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);
188 * Restoring the host value for MSRs that are only consumed when running in
189 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
190 * returns to userspace, i.e. the kernel can run with the guest's value.
192 #define KVM_MAX_NR_USER_RETURN_MSRS 16
194 struct kvm_user_return_msrs {
195 struct user_return_notifier urn;
197 struct kvm_user_return_msr_values {
200 } values[KVM_MAX_NR_USER_RETURN_MSRS];
203 u32 __read_mostly kvm_nr_uret_msrs;
204 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
205 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
206 static struct kvm_user_return_msrs __percpu *user_return_msrs;
208 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
209 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
210 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
211 | XFEATURE_MASK_PKRU)
213 u64 __read_mostly host_efer;
214 EXPORT_SYMBOL_GPL(host_efer);
216 bool __read_mostly allow_smaller_maxphyaddr = 0;
217 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
219 bool __read_mostly enable_apicv = true;
220 EXPORT_SYMBOL_GPL(enable_apicv);
222 u64 __read_mostly host_xss;
223 EXPORT_SYMBOL_GPL(host_xss);
224 u64 __read_mostly supported_xss;
225 EXPORT_SYMBOL_GPL(supported_xss);
227 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
228 KVM_GENERIC_VM_STATS(),
229 STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
230 STATS_DESC_COUNTER(VM, mmu_pte_write),
231 STATS_DESC_COUNTER(VM, mmu_pde_zapped),
232 STATS_DESC_COUNTER(VM, mmu_flooded),
233 STATS_DESC_COUNTER(VM, mmu_recycled),
234 STATS_DESC_COUNTER(VM, mmu_cache_miss),
235 STATS_DESC_ICOUNTER(VM, mmu_unsync),
236 STATS_DESC_ICOUNTER(VM, lpages),
237 STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
238 STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
239 STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
241 static_assert(ARRAY_SIZE(kvm_vm_stats_desc) ==
242 sizeof(struct kvm_vm_stat) / sizeof(u64));
244 const struct kvm_stats_header kvm_vm_stats_header = {
245 .name_size = KVM_STATS_NAME_SIZE,
246 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
247 .id_offset = sizeof(struct kvm_stats_header),
248 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
249 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
250 sizeof(kvm_vm_stats_desc),
253 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
254 KVM_GENERIC_VCPU_STATS(),
255 STATS_DESC_COUNTER(VCPU, pf_fixed),
256 STATS_DESC_COUNTER(VCPU, pf_guest),
257 STATS_DESC_COUNTER(VCPU, tlb_flush),
258 STATS_DESC_COUNTER(VCPU, invlpg),
259 STATS_DESC_COUNTER(VCPU, exits),
260 STATS_DESC_COUNTER(VCPU, io_exits),
261 STATS_DESC_COUNTER(VCPU, mmio_exits),
262 STATS_DESC_COUNTER(VCPU, signal_exits),
263 STATS_DESC_COUNTER(VCPU, irq_window_exits),
264 STATS_DESC_COUNTER(VCPU, nmi_window_exits),
265 STATS_DESC_COUNTER(VCPU, l1d_flush),
266 STATS_DESC_COUNTER(VCPU, halt_exits),
267 STATS_DESC_COUNTER(VCPU, request_irq_exits),
268 STATS_DESC_COUNTER(VCPU, irq_exits),
269 STATS_DESC_COUNTER(VCPU, host_state_reload),
270 STATS_DESC_COUNTER(VCPU, fpu_reload),
271 STATS_DESC_COUNTER(VCPU, insn_emulation),
272 STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
273 STATS_DESC_COUNTER(VCPU, hypercalls),
274 STATS_DESC_COUNTER(VCPU, irq_injections),
275 STATS_DESC_COUNTER(VCPU, nmi_injections),
276 STATS_DESC_COUNTER(VCPU, req_event),
277 STATS_DESC_COUNTER(VCPU, nested_run),
278 STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
279 STATS_DESC_COUNTER(VCPU, directed_yield_successful),
280 STATS_DESC_ICOUNTER(VCPU, guest_mode)
282 static_assert(ARRAY_SIZE(kvm_vcpu_stats_desc) ==
283 sizeof(struct kvm_vcpu_stat) / sizeof(u64));
285 const struct kvm_stats_header kvm_vcpu_stats_header = {
286 .name_size = KVM_STATS_NAME_SIZE,
287 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
288 .id_offset = sizeof(struct kvm_stats_header),
289 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
290 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
291 sizeof(kvm_vcpu_stats_desc),
294 u64 __read_mostly host_xcr0;
295 u64 __read_mostly supported_xcr0;
296 EXPORT_SYMBOL_GPL(supported_xcr0);
298 static struct kmem_cache *x86_fpu_cache;
300 static struct kmem_cache *x86_emulator_cache;
303 * When called, it means the previous get/set msr reached an invalid msr.
304 * Return true if we want to ignore/silent this failed msr access.
306 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
308 const char *op = write ? "wrmsr" : "rdmsr";
311 if (report_ignored_msrs)
312 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
317 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
323 static struct kmem_cache *kvm_alloc_emulator_cache(void)
325 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
326 unsigned int size = sizeof(struct x86_emulate_ctxt);
328 return kmem_cache_create_usercopy("x86_emulator", size,
329 __alignof__(struct x86_emulate_ctxt),
330 SLAB_ACCOUNT, useroffset,
331 size - useroffset, NULL);
334 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
336 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
339 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
340 vcpu->arch.apf.gfns[i] = ~0;
343 static void kvm_on_user_return(struct user_return_notifier *urn)
346 struct kvm_user_return_msrs *msrs
347 = container_of(urn, struct kvm_user_return_msrs, urn);
348 struct kvm_user_return_msr_values *values;
352 * Disabling irqs at this point since the following code could be
353 * interrupted and executed through kvm_arch_hardware_disable()
355 local_irq_save(flags);
356 if (msrs->registered) {
357 msrs->registered = false;
358 user_return_notifier_unregister(urn);
360 local_irq_restore(flags);
361 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
362 values = &msrs->values[slot];
363 if (values->host != values->curr) {
364 wrmsrl(kvm_uret_msrs_list[slot], values->host);
365 values->curr = values->host;
370 static int kvm_probe_user_return_msr(u32 msr)
376 ret = rdmsrl_safe(msr, &val);
379 ret = wrmsrl_safe(msr, val);
385 int kvm_add_user_return_msr(u32 msr)
387 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
389 if (kvm_probe_user_return_msr(msr))
392 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
393 return kvm_nr_uret_msrs++;
395 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
397 int kvm_find_user_return_msr(u32 msr)
401 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
402 if (kvm_uret_msrs_list[i] == msr)
407 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
409 static void kvm_user_return_msr_cpu_online(void)
411 unsigned int cpu = smp_processor_id();
412 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
416 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
417 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
418 msrs->values[i].host = value;
419 msrs->values[i].curr = value;
423 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
425 unsigned int cpu = smp_processor_id();
426 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
429 value = (value & mask) | (msrs->values[slot].host & ~mask);
430 if (value == msrs->values[slot].curr)
432 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
436 msrs->values[slot].curr = value;
437 if (!msrs->registered) {
438 msrs->urn.on_user_return = kvm_on_user_return;
439 user_return_notifier_register(&msrs->urn);
440 msrs->registered = true;
444 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
446 static void drop_user_return_notifiers(void)
448 unsigned int cpu = smp_processor_id();
449 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
451 if (msrs->registered)
452 kvm_on_user_return(&msrs->urn);
455 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
457 return vcpu->arch.apic_base;
459 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
461 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
463 return kvm_apic_mode(kvm_get_apic_base(vcpu));
465 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
467 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
469 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
470 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
471 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
472 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
474 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
476 if (!msr_info->host_initiated) {
477 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
479 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
483 kvm_lapic_set_base(vcpu, msr_info->data);
484 kvm_recalculate_apic_map(vcpu->kvm);
487 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
489 asmlinkage __visible noinstr void kvm_spurious_fault(void)
491 /* Fault while not rebooting. We want the trace. */
492 BUG_ON(!kvm_rebooting);
494 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
496 #define EXCPT_BENIGN 0
497 #define EXCPT_CONTRIBUTORY 1
500 static int exception_class(int vector)
510 return EXCPT_CONTRIBUTORY;
517 #define EXCPT_FAULT 0
519 #define EXCPT_ABORT 2
520 #define EXCPT_INTERRUPT 3
522 static int exception_type(int vector)
526 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
527 return EXCPT_INTERRUPT;
531 /* #DB is trap, as instruction watchpoints are handled elsewhere */
532 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
535 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
538 /* Reserved exceptions will result in fault */
542 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
544 unsigned nr = vcpu->arch.exception.nr;
545 bool has_payload = vcpu->arch.exception.has_payload;
546 unsigned long payload = vcpu->arch.exception.payload;
554 * "Certain debug exceptions may clear bit 0-3. The
555 * remaining contents of the DR6 register are never
556 * cleared by the processor".
558 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
560 * In order to reflect the #DB exception payload in guest
561 * dr6, three components need to be considered: active low
562 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
564 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
565 * In the target guest dr6:
566 * FIXED_1 bits should always be set.
567 * Active low bits should be cleared if 1-setting in payload.
568 * Active high bits should be set if 1-setting in payload.
570 * Note, the payload is compatible with the pending debug
571 * exceptions/exit qualification under VMX, that active_low bits
572 * are active high in payload.
573 * So they need to be flipped for DR6.
575 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
576 vcpu->arch.dr6 |= payload;
577 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
580 * The #DB payload is defined as compatible with the 'pending
581 * debug exceptions' field under VMX, not DR6. While bit 12 is
582 * defined in the 'pending debug exceptions' field (enabled
583 * breakpoint), it is reserved and must be zero in DR6.
585 vcpu->arch.dr6 &= ~BIT(12);
588 vcpu->arch.cr2 = payload;
592 vcpu->arch.exception.has_payload = false;
593 vcpu->arch.exception.payload = 0;
595 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
597 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
598 unsigned nr, bool has_error, u32 error_code,
599 bool has_payload, unsigned long payload, bool reinject)
604 kvm_make_request(KVM_REQ_EVENT, vcpu);
606 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
610 * On vmentry, vcpu->arch.exception.pending is only
611 * true if an event injection was blocked by
612 * nested_run_pending. In that case, however,
613 * vcpu_enter_guest requests an immediate exit,
614 * and the guest shouldn't proceed far enough to
617 WARN_ON_ONCE(vcpu->arch.exception.pending);
618 vcpu->arch.exception.injected = true;
619 if (WARN_ON_ONCE(has_payload)) {
621 * A reinjected event has already
622 * delivered its payload.
628 vcpu->arch.exception.pending = true;
629 vcpu->arch.exception.injected = false;
631 vcpu->arch.exception.has_error_code = has_error;
632 vcpu->arch.exception.nr = nr;
633 vcpu->arch.exception.error_code = error_code;
634 vcpu->arch.exception.has_payload = has_payload;
635 vcpu->arch.exception.payload = payload;
636 if (!is_guest_mode(vcpu))
637 kvm_deliver_exception_payload(vcpu);
641 /* to check exception */
642 prev_nr = vcpu->arch.exception.nr;
643 if (prev_nr == DF_VECTOR) {
644 /* triple fault -> shutdown */
645 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
648 class1 = exception_class(prev_nr);
649 class2 = exception_class(nr);
650 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
651 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
653 * Generate double fault per SDM Table 5-5. Set
654 * exception.pending = true so that the double fault
655 * can trigger a nested vmexit.
657 vcpu->arch.exception.pending = true;
658 vcpu->arch.exception.injected = false;
659 vcpu->arch.exception.has_error_code = true;
660 vcpu->arch.exception.nr = DF_VECTOR;
661 vcpu->arch.exception.error_code = 0;
662 vcpu->arch.exception.has_payload = false;
663 vcpu->arch.exception.payload = 0;
665 /* replace previous exception with a new one in a hope
666 that instruction re-execution will regenerate lost
671 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
673 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
675 EXPORT_SYMBOL_GPL(kvm_queue_exception);
677 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
679 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
681 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
683 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
684 unsigned long payload)
686 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
688 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
690 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
691 u32 error_code, unsigned long payload)
693 kvm_multiple_exception(vcpu, nr, true, error_code,
694 true, payload, false);
697 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
700 kvm_inject_gp(vcpu, 0);
702 return kvm_skip_emulated_instruction(vcpu);
706 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
708 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
710 ++vcpu->stat.pf_guest;
711 vcpu->arch.exception.nested_apf =
712 is_guest_mode(vcpu) && fault->async_page_fault;
713 if (vcpu->arch.exception.nested_apf) {
714 vcpu->arch.apf.nested_apf_token = fault->address;
715 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
717 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
721 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
723 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
724 struct x86_exception *fault)
726 struct kvm_mmu *fault_mmu;
727 WARN_ON_ONCE(fault->vector != PF_VECTOR);
729 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
733 * Invalidate the TLB entry for the faulting address, if it exists,
734 * else the access will fault indefinitely (and to emulate hardware).
736 if ((fault->error_code & PFERR_PRESENT_MASK) &&
737 !(fault->error_code & PFERR_RSVD_MASK))
738 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
739 fault_mmu->root_hpa);
741 fault_mmu->inject_page_fault(vcpu, fault);
742 return fault->nested_page_fault;
744 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
746 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
748 atomic_inc(&vcpu->arch.nmi_queued);
749 kvm_make_request(KVM_REQ_NMI, vcpu);
751 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
753 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
755 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
757 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
759 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
761 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
763 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
766 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
767 * a #GP and return false.
769 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
771 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
773 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
776 EXPORT_SYMBOL_GPL(kvm_require_cpl);
778 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
780 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
783 kvm_queue_exception(vcpu, UD_VECTOR);
786 EXPORT_SYMBOL_GPL(kvm_require_dr);
789 * This function will be used to read from the physical memory of the currently
790 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
791 * can read from guest physical or from the guest's guest physical memory.
793 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
794 gfn_t ngfn, void *data, int offset, int len,
797 struct x86_exception exception;
801 ngpa = gfn_to_gpa(ngfn);
802 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
803 if (real_gfn == UNMAPPED_GVA)
806 real_gfn = gpa_to_gfn(real_gfn);
808 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
810 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
812 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
814 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
818 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
820 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
822 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
823 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
826 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
828 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
829 offset * sizeof(u64), sizeof(pdpte),
830 PFERR_USER_MASK|PFERR_WRITE_MASK);
835 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
836 if ((pdpte[i] & PT_PRESENT_MASK) &&
837 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
844 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
845 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
846 vcpu->arch.pdptrs_from_userspace = false;
852 EXPORT_SYMBOL_GPL(load_pdptrs);
854 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
856 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
857 kvm_clear_async_pf_completion_queue(vcpu);
858 kvm_async_pf_hash_reset(vcpu);
861 if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
862 kvm_mmu_reset_context(vcpu);
864 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
865 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
866 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
867 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
869 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
871 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
873 unsigned long old_cr0 = kvm_read_cr0(vcpu);
874 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
879 if (cr0 & 0xffffffff00000000UL)
883 cr0 &= ~CR0_RESERVED_BITS;
885 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
888 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
892 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
893 (cr0 & X86_CR0_PG)) {
898 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
903 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
904 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
905 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
908 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
911 static_call(kvm_x86_set_cr0)(vcpu, cr0);
913 kvm_post_set_cr0(vcpu, old_cr0, cr0);
917 EXPORT_SYMBOL_GPL(kvm_set_cr0);
919 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
921 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
923 EXPORT_SYMBOL_GPL(kvm_lmsw);
925 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
927 if (vcpu->arch.guest_state_protected)
930 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
932 if (vcpu->arch.xcr0 != host_xcr0)
933 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
935 if (vcpu->arch.xsaves_enabled &&
936 vcpu->arch.ia32_xss != host_xss)
937 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
940 if (static_cpu_has(X86_FEATURE_PKU) &&
941 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
942 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
943 vcpu->arch.pkru != vcpu->arch.host_pkru)
944 write_pkru(vcpu->arch.pkru);
946 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
948 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
950 if (vcpu->arch.guest_state_protected)
953 if (static_cpu_has(X86_FEATURE_PKU) &&
954 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
955 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
956 vcpu->arch.pkru = rdpkru();
957 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
958 write_pkru(vcpu->arch.host_pkru);
961 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
963 if (vcpu->arch.xcr0 != host_xcr0)
964 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
966 if (vcpu->arch.xsaves_enabled &&
967 vcpu->arch.ia32_xss != host_xss)
968 wrmsrl(MSR_IA32_XSS, host_xss);
972 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
974 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
977 u64 old_xcr0 = vcpu->arch.xcr0;
980 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
981 if (index != XCR_XFEATURE_ENABLED_MASK)
983 if (!(xcr0 & XFEATURE_MASK_FP))
985 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
989 * Do not allow the guest to set bits that we do not support
990 * saving. However, xcr0 bit 0 is always set, even if the
991 * emulated CPU does not support XSAVE (see fx_init).
993 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
994 if (xcr0 & ~valid_bits)
997 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
998 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1001 if (xcr0 & XFEATURE_MASK_AVX512) {
1002 if (!(xcr0 & XFEATURE_MASK_YMM))
1004 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1007 vcpu->arch.xcr0 = xcr0;
1009 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1010 kvm_update_cpuid_runtime(vcpu);
1014 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1016 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1017 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1018 kvm_inject_gp(vcpu, 0);
1022 return kvm_skip_emulated_instruction(vcpu);
1024 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1026 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1028 if (cr4 & cr4_reserved_bits)
1031 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1034 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1036 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1038 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1040 if (((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS) ||
1041 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1042 kvm_mmu_reset_context(vcpu);
1044 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1046 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1048 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1049 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1052 if (!kvm_is_valid_cr4(vcpu, cr4))
1055 if (is_long_mode(vcpu)) {
1056 if (!(cr4 & X86_CR4_PAE))
1058 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1060 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1061 && ((cr4 ^ old_cr4) & pdptr_bits)
1062 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1063 kvm_read_cr3(vcpu)))
1066 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1067 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1070 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1071 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1075 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1077 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1081 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1083 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1085 struct kvm_mmu *mmu = vcpu->arch.mmu;
1086 unsigned long roots_to_free = 0;
1090 * If neither the current CR3 nor any of the prev_roots use the given
1091 * PCID, then nothing needs to be done here because a resync will
1092 * happen anyway before switching to any other CR3.
1094 if (kvm_get_active_pcid(vcpu) == pcid) {
1095 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1096 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1099 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1100 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1101 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1103 kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1106 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1108 bool skip_tlb_flush = false;
1109 unsigned long pcid = 0;
1110 #ifdef CONFIG_X86_64
1111 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1114 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1115 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1116 pcid = cr3 & X86_CR3_PCID_MASK;
1120 /* PDPTRs are always reloaded for PAE paging. */
1121 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1122 goto handle_tlb_flush;
1125 * Do not condition the GPA check on long mode, this helper is used to
1126 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1127 * the current vCPU mode is accurate.
1129 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1132 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1135 if (cr3 != kvm_read_cr3(vcpu))
1136 kvm_mmu_new_pgd(vcpu, cr3);
1138 vcpu->arch.cr3 = cr3;
1139 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1143 * A load of CR3 that flushes the TLB flushes only the current PCID,
1144 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1145 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1146 * and it's impossible to use a non-zero PCID when PCID is disabled,
1147 * i.e. only PCID=0 can be relevant.
1149 if (!skip_tlb_flush)
1150 kvm_invalidate_pcid(vcpu, pcid);
1154 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1156 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1158 if (cr8 & CR8_RESERVED_BITS)
1160 if (lapic_in_kernel(vcpu))
1161 kvm_lapic_set_tpr(vcpu, cr8);
1163 vcpu->arch.cr8 = cr8;
1166 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1168 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1170 if (lapic_in_kernel(vcpu))
1171 return kvm_lapic_get_cr8(vcpu);
1173 return vcpu->arch.cr8;
1175 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1177 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1181 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1182 for (i = 0; i < KVM_NR_DB_REGS; i++)
1183 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1184 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1188 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1192 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1193 dr7 = vcpu->arch.guest_debug_dr7;
1195 dr7 = vcpu->arch.dr7;
1196 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1197 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1198 if (dr7 & DR7_BP_EN_MASK)
1199 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1201 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1203 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1205 u64 fixed = DR6_FIXED_1;
1207 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1210 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1211 fixed |= DR6_BUS_LOCK;
1215 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1217 size_t size = ARRAY_SIZE(vcpu->arch.db);
1221 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1222 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1223 vcpu->arch.eff_db[dr] = val;
1227 if (!kvm_dr6_valid(val))
1229 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1233 if (!kvm_dr7_valid(val))
1235 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1236 kvm_update_dr7(vcpu);
1242 EXPORT_SYMBOL_GPL(kvm_set_dr);
1244 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1246 size_t size = ARRAY_SIZE(vcpu->arch.db);
1250 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1254 *val = vcpu->arch.dr6;
1258 *val = vcpu->arch.dr7;
1262 EXPORT_SYMBOL_GPL(kvm_get_dr);
1264 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1266 u32 ecx = kvm_rcx_read(vcpu);
1269 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1270 kvm_inject_gp(vcpu, 0);
1274 kvm_rax_write(vcpu, (u32)data);
1275 kvm_rdx_write(vcpu, data >> 32);
1276 return kvm_skip_emulated_instruction(vcpu);
1278 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1281 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1282 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1284 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1285 * extract the supported MSRs from the related const lists.
1286 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1287 * capabilities of the host cpu. This capabilities test skips MSRs that are
1288 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1289 * may depend on host virtualization features rather than host cpu features.
1292 static const u32 msrs_to_save_all[] = {
1293 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1295 #ifdef CONFIG_X86_64
1296 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1298 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1299 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1301 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1302 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1303 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1304 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1305 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1306 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1307 MSR_IA32_UMWAIT_CONTROL,
1309 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1310 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1311 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1312 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1313 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1314 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1315 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1316 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1317 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1318 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1319 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1320 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1321 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1322 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1323 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1324 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1325 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1326 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1327 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1328 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1329 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1330 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1333 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1334 static unsigned num_msrs_to_save;
1336 static const u32 emulated_msrs_all[] = {
1337 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1338 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1339 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1340 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1341 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1342 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1343 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1345 HV_X64_MSR_VP_INDEX,
1346 HV_X64_MSR_VP_RUNTIME,
1347 HV_X64_MSR_SCONTROL,
1348 HV_X64_MSR_STIMER0_CONFIG,
1349 HV_X64_MSR_VP_ASSIST_PAGE,
1350 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1351 HV_X64_MSR_TSC_EMULATION_STATUS,
1352 HV_X64_MSR_SYNDBG_OPTIONS,
1353 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1354 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1355 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1357 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1358 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1360 MSR_IA32_TSC_ADJUST,
1361 MSR_IA32_TSC_DEADLINE,
1362 MSR_IA32_ARCH_CAPABILITIES,
1363 MSR_IA32_PERF_CAPABILITIES,
1364 MSR_IA32_MISC_ENABLE,
1365 MSR_IA32_MCG_STATUS,
1367 MSR_IA32_MCG_EXT_CTL,
1371 MSR_MISC_FEATURES_ENABLES,
1372 MSR_AMD64_VIRT_SPEC_CTRL,
1377 * The following list leaves out MSRs whose values are determined
1378 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1379 * We always support the "true" VMX control MSRs, even if the host
1380 * processor does not, so I am putting these registers here rather
1381 * than in msrs_to_save_all.
1384 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1385 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1386 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1387 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1389 MSR_IA32_VMX_CR0_FIXED0,
1390 MSR_IA32_VMX_CR4_FIXED0,
1391 MSR_IA32_VMX_VMCS_ENUM,
1392 MSR_IA32_VMX_PROCBASED_CTLS2,
1393 MSR_IA32_VMX_EPT_VPID_CAP,
1394 MSR_IA32_VMX_VMFUNC,
1397 MSR_KVM_POLL_CONTROL,
1400 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1401 static unsigned num_emulated_msrs;
1404 * List of msr numbers which are used to expose MSR-based features that
1405 * can be used by a hypervisor to validate requested CPU features.
1407 static const u32 msr_based_features_all[] = {
1409 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1410 MSR_IA32_VMX_PINBASED_CTLS,
1411 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1412 MSR_IA32_VMX_PROCBASED_CTLS,
1413 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1414 MSR_IA32_VMX_EXIT_CTLS,
1415 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1416 MSR_IA32_VMX_ENTRY_CTLS,
1418 MSR_IA32_VMX_CR0_FIXED0,
1419 MSR_IA32_VMX_CR0_FIXED1,
1420 MSR_IA32_VMX_CR4_FIXED0,
1421 MSR_IA32_VMX_CR4_FIXED1,
1422 MSR_IA32_VMX_VMCS_ENUM,
1423 MSR_IA32_VMX_PROCBASED_CTLS2,
1424 MSR_IA32_VMX_EPT_VPID_CAP,
1425 MSR_IA32_VMX_VMFUNC,
1429 MSR_IA32_ARCH_CAPABILITIES,
1430 MSR_IA32_PERF_CAPABILITIES,
1433 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1434 static unsigned int num_msr_based_features;
1436 static u64 kvm_get_arch_capabilities(void)
1440 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1441 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1444 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1445 * the nested hypervisor runs with NX huge pages. If it is not,
1446 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1447 * L1 guests, so it need not worry about its own (L2) guests.
1449 data |= ARCH_CAP_PSCHANGE_MC_NO;
1452 * If we're doing cache flushes (either "always" or "cond")
1453 * we will do one whenever the guest does a vmlaunch/vmresume.
1454 * If an outer hypervisor is doing the cache flush for us
1455 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1456 * capability to the guest too, and if EPT is disabled we're not
1457 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1458 * require a nested hypervisor to do a flush of its own.
1460 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1461 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1463 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1464 data |= ARCH_CAP_RDCL_NO;
1465 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1466 data |= ARCH_CAP_SSB_NO;
1467 if (!boot_cpu_has_bug(X86_BUG_MDS))
1468 data |= ARCH_CAP_MDS_NO;
1470 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1472 * If RTM=0 because the kernel has disabled TSX, the host might
1473 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1474 * and therefore knows that there cannot be TAA) but keep
1475 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1476 * and we want to allow migrating those guests to tsx=off hosts.
1478 data &= ~ARCH_CAP_TAA_NO;
1479 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1480 data |= ARCH_CAP_TAA_NO;
1483 * Nothing to do here; we emulate TSX_CTRL if present on the
1484 * host so the guest can choose between disabling TSX or
1485 * using VERW to clear CPU buffers.
1492 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1494 switch (msr->index) {
1495 case MSR_IA32_ARCH_CAPABILITIES:
1496 msr->data = kvm_get_arch_capabilities();
1498 case MSR_IA32_UCODE_REV:
1499 rdmsrl_safe(msr->index, &msr->data);
1502 return static_call(kvm_x86_get_msr_feature)(msr);
1507 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1509 struct kvm_msr_entry msr;
1513 r = kvm_get_msr_feature(&msr);
1515 if (r == KVM_MSR_RET_INVALID) {
1516 /* Unconditionally clear the output for simplicity */
1518 if (kvm_msr_ignored_check(index, 0, false))
1530 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1532 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1535 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1538 if (efer & (EFER_LME | EFER_LMA) &&
1539 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1542 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1548 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1550 if (efer & efer_reserved_bits)
1553 return __kvm_valid_efer(vcpu, efer);
1555 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1557 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1559 u64 old_efer = vcpu->arch.efer;
1560 u64 efer = msr_info->data;
1563 if (efer & efer_reserved_bits)
1566 if (!msr_info->host_initiated) {
1567 if (!__kvm_valid_efer(vcpu, efer))
1570 if (is_paging(vcpu) &&
1571 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1576 efer |= vcpu->arch.efer & EFER_LMA;
1578 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1584 /* Update reserved bits */
1585 if ((efer ^ old_efer) & EFER_NX)
1586 kvm_mmu_reset_context(vcpu);
1591 void kvm_enable_efer_bits(u64 mask)
1593 efer_reserved_bits &= ~mask;
1595 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1597 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1599 struct kvm_x86_msr_filter *msr_filter;
1600 struct msr_bitmap_range *ranges;
1601 struct kvm *kvm = vcpu->kvm;
1606 /* x2APIC MSRs do not support filtering. */
1607 if (index >= 0x800 && index <= 0x8ff)
1610 idx = srcu_read_lock(&kvm->srcu);
1612 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1618 allowed = msr_filter->default_allow;
1619 ranges = msr_filter->ranges;
1621 for (i = 0; i < msr_filter->count; i++) {
1622 u32 start = ranges[i].base;
1623 u32 end = start + ranges[i].nmsrs;
1624 u32 flags = ranges[i].flags;
1625 unsigned long *bitmap = ranges[i].bitmap;
1627 if ((index >= start) && (index < end) && (flags & type)) {
1628 allowed = !!test_bit(index - start, bitmap);
1634 srcu_read_unlock(&kvm->srcu, idx);
1638 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1641 * Write @data into the MSR specified by @index. Select MSR specific fault
1642 * checks are bypassed if @host_initiated is %true.
1643 * Returns 0 on success, non-0 otherwise.
1644 * Assumes vcpu_load() was already called.
1646 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1647 bool host_initiated)
1649 struct msr_data msr;
1651 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1652 return KVM_MSR_RET_FILTERED;
1657 case MSR_KERNEL_GS_BASE:
1660 if (is_noncanonical_address(data, vcpu))
1663 case MSR_IA32_SYSENTER_EIP:
1664 case MSR_IA32_SYSENTER_ESP:
1666 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1667 * non-canonical address is written on Intel but not on
1668 * AMD (which ignores the top 32-bits, because it does
1669 * not implement 64-bit SYSENTER).
1671 * 64-bit code should hence be able to write a non-canonical
1672 * value on AMD. Making the address canonical ensures that
1673 * vmentry does not fail on Intel after writing a non-canonical
1674 * value, and that something deterministic happens if the guest
1675 * invokes 64-bit SYSENTER.
1677 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1680 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1683 if (!host_initiated &&
1684 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1685 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1689 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1690 * incomplete and conflicting architectural behavior. Current
1691 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1692 * reserved and always read as zeros. Enforce Intel's reserved
1693 * bits check if and only if the guest CPU is Intel, and clear
1694 * the bits in all other cases. This ensures cross-vendor
1695 * migration will provide consistent behavior for the guest.
1697 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1706 msr.host_initiated = host_initiated;
1708 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1711 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1712 u32 index, u64 data, bool host_initiated)
1714 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1716 if (ret == KVM_MSR_RET_INVALID)
1717 if (kvm_msr_ignored_check(index, data, true))
1724 * Read the MSR specified by @index into @data. Select MSR specific fault
1725 * checks are bypassed if @host_initiated is %true.
