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)
242 const struct kvm_stats_header kvm_vm_stats_header = {
243 .name_size = KVM_STATS_NAME_SIZE,
244 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
245 .id_offset = sizeof(struct kvm_stats_header),
246 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
247 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
248 sizeof(kvm_vm_stats_desc),
251 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
252 KVM_GENERIC_VCPU_STATS(),
253 STATS_DESC_COUNTER(VCPU, pf_fixed),
254 STATS_DESC_COUNTER(VCPU, pf_guest),
255 STATS_DESC_COUNTER(VCPU, tlb_flush),
256 STATS_DESC_COUNTER(VCPU, invlpg),
257 STATS_DESC_COUNTER(VCPU, exits),
258 STATS_DESC_COUNTER(VCPU, io_exits),
259 STATS_DESC_COUNTER(VCPU, mmio_exits),
260 STATS_DESC_COUNTER(VCPU, signal_exits),
261 STATS_DESC_COUNTER(VCPU, irq_window_exits),
262 STATS_DESC_COUNTER(VCPU, nmi_window_exits),
263 STATS_DESC_COUNTER(VCPU, l1d_flush),
264 STATS_DESC_COUNTER(VCPU, halt_exits),
265 STATS_DESC_COUNTER(VCPU, request_irq_exits),
266 STATS_DESC_COUNTER(VCPU, irq_exits),
267 STATS_DESC_COUNTER(VCPU, host_state_reload),
268 STATS_DESC_COUNTER(VCPU, fpu_reload),
269 STATS_DESC_COUNTER(VCPU, insn_emulation),
270 STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
271 STATS_DESC_COUNTER(VCPU, hypercalls),
272 STATS_DESC_COUNTER(VCPU, irq_injections),
273 STATS_DESC_COUNTER(VCPU, nmi_injections),
274 STATS_DESC_COUNTER(VCPU, req_event),
275 STATS_DESC_COUNTER(VCPU, nested_run),
276 STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
277 STATS_DESC_COUNTER(VCPU, directed_yield_successful),
278 STATS_DESC_ICOUNTER(VCPU, guest_mode)
281 const struct kvm_stats_header kvm_vcpu_stats_header = {
282 .name_size = KVM_STATS_NAME_SIZE,
283 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
284 .id_offset = sizeof(struct kvm_stats_header),
285 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
286 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
287 sizeof(kvm_vcpu_stats_desc),
290 u64 __read_mostly host_xcr0;
291 u64 __read_mostly supported_xcr0;
292 EXPORT_SYMBOL_GPL(supported_xcr0);
294 static struct kmem_cache *x86_fpu_cache;
296 static struct kmem_cache *x86_emulator_cache;
299 * When called, it means the previous get/set msr reached an invalid msr.
300 * Return true if we want to ignore/silent this failed msr access.
302 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
304 const char *op = write ? "wrmsr" : "rdmsr";
307 if (report_ignored_msrs)
308 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
313 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
319 static struct kmem_cache *kvm_alloc_emulator_cache(void)
321 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
322 unsigned int size = sizeof(struct x86_emulate_ctxt);
324 return kmem_cache_create_usercopy("x86_emulator", size,
325 __alignof__(struct x86_emulate_ctxt),
326 SLAB_ACCOUNT, useroffset,
327 size - useroffset, NULL);
330 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
332 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
335 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
336 vcpu->arch.apf.gfns[i] = ~0;
339 static void kvm_on_user_return(struct user_return_notifier *urn)
342 struct kvm_user_return_msrs *msrs
343 = container_of(urn, struct kvm_user_return_msrs, urn);
344 struct kvm_user_return_msr_values *values;
348 * Disabling irqs at this point since the following code could be
349 * interrupted and executed through kvm_arch_hardware_disable()
351 local_irq_save(flags);
352 if (msrs->registered) {
353 msrs->registered = false;
354 user_return_notifier_unregister(urn);
356 local_irq_restore(flags);
357 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
358 values = &msrs->values[slot];
359 if (values->host != values->curr) {
360 wrmsrl(kvm_uret_msrs_list[slot], values->host);
361 values->curr = values->host;
366 static int kvm_probe_user_return_msr(u32 msr)
372 ret = rdmsrl_safe(msr, &val);
375 ret = wrmsrl_safe(msr, val);
381 int kvm_add_user_return_msr(u32 msr)
383 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
385 if (kvm_probe_user_return_msr(msr))
388 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
389 return kvm_nr_uret_msrs++;
391 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
393 int kvm_find_user_return_msr(u32 msr)
397 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
398 if (kvm_uret_msrs_list[i] == msr)
403 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
405 static void kvm_user_return_msr_cpu_online(void)
407 unsigned int cpu = smp_processor_id();
408 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
412 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
413 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
414 msrs->values[i].host = value;
415 msrs->values[i].curr = value;
419 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
421 unsigned int cpu = smp_processor_id();
422 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
425 value = (value & mask) | (msrs->values[slot].host & ~mask);
426 if (value == msrs->values[slot].curr)
428 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
432 msrs->values[slot].curr = value;
433 if (!msrs->registered) {
434 msrs->urn.on_user_return = kvm_on_user_return;
435 user_return_notifier_register(&msrs->urn);
436 msrs->registered = true;
440 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
442 static void drop_user_return_notifiers(void)
444 unsigned int cpu = smp_processor_id();
445 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
447 if (msrs->registered)
448 kvm_on_user_return(&msrs->urn);
451 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
453 return vcpu->arch.apic_base;
455 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
457 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
459 return kvm_apic_mode(kvm_get_apic_base(vcpu));
461 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
463 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
465 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
466 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
467 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
468 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
470 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
472 if (!msr_info->host_initiated) {
473 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
475 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
479 kvm_lapic_set_base(vcpu, msr_info->data);
480 kvm_recalculate_apic_map(vcpu->kvm);
483 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
486 * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
488 * Hardware virtualization extension instructions may fault if a reboot turns
489 * off virtualization while processes are running. Usually after catching the
490 * fault we just panic; during reboot instead the instruction is ignored.
492 noinstr void kvm_spurious_fault(void)
494 /* Fault while not rebooting. We want the trace. */
495 BUG_ON(!kvm_rebooting);
497 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
499 #define EXCPT_BENIGN 0
500 #define EXCPT_CONTRIBUTORY 1
503 static int exception_class(int vector)
513 return EXCPT_CONTRIBUTORY;
520 #define EXCPT_FAULT 0
522 #define EXCPT_ABORT 2
523 #define EXCPT_INTERRUPT 3
525 static int exception_type(int vector)
529 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
530 return EXCPT_INTERRUPT;
534 /* #DB is trap, as instruction watchpoints are handled elsewhere */
535 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
538 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
541 /* Reserved exceptions will result in fault */
545 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
547 unsigned nr = vcpu->arch.exception.nr;
548 bool has_payload = vcpu->arch.exception.has_payload;
549 unsigned long payload = vcpu->arch.exception.payload;
557 * "Certain debug exceptions may clear bit 0-3. The
558 * remaining contents of the DR6 register are never
559 * cleared by the processor".
561 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
563 * In order to reflect the #DB exception payload in guest
564 * dr6, three components need to be considered: active low
565 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
567 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
568 * In the target guest dr6:
569 * FIXED_1 bits should always be set.
570 * Active low bits should be cleared if 1-setting in payload.
571 * Active high bits should be set if 1-setting in payload.
573 * Note, the payload is compatible with the pending debug
574 * exceptions/exit qualification under VMX, that active_low bits
575 * are active high in payload.
576 * So they need to be flipped for DR6.
578 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
579 vcpu->arch.dr6 |= payload;
580 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
583 * The #DB payload is defined as compatible with the 'pending
584 * debug exceptions' field under VMX, not DR6. While bit 12 is
585 * defined in the 'pending debug exceptions' field (enabled
586 * breakpoint), it is reserved and must be zero in DR6.
588 vcpu->arch.dr6 &= ~BIT(12);
591 vcpu->arch.cr2 = payload;
595 vcpu->arch.exception.has_payload = false;
596 vcpu->arch.exception.payload = 0;
598 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
600 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
601 unsigned nr, bool has_error, u32 error_code,
602 bool has_payload, unsigned long payload, bool reinject)
607 kvm_make_request(KVM_REQ_EVENT, vcpu);
609 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
613 * On vmentry, vcpu->arch.exception.pending is only
614 * true if an event injection was blocked by
615 * nested_run_pending. In that case, however,
616 * vcpu_enter_guest requests an immediate exit,
617 * and the guest shouldn't proceed far enough to
620 WARN_ON_ONCE(vcpu->arch.exception.pending);
621 vcpu->arch.exception.injected = true;
622 if (WARN_ON_ONCE(has_payload)) {
624 * A reinjected event has already
625 * delivered its payload.
631 vcpu->arch.exception.pending = true;
632 vcpu->arch.exception.injected = false;
634 vcpu->arch.exception.has_error_code = has_error;
635 vcpu->arch.exception.nr = nr;
636 vcpu->arch.exception.error_code = error_code;
637 vcpu->arch.exception.has_payload = has_payload;
638 vcpu->arch.exception.payload = payload;
639 if (!is_guest_mode(vcpu))
640 kvm_deliver_exception_payload(vcpu);
644 /* to check exception */
645 prev_nr = vcpu->arch.exception.nr;
646 if (prev_nr == DF_VECTOR) {
647 /* triple fault -> shutdown */
648 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
651 class1 = exception_class(prev_nr);
652 class2 = exception_class(nr);
653 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
654 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
656 * Generate double fault per SDM Table 5-5. Set
657 * exception.pending = true so that the double fault
658 * can trigger a nested vmexit.
660 vcpu->arch.exception.pending = true;
661 vcpu->arch.exception.injected = false;
662 vcpu->arch.exception.has_error_code = true;
663 vcpu->arch.exception.nr = DF_VECTOR;
664 vcpu->arch.exception.error_code = 0;
665 vcpu->arch.exception.has_payload = false;
666 vcpu->arch.exception.payload = 0;
668 /* replace previous exception with a new one in a hope
669 that instruction re-execution will regenerate lost
674 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
676 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
678 EXPORT_SYMBOL_GPL(kvm_queue_exception);
680 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
682 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
684 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
686 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
687 unsigned long payload)
689 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
691 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
693 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
694 u32 error_code, unsigned long payload)
696 kvm_multiple_exception(vcpu, nr, true, error_code,
697 true, payload, false);
700 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
703 kvm_inject_gp(vcpu, 0);
705 return kvm_skip_emulated_instruction(vcpu);
709 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
711 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
713 ++vcpu->stat.pf_guest;
714 vcpu->arch.exception.nested_apf =
715 is_guest_mode(vcpu) && fault->async_page_fault;
716 if (vcpu->arch.exception.nested_apf) {
717 vcpu->arch.apf.nested_apf_token = fault->address;
718 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
720 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
724 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
726 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
727 struct x86_exception *fault)
729 struct kvm_mmu *fault_mmu;
730 WARN_ON_ONCE(fault->vector != PF_VECTOR);
732 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
736 * Invalidate the TLB entry for the faulting address, if it exists,
737 * else the access will fault indefinitely (and to emulate hardware).
739 if ((fault->error_code & PFERR_PRESENT_MASK) &&
740 !(fault->error_code & PFERR_RSVD_MASK))
741 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
742 fault_mmu->root_hpa);
744 fault_mmu->inject_page_fault(vcpu, fault);
745 return fault->nested_page_fault;
747 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
749 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
751 atomic_inc(&vcpu->arch.nmi_queued);
752 kvm_make_request(KVM_REQ_NMI, vcpu);
754 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
756 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
758 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
760 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
762 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
764 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
766 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
769 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
770 * a #GP and return false.
772 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
774 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
776 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
779 EXPORT_SYMBOL_GPL(kvm_require_cpl);
781 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
783 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
786 kvm_queue_exception(vcpu, UD_VECTOR);
789 EXPORT_SYMBOL_GPL(kvm_require_dr);
792 * This function will be used to read from the physical memory of the currently
793 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
794 * can read from guest physical or from the guest's guest physical memory.
796 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
797 gfn_t ngfn, void *data, int offset, int len,
800 struct x86_exception exception;
804 ngpa = gfn_to_gpa(ngfn);
805 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
806 if (real_gfn == UNMAPPED_GVA)
809 real_gfn = gpa_to_gfn(real_gfn);
811 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
813 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
815 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
817 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
821 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
823 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
825 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
826 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
829 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
831 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
832 offset * sizeof(u64), sizeof(pdpte),
833 PFERR_USER_MASK|PFERR_WRITE_MASK);
838 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
839 if ((pdpte[i] & PT_PRESENT_MASK) &&
840 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
847 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
848 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
849 vcpu->arch.pdptrs_from_userspace = false;
855 EXPORT_SYMBOL_GPL(load_pdptrs);
857 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
859 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
860 kvm_clear_async_pf_completion_queue(vcpu);
861 kvm_async_pf_hash_reset(vcpu);
864 if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
865 kvm_mmu_reset_context(vcpu);
867 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
868 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
869 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
870 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
872 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
874 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
876 unsigned long old_cr0 = kvm_read_cr0(vcpu);
877 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
882 if (cr0 & 0xffffffff00000000UL)
886 cr0 &= ~CR0_RESERVED_BITS;
888 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
891 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
895 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
896 (cr0 & X86_CR0_PG)) {
901 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
906 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
907 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
908 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
911 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
914 static_call(kvm_x86_set_cr0)(vcpu, cr0);
916 kvm_post_set_cr0(vcpu, old_cr0, cr0);
920 EXPORT_SYMBOL_GPL(kvm_set_cr0);
922 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
924 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
926 EXPORT_SYMBOL_GPL(kvm_lmsw);
928 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
930 if (vcpu->arch.guest_state_protected)
933 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
935 if (vcpu->arch.xcr0 != host_xcr0)
936 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
938 if (vcpu->arch.xsaves_enabled &&
939 vcpu->arch.ia32_xss != host_xss)
940 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
943 if (static_cpu_has(X86_FEATURE_PKU) &&
944 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
945 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
946 vcpu->arch.pkru != vcpu->arch.host_pkru)
947 write_pkru(vcpu->arch.pkru);
949 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
951 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
953 if (vcpu->arch.guest_state_protected)
956 if (static_cpu_has(X86_FEATURE_PKU) &&
957 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
958 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
959 vcpu->arch.pkru = rdpkru();
960 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
961 write_pkru(vcpu->arch.host_pkru);
964 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
966 if (vcpu->arch.xcr0 != host_xcr0)
967 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
969 if (vcpu->arch.xsaves_enabled &&
970 vcpu->arch.ia32_xss != host_xss)
971 wrmsrl(MSR_IA32_XSS, host_xss);
975 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
977 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
980 u64 old_xcr0 = vcpu->arch.xcr0;
983 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
984 if (index != XCR_XFEATURE_ENABLED_MASK)
986 if (!(xcr0 & XFEATURE_MASK_FP))
988 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
992 * Do not allow the guest to set bits that we do not support
993 * saving. However, xcr0 bit 0 is always set, even if the
994 * emulated CPU does not support XSAVE (see fx_init).
996 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
997 if (xcr0 & ~valid_bits)
1000 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1001 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1004 if (xcr0 & XFEATURE_MASK_AVX512) {
1005 if (!(xcr0 & XFEATURE_MASK_YMM))
1007 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1010 vcpu->arch.xcr0 = xcr0;
1012 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1013 kvm_update_cpuid_runtime(vcpu);
1017 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1019 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1020 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1021 kvm_inject_gp(vcpu, 0);
1025 return kvm_skip_emulated_instruction(vcpu);
1027 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1029 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1031 if (cr4 & cr4_reserved_bits)
1034 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1037 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1039 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1041 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1043 if (((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS) ||
1044 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1045 kvm_mmu_reset_context(vcpu);
1047 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1049 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1051 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1052 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1055 if (!kvm_is_valid_cr4(vcpu, cr4))
1058 if (is_long_mode(vcpu)) {
1059 if (!(cr4 & X86_CR4_PAE))
1061 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1063 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1064 && ((cr4 ^ old_cr4) & pdptr_bits)
1065 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1066 kvm_read_cr3(vcpu)))
1069 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1070 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1073 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1074 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1078 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1080 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1084 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1086 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1088 struct kvm_mmu *mmu = vcpu->arch.mmu;
1089 unsigned long roots_to_free = 0;
1093 * If neither the current CR3 nor any of the prev_roots use the given
1094 * PCID, then nothing needs to be done here because a resync will
1095 * happen anyway before switching to any other CR3.
1097 if (kvm_get_active_pcid(vcpu) == pcid) {
1098 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1099 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1102 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1103 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1104 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1106 kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1109 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1111 bool skip_tlb_flush = false;
1112 unsigned long pcid = 0;
1113 #ifdef CONFIG_X86_64
1114 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1117 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1118 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1119 pcid = cr3 & X86_CR3_PCID_MASK;
1123 /* PDPTRs are always reloaded for PAE paging. */
1124 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1125 goto handle_tlb_flush;
1128 * Do not condition the GPA check on long mode, this helper is used to
1129 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1130 * the current vCPU mode is accurate.
1132 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1135 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1138 if (cr3 != kvm_read_cr3(vcpu))
1139 kvm_mmu_new_pgd(vcpu, cr3);
1141 vcpu->arch.cr3 = cr3;
1142 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1146 * A load of CR3 that flushes the TLB flushes only the current PCID,
1147 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1148 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1149 * and it's impossible to use a non-zero PCID when PCID is disabled,
1150 * i.e. only PCID=0 can be relevant.
1152 if (!skip_tlb_flush)
1153 kvm_invalidate_pcid(vcpu, pcid);
1157 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1159 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1161 if (cr8 & CR8_RESERVED_BITS)
1163 if (lapic_in_kernel(vcpu))
1164 kvm_lapic_set_tpr(vcpu, cr8);
1166 vcpu->arch.cr8 = cr8;
1169 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1171 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1173 if (lapic_in_kernel(vcpu))
1174 return kvm_lapic_get_cr8(vcpu);
1176 return vcpu->arch.cr8;
1178 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1180 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1184 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1185 for (i = 0; i < KVM_NR_DB_REGS; i++)
1186 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1190 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1194 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1195 dr7 = vcpu->arch.guest_debug_dr7;
1197 dr7 = vcpu->arch.dr7;
1198 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1199 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1200 if (dr7 & DR7_BP_EN_MASK)
1201 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1203 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1205 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1207 u64 fixed = DR6_FIXED_1;
1209 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1212 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1213 fixed |= DR6_BUS_LOCK;
1217 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1219 size_t size = ARRAY_SIZE(vcpu->arch.db);
1223 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1224 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1225 vcpu->arch.eff_db[dr] = val;
1229 if (!kvm_dr6_valid(val))
1231 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1235 if (!kvm_dr7_valid(val))
1237 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1238 kvm_update_dr7(vcpu);
1244 EXPORT_SYMBOL_GPL(kvm_set_dr);
1246 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1248 size_t size = ARRAY_SIZE(vcpu->arch.db);
1252 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1256 *val = vcpu->arch.dr6;
1260 *val = vcpu->arch.dr7;
1264 EXPORT_SYMBOL_GPL(kvm_get_dr);
1266 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1268 u32 ecx = kvm_rcx_read(vcpu);
1271 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1272 kvm_inject_gp(vcpu, 0);
1276 kvm_rax_write(vcpu, (u32)data);
1277 kvm_rdx_write(vcpu, data >> 32);
1278 return kvm_skip_emulated_instruction(vcpu);
1280 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1283 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1284 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1286 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1287 * extract the supported MSRs from the related const lists.
