_compatibility();
11492 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11494 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11496 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11498 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11500 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11503 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11504 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11506 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11508 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11510 vcpu->arch.l1tf_flush_l1d = true;
11511 if (pmu->version && unlikely(pmu->event_count)) {
11512 pmu->need_cleanup = true;
11513 kvm_make_request(KVM_REQ_PMU, vcpu);
11515 static_call(kvm_x86_sched_in)(vcpu, cpu);
11518 void kvm_arch_free_vm(struct kvm *kvm)
11520 kfree(to_kvm_hv(kvm)->hv_pa_pg);
11525 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11532 ret = kvm_page_track_init(kvm);
11536 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11537 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11538 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11539 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11540 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11541 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11543 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11544 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11545 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11546 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11547 &kvm->arch.irq_sources_bitmap);
11549 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11550 mutex_init(&kvm->arch.apic_map_lock);
11551 raw_spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
11553 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11554 pvclock_update_vm_gtod_copy(kvm);
11556 kvm->arch.guest_can_read_msr_platform_info = true;
11558 #if IS_ENABLED(CONFIG_HYPERV)
11559 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11560 kvm->arch.hv_root_tdp = INVALID_PAGE;
11563 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11564 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11566 kvm_apicv_init(kvm);
11567 kvm_hv_init_vm(kvm);
11568 kvm_mmu_init_vm(kvm);
11569 kvm_xen_init_vm(kvm);
11571 return static_call(kvm_x86_vm_init)(kvm);
11574 int kvm_arch_post_init_vm(struct kvm *kvm)
11576 return kvm_mmu_post_init_vm(kvm);
11579 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11582 kvm_mmu_unload(vcpu);
11586 static void kvm_free_vcpus(struct kvm *kvm)
11589 struct kvm_vcpu *vcpu;
11592 * Unpin any mmu pages first.
11594 kvm_for_each_vcpu(i, vcpu, kvm) {
11595 kvm_clear_async_pf_completion_queue(vcpu);
11596 kvm_unload_vcpu_mmu(vcpu);
11598 kvm_for_each_vcpu(i, vcpu, kvm)
11599 kvm_vcpu_destroy(vcpu);
11601 mutex_lock(&kvm->lock);
11602 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11603 kvm->vcpus[i] = NULL;
11605 atomic_set(&kvm->online_vcpus, 0);
11606 mutex_unlock(&kvm->lock);
11609 void kvm_arch_sync_events(struct kvm *kvm)
11611 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11612 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11616 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
11619 * __x86_set_memory_region: Setup KVM internal memory slot
11621 * @kvm: the kvm pointer to the VM.
11622 * @id: the slot ID to setup.
11623 * @gpa: the GPA to install the slot (unused when @size == 0).
11624 * @size: the size of the slot. Set to zero to uninstall a slot.
11626 * This function helps to setup a KVM internal memory slot. Specify
11627 * @size > 0 to install a new slot, while @size == 0 to uninstall a
11628 * slot. The return code can be one of the following:
11630 * HVA: on success (uninstall will return a bogus HVA)
11633 * The caller should always use IS_ERR() to check the return value
11634 * before use. Note, the KVM internal memory slots are guaranteed to
11635 * remain valid and unchanged until the VM is destroyed, i.e., the
11636 * GPA->HVA translation will not change. However, the HVA is a user
11637 * address, i.e. its accessibility is not guaranteed, and must be
11638 * accessed via __copy_{to,from}_user().