1726 * Returns 0 on success, non-0 otherwise.
1727 * Assumes vcpu_load() was already called.
1729 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1730 bool host_initiated)
1732 struct msr_data msr;
1735 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1736 return KVM_MSR_RET_FILTERED;
1740 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1743 if (!host_initiated &&
1744 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1745 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1751 msr.host_initiated = host_initiated;
1753 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1759 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1760 u32 index, u64 *data, bool host_initiated)
1762 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1764 if (ret == KVM_MSR_RET_INVALID) {
1765 /* Unconditionally clear *data for simplicity */
1767 if (kvm_msr_ignored_check(index, 0, false))
1774 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1776 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1778 EXPORT_SYMBOL_GPL(kvm_get_msr);
1780 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1782 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1784 EXPORT_SYMBOL_GPL(kvm_set_msr);
1786 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1788 int err = vcpu->run->msr.error;
1790 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1791 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1794 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1797 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1799 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1802 static u64 kvm_msr_reason(int r)
1805 case KVM_MSR_RET_INVALID:
1806 return KVM_MSR_EXIT_REASON_UNKNOWN;
1807 case KVM_MSR_RET_FILTERED:
1808 return KVM_MSR_EXIT_REASON_FILTER;
1810 return KVM_MSR_EXIT_REASON_INVAL;
1814 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1815 u32 exit_reason, u64 data,
1816 int (*completion)(struct kvm_vcpu *vcpu),
1819 u64 msr_reason = kvm_msr_reason(r);
1821 /* Check if the user wanted to know about this MSR fault */
1822 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1825 vcpu->run->exit_reason = exit_reason;
1826 vcpu->run->msr.error = 0;
1827 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1828 vcpu->run->msr.reason = msr_reason;
1829 vcpu->run->msr.index = index;
1830 vcpu->run->msr.data = data;
1831 vcpu->arch.complete_userspace_io = completion;
1836 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1838 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1839 complete_emulated_rdmsr, r);
1842 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1844 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1845 complete_emulated_wrmsr, r);
1848 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1850 u32 ecx = kvm_rcx_read(vcpu);
1854 r = kvm_get_msr(vcpu, ecx, &data);
1856 /* MSR read failed? See if we should ask user space */
1857 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1858 /* Bounce to user space */
1863 trace_kvm_msr_read(ecx, data);
1865 kvm_rax_write(vcpu, data & -1u);
1866 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1868 trace_kvm_msr_read_ex(ecx);
1871 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1873 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1875 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1877 u32 ecx = kvm_rcx_read(vcpu);
1878 u64 data = kvm_read_edx_eax(vcpu);
1881 r = kvm_set_msr(vcpu, ecx, data);
1883 /* MSR write failed? See if we should ask user space */
1884 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1885 /* Bounce to user space */
1888 /* Signal all other negative errors to userspace */
1893 trace_kvm_msr_write(ecx, data);
1895 trace_kvm_msr_write_ex(ecx, data);
1897 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1899 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1901 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1903 return kvm_skip_emulated_instruction(vcpu);
1905 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1907 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1909 /* Treat an INVD instruction as a NOP and just skip it. */
1910 return kvm_emulate_as_nop(vcpu);
1912 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1914 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1916 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1917 return kvm_emulate_as_nop(vcpu);
1919 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1921 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1923 kvm_queue_exception(vcpu, UD_VECTOR);
1926 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1928 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1930 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1931 return kvm_emulate_as_nop(vcpu);
1933 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1935 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1937 xfer_to_guest_mode_prepare();
1938 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1939 xfer_to_guest_mode_work_pending();
1943 * The fast path for frequent and performance sensitive wrmsr emulation,
1944 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1945 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1946 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1947 * other cases which must be called after interrupts are enabled on the host.
1949 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1951 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1954 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1955 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1956 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1957 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1960 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1961 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1962 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1963 trace_kvm_apic_write(APIC_ICR, (u32)data);
1970 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1972 if (!kvm_can_use_hv_timer(vcpu))
1975 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1979 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1981 u32 msr = kvm_rcx_read(vcpu);
1983 fastpath_t ret = EXIT_FASTPATH_NONE;
1986 case APIC_BASE_MSR + (APIC_ICR >> 4):
1987 data = kvm_read_edx_eax(vcpu);
1988 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1989 kvm_skip_emulated_instruction(vcpu);
1990 ret = EXIT_FASTPATH_EXIT_HANDLED;
1993 case MSR_IA32_TSC_DEADLINE:
1994 data = kvm_read_edx_eax(vcpu);
1995 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1996 kvm_skip_emulated_instruction(vcpu);
1997 ret = EXIT_FASTPATH_REENTER_GUEST;
2004 if (ret != EXIT_FASTPATH_NONE)
2005 trace_kvm_msr_write(msr, data);
2009 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2012 * Adapt set_msr() to msr_io()'s calling convention
2014 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2016 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2019 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2021 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2024 #ifdef CONFIG_X86_64
2025 struct pvclock_clock {
2035 struct pvclock_gtod_data {
2038 struct pvclock_clock clock; /* extract of a clocksource struct */
2039 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2045 static struct pvclock_gtod_data pvclock_gtod_data;
2047 static void update_pvclock_gtod(struct timekeeper *tk)
2049 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2051 write_seqcount_begin(&vdata->seq);
2053 /* copy pvclock gtod data */
2054 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2055 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2056 vdata->clock.mask = tk->tkr_mono.mask;
2057 vdata->clock.mult = tk->tkr_mono.mult;
2058 vdata->clock.shift = tk->tkr_mono.shift;
2059 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2060 vdata->clock.offset = tk->tkr_mono.base;
2062 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2063 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2064 vdata->raw_clock.mask = tk->tkr_raw.mask;
2065 vdata->raw_clock.mult = tk->tkr_raw.mult;
2066 vdata->raw_clock.shift = tk->tkr_raw.shift;
2067 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2068 vdata->raw_clock.offset = tk->tkr_raw.base;
2070 vdata->wall_time_sec = tk->xtime_sec;
2072 vdata->offs_boot = tk->offs_boot;
2074 write_seqcount_end(&vdata->seq);
2077 static s64 get_kvmclock_base_ns(void)
2079 /* Count up from boot time, but with the frequency of the raw clock. */
2080 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2083 static s64 get_kvmclock_base_ns(void)
2085 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2086 return ktime_get_boottime_ns();
2090 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2094 struct pvclock_wall_clock wc;
2101 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2106 ++version; /* first time write, random junk */
2110 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2114 * The guest calculates current wall clock time by adding
2115 * system time (updated by kvm_guest_time_update below) to the
2116 * wall clock specified here. We do the reverse here.
2118 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2120 wc.nsec = do_div(wall_nsec, 1000000000);
2121 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2122 wc.version = version;
2124 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2127 wc_sec_hi = wall_nsec >> 32;
2128 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2129 &wc_sec_hi, sizeof(wc_sec_hi));
2133 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2136 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2137 bool old_msr, bool host_initiated)
2139 struct kvm_arch *ka = &vcpu->kvm->arch;
2141 if (vcpu->vcpu_id == 0 && !host_initiated) {
2142 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2143 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2145 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2148 vcpu->arch.time = system_time;
2149 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2151 /* we verify if the enable bit is set... */
2152 vcpu->arch.pv_time_enabled = false;
2153 if (!(system_time & 1))
2156 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2157 &vcpu->arch.pv_time, system_time & ~1ULL,
2158 sizeof(struct pvclock_vcpu_time_info)))
2159 vcpu->arch.pv_time_enabled = true;
2164 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2166 do_shl32_div32(dividend, divisor);
2170 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2171 s8 *pshift, u32 *pmultiplier)
2179 scaled64 = scaled_hz;
2180 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2185 tps32 = (uint32_t)tps64;
2186 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2187 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2195 *pmultiplier = div_frac(scaled64, tps32);
2198 #ifdef CONFIG_X86_64
2199 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2202 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2203 static unsigned long max_tsc_khz;
2205 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2207 u64 v = (u64)khz * (1000000 + ppm);
2212 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2214 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2218 /* Guest TSC same frequency as host TSC? */
2220 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2224 /* TSC scaling supported? */
2225 if (!kvm_has_tsc_control) {
2226 if (user_tsc_khz > tsc_khz) {
2227 vcpu->arch.tsc_catchup = 1;
2228 vcpu->arch.tsc_always_catchup = 1;
2231 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2236 /* TSC scaling required - calculate ratio */
2237 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2238 user_tsc_khz, tsc_khz);
2240 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2241 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2246 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2250 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2252 u32 thresh_lo, thresh_hi;
2253 int use_scaling = 0;
2255 /* tsc_khz can be zero if TSC calibration fails */
2256 if (user_tsc_khz == 0) {
2257 /* set tsc_scaling_ratio to a safe value */
2258 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2262 /* Compute a scale to convert nanoseconds in TSC cycles */
2263 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2264 &vcpu->arch.virtual_tsc_shift,
2265 &vcpu->arch.virtual_tsc_mult);
2266 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2269 * Compute the variation in TSC rate which is acceptable
2270 * within the range of tolerance and decide if the
2271 * rate being applied is within that bounds of the hardware
2272 * rate. If so, no scaling or compensation need be done.
2274 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2275 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2276 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2277 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2280 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2283 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2285 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2286 vcpu->arch.virtual_tsc_mult,
2287 vcpu->arch.virtual_tsc_shift);
2288 tsc += vcpu->arch.this_tsc_write;
2292 static inline int gtod_is_based_on_tsc(int mode)
2294 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2297 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2299 #ifdef CONFIG_X86_64
2301 struct kvm_arch *ka = &vcpu->kvm->arch;
2302 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2304 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2305 atomic_read(&vcpu->kvm->online_vcpus));
2308 * Once the masterclock is enabled, always perform request in
2309 * order to update it.
2311 * In order to enable masterclock, the host clocksource must be TSC
2312 * and the vcpus need to have matched TSCs. When that happens,
2313 * perform request to enable masterclock.
2315 if (ka->use_master_clock ||
2316 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2317 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2319 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2320 atomic_read(&vcpu->kvm->online_vcpus),
2321 ka->use_master_clock, gtod->clock.vclock_mode);
2326 * Multiply tsc by a fixed point number represented by ratio.
2328 * The most significant 64-N bits (mult) of ratio represent the
2329 * integral part of the fixed point number; the remaining N bits
2330 * (frac) represent the fractional part, ie. ratio represents a fixed
2331 * point number (mult + frac * 2^(-N)).
2333 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2335 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2337 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2340 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2344 if (ratio != kvm_default_tsc_scaling_ratio)
2345 _tsc = __scale_tsc(ratio, tsc);
2349 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2351 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2355 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2357 return target_tsc - tsc;
2360 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2362 return vcpu->arch.l1_tsc_offset +
2363 kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2365 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2367 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2371 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2372 nested_offset = l1_offset;
2374 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2375 kvm_tsc_scaling_ratio_frac_bits);
2377 nested_offset += l2_offset;
2378 return nested_offset;
2380 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2382 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2384 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2385 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2386 kvm_tsc_scaling_ratio_frac_bits);
2388 return l1_multiplier;
2390 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2392 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2394 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2395 vcpu->arch.l1_tsc_offset,
2398 vcpu->arch.l1_tsc_offset = l1_offset;
2401 * If we are here because L1 chose not to trap WRMSR to TSC then
2402 * according to the spec this should set L1's TSC (as opposed to
2403 * setting L1's offset for L2).
2405 if (is_guest_mode(vcpu))
2406 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2408 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2409 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2411 vcpu->arch.tsc_offset = l1_offset;
2413 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2416 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2418 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2420 /* Userspace is changing the multiplier while L2 is active */
2421 if (is_guest_mode(vcpu))
2422 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2424 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2426 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2428 if (kvm_has_tsc_control)
2429 static_call(kvm_x86_write_tsc_multiplier)(
2430 vcpu, vcpu->arch.tsc_scaling_ratio);
2433 static inline bool kvm_check_tsc_unstable(void)
2435 #ifdef CONFIG_X86_64
2437 * TSC is marked unstable when we're running on Hyper-V,
2438 * 'TSC page' clocksource is good.
2440 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2443 return check_tsc_unstable();
2446 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2448 struct kvm *kvm = vcpu->kvm;
2449 u64 offset, ns, elapsed;
2450 unsigned long flags;
2452 bool already_matched;
2453 bool synchronizing = false;
2455 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2456 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2457 ns = get_kvmclock_base_ns();
2458 elapsed = ns - kvm->arch.last_tsc_nsec;
2460 if (vcpu->arch.virtual_tsc_khz) {
2463 * detection of vcpu initialization -- need to sync
2464 * with other vCPUs. This particularly helps to keep
2465 * kvm_clock stable after CPU hotplug
2467 synchronizing = true;
2469 u64 tsc_exp = kvm->arch.last_tsc_write +
2470 nsec_to_cycles(vcpu, elapsed);
2471 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2473 * Special case: TSC write with a small delta (1 second)
2474 * of virtual cycle time against real time is
2475 * interpreted as an attempt to synchronize the CPU.
2477 synchronizing = data < tsc_exp + tsc_hz &&
2478 data + tsc_hz > tsc_exp;
2483 * For a reliable TSC, we can match TSC offsets, and for an unstable
2484 * TSC, we add elapsed time in this computation. We could let the
2485 * compensation code attempt to catch up if we fall behind, but
2486 * it's better to try to match offsets from the beginning.
2488 if (synchronizing &&
2489 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2490 if (!kvm_check_tsc_unstable()) {
2491 offset = kvm->arch.cur_tsc_offset;
2493 u64 delta = nsec_to_cycles(vcpu, elapsed);
2495 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2498 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2501 * We split periods of matched TSC writes into generations.
2502 * For each generation, we track the original measured
2503 * nanosecond time, offset, and write, so if TSCs are in
2504 * sync, we can match exact offset, and if not, we can match
2505 * exact software computation in compute_guest_tsc()
2507 * These values are tracked in kvm->arch.cur_xxx variables.
2509 kvm->arch.cur_tsc_generation++;
2510 kvm->arch.cur_tsc_nsec = ns;
2511 kvm->arch.cur_tsc_write = data;
2512 kvm->arch.cur_tsc_offset = offset;
2517 * We also track th most recent recorded KHZ, write and time to
2518 * allow the matching interval to be extended at each write.
2520 kvm->arch.last_tsc_nsec = ns;
2521 kvm->arch.last_tsc_write = data;
2522 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2524 vcpu->arch.last_guest_tsc = data;
2526 /* Keep track of which generation this VCPU has synchronized to */
2527 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2528 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2529 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2531 kvm_vcpu_write_tsc_offset(vcpu, offset);
2532 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2534 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2536 kvm->arch.nr_vcpus_matched_tsc = 0;
2537 } else if (!already_matched) {
2538 kvm->arch.nr_vcpus_matched_tsc++;
2541 kvm_track_tsc_matching(vcpu);
2542 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2545 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2548 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2549 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2552 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2554 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2555 WARN_ON(adjustment < 0);
2556 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2557 vcpu->arch.l1_tsc_scaling_ratio);
2558 adjust_tsc_offset_guest(vcpu, adjustment);
2561 #ifdef CONFIG_X86_64
2563 static u64 read_tsc(void)
2565 u64 ret = (u64)rdtsc_ordered();
2566 u64 last = pvclock_gtod_data.clock.cycle_last;
2568 if (likely(ret >= last))
2572 * GCC likes to generate cmov here, but this branch is extremely
2573 * predictable (it's just a function of time and the likely is
2574 * very likely) and there's a data dependence, so force GCC
2575 * to generate a branch instead. I don't barrier() because
2576 * we don't actually need a barrier, and if this function
2577 * ever gets inlined it will generate worse code.
2583 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2589 switch (clock->vclock_mode) {
2590 case VDSO_CLOCKMODE_HVCLOCK:
2591 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2593 if (tsc_pg_val != U64_MAX) {
2594 /* TSC page valid */
2595 *mode = VDSO_CLOCKMODE_HVCLOCK;
2596 v = (tsc_pg_val - clock->cycle_last) &
2599 /* TSC page invalid */
2600 *mode = VDSO_CLOCKMODE_NONE;
2603 case VDSO_CLOCKMODE_TSC:
2604 *mode = VDSO_CLOCKMODE_TSC;
2605 *tsc_timestamp = read_tsc();
2606 v = (*tsc_timestamp - clock->cycle_last) &
2610 *mode = VDSO_CLOCKMODE_NONE;
2613 if (*mode == VDSO_CLOCKMODE_NONE)
2614 *tsc_timestamp = v = 0;
2616 return v * clock->mult;
2619 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2621 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2627 seq = read_seqcount_begin(>od->seq);
2628 ns = gtod->raw_clock.base_cycles;
2629 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2630 ns >>= gtod->raw_clock.shift;
2631 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2632 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2638 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2640 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2646 seq = read_seqcount_begin(>od->seq);
2647 ts->tv_sec = gtod->wall_time_sec;
2648 ns = gtod->clock.base_cycles;
2649 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2650 ns >>= gtod->clock.shift;
2651 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2653 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2659 /* returns true if host is using TSC based clocksource */
2660 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2662 /* checked again under seqlock below */
2663 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2666 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2670 /* returns true if host is using TSC based clocksource */
2671 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2674 /* checked again under seqlock below */
2675 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2678 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2684 * Assuming a stable TSC across physical CPUS, and a stable TSC
2685 * across virtual CPUs, the following condition is possible.
2686 * Each numbered line represents an event visible to both
2687 * CPUs at the next numbered event.
2689 * "timespecX" represents host monotonic time. "tscX" represents
2692 * VCPU0 on CPU0 | VCPU1 on CPU1
2694 * 1. read timespec0,tsc0
2695 * 2. | timespec1 = timespec0 + N
2697 * 3. transition to guest | transition to guest
2698 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2699 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2700 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2702 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2705 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2707 * - 0 < N - M => M < N
2709 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2710 * always the case (the difference between two distinct xtime instances
2711 * might be smaller then the difference between corresponding TSC reads,
2712 * when updating guest vcpus pvclock areas).
2714 * To avoid that problem, do not allow visibility of distinct
2715 * system_timestamp/tsc_timestamp values simultaneously: use a master
2716 * copy of host monotonic time values. Update that master copy
2719 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2723 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2725 #ifdef CONFIG_X86_64
2726 struct kvm_arch *ka = &kvm->arch;
2728 bool host_tsc_clocksource, vcpus_matched;
2730 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2731 atomic_read(&kvm->online_vcpus));
2734 * If the host uses TSC clock, then passthrough TSC as stable
2737 host_tsc_clocksource = kvm_get_time_and_clockread(
2738 &ka->master_kernel_ns,
2739 &ka->master_cycle_now);
2741 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2742 && !ka->backwards_tsc_observed
2743 && !ka->boot_vcpu_runs_old_kvmclock;
2745 if (ka->use_master_clock)
2746 atomic_set(&kvm_guest_has_master_clock, 1);
2748 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2749 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2754 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2756 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2759 static void kvm_gen_update_masterclock(struct kvm *kvm)
2761 #ifdef CONFIG_X86_64
2763 struct kvm_vcpu *vcpu;
2764 struct kvm_arch *ka = &kvm->arch;
2765 unsigned long flags;
2767 kvm_hv_invalidate_tsc_page(kvm);
2769 kvm_make_mclock_inprogress_request(kvm);
2771 /* no guest entries from this point */
2772 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2773 pvclock_update_vm_gtod_copy(kvm);
2774 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2776 kvm_for_each_vcpu(i, vcpu, kvm)
2777 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2779 /* guest entries allowed */
2780 kvm_for_each_vcpu(i, vcpu, kvm)
2781 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2785 u64 get_kvmclock_ns(struct kvm *kvm)
2787 struct kvm_arch *ka = &kvm->arch;
2788 struct pvclock_vcpu_time_info hv_clock;
2789 unsigned long flags;
2792 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2793 if (!ka->use_master_clock) {
2794 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2795 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2798 hv_clock.tsc_timestamp = ka->master_cycle_now;
2799 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2800 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2802 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2805 if (__this_cpu_read(cpu_tsc_khz)) {
2806 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2807 &hv_clock.tsc_shift,
2808 &hv_clock.tsc_to_system_mul);
2809 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2811 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2818 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2819 struct gfn_to_hva_cache *cache,
2820 unsigned int offset)
2822 struct kvm_vcpu_arch *vcpu = &v->arch;
2823 struct pvclock_vcpu_time_info guest_hv_clock;
2825 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2826 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2829 /* This VCPU is paused, but it's legal for a guest to read another
2830 * VCPU's kvmclock, so we really have to follow the specification where
2831 * it says that version is odd if data is being modified, and even after
2834 * Version field updates must be kept separate. This is because
2835 * kvm_write_guest_cached might use a "rep movs" instruction, and
2836 * writes within a string instruction are weakly ordered. So there
2837 * are three writes overall.
2839 * As a small optimization, only write the version field in the first
2840 * and third write. The vcpu->pv_time cache is still valid, because the
2841 * version field is the first in the struct.
2843 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2845 if (guest_hv_clock.version & 1)
2846 ++guest_hv_clock.version; /* first time write, random junk */
2848 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2849 kvm_write_guest_offset_cached(v->kvm, cache,
2850 &vcpu->hv_clock, offset,
2851 sizeof(vcpu->hv_clock.version));
2855 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2856 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2858 if (vcpu->pvclock_set_guest_stopped_request) {
2859 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2860 vcpu->pvclock_set_guest_stopped_request = false;
2863 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2865 kvm_write_guest_offset_cached(v->kvm, cache,
2866 &vcpu->hv_clock, offset,
2867 sizeof(vcpu->hv_clock));
2871 vcpu->hv_clock.version++;
2872 kvm_write_guest_offset_cached(v->kvm, cache,
2873 &vcpu->hv_clock, offset,
2874 sizeof(vcpu->hv_clock.version));
2877 static int kvm_guest_time_update(struct kvm_vcpu *v)
2879 unsigned long flags, tgt_tsc_khz;
2880 struct kvm_vcpu_arch *vcpu = &v->arch;
2881 struct kvm_arch *ka = &v->kvm->arch;
2883 u64 tsc_timestamp, host_tsc;
2885 bool use_master_clock;
2891 * If the host uses TSC clock, then passthrough TSC as stable
2894 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2895 use_master_clock = ka->use_master_clock;
2896 if (use_master_clock) {
2897 host_tsc = ka->master_cycle_now;
2898 kernel_ns = ka->master_kernel_ns;
2900 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2902 /* Keep irq disabled to prevent changes to the clock */
2903 local_irq_save(flags);
2904 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2905 if (unlikely(tgt_tsc_khz == 0)) {
2906 local_irq_restore(flags);
2907 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2910 if (!use_master_clock) {
2912 kernel_ns = get_kvmclock_base_ns();
2915 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2918 * We may have to catch up the TSC to match elapsed wall clock
2919 * time for two reasons, even if kvmclock is used.
2920 * 1) CPU could have been running below the maximum TSC rate
2921 * 2) Broken TSC compensation resets the base at each VCPU
2922 * entry to avoid unknown leaps of TSC even when running
2923 * again on the same CPU. This may cause apparent elapsed
2924 * time to disappear, and the guest to stand still or run
2927 if (vcpu->tsc_catchup) {
2928 u64 tsc = compute_guest_tsc(v, kernel_ns);
2929 if (tsc > tsc_timestamp) {
2930 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2931 tsc_timestamp = tsc;
2935 local_irq_restore(flags);
2937 /* With all the info we got, fill in the values */
2939 if (kvm_has_tsc_control)
2940 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
2941 v->arch.l1_tsc_scaling_ratio);
2943 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2944 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2945 &vcpu->hv_clock.tsc_shift,
2946 &vcpu->hv_clock.tsc_to_system_mul);
2947 vcpu->hw_tsc_khz = tgt_tsc_khz;
2950 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2951 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2952 vcpu->last_guest_tsc = tsc_timestamp;
2954 /* If the host uses TSC clocksource, then it is stable */
2956 if (use_master_clock)
2957 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2959 vcpu->hv_clock.flags = pvclock_flags;
2961 if (vcpu->pv_time_enabled)
2962 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2963 if (vcpu->xen.vcpu_info_set)
2964 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2965 offsetof(struct compat_vcpu_info, time));
2966 if (vcpu->xen.vcpu_time_info_set)
2967 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2968 if (v == kvm_get_vcpu(v->kvm, 0))
2969 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2974 * kvmclock updates which are isolated to a given vcpu, such as
2975 * vcpu->cpu migration, should not allow system_timestamp from
2976 * the rest of the vcpus to remain static. Otherwise ntp frequency
2977 * correction applies to one vcpu's system_timestamp but not
2980 * So in those cases, request a kvmclock update for all vcpus.
2981 * We need to rate-limit these requests though, as they can
2982 * considerably slow guests that have a large number of vcpus.
2983 * The time for a remote vcpu to update its kvmclock is bound
2984 * by the delay we use to rate-limit the updates.
2987 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2989 static void kvmclock_update_fn(struct work_struct *work)
2992 struct delayed_work *dwork = to_delayed_work(work);
2993 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2994 kvmclock_update_work);
2995 struct kvm *kvm = container_of(ka, struct kvm, arch);
2996 struct kvm_vcpu *vcpu;
2998 kvm_for_each_vcpu(i, vcpu, kvm) {
2999 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3000 kvm_vcpu_kick(vcpu);
3004 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3006 struct kvm *kvm = v->kvm;
3008 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3009 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3010 KVMCLOCK_UPDATE_DELAY);
3013 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3015 static void kvmclock_sync_fn(struct work_struct *work)
3017 struct delayed_work *dwork = to_delayed_work(work);
3018 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3019 kvmclock_sync_work);
3020 struct kvm *kvm = container_of(ka, struct kvm, arch);
3022 if (!kvmclock_periodic_sync)
3025 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3026 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3027 KVMCLOCK_SYNC_PERIOD);
3031 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3033 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3035 /* McStatusWrEn enabled? */
3036 if (guest_cpuid_is_amd_or_hygon(vcpu))
3037 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3042 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3044 u64 mcg_cap = vcpu->arch.mcg_cap;
3045 unsigned bank_num = mcg_cap & 0xff;
3046 u32 msr = msr_info->index;
3047 u64 data = msr_info->data;
3050 case MSR_IA32_MCG_STATUS:
3051 vcpu->arch.mcg_status = data;
3053 case MSR_IA32_MCG_CTL:
3054 if (!(mcg_cap & MCG_CTL_P) &&
3055 (data || !msr_info->host_initiated))
3057 if (data != 0 && data != ~(u64)0)
3059 vcpu->arch.mcg_ctl = data;
3062 if (msr >= MSR_IA32_MC0_CTL &&
3063 msr < MSR_IA32_MCx_CTL(bank_num)) {
3064 u32 offset = array_index_nospec(
3065 msr - MSR_IA32_MC0_CTL,
3066 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3068 /* only 0 or all 1s can be written to IA32_MCi_CTL
3069 * some Linux kernels though clear bit 10 in bank 4 to
3070 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3071 * this to avoid an uncatched #GP in the guest
3073 if ((offset & 0x3) == 0 &&
3074 data != 0 && (data | (1 << 10)) != ~(u64)0)
3078 if (!msr_info->host_initiated &&
3079 (offset & 0x3) == 1 && data != 0) {
3080 if (!can_set_mci_status(vcpu))
3084 vcpu->arch.mce_banks[offset] = data;
3092 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3094 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3096 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3099 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3101 gpa_t gpa = data & ~0x3f;
3103 /* Bits 4:5 are reserved, Should be zero */
3107 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3108 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3111 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3112 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3115 if (!lapic_in_kernel(vcpu))
3116 return data ? 1 : 0;
3118 vcpu->arch.apf.msr_en_val = data;
3120 if (!kvm_pv_async_pf_enabled(vcpu)) {
3121 kvm_clear_async_pf_completion_queue(vcpu);
3122 kvm_async_pf_hash_reset(vcpu);
3126 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3130 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3131 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3133 kvm_async_pf_wakeup_all(vcpu);
3138 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3140 /* Bits 8-63 are reserved */
3144 if (!lapic_in_kernel(vcpu))
3147 vcpu->arch.apf.msr_int_val = data;
3149 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3154 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3156 vcpu->arch.pv_time_enabled = false;
3157 vcpu->arch.time = 0;
3160 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3162 ++vcpu->stat.tlb_flush;
3163 static_call(kvm_x86_tlb_flush_all)(vcpu);
3166 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3168 ++vcpu->stat.tlb_flush;
3172 * A TLB flush on behalf of the guest is equivalent to
3173 * INVPCID(all), toggling CR4.PGE, etc., which requires
3174 * a forced sync of the shadow page tables. Unload the
3175 * entire MMU here and the subsequent load will sync the
3176 * shadow page tables, and also flush the TLB.