1288 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1289 * capabilities of the host cpu. This capabilities test skips MSRs that are
1290 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1291 * may depend on host virtualization features rather than host cpu features.
1294 static const u32 msrs_to_save_all[] = {
1295 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1297 #ifdef CONFIG_X86_64
1298 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1300 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1301 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1303 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1304 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1305 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1306 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1307 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1308 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1309 MSR_IA32_UMWAIT_CONTROL,
1311 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1312 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1313 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1314 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1315 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1316 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1317 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1318 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1319 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1320 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1321 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1322 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1323 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1324 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1325 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1326 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1327 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1328 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1329 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1330 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1331 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1332 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1335 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1336 static unsigned num_msrs_to_save;
1338 static const u32 emulated_msrs_all[] = {
1339 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1340 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1341 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1342 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1343 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1344 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1345 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1347 HV_X64_MSR_VP_INDEX,
1348 HV_X64_MSR_VP_RUNTIME,
1349 HV_X64_MSR_SCONTROL,
1350 HV_X64_MSR_STIMER0_CONFIG,
1351 HV_X64_MSR_VP_ASSIST_PAGE,
1352 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1353 HV_X64_MSR_TSC_EMULATION_STATUS,
1354 HV_X64_MSR_SYNDBG_OPTIONS,
1355 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1356 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1357 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1359 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1360 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1362 MSR_IA32_TSC_ADJUST,
1363 MSR_IA32_TSC_DEADLINE,
1364 MSR_IA32_ARCH_CAPABILITIES,
1365 MSR_IA32_PERF_CAPABILITIES,
1366 MSR_IA32_MISC_ENABLE,
1367 MSR_IA32_MCG_STATUS,
1369 MSR_IA32_MCG_EXT_CTL,
1373 MSR_MISC_FEATURES_ENABLES,
1374 MSR_AMD64_VIRT_SPEC_CTRL,
1379 * The following list leaves out MSRs whose values are determined
1380 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1381 * We always support the "true" VMX control MSRs, even if the host
1382 * processor does not, so I am putting these registers here rather
1383 * than in msrs_to_save_all.
1386 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1387 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1388 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1389 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1391 MSR_IA32_VMX_CR0_FIXED0,
1392 MSR_IA32_VMX_CR4_FIXED0,
1393 MSR_IA32_VMX_VMCS_ENUM,
1394 MSR_IA32_VMX_PROCBASED_CTLS2,
1395 MSR_IA32_VMX_EPT_VPID_CAP,
1396 MSR_IA32_VMX_VMFUNC,
1399 MSR_KVM_POLL_CONTROL,
1402 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1403 static unsigned num_emulated_msrs;
1406 * List of msr numbers which are used to expose MSR-based features that
1407 * can be used by a hypervisor to validate requested CPU features.
1409 static const u32 msr_based_features_all[] = {
1411 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1412 MSR_IA32_VMX_PINBASED_CTLS,
1413 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1414 MSR_IA32_VMX_PROCBASED_CTLS,
1415 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1416 MSR_IA32_VMX_EXIT_CTLS,
1417 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1418 MSR_IA32_VMX_ENTRY_CTLS,
1420 MSR_IA32_VMX_CR0_FIXED0,
1421 MSR_IA32_VMX_CR0_FIXED1,
1422 MSR_IA32_VMX_CR4_FIXED0,
1423 MSR_IA32_VMX_CR4_FIXED1,
1424 MSR_IA32_VMX_VMCS_ENUM,
1425 MSR_IA32_VMX_PROCBASED_CTLS2,
1426 MSR_IA32_VMX_EPT_VPID_CAP,
1427 MSR_IA32_VMX_VMFUNC,
1431 MSR_IA32_ARCH_CAPABILITIES,
1432 MSR_IA32_PERF_CAPABILITIES,
1435 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1436 static unsigned int num_msr_based_features;
1438 static u64 kvm_get_arch_capabilities(void)
1442 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1443 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1446 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1447 * the nested hypervisor runs with NX huge pages. If it is not,
1448 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1449 * L1 guests, so it need not worry about its own (L2) guests.
1451 data |= ARCH_CAP_PSCHANGE_MC_NO;
1454 * If we're doing cache flushes (either "always" or "cond")
1455 * we will do one whenever the guest does a vmlaunch/vmresume.
1456 * If an outer hypervisor is doing the cache flush for us
1457 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1458 * capability to the guest too, and if EPT is disabled we're not
1459 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1460 * require a nested hypervisor to do a flush of its own.
1462 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1463 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1465 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1466 data |= ARCH_CAP_RDCL_NO;
1467 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1468 data |= ARCH_CAP_SSB_NO;
1469 if (!boot_cpu_has_bug(X86_BUG_MDS))
1470 data |= ARCH_CAP_MDS_NO;
1472 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1474 * If RTM=0 because the kernel has disabled TSX, the host might
1475 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1476 * and therefore knows that there cannot be TAA) but keep
1477 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1478 * and we want to allow migrating those guests to tsx=off hosts.
1480 data &= ~ARCH_CAP_TAA_NO;
1481 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1482 data |= ARCH_CAP_TAA_NO;
1485 * Nothing to do here; we emulate TSX_CTRL if present on the
1486 * host so the guest can choose between disabling TSX or
1487 * using VERW to clear CPU buffers.
1494 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1496 switch (msr->index) {
1497 case MSR_IA32_ARCH_CAPABILITIES:
1498 msr->data = kvm_get_arch_capabilities();
1500 case MSR_IA32_UCODE_REV:
1501 rdmsrl_safe(msr->index, &msr->data);
1504 return static_call(kvm_x86_get_msr_feature)(msr);
1509 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1511 struct kvm_msr_entry msr;
1515 r = kvm_get_msr_feature(&msr);
1517 if (r == KVM_MSR_RET_INVALID) {
1518 /* Unconditionally clear the output for simplicity */
1520 if (kvm_msr_ignored_check(index, 0, false))
1532 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1534 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1537 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1540 if (efer & (EFER_LME | EFER_LMA) &&
1541 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1544 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1550 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1552 if (efer & efer_reserved_bits)
1555 return __kvm_valid_efer(vcpu, efer);
1557 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1559 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1561 u64 old_efer = vcpu->arch.efer;
1562 u64 efer = msr_info->data;
1565 if (efer & efer_reserved_bits)
1568 if (!msr_info->host_initiated) {
1569 if (!__kvm_valid_efer(vcpu, efer))
1572 if (is_paging(vcpu) &&
1573 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1578 efer |= vcpu->arch.efer & EFER_LMA;
1580 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1586 /* Update reserved bits */
1587 if ((efer ^ old_efer) & EFER_NX)
1588 kvm_mmu_reset_context(vcpu);
1593 void kvm_enable_efer_bits(u64 mask)
1595 efer_reserved_bits &= ~mask;
1597 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1599 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1601 struct kvm_x86_msr_filter *msr_filter;
1602 struct msr_bitmap_range *ranges;
1603 struct kvm *kvm = vcpu->kvm;
1608 /* x2APIC MSRs do not support filtering. */
1609 if (index >= 0x800 && index <= 0x8ff)
1612 idx = srcu_read_lock(&kvm->srcu);
1614 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1620 allowed = msr_filter->default_allow;
1621 ranges = msr_filter->ranges;
1623 for (i = 0; i < msr_filter->count; i++) {
1624 u32 start = ranges[i].base;
1625 u32 end = start + ranges[i].nmsrs;
1626 u32 flags = ranges[i].flags;
1627 unsigned long *bitmap = ranges[i].bitmap;
1629 if ((index >= start) && (index < end) && (flags & type)) {
1630 allowed = !!test_bit(index - start, bitmap);
1636 srcu_read_unlock(&kvm->srcu, idx);
1640 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1643 * Write @data into the MSR specified by @index. Select MSR specific fault
1644 * checks are bypassed if @host_initiated is %true.
1645 * Returns 0 on success, non-0 otherwise.
1646 * Assumes vcpu_load() was already called.
1648 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1649 bool host_initiated)
1651 struct msr_data msr;
1653 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1654 return KVM_MSR_RET_FILTERED;
1659 case MSR_KERNEL_GS_BASE:
1662 if (is_noncanonical_address(data, vcpu))
1665 case MSR_IA32_SYSENTER_EIP:
1666 case MSR_IA32_SYSENTER_ESP:
1668 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1669 * non-canonical address is written on Intel but not on
1670 * AMD (which ignores the top 32-bits, because it does
1671 * not implement 64-bit SYSENTER).
1673 * 64-bit code should hence be able to write a non-canonical
1674 * value on AMD. Making the address canonical ensures that
1675 * vmentry does not fail on Intel after writing a non-canonical
1676 * value, and that something deterministic happens if the guest
1677 * invokes 64-bit SYSENTER.
1679 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1682 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1685 if (!host_initiated &&
1686 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1687 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1691 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1692 * incomplete and conflicting architectural behavior. Current
1693 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1694 * reserved and always read as zeros. Enforce Intel's reserved
1695 * bits check if and only if the guest CPU is Intel, and clear
1696 * the bits in all other cases. This ensures cross-vendor
1697 * migration will provide consistent behavior for the guest.
1699 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1708 msr.host_initiated = host_initiated;
1710 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1713 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1714 u32 index, u64 data, bool host_initiated)
1716 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1718 if (ret == KVM_MSR_RET_INVALID)
1719 if (kvm_msr_ignored_check(index, data, true))
1726 * Read the MSR specified by @index into @data. Select MSR specific fault
1727 * checks are bypassed if @host_initiated is %true.
1728 * Returns 0 on success, non-0 otherwise.
1729 * Assumes vcpu_load() was already called.
1731 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1732 bool host_initiated)
1734 struct msr_data msr;
1737 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1738 return KVM_MSR_RET_FILTERED;
1742 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1745 if (!host_initiated &&
1746 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1747 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1753 msr.host_initiated = host_initiated;
1755 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1761 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1762 u32 index, u64 *data, bool host_initiated)
1764 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1766 if (ret == KVM_MSR_RET_INVALID) {
1767 /* Unconditionally clear *data for simplicity */
1769 if (kvm_msr_ignored_check(index, 0, false))
1776 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1778 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1780 EXPORT_SYMBOL_GPL(kvm_get_msr);
1782 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1784 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1786 EXPORT_SYMBOL_GPL(kvm_set_msr);
1788 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1790 int err = vcpu->run->msr.error;
1792 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1793 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1796 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1799 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1801 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1804 static u64 kvm_msr_reason(int r)
1807 case KVM_MSR_RET_INVALID:
1808 return KVM_MSR_EXIT_REASON_UNKNOWN;
1809 case KVM_MSR_RET_FILTERED:
1810 return KVM_MSR_EXIT_REASON_FILTER;
1812 return KVM_MSR_EXIT_REASON_INVAL;
1816 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1817 u32 exit_reason, u64 data,
1818 int (*completion)(struct kvm_vcpu *vcpu),
1821 u64 msr_reason = kvm_msr_reason(r);
1823 /* Check if the user wanted to know about this MSR fault */
1824 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1827 vcpu->run->exit_reason = exit_reason;
1828 vcpu->run->msr.error = 0;
1829 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1830 vcpu->run->msr.reason = msr_reason;
1831 vcpu->run->msr.index = index;
1832 vcpu->run->msr.data = data;
1833 vcpu->arch.complete_userspace_io = completion;
1838 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1840 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1841 complete_emulated_rdmsr, r);
1844 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1846 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1847 complete_emulated_wrmsr, r);
1850 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1852 u32 ecx = kvm_rcx_read(vcpu);
1856 r = kvm_get_msr(vcpu, ecx, &data);
1858 /* MSR read failed? See if we should ask user space */
1859 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1860 /* Bounce to user space */
1865 trace_kvm_msr_read(ecx, data);
1867 kvm_rax_write(vcpu, data & -1u);
1868 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1870 trace_kvm_msr_read_ex(ecx);
1873 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1875 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1877 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1879 u32 ecx = kvm_rcx_read(vcpu);
1880 u64 data = kvm_read_edx_eax(vcpu);
1883 r = kvm_set_msr(vcpu, ecx, data);
1885 /* MSR write failed? See if we should ask user space */
1886 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1887 /* Bounce to user space */
1890 /* Signal all other negative errors to userspace */
1895 trace_kvm_msr_write(ecx, data);
1897 trace_kvm_msr_write_ex(ecx, data);
1899 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1901 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1903 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1905 return kvm_skip_emulated_instruction(vcpu);
1907 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1909 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1911 /* Treat an INVD instruction as a NOP and just skip it. */
1912 return kvm_emulate_as_nop(vcpu);
1914 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1916 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1918 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1919 return kvm_emulate_as_nop(vcpu);
1921 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1923 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1925 kvm_queue_exception(vcpu, UD_VECTOR);
1928 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1930 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1932 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1933 return kvm_emulate_as_nop(vcpu);
1935 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1937 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1939 xfer_to_guest_mode_prepare();
1940 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1941 xfer_to_guest_mode_work_pending();
1945 * The fast path for frequent and performance sensitive wrmsr emulation,
1946 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1947 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1948 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1949 * other cases which must be called after interrupts are enabled on the host.
1951 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1953 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1956 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1957 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1958 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1959 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1962 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1963 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1964 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1965 trace_kvm_apic_write(APIC_ICR, (u32)data);
1972 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1974 if (!kvm_can_use_hv_timer(vcpu))
1977 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1981 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1983 u32 msr = kvm_rcx_read(vcpu);
1985 fastpath_t ret = EXIT_FASTPATH_NONE;
1988 case APIC_BASE_MSR + (APIC_ICR >> 4):
1989 data = kvm_read_edx_eax(vcpu);
1990 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1991 kvm_skip_emulated_instruction(vcpu);
1992 ret = EXIT_FASTPATH_EXIT_HANDLED;
1995 case MSR_IA32_TSC_DEADLINE:
1996 data = kvm_read_edx_eax(vcpu);
1997 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1998 kvm_skip_emulated_instruction(vcpu);
1999 ret = EXIT_FASTPATH_REENTER_GUEST;
2006 if (ret != EXIT_FASTPATH_NONE)
2007 trace_kvm_msr_write(msr, data);
2011 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2014 * Adapt set_msr() to msr_io()'s calling convention
2016 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2018 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2021 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2023 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2026 #ifdef CONFIG_X86_64
2027 struct pvclock_clock {
2037 struct pvclock_gtod_data {
2040 struct pvclock_clock clock; /* extract of a clocksource struct */
2041 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2047 static struct pvclock_gtod_data pvclock_gtod_data;
2049 static void update_pvclock_gtod(struct timekeeper *tk)
2051 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2053 write_seqcount_begin(&vdata->seq);
2055 /* copy pvclock gtod data */
2056 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2057 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2058 vdata->clock.mask = tk->tkr_mono.mask;
2059 vdata->clock.mult = tk->tkr_mono.mult;
2060 vdata->clock.shift = tk->tkr_mono.shift;
2061 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2062 vdata->clock.offset = tk->tkr_mono.base;
2064 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2065 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2066 vdata->raw_clock.mask = tk->tkr_raw.mask;
2067 vdata->raw_clock.mult = tk->tkr_raw.mult;
2068 vdata->raw_clock.shift = tk->tkr_raw.shift;
2069 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2070 vdata->raw_clock.offset = tk->tkr_raw.base;
2072 vdata->wall_time_sec = tk->xtime_sec;
2074 vdata->offs_boot = tk->offs_boot;
2076 write_seqcount_end(&vdata->seq);
2079 static s64 get_kvmclock_base_ns(void)
2081 /* Count up from boot time, but with the frequency of the raw clock. */
2082 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2085 static s64 get_kvmclock_base_ns(void)
2087 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2088 return ktime_get_boottime_ns();
2092 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2096 struct pvclock_wall_clock wc;
2103 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2108 ++version; /* first time write, random junk */
2112 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2116 * The guest calculates current wall clock time by adding
2117 * system time (updated by kvm_guest_time_update below) to the
2118 * wall clock specified here. We do the reverse here.
2120 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2122 wc.nsec = do_div(wall_nsec, 1000000000);
2123 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2124 wc.version = version;
2126 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2129 wc_sec_hi = wall_nsec >> 32;
2130 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2131 &wc_sec_hi, sizeof(wc_sec_hi));
2135 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2138 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2139 bool old_msr, bool host_initiated)
2141 struct kvm_arch *ka = &vcpu->kvm->arch;
2143 if (vcpu->vcpu_id == 0 && !host_initiated) {
2144 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2145 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2147 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2150 vcpu->arch.time = system_time;
2151 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2153 /* we verify if the enable bit is set... */
2154 vcpu->arch.pv_time_enabled = false;
2155 if (!(system_time & 1))
2158 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2159 &vcpu->arch.pv_time, system_time & ~1ULL,
2160 sizeof(struct pvclock_vcpu_time_info)))
2161 vcpu->arch.pv_time_enabled = true;
2166 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2168 do_shl32_div32(dividend, divisor);
2172 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2173 s8 *pshift, u32 *pmultiplier)
2181 scaled64 = scaled_hz;
2182 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2187 tps32 = (uint32_t)tps64;
2188 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2189 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2197 *pmultiplier = div_frac(scaled64, tps32);
2200 #ifdef CONFIG_X86_64
2201 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2204 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2205 static unsigned long max_tsc_khz;
2207 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2209 u64 v = (u64)khz * (1000000 + ppm);
2214 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2216 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2220 /* Guest TSC same frequency as host TSC? */
2222 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2226 /* TSC scaling supported? */
2227 if (!kvm_has_tsc_control) {
2228 if (user_tsc_khz > tsc_khz) {
2229 vcpu->arch.tsc_catchup = 1;
2230 vcpu->arch.tsc_always_catchup = 1;
2233 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2238 /* TSC scaling required - calculate ratio */
2239 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2240 user_tsc_khz, tsc_khz);
2242 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2243 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2248 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2252 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2254 u32 thresh_lo, thresh_hi;
2255 int use_scaling = 0;
2257 /* tsc_khz can be zero if TSC calibration fails */
2258 if (user_tsc_khz == 0) {
2259 /* set tsc_scaling_ratio to a safe value */
2260 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2264 /* Compute a scale to convert nanoseconds in TSC cycles */
2265 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2266 &vcpu->arch.virtual_tsc_shift,
2267 &vcpu->arch.virtual_tsc_mult);
2268 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2271 * Compute the variation in TSC rate which is acceptable
2272 * within the range of tolerance and decide if the
2273 * rate being applied is within that bounds of the hardware
2274 * rate. If so, no scaling or compensation need be done.
2276 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2277 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2278 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2279 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2282 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2285 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2287 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2288 vcpu->arch.virtual_tsc_mult,
2289 vcpu->arch.virtual_tsc_shift);
2290 tsc += vcpu->arch.this_tsc_write;
2294 static inline int gtod_is_based_on_tsc(int mode)
2296 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2299 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2301 #ifdef CONFIG_X86_64
2303 struct kvm_arch *ka = &vcpu->kvm->arch;
2304 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2306 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2307 atomic_read(&vcpu->kvm->online_vcpus));
2310 * Once the masterclock is enabled, always perform request in
2311 * order to update it.