11640 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11644 unsigned long hva, old_npages;
11645 struct kvm_memslots *slots = kvm_memslots(kvm);
11646 struct kvm_memory_slot *slot;
11648 /* Called with kvm->slots_lock held. */
11649 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11650 return ERR_PTR_USR(-EINVAL);
11652 slot = id_to_memslot(slots, id);
11654 if (slot && slot->npages)
11655 return ERR_PTR_USR(-EEXIST);
11658 * MAP_SHARED to prevent internal slot pages from being moved
11661 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11662 MAP_SHARED | MAP_ANONYMOUS, 0);
11663 if (IS_ERR((void *)hva))
11664 return (void __user *)hva;
11666 if (!slot || !slot->npages)
11669 old_npages = slot->npages;
11670 hva = slot->userspace_addr;
11673 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11674 struct kvm_userspace_memory_region m;
11676 m.slot = id | (i << 16);
11678 m.guest_phys_addr = gpa;
11679 m.userspace_addr = hva;
11680 m.memory_size = size;
11681 r = __kvm_set_memory_region(kvm, &m);
11683 return ERR_PTR_USR(r);
11687 vm_munmap(hva, old_npages * PAGE_SIZE);
11689 return (void __user *)hva;
11691 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11693 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11695 kvm_mmu_pre_destroy_vm(kvm);
11698 void kvm_arch_destroy_vm(struct kvm *kvm)
11700 if (current->mm == kvm->mm) {
11702 * Free memory regions allocated on behalf of userspace,
11703 * unless the the memory map has changed due to process exit
11706 mutex_lock(&kvm->slots_lock);
11707 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11709 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11711 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11712 mutex_unlock(&kvm->slots_lock);
11714 static_call_cond(kvm_x86_vm_destroy)(kvm);
11715 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11716 kvm_pic_destroy(kvm);
11717 kvm_ioapic_destroy(kvm);
11718 kvm_free_vcpus(kvm);
11719 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11720 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11721 kvm_mmu_uninit_vm(kvm);
11722 kvm_page_track_cleanup(kvm);
11723 kvm_xen_destroy_vm(kvm);
11724 kvm_hv_destroy_vm(kvm);
11727 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11731 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11732 kvfree(slot->arch.rmap[i]);
11733 slot->arch.rmap[i] = NULL;
11737 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11741 memslot_rmap_free(slot);
11743 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11744 kvfree(slot->arch.lpage_info[i - 1]);
11745 slot->arch.lpage_info[i - 1] = NULL;
11748 kvm_page_track_free_memslot(slot);
11751 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
11752 unsigned long npages)
11754 const int sz = sizeof(*slot->arch.rmap[0]);
11757 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11759 int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11761 if (slot->arch.rmap[i])
11764 slot->arch.rmap[i] = __vcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11765 if (!slot->arch.rmap[i]) {
11766 memslot_rmap_free(slot);
11774 int alloc_all_memslots_rmaps(struct kvm *kvm)
11776 struct kvm_memslots *slots;
11777 struct kvm_memory_slot *slot;
11781 * Check if memslots alreday have rmaps early before acquiring
11782 * the slots_arch_lock below.
11784 if (kvm_memslots_have_rmaps(kvm))
11787 mutex_lock(&kvm->slots_arch_lock);
11790 * Read memslots_have_rmaps again, under the slots arch lock,
11791 * before allocating the rmaps
11793 if (kvm_memslots_have_rmaps(kvm)) {
11794 mutex_unlock(&kvm->slots_arch_lock);
11798 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11799 slots = __kvm_memslots(kvm, i);
11800 kvm_for_each_memslot(slot, slots) {
11801 r = memslot_rmap_alloc(slot, slot->npages);
11803 mutex_unlock(&kvm->slots_arch_lock);
11810 * Ensure that memslots_have_rmaps becomes true strictly after
11811 * all the rmap pointers are set.
11813 smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11814 mutex_unlock(&kvm->slots_arch_lock);
11818 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11819 struct kvm_memory_slot *slot,
11820 unsigned long npages)
11825 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
11826 * old arrays will be freed by __kvm_set_memory_region() if installing
11827 * the new memslot is successful.
11829 memset(&slot->arch, 0, sizeof(slot->arch));
11831 if (kvm_memslots_have_rmaps(kvm)) {
11832 r = memslot_rmap_alloc(slot, npages);
11837 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11838 struct kvm_lpage_info *linfo;
11839 unsigned long ugfn;
11843 lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11845 linfo = __vcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11849 slot->arch.lpage_info[i - 1] = linfo;
11851 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11852 linfo[0].disallow_lpage = 1;
11853 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11854 linfo[lpages - 1].disallow_lpage = 1;
11855 ugfn = slot->userspace_addr >> PAGE_SHIFT;
11857 * If the gfn and userspace address are not aligned wrt each
11858 * other, disable large page support for this slot.