3178 kvm_mmu_unload(vcpu);
3182 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3185 static void record_steal_time(struct kvm_vcpu *vcpu)
3187 struct kvm_host_map map;
3188 struct kvm_steal_time *st;
3190 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3191 kvm_xen_runstate_set_running(vcpu);
3195 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3198 /* -EAGAIN is returned in atomic context so we can just return. */
3199 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3200 &map, &vcpu->arch.st.cache, false))
3204 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3207 * Doing a TLB flush here, on the guest's behalf, can avoid
3210 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3211 u8 st_preempted = xchg(&st->preempted, 0);
3213 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3214 st_preempted & KVM_VCPU_FLUSH_TLB);
3215 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3216 kvm_vcpu_flush_tlb_guest(vcpu);
3221 vcpu->arch.st.preempted = 0;
3223 if (st->version & 1)
3224 st->version += 1; /* first time write, random junk */
3230 st->steal += current->sched_info.run_delay -
3231 vcpu->arch.st.last_steal;
3232 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3238 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3241 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3244 u32 msr = msr_info->index;
3245 u64 data = msr_info->data;
3247 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3248 return kvm_xen_write_hypercall_page(vcpu, data);
3251 case MSR_AMD64_NB_CFG:
3252 case MSR_IA32_UCODE_WRITE:
3253 case MSR_VM_HSAVE_PA:
3254 case MSR_AMD64_PATCH_LOADER:
3255 case MSR_AMD64_BU_CFG2:
3256 case MSR_AMD64_DC_CFG:
3257 case MSR_F15H_EX_CFG:
3260 case MSR_IA32_UCODE_REV:
3261 if (msr_info->host_initiated)
3262 vcpu->arch.microcode_version = data;
3264 case MSR_IA32_ARCH_CAPABILITIES:
3265 if (!msr_info->host_initiated)
3267 vcpu->arch.arch_capabilities = data;
3269 case MSR_IA32_PERF_CAPABILITIES: {
3270 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3272 if (!msr_info->host_initiated)
3274 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3276 if (data & ~msr_ent.data)
3279 vcpu->arch.perf_capabilities = data;
3284 return set_efer(vcpu, msr_info);
3286 data &= ~(u64)0x40; /* ignore flush filter disable */
3287 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3288 data &= ~(u64)0x8; /* ignore TLB cache disable */
3290 /* Handle McStatusWrEn */
3291 if (data == BIT_ULL(18)) {
3292 vcpu->arch.msr_hwcr = data;
3293 } else if (data != 0) {
3294 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3299 case MSR_FAM10H_MMIO_CONF_BASE:
3301 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3306 case 0x200 ... 0x2ff:
3307 return kvm_mtrr_set_msr(vcpu, msr, data);
3308 case MSR_IA32_APICBASE:
3309 return kvm_set_apic_base(vcpu, msr_info);
3310 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3311 return kvm_x2apic_msr_write(vcpu, msr, data);
3312 case MSR_IA32_TSC_DEADLINE:
3313 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3315 case MSR_IA32_TSC_ADJUST:
3316 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3317 if (!msr_info->host_initiated) {
3318 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3319 adjust_tsc_offset_guest(vcpu, adj);
3321 vcpu->arch.ia32_tsc_adjust_msr = data;
3324 case MSR_IA32_MISC_ENABLE:
3325 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3326 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3327 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3329 vcpu->arch.ia32_misc_enable_msr = data;
3330 kvm_update_cpuid_runtime(vcpu);
3332 vcpu->arch.ia32_misc_enable_msr = data;
3335 case MSR_IA32_SMBASE:
3336 if (!msr_info->host_initiated)
3338 vcpu->arch.smbase = data;
3340 case MSR_IA32_POWER_CTL:
3341 vcpu->arch.msr_ia32_power_ctl = data;
3344 if (msr_info->host_initiated) {
3345 kvm_synchronize_tsc(vcpu, data);
3347 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3348 adjust_tsc_offset_guest(vcpu, adj);
3349 vcpu->arch.ia32_tsc_adjust_msr += adj;
3353 if (!msr_info->host_initiated &&
3354 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3357 * KVM supports exposing PT to the guest, but does not support
3358 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3359 * XSAVES/XRSTORS to save/restore PT MSRs.
3361 if (data & ~supported_xss)
3363 vcpu->arch.ia32_xss = data;
3366 if (!msr_info->host_initiated)
3368 vcpu->arch.smi_count = data;
3370 case MSR_KVM_WALL_CLOCK_NEW:
3371 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3374 vcpu->kvm->arch.wall_clock = data;
3375 kvm_write_wall_clock(vcpu->kvm, data, 0);
3377 case MSR_KVM_WALL_CLOCK:
3378 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3381 vcpu->kvm->arch.wall_clock = data;
3382 kvm_write_wall_clock(vcpu->kvm, data, 0);
3384 case MSR_KVM_SYSTEM_TIME_NEW:
3385 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3388 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3390 case MSR_KVM_SYSTEM_TIME:
3391 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3394 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3396 case MSR_KVM_ASYNC_PF_EN:
3397 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3400 if (kvm_pv_enable_async_pf(vcpu, data))
3403 case MSR_KVM_ASYNC_PF_INT:
3404 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3407 if (kvm_pv_enable_async_pf_int(vcpu, data))
3410 case MSR_KVM_ASYNC_PF_ACK:
3411 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3414 vcpu->arch.apf.pageready_pending = false;
3415 kvm_check_async_pf_completion(vcpu);
3418 case MSR_KVM_STEAL_TIME:
3419 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3422 if (unlikely(!sched_info_on()))
3425 if (data & KVM_STEAL_RESERVED_MASK)
3428 vcpu->arch.st.msr_val = data;
3430 if (!(data & KVM_MSR_ENABLED))
3433 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3436 case MSR_KVM_PV_EOI_EN:
3437 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3440 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3444 case MSR_KVM_POLL_CONTROL:
3445 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3448 /* only enable bit supported */
3449 if (data & (-1ULL << 1))
3452 vcpu->arch.msr_kvm_poll_control = data;
3455 case MSR_IA32_MCG_CTL:
3456 case MSR_IA32_MCG_STATUS:
3457 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3458 return set_msr_mce(vcpu, msr_info);
3460 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3461 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3464 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3465 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3466 if (kvm_pmu_is_valid_msr(vcpu, msr))
3467 return kvm_pmu_set_msr(vcpu, msr_info);
3469 if (pr || data != 0)
3470 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3471 "0x%x data 0x%llx\n", msr, data);
3473 case MSR_K7_CLK_CTL:
3475 * Ignore all writes to this no longer documented MSR.
3476 * Writes are only relevant for old K7 processors,
3477 * all pre-dating SVM, but a recommended workaround from
3478 * AMD for these chips. It is possible to specify the
3479 * affected processor models on the command line, hence
3480 * the need to ignore the workaround.
3483 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3484 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3485 case HV_X64_MSR_SYNDBG_OPTIONS:
3486 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3487 case HV_X64_MSR_CRASH_CTL:
3488 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3489 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3490 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3491 case HV_X64_MSR_TSC_EMULATION_STATUS:
3492 return kvm_hv_set_msr_common(vcpu, msr, data,
3493 msr_info->host_initiated);
3494 case MSR_IA32_BBL_CR_CTL3:
3495 /* Drop writes to this legacy MSR -- see rdmsr
3496 * counterpart for further detail.
3498 if (report_ignored_msrs)
3499 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3502 case MSR_AMD64_OSVW_ID_LENGTH:
3503 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3505 vcpu->arch.osvw.length = data;
3507 case MSR_AMD64_OSVW_STATUS:
3508 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3510 vcpu->arch.osvw.status = data;
3512 case MSR_PLATFORM_INFO:
3513 if (!msr_info->host_initiated ||
3514 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3515 cpuid_fault_enabled(vcpu)))
3517 vcpu->arch.msr_platform_info = data;
3519 case MSR_MISC_FEATURES_ENABLES:
3520 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3521 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3522 !supports_cpuid_fault(vcpu)))
3524 vcpu->arch.msr_misc_features_enables = data;
3527 if (kvm_pmu_is_valid_msr(vcpu, msr))
3528 return kvm_pmu_set_msr(vcpu, msr_info);
3529 return KVM_MSR_RET_INVALID;
3533 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3535 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3538 u64 mcg_cap = vcpu->arch.mcg_cap;
3539 unsigned bank_num = mcg_cap & 0xff;
3542 case MSR_IA32_P5_MC_ADDR:
3543 case MSR_IA32_P5_MC_TYPE:
3546 case MSR_IA32_MCG_CAP:
3547 data = vcpu->arch.mcg_cap;
3549 case MSR_IA32_MCG_CTL:
3550 if (!(mcg_cap & MCG_CTL_P) && !host)
3552 data = vcpu->arch.mcg_ctl;
3554 case MSR_IA32_MCG_STATUS:
3555 data = vcpu->arch.mcg_status;
3558 if (msr >= MSR_IA32_MC0_CTL &&
3559 msr < MSR_IA32_MCx_CTL(bank_num)) {
3560 u32 offset = array_index_nospec(
3561 msr - MSR_IA32_MC0_CTL,
3562 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3564 data = vcpu->arch.mce_banks[offset];
3573 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3575 switch (msr_info->index) {
3576 case MSR_IA32_PLATFORM_ID:
3577 case MSR_IA32_EBL_CR_POWERON:
3578 case MSR_IA32_LASTBRANCHFROMIP:
3579 case MSR_IA32_LASTBRANCHTOIP:
3580 case MSR_IA32_LASTINTFROMIP:
3581 case MSR_IA32_LASTINTTOIP:
3582 case MSR_AMD64_SYSCFG:
3583 case MSR_K8_TSEG_ADDR:
3584 case MSR_K8_TSEG_MASK:
3585 case MSR_VM_HSAVE_PA:
3586 case MSR_K8_INT_PENDING_MSG:
3587 case MSR_AMD64_NB_CFG:
3588 case MSR_FAM10H_MMIO_CONF_BASE:
3589 case MSR_AMD64_BU_CFG2:
3590 case MSR_IA32_PERF_CTL:
3591 case MSR_AMD64_DC_CFG:
3592 case MSR_F15H_EX_CFG:
3594 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3595 * limit) MSRs. Just return 0, as we do not want to expose the host
3596 * data here. Do not conditionalize this on CPUID, as KVM does not do
3597 * so for existing CPU-specific MSRs.
3599 case MSR_RAPL_POWER_UNIT:
3600 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3601 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3602 case MSR_PKG_ENERGY_STATUS: /* Total package */
3603 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3606 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3607 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3608 return kvm_pmu_get_msr(vcpu, msr_info);
3609 if (!msr_info->host_initiated)
3613 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3614 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3615 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3616 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3617 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3618 return kvm_pmu_get_msr(vcpu, msr_info);
3621 case MSR_IA32_UCODE_REV:
3622 msr_info->data = vcpu->arch.microcode_version;
3624 case MSR_IA32_ARCH_CAPABILITIES:
3625 if (!msr_info->host_initiated &&
3626 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3628 msr_info->data = vcpu->arch.arch_capabilities;
3630 case MSR_IA32_PERF_CAPABILITIES:
3631 if (!msr_info->host_initiated &&
3632 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3634 msr_info->data = vcpu->arch.perf_capabilities;
3636 case MSR_IA32_POWER_CTL:
3637 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3639 case MSR_IA32_TSC: {
3641 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3642 * even when not intercepted. AMD manual doesn't explicitly
3643 * state this but appears to behave the same.
3645 * On userspace reads and writes, however, we unconditionally
3646 * return L1's TSC value to ensure backwards-compatible
3647 * behavior for migration.
3651 if (msr_info->host_initiated) {
3652 offset = vcpu->arch.l1_tsc_offset;
3653 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3655 offset = vcpu->arch.tsc_offset;
3656 ratio = vcpu->arch.tsc_scaling_ratio;
3659 msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3663 case 0x200 ... 0x2ff:
3664 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3665 case 0xcd: /* fsb frequency */
3669 * MSR_EBC_FREQUENCY_ID
3670 * Conservative value valid for even the basic CPU models.
3671 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3672 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3673 * and 266MHz for model 3, or 4. Set Core Clock
3674 * Frequency to System Bus Frequency Ratio to 1 (bits
3675 * 31:24) even though these are only valid for CPU
3676 * models > 2, however guests may end up dividing or
3677 * multiplying by zero otherwise.
3679 case MSR_EBC_FREQUENCY_ID:
3680 msr_info->data = 1 << 24;
3682 case MSR_IA32_APICBASE:
3683 msr_info->data = kvm_get_apic_base(vcpu);
3685 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3686 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3687 case MSR_IA32_TSC_DEADLINE:
3688 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3690 case MSR_IA32_TSC_ADJUST:
3691 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3693 case MSR_IA32_MISC_ENABLE:
3694 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3696 case MSR_IA32_SMBASE:
3697 if (!msr_info->host_initiated)
3699 msr_info->data = vcpu->arch.smbase;
3702 msr_info->data = vcpu->arch.smi_count;
3704 case MSR_IA32_PERF_STATUS:
3705 /* TSC increment by tick */
3706 msr_info->data = 1000ULL;
3707 /* CPU multiplier */
3708 msr_info->data |= (((uint64_t)4ULL) << 40);
3711 msr_info->data = vcpu->arch.efer;
3713 case MSR_KVM_WALL_CLOCK:
3714 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3717 msr_info->data = vcpu->kvm->arch.wall_clock;
3719 case MSR_KVM_WALL_CLOCK_NEW:
3720 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3723 msr_info->data = vcpu->kvm->arch.wall_clock;
3725 case MSR_KVM_SYSTEM_TIME:
3726 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3729 msr_info->data = vcpu->arch.time;
3731 case MSR_KVM_SYSTEM_TIME_NEW:
3732 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3735 msr_info->data = vcpu->arch.time;
3737 case MSR_KVM_ASYNC_PF_EN:
3738 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3741 msr_info->data = vcpu->arch.apf.msr_en_val;
3743 case MSR_KVM_ASYNC_PF_INT:
3744 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3747 msr_info->data = vcpu->arch.apf.msr_int_val;
3749 case MSR_KVM_ASYNC_PF_ACK:
3750 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3755 case MSR_KVM_STEAL_TIME:
3756 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3759 msr_info->data = vcpu->arch.st.msr_val;
3761 case MSR_KVM_PV_EOI_EN:
3762 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3765 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3767 case MSR_KVM_POLL_CONTROL:
3768 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3771 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3773 case MSR_IA32_P5_MC_ADDR:
3774 case MSR_IA32_P5_MC_TYPE:
3775 case MSR_IA32_MCG_CAP:
3776 case MSR_IA32_MCG_CTL:
3777 case MSR_IA32_MCG_STATUS:
3778 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3779 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3780 msr_info->host_initiated);
3782 if (!msr_info->host_initiated &&
3783 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3785 msr_info->data = vcpu->arch.ia32_xss;
3787 case MSR_K7_CLK_CTL:
3789 * Provide expected ramp-up count for K7. All other
3790 * are set to zero, indicating minimum divisors for
3793 * This prevents guest kernels on AMD host with CPU
3794 * type 6, model 8 and higher from exploding due to
3795 * the rdmsr failing.
3797 msr_info->data = 0x20000000;
3799 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3800 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3801 case HV_X64_MSR_SYNDBG_OPTIONS:
3802 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3803 case HV_X64_MSR_CRASH_CTL:
3804 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3805 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3806 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3807 case HV_X64_MSR_TSC_EMULATION_STATUS:
3808 return kvm_hv_get_msr_common(vcpu,
3809 msr_info->index, &msr_info->data,
3810 msr_info->host_initiated);
3811 case MSR_IA32_BBL_CR_CTL3:
3812 /* This legacy MSR exists but isn't fully documented in current
3813 * silicon. It is however accessed by winxp in very narrow
3814 * scenarios where it sets bit #19, itself documented as
3815 * a "reserved" bit. Best effort attempt to source coherent
3816 * read data here should the balance of the register be
3817 * interpreted by the guest:
3819 * L2 cache control register 3: 64GB range, 256KB size,
3820 * enabled, latency 0x1, configured
3822 msr_info->data = 0xbe702111;
3824 case MSR_AMD64_OSVW_ID_LENGTH:
3825 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3827 msr_info->data = vcpu->arch.osvw.length;
3829 case MSR_AMD64_OSVW_STATUS:
3830 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3832 msr_info->data = vcpu->arch.osvw.status;
3834 case MSR_PLATFORM_INFO:
3835 if (!msr_info->host_initiated &&
3836 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3838 msr_info->data = vcpu->arch.msr_platform_info;
3840 case MSR_MISC_FEATURES_ENABLES:
3841 msr_info->data = vcpu->arch.msr_misc_features_enables;
3844 msr_info->data = vcpu->arch.msr_hwcr;
3847 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3848 return kvm_pmu_get_msr(vcpu, msr_info);
3849 return KVM_MSR_RET_INVALID;
3853 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3856 * Read or write a bunch of msrs. All parameters are kernel addresses.
3858 * @return number of msrs set successfully.
3860 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3861 struct kvm_msr_entry *entries,
3862 int (*do_msr)(struct kvm_vcpu *vcpu,
3863 unsigned index, u64 *data))
3867 for (i = 0; i < msrs->nmsrs; ++i)
3868 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3875 * Read or write a bunch of msrs. Parameters are user addresses.
3877 * @return number of msrs set successfully.
3879 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3880 int (*do_msr)(struct kvm_vcpu *vcpu,
3881 unsigned index, u64 *data),
3884 struct kvm_msrs msrs;
3885 struct kvm_msr_entry *entries;
3890 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3894 if (msrs.nmsrs >= MAX_IO_MSRS)
3897 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3898 entries = memdup_user(user_msrs->entries, size);
3899 if (IS_ERR(entries)) {
3900 r = PTR_ERR(entries);
3904 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3909 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3920 static inline bool kvm_can_mwait_in_guest(void)
3922 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3923 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3924 boot_cpu_has(X86_FEATURE_ARAT);
3927 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3928 struct kvm_cpuid2 __user *cpuid_arg)
3930 struct kvm_cpuid2 cpuid;
3934 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3937 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3942 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3948 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3953 case KVM_CAP_IRQCHIP:
3955 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3956 case KVM_CAP_SET_TSS_ADDR:
3957 case KVM_CAP_EXT_CPUID:
3958 case KVM_CAP_EXT_EMUL_CPUID:
3959 case KVM_CAP_CLOCKSOURCE:
3961 case KVM_CAP_NOP_IO_DELAY:
3962 case KVM_CAP_MP_STATE:
3963 case KVM_CAP_SYNC_MMU:
3964 case KVM_CAP_USER_NMI:
3965 case KVM_CAP_REINJECT_CONTROL:
3966 case KVM_CAP_IRQ_INJECT_STATUS:
3967 case KVM_CAP_IOEVENTFD:
3968 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3970 case KVM_CAP_PIT_STATE2:
3971 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3972 case KVM_CAP_VCPU_EVENTS:
3973 case KVM_CAP_HYPERV:
3974 case KVM_CAP_HYPERV_VAPIC:
3975 case KVM_CAP_HYPERV_SPIN:
3976 case KVM_CAP_HYPERV_SYNIC:
3977 case KVM_CAP_HYPERV_SYNIC2:
3978 case KVM_CAP_HYPERV_VP_INDEX:
3979 case KVM_CAP_HYPERV_EVENTFD:
3980 case KVM_CAP_HYPERV_TLBFLUSH:
3981 case KVM_CAP_HYPERV_SEND_IPI:
3982 case KVM_CAP_HYPERV_CPUID:
3983 case KVM_CAP_HYPERV_ENFORCE_CPUID:
3984 case KVM_CAP_SYS_HYPERV_CPUID:
3985 case KVM_CAP_PCI_SEGMENT:
3986 case KVM_CAP_DEBUGREGS:
3987 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3989 case KVM_CAP_ASYNC_PF:
3990 case KVM_CAP_ASYNC_PF_INT:
3991 case KVM_CAP_GET_TSC_KHZ:
3992 case KVM_CAP_KVMCLOCK_CTRL:
3993 case KVM_CAP_READONLY_MEM:
3994 case KVM_CAP_HYPERV_TIME:
3995 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3996 case KVM_CAP_TSC_DEADLINE_TIMER:
3997 case KVM_CAP_DISABLE_QUIRKS:
3998 case KVM_CAP_SET_BOOT_CPU_ID:
3999 case KVM_CAP_SPLIT_IRQCHIP:
4000 case KVM_CAP_IMMEDIATE_EXIT:
4001 case KVM_CAP_PMU_EVENT_FILTER:
4002 case KVM_CAP_GET_MSR_FEATURES:
4003 case KVM_CAP_MSR_PLATFORM_INFO:
4004 case KVM_CAP_EXCEPTION_PAYLOAD:
4005 case KVM_CAP_SET_GUEST_DEBUG:
4006 case KVM_CAP_LAST_CPU:
4007 case KVM_CAP_X86_USER_SPACE_MSR:
4008 case KVM_CAP_X86_MSR_FILTER:
4009 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4010 #ifdef CONFIG_X86_SGX_KVM
4011 case KVM_CAP_SGX_ATTRIBUTE:
4013 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4014 case KVM_CAP_SREGS2:
4015 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4018 case KVM_CAP_EXIT_HYPERCALL:
4019 r = KVM_EXIT_HYPERCALL_VALID_MASK;
4021 case KVM_CAP_SET_GUEST_DEBUG2:
4022 return KVM_GUESTDBG_VALID_MASK;
4023 #ifdef CONFIG_KVM_XEN
4024 case KVM_CAP_XEN_HVM:
4025 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4026 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4027 KVM_XEN_HVM_CONFIG_SHARED_INFO;
4028 if (sched_info_on())
4029 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4032 case KVM_CAP_SYNC_REGS:
4033 r = KVM_SYNC_X86_VALID_FIELDS;
4035 case KVM_CAP_ADJUST_CLOCK:
4036 r = KVM_CLOCK_TSC_STABLE;
4038 case KVM_CAP_X86_DISABLE_EXITS:
4039 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4040 KVM_X86_DISABLE_EXITS_CSTATE;
4041 if(kvm_can_mwait_in_guest())
4042 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4044 case KVM_CAP_X86_SMM:
4045 /* SMBASE is usually relocated above 1M on modern chipsets,
4046 * and SMM handlers might indeed rely on 4G segment limits,
4047 * so do not report SMM to be available if real mode is
4048 * emulated via vm86 mode. Still, do not go to great lengths
4049 * to avoid userspace's usage of the feature, because it is a
4050 * fringe case that is not enabled except via specific settings
4051 * of the module parameters.
4053 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4056 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4058 case KVM_CAP_NR_VCPUS:
4059 r = KVM_SOFT_MAX_VCPUS;
4061 case KVM_CAP_MAX_VCPUS:
4064 case KVM_CAP_MAX_VCPU_ID:
4065 r = KVM_MAX_VCPU_ID;
4067 case KVM_CAP_PV_MMU: /* obsolete */
4071 r = KVM_MAX_MCE_BANKS;
4074 r = boot_cpu_has(X86_FEATURE_XSAVE);
4076 case KVM_CAP_TSC_CONTROL:
4077 r = kvm_has_tsc_control;
4079 case KVM_CAP_X2APIC_API:
4080 r = KVM_X2APIC_API_VALID_FLAGS;
4082 case KVM_CAP_NESTED_STATE:
4083 r = kvm_x86_ops.nested_ops->get_state ?
4084 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4086 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4087 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4089 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4090 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4092 case KVM_CAP_SMALLER_MAXPHYADDR:
4093 r = (int) allow_smaller_maxphyaddr;
4095 case KVM_CAP_STEAL_TIME:
4096 r = sched_info_on();
4098 case KVM_CAP_X86_BUS_LOCK_EXIT:
4099 if (kvm_has_bus_lock_exit)
4100 r = KVM_BUS_LOCK_DETECTION_OFF |
4101 KVM_BUS_LOCK_DETECTION_EXIT;
4112 long kvm_arch_dev_ioctl(struct file *filp,
4113 unsigned int ioctl, unsigned long arg)
4115 void __user *argp = (void __user *)arg;
4119 case KVM_GET_MSR_INDEX_LIST: {
4120 struct kvm_msr_list __user *user_msr_list = argp;
4121 struct kvm_msr_list msr_list;
4125 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4128 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4129 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4132 if (n < msr_list.nmsrs)
4135 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4136 num_msrs_to_save * sizeof(u32)))
4138 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4140 num_emulated_msrs * sizeof(u32)))
4145 case KVM_GET_SUPPORTED_CPUID:
4146 case KVM_GET_EMULATED_CPUID: {
4147 struct kvm_cpuid2 __user *cpuid_arg = argp;
4148 struct kvm_cpuid2 cpuid;
4151 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4154 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4160 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4165 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4167 if (copy_to_user(argp, &kvm_mce_cap_supported,
4168 sizeof(kvm_mce_cap_supported)))
4172 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4173 struct kvm_msr_list __user *user_msr_list = argp;
4174 struct kvm_msr_list msr_list;
4178 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4181 msr_list.nmsrs = num_msr_based_features;
4182 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4185 if (n < msr_list.nmsrs)
4188 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4189 num_msr_based_features * sizeof(u32)))
4195 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4197 case KVM_GET_SUPPORTED_HV_CPUID:
4198 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4208 static void wbinvd_ipi(void *garbage)
4213 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4215 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4218 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4220 /* Address WBINVD may be executed by guest */
4221 if (need_emulate_wbinvd(vcpu)) {
4222 if (static_call(kvm_x86_has_wbinvd_exit)())
4223 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4224 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4225 smp_call_function_single(vcpu->cpu,
4226 wbinvd_ipi, NULL, 1);
4229 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4231 /* Save host pkru register if supported */
4232 vcpu->arch.host_pkru = read_pkru();
4234 /* Apply any externally detected TSC adjustments (due to suspend) */
4235 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4236 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4237 vcpu->arch.tsc_offset_adjustment = 0;
4238 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4241 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4242 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4243 rdtsc() - vcpu->arch.last_host_tsc;
4245 mark_tsc_unstable("KVM discovered backwards TSC");
4247 if (kvm_check_tsc_unstable()) {
4248 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4249 vcpu->arch.last_guest_tsc);
4250 kvm_vcpu_write_tsc_offset(vcpu, offset);
4251 vcpu->arch.tsc_catchup = 1;
4254 if (kvm_lapic_hv_timer_in_use(vcpu))
4255 kvm_lapic_restart_hv_timer(vcpu);
4258 * On a host with synchronized TSC, there is no need to update
4259 * kvmclock on vcpu->cpu migration
4261 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4262 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4263 if (vcpu->cpu != cpu)
4264 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4268 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4271 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4273 struct kvm_host_map map;
4274 struct kvm_steal_time *st;
4276 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4279 if (vcpu->arch.st.preempted)
4282 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4283 &vcpu->arch.st.cache, true))
4287 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4289 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4291 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4294 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4298 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4299 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4302 * Take the srcu lock as memslots will be accessed to check the gfn
4303 * cache generation against the memslots generation.
4305 idx = srcu_read_lock(&vcpu->kvm->srcu);
4306 if (kvm_xen_msr_enabled(vcpu->kvm))
4307 kvm_xen_runstate_set_preempted(vcpu);
4309 kvm_steal_time_set_preempted(vcpu);
4310 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4312 static_call(kvm_x86_vcpu_put)(vcpu);
4313 vcpu->arch.last_host_tsc = rdtsc();
4315 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4316 * on every vmexit, but if not, we might have a stale dr6 from the
4317 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4322 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4323 struct kvm_lapic_state *s)
4325 if (vcpu->arch.apicv_active)
4326 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4328 return kvm_apic_get_state(vcpu, s);
4331 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4332 struct kvm_lapic_state *s)
4336 r = kvm_apic_set_state(vcpu, s);
4339 update_cr8_intercept(vcpu);
4344 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4347 * We can accept userspace's request for interrupt injection
4348 * as long as we have a place to store the interrupt number.
4349 * The actual injection will happen when the CPU is able to
4350 * deliver the interrupt.
4352 if (kvm_cpu_has_extint(vcpu))
4355 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4356 return (!lapic_in_kernel(vcpu) ||
4357 kvm_apic_accept_pic_intr(vcpu));
4360 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4363 * Do not cause an interrupt window exit if an exception
4364 * is pending or an event needs reinjection; userspace
4365 * might want to inject the interrupt manually using KVM_SET_REGS
4366 * or KVM_SET_SREGS. For that to work, we must be at an
4367 * instruction boundary and with no events half-injected.
4369 return (kvm_arch_interrupt_allowed(vcpu) &&
4370 kvm_cpu_accept_dm_intr(vcpu) &&
4371 !kvm_event_needs_reinjection(vcpu) &&
4372 !vcpu->arch.exception.pending);
4375 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4376 struct kvm_interrupt *irq)
4378 if (irq->irq >= KVM_NR_INTERRUPTS)
4381 if (!irqchip_in_kernel(vcpu->kvm)) {
4382 kvm_queue_interrupt(vcpu, irq->irq, false);
4383 kvm_make_request(KVM_REQ_EVENT, vcpu);
4388 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4389 * fail for in-kernel 8259.
4391 if (pic_in_kernel(vcpu->kvm))
4394 if (vcpu->arch.pending_external_vector != -1)
4397 vcpu->arch.pending_external_vector = irq->irq;
4398 kvm_make_request(KVM_REQ_EVENT, vcpu);
4402 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4404 kvm_inject_nmi(vcpu);
4409 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4411 kvm_make_request(KVM_REQ_SMI, vcpu);
4416 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4417 struct kvm_tpr_access_ctl *tac)
4421 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4425 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4429 unsigned bank_num = mcg_cap & 0xff, bank;
4432 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4434 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4437 vcpu->arch.mcg_cap = mcg_cap;
4438 /* Init IA32_MCG_CTL to all 1s */
4439 if (mcg_cap & MCG_CTL_P)
4440 vcpu->arch.mcg_ctl = ~(u64)0;
4441 /* Init IA32_MCi_CTL to all 1s */
4442 for (bank = 0; bank < bank_num; bank++)
4443 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4445 static_call(kvm_x86_setup_mce)(vcpu);
4450 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4451 struct kvm_x86_mce *mce)
4453 u64 mcg_cap = vcpu->arch.mcg_cap;
4454 unsigned bank_num = mcg_cap & 0xff;
4455 u64 *banks = vcpu->arch.mce_banks;
4457 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4460 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4461 * reporting is disabled
4463 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4464 vcpu->arch.mcg_ctl != ~(u64)0)
4466 banks += 4 * mce->bank;
4468 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4469 * reporting is disabled for the bank
4471 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4473 if (mce->status & MCI_STATUS_UC) {
4474 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4475 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4476 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4479 if (banks[1] & MCI_STATUS_VAL)
4480 mce->status |= MCI_STATUS_OVER;
4481 banks[2] = mce->addr;
4482 banks[3] = mce->misc;
4483 vcpu->arch.mcg_status = mce->mcg_status;
4484 banks[1] = mce->status;
4485 kvm_queue_exception(vcpu, MC_VECTOR);
4486 } else if (!(banks[1] & MCI_STATUS_VAL)
4487 || !(banks[1] & MCI_STATUS_UC)) {
4488 if (banks[1] & MCI_STATUS_VAL)
4489 mce->status |= MCI_STATUS_OVER;
4490 banks[2] = mce->addr;
4491 banks[3] = mce->misc;
4492 banks[1] = mce->status;
4494 banks[1] |= MCI_STATUS_OVER;
4498 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4499 struct kvm_vcpu_events *events)
4503 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4507 * In guest mode, payload delivery should be deferred,
4508 * so that the L1 hypervisor can intercept #PF before
4509 * CR2 is modified (or intercept #DB before DR6 is
4510 * modified under nVMX). Unless the per-VM capability,
4511 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4512 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4513 * opportunistically defer the exception payload, deliver it if the
4514 * capability hasn't been requested before processing a
4515 * KVM_GET_VCPU_EVENTS.