2313 * In order to enable masterclock, the host clocksource must be TSC
2314 * and the vcpus need to have matched TSCs. When that happens,
2315 * perform request to enable masterclock.
2317 if (ka->use_master_clock ||
2318 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2319 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2321 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2322 atomic_read(&vcpu->kvm->online_vcpus),
2323 ka->use_master_clock, gtod->clock.vclock_mode);
2328 * Multiply tsc by a fixed point number represented by ratio.
2330 * The most significant 64-N bits (mult) of ratio represent the
2331 * integral part of the fixed point number; the remaining N bits
2332 * (frac) represent the fractional part, ie. ratio represents a fixed
2333 * point number (mult + frac * 2^(-N)).
2335 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2337 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2339 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2342 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2346 if (ratio != kvm_default_tsc_scaling_ratio)
2347 _tsc = __scale_tsc(ratio, tsc);
2351 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2353 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2357 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2359 return target_tsc - tsc;
2362 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2364 return vcpu->arch.l1_tsc_offset +
2365 kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2367 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2369 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2373 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2374 nested_offset = l1_offset;
2376 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2377 kvm_tsc_scaling_ratio_frac_bits);
2379 nested_offset += l2_offset;
2380 return nested_offset;
2382 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2384 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2386 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2387 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2388 kvm_tsc_scaling_ratio_frac_bits);
2390 return l1_multiplier;
2392 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2394 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2396 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2397 vcpu->arch.l1_tsc_offset,
2400 vcpu->arch.l1_tsc_offset = l1_offset;
2403 * If we are here because L1 chose not to trap WRMSR to TSC then
2404 * according to the spec this should set L1's TSC (as opposed to
2405 * setting L1's offset for L2).
2407 if (is_guest_mode(vcpu))
2408 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2410 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2411 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2413 vcpu->arch.tsc_offset = l1_offset;
2415 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2418 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2420 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2422 /* Userspace is changing the multiplier while L2 is active */
2423 if (is_guest_mode(vcpu))
2424 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2426 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2428 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2430 if (kvm_has_tsc_control)
2431 static_call(kvm_x86_write_tsc_multiplier)(
2432 vcpu, vcpu->arch.tsc_scaling_ratio);
2435 static inline bool kvm_check_tsc_unstable(void)
2437 #ifdef CONFIG_X86_64
2439 * TSC is marked unstable when we're running on Hyper-V,
2440 * 'TSC page' clocksource is good.
2442 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2445 return check_tsc_unstable();
2448 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2450 struct kvm *kvm = vcpu->kvm;
2451 u64 offset, ns, elapsed;
2452 unsigned long flags;
2454 bool already_matched;
2455 bool synchronizing = false;
2457 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2458 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2459 ns = get_kvmclock_base_ns();
2460 elapsed = ns - kvm->arch.last_tsc_nsec;
2462 if (vcpu->arch.virtual_tsc_khz) {
2465 * detection of vcpu initialization -- need to sync
2466 * with other vCPUs. This particularly helps to keep
2467 * kvm_clock stable after CPU hotplug
2469 synchronizing = true;
2471 u64 tsc_exp = kvm->arch.last_tsc_write +
2472 nsec_to_cycles(vcpu, elapsed);
2473 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2475 * Special case: TSC write with a small delta (1 second)
2476 * of virtual cycle time against real time is
2477 * interpreted as an attempt to synchronize the CPU.
2479 synchronizing = data < tsc_exp + tsc_hz &&
2480 data + tsc_hz > tsc_exp;
2485 * For a reliable TSC, we can match TSC offsets, and for an unstable
2486 * TSC, we add elapsed time in this computation. We could let the
2487 * compensation code attempt to catch up if we fall behind, but
2488 * it's better to try to match offsets from the beginning.
2490 if (synchronizing &&
2491 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2492 if (!kvm_check_tsc_unstable()) {
2493 offset = kvm->arch.cur_tsc_offset;
2495 u64 delta = nsec_to_cycles(vcpu, elapsed);
2497 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2500 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2503 * We split periods of matched TSC writes into generations.
2504 * For each generation, we track the original measured
2505 * nanosecond time, offset, and write, so if TSCs are in
2506 * sync, we can match exact offset, and if not, we can match
2507 * exact software computation in compute_guest_tsc()
2509 * These values are tracked in kvm->arch.cur_xxx variables.
2511 kvm->arch.cur_tsc_generation++;
2512 kvm->arch.cur_tsc_nsec = ns;
2513 kvm->arch.cur_tsc_write = data;
2514 kvm->arch.cur_tsc_offset = offset;
2519 * We also track th most recent recorded KHZ, write and time to
2520 * allow the matching interval to be extended at each write.
2522 kvm->arch.last_tsc_nsec = ns;
2523 kvm->arch.last_tsc_write = data;
2524 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2526 vcpu->arch.last_guest_tsc = data;
2528 /* Keep track of which generation this VCPU has synchronized to */
2529 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2530 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2531 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2533 kvm_vcpu_write_tsc_offset(vcpu, offset);
2534 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2536 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2538 kvm->arch.nr_vcpus_matched_tsc = 0;
2539 } else if (!already_matched) {
2540 kvm->arch.nr_vcpus_matched_tsc++;
2543 kvm_track_tsc_matching(vcpu);
2544 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2547 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2550 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2551 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2554 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2556 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2557 WARN_ON(adjustment < 0);
2558 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2559 vcpu->arch.l1_tsc_scaling_ratio);
2560 adjust_tsc_offset_guest(vcpu, adjustment);
2563 #ifdef CONFIG_X86_64
2565 static u64 read_tsc(void)
2567 u64 ret = (u64)rdtsc_ordered();
2568 u64 last = pvclock_gtod_data.clock.cycle_last;
2570 if (likely(ret >= last))
2574 * GCC likes to generate cmov here, but this branch is extremely
2575 * predictable (it's just a function of time and the likely is
2576 * very likely) and there's a data dependence, so force GCC
2577 * to generate a branch instead. I don't barrier() because
2578 * we don't actually need a barrier, and if this function
2579 * ever gets inlined it will generate worse code.
2585 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2591 switch (clock->vclock_mode) {
2592 case VDSO_CLOCKMODE_HVCLOCK:
2593 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2595 if (tsc_pg_val != U64_MAX) {
2596 /* TSC page valid */
2597 *mode = VDSO_CLOCKMODE_HVCLOCK;
2598 v = (tsc_pg_val - clock->cycle_last) &
2601 /* TSC page invalid */
2602 *mode = VDSO_CLOCKMODE_NONE;
2605 case VDSO_CLOCKMODE_TSC:
2606 *mode = VDSO_CLOCKMODE_TSC;
2607 *tsc_timestamp = read_tsc();
2608 v = (*tsc_timestamp - clock->cycle_last) &
2612 *mode = VDSO_CLOCKMODE_NONE;
2615 if (*mode == VDSO_CLOCKMODE_NONE)
2616 *tsc_timestamp = v = 0;
2618 return v * clock->mult;
2621 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2623 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2629 seq = read_seqcount_begin(>od->seq);
2630 ns = gtod->raw_clock.base_cycles;
2631 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2632 ns >>= gtod->raw_clock.shift;
2633 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2634 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2640 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2642 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2648 seq = read_seqcount_begin(>od->seq);
2649 ts->tv_sec = gtod->wall_time_sec;
2650 ns = gtod->clock.base_cycles;
2651 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2652 ns >>= gtod->clock.shift;
2653 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2655 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2661 /* returns true if host is using TSC based clocksource */
2662 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2664 /* checked again under seqlock below */
2665 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2668 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2672 /* returns true if host is using TSC based clocksource */
2673 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2676 /* checked again under seqlock below */
2677 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2680 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2686 * Assuming a stable TSC across physical CPUS, and a stable TSC
2687 * across virtual CPUs, the following condition is possible.
2688 * Each numbered line represents an event visible to both
2689 * CPUs at the next numbered event.
2691 * "timespecX" represents host monotonic time. "tscX" represents
2694 * VCPU0 on CPU0 | VCPU1 on CPU1
2696 * 1. read timespec0,tsc0
2697 * 2. | timespec1 = timespec0 + N
2699 * 3. transition to guest | transition to guest
2700 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2701 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2702 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2704 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2707 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2709 * - 0 < N - M => M < N
2711 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2712 * always the case (the difference between two distinct xtime instances
2713 * might be smaller then the difference between corresponding TSC reads,
2714 * when updating guest vcpus pvclock areas).
2716 * To avoid that problem, do not allow visibility of distinct
2717 * system_timestamp/tsc_timestamp values simultaneously: use a master
2718 * copy of host monotonic time values. Update that master copy
2721 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2725 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2727 #ifdef CONFIG_X86_64
2728 struct kvm_arch *ka = &kvm->arch;
2730 bool host_tsc_clocksource, vcpus_matched;
2732 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2733 atomic_read(&kvm->online_vcpus));
2736 * If the host uses TSC clock, then passthrough TSC as stable
2739 host_tsc_clocksource = kvm_get_time_and_clockread(
2740 &ka->master_kernel_ns,
2741 &ka->master_cycle_now);
2743 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2744 && !ka->backwards_tsc_observed
2745 && !ka->boot_vcpu_runs_old_kvmclock;
2747 if (ka->use_master_clock)
2748 atomic_set(&kvm_guest_has_master_clock, 1);
2750 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2751 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2756 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2758 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2761 static void kvm_gen_update_masterclock(struct kvm *kvm)
2763 #ifdef CONFIG_X86_64
2765 struct kvm_vcpu *vcpu;
2766 struct kvm_arch *ka = &kvm->arch;
2767 unsigned long flags;
2769 kvm_hv_invalidate_tsc_page(kvm);
2771 kvm_make_mclock_inprogress_request(kvm);
2773 /* no guest entries from this point */
2774 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2775 pvclock_update_vm_gtod_copy(kvm);
2776 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2778 kvm_for_each_vcpu(i, vcpu, kvm)
2779 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2781 /* guest entries allowed */
2782 kvm_for_each_vcpu(i, vcpu, kvm)
2783 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2787 u64 get_kvmclock_ns(struct kvm *kvm)
2789 struct kvm_arch *ka = &kvm->arch;
2790 struct pvclock_vcpu_time_info hv_clock;
2791 unsigned long flags;
2794 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2795 if (!ka->use_master_clock) {
2796 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2797 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2800 hv_clock.tsc_timestamp = ka->master_cycle_now;
2801 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2802 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2804 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2807 if (__this_cpu_read(cpu_tsc_khz)) {
2808 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2809 &hv_clock.tsc_shift,
2810 &hv_clock.tsc_to_system_mul);
2811 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2813 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2820 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2821 struct gfn_to_hva_cache *cache,
2822 unsigned int offset)
2824 struct kvm_vcpu_arch *vcpu = &v->arch;
2825 struct pvclock_vcpu_time_info guest_hv_clock;
2827 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2828 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2831 /* This VCPU is paused, but it's legal for a guest to read another
2832 * VCPU's kvmclock, so we really have to follow the specification where
2833 * it says that version is odd if data is being modified, and even after
2836 * Version field updates must be kept separate. This is because
2837 * kvm_write_guest_cached might use a "rep movs" instruction, and
2838 * writes within a string instruction are weakly ordered. So there
2839 * are three writes overall.
2841 * As a small optimization, only write the version field in the first
2842 * and third write. The vcpu->pv_time cache is still valid, because the
2843 * version field is the first in the struct.
2845 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2847 if (guest_hv_clock.version & 1)
2848 ++guest_hv_clock.version; /* first time write, random junk */
2850 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2851 kvm_write_guest_offset_cached(v->kvm, cache,
2852 &vcpu->hv_clock, offset,
2853 sizeof(vcpu->hv_clock.version));
2857 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2858 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2860 if (vcpu->pvclock_set_guest_stopped_request) {
2861 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2862 vcpu->pvclock_set_guest_stopped_request = false;
2865 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2867 kvm_write_guest_offset_cached(v->kvm, cache,
2868 &vcpu->hv_clock, offset,
2869 sizeof(vcpu->hv_clock));
2873 vcpu->hv_clock.version++;
2874 kvm_write_guest_offset_cached(v->kvm, cache,
2875 &vcpu->hv_clock, offset,
2876 sizeof(vcpu->hv_clock.version));
2879 static int kvm_guest_time_update(struct kvm_vcpu *v)
2881 unsigned long flags, tgt_tsc_khz;
2882 struct kvm_vcpu_arch *vcpu = &v->arch;
2883 struct kvm_arch *ka = &v->kvm->arch;
2885 u64 tsc_timestamp, host_tsc;
2887 bool use_master_clock;
2893 * If the host uses TSC clock, then passthrough TSC as stable
2896 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2897 use_master_clock = ka->use_master_clock;
2898 if (use_master_clock) {
2899 host_tsc = ka->master_cycle_now;
2900 kernel_ns = ka->master_kernel_ns;
2902 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2904 /* Keep irq disabled to prevent changes to the clock */
2905 local_irq_save(flags);
2906 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2907 if (unlikely(tgt_tsc_khz == 0)) {
2908 local_irq_restore(flags);
2909 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2912 if (!use_master_clock) {
2914 kernel_ns = get_kvmclock_base_ns();
2917 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2920 * We may have to catch up the TSC to match elapsed wall clock
2921 * time for two reasons, even if kvmclock is used.
2922 * 1) CPU could have been running below the maximum TSC rate
2923 * 2) Broken TSC compensation resets the base at each VCPU
2924 * entry to avoid unknown leaps of TSC even when running
2925 * again on the same CPU. This may cause apparent elapsed
2926 * time to disappear, and the guest to stand still or run
2929 if (vcpu->tsc_catchup) {
2930 u64 tsc = compute_guest_tsc(v, kernel_ns);
2931 if (tsc > tsc_timestamp) {
2932 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2933 tsc_timestamp = tsc;
2937 local_irq_restore(flags);
2939 /* With all the info we got, fill in the values */
2941 if (kvm_has_tsc_control)
2942 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
2943 v->arch.l1_tsc_scaling_ratio);
2945 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2946 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2947 &vcpu->hv_clock.tsc_shift,
2948 &vcpu->hv_clock.tsc_to_system_mul);
2949 vcpu->hw_tsc_khz = tgt_tsc_khz;
2952 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2953 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2954 vcpu->last_guest_tsc = tsc_timestamp;
2956 /* If the host uses TSC clocksource, then it is stable */
2958 if (use_master_clock)
2959 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2961 vcpu->hv_clock.flags = pvclock_flags;
2963 if (vcpu->pv_time_enabled)
2964 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2965 if (vcpu->xen.vcpu_info_set)
2966 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2967 offsetof(struct compat_vcpu_info, time));
2968 if (vcpu->xen.vcpu_time_info_set)
2969 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2970 if (v == kvm_get_vcpu(v->kvm, 0))
2971 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2976 * kvmclock updates which are isolated to a given vcpu, such as
2977 * vcpu->cpu migration, should not allow system_timestamp from
2978 * the rest of the vcpus to remain static. Otherwise ntp frequency
2979 * correction applies to one vcpu's system_timestamp but not
2982 * So in those cases, request a kvmclock update for all vcpus.
2983 * We need to rate-limit these requests though, as they can
2984 * considerably slow guests that have a large number of vcpus.
2985 * The time for a remote vcpu to update its kvmclock is bound
2986 * by the delay we use to rate-limit the updates.
2989 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2991 static void kvmclock_update_fn(struct work_struct *work)
2994 struct delayed_work *dwork = to_delayed_work(work);
2995 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2996 kvmclock_update_work);
2997 struct kvm *kvm = container_of(ka, struct kvm, arch);
2998 struct kvm_vcpu *vcpu;
3000 kvm_for_each_vcpu(i, vcpu, kvm) {
3001 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3002 kvm_vcpu_kick(vcpu);
3006 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3008 struct kvm *kvm = v->kvm;
3010 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3011 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3012 KVMCLOCK_UPDATE_DELAY);
3015 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3017 static void kvmclock_sync_fn(struct work_struct *work)
3019 struct delayed_work *dwork = to_delayed_work(work);
3020 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3021 kvmclock_sync_work);
3022 struct kvm *kvm = container_of(ka, struct kvm, arch);
3024 if (!kvmclock_periodic_sync)
3027 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3028 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3029 KVMCLOCK_SYNC_PERIOD);
3033 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3035 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3037 /* McStatusWrEn enabled? */
3038 if (guest_cpuid_is_amd_or_hygon(vcpu))
3039 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3044 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3046 u64 mcg_cap = vcpu->arch.mcg_cap;
3047 unsigned bank_num = mcg_cap & 0xff;
3048 u32 msr = msr_info->index;
3049 u64 data = msr_info->data;
3052 case MSR_IA32_MCG_STATUS:
3053 vcpu->arch.mcg_status = data;
3055 case MSR_IA32_MCG_CTL:
3056 if (!(mcg_cap & MCG_CTL_P) &&
3057 (data || !msr_info->host_initiated))
3059 if (data != 0 && data != ~(u64)0)
3061 vcpu->arch.mcg_ctl = data;
3064 if (msr >= MSR_IA32_MC0_CTL &&
3065 msr < MSR_IA32_MCx_CTL(bank_num)) {
3066 u32 offset = array_index_nospec(
3067 msr - MSR_IA32_MC0_CTL,
3068 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3070 /* only 0 or all 1s can be written to IA32_MCi_CTL
3071 * some Linux kernels though clear bit 10 in bank 4 to
3072 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3073 * this to avoid an uncatched #GP in the guest
3075 if ((offset & 0x3) == 0 &&
3076 data != 0 && (data | (1 << 10)) != ~(u64)0)
3080 if (!msr_info->host_initiated &&
3081 (offset & 0x3) == 1 && data != 0) {
3082 if (!can_set_mci_status(vcpu))
3086 vcpu->arch.mce_banks[offset] = data;
3094 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3096 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3098 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3101 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3103 gpa_t gpa = data & ~0x3f;
3105 /* Bits 4:5 are reserved, Should be zero */
3109 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3110 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3113 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3114 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3117 if (!lapic_in_kernel(vcpu))
3118 return data ? 1 : 0;
3120 vcpu->arch.apf.msr_en_val = data;
3122 if (!kvm_pv_async_pf_enabled(vcpu)) {
3123 kvm_clear_async_pf_completion_queue(vcpu);
3124 kvm_async_pf_hash_reset(vcpu);
3128 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3132 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3133 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3135 kvm_async_pf_wakeup_all(vcpu);
3140 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3142 /* Bits 8-63 are reserved */
3146 if (!lapic_in_kernel(vcpu))
3149 vcpu->arch.apf.msr_int_val = data;
3151 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3156 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3158 vcpu->arch.pv_time_enabled = false;
3159 vcpu->arch.time = 0;
3162 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3164 ++vcpu->stat.tlb_flush;
3165 static_call(kvm_x86_tlb_flush_all)(vcpu);
3168 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3170 ++vcpu->stat.tlb_flush;
3174 * A TLB flush on behalf of the guest is equivalent to
3175 * INVPCID(all), toggling CR4.PGE, etc., which requires
3176 * a forced sync of the shadow page tables. Unload the
3177 * entire MMU here and the subsequent load will sync the
3178 * shadow page tables, and also flush the TLB.