11860 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11863 for (j = 0; j < lpages; ++j)
11864 linfo[j].disallow_lpage = 1;
11868 if (kvm_page_track_create_memslot(slot, npages))
11874 memslot_rmap_free(slot);
11876 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11877 kvfree(slot->arch.lpage_info[i - 1]);
11878 slot->arch.lpage_info[i - 1] = NULL;
11883 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11885 struct kvm_vcpu *vcpu;
11889 * memslots->generation has been incremented.
11890 * mmio generation may have reached its maximum value.
11892 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11894 /* Force re-initialization of steal_time cache */
11895 kvm_for_each_vcpu(i, vcpu, kvm)
11896 kvm_vcpu_kick(vcpu);
11899 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11900 struct kvm_memory_slot *memslot,
11901 const struct kvm_userspace_memory_region *mem,
11902 enum kvm_mr_change change)
11904 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11905 return kvm_alloc_memslot_metadata(kvm, memslot,
11906 mem->memory_size >> PAGE_SHIFT);
11911 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11913 struct kvm_arch *ka = &kvm->arch;
11915 if (!kvm_x86_ops.cpu_dirty_log_size)
11918 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11919 (!enable && --ka->cpu_dirty_logging_count == 0))
11920 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11922 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11925 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11926 struct kvm_memory_slot *old,
11927 const struct kvm_memory_slot *new,
11928 enum kvm_mr_change change)
11930 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11933 * Update CPU dirty logging if dirty logging is being toggled. This
11934 * applies to all operations.
11936 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11937 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11940 * Nothing more to do for RO slots (which can't be dirtied and can't be
11941 * made writable) or CREATE/MOVE/DELETE of a slot.
11943 * For a memslot with dirty logging disabled:
11944 * CREATE: No dirty mappings will already exist.
11945 * MOVE/DELETE: The old mappings will already have been cleaned up by
11946 * kvm_arch_flush_shadow_memslot()
11948 * For a memslot with dirty logging enabled:
11949 * CREATE: No shadow pages exist, thus nothing to write-protect
11950 * and no dirty bits to clear.
11951 * MOVE/DELETE: The old mappings will already have been cleaned up by
11952 * kvm_arch_flush_shadow_memslot().
11954 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11958 * READONLY and non-flags changes were filtered out above, and the only
11959 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11960 * logging isn't being toggled on or off.
11962 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11965 if (!log_dirty_pages) {
11967 * Dirty logging tracks sptes in 4k granularity, meaning that
11968 * large sptes have to be split. If live migration succeeds,
11969 * the guest in the source machine will be destroyed and large
11970 * sptes will be created in the destination. However, if the
11971 * guest continues to run in the source machine (for example if
11972 * live migration fails), small sptes will remain around and
11973 * cause bad performance.
11975 * Scan sptes if dirty logging has been stopped, dropping those
11976 * which can be collapsed into a single large-page spte. Later
11977 * page faults will create the large-page sptes.
11979 kvm_mmu_zap_collapsible_sptes(kvm, new);
11982 * Initially-all-set does not require write protecting any page,
11983 * because they're all assumed to be dirty.