4517 if (!vcpu->kvm->arch.exception_payload_enabled &&
4518 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4519 kvm_deliver_exception_payload(vcpu);
4522 * The API doesn't provide the instruction length for software
4523 * exceptions, so don't report them. As long as the guest RIP
4524 * isn't advanced, we should expect to encounter the exception
4527 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4528 events->exception.injected = 0;
4529 events->exception.pending = 0;
4531 events->exception.injected = vcpu->arch.exception.injected;
4532 events->exception.pending = vcpu->arch.exception.pending;
4534 * For ABI compatibility, deliberately conflate
4535 * pending and injected exceptions when
4536 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4538 if (!vcpu->kvm->arch.exception_payload_enabled)
4539 events->exception.injected |=
4540 vcpu->arch.exception.pending;
4542 events->exception.nr = vcpu->arch.exception.nr;
4543 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4544 events->exception.error_code = vcpu->arch.exception.error_code;
4545 events->exception_has_payload = vcpu->arch.exception.has_payload;
4546 events->exception_payload = vcpu->arch.exception.payload;
4548 events->interrupt.injected =
4549 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4550 events->interrupt.nr = vcpu->arch.interrupt.nr;
4551 events->interrupt.soft = 0;
4552 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4554 events->nmi.injected = vcpu->arch.nmi_injected;
4555 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4556 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4557 events->nmi.pad = 0;
4559 events->sipi_vector = 0; /* never valid when reporting to user space */
4561 events->smi.smm = is_smm(vcpu);
4562 events->smi.pending = vcpu->arch.smi_pending;
4563 events->smi.smm_inside_nmi =
4564 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4565 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4567 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4568 | KVM_VCPUEVENT_VALID_SHADOW
4569 | KVM_VCPUEVENT_VALID_SMM);
4570 if (vcpu->kvm->arch.exception_payload_enabled)
4571 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4573 memset(&events->reserved, 0, sizeof(events->reserved));
4576 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4578 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4579 struct kvm_vcpu_events *events)
4581 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4582 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4583 | KVM_VCPUEVENT_VALID_SHADOW
4584 | KVM_VCPUEVENT_VALID_SMM
4585 | KVM_VCPUEVENT_VALID_PAYLOAD))
4588 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4589 if (!vcpu->kvm->arch.exception_payload_enabled)
4591 if (events->exception.pending)
4592 events->exception.injected = 0;
4594 events->exception_has_payload = 0;
4596 events->exception.pending = 0;
4597 events->exception_has_payload = 0;
4600 if ((events->exception.injected || events->exception.pending) &&
4601 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4604 /* INITs are latched while in SMM */
4605 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4606 (events->smi.smm || events->smi.pending) &&
4607 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4611 vcpu->arch.exception.injected = events->exception.injected;
4612 vcpu->arch.exception.pending = events->exception.pending;
4613 vcpu->arch.exception.nr = events->exception.nr;
4614 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4615 vcpu->arch.exception.error_code = events->exception.error_code;
4616 vcpu->arch.exception.has_payload = events->exception_has_payload;
4617 vcpu->arch.exception.payload = events->exception_payload;
4619 vcpu->arch.interrupt.injected = events->interrupt.injected;
4620 vcpu->arch.interrupt.nr = events->interrupt.nr;
4621 vcpu->arch.interrupt.soft = events->interrupt.soft;
4622 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4623 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4624 events->interrupt.shadow);
4626 vcpu->arch.nmi_injected = events->nmi.injected;
4627 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4628 vcpu->arch.nmi_pending = events->nmi.pending;
4629 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4631 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4632 lapic_in_kernel(vcpu))
4633 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4635 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4636 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4637 kvm_smm_changed(vcpu, events->smi.smm);
4639 vcpu->arch.smi_pending = events->smi.pending;
4641 if (events->smi.smm) {
4642 if (events->smi.smm_inside_nmi)
4643 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4645 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4648 if (lapic_in_kernel(vcpu)) {
4649 if (events->smi.latched_init)
4650 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4652 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4656 kvm_make_request(KVM_REQ_EVENT, vcpu);
4661 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4662 struct kvm_debugregs *dbgregs)
4666 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4667 kvm_get_dr(vcpu, 6, &val);
4669 dbgregs->dr7 = vcpu->arch.dr7;
4671 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4674 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4675 struct kvm_debugregs *dbgregs)
4680 if (!kvm_dr6_valid(dbgregs->dr6))
4682 if (!kvm_dr7_valid(dbgregs->dr7))
4685 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4686 kvm_update_dr0123(vcpu);
4687 vcpu->arch.dr6 = dbgregs->dr6;
4688 vcpu->arch.dr7 = dbgregs->dr7;
4689 kvm_update_dr7(vcpu);
4694 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4696 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4698 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4699 u64 xstate_bv = xsave->header.xfeatures;
4703 * Copy legacy XSAVE area, to avoid complications with CPUID
4704 * leaves 0 and 1 in the loop below.
4706 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4709 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4710 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4713 * Copy each region from the possibly compacted offset to the
4714 * non-compacted offset.
4716 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4718 u32 size, offset, ecx, edx;
4719 u64 xfeature_mask = valid & -valid;
4720 int xfeature_nr = fls64(xfeature_mask) - 1;
4723 cpuid_count(XSTATE_CPUID, xfeature_nr,
4724 &size, &offset, &ecx, &edx);
4726 if (xfeature_nr == XFEATURE_PKRU) {
4727 memcpy(dest + offset, &vcpu->arch.pkru,
4728 sizeof(vcpu->arch.pkru));
4730 src = get_xsave_addr(xsave, xfeature_nr);
4732 memcpy(dest + offset, src, size);
4735 valid -= xfeature_mask;
4739 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4741 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4742 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4746 * Copy legacy XSAVE area, to avoid complications with CPUID
4747 * leaves 0 and 1 in the loop below.
4749 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4751 /* Set XSTATE_BV and possibly XCOMP_BV. */
4752 xsave->header.xfeatures = xstate_bv;
4753 if (boot_cpu_has(X86_FEATURE_XSAVES))
4754 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4757 * Copy each region from the non-compacted offset to the
4758 * possibly compacted offset.
4760 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4762 u32 size, offset, ecx, edx;
4763 u64 xfeature_mask = valid & -valid;
4764 int xfeature_nr = fls64(xfeature_mask) - 1;
4766 cpuid_count(XSTATE_CPUID, xfeature_nr,
4767 &size, &offset, &ecx, &edx);
4769 if (xfeature_nr == XFEATURE_PKRU) {
4770 memcpy(&vcpu->arch.pkru, src + offset,
4771 sizeof(vcpu->arch.pkru));
4773 void *dest = get_xsave_addr(xsave, xfeature_nr);
4776 memcpy(dest, src + offset, size);
4779 valid -= xfeature_mask;
4783 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4784 struct kvm_xsave *guest_xsave)
4786 if (!vcpu->arch.guest_fpu)
4789 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4790 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4791 fill_xsave((u8 *) guest_xsave->region, vcpu);
4793 memcpy(guest_xsave->region,
4794 &vcpu->arch.guest_fpu->state.fxsave,
4795 sizeof(struct fxregs_state));
4796 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4797 XFEATURE_MASK_FPSSE;
4801 #define XSAVE_MXCSR_OFFSET 24
4803 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4804 struct kvm_xsave *guest_xsave)
4809 if (!vcpu->arch.guest_fpu)
4812 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4813 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4815 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4817 * Here we allow setting states that are not present in
4818 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4819 * with old userspace.
4821 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4823 load_xsave(vcpu, (u8 *)guest_xsave->region);
4825 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4826 mxcsr & ~mxcsr_feature_mask)
4828 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4829 guest_xsave->region, sizeof(struct fxregs_state));
4834 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4835 struct kvm_xcrs *guest_xcrs)
4837 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4838 guest_xcrs->nr_xcrs = 0;
4842 guest_xcrs->nr_xcrs = 1;
4843 guest_xcrs->flags = 0;
4844 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4845 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4848 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4849 struct kvm_xcrs *guest_xcrs)
4853 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4856 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4859 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4860 /* Only support XCR0 currently */
4861 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4862 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4863 guest_xcrs->xcrs[i].value);
4872 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4873 * stopped by the hypervisor. This function will be called from the host only.
4874 * EINVAL is returned when the host attempts to set the flag for a guest that
4875 * does not support pv clocks.
4877 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4879 if (!vcpu->arch.pv_time_enabled)
4881 vcpu->arch.pvclock_set_guest_stopped_request = true;
4882 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4886 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4887 struct kvm_enable_cap *cap)
4890 uint16_t vmcs_version;
4891 void __user *user_ptr;
4897 case KVM_CAP_HYPERV_SYNIC2:
4902 case KVM_CAP_HYPERV_SYNIC:
4903 if (!irqchip_in_kernel(vcpu->kvm))
4905 return kvm_hv_activate_synic(vcpu, cap->cap ==
4906 KVM_CAP_HYPERV_SYNIC2);
4907 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4908 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4910 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4912 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4913 if (copy_to_user(user_ptr, &vmcs_version,
4914 sizeof(vmcs_version)))
4918 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4919 if (!kvm_x86_ops.enable_direct_tlbflush)
4922 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4924 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4925 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
4927 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4928 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4929 if (vcpu->arch.pv_cpuid.enforce)
4930 kvm_update_pv_runtime(vcpu);
4938 long kvm_arch_vcpu_ioctl(struct file *filp,
4939 unsigned int ioctl, unsigned long arg)
4941 struct kvm_vcpu *vcpu = filp->private_data;
4942 void __user *argp = (void __user *)arg;
4945 struct kvm_sregs2 *sregs2;
4946 struct kvm_lapic_state *lapic;
4947 struct kvm_xsave *xsave;
4948 struct kvm_xcrs *xcrs;
4956 case KVM_GET_LAPIC: {
4958 if (!lapic_in_kernel(vcpu))
4960 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4961 GFP_KERNEL_ACCOUNT);
4966 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4970 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4975 case KVM_SET_LAPIC: {
4977 if (!lapic_in_kernel(vcpu))
4979 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4980 if (IS_ERR(u.lapic)) {
4981 r = PTR_ERR(u.lapic);
4985 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4988 case KVM_INTERRUPT: {
4989 struct kvm_interrupt irq;
4992 if (copy_from_user(&irq, argp, sizeof(irq)))
4994 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4998 r = kvm_vcpu_ioctl_nmi(vcpu);
5002 r = kvm_vcpu_ioctl_smi(vcpu);
5005 case KVM_SET_CPUID: {
5006 struct kvm_cpuid __user *cpuid_arg = argp;
5007 struct kvm_cpuid cpuid;
5010 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5012 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5015 case KVM_SET_CPUID2: {
5016 struct kvm_cpuid2 __user *cpuid_arg = argp;
5017 struct kvm_cpuid2 cpuid;
5020 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5022 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5023 cpuid_arg->entries);
5026 case KVM_GET_CPUID2: {
5027 struct kvm_cpuid2 __user *cpuid_arg = argp;
5028 struct kvm_cpuid2 cpuid;
5031 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5033 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5034 cpuid_arg->entries);
5038 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5043 case KVM_GET_MSRS: {
5044 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5045 r = msr_io(vcpu, argp, do_get_msr, 1);
5046 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5049 case KVM_SET_MSRS: {
5050 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5051 r = msr_io(vcpu, argp, do_set_msr, 0);
5052 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5055 case KVM_TPR_ACCESS_REPORTING: {
5056 struct kvm_tpr_access_ctl tac;
5059 if (copy_from_user(&tac, argp, sizeof(tac)))
5061 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5065 if (copy_to_user(argp, &tac, sizeof(tac)))
5070 case KVM_SET_VAPIC_ADDR: {
5071 struct kvm_vapic_addr va;
5075 if (!lapic_in_kernel(vcpu))
5078 if (copy_from_user(&va, argp, sizeof(va)))
5080 idx = srcu_read_lock(&vcpu->kvm->srcu);
5081 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5082 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5085 case KVM_X86_SETUP_MCE: {
5089 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5091 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5094 case KVM_X86_SET_MCE: {
5095 struct kvm_x86_mce mce;
5098 if (copy_from_user(&mce, argp, sizeof(mce)))
5100 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5103 case KVM_GET_VCPU_EVENTS: {
5104 struct kvm_vcpu_events events;
5106 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5109 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5114 case KVM_SET_VCPU_EVENTS: {
5115 struct kvm_vcpu_events events;
5118 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5121 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5124 case KVM_GET_DEBUGREGS: {
5125 struct kvm_debugregs dbgregs;
5127 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5130 if (copy_to_user(argp, &dbgregs,
5131 sizeof(struct kvm_debugregs)))
5136 case KVM_SET_DEBUGREGS: {
5137 struct kvm_debugregs dbgregs;
5140 if (copy_from_user(&dbgregs, argp,
5141 sizeof(struct kvm_debugregs)))
5144 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5147 case KVM_GET_XSAVE: {
5148 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5153 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5156 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5161 case KVM_SET_XSAVE: {
5162 u.xsave = memdup_user(argp, sizeof(*u.xsave));
5163 if (IS_ERR(u.xsave)) {
5164 r = PTR_ERR(u.xsave);
5168 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5171 case KVM_GET_XCRS: {
5172 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5177 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5180 if (copy_to_user(argp, u.xcrs,
5181 sizeof(struct kvm_xcrs)))
5186 case KVM_SET_XCRS: {
5187 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5188 if (IS_ERR(u.xcrs)) {
5189 r = PTR_ERR(u.xcrs);
5193 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5196 case KVM_SET_TSC_KHZ: {
5200 user_tsc_khz = (u32)arg;
5202 if (kvm_has_tsc_control &&
5203 user_tsc_khz >= kvm_max_guest_tsc_khz)
5206 if (user_tsc_khz == 0)
5207 user_tsc_khz = tsc_khz;
5209 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5214 case KVM_GET_TSC_KHZ: {
5215 r = vcpu->arch.virtual_tsc_khz;
5218 case KVM_KVMCLOCK_CTRL: {
5219 r = kvm_set_guest_paused(vcpu);
5222 case KVM_ENABLE_CAP: {
5223 struct kvm_enable_cap cap;
5226 if (copy_from_user(&cap, argp, sizeof(cap)))
5228 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5231 case KVM_GET_NESTED_STATE: {
5232 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5236 if (!kvm_x86_ops.nested_ops->get_state)
5239 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5241 if (get_user(user_data_size, &user_kvm_nested_state->size))
5244 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5249 if (r > user_data_size) {
5250 if (put_user(r, &user_kvm_nested_state->size))
5260 case KVM_SET_NESTED_STATE: {
5261 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5262 struct kvm_nested_state kvm_state;
5266 if (!kvm_x86_ops.nested_ops->set_state)
5270 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5274 if (kvm_state.size < sizeof(kvm_state))
5277 if (kvm_state.flags &
5278 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5279 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5280 | KVM_STATE_NESTED_GIF_SET))
5283 /* nested_run_pending implies guest_mode. */
5284 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5285 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5288 idx = srcu_read_lock(&vcpu->kvm->srcu);
5289 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5290 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5293 case KVM_GET_SUPPORTED_HV_CPUID:
5294 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5296 #ifdef CONFIG_KVM_XEN
5297 case KVM_XEN_VCPU_GET_ATTR: {
5298 struct kvm_xen_vcpu_attr xva;
5301 if (copy_from_user(&xva, argp, sizeof(xva)))
5303 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5304 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5308 case KVM_XEN_VCPU_SET_ATTR: {
5309 struct kvm_xen_vcpu_attr xva;
5312 if (copy_from_user(&xva, argp, sizeof(xva)))
5314 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5318 case KVM_GET_SREGS2: {
5319 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5323 __get_sregs2(vcpu, u.sregs2);
5325 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5330 case KVM_SET_SREGS2: {
5331 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5332 if (IS_ERR(u.sregs2)) {
5333 r = PTR_ERR(u.sregs2);
5337 r = __set_sregs2(vcpu, u.sregs2);
5350 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5352 return VM_FAULT_SIGBUS;
5355 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5359 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5361 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5365 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5368 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5371 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5372 unsigned long kvm_nr_mmu_pages)
5374 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5377 mutex_lock(&kvm->slots_lock);
5379 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5380 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5382 mutex_unlock(&kvm->slots_lock);
5386 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5388 return kvm->arch.n_max_mmu_pages;
5391 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5393 struct kvm_pic *pic = kvm->arch.vpic;
5397 switch (chip->chip_id) {
5398 case KVM_IRQCHIP_PIC_MASTER:
5399 memcpy(&chip->chip.pic, &pic->pics[0],
5400 sizeof(struct kvm_pic_state));
5402 case KVM_IRQCHIP_PIC_SLAVE:
5403 memcpy(&chip->chip.pic, &pic->pics[1],
5404 sizeof(struct kvm_pic_state));
5406 case KVM_IRQCHIP_IOAPIC:
5407 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5416 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5418 struct kvm_pic *pic = kvm->arch.vpic;
5422 switch (chip->chip_id) {
5423 case KVM_IRQCHIP_PIC_MASTER:
5424 spin_lock(&pic->lock);
5425 memcpy(&pic->pics[0], &chip->chip.pic,
5426 sizeof(struct kvm_pic_state));
5427 spin_unlock(&pic->lock);
5429 case KVM_IRQCHIP_PIC_SLAVE:
5430 spin_lock(&pic->lock);
5431 memcpy(&pic->pics[1], &chip->chip.pic,
5432 sizeof(struct kvm_pic_state));
5433 spin_unlock(&pic->lock);
5435 case KVM_IRQCHIP_IOAPIC:
5436 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5442 kvm_pic_update_irq(pic);
5446 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5448 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5450 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5452 mutex_lock(&kps->lock);
5453 memcpy(ps, &kps->channels, sizeof(*ps));
5454 mutex_unlock(&kps->lock);
5458 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5461 struct kvm_pit *pit = kvm->arch.vpit;
5463 mutex_lock(&pit->pit_state.lock);
5464 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5465 for (i = 0; i < 3; i++)
5466 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5467 mutex_unlock(&pit->pit_state.lock);
5471 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5473 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5474 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5475 sizeof(ps->channels));
5476 ps->flags = kvm->arch.vpit->pit_state.flags;
5477 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5478 memset(&ps->reserved, 0, sizeof(ps->reserved));
5482 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5486 u32 prev_legacy, cur_legacy;
5487 struct kvm_pit *pit = kvm->arch.vpit;
5489 mutex_lock(&pit->pit_state.lock);
5490 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5491 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5492 if (!prev_legacy && cur_legacy)
5494 memcpy(&pit->pit_state.channels, &ps->channels,
5495 sizeof(pit->pit_state.channels));
5496 pit->pit_state.flags = ps->flags;
5497 for (i = 0; i < 3; i++)
5498 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5500 mutex_unlock(&pit->pit_state.lock);
5504 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5505 struct kvm_reinject_control *control)
5507 struct kvm_pit *pit = kvm->arch.vpit;
5509 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5510 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5511 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5513 mutex_lock(&pit->pit_state.lock);
5514 kvm_pit_set_reinject(pit, control->pit_reinject);
5515 mutex_unlock(&pit->pit_state.lock);
5520 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5524 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5525 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5526 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5529 struct kvm_vcpu *vcpu;
5532 kvm_for_each_vcpu(i, vcpu, kvm)
5533 kvm_vcpu_kick(vcpu);
5536 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5539 if (!irqchip_in_kernel(kvm))
5542 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5543 irq_event->irq, irq_event->level,
5548 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5549 struct kvm_enable_cap *cap)
5557 case KVM_CAP_DISABLE_QUIRKS:
5558 kvm->arch.disabled_quirks = cap->args[0];
5561 case KVM_CAP_SPLIT_IRQCHIP: {
5562 mutex_lock(&kvm->lock);
5564 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5565 goto split_irqchip_unlock;
5567 if (irqchip_in_kernel(kvm))
5568 goto split_irqchip_unlock;
5569 if (kvm->created_vcpus)
5570 goto split_irqchip_unlock;
5571 r = kvm_setup_empty_irq_routing(kvm);
5573 goto split_irqchip_unlock;
5574 /* Pairs with irqchip_in_kernel. */
5576 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5577 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5579 split_irqchip_unlock:
5580 mutex_unlock(&kvm->lock);
5583 case KVM_CAP_X2APIC_API:
5585 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5588 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5589 kvm->arch.x2apic_format = true;
5590 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5591 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5595 case KVM_CAP_X86_DISABLE_EXITS:
5597 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5600 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5601 kvm_can_mwait_in_guest())
5602 kvm->arch.mwait_in_guest = true;
5603 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5604 kvm->arch.hlt_in_guest = true;
5605 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5606 kvm->arch.pause_in_guest = true;
5607 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5608 kvm->arch.cstate_in_guest = true;
5611 case KVM_CAP_MSR_PLATFORM_INFO:
5612 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5615 case KVM_CAP_EXCEPTION_PAYLOAD:
5616 kvm->arch.exception_payload_enabled = cap->args[0];
5619 case KVM_CAP_X86_USER_SPACE_MSR:
5620 kvm->arch.user_space_msr_mask = cap->args[0];
5623 case KVM_CAP_X86_BUS_LOCK_EXIT:
5625 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5628 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5629 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5632 if (kvm_has_bus_lock_exit &&
5633 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5634 kvm->arch.bus_lock_detection_enabled = true;
5637 #ifdef CONFIG_X86_SGX_KVM
5638 case KVM_CAP_SGX_ATTRIBUTE: {
5639 unsigned long allowed_attributes = 0;
5641 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5645 /* KVM only supports the PROVISIONKEY privileged attribute. */
5646 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5647 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5648 kvm->arch.sgx_provisioning_allowed = true;
5654 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5656 if (kvm_x86_ops.vm_copy_enc_context_from)
5657 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5659 case KVM_CAP_EXIT_HYPERCALL:
5660 if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5664 kvm->arch.hypercall_exit_enabled = cap->args[0];
5667 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5669 if (cap->args[0] & ~1)
5671 kvm->arch.exit_on_emulation_error = cap->args[0];
5681 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5683 struct kvm_x86_msr_filter *msr_filter;
5685 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5689 msr_filter->default_allow = default_allow;
5693 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5700 for (i = 0; i < msr_filter->count; i++)
5701 kfree(msr_filter->ranges[i].bitmap);
5706 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5707 struct kvm_msr_filter_range *user_range)
5709 unsigned long *bitmap = NULL;
5712 if (!user_range->nmsrs)
5715 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5718 if (!user_range->flags)
5721 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5722 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5725 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5727 return PTR_ERR(bitmap);
5729 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5730 .flags = user_range->flags,
5731 .base = user_range->base,
5732 .nmsrs = user_range->nmsrs,
5736 msr_filter->count++;
5740 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5742 struct kvm_msr_filter __user *user_msr_filter = argp;
5743 struct kvm_x86_msr_filter *new_filter, *old_filter;
5744 struct kvm_msr_filter filter;
5750 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5753 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5754 empty &= !filter.ranges[i].nmsrs;
5756 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5757 if (empty && !default_allow)
5760 new_filter = kvm_alloc_msr_filter(default_allow);
5764 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5765 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5767 kvm_free_msr_filter(new_filter);
5772 mutex_lock(&kvm->lock);
5774 /* The per-VM filter is protected by kvm->lock... */
5775 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5777 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5778 synchronize_srcu(&kvm->srcu);
5780 kvm_free_msr_filter(old_filter);
5782 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5783 mutex_unlock(&kvm->lock);
5788 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5789 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5791 struct kvm_vcpu *vcpu;
5794 mutex_lock(&kvm->lock);
5795 kvm_for_each_vcpu(i, vcpu, kvm) {
5796 if (!vcpu->arch.pv_time_enabled)
5799 ret = kvm_set_guest_paused(vcpu);
5801 kvm_err("Failed to pause guest VCPU%d: %d\n",
5802 vcpu->vcpu_id, ret);
5806 mutex_unlock(&kvm->lock);
5808 return ret ? NOTIFY_BAD : NOTIFY_DONE;
5811 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5814 case PM_HIBERNATION_PREPARE:
5815 case PM_SUSPEND_PREPARE:
5816 return kvm_arch_suspend_notifier(kvm);
5821 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5823 long kvm_arch_vm_ioctl(struct file *filp,
5824 unsigned int ioctl, unsigned long arg)
5826 struct kvm *kvm = filp->private_data;
5827 void __user *argp = (void __user *)arg;
5830 * This union makes it completely explicit to gcc-3.x
5831 * that these two variables' stack usage should be
5832 * combined, not added together.
5835 struct kvm_pit_state ps;
5836 struct kvm_pit_state2 ps2;
5837 struct kvm_pit_config pit_config;
5841 case KVM_SET_TSS_ADDR:
5842 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5844 case KVM_SET_IDENTITY_MAP_ADDR: {
5847 mutex_lock(&kvm->lock);
5849 if (kvm->created_vcpus)
5850 goto set_identity_unlock;
5852 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5853 goto set_identity_unlock;
5854 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5855 set_identity_unlock:
5856 mutex_unlock(&kvm->lock);
5859 case KVM_SET_NR_MMU_PAGES:
5860 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5862 case KVM_GET_NR_MMU_PAGES:
5863 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5865 case KVM_CREATE_IRQCHIP: {
5866 mutex_lock(&kvm->lock);
5869 if (irqchip_in_kernel(kvm))
5870 goto create_irqchip_unlock;
5873 if (kvm->created_vcpus)
5874 goto create_irqchip_unlock;
5876 r = kvm_pic_init(kvm);
5878 goto create_irqchip_unlock;
5880 r = kvm_ioapic_init(kvm);
5882 kvm_pic_destroy(kvm);
5883 goto create_irqchip_unlock;
5886 r = kvm_setup_default_irq_routing(kvm);
5888 kvm_ioapic_destroy(kvm);
5889 kvm_pic_destroy(kvm);
5890 goto create_irqchip_unlock;
5892 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5894 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5895 create_irqchip_unlock:
5896 mutex_unlock(&kvm->lock);
5899 case KVM_CREATE_PIT:
5900 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5902 case KVM_CREATE_PIT2:
5904 if (copy_from_user(&u.pit_config, argp,
5905 sizeof(struct kvm_pit_config)))
5908 mutex_lock(&kvm->lock);
5911 goto create_pit_unlock;
5913 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5917 mutex_unlock(&kvm->lock);
5919 case KVM_GET_IRQCHIP: {
5920 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5921 struct kvm_irqchip *chip;
5923 chip = memdup_user(argp, sizeof(*chip));
5930 if (!irqchip_kernel(kvm))
5931 goto get_irqchip_out;
5932 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5934 goto get_irqchip_out;
5936 if (copy_to_user(argp, chip, sizeof(*chip)))
5937 goto get_irqchip_out;
5943 case KVM_SET_IRQCHIP: {
5944 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5945 struct kvm_irqchip *chip;
5947 chip = memdup_user(argp, sizeof(*chip));
5954 if (!irqchip_kernel(kvm))
5955 goto set_irqchip_out;
5956 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5963 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5966 if (!kvm->arch.vpit)
5968 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5972 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5979 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5981 mutex_lock(&kvm->lock);
5983 if (!kvm->arch.vpit)
5985 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5987 mutex_unlock(&kvm->lock);
5990 case KVM_GET_PIT2: {
5992 if (!kvm->arch.vpit)
5994 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5998 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6003 case KVM_SET_PIT2: {
6005 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6007 mutex_lock(&kvm->lock);
6009 if (!kvm->arch.vpit)
6011 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6013 mutex_unlock(&kvm->lock);
6016 case KVM_REINJECT_CONTROL: {
6017 struct kvm_reinject_control control;
6019 if (copy_from_user(&control, argp, sizeof(control)))
6022 if (!kvm->arch.vpit)
6024 r = kvm_vm_ioctl_reinject(kvm, &control);
6027 case KVM_SET_BOOT_CPU_ID:
6029 mutex_lock(&kvm->lock);
6030 if (kvm->created_vcpus)
6033 kvm->arch.bsp_vcpu_id = arg;
6034 mutex_unlock(&kvm->lock);
6036 #ifdef CONFIG_KVM_XEN
6037 case KVM_XEN_HVM_CONFIG: {
6038 struct kvm_xen_hvm_config xhc;
6040 if (copy_from_user(&xhc, argp, sizeof(xhc)))
6042 r = kvm_xen_hvm_config(kvm, &xhc);
6045 case KVM_XEN_HVM_GET_ATTR: {
6046 struct kvm_xen_hvm_attr xha;
6049 if (copy_from_user(&xha, argp, sizeof(xha)))
6051 r = kvm_xen_hvm_get_attr(kvm, &xha);
6052 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6056 case KVM_XEN_HVM_SET_ATTR: {
6057 struct kvm_xen_hvm_attr xha;
6060 if (copy_from_user(&xha, argp, sizeof(xha)))
6062 r = kvm_xen_hvm_set_attr(kvm, &xha);
6066 case KVM_SET_CLOCK: {
6067 struct kvm_arch *ka = &kvm->arch;
6068 struct kvm_clock_data user_ns;
6072 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
6081 * TODO: userspace has to take care of races with VCPU_RUN, so
6082 * kvm_gen_update_masterclock() can be cut down to locked
6083 * pvclock_update_vm_gtod_copy().
6085 kvm_gen_update_masterclock(kvm);
6088 * This pairs with kvm_guest_time_update(): when masterclock is
6089 * in use, we use master_kernel_ns + kvmclock_offset to set
6090 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6091 * is slightly ahead) here we risk going negative on unsigned
6092 * 'system_time' when 'user_ns.clock' is very small.
6094 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
6095 if (kvm->arch.use_master_clock)
6096 now_ns = ka->master_kernel_ns;
6098 now_ns = get_kvmclock_base_ns();
6099 ka->kvmclock_offset = user_ns.clock - now_ns;
6100 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
6102 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
6105 case KVM_GET_CLOCK: {
6106 struct kvm_clock_data user_ns;
6109 now_ns = get_kvmclock_ns(kvm);
6110 user_ns.clock = now_ns;
6111 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
6112 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
6115 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
6120 case KVM_MEMORY_ENCRYPT_OP: {
6122 if (kvm_x86_ops.mem_enc_op)
6123 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6126 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6127 struct kvm_enc_region region;
6130 if (copy_from_user(®ion, argp, sizeof(region)))
6134 if (kvm_x86_ops.mem_enc_reg_region)
6135 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
6138 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6139 struct kvm_enc_region region;
6142 if (copy_from_user(®ion, argp, sizeof(region)))
6146 if (kvm_x86_ops.mem_enc_unreg_region)
6147 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
6150 case KVM_HYPERV_EVENTFD: {
6151 struct kvm_hyperv_eventfd hvevfd;
6154 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6156 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6159 case KVM_SET_PMU_EVENT_FILTER:
6160 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6162 case KVM_X86_SET_MSR_FILTER:
6163 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6172 static void kvm_init_msr_list(void)
6174 struct x86_pmu_capability x86_pmu;
6178 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6179 "Please update the fixed PMCs in msrs_to_saved_all[]");
6181 perf_get_x86_pmu_capability(&x86_pmu);
6183 num_msrs_to_save = 0;
6184 num_emulated_msrs = 0;
6185 num_msr_based_features = 0;
6187 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6188 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6192 * Even MSRs that are valid in the host may not be exposed
6193 * to the guests in some cases.