3180 kvm_mmu_unload(vcpu);
3184 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3187 static void record_steal_time(struct kvm_vcpu *vcpu)
3189 struct kvm_host_map map;
3190 struct kvm_steal_time *st;
3192 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3193 kvm_xen_runstate_set_running(vcpu);
3197 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3200 /* -EAGAIN is returned in atomic context so we can just return. */
3201 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3202 &map, &vcpu->arch.st.cache, false))
3206 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3209 * Doing a TLB flush here, on the guest's behalf, can avoid
3212 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3213 u8 st_preempted = xchg(&st->preempted, 0);
3215 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3216 st_preempted & KVM_VCPU_FLUSH_TLB);
3217 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3218 kvm_vcpu_flush_tlb_guest(vcpu);
3223 vcpu->arch.st.preempted = 0;
3225 if (st->version & 1)
3226 st->version += 1; /* first time write, random junk */
3232 st->steal += current->sched_info.run_delay -
3233 vcpu->arch.st.last_steal;
3234 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3240 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3243 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3246 u32 msr = msr_info->index;
3247 u64 data = msr_info->data;
3249 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3250 return kvm_xen_write_hypercall_page(vcpu, data);
3253 case MSR_AMD64_NB_CFG:
3254 case MSR_IA32_UCODE_WRITE:
3255 case MSR_VM_HSAVE_PA:
3256 case MSR_AMD64_PATCH_LOADER:
3257 case MSR_AMD64_BU_CFG2:
3258 case MSR_AMD64_DC_CFG:
3259 case MSR_F15H_EX_CFG:
3262 case MSR_IA32_UCODE_REV:
3263 if (msr_info->host_initiated)
3264 vcpu->arch.microcode_version = data;
3266 case MSR_IA32_ARCH_CAPABILITIES:
3267 if (!msr_info->host_initiated)
3269 vcpu->arch.arch_capabilities = data;
3271 case MSR_IA32_PERF_CAPABILITIES: {
3272 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3274 if (!msr_info->host_initiated)
3276 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3278 if (data & ~msr_ent.data)
3281 vcpu->arch.perf_capabilities = data;
3286 return set_efer(vcpu, msr_info);
3288 data &= ~(u64)0x40; /* ignore flush filter disable */
3289 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3290 data &= ~(u64)0x8; /* ignore TLB cache disable */
3292 /* Handle McStatusWrEn */
3293 if (data == BIT_ULL(18)) {
3294 vcpu->arch.msr_hwcr = data;
3295 } else if (data != 0) {
3296 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3301 case MSR_FAM10H_MMIO_CONF_BASE:
3303 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3308 case 0x200 ... 0x2ff:
3309 return kvm_mtrr_set_msr(vcpu, msr, data);
3310 case MSR_IA32_APICBASE:
3311 return kvm_set_apic_base(vcpu, msr_info);
3312 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3313 return kvm_x2apic_msr_write(vcpu, msr, data);
3314 case MSR_IA32_TSC_DEADLINE:
3315 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3317 case MSR_IA32_TSC_ADJUST:
3318 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3319 if (!msr_info->host_initiated) {
3320 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3321 adjust_tsc_offset_guest(vcpu, adj);
3323 vcpu->arch.ia32_tsc_adjust_msr = data;
3326 case MSR_IA32_MISC_ENABLE:
3327 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3328 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3329 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3331 vcpu->arch.ia32_misc_enable_msr = data;
3332 kvm_update_cpuid_runtime(vcpu);
3334 vcpu->arch.ia32_misc_enable_msr = data;
3337 case MSR_IA32_SMBASE:
3338 if (!msr_info->host_initiated)
3340 vcpu->arch.smbase = data;
3342 case MSR_IA32_POWER_CTL:
3343 vcpu->arch.msr_ia32_power_ctl = data;
3346 if (msr_info->host_initiated) {
3347 kvm_synchronize_tsc(vcpu, data);
3349 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3350 adjust_tsc_offset_guest(vcpu, adj);
3351 vcpu->arch.ia32_tsc_adjust_msr += adj;
3355 if (!msr_info->host_initiated &&
3356 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3359 * KVM supports exposing PT to the guest, but does not support
3360 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3361 * XSAVES/XRSTORS to save/restore PT MSRs.
3363 if (data & ~supported_xss)
3365 vcpu->arch.ia32_xss = data;
3368 if (!msr_info->host_initiated)
3370 vcpu->arch.smi_count = data;
3372 case MSR_KVM_WALL_CLOCK_NEW:
3373 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3376 vcpu->kvm->arch.wall_clock = data;
3377 kvm_write_wall_clock(vcpu->kvm, data, 0);
3379 case MSR_KVM_WALL_CLOCK:
3380 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3383 vcpu->kvm->arch.wall_clock = data;
3384 kvm_write_wall_clock(vcpu->kvm, data, 0);
3386 case MSR_KVM_SYSTEM_TIME_NEW:
3387 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3390 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3392 case MSR_KVM_SYSTEM_TIME:
3393 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3396 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3398 case MSR_KVM_ASYNC_PF_EN:
3399 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3402 if (kvm_pv_enable_async_pf(vcpu, data))
3405 case MSR_KVM_ASYNC_PF_INT:
3406 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3409 if (kvm_pv_enable_async_pf_int(vcpu, data))
3412 case MSR_KVM_ASYNC_PF_ACK:
3413 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3416 vcpu->arch.apf.pageready_pending = false;
3417 kvm_check_async_pf_completion(vcpu);
3420 case MSR_KVM_STEAL_TIME:
3421 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3424 if (unlikely(!sched_info_on()))
3427 if (data & KVM_STEAL_RESERVED_MASK)
3430 vcpu->arch.st.msr_val = data;
3432 if (!(data & KVM_MSR_ENABLED))
3435 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3438 case MSR_KVM_PV_EOI_EN:
3439 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3442 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3446 case MSR_KVM_POLL_CONTROL:
3447 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3450 /* only enable bit supported */
3451 if (data & (-1ULL << 1))
3454 vcpu->arch.msr_kvm_poll_control = data;
3457 case MSR_IA32_MCG_CTL:
3458 case MSR_IA32_MCG_STATUS:
3459 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3460 return set_msr_mce(vcpu, msr_info);
3462 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3463 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3466 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3467 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3468 if (kvm_pmu_is_valid_msr(vcpu, msr))
3469 return kvm_pmu_set_msr(vcpu, msr_info);
3471 if (pr || data != 0)
3472 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3473 "0x%x data 0x%llx\n", msr, data);
3475 case MSR_K7_CLK_CTL:
3477 * Ignore all writes to this no longer documented MSR.
3478 * Writes are only relevant for old K7 processors,
3479 * all pre-dating SVM, but a recommended workaround from
3480 * AMD for these chips. It is possible to specify the
3481 * affected processor models on the command line, hence
3482 * the need to ignore the workaround.
3485 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3486 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3487 case HV_X64_MSR_SYNDBG_OPTIONS:
3488 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3489 case HV_X64_MSR_CRASH_CTL:
3490 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3491 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3492 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3493 case HV_X64_MSR_TSC_EMULATION_STATUS:
3494 return kvm_hv_set_msr_common(vcpu, msr, data,
3495 msr_info->host_initiated);
3496 case MSR_IA32_BBL_CR_CTL3:
3497 /* Drop writes to this legacy MSR -- see rdmsr
3498 * counterpart for further detail.
3500 if (report_ignored_msrs)
3501 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3504 case MSR_AMD64_OSVW_ID_LENGTH:
3505 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3507 vcpu->arch.osvw.length = data;
3509 case MSR_AMD64_OSVW_STATUS:
3510 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3512 vcpu->arch.osvw.status = data;
3514 case MSR_PLATFORM_INFO:
3515 if (!msr_info->host_initiated ||
3516 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3517 cpuid_fault_enabled(vcpu)))
3519 vcpu->arch.msr_platform_info = data;
3521 case MSR_MISC_FEATURES_ENABLES:
3522 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3523 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3524 !supports_cpuid_fault(vcpu)))
3526 vcpu->arch.msr_misc_features_enables = data;
3529 if (kvm_pmu_is_valid_msr(vcpu, msr))
3530 return kvm_pmu_set_msr(vcpu, msr_info);
3531 return KVM_MSR_RET_INVALID;
3535 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3537 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3540 u64 mcg_cap = vcpu->arch.mcg_cap;
3541 unsigned bank_num = mcg_cap & 0xff;
3544 case MSR_IA32_P5_MC_ADDR:
3545 case MSR_IA32_P5_MC_TYPE:
3548 case MSR_IA32_MCG_CAP:
3549 data = vcpu->arch.mcg_cap;
3551 case MSR_IA32_MCG_CTL:
3552 if (!(mcg_cap & MCG_CTL_P) && !host)
3554 data = vcpu->arch.mcg_ctl;
3556 case MSR_IA32_MCG_STATUS:
3557 data = vcpu->arch.mcg_status;
3560 if (msr >= MSR_IA32_MC0_CTL &&
3561 msr < MSR_IA32_MCx_CTL(bank_num)) {
3562 u32 offset = array_index_nospec(
3563 msr - MSR_IA32_MC0_CTL,
3564 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3566 data = vcpu->arch.mce_banks[offset];
3575 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3577 switch (msr_info->index) {
3578 case MSR_IA32_PLATFORM_ID:
3579 case MSR_IA32_EBL_CR_POWERON:
3580 case MSR_IA32_LASTBRANCHFROMIP:
3581 case MSR_IA32_LASTBRANCHTOIP:
3582 case MSR_IA32_LASTINTFROMIP:
3583 case MSR_IA32_LASTINTTOIP:
3584 case MSR_AMD64_SYSCFG:
3585 case MSR_K8_TSEG_ADDR:
3586 case MSR_K8_TSEG_MASK:
3587 case MSR_VM_HSAVE_PA:
3588 case MSR_K8_INT_PENDING_MSG:
3589 case MSR_AMD64_NB_CFG:
3590 case MSR_FAM10H_MMIO_CONF_BASE:
3591 case MSR_AMD64_BU_CFG2:
3592 case MSR_IA32_PERF_CTL:
3593 case MSR_AMD64_DC_CFG:
3594 case MSR_F15H_EX_CFG:
3596 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3597 * limit) MSRs. Just return 0, as we do not want to expose the host
3598 * data here. Do not conditionalize this on CPUID, as KVM does not do
3599 * so for existing CPU-specific MSRs.
3601 case MSR_RAPL_POWER_UNIT:
3602 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3603 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3604 case MSR_PKG_ENERGY_STATUS: /* Total package */
3605 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3608 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3609 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3610 return kvm_pmu_get_msr(vcpu, msr_info);
3611 if (!msr_info->host_initiated)
3615 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3616 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3617 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3618 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3619 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3620 return kvm_pmu_get_msr(vcpu, msr_info);
3623 case MSR_IA32_UCODE_REV:
3624 msr_info->data = vcpu->arch.microcode_version;
3626 case MSR_IA32_ARCH_CAPABILITIES:
3627 if (!msr_info->host_initiated &&
3628 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3630 msr_info->data = vcpu->arch.arch_capabilities;
3632 case MSR_IA32_PERF_CAPABILITIES:
3633 if (!msr_info->host_initiated &&
3634 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3636 msr_info->data = vcpu->arch.perf_capabilities;
3638 case MSR_IA32_POWER_CTL:
3639 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3641 case MSR_IA32_TSC: {
3643 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3644 * even when not intercepted. AMD manual doesn't explicitly
3645 * state this but appears to behave the same.
3647 * On userspace reads and writes, however, we unconditionally
3648 * return L1's TSC value to ensure backwards-compatible
3649 * behavior for migration.
3653 if (msr_info->host_initiated) {
3654 offset = vcpu->arch.l1_tsc_offset;
3655 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3657 offset = vcpu->arch.tsc_offset;
3658 ratio = vcpu->arch.tsc_scaling_ratio;
3661 msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3665 case 0x200 ... 0x2ff:
3666 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3667 case 0xcd: /* fsb frequency */
3671 * MSR_EBC_FREQUENCY_ID
3672 * Conservative value valid for even the basic CPU models.
3673 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3674 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3675 * and 266MHz for model 3, or 4. Set Core Clock
3676 * Frequency to System Bus Frequency Ratio to 1 (bits
3677 * 31:24) even though these are only valid for CPU
3678 * models > 2, however guests may end up dividing or
3679 * multiplying by zero otherwise.
3681 case MSR_EBC_FREQUENCY_ID:
3682 msr_info->data = 1 << 24;
3684 case MSR_IA32_APICBASE:
3685 msr_info->data = kvm_get_apic_base(vcpu);
3687 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3688 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3689 case MSR_IA32_TSC_DEADLINE:
3690 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3692 case MSR_IA32_TSC_ADJUST:
3693 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3695 case MSR_IA32_MISC_ENABLE:
3696 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3698 case MSR_IA32_SMBASE:
3699 if (!msr_info->host_initiated)
3701 msr_info->data = vcpu->arch.smbase;
3704 msr_info->data = vcpu->arch.smi_count;
3706 case MSR_IA32_PERF_STATUS:
3707 /* TSC increment by tick */
3708 msr_info->data = 1000ULL;
3709 /* CPU multiplier */
3710 msr_info->data |= (((uint64_t)4ULL) << 40);
3713 msr_info->data = vcpu->arch.efer;
3715 case MSR_KVM_WALL_CLOCK:
3716 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3719 msr_info->data = vcpu->kvm->arch.wall_clock;
3721 case MSR_KVM_WALL_CLOCK_NEW:
3722 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3725 msr_info->data = vcpu->kvm->arch.wall_clock;
3727 case MSR_KVM_SYSTEM_TIME:
3728 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3731 msr_info->data = vcpu->arch.time;
3733 case MSR_KVM_SYSTEM_TIME_NEW:
3734 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3737 msr_info->data = vcpu->arch.time;
3739 case MSR_KVM_ASYNC_PF_EN:
3740 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3743 msr_info->data = vcpu->arch.apf.msr_en_val;
3745 case MSR_KVM_ASYNC_PF_INT:
3746 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3749 msr_info->data = vcpu->arch.apf.msr_int_val;
3751 case MSR_KVM_ASYNC_PF_ACK:
3752 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3757 case MSR_KVM_STEAL_TIME:
3758 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3761 msr_info->data = vcpu->arch.st.msr_val;
3763 case MSR_KVM_PV_EOI_EN:
3764 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3767 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3769 case MSR_KVM_POLL_CONTROL:
3770 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3773 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3775 case MSR_IA32_P5_MC_ADDR:
3776 case MSR_IA32_P5_MC_TYPE:
3777 case MSR_IA32_MCG_CAP:
3778 case MSR_IA32_MCG_CTL:
3779 case MSR_IA32_MCG_STATUS:
3780 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3781 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3782 msr_info->host_initiated);
3784 if (!msr_info->host_initiated &&
3785 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3787 msr_info->data = vcpu->arch.ia32_xss;
3789 case MSR_K7_CLK_CTL:
3791 * Provide expected ramp-up count for K7. All other
3792 * are set to zero, indicating minimum divisors for
3795 * This prevents guest kernels on AMD host with CPU
3796 * type 6, model 8 and higher from exploding due to
3797 * the rdmsr failing.
3799 msr_info->data = 0x20000000;
3801 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3802 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3803 case HV_X64_MSR_SYNDBG_OPTIONS:
3804 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3805 case HV_X64_MSR_CRASH_CTL:
3806 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3807 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3808 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3809 case HV_X64_MSR_TSC_EMULATION_STATUS:
3810 return kvm_hv_get_msr_common(vcpu,
3811 msr_info->index, &msr_info->data,
3812 msr_info->host_initiated);
3813 case MSR_IA32_BBL_CR_CTL3:
3814 /* This legacy MSR exists but isn't fully documented in current
3815 * silicon. It is however accessed by winxp in very narrow
3816 * scenarios where it sets bit #19, itself documented as
3817 * a "reserved" bit. Best effort attempt to source coherent
3818 * read data here should the balance of the register be
3819 * interpreted by the guest:
3821 * L2 cache control register 3: 64GB range, 256KB size,
3822 * enabled, latency 0x1, configured
3824 msr_info->data = 0xbe702111;
3826 case MSR_AMD64_OSVW_ID_LENGTH:
3827 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3829 msr_info->data = vcpu->arch.osvw.length;
3831 case MSR_AMD64_OSVW_STATUS:
3832 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3834 msr_info->data = vcpu->arch.osvw.status;
3836 case MSR_PLATFORM_INFO:
3837 if (!msr_info->host_initiated &&
3838 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3840 msr_info->data = vcpu->arch.msr_platform_info;
3842 case MSR_MISC_FEATURES_ENABLES:
3843 msr_info->data = vcpu->arch.msr_misc_features_enables;
3846 msr_info->data = vcpu->arch.msr_hwcr;
3849 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3850 return kvm_pmu_get_msr(vcpu, msr_info);
3851 return KVM_MSR_RET_INVALID;
3855 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3858 * Read or write a bunch of msrs. All parameters are kernel addresses.
3860 * @return number of msrs set successfully.
3862 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3863 struct kvm_msr_entry *entries,
3864 int (*do_msr)(struct kvm_vcpu *vcpu,
3865 unsigned index, u64 *data))
3869 for (i = 0; i < msrs->nmsrs; ++i)
3870 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3877 * Read or write a bunch of msrs. Parameters are user addresses.
3879 * @return number of msrs set successfully.