11985 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11988 if (kvm_x86_ops.cpu_dirty_log_size) {
11989 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11990 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11992 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11997 void kvm_arch_commit_memory_region(struct kvm *kvm,
11998 const struct kvm_userspace_memory_region *mem,
11999 struct kvm_memory_slot *old,
12000 const struct kvm_memory_slot *new,
12001 enum kvm_mr_change change)
12003 if (!kvm->arch.n_requested_mmu_pages)
12004 kvm_mmu_change_mmu_pages(kvm,
12005 kvm_mmu_calculate_default_mmu_pages(kvm));
12007 kvm_mmu_slot_apply_flags(kvm, old, new, change);
12009 /* Free the arrays associated with the old memslot. */
12010 if (change == KVM_MR_MOVE)
12011 kvm_arch_free_memslot(kvm, old);
12014 void kvm_arch_flush_shadow_all(struct kvm *kvm)
12016 kvm_mmu_zap_all(kvm);
12019 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
12020 struct kvm_memory_slot *slot)
12022 kvm_page_track_flush_slot(kvm, slot);
12025 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
12027 return (is_guest_mode(vcpu) &&
12028 kvm_x86_ops.guest_apic_has_interrupt &&
12029 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
12032 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
12034 if (!list_empty_careful(&vcpu->async_pf.done))
12037 if (kvm_apic_has_events(vcpu))
12040 if (vcpu->arch.pv.pv_unhalted)
12043 if (vcpu->arch.exception.pending)
12046 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12047 (vcpu->arch.nmi_pending &&
12048 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
12051 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
12052 (vcpu->arch.smi_pending &&
12053 static_call(kvm_x86_smi_allowed)(vcpu, false)))
12056 if (kvm_arch_interrupt_allowed(vcpu) &&
12057 (kvm_cpu_has_interrupt(vcpu) ||
12058 kvm_guest_apic_has_interrupt(vcpu)))
12061 if (kvm_hv_has_stimer_pending(vcpu))
12064 if (is_guest_mode(vcpu) &&
12065 kvm_x86_ops.nested_ops->hv_timer_pending &&
12066 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
12072 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
12074 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
12077 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
12079 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
12085 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
12087 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
12090 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12091 kvm_test_request(KVM_REQ_SMI, vcpu) ||
12092 kvm_test_request(KVM_REQ_EVENT, vcpu))
12095 return kvm_arch_dy_has_pending_interrupt(vcpu);
12098 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
12100 if (vcpu->arch.guest_state_protected)
12103 return vcpu->arch.preempted_in_kernel;
12106 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
12108 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
12111 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
12113 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
12116 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
12118 /* Can't read the RIP when guest state is protected, just return 0 */
12119 if (vcpu->arch.guest_state_protected)
12122 if (is_64_bit_mode(vcpu))
12123 return kvm_rip_read(vcpu);
12124 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
12125 kvm_rip_read(vcpu));
12127 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
12129 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
12131 return kvm_get_linear_rip(vcpu) == linear_rip;
12133 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
12135 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
12137 unsigned long rflags;
12139 rflags = static_call(kvm_x86_get_rflags)(vcpu);
12140 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
12141 rflags &= ~X86_EFLAGS_TF;
12144 EXPORT_SYMBOL_GPL(kvm_get_rflags);
12146 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12148 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
12149 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
12150 rflags |= X86_EFLAGS_TF;
12151 static_call(kvm_x86_set_rflags)(vcpu, rflags);
12154 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12156 __kvm_set_rflags(vcpu, rflags);
12157 kvm_make_request(KVM_REQ_EVENT, vcpu);
12159 EXPORT_SYMBOL_GPL(kvm_set_rflags);
12161 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
12165 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
12169 r = kvm_mmu_reload(vcpu);
12173 if (!vcpu->arch.mmu->direct_map &&
12174 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
12177 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
12180 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
12182 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
12184 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
12187 static inline u32 kvm_async_pf_next_probe(u32 key)
12189 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
12192 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12194 u32 key = kvm_async_pf_hash_fn(gfn);
12196 while (vcpu->arch.apf.gfns[key] != ~0)
12197 key = kvm_async_pf_next_probe(key);
12199 vcpu->arch.apf.gfns[key] = gfn;
12202 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
12205 u32 key = kvm_async_pf_hash_fn(gfn);
12207 for (i = 0; i < ASYNC_PF_PER_VCPU &&
12208 (vcpu->arch.apf.gfns[key] != gfn &&
12209 vcpu->arch.apf.gfns[key] != ~0); i++)
12210 key = kvm_async_pf_next_probe(key);
12215 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12217 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
12220 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12224 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
12226 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
12230 vcpu->arch.apf.gfns[i] = ~0;