6195 switch (msrs_to_save_all[i]) {
6196 case MSR_IA32_BNDCFGS:
6197 if (!kvm_mpx_supported())
6201 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6202 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6205 case MSR_IA32_UMWAIT_CONTROL:
6206 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6209 case MSR_IA32_RTIT_CTL:
6210 case MSR_IA32_RTIT_STATUS:
6211 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6214 case MSR_IA32_RTIT_CR3_MATCH:
6215 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6216 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6219 case MSR_IA32_RTIT_OUTPUT_BASE:
6220 case MSR_IA32_RTIT_OUTPUT_MASK:
6221 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6222 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6223 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6226 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6227 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6228 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6229 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6232 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6233 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6234 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6237 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6238 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6239 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6246 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6249 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6250 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6253 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6256 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6257 struct kvm_msr_entry msr;
6259 msr.index = msr_based_features_all[i];
6260 if (kvm_get_msr_feature(&msr))
6263 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6267 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6275 if (!(lapic_in_kernel(vcpu) &&
6276 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6277 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6288 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6295 if (!(lapic_in_kernel(vcpu) &&
6296 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6298 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6300 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6310 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6311 struct kvm_segment *var, int seg)
6313 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6316 void kvm_get_segment(struct kvm_vcpu *vcpu,
6317 struct kvm_segment *var, int seg)
6319 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6322 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6323 struct x86_exception *exception)
6327 BUG_ON(!mmu_is_nested(vcpu));
6329 /* NPT walks are always user-walks */
6330 access |= PFERR_USER_MASK;
6331 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6336 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6337 struct x86_exception *exception)
6339 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6340 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6342 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6344 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6345 struct x86_exception *exception)
6347 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6348 access |= PFERR_FETCH_MASK;
6349 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6352 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6353 struct x86_exception *exception)
6355 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6356 access |= PFERR_WRITE_MASK;
6357 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6359 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6361 /* uses this to access any guest's mapped memory without checking CPL */
6362 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6363 struct x86_exception *exception)
6365 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6368 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6369 struct kvm_vcpu *vcpu, u32 access,
6370 struct x86_exception *exception)
6373 int r = X86EMUL_CONTINUE;
6376 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6378 unsigned offset = addr & (PAGE_SIZE-1);
6379 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6382 if (gpa == UNMAPPED_GVA)
6383 return X86EMUL_PROPAGATE_FAULT;
6384 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6387 r = X86EMUL_IO_NEEDED;
6399 /* used for instruction fetching */
6400 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6401 gva_t addr, void *val, unsigned int bytes,
6402 struct x86_exception *exception)
6404 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6405 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6409 /* Inline kvm_read_guest_virt_helper for speed. */
6410 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6412 if (unlikely(gpa == UNMAPPED_GVA))
6413 return X86EMUL_PROPAGATE_FAULT;
6415 offset = addr & (PAGE_SIZE-1);
6416 if (WARN_ON(offset + bytes > PAGE_SIZE))
6417 bytes = (unsigned)PAGE_SIZE - offset;
6418 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6420 if (unlikely(ret < 0))
6421 return X86EMUL_IO_NEEDED;
6423 return X86EMUL_CONTINUE;
6426 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6427 gva_t addr, void *val, unsigned int bytes,
6428 struct x86_exception *exception)
6430 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6433 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6434 * is returned, but our callers are not ready for that and they blindly
6435 * call kvm_inject_page_fault. Ensure that they at least do not leak
6436 * uninitialized kernel stack memory into cr2 and error code.
6438 memset(exception, 0, sizeof(*exception));
6439 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6442 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6444 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6445 gva_t addr, void *val, unsigned int bytes,
6446 struct x86_exception *exception, bool system)
6448 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6451 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6452 access |= PFERR_USER_MASK;
6454 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6457 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6458 unsigned long addr, void *val, unsigned int bytes)
6460 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6461 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6463 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6466 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6467 struct kvm_vcpu *vcpu, u32 access,
6468 struct x86_exception *exception)
6471 int r = X86EMUL_CONTINUE;
6474 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6477 unsigned offset = addr & (PAGE_SIZE-1);
6478 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6481 if (gpa == UNMAPPED_GVA)
6482 return X86EMUL_PROPAGATE_FAULT;
6483 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6485 r = X86EMUL_IO_NEEDED;
6497 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6498 unsigned int bytes, struct x86_exception *exception,
6501 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6502 u32 access = PFERR_WRITE_MASK;
6504 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6505 access |= PFERR_USER_MASK;
6507 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6511 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6512 unsigned int bytes, struct x86_exception *exception)
6514 /* kvm_write_guest_virt_system can pull in tons of pages. */
6515 vcpu->arch.l1tf_flush_l1d = true;
6517 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6518 PFERR_WRITE_MASK, exception);
6520 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6522 int handle_ud(struct kvm_vcpu *vcpu)
6524 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6525 int emul_type = EMULTYPE_TRAP_UD;
6526 char sig[5]; /* ud2; .ascii "kvm" */
6527 struct x86_exception e;
6529 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6532 if (force_emulation_prefix &&
6533 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6534 sig, sizeof(sig), &e) == 0 &&
6535 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6536 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6537 emul_type = EMULTYPE_TRAP_UD_FORCED;
6540 return kvm_emulate_instruction(vcpu, emul_type);
6542 EXPORT_SYMBOL_GPL(handle_ud);
6544 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6545 gpa_t gpa, bool write)
6547 /* For APIC access vmexit */
6548 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6551 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6552 trace_vcpu_match_mmio(gva, gpa, write, true);
6559 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6560 gpa_t *gpa, struct x86_exception *exception,
6563 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6564 | (write ? PFERR_WRITE_MASK : 0);
6567 * currently PKRU is only applied to ept enabled guest so
6568 * there is no pkey in EPT page table for L1 guest or EPT
6569 * shadow page table for L2 guest.
6571 if (vcpu_match_mmio_gva(vcpu, gva)
6572 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6573 vcpu->arch.mmio_access, 0, access)) {
6574 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6575 (gva & (PAGE_SIZE - 1));
6576 trace_vcpu_match_mmio(gva, *gpa, write, false);
6580 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6582 if (*gpa == UNMAPPED_GVA)
6585 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6588 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6589 const void *val, int bytes)
6593 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6596 kvm_page_track_write(vcpu, gpa, val, bytes);
6600 struct read_write_emulator_ops {
6601 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6603 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6604 void *val, int bytes);
6605 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6606 int bytes, void *val);
6607 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6608 void *val, int bytes);
6612 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6614 if (vcpu->mmio_read_completed) {
6615 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6616 vcpu->mmio_fragments[0].gpa, val);
6617 vcpu->mmio_read_completed = 0;
6624 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6625 void *val, int bytes)
6627 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6630 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6631 void *val, int bytes)
6633 return emulator_write_phys(vcpu, gpa, val, bytes);
6636 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6638 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6639 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6642 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6643 void *val, int bytes)
6645 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6646 return X86EMUL_IO_NEEDED;
6649 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6650 void *val, int bytes)
6652 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6654 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6655 return X86EMUL_CONTINUE;
6658 static const struct read_write_emulator_ops read_emultor = {
6659 .read_write_prepare = read_prepare,
6660 .read_write_emulate = read_emulate,
6661 .read_write_mmio = vcpu_mmio_read,
6662 .read_write_exit_mmio = read_exit_mmio,
6665 static const struct read_write_emulator_ops write_emultor = {
6666 .read_write_emulate = write_emulate,
6667 .read_write_mmio = write_mmio,
6668 .read_write_exit_mmio = write_exit_mmio,
6672 static int emulator_read_write_onepage(unsigned long addr, void *val,
6674 struct x86_exception *exception,
6675 struct kvm_vcpu *vcpu,
6676 const struct read_write_emulator_ops *ops)
6680 bool write = ops->write;
6681 struct kvm_mmio_fragment *frag;
6682 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6685 * If the exit was due to a NPF we may already have a GPA.
6686 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6687 * Note, this cannot be used on string operations since string
6688 * operation using rep will only have the initial GPA from the NPF
6691 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6692 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6693 gpa = ctxt->gpa_val;
6694 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6696 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6698 return X86EMUL_PROPAGATE_FAULT;
6701 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6702 return X86EMUL_CONTINUE;
6705 * Is this MMIO handled locally?
6707 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6708 if (handled == bytes)
6709 return X86EMUL_CONTINUE;
6715 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6716 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6720 return X86EMUL_CONTINUE;
6723 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6725 void *val, unsigned int bytes,
6726 struct x86_exception *exception,
6727 const struct read_write_emulator_ops *ops)
6729 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6733 if (ops->read_write_prepare &&
6734 ops->read_write_prepare(vcpu, val, bytes))
6735 return X86EMUL_CONTINUE;
6737 vcpu->mmio_nr_fragments = 0;
6739 /* Crossing a page boundary? */
6740 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6743 now = -addr & ~PAGE_MASK;
6744 rc = emulator_read_write_onepage(addr, val, now, exception,
6747 if (rc != X86EMUL_CONTINUE)
6750 if (ctxt->mode != X86EMUL_MODE_PROT64)
6756 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6758 if (rc != X86EMUL_CONTINUE)
6761 if (!vcpu->mmio_nr_fragments)
6764 gpa = vcpu->mmio_fragments[0].gpa;
6766 vcpu->mmio_needed = 1;
6767 vcpu->mmio_cur_fragment = 0;
6769 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6770 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6771 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6772 vcpu->run->mmio.phys_addr = gpa;
6774 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6777 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6781 struct x86_exception *exception)
6783 return emulator_read_write(ctxt, addr, val, bytes,
6784 exception, &read_emultor);
6787 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6791 struct x86_exception *exception)
6793 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6794 exception, &write_emultor);
6797 #define CMPXCHG_TYPE(t, ptr, old, new) \
6798 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6800 #ifdef CONFIG_X86_64
6801 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6803 # define CMPXCHG64(ptr, old, new) \
6804 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6807 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6812 struct x86_exception *exception)
6814 struct kvm_host_map map;
6815 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6821 /* guests cmpxchg8b have to be emulated atomically */
6822 if (bytes > 8 || (bytes & (bytes - 1)))
6825 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6827 if (gpa == UNMAPPED_GVA ||
6828 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6832 * Emulate the atomic as a straight write to avoid #AC if SLD is
6833 * enabled in the host and the access splits a cache line.
6835 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6836 page_line_mask = ~(cache_line_size() - 1);
6838 page_line_mask = PAGE_MASK;
6840 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6843 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6846 kaddr = map.hva + offset_in_page(gpa);
6850 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6853 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6856 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6859 exchanged = CMPXCHG64(kaddr, old, new);
6865 kvm_vcpu_unmap(vcpu, &map, true);
6868 return X86EMUL_CMPXCHG_FAILED;
6870 kvm_page_track_write(vcpu, gpa, new, bytes);
6872 return X86EMUL_CONTINUE;
6875 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6877 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6880 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6884 for (i = 0; i < vcpu->arch.pio.count; i++) {
6885 if (vcpu->arch.pio.in)
6886 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6887 vcpu->arch.pio.size, pd);
6889 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6890 vcpu->arch.pio.port, vcpu->arch.pio.size,
6894 pd += vcpu->arch.pio.size;
6899 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6900 unsigned short port, void *val,
6901 unsigned int count, bool in)
6903 vcpu->arch.pio.port = port;
6904 vcpu->arch.pio.in = in;
6905 vcpu->arch.pio.count = count;
6906 vcpu->arch.pio.size = size;
6908 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6909 vcpu->arch.pio.count = 0;
6913 vcpu->run->exit_reason = KVM_EXIT_IO;
6914 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6915 vcpu->run->io.size = size;
6916 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6917 vcpu->run->io.count = count;
6918 vcpu->run->io.port = port;
6923 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6924 unsigned short port, void *val, unsigned int count)
6928 if (vcpu->arch.pio.count)
6931 memset(vcpu->arch.pio_data, 0, size * count);
6933 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6936 memcpy(val, vcpu->arch.pio_data, size * count);
6937 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6938 vcpu->arch.pio.count = 0;
6945 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6946 int size, unsigned short port, void *val,
6949 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6953 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6954 unsigned short port, const void *val,
6957 memcpy(vcpu->arch.pio_data, val, size * count);
6958 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6959 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6962 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6963 int size, unsigned short port,
6964 const void *val, unsigned int count)
6966 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6969 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6971 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6974 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6976 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6979 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6981 if (!need_emulate_wbinvd(vcpu))
6982 return X86EMUL_CONTINUE;
6984 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6985 int cpu = get_cpu();
6987 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6988 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6989 wbinvd_ipi, NULL, 1);
6991 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6994 return X86EMUL_CONTINUE;
6997 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6999 kvm_emulate_wbinvd_noskip(vcpu);
7000 return kvm_skip_emulated_instruction(vcpu);
7002 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7006 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7008 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7011 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7012 unsigned long *dest)
7014 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7017 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7018 unsigned long value)
7021 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7024 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7026 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7029 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7031 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7032 unsigned long value;
7036 value = kvm_read_cr0(vcpu);
7039 value = vcpu->arch.cr2;
7042 value = kvm_read_cr3(vcpu);
7045 value = kvm_read_cr4(vcpu);
7048 value = kvm_get_cr8(vcpu);
7051 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7058 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7060 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7065 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7068 vcpu->arch.cr2 = val;
7071 res = kvm_set_cr3(vcpu, val);
7074 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7077 res = kvm_set_cr8(vcpu, val);
7080 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7087 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7089 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7092 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7094 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7097 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7099 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7102 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7104 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7107 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7109 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7112 static unsigned long emulator_get_cached_segment_base(
7113 struct x86_emulate_ctxt *ctxt, int seg)
7115 return get_segment_base(emul_to_vcpu(ctxt), seg);
7118 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7119 struct desc_struct *desc, u32 *base3,
7122 struct kvm_segment var;
7124 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7125 *selector = var.selector;
7128 memset(desc, 0, sizeof(*desc));
7136 set_desc_limit(desc, var.limit);
7137 set_desc_base(desc, (unsigned long)var.base);
7138 #ifdef CONFIG_X86_64
7140 *base3 = var.base >> 32;
7142 desc->type = var.type;
7144 desc->dpl = var.dpl;
7145 desc->p = var.present;
7146 desc->avl = var.avl;
7154 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7155 struct desc_struct *desc, u32 base3,
7158 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7159 struct kvm_segment var;
7161 var.selector = selector;
7162 var.base = get_desc_base(desc);
7163 #ifdef CONFIG_X86_64
7164 var.base |= ((u64)base3) << 32;
7166 var.limit = get_desc_limit(desc);
7168 var.limit = (var.limit << 12) | 0xfff;
7169 var.type = desc->type;
7170 var.dpl = desc->dpl;
7175 var.avl = desc->avl;
7176 var.present = desc->p;
7177 var.unusable = !var.present;
7180 kvm_set_segment(vcpu, &var, seg);
7184 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7185 u32 msr_index, u64 *pdata)
7187 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7190 r = kvm_get_msr(vcpu, msr_index, pdata);
7192 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7193 /* Bounce to user space */
7194 return X86EMUL_IO_NEEDED;
7200 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7201 u32 msr_index, u64 data)
7203 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7206 r = kvm_set_msr(vcpu, msr_index, data);
7208 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7209 /* Bounce to user space */
7210 return X86EMUL_IO_NEEDED;
7216 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7218 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7220 return vcpu->arch.smbase;
7223 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7225 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7227 vcpu->arch.smbase = smbase;
7230 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7233 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7236 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7237 u32 pmc, u64 *pdata)
7239 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7242 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7244 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7247 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7248 struct x86_instruction_info *info,
7249 enum x86_intercept_stage stage)
7251 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7255 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7256 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7259 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7262 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7264 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7267 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7269 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7272 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7274 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7277 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7279 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7282 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7284 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7287 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7289 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7292 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7294 return emul_to_vcpu(ctxt)->arch.hflags;
7297 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7299 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7301 kvm_smm_changed(vcpu, false);
7304 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7305 const char *smstate)
7307 return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7310 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7312 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7315 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7317 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7320 static const struct x86_emulate_ops emulate_ops = {
7321 .read_gpr = emulator_read_gpr,
7322 .write_gpr = emulator_write_gpr,
7323 .read_std = emulator_read_std,
7324 .write_std = emulator_write_std,
7325 .read_phys = kvm_read_guest_phys_system,
7326 .fetch = kvm_fetch_guest_virt,
7327 .read_emulated = emulator_read_emulated,
7328 .write_emulated = emulator_write_emulated,
7329 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7330 .invlpg = emulator_invlpg,
7331 .pio_in_emulated = emulator_pio_in_emulated,
7332 .pio_out_emulated = emulator_pio_out_emulated,
7333 .get_segment = emulator_get_segment,
7334 .set_segment = emulator_set_segment,
7335 .get_cached_segment_base = emulator_get_cached_segment_base,
7336 .get_gdt = emulator_get_gdt,
7337 .get_idt = emulator_get_idt,
7338 .set_gdt = emulator_set_gdt,
7339 .set_idt = emulator_set_idt,
7340 .get_cr = emulator_get_cr,
7341 .set_cr = emulator_set_cr,
7342 .cpl = emulator_get_cpl,
7343 .get_dr = emulator_get_dr,
7344 .set_dr = emulator_set_dr,
7345 .get_smbase = emulator_get_smbase,
7346 .set_smbase = emulator_set_smbase,
7347 .set_msr = emulator_set_msr,
7348 .get_msr = emulator_get_msr,
7349 .check_pmc = emulator_check_pmc,
7350 .read_pmc = emulator_read_pmc,
7351 .halt = emulator_halt,
7352 .wbinvd = emulator_wbinvd,
7353 .fix_hypercall = emulator_fix_hypercall,
7354 .intercept = emulator_intercept,
7355 .get_cpuid = emulator_get_cpuid,
7356 .guest_has_long_mode = emulator_guest_has_long_mode,
7357 .guest_has_movbe = emulator_guest_has_movbe,
7358 .guest_has_fxsr = emulator_guest_has_fxsr,
7359 .set_nmi_mask = emulator_set_nmi_mask,
7360 .get_hflags = emulator_get_hflags,
7361 .exiting_smm = emulator_exiting_smm,
7362 .leave_smm = emulator_leave_smm,
7363 .triple_fault = emulator_triple_fault,
7364 .set_xcr = emulator_set_xcr,
7367 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7369 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7371 * an sti; sti; sequence only disable interrupts for the first
7372 * instruction. So, if the last instruction, be it emulated or
7373 * not, left the system with the INT_STI flag enabled, it
7374 * means that the last instruction is an sti. We should not
7375 * leave the flag on in this case. The same goes for mov ss
7377 if (int_shadow & mask)
7379 if (unlikely(int_shadow || mask)) {
7380 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7382 kvm_make_request(KVM_REQ_EVENT, vcpu);
7386 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7388 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7389 if (ctxt->exception.vector == PF_VECTOR)
7390 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7392 if (ctxt->exception.error_code_valid)
7393 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7394 ctxt->exception.error_code);
7396 kvm_queue_exception(vcpu, ctxt->exception.vector);
7400 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7402 struct x86_emulate_ctxt *ctxt;
7404 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7406 pr_err("kvm: failed to allocate vcpu's emulator\n");
7411 ctxt->ops = &emulate_ops;
7412 vcpu->arch.emulate_ctxt = ctxt;
7417 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7419 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7422 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7424 ctxt->gpa_available = false;
7425 ctxt->eflags = kvm_get_rflags(vcpu);
7426 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7428 ctxt->eip = kvm_rip_read(vcpu);
7429 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7430 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7431 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7432 cs_db ? X86EMUL_MODE_PROT32 :
7433 X86EMUL_MODE_PROT16;
7434 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7435 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7436 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7438 ctxt->interruptibility = 0;
7439 ctxt->have_exception = false;
7440 ctxt->exception.vector = -1;
7441 ctxt->perm_ok = false;
7443 init_decode_cache(ctxt);
7444 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7447 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7449 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7452 init_emulate_ctxt(vcpu);
7456 ctxt->_eip = ctxt->eip + inc_eip;
7457 ret = emulate_int_real(ctxt, irq);
7459 if (ret != X86EMUL_CONTINUE) {
7460 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7462 ctxt->eip = ctxt->_eip;
7463 kvm_rip_write(vcpu, ctxt->eip);
7464 kvm_set_rflags(vcpu, ctxt->eflags);
7467 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7469 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7471 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7472 u32 insn_size = ctxt->fetch.end - ctxt->fetch.data;
7473 struct kvm_run *run = vcpu->run;
7475 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7476 run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7477 run->emulation_failure.ndata = 0;
7478 run->emulation_failure.flags = 0;
7481 run->emulation_failure.ndata = 3;
7482 run->emulation_failure.flags |=
7483 KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7484 run->emulation_failure.insn_size = insn_size;
7485 memset(run->emulation_failure.insn_bytes, 0x90,
7486 sizeof(run->emulation_failure.insn_bytes));
7487 memcpy(run->emulation_failure.insn_bytes,
7488 ctxt->fetch.data, insn_size);
7492 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7494 struct kvm *kvm = vcpu->kvm;
7496 ++vcpu->stat.insn_emulation_fail;
7497 trace_kvm_emulate_insn_failed(vcpu);
7499 if (emulation_type & EMULTYPE_VMWARE_GP) {
7500 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7504 if (kvm->arch.exit_on_emulation_error ||
7505 (emulation_type & EMULTYPE_SKIP)) {
7506 prepare_emulation_failure_exit(vcpu);
7510 kvm_queue_exception(vcpu, UD_VECTOR);
7512 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7513 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7514 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7515 vcpu->run->internal.ndata = 0;
7522 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7523 bool write_fault_to_shadow_pgtable,
7526 gpa_t gpa = cr2_or_gpa;
7529 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7532 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7533 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7536 if (!vcpu->arch.mmu->direct_map) {
7538 * Write permission should be allowed since only
7539 * write access need to be emulated.
7541 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7544 * If the mapping is invalid in guest, let cpu retry
7545 * it to generate fault.
7547 if (gpa == UNMAPPED_GVA)
7552 * Do not retry the unhandleable instruction if it faults on the
7553 * readonly host memory, otherwise it will goto a infinite loop:
7554 * retry instruction -> write #PF -> emulation fail -> retry
7555 * instruction -> ...
7557 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7560 * If the instruction failed on the error pfn, it can not be fixed,
7561 * report the error to userspace.
7563 if (is_error_noslot_pfn(pfn))
7566 kvm_release_pfn_clean(pfn);
7568 /* The instructions are well-emulated on direct mmu. */
7569 if (vcpu->arch.mmu->direct_map) {
7570 unsigned int indirect_shadow_pages;
7572 write_lock(&vcpu->kvm->mmu_lock);
7573 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7574 write_unlock(&vcpu->kvm->mmu_lock);
7576 if (indirect_shadow_pages)
7577 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7583 * if emulation was due to access to shadowed page table
7584 * and it failed try to unshadow page and re-enter the
7585 * guest to let CPU execute the instruction.
7587 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7590 * If the access faults on its page table, it can not
7591 * be fixed by unprotecting shadow page and it should
7592 * be reported to userspace.
7594 return !write_fault_to_shadow_pgtable;
7597 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7598 gpa_t cr2_or_gpa, int emulation_type)
7600 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7601 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7603 last_retry_eip = vcpu->arch.last_retry_eip;
7604 last_retry_addr = vcpu->arch.last_retry_addr;
7607 * If the emulation is caused by #PF and it is non-page_table
7608 * writing instruction, it means the VM-EXIT is caused by shadow
7609 * page protected, we can zap the shadow page and retry this
7610 * instruction directly.
7612 * Note: if the guest uses a non-page-table modifying instruction
7613 * on the PDE that points to the instruction, then we will unmap
7614 * the instruction and go to an infinite loop. So, we cache the
7615 * last retried eip and the last fault address, if we meet the eip
7616 * and the address again, we can break out of the potential infinite
7619 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7621 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7624 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7625 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7628 if (x86_page_table_writing_insn(ctxt))
7631 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7634 vcpu->arch.last_retry_eip = ctxt->eip;
7635 vcpu->arch.last_retry_addr = cr2_or_gpa;
7637 if (!vcpu->arch.mmu->direct_map)
7638 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7640 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7645 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7646 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7648 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7650 trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7653 vcpu->arch.hflags |= HF_SMM_MASK;
7655 vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7657 /* Process a latched INIT or SMI, if any. */
7658 kvm_make_request(KVM_REQ_EVENT, vcpu);
7661 kvm_mmu_reset_context(vcpu);
7664 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7673 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7674 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7679 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7681 struct kvm_run *kvm_run = vcpu->run;
7683 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7684 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7685 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7686 kvm_run->debug.arch.exception = DB_VECTOR;
7687 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7690 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7694 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7696 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7699 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7704 * rflags is the old, "raw" value of the flags. The new value has
7705 * not been saved yet.
7707 * This is correct even for TF set by the guest, because "the
7708 * processor will not generate this exception after the instruction
7709 * that sets the TF flag".
7711 if (unlikely(rflags & X86_EFLAGS_TF))
7712 r = kvm_vcpu_do_singlestep(vcpu);
7715 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7717 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7719 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7720 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7721 struct kvm_run *kvm_run = vcpu->run;
7722 unsigned long eip = kvm_get_linear_rip(vcpu);
7723 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7724 vcpu->arch.guest_debug_dr7,
7728 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7729 kvm_run->debug.arch.pc = eip;
7730 kvm_run->debug.arch.exception = DB_VECTOR;
7731 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7737 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7738 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7739 unsigned long eip = kvm_get_linear_rip(vcpu);
7740 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7745 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7754 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7756 switch (ctxt->opcode_len) {
7763 case 0xe6: /* OUT */
7767 case 0x6c: /* INS */
7769 case 0x6e: /* OUTS */
7776 case 0x33: /* RDPMC */
7786 * Decode to be emulated instruction. Return EMULATION_OK if success.
7788 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7789 void *insn, int insn_len)
7791 int r = EMULATION_OK;
7792 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7794 init_emulate_ctxt(vcpu);
7797 * We will reenter on the same instruction since we do not set
7798 * complete_userspace_io. This does not handle watchpoints yet,
7799 * those would be handled in the emulate_ops.
7801 if (!(emulation_type & EMULTYPE_SKIP) &&
7802 kvm_vcpu_check_breakpoint(vcpu, &r))
7805 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7807 trace_kvm_emulate_insn_start(vcpu);
7808 ++vcpu->stat.insn_emulation;
7812 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7814 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7815 int emulation_type, void *insn, int insn_len)
7818 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7819 bool writeback = true;
7820 bool write_fault_to_spt;
7822 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7825 vcpu->arch.l1tf_flush_l1d = true;
7828 * Clear write_fault_to_shadow_pgtable here to ensure it is
7831 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7832 vcpu->arch.write_fault_to_shadow_pgtable = false;
7834 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7835 kvm_clear_exception_queue(vcpu);
7837 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7839 if (r != EMULATION_OK) {
7840 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7841 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7842 kvm_queue_exception(vcpu, UD_VECTOR);
7845 if (reexecute_instruction(vcpu, cr2_or_gpa,
7849 if (ctxt->have_exception) {
7851 * #UD should result in just EMULATION_FAILED, and trap-like
7852 * exception should not be encountered during decode.
7854 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7855 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7856 inject_emulated_exception(vcpu);
7859 return handle_emulation_failure(vcpu, emulation_type);
7863 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7864 !is_vmware_backdoor_opcode(ctxt)) {
7865 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7870 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7871 * for kvm_skip_emulated_instruction(). The caller is responsible for
7872 * updating interruptibility state and injecting single-step #DBs.
7874 if (emulation_type & EMULTYPE_SKIP) {
7875 kvm_rip_write(vcpu, ctxt->_eip);
7876 if (ctxt->eflags & X86_EFLAGS_RF)
7877 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7881 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7884 /* this is needed for vmware backdoor interface to work since it
7885 changes registers values during IO operation */
7886 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7887 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7888 emulator_invalidate_register_cache(ctxt);
7892 if (emulation_type & EMULTYPE_PF) {
7893 /* Save the faulting GPA (cr2) in the address field */
7894 ctxt->exception.address = cr2_or_gpa;
7896 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7897 if (vcpu->arch.mmu->direct_map) {
7898 ctxt->gpa_available = true;
7899 ctxt->gpa_val = cr2_or_gpa;
7902 /* Sanitize the address out of an abundance of paranoia. */
7903 ctxt->exception.address = 0;
7906 r = x86_emulate_insn(ctxt);
7908 if (r == EMULATION_INTERCEPTED)
7911 if (r == EMULATION_FAILED) {
7912 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7916 return handle_emulation_failure(vcpu, emulation_type);
7919 if (ctxt->have_exception) {
7921 if (inject_emulated_exception(vcpu))
7923 } else if (vcpu->arch.pio.count) {
7924 if (!vcpu->arch.pio.in) {
7925 /* FIXME: return into emulator if single-stepping. */
7926 vcpu->arch.pio.count = 0;
7929 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7932 } else if (vcpu->mmio_needed) {
7933 ++vcpu->stat.mmio_exits;
7935 if (!vcpu->mmio_is_write)
7938 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7939 } else if (r == EMULATION_RESTART)
7945 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7946 toggle_interruptibility(vcpu, ctxt->interruptibility);
7947 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7948 if (!ctxt->have_exception ||
7949 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7950 kvm_rip_write(vcpu, ctxt->eip);
7951 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7952 r = kvm_vcpu_do_singlestep(vcpu);
7953 if (kvm_x86_ops.update_emulated_instruction)
7954 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7955 __kvm_set_rflags(vcpu, ctxt->eflags);
7959 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7960 * do nothing, and it will be requested again as soon as
7961 * the shadow expires. But we still need to check here,
7962 * because POPF has no interrupt shadow.