3881 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3882 int (*do_msr)(struct kvm_vcpu *vcpu,
3883 unsigned index, u64 *data),
3886 struct kvm_msrs msrs;
3887 struct kvm_msr_entry *entries;
3892 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3896 if (msrs.nmsrs >= MAX_IO_MSRS)
3899 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3900 entries = memdup_user(user_msrs->entries, size);
3901 if (IS_ERR(entries)) {
3902 r = PTR_ERR(entries);
3906 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3911 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3922 static inline bool kvm_can_mwait_in_guest(void)
3924 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3925 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3926 boot_cpu_has(X86_FEATURE_ARAT);
3929 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3930 struct kvm_cpuid2 __user *cpuid_arg)
3932 struct kvm_cpuid2 cpuid;
3936 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3939 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3944 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3950 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3955 case KVM_CAP_IRQCHIP:
3957 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3958 case KVM_CAP_SET_TSS_ADDR:
3959 case KVM_CAP_EXT_CPUID:
3960 case KVM_CAP_EXT_EMUL_CPUID:
3961 case KVM_CAP_CLOCKSOURCE:
3963 case KVM_CAP_NOP_IO_DELAY:
3964 case KVM_CAP_MP_STATE:
3965 case KVM_CAP_SYNC_MMU:
3966 case KVM_CAP_USER_NMI:
3967 case KVM_CAP_REINJECT_CONTROL:
3968 case KVM_CAP_IRQ_INJECT_STATUS:
3969 case KVM_CAP_IOEVENTFD:
3970 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3972 case KVM_CAP_PIT_STATE2:
3973 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3974 case KVM_CAP_VCPU_EVENTS:
3975 case KVM_CAP_HYPERV:
3976 case KVM_CAP_HYPERV_VAPIC:
3977 case KVM_CAP_HYPERV_SPIN:
3978 case KVM_CAP_HYPERV_SYNIC:
3979 case KVM_CAP_HYPERV_SYNIC2:
3980 case KVM_CAP_HYPERV_VP_INDEX:
3981 case KVM_CAP_HYPERV_EVENTFD:
3982 case KVM_CAP_HYPERV_TLBFLUSH:
3983 case KVM_CAP_HYPERV_SEND_IPI:
3984 case KVM_CAP_HYPERV_CPUID:
3985 case KVM_CAP_HYPERV_ENFORCE_CPUID:
3986 case KVM_CAP_SYS_HYPERV_CPUID:
3987 case KVM_CAP_PCI_SEGMENT:
3988 case KVM_CAP_DEBUGREGS:
3989 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3991 case KVM_CAP_ASYNC_PF:
3992 case KVM_CAP_ASYNC_PF_INT:
3993 case KVM_CAP_GET_TSC_KHZ:
3994 case KVM_CAP_KVMCLOCK_CTRL:
3995 case KVM_CAP_READONLY_MEM:
3996 case KVM_CAP_HYPERV_TIME:
3997 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3998 case KVM_CAP_TSC_DEADLINE_TIMER:
3999 case KVM_CAP_DISABLE_QUIRKS:
4000 case KVM_CAP_SET_BOOT_CPU_ID:
4001 case KVM_CAP_SPLIT_IRQCHIP:
4002 case KVM_CAP_IMMEDIATE_EXIT:
4003 case KVM_CAP_PMU_EVENT_FILTER:
4004 case KVM_CAP_GET_MSR_FEATURES:
4005 case KVM_CAP_MSR_PLATFORM_INFO:
4006 case KVM_CAP_EXCEPTION_PAYLOAD:
4007 case KVM_CAP_SET_GUEST_DEBUG:
4008 case KVM_CAP_LAST_CPU:
4009 case KVM_CAP_X86_USER_SPACE_MSR:
4010 case KVM_CAP_X86_MSR_FILTER:
4011 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4012 #ifdef CONFIG_X86_SGX_KVM
4013 case KVM_CAP_SGX_ATTRIBUTE:
4015 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4016 case KVM_CAP_SREGS2:
4017 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4020 case KVM_CAP_EXIT_HYPERCALL:
4021 r = KVM_EXIT_HYPERCALL_VALID_MASK;
4023 case KVM_CAP_SET_GUEST_DEBUG2:
4024 return KVM_GUESTDBG_VALID_MASK;
4025 #ifdef CONFIG_KVM_XEN
4026 case KVM_CAP_XEN_HVM:
4027 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4028 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4029 KVM_XEN_HVM_CONFIG_SHARED_INFO;
4030 if (sched_info_on())
4031 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4034 case KVM_CAP_SYNC_REGS:
4035 r = KVM_SYNC_X86_VALID_FIELDS;
4037 case KVM_CAP_ADJUST_CLOCK:
4038 r = KVM_CLOCK_TSC_STABLE;
4040 case KVM_CAP_X86_DISABLE_EXITS:
4041 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4042 KVM_X86_DISABLE_EXITS_CSTATE;
4043 if(kvm_can_mwait_in_guest())
4044 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4046 case KVM_CAP_X86_SMM:
4047 /* SMBASE is usually relocated above 1M on modern chipsets,
4048 * and SMM handlers might indeed rely on 4G segment limits,
4049 * so do not report SMM to be available if real mode is
4050 * emulated via vm86 mode. Still, do not go to great lengths
4051 * to avoid userspace's usage of the feature, because it is a
4052 * fringe case that is not enabled except via specific settings
4053 * of the module parameters.
4055 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4058 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4060 case KVM_CAP_NR_VCPUS:
4061 r = KVM_SOFT_MAX_VCPUS;
4063 case KVM_CAP_MAX_VCPUS:
4066 case KVM_CAP_MAX_VCPU_ID:
4067 r = KVM_MAX_VCPU_ID;
4069 case KVM_CAP_PV_MMU: /* obsolete */
4073 r = KVM_MAX_MCE_BANKS;
4076 r = boot_cpu_has(X86_FEATURE_XSAVE);
4078 case KVM_CAP_TSC_CONTROL:
4079 r = kvm_has_tsc_control;
4081 case KVM_CAP_X2APIC_API:
4082 r = KVM_X2APIC_API_VALID_FLAGS;
4084 case KVM_CAP_NESTED_STATE:
4085 r = kvm_x86_ops.nested_ops->get_state ?
4086 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4088 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4089 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4091 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4092 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4094 case KVM_CAP_SMALLER_MAXPHYADDR:
4095 r = (int) allow_smaller_maxphyaddr;
4097 case KVM_CAP_STEAL_TIME:
4098 r = sched_info_on();
4100 case KVM_CAP_X86_BUS_LOCK_EXIT:
4101 if (kvm_has_bus_lock_exit)
4102 r = KVM_BUS_LOCK_DETECTION_OFF |
4103 KVM_BUS_LOCK_DETECTION_EXIT;
4114 long kvm_arch_dev_ioctl(struct file *filp,
4115 unsigned int ioctl, unsigned long arg)
4117 void __user *argp = (void __user *)arg;
4121 case KVM_GET_MSR_INDEX_LIST: {
4122 struct kvm_msr_list __user *user_msr_list = argp;
4123 struct kvm_msr_list msr_list;
4127 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4130 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4131 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4134 if (n < msr_list.nmsrs)
4137 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4138 num_msrs_to_save * sizeof(u32)))
4140 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4142 num_emulated_msrs * sizeof(u32)))
4147 case KVM_GET_SUPPORTED_CPUID:
4148 case KVM_GET_EMULATED_CPUID: {
4149 struct kvm_cpuid2 __user *cpuid_arg = argp;
4150 struct kvm_cpuid2 cpuid;
4153 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4156 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4162 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4167 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4169 if (copy_to_user(argp, &kvm_mce_cap_supported,
4170 sizeof(kvm_mce_cap_supported)))
4174 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4175 struct kvm_msr_list __user *user_msr_list = argp;
4176 struct kvm_msr_list msr_list;
4180 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4183 msr_list.nmsrs = num_msr_based_features;
4184 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4187 if (n < msr_list.nmsrs)
4190 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4191 num_msr_based_features * sizeof(u32)))
4197 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4199 case KVM_GET_SUPPORTED_HV_CPUID:
4200 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4210 static void wbinvd_ipi(void *garbage)
4215 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4217 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4220 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4222 /* Address WBINVD may be executed by guest */
4223 if (need_emulate_wbinvd(vcpu)) {
4224 if (static_call(kvm_x86_has_wbinvd_exit)())
4225 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4226 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4227 smp_call_function_single(vcpu->cpu,
4228 wbinvd_ipi, NULL, 1);
4231 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4233 /* Save host pkru register if supported */
4234 vcpu->arch.host_pkru = read_pkru();
4236 /* Apply any externally detected TSC adjustments (due to suspend) */
4237 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4238 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4239 vcpu->arch.tsc_offset_adjustment = 0;
4240 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4243 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4244 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4245 rdtsc() - vcpu->arch.last_host_tsc;
4247 mark_tsc_unstable("KVM discovered backwards TSC");
4249 if (kvm_check_tsc_unstable()) {
4250 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4251 vcpu->arch.last_guest_tsc);
4252 kvm_vcpu_write_tsc_offset(vcpu, offset);
4253 vcpu->arch.tsc_catchup = 1;
4256 if (kvm_lapic_hv_timer_in_use(vcpu))
4257 kvm_lapic_restart_hv_timer(vcpu);
4260 * On a host with synchronized TSC, there is no need to update
4261 * kvmclock on vcpu->cpu migration
4263 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4264 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4265 if (vcpu->cpu != cpu)
4266 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4270 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4273 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4275 struct kvm_host_map map;
4276 struct kvm_steal_time *st;
4278 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4281 if (vcpu->arch.st.preempted)
4284 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4285 &vcpu->arch.st.cache, true))
4289 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4291 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4293 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4296 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4300 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4301 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4304 * Take the srcu lock as memslots will be accessed to check the gfn
4305 * cache generation against the memslots generation.
4307 idx = srcu_read_lock(&vcpu->kvm->srcu);
4308 if (kvm_xen_msr_enabled(vcpu->kvm))
4309 kvm_xen_runstate_set_preempted(vcpu);
4311 kvm_steal_time_set_preempted(vcpu);
4312 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4314 static_call(kvm_x86_vcpu_put)(vcpu);
4315 vcpu->arch.last_host_tsc = rdtsc();
4318 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4319 struct kvm_lapic_state *s)
4321 if (vcpu->arch.apicv_active)
4322 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4324 return kvm_apic_get_state(vcpu, s);
4327 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4328 struct kvm_lapic_state *s)
4332 r = kvm_apic_set_state(vcpu, s);
4335 update_cr8_intercept(vcpu);
4340 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4343 * We can accept userspace's request for interrupt injection
4344 * as long as we have a place to store the interrupt number.
4345 * The actual injection will happen when the CPU is able to
4346 * deliver the interrupt.
4348 if (kvm_cpu_has_extint(vcpu))
4351 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4352 return (!lapic_in_kernel(vcpu) ||
4353 kvm_apic_accept_pic_intr(vcpu));
4356 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4359 * Do not cause an interrupt window exit if an exception
4360 * is pending or an event needs reinjection; userspace
4361 * might want to inject the interrupt manually using KVM_SET_REGS
4362 * or KVM_SET_SREGS. For that to work, we must be at an
4363 * instruction boundary and with no events half-injected.
4365 return (kvm_arch_interrupt_allowed(vcpu) &&
4366 kvm_cpu_accept_dm_intr(vcpu) &&
4367 !kvm_event_needs_reinjection(vcpu) &&
4368 !vcpu->arch.exception.pending);
4371 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4372 struct kvm_interrupt *irq)
4374 if (irq->irq >= KVM_NR_INTERRUPTS)
4377 if (!irqchip_in_kernel(vcpu->kvm)) {
4378 kvm_queue_interrupt(vcpu, irq->irq, false);
4379 kvm_make_request(KVM_REQ_EVENT, vcpu);
4384 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4385 * fail for in-kernel 8259.
4387 if (pic_in_kernel(vcpu->kvm))
4390 if (vcpu->arch.pending_external_vector != -1)
4393 vcpu->arch.pending_external_vector = irq->irq;
4394 kvm_make_request(KVM_REQ_EVENT, vcpu);
4398 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4400 kvm_inject_nmi(vcpu);
4405 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4407 kvm_make_request(KVM_REQ_SMI, vcpu);
4412 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4413 struct kvm_tpr_access_ctl *tac)
4417 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4421 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4425 unsigned bank_num = mcg_cap & 0xff, bank;
4428 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4430 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4433 vcpu->arch.mcg_cap = mcg_cap;
4434 /* Init IA32_MCG_CTL to all 1s */
4435 if (mcg_cap & MCG_CTL_P)
4436 vcpu->arch.mcg_ctl = ~(u64)0;
4437 /* Init IA32_MCi_CTL to all 1s */
4438 for (bank = 0; bank < bank_num; bank++)
4439 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4441 static_call(kvm_x86_setup_mce)(vcpu);
4446 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4447 struct kvm_x86_mce *mce)
4449 u64 mcg_cap = vcpu->arch.mcg_cap;
4450 unsigned bank_num = mcg_cap & 0xff;
4451 u64 *banks = vcpu->arch.mce_banks;
4453 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4456 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4457 * reporting is disabled
4459 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4460 vcpu->arch.mcg_ctl != ~(u64)0)
4462 banks += 4 * mce->bank;
4464 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4465 * reporting is disabled for the bank
4467 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4469 if (mce->status & MCI_STATUS_UC) {
4470 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4471 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4472 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4475 if (banks[1] & MCI_STATUS_VAL)
4476 mce->status |= MCI_STATUS_OVER;
4477 banks[2] = mce->addr;
4478 banks[3] = mce->misc;
4479 vcpu->arch.mcg_status = mce->mcg_status;
4480 banks[1] = mce->status;
4481 kvm_queue_exception(vcpu, MC_VECTOR);
4482 } else if (!(banks[1] & MCI_STATUS_VAL)
4483 || !(banks[1] & MCI_STATUS_UC)) {
4484 if (banks[1] & MCI_STATUS_VAL)
4485 mce->status |= MCI_STATUS_OVER;
4486 banks[2] = mce->addr;
4487 banks[3] = mce->misc;
4488 banks[1] = mce->status;
4490 banks[1] |= MCI_STATUS_OVER;
4494 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4495 struct kvm_vcpu_events *events)
4499 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4503 * In guest mode, payload delivery should be deferred,
4504 * so that the L1 hypervisor can intercept #PF before
4505 * CR2 is modified (or intercept #DB before DR6 is
4506 * modified under nVMX). Unless the per-VM capability,
4507 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4508 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4509 * opportunistically defer the exception payload, deliver it if the
4510 * capability hasn't been requested before processing a
4511 * KVM_GET_VCPU_EVENTS.
4513 if (!vcpu->kvm->arch.exception_payload_enabled &&
4514 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4515 kvm_deliver_exception_payload(vcpu);
4518 * The API doesn't provide the instruction length for software
4519 * exceptions, so don't report them. As long as the guest RIP
4520 * isn't advanced, we should expect to encounter the exception
4523 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4524 events->exception.injected = 0;
4525 events->exception.pending = 0;
4527 events->exception.injected = vcpu->arch.exception.injected;
4528 events->exception.pending = vcpu->arch.exception.pending;
4530 * For ABI compatibility, deliberately conflate
4531 * pending and injected exceptions when
4532 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4534 if (!vcpu->kvm->arch.exception_payload_enabled)
4535 events->exception.injected |=
4536 vcpu->arch.exception.pending;
4538 events->exception.nr = vcpu->arch.exception.nr;
4539 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4540 events->exception.error_code = vcpu->arch.exception.error_code;
4541 events->exception_has_payload = vcpu->arch.exception.has_payload;
4542 events->exception_payload = vcpu->arch.exception.payload;
4544 events->interrupt.injected =
4545 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4546 events->interrupt.nr = vcpu->arch.interrupt.nr;
4547 events->interrupt.soft = 0;
4548 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4550 events->nmi.injected = vcpu->arch.nmi_injected;
4551 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4552 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4553 events->nmi.pad = 0;
4555 events->sipi_vector = 0; /* never valid when reporting to user space */
4557 events->smi.smm = is_smm(vcpu);
4558 events->smi.pending = vcpu->arch.smi_pending;
4559 events->smi.smm_inside_nmi =
4560 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4561 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4563 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4564 | KVM_VCPUEVENT_VALID_SHADOW
4565 | KVM_VCPUEVENT_VALID_SMM);
4566 if (vcpu->kvm->arch.exception_payload_enabled)
4567 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4569 memset(&events->reserved, 0, sizeof(events->reserved));
4572 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4574 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4575 struct kvm_vcpu_events *events)
4577 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4578 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4579 | KVM_VCPUEVENT_VALID_SHADOW
4580 | KVM_VCPUEVENT_VALID_SMM
4581 | KVM_VCPUEVENT_VALID_PAYLOAD))
4584 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4585 if (!vcpu->kvm->arch.exception_payload_enabled)
4587 if (events->exception.pending)
4588 events->exception.injected = 0;
4590 events->exception_has_payload = 0;
4592 events->exception.pending = 0;
4593 events->exception_has_payload = 0;
4596 if ((events->exception.injected || events->exception.pending) &&
4597 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4600 /* INITs are latched while in SMM */
4601 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4602 (events->smi.smm || events->smi.pending) &&
4603 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4607 vcpu->arch.exception.injected = events->exception.injected;
4608 vcpu->arch.exception.pending = events->exception.pending;
4609 vcpu->arch.exception.nr = events->exception.nr;
4610 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4611 vcpu->arch.exception.error_code = events->exception.error_code;
4612 vcpu->arch.exception.has_payload = events->exception_has_payload;
4613 vcpu->arch.exception.payload = events->exception_payload;
4615 vcpu->arch.interrupt.injected = events->interrupt.injected;
4616 vcpu->arch.interrupt.nr = events->interrupt.nr;
4617 vcpu->arch.interrupt.soft = events->interrupt.soft;
4618 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4619 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4620 events->interrupt.shadow);
4622 vcpu->arch.nmi_injected = events->nmi.injected;
4623 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4624 vcpu->arch.nmi_pending = events->nmi.pending;
4625 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4627 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4628 lapic_in_kernel(vcpu))
4629 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4631 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4632 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4633 kvm_smm_changed(vcpu, events->smi.smm);
4635 vcpu->arch.smi_pending = events->smi.pending;
4637 if (events->smi.smm) {
4638 if (events->smi.smm_inside_nmi)
4639 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4641 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4644 if (lapic_in_kernel(vcpu)) {
4645 if (events->smi.latched_init)
4646 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4648 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4652 kvm_make_request(KVM_REQ_EVENT, vcpu);
4657 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4658 struct kvm_debugregs *dbgregs)
4662 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4663 kvm_get_dr(vcpu, 6, &val);
4665 dbgregs->dr7 = vcpu->arch.dr7;
4667 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4670 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4671 struct kvm_debugregs *dbgregs)
4676 if (!kvm_dr6_valid(dbgregs->dr6))
4678 if (!kvm_dr7_valid(dbgregs->dr7))
4681 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4682 kvm_update_dr0123(vcpu);
4683 vcpu->arch.dr6 = dbgregs->dr6;
4684 vcpu->arch.dr7 = dbgregs->dr7;
4685 kvm_update_dr7(vcpu);
4690 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4692 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4694 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4695 u64 xstate_bv = xsave->header.xfeatures;
4699 * Copy legacy XSAVE area, to avoid complications with CPUID
4700 * leaves 0 and 1 in the loop below.
4702 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4705 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4706 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4709 * Copy each region from the possibly compacted offset to the
4710 * non-compacted offset.
4712 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4714 u32 size, offset, ecx, edx;
4715 u64 xfeature_mask = valid & -valid;
4716 int xfeature_nr = fls64(xfeature_mask) - 1;
4719 cpuid_count(XSTATE_CPUID, xfeature_nr,
4720 &size, &offset, &ecx, &edx);
4722 if (xfeature_nr == XFEATURE_PKRU) {
4723 memcpy(dest + offset, &vcpu->arch.pkru,
4724 sizeof(vcpu->arch.pkru));
4726 src = get_xsave_addr(xsave, xfeature_nr);
4728 memcpy(dest + offset, src, size);
4731 valid -= xfeature_mask;
4735 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4737 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4738 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4742 * Copy legacy XSAVE area, to avoid complications with CPUID
4743 * leaves 0 and 1 in the loop below.
4745 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4747 /* Set XSTATE_BV and possibly XCOMP_BV. */
4748 xsave->header.xfeatures = xstate_bv;
4749 if (boot_cpu_has(X86_FEATURE_XSAVES))
4750 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4753 * Copy each region from the non-compacted offset to the
4754 * possibly compacted offset.