12232 j = kvm_async_pf_next_probe(j);
12233 if (vcpu->arch.apf.gfns[j] == ~0)
12235 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
12237 * k lies cyclically in ]i,j]
12239 * |....j i.k.| or |.k..j i...|
12241 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
12242 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
12247 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
12249 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
12251 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
12255 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
12257 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12259 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12260 &token, offset, sizeof(token));
12263 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
12265 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12268 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12269 &val, offset, sizeof(val)))
12275 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
12277 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
12280 if (!kvm_pv_async_pf_enabled(vcpu) ||
12281 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
12287 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
12289 if (unlikely(!lapic_in_kernel(vcpu) ||
12290 kvm_event_needs_reinjection(vcpu) ||
12291 vcpu->arch.exception.pending))
12294 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
12298 * If interrupts are off we cannot even use an artificial
12301 return kvm_arch_interrupt_allowed(vcpu);
12304 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
12305 struct kvm_async_pf *work)
12307 struct x86_exception fault;
12309 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
12310 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
12312 if (kvm_can_deliver_async_pf(vcpu) &&
12313 !apf_put_user_notpresent(vcpu)) {
12314 fault.vector = PF_VECTOR;
12315 fault.error_code_valid = true;
12316 fault.error_code = 0;
12317 fault.nested_page_fault = false;
12318 fault.address = work->arch.token;
12319 fault.async_page_fault = true;
12320 kvm_inject_page_fault(vcpu, &fault);
12324 * It is not possible to deliver a paravirtualized asynchronous
12325 * page fault, but putting the guest in an artificial halt state
12326 * can be beneficial nevertheless: if an interrupt arrives, we
12327 * can deliver it timely and perhaps the guest will schedule
12328 * another process. When the instruction that triggered a page
12329 * fault is retried, hopefully the page will be ready in the host.
12331 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
12336 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
12337 struct kvm_async_pf *work)
12339 struct kvm_lapic_irq irq = {
12340 .delivery_mode = APIC_DM_FIXED,
12341 .vector = vcpu->arch.apf.vec
12344 if (work->wakeup_all)
12345 work->arch.token = ~0; /* broadcast wakeup */
12347 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12348 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12350 if ((work->wakeup_all || work->notpresent_injected) &&
12351 kvm_pv_async_pf_enabled(vcpu) &&
12352 !apf_put_user_ready(vcpu, work->arch.token)) {
12353 vcpu->arch.apf.pageready_pending = true;
12354 kvm_apic_set_irq(vcpu, &irq, NULL);
12357 vcpu->arch.apf.halted = false;
12358 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
12361 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
12363 kvm_make_request(KVM_REQ_APF_READY, vcpu);
12364 if (!vcpu->arch.apf.pageready_pending)
12365 kvm_vcpu_kick(vcpu);
12368 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
12370 if (!kvm_pv_async_pf_enabled(vcpu))
12373 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
12376 void kvm_arch_start_assignment(struct kvm *kvm)
12378 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
12379 static_call_cond(kvm_x86_start_assignment)(kvm);
12381 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
12383 void kvm_arch_end_assignment(struct kvm *kvm)
12385 atomic_dec(&kvm->arch.assigned_device_count);
12387 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
12389 bool noinstr kvm_arch_has_assigned_device(struct kvm *kvm)
12391 return arch_atomic_read(&kvm->arch.assigned_device_count);
12393 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
12395 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12397 atomic_inc(&kvm->arch.noncoherent_dma_count);
12399 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12401 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12403 atomic_dec(&kvm->arch.noncoherent_dma_count);
12405 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12407 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12409 return atomic_read(&kvm->arch.noncoherent_dma_count);
12411 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12413 bool kvm_arch_has_irq_bypass(void)
12418 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12419 struct irq_bypass_producer *prod)
12421 struct kvm_kernel_irqfd *irqfd =
12422 container_of(cons, struct kvm_kernel_irqfd, consumer);
12425 irqfd->producer = prod;
12426 kvm_arch_start_assignment(irqfd->kvm);
12427 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
12428 prod->irq, irqfd->gsi, 1);
12431 kvm_arch_end_assignment(irqfd->kvm);
12436 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12437 struct irq_bypass_producer *prod)
12440 struct kvm_kernel_irqfd *irqfd =
12441 container_of(cons, struct kvm_kernel_irqfd, consumer);
12443 WARN_ON(irqfd->producer != prod);
12444 irqfd->producer = NULL;
12447 * When producer of consumer is unregistered, we change back to
12448 * remapped mode, so we can re-use the current implementation
12449 * when the irq is masked/disabled or the consumer side (KVM
12450 * int this case doesn't want to receive the interrupts.