7964 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7965 kvm_make_request(KVM_REQ_EVENT, vcpu);
7967 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7972 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7974 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7976 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7978 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7979 void *insn, int insn_len)
7981 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7983 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7985 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7987 vcpu->arch.pio.count = 0;
7991 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7993 vcpu->arch.pio.count = 0;
7995 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7998 return kvm_skip_emulated_instruction(vcpu);
8001 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8002 unsigned short port)
8004 unsigned long val = kvm_rax_read(vcpu);
8005 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8011 * Workaround userspace that relies on old KVM behavior of %rip being
8012 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8015 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8016 vcpu->arch.complete_userspace_io =
8017 complete_fast_pio_out_port_0x7e;
8018 kvm_skip_emulated_instruction(vcpu);
8020 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8021 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8026 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8030 /* We should only ever be called with arch.pio.count equal to 1 */
8031 BUG_ON(vcpu->arch.pio.count != 1);
8033 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8034 vcpu->arch.pio.count = 0;
8038 /* For size less than 4 we merge, else we zero extend */
8039 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8042 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8043 * the copy and tracing
8045 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8046 kvm_rax_write(vcpu, val);
8048 return kvm_skip_emulated_instruction(vcpu);
8051 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8052 unsigned short port)
8057 /* For size less than 4 we merge, else we zero extend */
8058 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8060 ret = emulator_pio_in(vcpu, size, port, &val, 1);
8062 kvm_rax_write(vcpu, val);
8066 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8067 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8072 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8077 ret = kvm_fast_pio_in(vcpu, size, port);
8079 ret = kvm_fast_pio_out(vcpu, size, port);
8080 return ret && kvm_skip_emulated_instruction(vcpu);
8082 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8084 static int kvmclock_cpu_down_prep(unsigned int cpu)
8086 __this_cpu_write(cpu_tsc_khz, 0);
8090 static void tsc_khz_changed(void *data)
8092 struct cpufreq_freqs *freq = data;
8093 unsigned long khz = 0;
8097 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8098 khz = cpufreq_quick_get(raw_smp_processor_id());
8101 __this_cpu_write(cpu_tsc_khz, khz);
8104 #ifdef CONFIG_X86_64
8105 static void kvm_hyperv_tsc_notifier(void)
8108 struct kvm_vcpu *vcpu;
8110 unsigned long flags;
8112 mutex_lock(&kvm_lock);
8113 list_for_each_entry(kvm, &vm_list, vm_list)
8114 kvm_make_mclock_inprogress_request(kvm);
8116 hyperv_stop_tsc_emulation();
8118 /* TSC frequency always matches when on Hyper-V */
8119 for_each_present_cpu(cpu)
8120 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8121 kvm_max_guest_tsc_khz = tsc_khz;
8123 list_for_each_entry(kvm, &vm_list, vm_list) {
8124 struct kvm_arch *ka = &kvm->arch;
8126 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
8127 pvclock_update_vm_gtod_copy(kvm);
8128 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
8130 kvm_for_each_vcpu(cpu, vcpu, kvm)
8131 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8133 kvm_for_each_vcpu(cpu, vcpu, kvm)
8134 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
8136 mutex_unlock(&kvm_lock);
8140 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8143 struct kvm_vcpu *vcpu;
8144 int i, send_ipi = 0;
8147 * We allow guests to temporarily run on slowing clocks,
8148 * provided we notify them after, or to run on accelerating
8149 * clocks, provided we notify them before. Thus time never
8152 * However, we have a problem. We can't atomically update
8153 * the frequency of a given CPU from this function; it is
8154 * merely a notifier, which can be called from any CPU.
8155 * Changing the TSC frequency at arbitrary points in time
8156 * requires a recomputation of local variables related to
8157 * the TSC for each VCPU. We must flag these local variables
8158 * to be updated and be sure the update takes place with the
8159 * new frequency before any guests proceed.
8161 * Unfortunately, the combination of hotplug CPU and frequency
8162 * change creates an intractable locking scenario; the order
8163 * of when these callouts happen is undefined with respect to
8164 * CPU hotplug, and they can race with each other. As such,
8165 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8166 * undefined; you can actually have a CPU frequency change take
8167 * place in between the computation of X and the setting of the
8168 * variable. To protect against this problem, all updates of
8169 * the per_cpu tsc_khz variable are done in an interrupt
8170 * protected IPI, and all callers wishing to update the value
8171 * must wait for a synchronous IPI to complete (which is trivial
8172 * if the caller is on the CPU already). This establishes the
8173 * necessary total order on variable updates.
8175 * Note that because a guest time update may take place
8176 * anytime after the setting of the VCPU's request bit, the
8177 * correct TSC value must be set before the request. However,
8178 * to ensure the update actually makes it to any guest which
8179 * starts running in hardware virtualization between the set
8180 * and the acquisition of the spinlock, we must also ping the
8181 * CPU after setting the request bit.
8185 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8187 mutex_lock(&kvm_lock);
8188 list_for_each_entry(kvm, &vm_list, vm_list) {
8189 kvm_for_each_vcpu(i, vcpu, kvm) {
8190 if (vcpu->cpu != cpu)
8192 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8193 if (vcpu->cpu != raw_smp_processor_id())
8197 mutex_unlock(&kvm_lock);
8199 if (freq->old < freq->new && send_ipi) {
8201 * We upscale the frequency. Must make the guest
8202 * doesn't see old kvmclock values while running with
8203 * the new frequency, otherwise we risk the guest sees
8204 * time go backwards.
8206 * In case we update the frequency for another cpu
8207 * (which might be in guest context) send an interrupt
8208 * to kick the cpu out of guest context. Next time
8209 * guest context is entered kvmclock will be updated,
8210 * so the guest will not see stale values.
8212 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8216 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8219 struct cpufreq_freqs *freq = data;
8222 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8224 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8227 for_each_cpu(cpu, freq->policy->cpus)
8228 __kvmclock_cpufreq_notifier(freq, cpu);
8233 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8234 .notifier_call = kvmclock_cpufreq_notifier
8237 static int kvmclock_cpu_online(unsigned int cpu)
8239 tsc_khz_changed(NULL);
8243 static void kvm_timer_init(void)
8245 max_tsc_khz = tsc_khz;
8247 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8248 #ifdef CONFIG_CPU_FREQ
8249 struct cpufreq_policy *policy;
8253 policy = cpufreq_cpu_get(cpu);
8255 if (policy->cpuinfo.max_freq)
8256 max_tsc_khz = policy->cpuinfo.max_freq;
8257 cpufreq_cpu_put(policy);
8261 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8262 CPUFREQ_TRANSITION_NOTIFIER);
8265 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8266 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8269 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8270 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8272 int kvm_is_in_guest(void)
8274 return __this_cpu_read(current_vcpu) != NULL;
8277 static int kvm_is_user_mode(void)
8281 if (__this_cpu_read(current_vcpu))
8282 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8284 return user_mode != 0;
8287 static unsigned long kvm_get_guest_ip(void)
8289 unsigned long ip = 0;
8291 if (__this_cpu_read(current_vcpu))
8292 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8297 static void kvm_handle_intel_pt_intr(void)
8299 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8301 kvm_make_request(KVM_REQ_PMI, vcpu);
8302 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8303 (unsigned long *)&vcpu->arch.pmu.global_status);
8306 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8307 .is_in_guest = kvm_is_in_guest,
8308 .is_user_mode = kvm_is_user_mode,
8309 .get_guest_ip = kvm_get_guest_ip,
8310 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8313 #ifdef CONFIG_X86_64
8314 static void pvclock_gtod_update_fn(struct work_struct *work)
8318 struct kvm_vcpu *vcpu;
8321 mutex_lock(&kvm_lock);
8322 list_for_each_entry(kvm, &vm_list, vm_list)
8323 kvm_for_each_vcpu(i, vcpu, kvm)
8324 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8325 atomic_set(&kvm_guest_has_master_clock, 0);
8326 mutex_unlock(&kvm_lock);
8329 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8332 * Indirection to move queue_work() out of the tk_core.seq write held
8333 * region to prevent possible deadlocks against time accessors which
8334 * are invoked with work related locks held.
8336 static void pvclock_irq_work_fn(struct irq_work *w)
8338 queue_work(system_long_wq, &pvclock_gtod_work);
8341 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8344 * Notification about pvclock gtod data update.
8346 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8349 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8350 struct timekeeper *tk = priv;
8352 update_pvclock_gtod(tk);
8355 * Disable master clock if host does not trust, or does not use,
8356 * TSC based clocksource. Delegate queue_work() to irq_work as
8357 * this is invoked with tk_core.seq write held.
8359 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8360 atomic_read(&kvm_guest_has_master_clock) != 0)
8361 irq_work_queue(&pvclock_irq_work);
8365 static struct notifier_block pvclock_gtod_notifier = {
8366 .notifier_call = pvclock_gtod_notify,
8370 int kvm_arch_init(void *opaque)
8372 struct kvm_x86_init_ops *ops = opaque;
8375 if (kvm_x86_ops.hardware_enable) {
8376 printk(KERN_ERR "kvm: already loaded the other module\n");
8381 if (!ops->cpu_has_kvm_support()) {
8382 pr_err_ratelimited("kvm: no hardware support\n");
8386 if (ops->disabled_by_bios()) {
8387 pr_err_ratelimited("kvm: disabled by bios\n");
8393 * KVM explicitly assumes that the guest has an FPU and
8394 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8395 * vCPU's FPU state as a fxregs_state struct.
8397 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8398 printk(KERN_ERR "kvm: inadequate fpu\n");
8404 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8405 __alignof__(struct fpu), SLAB_ACCOUNT,
8407 if (!x86_fpu_cache) {
8408 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8412 x86_emulator_cache = kvm_alloc_emulator_cache();
8413 if (!x86_emulator_cache) {
8414 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8415 goto out_free_x86_fpu_cache;
8418 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8419 if (!user_return_msrs) {
8420 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8421 goto out_free_x86_emulator_cache;
8423 kvm_nr_uret_msrs = 0;
8425 r = kvm_mmu_module_init();
8427 goto out_free_percpu;
8431 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8433 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8434 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8435 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8438 if (pi_inject_timer == -1)
8439 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8440 #ifdef CONFIG_X86_64
8441 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8443 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8444 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8450 free_percpu(user_return_msrs);
8451 out_free_x86_emulator_cache:
8452 kmem_cache_destroy(x86_emulator_cache);
8453 out_free_x86_fpu_cache:
8454 kmem_cache_destroy(x86_fpu_cache);
8459 void kvm_arch_exit(void)
8461 #ifdef CONFIG_X86_64
8462 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8463 clear_hv_tscchange_cb();
8466 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8468 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8469 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8470 CPUFREQ_TRANSITION_NOTIFIER);
8471 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8472 #ifdef CONFIG_X86_64
8473 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8474 irq_work_sync(&pvclock_irq_work);
8475 cancel_work_sync(&pvclock_gtod_work);
8477 kvm_x86_ops.hardware_enable = NULL;
8478 kvm_mmu_module_exit();
8479 free_percpu(user_return_msrs);
8480 kmem_cache_destroy(x86_emulator_cache);
8481 kmem_cache_destroy(x86_fpu_cache);
8482 #ifdef CONFIG_KVM_XEN
8483 static_key_deferred_flush(&kvm_xen_enabled);
8484 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8488 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8490 ++vcpu->stat.halt_exits;
8491 if (lapic_in_kernel(vcpu)) {
8492 vcpu->arch.mp_state = state;
8495 vcpu->run->exit_reason = reason;
8500 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8502 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8504 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8506 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8508 int ret = kvm_skip_emulated_instruction(vcpu);
8510 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8511 * KVM_EXIT_DEBUG here.
8513 return kvm_vcpu_halt(vcpu) && ret;
8515 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8517 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8519 int ret = kvm_skip_emulated_instruction(vcpu);
8521 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8523 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8525 #ifdef CONFIG_X86_64
8526 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8527 unsigned long clock_type)
8529 struct kvm_clock_pairing clock_pairing;
8530 struct timespec64 ts;
8534 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8535 return -KVM_EOPNOTSUPP;
8537 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8538 return -KVM_EOPNOTSUPP;
8540 clock_pairing.sec = ts.tv_sec;
8541 clock_pairing.nsec = ts.tv_nsec;
8542 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8543 clock_pairing.flags = 0;
8544 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8547 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8548 sizeof(struct kvm_clock_pairing)))
8556 * kvm_pv_kick_cpu_op: Kick a vcpu.
8558 * @apicid - apicid of vcpu to be kicked.
8560 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8562 struct kvm_lapic_irq lapic_irq;
8564 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8565 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8566 lapic_irq.level = 0;
8567 lapic_irq.dest_id = apicid;
8568 lapic_irq.msi_redir_hint = false;
8570 lapic_irq.delivery_mode = APIC_DM_REMRD;
8571 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8574 bool kvm_apicv_activated(struct kvm *kvm)
8576 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8578 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8580 static void kvm_apicv_init(struct kvm *kvm)
8583 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8584 &kvm->arch.apicv_inhibit_reasons);
8586 set_bit(APICV_INHIBIT_REASON_DISABLE,
8587 &kvm->arch.apicv_inhibit_reasons);
8590 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8592 struct kvm_vcpu *target = NULL;
8593 struct kvm_apic_map *map;
8595 vcpu->stat.directed_yield_attempted++;
8597 if (single_task_running())
8601 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8603 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8604 target = map->phys_map[dest_id]->vcpu;
8608 if (!target || !READ_ONCE(target->ready))
8611 /* Ignore requests to yield to self */
8615 if (kvm_vcpu_yield_to(target) <= 0)
8618 vcpu->stat.directed_yield_successful++;
8624 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8626 u64 ret = vcpu->run->hypercall.ret;
8628 if (!is_64_bit_mode(vcpu))
8630 kvm_rax_write(vcpu, ret);
8631 ++vcpu->stat.hypercalls;
8632 return kvm_skip_emulated_instruction(vcpu);
8635 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8637 unsigned long nr, a0, a1, a2, a3, ret;
8640 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8641 return kvm_xen_hypercall(vcpu);
8643 if (kvm_hv_hypercall_enabled(vcpu))
8644 return kvm_hv_hypercall(vcpu);
8646 nr = kvm_rax_read(vcpu);
8647 a0 = kvm_rbx_read(vcpu);
8648 a1 = kvm_rcx_read(vcpu);
8649 a2 = kvm_rdx_read(vcpu);
8650 a3 = kvm_rsi_read(vcpu);
8652 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8654 op_64_bit = is_64_bit_mode(vcpu);
8663 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8671 case KVM_HC_VAPIC_POLL_IRQ:
8674 case KVM_HC_KICK_CPU:
8675 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8678 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8679 kvm_sched_yield(vcpu, a1);
8682 #ifdef CONFIG_X86_64
8683 case KVM_HC_CLOCK_PAIRING:
8684 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8687 case KVM_HC_SEND_IPI:
8688 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8691 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8693 case KVM_HC_SCHED_YIELD:
8694 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8697 kvm_sched_yield(vcpu, a0);
8700 case KVM_HC_MAP_GPA_RANGE: {
8701 u64 gpa = a0, npages = a1, attrs = a2;
8704 if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
8707 if (!PAGE_ALIGNED(gpa) || !npages ||
8708 gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
8713 vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
8714 vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
8715 vcpu->run->hypercall.args[0] = gpa;
8716 vcpu->run->hypercall.args[1] = npages;
8717 vcpu->run->hypercall.args[2] = attrs;
8718 vcpu->run->hypercall.longmode = op_64_bit;
8719 vcpu->arch.complete_userspace_io = complete_hypercall_exit;
8729 kvm_rax_write(vcpu, ret);
8731 ++vcpu->stat.hypercalls;
8732 return kvm_skip_emulated_instruction(vcpu);
8734 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8736 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8738 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8739 char instruction[3];
8740 unsigned long rip = kvm_rip_read(vcpu);
8742 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8744 return emulator_write_emulated(ctxt, rip, instruction, 3,
8748 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8750 return vcpu->run->request_interrupt_window &&
8751 likely(!pic_in_kernel(vcpu->kvm));
8754 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8756 struct kvm_run *kvm_run = vcpu->run;
8759 * if_flag is obsolete and useless, so do not bother
8760 * setting it for SEV-ES guests. Userspace can just
8761 * use kvm_run->ready_for_interrupt_injection.
8763 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8764 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8766 kvm_run->cr8 = kvm_get_cr8(vcpu);
8767 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8768 kvm_run->ready_for_interrupt_injection =
8769 pic_in_kernel(vcpu->kvm) ||
8770 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8773 kvm_run->flags |= KVM_RUN_X86_SMM;
8776 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8780 if (!kvm_x86_ops.update_cr8_intercept)
8783 if (!lapic_in_kernel(vcpu))
8786 if (vcpu->arch.apicv_active)
8789 if (!vcpu->arch.apic->vapic_addr)
8790 max_irr = kvm_lapic_find_highest_irr(vcpu);
8797 tpr = kvm_lapic_get_cr8(vcpu);
8799 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8803 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8805 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8806 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8810 return kvm_x86_ops.nested_ops->check_events(vcpu);
8813 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8815 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8816 vcpu->arch.exception.error_code = false;
8817 static_call(kvm_x86_queue_exception)(vcpu);
8820 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8823 bool can_inject = true;
8825 /* try to reinject previous events if any */
8827 if (vcpu->arch.exception.injected) {
8828 kvm_inject_exception(vcpu);
8832 * Do not inject an NMI or interrupt if there is a pending
8833 * exception. Exceptions and interrupts are recognized at
8834 * instruction boundaries, i.e. the start of an instruction.
8835 * Trap-like exceptions, e.g. #DB, have higher priority than
8836 * NMIs and interrupts, i.e. traps are recognized before an
8837 * NMI/interrupt that's pending on the same instruction.
8838 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8839 * priority, but are only generated (pended) during instruction
8840 * execution, i.e. a pending fault-like exception means the
8841 * fault occurred on the *previous* instruction and must be
8842 * serviced prior to recognizing any new events in order to
8843 * fully complete the previous instruction.
8845 else if (!vcpu->arch.exception.pending) {
8846 if (vcpu->arch.nmi_injected) {
8847 static_call(kvm_x86_set_nmi)(vcpu);
8849 } else if (vcpu->arch.interrupt.injected) {
8850 static_call(kvm_x86_set_irq)(vcpu);
8855 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8856 vcpu->arch.exception.pending);
8859 * Call check_nested_events() even if we reinjected a previous event
8860 * in order for caller to determine if it should require immediate-exit
8861 * from L2 to L1 due to pending L1 events which require exit
8864 if (is_guest_mode(vcpu)) {
8865 r = kvm_check_nested_events(vcpu);
8870 /* try to inject new event if pending */
8871 if (vcpu->arch.exception.pending) {
8872 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8873 vcpu->arch.exception.has_error_code,
8874 vcpu->arch.exception.error_code);
8876 vcpu->arch.exception.pending = false;
8877 vcpu->arch.exception.injected = true;
8879 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8880 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8883 if (vcpu->arch.exception.nr == DB_VECTOR) {
8884 kvm_deliver_exception_payload(vcpu);
8885 if (vcpu->arch.dr7 & DR7_GD) {
8886 vcpu->arch.dr7 &= ~DR7_GD;
8887 kvm_update_dr7(vcpu);
8891 kvm_inject_exception(vcpu);
8896 * Finally, inject interrupt events. If an event cannot be injected
8897 * due to architectural conditions (e.g. IF=0) a window-open exit
8898 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8899 * and can architecturally be injected, but we cannot do it right now:
8900 * an interrupt could have arrived just now and we have to inject it
8901 * as a vmexit, or there could already an event in the queue, which is
8902 * indicated by can_inject. In that case we request an immediate exit
8903 * in order to make progress and get back here for another iteration.
8904 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8906 if (vcpu->arch.smi_pending) {
8907 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8911 vcpu->arch.smi_pending = false;
8912 ++vcpu->arch.smi_count;
8916 static_call(kvm_x86_enable_smi_window)(vcpu);
8919 if (vcpu->arch.nmi_pending) {
8920 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8924 --vcpu->arch.nmi_pending;
8925 vcpu->arch.nmi_injected = true;
8926 static_call(kvm_x86_set_nmi)(vcpu);
8928 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8930 if (vcpu->arch.nmi_pending)
8931 static_call(kvm_x86_enable_nmi_window)(vcpu);
8934 if (kvm_cpu_has_injectable_intr(vcpu)) {
8935 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8939 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8940 static_call(kvm_x86_set_irq)(vcpu);
8941 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8943 if (kvm_cpu_has_injectable_intr(vcpu))
8944 static_call(kvm_x86_enable_irq_window)(vcpu);
8947 if (is_guest_mode(vcpu) &&
8948 kvm_x86_ops.nested_ops->hv_timer_pending &&
8949 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8950 *req_immediate_exit = true;
8952 WARN_ON(vcpu->arch.exception.pending);
8957 *req_immediate_exit = true;
8963 static void process_nmi(struct kvm_vcpu *vcpu)
8968 * x86 is limited to one NMI running, and one NMI pending after it.
8969 * If an NMI is already in progress, limit further NMIs to just one.
8970 * Otherwise, allow two (and we'll inject the first one immediately).
8972 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8975 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8976 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8977 kvm_make_request(KVM_REQ_EVENT, vcpu);
8980 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8983 flags |= seg->g << 23;
8984 flags |= seg->db << 22;
8985 flags |= seg->l << 21;
8986 flags |= seg->avl << 20;
8987 flags |= seg->present << 15;
8988 flags |= seg->dpl << 13;
8989 flags |= seg->s << 12;
8990 flags |= seg->type << 8;
8994 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8996 struct kvm_segment seg;
8999 kvm_get_segment(vcpu, &seg, n);
9000 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9003 offset = 0x7f84 + n * 12;
9005 offset = 0x7f2c + (n - 3) * 12;
9007 put_smstate(u32, buf, offset + 8, seg.base);
9008 put_smstate(u32, buf, offset + 4, seg.limit);
9009 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9012 #ifdef CONFIG_X86_64
9013 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9015 struct kvm_segment seg;
9019 kvm_get_segment(vcpu, &seg, n);
9020 offset = 0x7e00 + n * 16;
9022 flags = enter_smm_get_segment_flags(&seg) >> 8;
9023 put_smstate(u16, buf, offset, seg.selector);
9024 put_smstate(u16, buf, offset + 2, flags);
9025 put_smstate(u32, buf, offset + 4, seg.limit);
9026 put_smstate(u64, buf, offset + 8, seg.base);
9030 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9033 struct kvm_segment seg;
9037 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9038 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9039 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9040 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9042 for (i = 0; i < 8; i++)
9043 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9045 kvm_get_dr(vcpu, 6, &val);
9046 put_smstate(u32, buf, 0x7fcc, (u32)val);
9047 kvm_get_dr(vcpu, 7, &val);
9048 put_smstate(u32, buf, 0x7fc8, (u32)val);
9050 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9051 put_smstate(u32, buf, 0x7fc4, seg.selector);
9052 put_smstate(u32, buf, 0x7f64, seg.base);
9053 put_smstate(u32, buf, 0x7f60, seg.limit);
9054 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9056 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9057 put_smstate(u32, buf, 0x7fc0, seg.selector);
9058 put_smstate(u32, buf, 0x7f80, seg.base);
9059 put_smstate(u32, buf, 0x7f7c, seg.limit);
9060 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9062 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9063 put_smstate(u32, buf, 0x7f74, dt.address);
9064 put_smstate(u32, buf, 0x7f70, dt.size);
9066 static_call(kvm_x86_get_idt)(vcpu, &dt);
9067 put_smstate(u32, buf, 0x7f58, dt.address);
9068 put_smstate(u32, buf, 0x7f54, dt.size);
9070 for (i = 0; i < 6; i++)
9071 enter_smm_save_seg_32(vcpu, buf, i);
9073 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9076 put_smstate(u32, buf, 0x7efc, 0x00020000);
9077 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9080 #ifdef CONFIG_X86_64
9081 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9084 struct kvm_segment seg;
9088 for (i = 0; i < 16; i++)
9089 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9091 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9092 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9094 kvm_get_dr(vcpu, 6, &val);
9095 put_smstate(u64, buf, 0x7f68, val);
9096 kvm_get_dr(vcpu, 7, &val);
9097 put_smstate(u64, buf, 0x7f60, val);
9099 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9100 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9101 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9103 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9106 put_smstate(u32, buf, 0x7efc, 0x00020064);
9108 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9110 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9111 put_smstate(u16, buf, 0x7e90, seg.selector);
9112 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9113 put_smstate(u32, buf, 0x7e94, seg.limit);
9114 put_smstate(u64, buf, 0x7e98, seg.base);
9116 static_call(kvm_x86_get_idt)(vcpu, &dt);
9117 put_smstate(u32, buf, 0x7e84, dt.size);
9118 put_smstate(u64, buf, 0x7e88, dt.address);
9120 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9121 put_smstate(u16, buf, 0x7e70, seg.selector);
9122 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9123 put_smstate(u32, buf, 0x7e74, seg.limit);
9124 put_smstate(u64, buf, 0x7e78, seg.base);
9126 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9127 put_smstate(u32, buf, 0x7e64, dt.size);
9128 put_smstate(u64, buf, 0x7e68, dt.address);
9130 for (i = 0; i < 6; i++)
9131 enter_smm_save_seg_64(vcpu, buf, i);
9135 static void enter_smm(struct kvm_vcpu *vcpu)
9137 struct kvm_segment cs, ds;
9142 memset(buf, 0, 512);
9143 #ifdef CONFIG_X86_64
9144 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9145 enter_smm_save_state_64(vcpu, buf);
9148 enter_smm_save_state_32(vcpu, buf);
9151 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9152 * state (e.g. leave guest mode) after we've saved the state into the
9153 * SMM state-save area.
9155 static_call(kvm_x86_enter_smm)(vcpu, buf);
9157 kvm_smm_changed(vcpu, true);
9158 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9160 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9161 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9163 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9165 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9166 kvm_rip_write(vcpu, 0x8000);
9168 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9169 static_call(kvm_x86_set_cr0)(vcpu, cr0);
9170 vcpu->arch.cr0 = cr0;
9172 static_call(kvm_x86_set_cr4)(vcpu, 0);
9174 /* Undocumented: IDT limit is set to zero on entry to SMM. */
9175 dt.address = dt.size = 0;
9176 static_call(kvm_x86_set_idt)(vcpu, &dt);
9178 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9180 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9181 cs.base = vcpu->arch.smbase;
9186 cs.limit = ds.limit = 0xffffffff;
9187 cs.type = ds.type = 0x3;
9188 cs.dpl = ds.dpl = 0;
9193 cs.avl = ds.avl = 0;
9194 cs.present = ds.present = 1;
9195 cs.unusable = ds.unusable = 0;
9196 cs.padding = ds.padding = 0;
9198 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9199 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9200 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9201 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9202 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9203 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9205 #ifdef CONFIG_X86_64
9206 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9207 static_call(kvm_x86_set_efer)(vcpu, 0);
9210 kvm_update_cpuid_runtime(vcpu);
9211 kvm_mmu_reset_context(vcpu);
9214 static void process_smi(struct kvm_vcpu *vcpu)
9216 vcpu->arch.smi_pending = true;
9217 kvm_make_request(KVM_REQ_EVENT, vcpu);
9220 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9221 unsigned long *vcpu_bitmap)
9225 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
9227 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
9228 NULL, vcpu_bitmap, cpus);
9230 free_cpumask_var(cpus);
9233 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9235 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9238 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9240 if (!lapic_in_kernel(vcpu))
9243 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
9244 kvm_apic_update_apicv(vcpu);
9245 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9248 * When APICv gets disabled, we may still have injected interrupts
9249 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9250 * still active when the interrupt got accepted. Make sure
9251 * inject_pending_event() is called to check for that.
9253 if (!vcpu->arch.apicv_active)
9254 kvm_make_request(KVM_REQ_EVENT, vcpu);
9256 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9259 * NOTE: Do not hold any lock prior to calling this.
9261 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
9262 * locked, because it calls __x86_set_memory_region() which does
9263 * synchronize_srcu(&kvm->srcu).
9265 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9267 struct kvm_vcpu *except;
9268 unsigned long old, new, expected;
9270 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9271 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9274 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
9276 expected = new = old;
9278 __clear_bit(bit, &new);
9280 __set_bit(bit, &new);
9283 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
9284 } while (old != expected);
9289 trace_kvm_apicv_update_request(activate, bit);
9290 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
9291 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
9294 * Sending request to update APICV for all other vcpus,
9295 * while update the calling vcpu immediately instead of
9296 * waiting for another #VMEXIT to handle the request.
9298 except = kvm_get_running_vcpu();
9299 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
9302 kvm_vcpu_update_apicv(except);
9304 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9306 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9308 if (!kvm_apic_present(vcpu))
9311 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9313 if (irqchip_split(vcpu->kvm))
9314 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9316 if (vcpu->arch.apicv_active)
9317 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9318 if (ioapic_in_kernel(vcpu->kvm))
9319 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9322 if (is_guest_mode(vcpu))
9323 vcpu->arch.load_eoi_exitmap_pending = true;
9325 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9328 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9330 u64 eoi_exit_bitmap[4];
9332 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9335 if (to_hv_vcpu(vcpu))
9336 bitmap_or((ulong *)eoi_exit_bitmap,
9337 vcpu->arch.ioapic_handled_vectors,
9338 to_hv_synic(vcpu)->vec_bitmap, 256);
9340 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9343 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9344 unsigned long start, unsigned long end)
9346 unsigned long apic_address;
9349 * The physical address of apic access page is stored in the VMCS.
9350 * Update it when it becomes invalid.