4756 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4758 u32 size, offset, ecx, edx;
4759 u64 xfeature_mask = valid & -valid;
4760 int xfeature_nr = fls64(xfeature_mask) - 1;
4762 cpuid_count(XSTATE_CPUID, xfeature_nr,
4763 &size, &offset, &ecx, &edx);
4765 if (xfeature_nr == XFEATURE_PKRU) {
4766 memcpy(&vcpu->arch.pkru, src + offset,
4767 sizeof(vcpu->arch.pkru));
4769 void *dest = get_xsave_addr(xsave, xfeature_nr);
4772 memcpy(dest, src + offset, size);
4775 valid -= xfeature_mask;
4779 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4780 struct kvm_xsave *guest_xsave)
4782 if (!vcpu->arch.guest_fpu)
4785 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4786 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4787 fill_xsave((u8 *) guest_xsave->region, vcpu);
4789 memcpy(guest_xsave->region,
4790 &vcpu->arch.guest_fpu->state.fxsave,
4791 sizeof(struct fxregs_state));
4792 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4793 XFEATURE_MASK_FPSSE;
4797 #define XSAVE_MXCSR_OFFSET 24
4799 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4800 struct kvm_xsave *guest_xsave)
4805 if (!vcpu->arch.guest_fpu)
4808 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4809 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4811 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4813 * Here we allow setting states that are not present in
4814 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4815 * with old userspace.
4817 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4819 load_xsave(vcpu, (u8 *)guest_xsave->region);
4821 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4822 mxcsr & ~mxcsr_feature_mask)
4824 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4825 guest_xsave->region, sizeof(struct fxregs_state));
4830 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4831 struct kvm_xcrs *guest_xcrs)
4833 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4834 guest_xcrs->nr_xcrs = 0;
4838 guest_xcrs->nr_xcrs = 1;
4839 guest_xcrs->flags = 0;
4840 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4841 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4844 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4845 struct kvm_xcrs *guest_xcrs)
4849 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4852 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4855 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4856 /* Only support XCR0 currently */
4857 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4858 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4859 guest_xcrs->xcrs[i].value);
4868 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4869 * stopped by the hypervisor. This function will be called from the host only.
4870 * EINVAL is returned when the host attempts to set the flag for a guest that
4871 * does not support pv clocks.
4873 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4875 if (!vcpu->arch.pv_time_enabled)
4877 vcpu->arch.pvclock_set_guest_stopped_request = true;
4878 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4882 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4883 struct kvm_enable_cap *cap)
4886 uint16_t vmcs_version;
4887 void __user *user_ptr;
4893 case KVM_CAP_HYPERV_SYNIC2:
4898 case KVM_CAP_HYPERV_SYNIC:
4899 if (!irqchip_in_kernel(vcpu->kvm))
4901 return kvm_hv_activate_synic(vcpu, cap->cap ==
4902 KVM_CAP_HYPERV_SYNIC2);
4903 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4904 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4906 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4908 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4909 if (copy_to_user(user_ptr, &vmcs_version,
4910 sizeof(vmcs_version)))
4914 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4915 if (!kvm_x86_ops.enable_direct_tlbflush)
4918 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4920 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4921 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
4923 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4924 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4925 if (vcpu->arch.pv_cpuid.enforce)
4926 kvm_update_pv_runtime(vcpu);
4934 long kvm_arch_vcpu_ioctl(struct file *filp,
4935 unsigned int ioctl, unsigned long arg)
4937 struct kvm_vcpu *vcpu = filp->private_data;
4938 void __user *argp = (void __user *)arg;
4941 struct kvm_sregs2 *sregs2;
4942 struct kvm_lapic_state *lapic;
4943 struct kvm_xsave *xsave;
4944 struct kvm_xcrs *xcrs;
4952 case KVM_GET_LAPIC: {
4954 if (!lapic_in_kernel(vcpu))
4956 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4957 GFP_KERNEL_ACCOUNT);
4962 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4966 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4971 case KVM_SET_LAPIC: {
4973 if (!lapic_in_kernel(vcpu))
4975 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4976 if (IS_ERR(u.lapic)) {
4977 r = PTR_ERR(u.lapic);
4981 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4984 case KVM_INTERRUPT: {
4985 struct kvm_interrupt irq;
4988 if (copy_from_user(&irq, argp, sizeof(irq)))
4990 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4994 r = kvm_vcpu_ioctl_nmi(vcpu);
4998 r = kvm_vcpu_ioctl_smi(vcpu);
5001 case KVM_SET_CPUID: {
5002 struct kvm_cpuid __user *cpuid_arg = argp;
5003 struct kvm_cpuid cpuid;
5006 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5008 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5011 case KVM_SET_CPUID2: {
5012 struct kvm_cpuid2 __user *cpuid_arg = argp;
5013 struct kvm_cpuid2 cpuid;
5016 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5018 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5019 cpuid_arg->entries);
5022 case KVM_GET_CPUID2: {
5023 struct kvm_cpuid2 __user *cpuid_arg = argp;
5024 struct kvm_cpuid2 cpuid;
5027 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5029 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5030 cpuid_arg->entries);
5034 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5039 case KVM_GET_MSRS: {
5040 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5041 r = msr_io(vcpu, argp, do_get_msr, 1);
5042 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5045 case KVM_SET_MSRS: {
5046 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5047 r = msr_io(vcpu, argp, do_set_msr, 0);
5048 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5051 case KVM_TPR_ACCESS_REPORTING: {
5052 struct kvm_tpr_access_ctl tac;
5055 if (copy_from_user(&tac, argp, sizeof(tac)))
5057 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5061 if (copy_to_user(argp, &tac, sizeof(tac)))
5066 case KVM_SET_VAPIC_ADDR: {
5067 struct kvm_vapic_addr va;
5071 if (!lapic_in_kernel(vcpu))
5074 if (copy_from_user(&va, argp, sizeof(va)))
5076 idx = srcu_read_lock(&vcpu->kvm->srcu);
5077 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5078 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5081 case KVM_X86_SETUP_MCE: {
5085 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5087 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5090 case KVM_X86_SET_MCE: {
5091 struct kvm_x86_mce mce;
5094 if (copy_from_user(&mce, argp, sizeof(mce)))
5096 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5099 case KVM_GET_VCPU_EVENTS: {
5100 struct kvm_vcpu_events events;
5102 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5105 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5110 case KVM_SET_VCPU_EVENTS: {
5111 struct kvm_vcpu_events events;
5114 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5117 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5120 case KVM_GET_DEBUGREGS: {
5121 struct kvm_debugregs dbgregs;
5123 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5126 if (copy_to_user(argp, &dbgregs,
5127 sizeof(struct kvm_debugregs)))
5132 case KVM_SET_DEBUGREGS: {
5133 struct kvm_debugregs dbgregs;
5136 if (copy_from_user(&dbgregs, argp,
5137 sizeof(struct kvm_debugregs)))
5140 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5143 case KVM_GET_XSAVE: {
5144 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5149 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5152 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5157 case KVM_SET_XSAVE: {
5158 u.xsave = memdup_user(argp, sizeof(*u.xsave));
5159 if (IS_ERR(u.xsave)) {
5160 r = PTR_ERR(u.xsave);
5164 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5167 case KVM_GET_XCRS: {
5168 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5173 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5176 if (copy_to_user(argp, u.xcrs,
5177 sizeof(struct kvm_xcrs)))
5182 case KVM_SET_XCRS: {
5183 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5184 if (IS_ERR(u.xcrs)) {
5185 r = PTR_ERR(u.xcrs);
5189 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5192 case KVM_SET_TSC_KHZ: {
5196 user_tsc_khz = (u32)arg;
5198 if (kvm_has_tsc_control &&
5199 user_tsc_khz >= kvm_max_guest_tsc_khz)
5202 if (user_tsc_khz == 0)
5203 user_tsc_khz = tsc_khz;
5205 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5210 case KVM_GET_TSC_KHZ: {
5211 r = vcpu->arch.virtual_tsc_khz;
5214 case KVM_KVMCLOCK_CTRL: {
5215 r = kvm_set_guest_paused(vcpu);
5218 case KVM_ENABLE_CAP: {
5219 struct kvm_enable_cap cap;
5222 if (copy_from_user(&cap, argp, sizeof(cap)))
5224 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5227 case KVM_GET_NESTED_STATE: {
5228 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5232 if (!kvm_x86_ops.nested_ops->get_state)
5235 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5237 if (get_user(user_data_size, &user_kvm_nested_state->size))
5240 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5245 if (r > user_data_size) {
5246 if (put_user(r, &user_kvm_nested_state->size))
5256 case KVM_SET_NESTED_STATE: {
5257 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5258 struct kvm_nested_state kvm_state;
5262 if (!kvm_x86_ops.nested_ops->set_state)
5266 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5270 if (kvm_state.size < sizeof(kvm_state))
5273 if (kvm_state.flags &
5274 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5275 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5276 | KVM_STATE_NESTED_GIF_SET))
5279 /* nested_run_pending implies guest_mode. */
5280 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5281 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5284 idx = srcu_read_lock(&vcpu->kvm->srcu);
5285 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5286 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5289 case KVM_GET_SUPPORTED_HV_CPUID:
5290 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5292 #ifdef CONFIG_KVM_XEN
5293 case KVM_XEN_VCPU_GET_ATTR: {
5294 struct kvm_xen_vcpu_attr xva;
5297 if (copy_from_user(&xva, argp, sizeof(xva)))
5299 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5300 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5304 case KVM_XEN_VCPU_SET_ATTR: {
5305 struct kvm_xen_vcpu_attr xva;
5308 if (copy_from_user(&xva, argp, sizeof(xva)))
5310 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5314 case KVM_GET_SREGS2: {
5315 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5319 __get_sregs2(vcpu, u.sregs2);
5321 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5326 case KVM_SET_SREGS2: {
5327 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5328 if (IS_ERR(u.sregs2)) {
5329 r = PTR_ERR(u.sregs2);
5333 r = __set_sregs2(vcpu, u.sregs2);
5346 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5348 return VM_FAULT_SIGBUS;
5351 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5355 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5357 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5361 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5364 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5367 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5368 unsigned long kvm_nr_mmu_pages)
5370 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5373 mutex_lock(&kvm->slots_lock);
5375 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5376 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5378 mutex_unlock(&kvm->slots_lock);
5382 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5384 return kvm->arch.n_max_mmu_pages;
5387 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5389 struct kvm_pic *pic = kvm->arch.vpic;
5393 switch (chip->chip_id) {
5394 case KVM_IRQCHIP_PIC_MASTER:
5395 memcpy(&chip->chip.pic, &pic->pics[0],
5396 sizeof(struct kvm_pic_state));
5398 case KVM_IRQCHIP_PIC_SLAVE:
5399 memcpy(&chip->chip.pic, &pic->pics[1],
5400 sizeof(struct kvm_pic_state));
5402 case KVM_IRQCHIP_IOAPIC:
5403 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5412 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5414 struct kvm_pic *pic = kvm->arch.vpic;
5418 switch (chip->chip_id) {
5419 case KVM_IRQCHIP_PIC_MASTER:
5420 spin_lock(&pic->lock);
5421 memcpy(&pic->pics[0], &chip->chip.pic,
5422 sizeof(struct kvm_pic_state));
5423 spin_unlock(&pic->lock);
5425 case KVM_IRQCHIP_PIC_SLAVE:
5426 spin_lock(&pic->lock);
5427 memcpy(&pic->pics[1], &chip->chip.pic,
5428 sizeof(struct kvm_pic_state));
5429 spin_unlock(&pic->lock);
5431 case KVM_IRQCHIP_IOAPIC:
5432 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5438 kvm_pic_update_irq(pic);
5442 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5444 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5446 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5448 mutex_lock(&kps->lock);
5449 memcpy(ps, &kps->channels, sizeof(*ps));
5450 mutex_unlock(&kps->lock);
5454 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5457 struct kvm_pit *pit = kvm->arch.vpit;
5459 mutex_lock(&pit->pit_state.lock);
5460 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5461 for (i = 0; i < 3; i++)
5462 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5463 mutex_unlock(&pit->pit_state.lock);
5467 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5469 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5470 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5471 sizeof(ps->channels));
5472 ps->flags = kvm->arch.vpit->pit_state.flags;
5473 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5474 memset(&ps->reserved, 0, sizeof(ps->reserved));
5478 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5482 u32 prev_legacy, cur_legacy;
5483 struct kvm_pit *pit = kvm->arch.vpit;
5485 mutex_lock(&pit->pit_state.lock);
5486 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5487 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5488 if (!prev_legacy && cur_legacy)
5490 memcpy(&pit->pit_state.channels, &ps->channels,
5491 sizeof(pit->pit_state.channels));
5492 pit->pit_state.flags = ps->flags;
5493 for (i = 0; i < 3; i++)
5494 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5496 mutex_unlock(&pit->pit_state.lock);
5500 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5501 struct kvm_reinject_control *control)
5503 struct kvm_pit *pit = kvm->arch.vpit;
5505 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5506 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5507 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5509 mutex_lock(&pit->pit_state.lock);
5510 kvm_pit_set_reinject(pit, control->pit_reinject);
5511 mutex_unlock(&pit->pit_state.lock);
5516 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5520 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5521 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5522 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5525 struct kvm_vcpu *vcpu;
5528 kvm_for_each_vcpu(i, vcpu, kvm)
5529 kvm_vcpu_kick(vcpu);
5532 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5535 if (!irqchip_in_kernel(kvm))
5538 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5539 irq_event->irq, irq_event->level,
5544 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5545 struct kvm_enable_cap *cap)
5553 case KVM_CAP_DISABLE_QUIRKS:
5554 kvm->arch.disabled_quirks = cap->args[0];
5557 case KVM_CAP_SPLIT_IRQCHIP: {
5558 mutex_lock(&kvm->lock);
5560 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5561 goto split_irqchip_unlock;
5563 if (irqchip_in_kernel(kvm))
5564 goto split_irqchip_unlock;
5565 if (kvm->created_vcpus)
5566 goto split_irqchip_unlock;
5567 r = kvm_setup_empty_irq_routing(kvm);
5569 goto split_irqchip_unlock;
5570 /* Pairs with irqchip_in_kernel. */
5572 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5573 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5575 split_irqchip_unlock:
5576 mutex_unlock(&kvm->lock);
5579 case KVM_CAP_X2APIC_API:
5581 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5584 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5585 kvm->arch.x2apic_format = true;
5586 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5587 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5591 case KVM_CAP_X86_DISABLE_EXITS:
5593 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5596 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5597 kvm_can_mwait_in_guest())
5598 kvm->arch.mwait_in_guest = true;
5599 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5600 kvm->arch.hlt_in_guest = true;
5601 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5602 kvm->arch.pause_in_guest = true;
5603 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5604 kvm->arch.cstate_in_guest = true;
5607 case KVM_CAP_MSR_PLATFORM_INFO:
5608 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5611 case KVM_CAP_EXCEPTION_PAYLOAD:
5612 kvm->arch.exception_payload_enabled = cap->args[0];
5615 case KVM_CAP_X86_USER_SPACE_MSR:
5616 kvm->arch.user_space_msr_mask = cap->args[0];
5619 case KVM_CAP_X86_BUS_LOCK_EXIT:
5621 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5624 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5625 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5628 if (kvm_has_bus_lock_exit &&
5629 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5630 kvm->arch.bus_lock_detection_enabled = true;
5633 #ifdef CONFIG_X86_SGX_KVM
5634 case KVM_CAP_SGX_ATTRIBUTE: {
5635 unsigned long allowed_attributes = 0;
5637 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5641 /* KVM only supports the PROVISIONKEY privileged attribute. */
5642 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5643 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5644 kvm->arch.sgx_provisioning_allowed = true;
5650 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5652 if (kvm_x86_ops.vm_copy_enc_context_from)
5653 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5655 case KVM_CAP_EXIT_HYPERCALL:
5656 if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5660 kvm->arch.hypercall_exit_enabled = cap->args[0];
5663 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5665 if (cap->args[0] & ~1)
5667 kvm->arch.exit_on_emulation_error = cap->args[0];
5677 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5679 struct kvm_x86_msr_filter *msr_filter;
5681 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5685 msr_filter->default_allow = default_allow;
5689 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5696 for (i = 0; i < msr_filter->count; i++)
5697 kfree(msr_filter->ranges[i].bitmap);
5702 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5703 struct kvm_msr_filter_range *user_range)
5705 unsigned long *bitmap = NULL;
5708 if (!user_range->nmsrs)
5711 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5714 if (!user_range->flags)
5717 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5718 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5721 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5723 return PTR_ERR(bitmap);
5725 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5726 .flags = user_range->flags,
5727 .base = user_range->base,
5728 .nmsrs = user_range->nmsrs,
5732 msr_filter->count++;
5736 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5738 struct kvm_msr_filter __user *user_msr_filter = argp;
5739 struct kvm_x86_msr_filter *new_filter, *old_filter;
5740 struct kvm_msr_filter filter;
5746 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5749 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5750 empty &= !filter.ranges[i].nmsrs;
5752 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5753 if (empty && !default_allow)
5756 new_filter = kvm_alloc_msr_filter(default_allow);
5760 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5761 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5763 kvm_free_msr_filter(new_filter);
5768 mutex_lock(&kvm->lock);
5770 /* The per-VM filter is protected by kvm->lock... */
5771 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5773 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5774 synchronize_srcu(&kvm->srcu);
5776 kvm_free_msr_filter(old_filter);
5778 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5779 mutex_unlock(&kvm->lock);
5784 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5785 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5787 struct kvm_vcpu *vcpu;
5790 mutex_lock(&kvm->lock);
5791 kvm_for_each_vcpu(i, vcpu, kvm) {
5792 if (!vcpu->arch.pv_time_enabled)
5795 ret = kvm_set_guest_paused(vcpu);
5797 kvm_err("Failed to pause guest VCPU%d: %d\n",
5798 vcpu->vcpu_id, ret);
5802 mutex_unlock(&kvm->lock);
5804 return ret ? NOTIFY_BAD : NOTIFY_DONE;
5807 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5810 case PM_HIBERNATION_PREPARE:
5811 case PM_SUSPEND_PREPARE:
5812 return kvm_arch_suspend_notifier(kvm);
5817 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5819 long kvm_arch_vm_ioctl(struct file *filp,
5820 unsigned int ioctl, unsigned long arg)
5822 struct kvm *kvm = filp->private_data;
5823 void __user *argp = (void __user *)arg;
5826 * This union makes it completely explicit to gcc-3.x
5827 * that these two variables' stack usage should be
5828 * combined, not added together.