12452 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12454 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12455 " fails: %d\n", irqfd->consumer.token, ret);
12457 kvm_arch_end_assignment(irqfd->kvm);
12460 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12461 uint32_t guest_irq, bool set)
12463 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12466 bool kvm_vector_hashing_enabled(void)
12468 return vector_hashing;
12471 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12473 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12475 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12478 int kvm_spec_ctrl_test_value(u64 value)
12481 * test that setting IA32_SPEC_CTRL to given value
12482 * is allowed by the host processor
12486 unsigned long flags;
12489 local_irq_save(flags);
12491 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12493 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12496 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12498 local_irq_restore(flags);
12502 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12504 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12506 struct x86_exception fault;
12507 u32 access = error_code &
12508 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12510 if (!(error_code & PFERR_PRESENT_MASK) ||
12511 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12513 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12514 * tables probably do not match the TLB. Just proceed
12515 * with the error code that the processor gave.
12517 fault.vector = PF_VECTOR;
12518 fault.error_code_valid = true;
12519 fault.error_code = error_code;
12520 fault.nested_page_fault = false;
12521 fault.address = gva;
12523 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12525 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12528 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12529 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12530 * indicates whether exit to userspace is needed.
12532 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12533 struct x86_exception *e)
12535 if (r == X86EMUL_PROPAGATE_FAULT) {
12536 kvm_inject_emulated_page_fault(vcpu, e);
12541 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12542 * while handling a VMX instruction KVM could've handled the request
12543 * correctly by exiting to userspace and performing I/O but there
12544 * doesn't seem to be a real use-case behind such requests, just return
12545 * KVM_EXIT_INTERNAL_ERROR for now.
12547 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
12548 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
12549 vcpu->run->internal.ndata = 0;
12553 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12555 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12558 struct x86_exception e;
12565 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12566 if (r != X86EMUL_CONTINUE)
12567 return kvm_handle_memory_failure(vcpu, r, &e);
12569 if (operand.pcid >> 12 != 0) {
12570 kvm_inject_gp(vcpu, 0);
12574 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12577 case INVPCID_TYPE_INDIV_ADDR:
12578 if ((!pcid_enabled && (operand.pcid != 0)) ||
12579 is_noncanonical_address(operand.gla, vcpu)) {
12580 kvm_inject_gp(vcpu, 0);
12583 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12584 return kvm_skip_emulated_instruction(vcpu);
12586 case INVPCID_TYPE_SINGLE_CTXT:
12587 if (!pcid_enabled && (operand.pcid != 0)) {
12588 kvm_inject_gp(vcpu, 0);
12592 kvm_invalidate_pcid(vcpu, operand.pcid);
12593 return kvm_skip_emulated_instruction(vcpu);
12595 case INVPCID_TYPE_ALL_NON_GLOBAL:
12597 * Currently, KVM doesn't mark global entries in the shadow
12598 * page tables, so a non-global flush just degenerates to a
12599 * global flush. If needed, we could optimize this later by
12600 * keeping track of global entries in shadow page tables.