9352 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9353 if (start <= apic_address && apic_address < end)
9354 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9357 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9359 if (!lapic_in_kernel(vcpu))
9362 if (!kvm_x86_ops.set_apic_access_page_addr)
9365 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9368 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9370 smp_send_reschedule(vcpu->cpu);
9372 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9375 * Returns 1 to let vcpu_run() continue the guest execution loop without
9376 * exiting to the userspace. Otherwise, the value will be returned to the
9379 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9383 dm_request_for_irq_injection(vcpu) &&
9384 kvm_cpu_accept_dm_intr(vcpu);
9385 fastpath_t exit_fastpath;
9387 bool req_immediate_exit = false;
9389 /* Forbid vmenter if vcpu dirty ring is soft-full */
9390 if (unlikely(vcpu->kvm->dirty_ring_size &&
9391 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9392 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9393 trace_kvm_dirty_ring_exit(vcpu);
9398 if (kvm_request_pending(vcpu)) {
9399 if (kvm_check_request(KVM_REQ_VM_BUGGED, vcpu)) {
9403 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9404 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9409 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9410 kvm_mmu_unload(vcpu);
9411 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9412 __kvm_migrate_timers(vcpu);
9413 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9414 kvm_gen_update_masterclock(vcpu->kvm);
9415 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9416 kvm_gen_kvmclock_update(vcpu);
9417 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9418 r = kvm_guest_time_update(vcpu);
9422 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9423 kvm_mmu_sync_roots(vcpu);
9424 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9425 kvm_mmu_load_pgd(vcpu);
9426 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9427 kvm_vcpu_flush_tlb_all(vcpu);
9429 /* Flushing all ASIDs flushes the current ASID... */
9430 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9432 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9433 kvm_vcpu_flush_tlb_current(vcpu);
9434 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9435 kvm_vcpu_flush_tlb_guest(vcpu);
9437 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9438 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9442 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9443 if (is_guest_mode(vcpu)) {
9444 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9446 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9447 vcpu->mmio_needed = 0;
9452 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9453 /* Page is swapped out. Do synthetic halt */
9454 vcpu->arch.apf.halted = true;
9458 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9459 record_steal_time(vcpu);
9460 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9462 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9464 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9465 kvm_pmu_handle_event(vcpu);
9466 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9467 kvm_pmu_deliver_pmi(vcpu);
9468 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9469 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9470 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9471 vcpu->arch.ioapic_handled_vectors)) {
9472 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9473 vcpu->run->eoi.vector =
9474 vcpu->arch.pending_ioapic_eoi;
9479 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9480 vcpu_scan_ioapic(vcpu);
9481 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9482 vcpu_load_eoi_exitmap(vcpu);
9483 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9484 kvm_vcpu_reload_apic_access_page(vcpu);
9485 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9486 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9487 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9491 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9492 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9493 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9497 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9498 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9500 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9501 vcpu->run->hyperv = hv_vcpu->exit;
9507 * KVM_REQ_HV_STIMER has to be processed after
9508 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9509 * depend on the guest clock being up-to-date
9511 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9512 kvm_hv_process_stimers(vcpu);
9513 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9514 kvm_vcpu_update_apicv(vcpu);
9515 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9516 kvm_check_async_pf_completion(vcpu);
9517 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9518 static_call(kvm_x86_msr_filter_changed)(vcpu);
9520 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9521 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9524 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9525 kvm_xen_has_interrupt(vcpu)) {
9526 ++vcpu->stat.req_event;
9527 r = kvm_apic_accept_events(vcpu);
9532 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9537 r = inject_pending_event(vcpu, &req_immediate_exit);
9543 static_call(kvm_x86_enable_irq_window)(vcpu);
9545 if (kvm_lapic_enabled(vcpu)) {
9546 update_cr8_intercept(vcpu);
9547 kvm_lapic_sync_to_vapic(vcpu);
9551 r = kvm_mmu_reload(vcpu);
9553 goto cancel_injection;
9558 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9561 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9562 * IPI are then delayed after guest entry, which ensures that they
9563 * result in virtual interrupt delivery.
9565 local_irq_disable();
9566 vcpu->mode = IN_GUEST_MODE;
9568 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9571 * 1) We should set ->mode before checking ->requests. Please see
9572 * the comment in kvm_vcpu_exiting_guest_mode().
9574 * 2) For APICv, we should set ->mode before checking PID.ON. This
9575 * pairs with the memory barrier implicit in pi_test_and_set_on
9576 * (see vmx_deliver_posted_interrupt).
9578 * 3) This also orders the write to mode from any reads to the page
9579 * tables done while the VCPU is running. Please see the comment
9580 * in kvm_flush_remote_tlbs.
9582 smp_mb__after_srcu_read_unlock();
9585 * This handles the case where a posted interrupt was
9586 * notified with kvm_vcpu_kick.
9588 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9589 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9591 if (kvm_vcpu_exit_request(vcpu)) {
9592 vcpu->mode = OUTSIDE_GUEST_MODE;
9596 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9598 goto cancel_injection;
9601 if (req_immediate_exit) {
9602 kvm_make_request(KVM_REQ_EVENT, vcpu);
9603 static_call(kvm_x86_request_immediate_exit)(vcpu);
9606 fpregs_assert_state_consistent();
9607 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9608 switch_fpu_return();
9610 if (unlikely(vcpu->arch.switch_db_regs)) {
9612 set_debugreg(vcpu->arch.eff_db[0], 0);
9613 set_debugreg(vcpu->arch.eff_db[1], 1);
9614 set_debugreg(vcpu->arch.eff_db[2], 2);
9615 set_debugreg(vcpu->arch.eff_db[3], 3);
9616 set_debugreg(vcpu->arch.dr6, 6);
9617 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9618 } else if (unlikely(hw_breakpoint_active())) {
9623 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9624 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9627 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9628 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9632 if (vcpu->arch.apicv_active)
9633 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9637 * Do this here before restoring debug registers on the host. And
9638 * since we do this before handling the vmexit, a DR access vmexit
9639 * can (a) read the correct value of the debug registers, (b) set
9640 * KVM_DEBUGREG_WONT_EXIT again.
9642 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9643 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9644 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9645 kvm_update_dr0123(vcpu);
9646 kvm_update_dr7(vcpu);
9647 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9651 * If the guest has used debug registers, at least dr7
9652 * will be disabled while returning to the host.
9653 * If we don't have active breakpoints in the host, we don't
9654 * care about the messed up debug address registers. But if
9655 * we have some of them active, restore the old state.
9657 if (hw_breakpoint_active())
9658 hw_breakpoint_restore();
9660 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9661 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9663 vcpu->mode = OUTSIDE_GUEST_MODE;
9666 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9669 * Consume any pending interrupts, including the possible source of
9670 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9671 * An instruction is required after local_irq_enable() to fully unblock
9672 * interrupts on processors that implement an interrupt shadow, the
9673 * stat.exits increment will do nicely.
9675 kvm_before_interrupt(vcpu);
9678 local_irq_disable();
9679 kvm_after_interrupt(vcpu);
9682 * Wait until after servicing IRQs to account guest time so that any
9683 * ticks that occurred while running the guest are properly accounted
9684 * to the guest. Waiting until IRQs are enabled degrades the accuracy
9685 * of accounting via context tracking, but the loss of accuracy is
9686 * acceptable for all known use cases.
9688 vtime_account_guest_exit();
9690 if (lapic_in_kernel(vcpu)) {
9691 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9692 if (delta != S64_MIN) {
9693 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9694 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9701 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9704 * Profile KVM exit RIPs:
9706 if (unlikely(prof_on == KVM_PROFILING)) {
9707 unsigned long rip = kvm_rip_read(vcpu);
9708 profile_hit(KVM_PROFILING, (void *)rip);
9711 if (unlikely(vcpu->arch.tsc_always_catchup))
9712 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9714 if (vcpu->arch.apic_attention)
9715 kvm_lapic_sync_from_vapic(vcpu);
9717 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9721 if (req_immediate_exit)
9722 kvm_make_request(KVM_REQ_EVENT, vcpu);
9723 static_call(kvm_x86_cancel_injection)(vcpu);
9724 if (unlikely(vcpu->arch.apic_attention))
9725 kvm_lapic_sync_from_vapic(vcpu);
9730 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9732 if (!kvm_arch_vcpu_runnable(vcpu) &&
9733 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9734 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9735 kvm_vcpu_block(vcpu);
9736 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9738 if (kvm_x86_ops.post_block)
9739 static_call(kvm_x86_post_block)(vcpu);
9741 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9745 if (kvm_apic_accept_events(vcpu) < 0)
9747 switch(vcpu->arch.mp_state) {
9748 case KVM_MP_STATE_HALTED:
9749 case KVM_MP_STATE_AP_RESET_HOLD:
9750 vcpu->arch.pv.pv_unhalted = false;
9751 vcpu->arch.mp_state =
9752 KVM_MP_STATE_RUNNABLE;
9754 case KVM_MP_STATE_RUNNABLE:
9755 vcpu->arch.apf.halted = false;
9757 case KVM_MP_STATE_INIT_RECEIVED:
9765 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9767 if (is_guest_mode(vcpu))
9768 kvm_check_nested_events(vcpu);
9770 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9771 !vcpu->arch.apf.halted);
9774 static int vcpu_run(struct kvm_vcpu *vcpu)
9777 struct kvm *kvm = vcpu->kvm;
9779 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9780 vcpu->arch.l1tf_flush_l1d = true;
9783 if (kvm_vcpu_running(vcpu)) {
9784 r = vcpu_enter_guest(vcpu);
9786 r = vcpu_block(kvm, vcpu);
9792 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9793 if (kvm_cpu_has_pending_timer(vcpu))
9794 kvm_inject_pending_timer_irqs(vcpu);
9796 if (dm_request_for_irq_injection(vcpu) &&
9797 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9799 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9800 ++vcpu->stat.request_irq_exits;
9804 if (__xfer_to_guest_mode_work_pending()) {
9805 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9806 r = xfer_to_guest_mode_handle_work(vcpu);
9809 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9813 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9818 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9822 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9823 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9824 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9828 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9830 BUG_ON(!vcpu->arch.pio.count);
9832 return complete_emulated_io(vcpu);
9836 * Implements the following, as a state machine:
9840 * for each mmio piece in the fragment
9848 * for each mmio piece in the fragment
9853 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9855 struct kvm_run *run = vcpu->run;
9856 struct kvm_mmio_fragment *frag;
9859 BUG_ON(!vcpu->mmio_needed);
9861 /* Complete previous fragment */
9862 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9863 len = min(8u, frag->len);
9864 if (!vcpu->mmio_is_write)
9865 memcpy(frag->data, run->mmio.data, len);
9867 if (frag->len <= 8) {
9868 /* Switch to the next fragment. */
9870 vcpu->mmio_cur_fragment++;
9872 /* Go forward to the next mmio piece. */
9878 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9879 vcpu->mmio_needed = 0;
9881 /* FIXME: return into emulator if single-stepping. */
9882 if (vcpu->mmio_is_write)
9884 vcpu->mmio_read_completed = 1;
9885 return complete_emulated_io(vcpu);
9888 run->exit_reason = KVM_EXIT_MMIO;
9889 run->mmio.phys_addr = frag->gpa;
9890 if (vcpu->mmio_is_write)
9891 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9892 run->mmio.len = min(8u, frag->len);
9893 run->mmio.is_write = vcpu->mmio_is_write;
9894 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9898 static void kvm_save_current_fpu(struct fpu *fpu)
9901 * If the target FPU state is not resident in the CPU registers, just
9902 * memcpy() from current, else save CPU state directly to the target.
9904 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9905 memcpy(&fpu->state, ¤t->thread.fpu.state,
9906 fpu_kernel_xstate_size);
9908 save_fpregs_to_fpstate(fpu);
9911 /* Swap (qemu) user FPU context for the guest FPU context. */
9912 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9916 kvm_save_current_fpu(vcpu->arch.user_fpu);
9919 * Guests with protected state can't have it set by the hypervisor,
9920 * so skip trying to set it.
9922 if (vcpu->arch.guest_fpu)
9923 /* PKRU is separately restored in kvm_x86_ops.run. */
9924 __restore_fpregs_from_fpstate(&vcpu->arch.guest_fpu->state,
9925 ~XFEATURE_MASK_PKRU);
9927 fpregs_mark_activate();
9933 /* When vcpu_run ends, restore user space FPU context. */
9934 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9939 * Guests with protected state can't have it read by the hypervisor,
9940 * so skip trying to save it.
9942 if (vcpu->arch.guest_fpu)
9943 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9945 restore_fpregs_from_fpstate(&vcpu->arch.user_fpu->state);
9947 fpregs_mark_activate();
9950 ++vcpu->stat.fpu_reload;
9954 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9956 struct kvm_run *kvm_run = vcpu->run;
9960 kvm_sigset_activate(vcpu);
9962 kvm_load_guest_fpu(vcpu);
9964 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9965 if (kvm_run->immediate_exit) {
9969 kvm_vcpu_block(vcpu);
9970 if (kvm_apic_accept_events(vcpu) < 0) {
9974 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9976 if (signal_pending(current)) {
9978 kvm_run->exit_reason = KVM_EXIT_INTR;
9979 ++vcpu->stat.signal_exits;
9984 if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
9985 (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
9990 if (kvm_run->kvm_dirty_regs) {
9991 r = sync_regs(vcpu);
9996 /* re-sync apic's tpr */
9997 if (!lapic_in_kernel(vcpu)) {
9998 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10004 if (unlikely(vcpu->arch.complete_userspace_io)) {
10005 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10006 vcpu->arch.complete_userspace_io = NULL;
10011 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10013 if (kvm_run->immediate_exit)
10016 r = vcpu_run(vcpu);
10019 kvm_put_guest_fpu(vcpu);
10020 if (kvm_run->kvm_valid_regs)
10022 post_kvm_run_save(vcpu);
10023 kvm_sigset_deactivate(vcpu);
10029 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10031 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10033 * We are here if userspace calls get_regs() in the middle of
10034 * instruction emulation. Registers state needs to be copied
10035 * back from emulation context to vcpu. Userspace shouldn't do
10036 * that usually, but some bad designed PV devices (vmware
10037 * backdoor interface) need this to work
10039 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10040 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10042 regs->rax = kvm_rax_read(vcpu);
10043 regs->rbx = kvm_rbx_read(vcpu);
10044 regs->rcx = kvm_rcx_read(vcpu);
10045 regs->rdx = kvm_rdx_read(vcpu);
10046 regs->rsi = kvm_rsi_read(vcpu);
10047 regs->rdi = kvm_rdi_read(vcpu);
10048 regs->rsp = kvm_rsp_read(vcpu);
10049 regs->rbp = kvm_rbp_read(vcpu);
10050 #ifdef CONFIG_X86_64
10051 regs->r8 = kvm_r8_read(vcpu);
10052 regs->r9 = kvm_r9_read(vcpu);
10053 regs->r10 = kvm_r10_read(vcpu);
10054 regs->r11 = kvm_r11_read(vcpu);
10055 regs->r12 = kvm_r12_read(vcpu);
10056 regs->r13 = kvm_r13_read(vcpu);
10057 regs->r14 = kvm_r14_read(vcpu);
10058 regs->r15 = kvm_r15_read(vcpu);
10061 regs->rip = kvm_rip_read(vcpu);
10062 regs->rflags = kvm_get_rflags(vcpu);
10065 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10068 __get_regs(vcpu, regs);
10073 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10075 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10076 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10078 kvm_rax_write(vcpu, regs->rax);
10079 kvm_rbx_write(vcpu, regs->rbx);
10080 kvm_rcx_write(vcpu, regs->rcx);
10081 kvm_rdx_write(vcpu, regs->rdx);
10082 kvm_rsi_write(vcpu, regs->rsi);
10083 kvm_rdi_write(vcpu, regs->rdi);
10084 kvm_rsp_write(vcpu, regs->rsp);
10085 kvm_rbp_write(vcpu, regs->rbp);
10086 #ifdef CONFIG_X86_64
10087 kvm_r8_write(vcpu, regs->r8);
10088 kvm_r9_write(vcpu, regs->r9);
10089 kvm_r10_write(vcpu, regs->r10);
10090 kvm_r11_write(vcpu, regs->r11);
10091 kvm_r12_write(vcpu, regs->r12);
10092 kvm_r13_write(vcpu, regs->r13);
10093 kvm_r14_write(vcpu, regs->r14);
10094 kvm_r15_write(vcpu, regs->r15);
10097 kvm_rip_write(vcpu, regs->rip);
10098 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10100 vcpu->arch.exception.pending = false;
10102 kvm_make_request(KVM_REQ_EVENT, vcpu);
10105 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10108 __set_regs(vcpu, regs);
10113 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10115 struct kvm_segment cs;
10117 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10121 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10123 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10125 struct desc_ptr dt;
10127 if (vcpu->arch.guest_state_protected)
10128 goto skip_protected_regs;
10130 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10131 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10132 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10133 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10134 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10135 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10137 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10138 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10140 static_call(kvm_x86_get_idt)(vcpu, &dt);
10141 sregs->idt.limit = dt.size;
10142 sregs->idt.base = dt.address;
10143 static_call(kvm_x86_get_gdt)(vcpu, &dt);
10144 sregs->gdt.limit = dt.size;
10145 sregs->gdt.base = dt.address;
10147 sregs->cr2 = vcpu->arch.cr2;
10148 sregs->cr3 = kvm_read_cr3(vcpu);
10150 skip_protected_regs:
10151 sregs->cr0 = kvm_read_cr0(vcpu);
10152 sregs->cr4 = kvm_read_cr4(vcpu);
10153 sregs->cr8 = kvm_get_cr8(vcpu);
10154 sregs->efer = vcpu->arch.efer;
10155 sregs->apic_base = kvm_get_apic_base(vcpu);
10158 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10160 __get_sregs_common(vcpu, sregs);
10162 if (vcpu->arch.guest_state_protected)
10165 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10166 set_bit(vcpu->arch.interrupt.nr,
10167 (unsigned long *)sregs->interrupt_bitmap);
10170 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10174 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10176 if (vcpu->arch.guest_state_protected)
10179 if (is_pae_paging(vcpu)) {
10180 for (i = 0 ; i < 4 ; i++)
10181 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10182 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10186 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10187 struct kvm_sregs *sregs)
10190 __get_sregs(vcpu, sregs);
10195 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10196 struct kvm_mp_state *mp_state)
10201 if (kvm_mpx_supported())
10202 kvm_load_guest_fpu(vcpu);
10204 r = kvm_apic_accept_events(vcpu);
10209 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10210 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10211 vcpu->arch.pv.pv_unhalted)
10212 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10214 mp_state->mp_state = vcpu->arch.mp_state;
10217 if (kvm_mpx_supported())
10218 kvm_put_guest_fpu(vcpu);
10223 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10224 struct kvm_mp_state *mp_state)
10230 if (!lapic_in_kernel(vcpu) &&
10231 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10235 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10236 * INIT state; latched init should be reported using
10237 * KVM_SET_VCPU_EVENTS, so reject it here.
10239 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10240 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10241 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10244 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10245 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10246 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10248 vcpu->arch.mp_state = mp_state->mp_state;
10249 kvm_make_request(KVM_REQ_EVENT, vcpu);
10257 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10258 int reason, bool has_error_code, u32 error_code)
10260 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10263 init_emulate_ctxt(vcpu);
10265 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10266 has_error_code, error_code);
10268 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10269 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10270 vcpu->run->internal.ndata = 0;
10274 kvm_rip_write(vcpu, ctxt->eip);
10275 kvm_set_rflags(vcpu, ctxt->eflags);
10278 EXPORT_SYMBOL_GPL(kvm_task_switch);
10280 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10282 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10284 * When EFER.LME and CR0.PG are set, the processor is in
10285 * 64-bit mode (though maybe in a 32-bit code segment).
10286 * CR4.PAE and EFER.LMA must be set.
10288 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10290 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10294 * Not in 64-bit mode: EFER.LMA is clear and the code
10295 * segment cannot be 64-bit.
10297 if (sregs->efer & EFER_LMA || sregs->cs.l)
10301 return kvm_is_valid_cr4(vcpu, sregs->cr4);
10304 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10305 int *mmu_reset_needed, bool update_pdptrs)
10307 struct msr_data apic_base_msr;
10309 struct desc_ptr dt;
10311 if (!kvm_is_valid_sregs(vcpu, sregs))
10314 apic_base_msr.data = sregs->apic_base;
10315 apic_base_msr.host_initiated = true;
10316 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10319 if (vcpu->arch.guest_state_protected)
10322 dt.size = sregs->idt.limit;
10323 dt.address = sregs->idt.base;
10324 static_call(kvm_x86_set_idt)(vcpu, &dt);
10325 dt.size = sregs->gdt.limit;
10326 dt.address = sregs->gdt.base;
10327 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10329 vcpu->arch.cr2 = sregs->cr2;
10330 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10331 vcpu->arch.cr3 = sregs->cr3;
10332 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10334 kvm_set_cr8(vcpu, sregs->cr8);
10336 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10337 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10339 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10340 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10341 vcpu->arch.cr0 = sregs->cr0;
10343 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10344 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10346 if (update_pdptrs) {
10347 idx = srcu_read_lock(&vcpu->kvm->srcu);
10348 if (is_pae_paging(vcpu)) {
10349 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10350 *mmu_reset_needed = 1;
10352 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10355 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10356 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10357 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10358 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10359 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10360 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10362 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10363 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10365 update_cr8_intercept(vcpu);
10367 /* Older userspace won't unhalt the vcpu on reset. */
10368 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10369 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10370 !is_protmode(vcpu))
10371 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10376 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10378 int pending_vec, max_bits;
10379 int mmu_reset_needed = 0;
10380 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10385 if (mmu_reset_needed)
10386 kvm_mmu_reset_context(vcpu);
10388 max_bits = KVM_NR_INTERRUPTS;
10389 pending_vec = find_first_bit(
10390 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10392 if (pending_vec < max_bits) {
10393 kvm_queue_interrupt(vcpu, pending_vec, false);
10394 pr_debug("Set back pending irq %d\n", pending_vec);
10395 kvm_make_request(KVM_REQ_EVENT, vcpu);
10400 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10402 int mmu_reset_needed = 0;
10403 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10404 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10405 !(sregs2->efer & EFER_LMA);
10408 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10411 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10414 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10415 &mmu_reset_needed, !valid_pdptrs);
10419 if (valid_pdptrs) {
10420 for (i = 0; i < 4 ; i++)
10421 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10423 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10424 mmu_reset_needed = 1;
10425 vcpu->arch.pdptrs_from_userspace = true;
10427 if (mmu_reset_needed)
10428 kvm_mmu_reset_context(vcpu);
10432 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10433 struct kvm_sregs *sregs)
10438 ret = __set_sregs(vcpu, sregs);
10443 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10444 struct kvm_guest_debug *dbg)
10446 unsigned long rflags;
10449 if (vcpu->arch.guest_state_protected)
10454 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10456 if (vcpu->arch.exception.pending)
10458 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10459 kvm_queue_exception(vcpu, DB_VECTOR);
10461 kvm_queue_exception(vcpu, BP_VECTOR);
10465 * Read rflags as long as potentially injected trace flags are still
10468 rflags = kvm_get_rflags(vcpu);
10470 vcpu->guest_debug = dbg->control;
10471 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10472 vcpu->guest_debug = 0;
10474 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10475 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10476 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10477 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10479 for (i = 0; i < KVM_NR_DB_REGS; i++)
10480 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10482 kvm_update_dr7(vcpu);
10484 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10485 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10488 * Trigger an rflags update that will inject or remove the trace
10491 kvm_set_rflags(vcpu, rflags);
10493 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10503 * Translate a guest virtual address to a guest physical address.
10505 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10506 struct kvm_translation *tr)
10508 unsigned long vaddr = tr->linear_address;
10514 idx = srcu_read_lock(&vcpu->kvm->srcu);
10515 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10516 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10517 tr->physical_address = gpa;
10518 tr->valid = gpa != UNMAPPED_GVA;
10526 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10528 struct fxregs_state *fxsave;
10530 if (!vcpu->arch.guest_fpu)
10535 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10536 memcpy(fpu->fpr, fxsave->st_space, 128);
10537 fpu->fcw = fxsave->cwd;
10538 fpu->fsw = fxsave->swd;
10539 fpu->ftwx = fxsave->twd;
10540 fpu->last_opcode = fxsave->fop;
10541 fpu->last_ip = fxsave->rip;
10542 fpu->last_dp = fxsave->rdp;
10543 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10549 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10551 struct fxregs_state *fxsave;
10553 if (!vcpu->arch.guest_fpu)
10558 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10560 memcpy(fxsave->st_space, fpu->fpr, 128);
10561 fxsave->cwd = fpu->fcw;
10562 fxsave->swd = fpu->fsw;
10563 fxsave->twd = fpu->ftwx;
10564 fxsave->fop = fpu->last_opcode;
10565 fxsave->rip = fpu->last_ip;
10566 fxsave->rdp = fpu->last_dp;
10567 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10573 static void store_regs(struct kvm_vcpu *vcpu)
10575 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10577 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10578 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10580 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10581 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10583 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10584 kvm_vcpu_ioctl_x86_get_vcpu_events(
10585 vcpu, &vcpu->run->s.regs.events);
10588 static int sync_regs(struct kvm_vcpu *vcpu)
10590 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10591 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10592 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10594 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10595 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10597 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10599 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10600 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10601 vcpu, &vcpu->run->s.regs.events))
10603 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10609 static void fx_init(struct kvm_vcpu *vcpu)
10611 if (!vcpu->arch.guest_fpu)
10614 fpstate_init(&vcpu->arch.guest_fpu->state);
10615 if (boot_cpu_has(X86_FEATURE_XSAVES))
10616 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10617 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10620 * Ensure guest xcr0 is valid for loading
10622 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10624 vcpu->arch.cr0 |= X86_CR0_ET;
10627 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10629 if (vcpu->arch.guest_fpu) {
10630 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10631 vcpu->arch.guest_fpu = NULL;
10634 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10636 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10638 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10639 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10640 "guest TSC will not be reliable\n");
10645 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10650 vcpu->arch.last_vmentry_cpu = -1;
10652 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10653 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10655 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10657 r = kvm_mmu_create(vcpu);
10661 if (irqchip_in_kernel(vcpu->kvm)) {
10662 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10664 goto fail_mmu_destroy;
10665 if (kvm_apicv_activated(vcpu->kvm))
10666 vcpu->arch.apicv_active = true;
10668 static_branch_inc(&kvm_has_noapic_vcpu);
10672 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10674 goto fail_free_lapic;
10675 vcpu->arch.pio_data = page_address(page);
10677 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10678 GFP_KERNEL_ACCOUNT);
10679 if (!vcpu->arch.mce_banks)
10680 goto fail_free_pio_data;
10681 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10683 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10684 GFP_KERNEL_ACCOUNT))
10685 goto fail_free_mce_banks;
10687 if (!alloc_emulate_ctxt(vcpu))
10688 goto free_wbinvd_dirty_mask;
10690 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10691 GFP_KERNEL_ACCOUNT);
10692 if (!vcpu->arch.user_fpu) {
10693 pr_err("kvm: failed to allocate userspace's fpu\n");
10694 goto free_emulate_ctxt;
10697 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10698 GFP_KERNEL_ACCOUNT);
10699 if (!vcpu->arch.guest_fpu) {
10700 pr_err("kvm: failed to allocate vcpu's fpu\n");
10701 goto free_user_fpu;
10705 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10706 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10708 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10710 kvm_async_pf_hash_reset(vcpu);
10711 kvm_pmu_init(vcpu);
10713 vcpu->arch.pending_external_vector = -1;
10714 vcpu->arch.preempted_in_kernel = false;
10716 #if IS_ENABLED(CONFIG_HYPERV)
10717 vcpu->arch.hv_root_tdp = INVALID_PAGE;
10720 r = static_call(kvm_x86_vcpu_create)(vcpu);
10722 goto free_guest_fpu;
10724 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10725 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10726 kvm_vcpu_mtrr_init(vcpu);
10728 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10729 kvm_vcpu_reset(vcpu, false);
10730 kvm_init_mmu(vcpu);
10735 kvm_free_guest_fpu(vcpu);
10737 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10739 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10740 free_wbinvd_dirty_mask:
10741 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10742 fail_free_mce_banks:
10743 kfree(vcpu->arch.mce_banks);
10744 fail_free_pio_data:
10745 free_page((unsigned long)vcpu->arch.pio_data);
10747 kvm_free_lapic(vcpu);
10749 kvm_mmu_destroy(vcpu);
10753 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10755 struct kvm *kvm = vcpu->kvm;
10757 if (mutex_lock_killable(&vcpu->mutex))
10760 kvm_synchronize_tsc(vcpu, 0);
10763 /* poll control enabled by default */
10764 vcpu->arch.msr_kvm_poll_control = 1;
10766 mutex_unlock(&vcpu->mutex);
10768 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10769 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10770 KVMCLOCK_SYNC_PERIOD);
10773 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10775 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10778 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10780 kvmclock_reset(vcpu);
10782 static_call(kvm_x86_vcpu_free)(vcpu);
10784 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10785 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10786 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10787 kvm_free_guest_fpu(vcpu);
10789 kvm_hv_vcpu_uninit(vcpu);
10790 kvm_pmu_destroy(vcpu);
10791 kfree(vcpu->arch.mce_banks);
10792 kvm_free_lapic(vcpu);
10793 idx = srcu_read_lock(&vcpu->kvm->srcu);
10794 kvm_mmu_destroy(vcpu);
10795 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10796 free_page((unsigned long)vcpu->arch.pio_data);
10797 kvfree(vcpu->arch.cpuid_entries);
10798 if (!lapic_in_kernel(vcpu))
10799 static_branch_dec(&kvm_has_noapic_vcpu);
10802 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10804 unsigned long old_cr0 = kvm_read_cr0(vcpu);
10806 kvm_lapic_reset(vcpu, init_event);
10808 vcpu->arch.hflags = 0;
10810 vcpu->arch.smi_pending = 0;
10811 vcpu->arch.smi_count = 0;
10812 atomic_set(&vcpu->arch.nmi_queued, 0);
10813 vcpu->arch.nmi_pending = 0;
10814 vcpu->arch.nmi_injected = false;
10815 kvm_clear_interrupt_queue(vcpu);
10816 kvm_clear_exception_queue(vcpu);
10818 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10819 kvm_update_dr0123(vcpu);
10820 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10821 vcpu->arch.dr7 = DR7_FIXED_1;
10822 kvm_update_dr7(vcpu);
10824 vcpu->arch.cr2 = 0;
10826 kvm_make_request(KVM_REQ_EVENT, vcpu);
10827 vcpu->arch.apf.msr_en_val = 0;
10828 vcpu->arch.apf.msr_int_val = 0;
10829 vcpu->arch.st.msr_val = 0;
10831 kvmclock_reset(vcpu);
10833 kvm_clear_async_pf_completion_queue(vcpu);
10834 kvm_async_pf_hash_reset(vcpu);
10835 vcpu->arch.apf.halted = false;
10837 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10838 void *mpx_state_buffer;
10841 * To avoid have the INIT path from kvm_apic_has_events() that be
10842 * called with loaded FPU and does not let userspace fix the state.