5831 struct kvm_pit_state ps;
5832 struct kvm_pit_state2 ps2;
5833 struct kvm_pit_config pit_config;
5837 case KVM_SET_TSS_ADDR:
5838 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5840 case KVM_SET_IDENTITY_MAP_ADDR: {
5843 mutex_lock(&kvm->lock);
5845 if (kvm->created_vcpus)
5846 goto set_identity_unlock;
5848 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5849 goto set_identity_unlock;
5850 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5851 set_identity_unlock:
5852 mutex_unlock(&kvm->lock);
5855 case KVM_SET_NR_MMU_PAGES:
5856 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5858 case KVM_GET_NR_MMU_PAGES:
5859 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5861 case KVM_CREATE_IRQCHIP: {
5862 mutex_lock(&kvm->lock);
5865 if (irqchip_in_kernel(kvm))
5866 goto create_irqchip_unlock;
5869 if (kvm->created_vcpus)
5870 goto create_irqchip_unlock;
5872 r = kvm_pic_init(kvm);
5874 goto create_irqchip_unlock;
5876 r = kvm_ioapic_init(kvm);
5878 kvm_pic_destroy(kvm);
5879 goto create_irqchip_unlock;
5882 r = kvm_setup_default_irq_routing(kvm);
5884 kvm_ioapic_destroy(kvm);
5885 kvm_pic_destroy(kvm);
5886 goto create_irqchip_unlock;
5888 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5890 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5891 create_irqchip_unlock:
5892 mutex_unlock(&kvm->lock);
5895 case KVM_CREATE_PIT:
5896 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5898 case KVM_CREATE_PIT2:
5900 if (copy_from_user(&u.pit_config, argp,
5901 sizeof(struct kvm_pit_config)))
5904 mutex_lock(&kvm->lock);
5907 goto create_pit_unlock;
5909 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5913 mutex_unlock(&kvm->lock);
5915 case KVM_GET_IRQCHIP: {
5916 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5917 struct kvm_irqchip *chip;
5919 chip = memdup_user(argp, sizeof(*chip));
5926 if (!irqchip_kernel(kvm))
5927 goto get_irqchip_out;
5928 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5930 goto get_irqchip_out;
5932 if (copy_to_user(argp, chip, sizeof(*chip)))
5933 goto get_irqchip_out;
5939 case KVM_SET_IRQCHIP: {
5940 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5941 struct kvm_irqchip *chip;
5943 chip = memdup_user(argp, sizeof(*chip));
5950 if (!irqchip_kernel(kvm))
5951 goto set_irqchip_out;
5952 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5959 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5962 if (!kvm->arch.vpit)
5964 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5968 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5975 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5977 mutex_lock(&kvm->lock);
5979 if (!kvm->arch.vpit)
5981 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5983 mutex_unlock(&kvm->lock);
5986 case KVM_GET_PIT2: {
5988 if (!kvm->arch.vpit)
5990 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5994 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5999 case KVM_SET_PIT2: {
6001 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6003 mutex_lock(&kvm->lock);
6005 if (!kvm->arch.vpit)
6007 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6009 mutex_unlock(&kvm->lock);
6012 case KVM_REINJECT_CONTROL: {
6013 struct kvm_reinject_control control;
6015 if (copy_from_user(&control, argp, sizeof(control)))
6018 if (!kvm->arch.vpit)
6020 r = kvm_vm_ioctl_reinject(kvm, &control);
6023 case KVM_SET_BOOT_CPU_ID:
6025 mutex_lock(&kvm->lock);
6026 if (kvm->created_vcpus)
6029 kvm->arch.bsp_vcpu_id = arg;
6030 mutex_unlock(&kvm->lock);
6032 #ifdef CONFIG_KVM_XEN
6033 case KVM_XEN_HVM_CONFIG: {
6034 struct kvm_xen_hvm_config xhc;
6036 if (copy_from_user(&xhc, argp, sizeof(xhc)))
6038 r = kvm_xen_hvm_config(kvm, &xhc);
6041 case KVM_XEN_HVM_GET_ATTR: {
6042 struct kvm_xen_hvm_attr xha;
6045 if (copy_from_user(&xha, argp, sizeof(xha)))
6047 r = kvm_xen_hvm_get_attr(kvm, &xha);
6048 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6052 case KVM_XEN_HVM_SET_ATTR: {
6053 struct kvm_xen_hvm_attr xha;
6056 if (copy_from_user(&xha, argp, sizeof(xha)))
6058 r = kvm_xen_hvm_set_attr(kvm, &xha);
6062 case KVM_SET_CLOCK: {
6063 struct kvm_arch *ka = &kvm->arch;
6064 struct kvm_clock_data user_ns;
6068 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
6077 * TODO: userspace has to take care of races with VCPU_RUN, so
6078 * kvm_gen_update_masterclock() can be cut down to locked
6079 * pvclock_update_vm_gtod_copy().
6081 kvm_gen_update_masterclock(kvm);
6084 * This pairs with kvm_guest_time_update(): when masterclock is
6085 * in use, we use master_kernel_ns + kvmclock_offset to set
6086 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6087 * is slightly ahead) here we risk going negative on unsigned
6088 * 'system_time' when 'user_ns.clock' is very small.
6090 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
6091 if (kvm->arch.use_master_clock)
6092 now_ns = ka->master_kernel_ns;
6094 now_ns = get_kvmclock_base_ns();
6095 ka->kvmclock_offset = user_ns.clock - now_ns;
6096 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
6098 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
6101 case KVM_GET_CLOCK: {
6102 struct kvm_clock_data user_ns;
6105 now_ns = get_kvmclock_ns(kvm);
6106 user_ns.clock = now_ns;
6107 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
6108 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
6111 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
6116 case KVM_MEMORY_ENCRYPT_OP: {
6118 if (kvm_x86_ops.mem_enc_op)
6119 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6122 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6123 struct kvm_enc_region region;
6126 if (copy_from_user(®ion, argp, sizeof(region)))
6130 if (kvm_x86_ops.mem_enc_reg_region)
6131 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
6134 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6135 struct kvm_enc_region region;
6138 if (copy_from_user(®ion, argp, sizeof(region)))
6142 if (kvm_x86_ops.mem_enc_unreg_region)
6143 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
6146 case KVM_HYPERV_EVENTFD: {
6147 struct kvm_hyperv_eventfd hvevfd;
6150 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6152 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6155 case KVM_SET_PMU_EVENT_FILTER:
6156 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6158 case KVM_X86_SET_MSR_FILTER:
6159 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6168 static void kvm_init_msr_list(void)
6170 struct x86_pmu_capability x86_pmu;
6174 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6175 "Please update the fixed PMCs in msrs_to_saved_all[]");
6177 perf_get_x86_pmu_capability(&x86_pmu);
6179 num_msrs_to_save = 0;
6180 num_emulated_msrs = 0;
6181 num_msr_based_features = 0;
6183 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6184 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6188 * Even MSRs that are valid in the host may not be exposed
6189 * to the guests in some cases.
6191 switch (msrs_to_save_all[i]) {
6192 case MSR_IA32_BNDCFGS:
6193 if (!kvm_mpx_supported())
6197 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6198 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6201 case MSR_IA32_UMWAIT_CONTROL:
6202 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6205 case MSR_IA32_RTIT_CTL:
6206 case MSR_IA32_RTIT_STATUS:
6207 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6210 case MSR_IA32_RTIT_CR3_MATCH:
6211 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6212 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6215 case MSR_IA32_RTIT_OUTPUT_BASE:
6216 case MSR_IA32_RTIT_OUTPUT_MASK:
6217 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6218 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6219 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6222 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6223 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6224 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6225 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6228 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6229 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6230 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6233 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6234 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6235 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6242 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6245 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6246 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6249 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6252 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6253 struct kvm_msr_entry msr;
6255 msr.index = msr_based_features_all[i];
6256 if (kvm_get_msr_feature(&msr))
6259 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6263 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6271 if (!(lapic_in_kernel(vcpu) &&
6272 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6273 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6284 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6291 if (!(lapic_in_kernel(vcpu) &&
6292 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6294 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6296 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6306 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6307 struct kvm_segment *var, int seg)
6309 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6312 void kvm_get_segment(struct kvm_vcpu *vcpu,
6313 struct kvm_segment *var, int seg)
6315 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6318 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6319 struct x86_exception *exception)
6323 BUG_ON(!mmu_is_nested(vcpu));
6325 /* NPT walks are always user-walks */
6326 access |= PFERR_USER_MASK;
6327 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6332 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6333 struct x86_exception *exception)
6335 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6336 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6338 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6340 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6341 struct x86_exception *exception)
6343 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6344 access |= PFERR_FETCH_MASK;
6345 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6348 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6349 struct x86_exception *exception)
6351 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6352 access |= PFERR_WRITE_MASK;
6353 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6355 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6357 /* uses this to access any guest's mapped memory without checking CPL */
6358 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6359 struct x86_exception *exception)
6361 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6364 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6365 struct kvm_vcpu *vcpu, u32 access,
6366 struct x86_exception *exception)
6369 int r = X86EMUL_CONTINUE;
6372 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6374 unsigned offset = addr & (PAGE_SIZE-1);
6375 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6378 if (gpa == UNMAPPED_GVA)
6379 return X86EMUL_PROPAGATE_FAULT;
6380 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6383 r = X86EMUL_IO_NEEDED;
6395 /* used for instruction fetching */
6396 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6397 gva_t addr, void *val, unsigned int bytes,
6398 struct x86_exception *exception)
6400 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6401 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6405 /* Inline kvm_read_guest_virt_helper for speed. */
6406 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6408 if (unlikely(gpa == UNMAPPED_GVA))
6409 return X86EMUL_PROPAGATE_FAULT;
6411 offset = addr & (PAGE_SIZE-1);
6412 if (WARN_ON(offset + bytes > PAGE_SIZE))
6413 bytes = (unsigned)PAGE_SIZE - offset;
6414 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6416 if (unlikely(ret < 0))
6417 return X86EMUL_IO_NEEDED;
6419 return X86EMUL_CONTINUE;
6422 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6423 gva_t addr, void *val, unsigned int bytes,
6424 struct x86_exception *exception)
6426 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6429 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6430 * is returned, but our callers are not ready for that and they blindly
6431 * call kvm_inject_page_fault. Ensure that they at least do not leak
6432 * uninitialized kernel stack memory into cr2 and error code.
6434 memset(exception, 0, sizeof(*exception));
6435 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6438 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6440 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6441 gva_t addr, void *val, unsigned int bytes,
6442 struct x86_exception *exception, bool system)
6444 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6447 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6448 access |= PFERR_USER_MASK;
6450 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6453 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6454 unsigned long addr, void *val, unsigned int bytes)
6456 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6457 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6459 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6462 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6463 struct kvm_vcpu *vcpu, u32 access,
6464 struct x86_exception *exception)
6467 int r = X86EMUL_CONTINUE;
6470 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6473 unsigned offset = addr & (PAGE_SIZE-1);
6474 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6477 if (gpa == UNMAPPED_GVA)
6478 return X86EMUL_PROPAGATE_FAULT;
6479 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6481 r = X86EMUL_IO_NEEDED;
6493 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6494 unsigned int bytes, struct x86_exception *exception,
6497 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6498 u32 access = PFERR_WRITE_MASK;
6500 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6501 access |= PFERR_USER_MASK;
6503 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6507 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6508 unsigned int bytes, struct x86_exception *exception)
6510 /* kvm_write_guest_virt_system can pull in tons of pages. */
6511 vcpu->arch.l1tf_flush_l1d = true;
6513 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6514 PFERR_WRITE_MASK, exception);
6516 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6518 int handle_ud(struct kvm_vcpu *vcpu)
6520 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6521 int emul_type = EMULTYPE_TRAP_UD;
6522 char sig[5]; /* ud2; .ascii "kvm" */
6523 struct x86_exception e;
6525 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6528 if (force_emulation_prefix &&
6529 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6530 sig, sizeof(sig), &e) == 0 &&
6531 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6532 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6533 emul_type = EMULTYPE_TRAP_UD_FORCED;
6536 return kvm_emulate_instruction(vcpu, emul_type);
6538 EXPORT_SYMBOL_GPL(handle_ud);
6540 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6541 gpa_t gpa, bool write)
6543 /* For APIC access vmexit */
6544 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6547 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6548 trace_vcpu_match_mmio(gva, gpa, write, true);
6555 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6556 gpa_t *gpa, struct x86_exception *exception,
6559 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6560 | (write ? PFERR_WRITE_MASK : 0);
6563 * currently PKRU is only applied to ept enabled guest so
6564 * there is no pkey in EPT page table for L1 guest or EPT
6565 * shadow page table for L2 guest.
6567 if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6568 !permission_fault(vcpu, vcpu->arch.walk_mmu,
6569 vcpu->arch.mmio_access, 0, access))) {
6570 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6571 (gva & (PAGE_SIZE - 1));
6572 trace_vcpu_match_mmio(gva, *gpa, write, false);
6576 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6578 if (*gpa == UNMAPPED_GVA)
6581 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6584 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6585 const void *val, int bytes)
6589 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6592 kvm_page_track_write(vcpu, gpa, val, bytes);
6596 struct read_write_emulator_ops {
6597 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6599 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6600 void *val, int bytes);
6601 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6602 int bytes, void *val);
6603 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6604 void *val, int bytes);
6608 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6610 if (vcpu->mmio_read_completed) {
6611 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6612 vcpu->mmio_fragments[0].gpa, val);
6613 vcpu->mmio_read_completed = 0;
6620 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6621 void *val, int bytes)
6623 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6626 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6627 void *val, int bytes)
6629 return emulator_write_phys(vcpu, gpa, val, bytes);
6632 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6634 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6635 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6638 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6639 void *val, int bytes)
6641 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6642 return X86EMUL_IO_NEEDED;
6645 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6646 void *val, int bytes)
6648 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6650 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6651 return X86EMUL_CONTINUE;
6654 static const struct read_write_emulator_ops read_emultor = {
6655 .read_write_prepare = read_prepare,
6656 .read_write_emulate = read_emulate,
6657 .read_write_mmio = vcpu_mmio_read,
6658 .read_write_exit_mmio = read_exit_mmio,
6661 static const struct read_write_emulator_ops write_emultor = {
6662 .read_write_emulate = write_emulate,
6663 .read_write_mmio = write_mmio,
6664 .read_write_exit_mmio = write_exit_mmio,
6668 static int emulator_read_write_onepage(unsigned long addr, void *val,
6670 struct x86_exception *exception,
6671 struct kvm_vcpu *vcpu,
6672 const struct read_write_emulator_ops *ops)
6676 bool write = ops->write;
6677 struct kvm_mmio_fragment *frag;
6678 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6681 * If the exit was due to a NPF we may already have a GPA.
6682 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6683 * Note, this cannot be used on string operations since string
6684 * operation using rep will only have the initial GPA from the NPF
6687 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6688 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6689 gpa = ctxt->gpa_val;
6690 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6692 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6694 return X86EMUL_PROPAGATE_FAULT;
6697 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6698 return X86EMUL_CONTINUE;
6701 * Is this MMIO handled locally?
6703 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6704 if (handled == bytes)
6705 return X86EMUL_CONTINUE;
6711 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6712 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6716 return X86EMUL_CONTINUE;
6719 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6721 void *val, unsigned int bytes,
6722 struct x86_exception *exception,
6723 const struct read_write_emulator_ops *ops)
6725 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6729 if (ops->read_write_prepare &&
6730 ops->read_write_prepare(vcpu, val, bytes))
6731 return X86EMUL_CONTINUE;
6733 vcpu->mmio_nr_fragments = 0;
6735 /* Crossing a page boundary? */
6736 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6739 now = -addr & ~PAGE_MASK;
6740 rc = emulator_read_write_onepage(addr, val, now, exception,
6743 if (rc != X86EMUL_CONTINUE)
6746 if (ctxt->mode != X86EMUL_MODE_PROT64)
6752 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6754 if (rc != X86EMUL_CONTINUE)
6757 if (!vcpu->mmio_nr_fragments)
6760 gpa = vcpu->mmio_fragments[0].gpa;
6762 vcpu->mmio_needed = 1;
6763 vcpu->mmio_cur_fragment = 0;
6765 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6766 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6767 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6768 vcpu->run->mmio.phys_addr = gpa;
6770 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6773 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6777 struct x86_exception *exception)
6779 return emulator_read_write(ctxt, addr, val, bytes,
6780 exception, &read_emultor);
6783 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6787 struct x86_exception *exception)
6789 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6790 exception, &write_emultor);
6793 #define CMPXCHG_TYPE(t, ptr, old, new) \
6794 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6796 #ifdef CONFIG_X86_64
6797 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6799 # define CMPXCHG64(ptr, old, new) \
6800 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6803 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6808 struct x86_exception *exception)
6810 struct kvm_host_map map;
6811 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6817 /* guests cmpxchg8b have to be emulated atomically */
6818 if (bytes > 8 || (bytes & (bytes - 1)))
6821 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6823 if (gpa == UNMAPPED_GVA ||
6824 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6828 * Emulate the atomic as a straight write to avoid #AC if SLD is
6829 * enabled in the host and the access splits a cache line.