12604 case INVPCID_TYPE_ALL_INCL_GLOBAL:
12605 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12606 return kvm_skip_emulated_instruction(vcpu);
12609 BUG(); /* We have already checked above that type <= 3 */
12612 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12614 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12616 struct kvm_run *run = vcpu->run;
12617 struct kvm_mmio_fragment *frag;
12620 BUG_ON(!vcpu->mmio_needed);
12622 /* Complete previous fragment */
12623 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12624 len = min(8u, frag->len);
12625 if (!vcpu->mmio_is_write)
12626 memcpy(frag->data, run->mmio.data, len);
12628 if (frag->len <= 8) {
12629 /* Switch to the next fragment. */
12631 vcpu->mmio_cur_fragment++;
12633 /* Go forward to the next mmio piece. */
12639 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12640 vcpu->mmio_needed = 0;
12642 // VMG change, at this point, we're always done
12643 // RIP has already been advanced
12647 // More MMIO is needed
12648 run->mmio.phys_addr = frag->gpa;
12649 run->mmio.len = min(8u, frag->len);
12650 run->mmio.is_write = vcpu->mmio_is_write;
12651 if (run->mmio.is_write)
12652 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12653 run->exit_reason = KVM_EXIT_MMIO;
12655 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12660 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12664 struct kvm_mmio_fragment *frag;
12669 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12670 if (handled == bytes)
12677 /*TODO: Check if need to increment number of frags */
12678 frag = vcpu->mmio_fragments;
12679 vcpu->mmio_nr_fragments = 1;
12684 vcpu->mmio_needed = 1;
12685 vcpu->mmio_cur_fragment = 0;
12687 vcpu->run->mmio.phys_addr = gpa;
12688 vcpu->run->mmio.len = min(8u, frag->len);
12689 vcpu->run->mmio.is_write = 1;
12690 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12691 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12693 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12697 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12699 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12703 struct kvm_mmio_fragment *frag;
12708 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12709 if (handled == bytes)
12716 /*TODO: Check if need to increment number of frags */
12717 frag = vcpu->mmio_fragments;
12718 vcpu->mmio_nr_fragments = 1;
12723 vcpu->mmio_needed = 1;
12724 vcpu->mmio_cur_fragment = 0;
12726 vcpu->run->mmio.phys_addr = gpa;
12727 vcpu->run->mmio.len = min(8u, frag->len);
12728 vcpu->run->mmio.is_write = 0;
12729 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12731 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12735 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12737 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12738 unsigned int port);
12740 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
12742 int size = vcpu->arch.pio.size;
12743 int port = vcpu->arch.pio.port;
12745 vcpu->arch.pio.count = 0;
12746 if (vcpu->arch.sev_pio_count)
12747 return kvm_sev_es_outs(vcpu, size, port);
12751 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12755 unsigned int count =
12756 min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12757 int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
12759 /* memcpy done already by emulator_pio_out. */
12760 vcpu->arch.sev_pio_count -= count;
12761 vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12765 /* Emulation done by the kernel. */
12766 if (!vcpu->arch.sev_pio_count)
12770 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
12774 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12775 unsigned int port);
12777 static void advance_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12779 unsigned count = vcpu->arch.pio.count;
12780 complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
12781 vcpu->arch.sev_pio_count -= count;
12782 vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12785 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12787 int size = vcpu->arch.pio.size;
12788 int port = vcpu->arch.pio.port;
12790 advance_sev_es_emulated_ins(vcpu);
12791 if (vcpu->arch.sev_pio_count)
12792 return kvm_sev_es_ins(vcpu, size, port);
12796 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12800 unsigned int count =
12801 min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12802 if (!__emulator_pio_in(vcpu, size, port, count))
12805 /* Emulation done by the kernel. */
12806 advance_sev_es_emulated_ins(vcpu);
12807 if (!vcpu->arch.sev_pio_count)
12811 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12815 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12816 unsigned int port, void *data, unsigned int count,
12819 vcpu->arch.sev_pio_data = data;
12820 vcpu->arch.sev_pio_count = count;
12821 return in ? kvm_sev_es_ins(vcpu, size, port)
12822 : kvm_sev_es_outs(vcpu, size, port);
12824 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12826 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12827 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12828 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12829 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12830 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12831 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12832 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12833 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12834 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12835 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12836 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12837 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12838 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12839 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12840 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12841 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12842 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12843 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12844 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12845 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12846 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12847 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12848 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12849 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12850 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12851 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12852 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
12854 static int __init kvm_x86_init(void)
12856 kvm_mmu_x86_module_init();
12857 mitigate_smt_rsb &= boot_cpu_has_bug(X86_BUG_SMT_RSB) && cpu_smt_possible();
12860 module_init(kvm_x86_init);
12862 static void __exit kvm_x86_exit(void)
12865 * If module_init() is implemented, module_exit() must also be
12866 * implemented to allow module unload.
12869 module_exit(kvm_x86_exit);