10845 kvm_put_guest_fpu(vcpu);
10846 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10848 if (mpx_state_buffer)
10849 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10850 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10852 if (mpx_state_buffer)
10853 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10855 kvm_load_guest_fpu(vcpu);
10859 kvm_pmu_reset(vcpu);
10860 vcpu->arch.smbase = 0x30000;
10862 vcpu->arch.msr_misc_features_enables = 0;
10864 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10867 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10868 vcpu->arch.regs_avail = ~0;
10869 vcpu->arch.regs_dirty = ~0;
10871 vcpu->arch.ia32_xss = 0;
10873 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10876 * Reset the MMU context if paging was enabled prior to INIT (which is
10877 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET). Unlike the
10878 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
10879 * checked because it is unconditionally cleared on INIT and all other
10880 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
10881 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
10883 if (old_cr0 & X86_CR0_PG)
10884 kvm_mmu_reset_context(vcpu);
10887 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10889 struct kvm_segment cs;
10891 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10892 cs.selector = vector << 8;
10893 cs.base = vector << 12;
10894 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10895 kvm_rip_write(vcpu, 0);
10897 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10899 int kvm_arch_hardware_enable(void)
10902 struct kvm_vcpu *vcpu;
10907 bool stable, backwards_tsc = false;
10909 kvm_user_return_msr_cpu_online();
10910 ret = static_call(kvm_x86_hardware_enable)();
10914 local_tsc = rdtsc();
10915 stable = !kvm_check_tsc_unstable();
10916 list_for_each_entry(kvm, &vm_list, vm_list) {
10917 kvm_for_each_vcpu(i, vcpu, kvm) {
10918 if (!stable && vcpu->cpu == smp_processor_id())
10919 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10920 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10921 backwards_tsc = true;
10922 if (vcpu->arch.last_host_tsc > max_tsc)
10923 max_tsc = vcpu->arch.last_host_tsc;
10929 * Sometimes, even reliable TSCs go backwards. This happens on
10930 * platforms that reset TSC during suspend or hibernate actions, but
10931 * maintain synchronization. We must compensate. Fortunately, we can
10932 * detect that condition here, which happens early in CPU bringup,
10933 * before any KVM threads can be running. Unfortunately, we can't
10934 * bring the TSCs fully up to date with real time, as we aren't yet far
10935 * enough into CPU bringup that we know how much real time has actually
10936 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10937 * variables that haven't been updated yet.
10939 * So we simply find the maximum observed TSC above, then record the
10940 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10941 * the adjustment will be applied. Note that we accumulate
10942 * adjustments, in case multiple suspend cycles happen before some VCPU
10943 * gets a chance to run again. In the event that no KVM threads get a
10944 * chance to run, we will miss the entire elapsed period, as we'll have
10945 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10946 * loose cycle time. This isn't too big a deal, since the loss will be
10947 * uniform across all VCPUs (not to mention the scenario is extremely
10948 * unlikely). It is possible that a second hibernate recovery happens
10949 * much faster than a first, causing the observed TSC here to be
10950 * smaller; this would require additional padding adjustment, which is
10951 * why we set last_host_tsc to the local tsc observed here.
10953 * N.B. - this code below runs only on platforms with reliable TSC,
10954 * as that is the only way backwards_tsc is set above. Also note
10955 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10956 * have the same delta_cyc adjustment applied if backwards_tsc
10957 * is detected. Note further, this adjustment is only done once,
10958 * as we reset last_host_tsc on all VCPUs to stop this from being
10959 * called multiple times (one for each physical CPU bringup).
10961 * Platforms with unreliable TSCs don't have to deal with this, they
10962 * will be compensated by the logic in vcpu_load, which sets the TSC to
10963 * catchup mode. This will catchup all VCPUs to real time, but cannot
10964 * guarantee that they stay in perfect synchronization.
10966 if (backwards_tsc) {
10967 u64 delta_cyc = max_tsc - local_tsc;
10968 list_for_each_entry(kvm, &vm_list, vm_list) {
10969 kvm->arch.backwards_tsc_observed = true;
10970 kvm_for_each_vcpu(i, vcpu, kvm) {
10971 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10972 vcpu->arch.last_host_tsc = local_tsc;
10973 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10977 * We have to disable TSC offset matching.. if you were
10978 * booting a VM while issuing an S4 host suspend....
10979 * you may have some problem. Solving this issue is
10980 * left as an exercise to the reader.
10982 kvm->arch.last_tsc_nsec = 0;
10983 kvm->arch.last_tsc_write = 0;
10990 void kvm_arch_hardware_disable(void)
10992 static_call(kvm_x86_hardware_disable)();
10993 drop_user_return_notifiers();
10996 int kvm_arch_hardware_setup(void *opaque)
10998 struct kvm_x86_init_ops *ops = opaque;
11001 rdmsrl_safe(MSR_EFER, &host_efer);
11003 if (boot_cpu_has(X86_FEATURE_XSAVES))
11004 rdmsrl(MSR_IA32_XSS, host_xss);
11006 r = ops->hardware_setup();
11010 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11011 kvm_ops_static_call_update();
11013 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11016 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11017 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11018 #undef __kvm_cpu_cap_has
11020 if (kvm_has_tsc_control) {
11022 * Make sure the user can only configure tsc_khz values that
11023 * fit into a signed integer.
11024 * A min value is not calculated because it will always
11025 * be 1 on all machines.
11027 u64 max = min(0x7fffffffULL,
11028 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11029 kvm_max_guest_tsc_khz = max;
11031 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11034 kvm_init_msr_list();
11038 void kvm_arch_hardware_unsetup(void)
11040 static_call(kvm_x86_hardware_unsetup)();
11043 int kvm_arch_check_processor_compat(void *opaque)
11045 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11046 struct kvm_x86_init_ops *ops = opaque;
11048 WARN_ON(!irqs_disabled());
11050 if (__cr4_reserved_bits(cpu_has, c) !=
11051 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11054 return ops->check_processor_compatibility();
11057 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11059 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11061 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11063 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11065 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11068 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11069 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11071 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11073 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11075 vcpu->arch.l1tf_flush_l1d = true;
11076 if (pmu->version && unlikely(pmu->event_count)) {
11077 pmu->need_cleanup = true;
11078 kvm_make_request(KVM_REQ_PMU, vcpu);
11080 static_call(kvm_x86_sched_in)(vcpu, cpu);
11083 void kvm_arch_free_vm(struct kvm *kvm)
11085 kfree(to_kvm_hv(kvm)->hv_pa_pg);
11090 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11095 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11096 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11097 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11098 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11099 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11100 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11102 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11103 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11104 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11105 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11106 &kvm->arch.irq_sources_bitmap);
11108 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11109 mutex_init(&kvm->arch.apic_map_lock);
11110 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
11112 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11113 pvclock_update_vm_gtod_copy(kvm);
11115 kvm->arch.guest_can_read_msr_platform_info = true;
11117 #if IS_ENABLED(CONFIG_HYPERV)
11118 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11119 kvm->arch.hv_root_tdp = INVALID_PAGE;
11122 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11123 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11125 kvm_apicv_init(kvm);
11126 kvm_hv_init_vm(kvm);
11127 kvm_page_track_init(kvm);
11128 kvm_mmu_init_vm(kvm);
11130 return static_call(kvm_x86_vm_init)(kvm);
11133 int kvm_arch_post_init_vm(struct kvm *kvm)
11135 return kvm_mmu_post_init_vm(kvm);
11138 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11141 kvm_mmu_unload(vcpu);
11145 static void kvm_free_vcpus(struct kvm *kvm)
11148 struct kvm_vcpu *vcpu;
11151 * Unpin any mmu pages first.
11153 kvm_for_each_vcpu(i, vcpu, kvm) {
11154 kvm_clear_async_pf_completion_queue(vcpu);
11155 kvm_unload_vcpu_mmu(vcpu);
11157 kvm_for_each_vcpu(i, vcpu, kvm)
11158 kvm_vcpu_destroy(vcpu);
11160 mutex_lock(&kvm->lock);
11161 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11162 kvm->vcpus[i] = NULL;
11164 atomic_set(&kvm->online_vcpus, 0);
11165 mutex_unlock(&kvm->lock);
11168 void kvm_arch_sync_events(struct kvm *kvm)
11170 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11171 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11175 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
11178 * __x86_set_memory_region: Setup KVM internal memory slot
11180 * @kvm: the kvm pointer to the VM.
11181 * @id: the slot ID to setup.
11182 * @gpa: the GPA to install the slot (unused when @size == 0).
11183 * @size: the size of the slot. Set to zero to uninstall a slot.
11185 * This function helps to setup a KVM internal memory slot. Specify
11186 * @size > 0 to install a new slot, while @size == 0 to uninstall a
11187 * slot. The return code can be one of the following:
11189 * HVA: on success (uninstall will return a bogus HVA)
11192 * The caller should always use IS_ERR() to check the return value
11193 * before use. Note, the KVM internal memory slots are guaranteed to
11194 * remain valid and unchanged until the VM is destroyed, i.e., the
11195 * GPA->HVA translation will not change. However, the HVA is a user
11196 * address, i.e. its accessibility is not guaranteed, and must be
11197 * accessed via __copy_{to,from}_user().
11199 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11203 unsigned long hva, old_npages;
11204 struct kvm_memslots *slots = kvm_memslots(kvm);
11205 struct kvm_memory_slot *slot;
11207 /* Called with kvm->slots_lock held. */
11208 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11209 return ERR_PTR_USR(-EINVAL);
11211 slot = id_to_memslot(slots, id);
11213 if (slot && slot->npages)
11214 return ERR_PTR_USR(-EEXIST);
11217 * MAP_SHARED to prevent internal slot pages from being moved
11220 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11221 MAP_SHARED | MAP_ANONYMOUS, 0);
11222 if (IS_ERR((void *)hva))
11223 return (void __user *)hva;
11225 if (!slot || !slot->npages)
11228 old_npages = slot->npages;
11229 hva = slot->userspace_addr;
11232 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11233 struct kvm_userspace_memory_region m;
11235 m.slot = id | (i << 16);
11237 m.guest_phys_addr = gpa;
11238 m.userspace_addr = hva;
11239 m.memory_size = size;
11240 r = __kvm_set_memory_region(kvm, &m);
11242 return ERR_PTR_USR(r);
11246 vm_munmap(hva, old_npages * PAGE_SIZE);
11248 return (void __user *)hva;
11250 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11252 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11254 kvm_mmu_pre_destroy_vm(kvm);
11257 void kvm_arch_destroy_vm(struct kvm *kvm)
11259 if (current->mm == kvm->mm) {
11261 * Free memory regions allocated on behalf of userspace,
11262 * unless the the memory map has changed due to process exit
11265 mutex_lock(&kvm->slots_lock);
11266 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11268 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11270 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11271 mutex_unlock(&kvm->slots_lock);
11273 static_call_cond(kvm_x86_vm_destroy)(kvm);
11274 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11275 kvm_pic_destroy(kvm);
11276 kvm_ioapic_destroy(kvm);
11277 kvm_free_vcpus(kvm);
11278 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11279 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11280 kvm_mmu_uninit_vm(kvm);
11281 kvm_page_track_cleanup(kvm);
11282 kvm_xen_destroy_vm(kvm);
11283 kvm_hv_destroy_vm(kvm);
11286 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11290 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11291 kvfree(slot->arch.rmap[i]);
11292 slot->arch.rmap[i] = NULL;
11296 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11300 memslot_rmap_free(slot);
11302 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11303 kvfree(slot->arch.lpage_info[i - 1]);
11304 slot->arch.lpage_info[i - 1] = NULL;
11307 kvm_page_track_free_memslot(slot);
11310 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
11311 unsigned long npages)
11313 const int sz = sizeof(*slot->arch.rmap[0]);
11316 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11318 int lpages = gfn_to_index(slot->base_gfn + npages - 1,
11319 slot->base_gfn, level) + 1;
11321 WARN_ON(slot->arch.rmap[i]);
11323 slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11324 if (!slot->arch.rmap[i]) {
11325 memslot_rmap_free(slot);
11333 int alloc_all_memslots_rmaps(struct kvm *kvm)
11335 struct kvm_memslots *slots;
11336 struct kvm_memory_slot *slot;
11340 * Check if memslots alreday have rmaps early before acquiring
11341 * the slots_arch_lock below.
11343 if (kvm_memslots_have_rmaps(kvm))
11346 mutex_lock(&kvm->slots_arch_lock);
11349 * Read memslots_have_rmaps again, under the slots arch lock,
11350 * before allocating the rmaps
11352 if (kvm_memslots_have_rmaps(kvm)) {
11353 mutex_unlock(&kvm->slots_arch_lock);
11357 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11358 slots = __kvm_memslots(kvm, i);
11359 kvm_for_each_memslot(slot, slots) {
11360 r = memslot_rmap_alloc(slot, slot->npages);
11362 mutex_unlock(&kvm->slots_arch_lock);
11369 * Ensure that memslots_have_rmaps becomes true strictly after
11370 * all the rmap pointers are set.
11372 smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11373 mutex_unlock(&kvm->slots_arch_lock);
11377 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11378 struct kvm_memory_slot *slot,
11379 unsigned long npages)
11384 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
11385 * old arrays will be freed by __kvm_set_memory_region() if installing
11386 * the new memslot is successful.
11388 memset(&slot->arch, 0, sizeof(slot->arch));
11390 if (kvm_memslots_have_rmaps(kvm)) {
11391 r = memslot_rmap_alloc(slot, npages);
11396 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11397 struct kvm_lpage_info *linfo;
11398 unsigned long ugfn;
11402 lpages = gfn_to_index(slot->base_gfn + npages - 1,
11403 slot->base_gfn, level) + 1;
11405 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11409 slot->arch.lpage_info[i - 1] = linfo;
11411 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11412 linfo[0].disallow_lpage = 1;
11413 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11414 linfo[lpages - 1].disallow_lpage = 1;
11415 ugfn = slot->userspace_addr >> PAGE_SHIFT;
11417 * If the gfn and userspace address are not aligned wrt each
11418 * other, disable large page support for this slot.
11420 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11423 for (j = 0; j < lpages; ++j)
11424 linfo[j].disallow_lpage = 1;
11428 if (kvm_page_track_create_memslot(slot, npages))
11434 memslot_rmap_free(slot);
11436 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11437 kvfree(slot->arch.lpage_info[i - 1]);
11438 slot->arch.lpage_info[i - 1] = NULL;
11443 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11445 struct kvm_vcpu *vcpu;
11449 * memslots->generation has been incremented.
11450 * mmio generation may have reached its maximum value.
11452 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11454 /* Force re-initialization of steal_time cache */
11455 kvm_for_each_vcpu(i, vcpu, kvm)
11456 kvm_vcpu_kick(vcpu);
11459 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11460 struct kvm_memory_slot *memslot,
11461 const struct kvm_userspace_memory_region *mem,
11462 enum kvm_mr_change change)
11464 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11465 return kvm_alloc_memslot_metadata(kvm, memslot,
11466 mem->memory_size >> PAGE_SHIFT);
11471 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11473 struct kvm_arch *ka = &kvm->arch;
11475 if (!kvm_x86_ops.cpu_dirty_log_size)
11478 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11479 (!enable && --ka->cpu_dirty_logging_count == 0))
11480 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11482 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11485 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11486 struct kvm_memory_slot *old,
11487 struct kvm_memory_slot *new,
11488 enum kvm_mr_change change)
11490 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11493 * Update CPU dirty logging if dirty logging is being toggled. This
11494 * applies to all operations.
11496 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11497 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11500 * Nothing more to do for RO slots (which can't be dirtied and can't be
11501 * made writable) or CREATE/MOVE/DELETE of a slot.
11503 * For a memslot with dirty logging disabled:
11504 * CREATE: No dirty mappings will already exist.
11505 * MOVE/DELETE: The old mappings will already have been cleaned up by
11506 * kvm_arch_flush_shadow_memslot()
11508 * For a memslot with dirty logging enabled:
11509 * CREATE: No shadow pages exist, thus nothing to write-protect
11510 * and no dirty bits to clear.
11511 * MOVE/DELETE: The old mappings will already have been cleaned up by
11512 * kvm_arch_flush_shadow_memslot().
11514 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11518 * READONLY and non-flags changes were filtered out above, and the only
11519 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11520 * logging isn't being toggled on or off.
11522 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11525 if (!log_dirty_pages) {
11527 * Dirty logging tracks sptes in 4k granularity, meaning that
11528 * large sptes have to be split. If live migration succeeds,
11529 * the guest in the source machine will be destroyed and large
11530 * sptes will be created in the destination. However, if the
11531 * guest continues to run in the source machine (for example if
11532 * live migration fails), small sptes will remain around and
11533 * cause bad performance.
11535 * Scan sptes if dirty logging has been stopped, dropping those
11536 * which can be collapsed into a single large-page spte. Later
11537 * page faults will create the large-page sptes.
11539 kvm_mmu_zap_collapsible_sptes(kvm, new);
11542 * Initially-all-set does not require write protecting any page,
11543 * because they're all assumed to be dirty.
11545 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11548 if (kvm_x86_ops.cpu_dirty_log_size) {
11549 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11550 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11552 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11557 void kvm_arch_commit_memory_region(struct kvm *kvm,
11558 const struct kvm_userspace_memory_region *mem,
11559 struct kvm_memory_slot *old,
11560 const struct kvm_memory_slot *new,
11561 enum kvm_mr_change change)
11563 if (!kvm->arch.n_requested_mmu_pages)
11564 kvm_mmu_change_mmu_pages(kvm,
11565 kvm_mmu_calculate_default_mmu_pages(kvm));
11568 * FIXME: const-ify all uses of struct kvm_memory_slot.
11570 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11572 /* Free the arrays associated with the old memslot. */
11573 if (change == KVM_MR_MOVE)
11574 kvm_arch_free_memslot(kvm, old);
11577 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11579 kvm_mmu_zap_all(kvm);
11582 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11583 struct kvm_memory_slot *slot)
11585 kvm_page_track_flush_slot(kvm, slot);
11588 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11590 return (is_guest_mode(vcpu) &&
11591 kvm_x86_ops.guest_apic_has_interrupt &&
11592 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11595 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11597 if (!list_empty_careful(&vcpu->async_pf.done))
11600 if (kvm_apic_has_events(vcpu))
11603 if (vcpu->arch.pv.pv_unhalted)
11606 if (vcpu->arch.exception.pending)
11609 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11610 (vcpu->arch.nmi_pending &&
11611 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11614 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11615 (vcpu->arch.smi_pending &&
11616 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11619 if (kvm_arch_interrupt_allowed(vcpu) &&
11620 (kvm_cpu_has_interrupt(vcpu) ||
11621 kvm_guest_apic_has_interrupt(vcpu)))
11624 if (kvm_hv_has_stimer_pending(vcpu))
11627 if (is_guest_mode(vcpu) &&
11628 kvm_x86_ops.nested_ops->hv_timer_pending &&
11629 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11635 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11637 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11640 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11642 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11648 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11650 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11653 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11654 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11655 kvm_test_request(KVM_REQ_EVENT, vcpu))
11658 return kvm_arch_dy_has_pending_interrupt(vcpu);
11661 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11663 if (vcpu->arch.guest_state_protected)
11666 return vcpu->arch.preempted_in_kernel;
11669 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11671 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11674 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11676 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11679 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11681 /* Can't read the RIP when guest state is protected, just return 0 */
11682 if (vcpu->arch.guest_state_protected)
11685 if (is_64_bit_mode(vcpu))
11686 return kvm_rip_read(vcpu);
11687 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11688 kvm_rip_read(vcpu));
11690 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11692 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11694 return kvm_get_linear_rip(vcpu) == linear_rip;
11696 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11698 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11700 unsigned long rflags;
11702 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11703 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11704 rflags &= ~X86_EFLAGS_TF;
11707 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11709 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11711 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11712 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11713 rflags |= X86_EFLAGS_TF;
11714 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11717 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11719 __kvm_set_rflags(vcpu, rflags);
11720 kvm_make_request(KVM_REQ_EVENT, vcpu);
11722 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11724 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11728 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11732 r = kvm_mmu_reload(vcpu);
11736 if (!vcpu->arch.mmu->direct_map &&
11737 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11740 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11743 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11745 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11747 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11750 static inline u32 kvm_async_pf_next_probe(u32 key)
11752 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11755 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11757 u32 key = kvm_async_pf_hash_fn(gfn);
11759 while (vcpu->arch.apf.gfns[key] != ~0)
11760 key = kvm_async_pf_next_probe(key);
11762 vcpu->arch.apf.gfns[key] = gfn;
11765 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11768 u32 key = kvm_async_pf_hash_fn(gfn);
11770 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11771 (vcpu->arch.apf.gfns[key] != gfn &&
11772 vcpu->arch.apf.gfns[key] != ~0); i++)
11773 key = kvm_async_pf_next_probe(key);
11778 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11780 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11783 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11787 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11789 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11793 vcpu->arch.apf.gfns[i] = ~0;
11795 j = kvm_async_pf_next_probe(j);
11796 if (vcpu->arch.apf.gfns[j] == ~0)
11798 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11800 * k lies cyclically in ]i,j]
11802 * |....j i.k.| or |.k..j i...|
11804 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11805 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11810 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11812 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11814 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11818 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11820 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11822 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11823 &token, offset, sizeof(token));
11826 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11828 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11831 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11832 &val, offset, sizeof(val)))
11838 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11840 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11843 if (!kvm_pv_async_pf_enabled(vcpu) ||
11844 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11850 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11852 if (unlikely(!lapic_in_kernel(vcpu) ||
11853 kvm_event_needs_reinjection(vcpu) ||
11854 vcpu->arch.exception.pending))
11857 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11861 * If interrupts are off we cannot even use an artificial
11864 return kvm_arch_interrupt_allowed(vcpu);
11867 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11868 struct kvm_async_pf *work)
11870 struct x86_exception fault;
11872 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11873 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11875 if (kvm_can_deliver_async_pf(vcpu) &&
11876 !apf_put_user_notpresent(vcpu)) {
11877 fault.vector = PF_VECTOR;
11878 fault.error_code_valid = true;
11879 fault.error_code = 0;
11880 fault.nested_page_fault = false;
11881 fault.address = work->arch.token;
11882 fault.async_page_fault = true;
11883 kvm_inject_page_fault(vcpu, &fault);
11887 * It is not possible to deliver a paravirtualized asynchronous
11888 * page fault, but putting the guest in an artificial halt state
11889 * can be beneficial nevertheless: if an interrupt arrives, we
11890 * can deliver it timely and perhaps the guest will schedule
11891 * another process. When the instruction that triggered a page
11892 * fault is retried, hopefully the page will be ready in the host.
11894 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11899 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11900 struct kvm_async_pf *work)
11902 struct kvm_lapic_irq irq = {
11903 .delivery_mode = APIC_DM_FIXED,
11904 .vector = vcpu->arch.apf.vec
11907 if (work->wakeup_all)
11908 work->arch.token = ~0; /* broadcast wakeup */
11910 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11911 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11913 if ((work->wakeup_all || work->notpresent_injected) &&
11914 kvm_pv_async_pf_enabled(vcpu) &&
11915 !apf_put_user_ready(vcpu, work->arch.token)) {
11916 vcpu->arch.apf.pageready_pending = true;
11917 kvm_apic_set_irq(vcpu, &irq, NULL);
11920 vcpu->arch.apf.halted = false;
11921 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11924 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11926 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11927 if (!vcpu->arch.apf.pageready_pending)
11928 kvm_vcpu_kick(vcpu);
11931 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11933 if (!kvm_pv_async_pf_enabled(vcpu))
11936 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11939 void kvm_arch_start_assignment(struct kvm *kvm)
11941 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
11942 static_call_cond(kvm_x86_start_assignment)(kvm);
11944 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11946 void kvm_arch_end_assignment(struct kvm *kvm)
11948 atomic_dec(&kvm->arch.assigned_device_count);
11950 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11952 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11954 return atomic_read(&kvm->arch.assigned_device_count);
11956 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11958 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11960 atomic_inc(&kvm->arch.noncoherent_dma_count);
11962 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11964 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11966 atomic_dec(&kvm->arch.noncoherent_dma_count);
11968 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11970 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11972 return atomic_read(&kvm->arch.noncoherent_dma_count);
11974 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11976 bool kvm_arch_has_irq_bypass(void)
11981 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11982 struct irq_bypass_producer *prod)
11984 struct kvm_kernel_irqfd *irqfd =
11985 container_of(cons, struct kvm_kernel_irqfd, consumer);
11988 irqfd->producer = prod;
11989 kvm_arch_start_assignment(irqfd->kvm);
11990 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11991 prod->irq, irqfd->gsi, 1);
11994 kvm_arch_end_assignment(irqfd->kvm);
11999 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12000 struct irq_bypass_producer *prod)
12003 struct kvm_kernel_irqfd *irqfd =
12004 container_of(cons, struct kvm_kernel_irqfd, consumer);
12006 WARN_ON(irqfd->producer != prod);
12007 irqfd->producer = NULL;
12010 * When producer of consumer is unregistered, we change back to
12011 * remapped mode, so we can re-use the current implementation
12012 * when the irq is masked/disabled or the consumer side (KVM
12013 * int this case doesn't want to receive the interrupts.
12015 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12017 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12018 " fails: %d\n", irqfd->consumer.token, ret);
12020 kvm_arch_end_assignment(irqfd->kvm);
12023 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12024 uint32_t guest_irq, bool set)
12026 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12029 bool kvm_vector_hashing_enabled(void)
12031 return vector_hashing;
12034 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12036 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12038 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12041 int kvm_spec_ctrl_test_value(u64 value)
12044 * test that setting IA32_SPEC_CTRL to given value
12045 * is allowed by the host processor
12049 unsigned long flags;
12052 local_irq_save(flags);
12054 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12056 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12059 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12061 local_irq_restore(flags);
12065 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12067 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12069 struct x86_exception fault;
12070 u32 access = error_code &
12071 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12073 if (!(error_code & PFERR_PRESENT_MASK) ||
12074 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12076 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12077 * tables probably do not match the TLB. Just proceed
12078 * with the error code that the processor gave.
12080 fault.vector = PF_VECTOR;
12081 fault.error_code_valid = true;
12082 fault.error_code = error_code;
12083 fault.nested_page_fault = false;
12084 fault.address = gva;
12086 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12088 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12091 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12092 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12093 * indicates whether exit to userspace is needed.
12095 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12096 struct x86_exception *e)
12098 if (r == X86EMUL_PROPAGATE_FAULT) {
12099 kvm_inject_emulated_page_fault(vcpu, e);
12104 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12105 * while handling a VMX instruction KVM could've handled the request
12106 * correctly by exiting to userspace and performing I/O but there
12107 * doesn't seem to be a real use-case behind such requests, just return
12108 * KVM_EXIT_INTERNAL_ERROR for now.
12110 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
12111 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
12112 vcpu->run->internal.ndata = 0;
12116 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12118 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12121 struct x86_exception e;
12128 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12129 if (r != X86EMUL_CONTINUE)
12130 return kvm_handle_memory_failure(vcpu, r, &e);
12132 if (operand.pcid >> 12 != 0) {
12133 kvm_inject_gp(vcpu, 0);
12137 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12140 case INVPCID_TYPE_INDIV_ADDR:
12141 if ((!pcid_enabled && (operand.pcid != 0)) ||
12142 is_noncanonical_address(operand.gla, vcpu)) {
12143 kvm_inject_gp(vcpu, 0);
12146 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12147 return kvm_skip_emulated_instruction(vcpu);
12149 case INVPCID_TYPE_SINGLE_CTXT:
12150 if (!pcid_enabled && (operand.pcid != 0)) {
12151 kvm_inject_gp(vcpu, 0);
12155 kvm_invalidate_pcid(vcpu, operand.pcid);
12156 return kvm_skip_emulated_instruction(vcpu);
12158 case INVPCID_TYPE_ALL_NON_GLOBAL:
12160 * Currently, KVM doesn't mark global entries in the shadow
12161 * page tables, so a non-global flush just degenerates to a
12162 * global flush. If needed, we could optimize this later by
12163 * keeping track of global entries in shadow page tables.
12167 case INVPCID_TYPE_ALL_INCL_GLOBAL:
12168 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12169 return kvm_skip_emulated_instruction(vcpu);
12172 BUG(); /* We have already checked above that type <= 3 */
12175 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12177 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12179 struct kvm_run *run = vcpu->run;
12180 struct kvm_mmio_fragment *frag;
12183 BUG_ON(!vcpu->mmio_needed);
12185 /* Complete previous fragment */
12186 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12187 len = min(8u, frag->len);
12188 if (!vcpu->mmio_is_write)
12189 memcpy(frag->data, run->mmio.data, len);
12191 if (frag->len <= 8) {
12192 /* Switch to the next fragment. */
12194 vcpu->mmio_cur_fragment++;
12196 /* Go forward to the next mmio piece. */
12202 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12203 vcpu->mmio_needed = 0;
12205 // VMG change, at this point, we're always done
12206 // RIP has already been advanced
12210 // More MMIO is needed
12211 run->mmio.phys_addr = frag->gpa;
12212 run->mmio.len = min(8u, frag->len);
12213 run->mmio.is_write = vcpu->mmio_is_write;
12214 if (run->mmio.is_write)
12215 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12216 run->exit_reason = KVM_EXIT_MMIO;
12218 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12223 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12227 struct kvm_mmio_fragment *frag;
12232 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12233 if (handled == bytes)
12240 /*TODO: Check if need to increment number of frags */
12241 frag = vcpu->mmio_fragments;
12242 vcpu->mmio_nr_fragments = 1;
12247 vcpu->mmio_needed = 1;
12248 vcpu->mmio_cur_fragment = 0;
12250 vcpu->run->mmio.phys_addr = gpa;
12251 vcpu->run->mmio.len = min(8u, frag->len);
12252 vcpu->run->mmio.is_write = 1;
12253 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12254 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12256 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12260 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12262 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12266 struct kvm_mmio_fragment *frag;
12271 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12272 if (handled == bytes)
12279 /*TODO: Check if need to increment number of frags */
12280 frag = vcpu->mmio_fragments;
12281 vcpu->mmio_nr_fragments = 1;
12286 vcpu->mmio_needed = 1;
12287 vcpu->mmio_cur_fragment = 0;
12289 vcpu->run->mmio.phys_addr = gpa;
12290 vcpu->run->mmio.len = min(8u, frag->len);
12291 vcpu->run->mmio.is_write = 0;
12292 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12294 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12298 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12300 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12302 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
12303 vcpu->arch.pio.count * vcpu->arch.pio.size);
12304 vcpu->arch.pio.count = 0;
12309 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12310 unsigned int port, void *data, unsigned int count)
12314 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
12319 vcpu->arch.pio.count = 0;
12324 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12325 unsigned int port, void *data, unsigned int count)
12329 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
12332 vcpu->arch.pio.count = 0;
12334 vcpu->arch.guest_ins_data = data;
12335 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12341 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12342 unsigned int port, void *data, unsigned int count,
12345 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
12346 : kvm_sev_es_outs(vcpu, size, port, data, count);
12348 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12350 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12351 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12352 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12353 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12354 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12355 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12356 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12357 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12358 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12359 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12360 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12361 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12362 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12363 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12364 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12365 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12366 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12367 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12368 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12369 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12370 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12371 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12372 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12373 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12374 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12375 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12376 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);