6831 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6832 page_line_mask = ~(cache_line_size() - 1);
6834 page_line_mask = PAGE_MASK;
6836 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6839 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6842 kaddr = map.hva + offset_in_page(gpa);
6846 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6849 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6852 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6855 exchanged = CMPXCHG64(kaddr, old, new);
6861 kvm_vcpu_unmap(vcpu, &map, true);
6864 return X86EMUL_CMPXCHG_FAILED;
6866 kvm_page_track_write(vcpu, gpa, new, bytes);
6868 return X86EMUL_CONTINUE;
6871 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6873 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6876 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6880 for (i = 0; i < vcpu->arch.pio.count; i++) {
6881 if (vcpu->arch.pio.in)
6882 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6883 vcpu->arch.pio.size, pd);
6885 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6886 vcpu->arch.pio.port, vcpu->arch.pio.size,
6890 pd += vcpu->arch.pio.size;
6895 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6896 unsigned short port, void *val,
6897 unsigned int count, bool in)
6899 vcpu->arch.pio.port = port;
6900 vcpu->arch.pio.in = in;
6901 vcpu->arch.pio.count = count;
6902 vcpu->arch.pio.size = size;
6904 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6905 vcpu->arch.pio.count = 0;
6909 vcpu->run->exit_reason = KVM_EXIT_IO;
6910 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6911 vcpu->run->io.size = size;
6912 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6913 vcpu->run->io.count = count;
6914 vcpu->run->io.port = port;
6919 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6920 unsigned short port, void *val, unsigned int count)
6924 if (vcpu->arch.pio.count)
6927 memset(vcpu->arch.pio_data, 0, size * count);
6929 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6932 memcpy(val, vcpu->arch.pio_data, size * count);
6933 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6934 vcpu->arch.pio.count = 0;
6941 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6942 int size, unsigned short port, void *val,
6945 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6949 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6950 unsigned short port, const void *val,
6953 memcpy(vcpu->arch.pio_data, val, size * count);
6954 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6955 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6958 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6959 int size, unsigned short port,
6960 const void *val, unsigned int count)
6962 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6965 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6967 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6970 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6972 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6975 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6977 if (!need_emulate_wbinvd(vcpu))
6978 return X86EMUL_CONTINUE;
6980 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6981 int cpu = get_cpu();
6983 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6984 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6985 wbinvd_ipi, NULL, 1);
6987 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6990 return X86EMUL_CONTINUE;
6993 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6995 kvm_emulate_wbinvd_noskip(vcpu);
6996 return kvm_skip_emulated_instruction(vcpu);
6998 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7002 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7004 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7007 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7008 unsigned long *dest)
7010 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7013 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7014 unsigned long value)
7017 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7020 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7022 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7025 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7027 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7028 unsigned long value;
7032 value = kvm_read_cr0(vcpu);
7035 value = vcpu->arch.cr2;
7038 value = kvm_read_cr3(vcpu);
7041 value = kvm_read_cr4(vcpu);
7044 value = kvm_get_cr8(vcpu);
7047 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7054 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7056 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7061 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7064 vcpu->arch.cr2 = val;
7067 res = kvm_set_cr3(vcpu, val);
7070 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7073 res = kvm_set_cr8(vcpu, val);
7076 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7083 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7085 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7088 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7090 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7093 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7095 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7098 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7100 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7103 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7105 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7108 static unsigned long emulator_get_cached_segment_base(
7109 struct x86_emulate_ctxt *ctxt, int seg)
7111 return get_segment_base(emul_to_vcpu(ctxt), seg);
7114 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7115 struct desc_struct *desc, u32 *base3,
7118 struct kvm_segment var;
7120 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7121 *selector = var.selector;
7124 memset(desc, 0, sizeof(*desc));
7132 set_desc_limit(desc, var.limit);
7133 set_desc_base(desc, (unsigned long)var.base);
7134 #ifdef CONFIG_X86_64
7136 *base3 = var.base >> 32;
7138 desc->type = var.type;
7140 desc->dpl = var.dpl;
7141 desc->p = var.present;
7142 desc->avl = var.avl;
7150 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7151 struct desc_struct *desc, u32 base3,
7154 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7155 struct kvm_segment var;
7157 var.selector = selector;
7158 var.base = get_desc_base(desc);
7159 #ifdef CONFIG_X86_64
7160 var.base |= ((u64)base3) << 32;
7162 var.limit = get_desc_limit(desc);
7164 var.limit = (var.limit << 12) | 0xfff;
7165 var.type = desc->type;
7166 var.dpl = desc->dpl;
7171 var.avl = desc->avl;
7172 var.present = desc->p;
7173 var.unusable = !var.present;
7176 kvm_set_segment(vcpu, &var, seg);
7180 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7181 u32 msr_index, u64 *pdata)
7183 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7186 r = kvm_get_msr(vcpu, msr_index, pdata);
7188 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7189 /* Bounce to user space */
7190 return X86EMUL_IO_NEEDED;
7196 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7197 u32 msr_index, u64 data)
7199 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7202 r = kvm_set_msr(vcpu, msr_index, data);
7204 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7205 /* Bounce to user space */
7206 return X86EMUL_IO_NEEDED;
7212 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7214 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7216 return vcpu->arch.smbase;
7219 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7221 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7223 vcpu->arch.smbase = smbase;
7226 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7229 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7232 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7233 u32 pmc, u64 *pdata)
7235 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7238 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7240 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7243 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7244 struct x86_instruction_info *info,
7245 enum x86_intercept_stage stage)
7247 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7251 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7252 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7255 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7258 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7260 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7263 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7265 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7268 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7270 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7273 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7275 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7278 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7280 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7283 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7285 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7288 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7290 return emul_to_vcpu(ctxt)->arch.hflags;
7293 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7295 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7297 kvm_smm_changed(vcpu, false);
7300 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7301 const char *smstate)
7303 return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7306 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7308 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7311 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7313 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7316 static const struct x86_emulate_ops emulate_ops = {
7317 .read_gpr = emulator_read_gpr,
7318 .write_gpr = emulator_write_gpr,
7319 .read_std = emulator_read_std,
7320 .write_std = emulator_write_std,
7321 .read_phys = kvm_read_guest_phys_system,
7322 .fetch = kvm_fetch_guest_virt,
7323 .read_emulated = emulator_read_emulated,
7324 .write_emulated = emulator_write_emulated,
7325 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7326 .invlpg = emulator_invlpg,
7327 .pio_in_emulated = emulator_pio_in_emulated,
7328 .pio_out_emulated = emulator_pio_out_emulated,
7329 .get_segment = emulator_get_segment,
7330 .set_segment = emulator_set_segment,
7331 .get_cached_segment_base = emulator_get_cached_segment_base,
7332 .get_gdt = emulator_get_gdt,
7333 .get_idt = emulator_get_idt,
7334 .set_gdt = emulator_set_gdt,
7335 .set_idt = emulator_set_idt,
7336 .get_cr = emulator_get_cr,
7337 .set_cr = emulator_set_cr,
7338 .cpl = emulator_get_cpl,
7339 .get_dr = emulator_get_dr,
7340 .set_dr = emulator_set_dr,
7341 .get_smbase = emulator_get_smbase,
7342 .set_smbase = emulator_set_smbase,
7343 .set_msr = emulator_set_msr,
7344 .get_msr = emulator_get_msr,
7345 .check_pmc = emulator_check_pmc,
7346 .read_pmc = emulator_read_pmc,
7347 .halt = emulator_halt,
7348 .wbinvd = emulator_wbinvd,
7349 .fix_hypercall = emulator_fix_hypercall,
7350 .intercept = emulator_intercept,
7351 .get_cpuid = emulator_get_cpuid,
7352 .guest_has_long_mode = emulator_guest_has_long_mode,
7353 .guest_has_movbe = emulator_guest_has_movbe,
7354 .guest_has_fxsr = emulator_guest_has_fxsr,
7355 .set_nmi_mask = emulator_set_nmi_mask,
7356 .get_hflags = emulator_get_hflags,
7357 .exiting_smm = emulator_exiting_smm,
7358 .leave_smm = emulator_leave_smm,
7359 .triple_fault = emulator_triple_fault,
7360 .set_xcr = emulator_set_xcr,
7363 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7365 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7367 * an sti; sti; sequence only disable interrupts for the first
7368 * instruction. So, if the last instruction, be it emulated or
7369 * not, left the system with the INT_STI flag enabled, it
7370 * means that the last instruction is an sti. We should not
7371 * leave the flag on in this case. The same goes for mov ss
7373 if (int_shadow & mask)
7375 if (unlikely(int_shadow || mask)) {
7376 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7378 kvm_make_request(KVM_REQ_EVENT, vcpu);
7382 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7384 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7385 if (ctxt->exception.vector == PF_VECTOR)
7386 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7388 if (ctxt->exception.error_code_valid)
7389 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7390 ctxt->exception.error_code);
7392 kvm_queue_exception(vcpu, ctxt->exception.vector);
7396 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7398 struct x86_emulate_ctxt *ctxt;
7400 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7402 pr_err("kvm: failed to allocate vcpu's emulator\n");
7407 ctxt->ops = &emulate_ops;
7408 vcpu->arch.emulate_ctxt = ctxt;
7413 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7415 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7418 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7420 ctxt->gpa_available = false;
7421 ctxt->eflags = kvm_get_rflags(vcpu);
7422 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7424 ctxt->eip = kvm_rip_read(vcpu);
7425 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7426 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7427 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7428 cs_db ? X86EMUL_MODE_PROT32 :
7429 X86EMUL_MODE_PROT16;
7430 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7431 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7432 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7434 ctxt->interruptibility = 0;
7435 ctxt->have_exception = false;
7436 ctxt->exception.vector = -1;
7437 ctxt->perm_ok = false;
7439 init_decode_cache(ctxt);
7440 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7443 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7445 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7448 init_emulate_ctxt(vcpu);
7452 ctxt->_eip = ctxt->eip + inc_eip;
7453 ret = emulate_int_real(ctxt, irq);
7455 if (ret != X86EMUL_CONTINUE) {
7456 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7458 ctxt->eip = ctxt->_eip;
7459 kvm_rip_write(vcpu, ctxt->eip);
7460 kvm_set_rflags(vcpu, ctxt->eflags);
7463 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7465 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7467 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7468 u32 insn_size = ctxt->fetch.end - ctxt->fetch.data;
7469 struct kvm_run *run = vcpu->run;
7471 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7472 run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7473 run->emulation_failure.ndata = 0;
7474 run->emulation_failure.flags = 0;
7477 run->emulation_failure.ndata = 3;
7478 run->emulation_failure.flags |=
7479 KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7480 run->emulation_failure.insn_size = insn_size;
7481 memset(run->emulation_failure.insn_bytes, 0x90,
7482 sizeof(run->emulation_failure.insn_bytes));
7483 memcpy(run->emulation_failure.insn_bytes,
7484 ctxt->fetch.data, insn_size);
7488 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7490 struct kvm *kvm = vcpu->kvm;
7492 ++vcpu->stat.insn_emulation_fail;
7493 trace_kvm_emulate_insn_failed(vcpu);
7495 if (emulation_type & EMULTYPE_VMWARE_GP) {
7496 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7500 if (kvm->arch.exit_on_emulation_error ||
7501 (emulation_type & EMULTYPE_SKIP)) {
7502 prepare_emulation_failure_exit(vcpu);
7506 kvm_queue_exception(vcpu, UD_VECTOR);
7508 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7509 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7510 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7511 vcpu->run->internal.ndata = 0;
7518 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7519 bool write_fault_to_shadow_pgtable,
7522 gpa_t gpa = cr2_or_gpa;
7525 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7528 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7529 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7532 if (!vcpu->arch.mmu->direct_map) {
7534 * Write permission should be allowed since only
7535 * write access need to be emulated.
7537 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7540 * If the mapping is invalid in guest, let cpu retry
7541 * it to generate fault.
7543 if (gpa == UNMAPPED_GVA)
7548 * Do not retry the unhandleable instruction if it faults on the
7549 * readonly host memory, otherwise it will goto a infinite loop:
7550 * retry instruction -> write #PF -> emulation fail -> retry
7551 * instruction -> ...
7553 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7556 * If the instruction failed on the error pfn, it can not be fixed,
7557 * report the error to userspace.
7559 if (is_error_noslot_pfn(pfn))
7562 kvm_release_pfn_clean(pfn);
7564 /* The instructions are well-emulated on direct mmu. */
7565 if (vcpu->arch.mmu->direct_map) {
7566 unsigned int indirect_shadow_pages;
7568 write_lock(&vcpu->kvm->mmu_lock);
7569 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7570 write_unlock(&vcpu->kvm->mmu_lock);
7572 if (indirect_shadow_pages)
7573 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7579 * if emulation was due to access to shadowed page table
7580 * and it failed try to unshadow page and re-enter the
7581 * guest to let CPU execute the instruction.
7583 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7586 * If the access faults on its page table, it can not
7587 * be fixed by unprotecting shadow page and it should
7588 * be reported to userspace.
7590 return !write_fault_to_shadow_pgtable;
7593 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7594 gpa_t cr2_or_gpa, int emulation_type)
7596 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7597 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7599 last_retry_eip = vcpu->arch.last_retry_eip;
7600 last_retry_addr = vcpu->arch.last_retry_addr;
7603 * If the emulation is caused by #PF and it is non-page_table
7604 * writing instruction, it means the VM-EXIT is caused by shadow
7605 * page protected, we can zap the shadow page and retry this
7606 * instruction directly.
7608 * Note: if the guest uses a non-page-table modifying instruction
7609 * on the PDE that points to the instruction, then we will unmap
7610 * the instruction and go to an infinite loop. So, we cache the
7611 * last retried eip and the last fault address, if we meet the eip
7612 * and the address again, we can break out of the potential infinite
7615 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7617 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7620 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7621 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7624 if (x86_page_table_writing_insn(ctxt))
7627 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7630 vcpu->arch.last_retry_eip = ctxt->eip;
7631 vcpu->arch.last_retry_addr = cr2_or_gpa;
7633 if (!vcpu->arch.mmu->direct_map)
7634 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7636 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7641 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7642 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7644 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7646 trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7649 vcpu->arch.hflags |= HF_SMM_MASK;
7651 vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7653 /* Process a latched INIT or SMI, if any. */
7654 kvm_make_request(KVM_REQ_EVENT, vcpu);
7657 kvm_mmu_reset_context(vcpu);
7660 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7669 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7670 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7675 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7677 struct kvm_run *kvm_run = vcpu->run;
7679 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7680 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7681 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7682 kvm_run->debug.arch.exception = DB_VECTOR;
7683 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7686 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7690 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7692 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7695 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7700 * rflags is the old, "raw" value of the flags. The new value has
7701 * not been saved yet.
7703 * This is correct even for TF set by the guest, because "the
7704 * processor will not generate this exception after the instruction
7705 * that sets the TF flag".
7707 if (unlikely(rflags & X86_EFLAGS_TF))
7708 r = kvm_vcpu_do_singlestep(vcpu);
7711 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7713 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7715 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7716 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7717 struct kvm_run *kvm_run = vcpu->run;
7718 unsigned long eip = kvm_get_linear_rip(vcpu);
7719 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7720 vcpu->arch.guest_debug_dr7,
7724 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7725 kvm_run->debug.arch.pc = eip;
7726 kvm_run->debug.arch.exception = DB_VECTOR;
7727 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7733 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7734 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7735 unsigned long eip = kvm_get_linear_rip(vcpu);
7736 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7741 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7750 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7752 switch (ctxt->opcode_len) {
7759 case 0xe6: /* OUT */
7763 case 0x6c: /* INS */
7765 case 0x6e: /* OUTS */
7772 case 0x33: /* RDPMC */
7782 * Decode to be emulated instruction. Return EMULATION_OK if success.
7784 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7785 void *insn, int insn_len)
7787 int r = EMULATION_OK;
7788 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7790 init_emulate_ctxt(vcpu);
7793 * We will reenter on the same instruction since we do not set
7794 * complete_userspace_io. This does not handle watchpoints yet,
7795 * those would be handled in the emulate_ops.
7797 if (!(emulation_type & EMULTYPE_SKIP) &&
7798 kvm_vcpu_check_breakpoint(vcpu, &r))
7801 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7803 trace_kvm_emulate_insn_start(vcpu);
7804 ++vcpu->stat.insn_emulation;
7808 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7810 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7811 int emulation_type, void *insn, int insn_len)
7814 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7815 bool writeback = true;
7816 bool write_fault_to_spt;
7818 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7821 vcpu->arch.l1tf_flush_l1d = true;
7824 * Clear write_fault_to_shadow_pgtable here to ensure it is
7827 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7828 vcpu->arch.write_fault_to_shadow_pgtable = false;
7830 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7831 kvm_clear_exception_queue(vcpu);
7833 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7835 if (r != EMULATION_OK) {
7836 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7837 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7838 kvm_queue_exception(vcpu, UD_VECTOR);
7841 if (reexecute_instruction(vcpu, cr2_or_gpa,
7845 if (ctxt->have_exception) {
7847 * #UD should result in just EMULATION_FAILED, and trap-like
7848 * exception should not be encountered during decode.
7850 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7851 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7852 inject_emulated_exception(vcpu);
7855 return handle_emulation_failure(vcpu, emulation_type);
7859 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7860 !is_vmware_backdoor_opcode(ctxt)) {
7861 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7866 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7867 * for kvm_skip_emulated_instruction(). The caller is responsible for
7868 * updating interruptibility state and injecting single-step #DBs.
7870 if (emulation_type & EMULTYPE_SKIP) {
7871 kvm_rip_write(vcpu, ctxt->_eip);
7872 if (ctxt->eflags & X86_EFLAGS_RF)
7873 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7877 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7880 /* this is needed for vmware backdoor interface to work since it
7881 changes registers values during IO operation */
7882 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7883 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7884 emulator_invalidate_register_cache(ctxt);
7888 if (emulation_type & EMULTYPE_PF) {
7889 /* Save the faulting GPA (cr2) in the address field */
7890 ctxt->exception.address = cr2_or_gpa;
7892 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7893 if (vcpu->arch.mmu->direct_map) {
7894 ctxt->gpa_available = true;
7895 ctxt->gpa_val = cr2_or_gpa;
7898 /* Sanitize the address out of an abundance of paranoia. */
7899 ctxt->exception.address = 0;
7902 r = x86_emulate_insn(ctxt);
7904 if (r == EMULATION_INTERCEPTED)
7907 if (r == EMULATION_FAILED) {
7908 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7912 return handle_emulation_failure(vcpu, emulation_type);
7915 if (ctxt->have_exception) {
7917 if (inject_emulated_exception(vcpu))
7919 } else if (vcpu->arch.pio.count) {
7920 if (!vcpu->arch.pio.in) {
7921 /* FIXME: return into emulator if single-stepping. */
7922 vcpu->arch.pio.count = 0;
7925 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7928 } else if (vcpu->mmio_needed) {
7929 ++vcpu->stat.mmio_exits;
7931 if (!vcpu->mmio_is_write)
7934 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7935 } else if (r == EMULATION_RESTART)
7941 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7942 toggle_interruptibility(vcpu, ctxt->interruptibility);
7943 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7944 if (!ctxt->have_exception ||
7945 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7946 kvm_rip_write(vcpu, ctxt->eip);
7947 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7948 r = kvm_vcpu_do_singlestep(vcpu);
7949 if (kvm_x86_ops.update_emulated_instruction)
7950 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7951 __kvm_set_rflags(vcpu, ctxt->eflags);
7955 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7956 * do nothing, and it will be requested again as soon as
7957 * the shadow expires. But we still need to check here,
7958 * because POPF has no interrupt shadow.
7960 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7961 kvm_make_request(KVM_REQ_EVENT, vcpu);
7963 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7968 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7970 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7972 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7974 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7975 void *insn, int insn_len)
7977 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7979 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7981 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7983 vcpu->arch.pio.count = 0;
7987 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7989 vcpu->arch.pio.count = 0;
7991 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7994 return kvm_skip_emulated_instruction(vcpu);
7997 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7998 unsigned short port)
8000 unsigned long val = kvm_rax_read(vcpu);
8001 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8007 * Workaround userspace that relies on old KVM behavior of %rip being
8008 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8011 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8012 vcpu->arch.complete_userspace_io =
8013 complete_fast_pio_out_port_0x7e;
8014 kvm_skip_emulated_instruction(vcpu);
8016 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8017 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8022 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8026 /* We should only ever be called with arch.pio.count equal to 1 */
8027 BUG_ON(vcpu->arch.pio.count != 1);
8029 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8030 vcpu->arch.pio.count = 0;
8034 /* For size less than 4 we merge, else we zero extend */
8035 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8038 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8039 * the copy and tracing
8041 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8042 kvm_rax_write(vcpu, val);
8044 return kvm_skip_emulated_instruction(vcpu);
8047 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8048 unsigned short port)
8053 /* For size less than 4 we merge, else we zero extend */
8054 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8056 ret = emulator_pio_in(vcpu, size, port, &val, 1);
8058 kvm_rax_write(vcpu, val);
8062 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8063 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8068 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8073 ret = kvm_fast_pio_in(vcpu, size, port);
8075 ret = kvm_fast_pio_out(vcpu, size, port);
8076 return ret && kvm_skip_emulated_instruction(vcpu);
8078 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8080 static int kvmclock_cpu_down_prep(unsigned int cpu)
8082 __this_cpu_write(cpu_tsc_khz, 0);
8086 static void tsc_khz_changed(void *data)
8088 struct cpufreq_freqs *freq = data;
8089 unsigned long khz = 0;
8093 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8094 khz = cpufreq_quick_get(raw_smp_processor_id());
8097 __this_cpu_write(cpu_tsc_khz, khz);
8100 #ifdef CONFIG_X86_64
8101 static void kvm_hyperv_tsc_notifier(void)
8104 struct kvm_vcpu *vcpu;
8106 unsigned long flags;
8108 mutex_lock(&kvm_lock);
8109 list_for_each_entry(kvm, &vm_list, vm_list)
8110 kvm_make_mclock_inprogress_request(kvm);
8112 hyperv_stop_tsc_emulation();
8114 /* TSC frequency always matches when on Hyper-V */
8115 for_each_present_cpu(cpu)
8116 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8117 kvm_max_guest_tsc_khz = tsc_khz;
8119 list_for_each_entry(kvm, &vm_list, vm_list) {
8120 struct kvm_arch *ka = &kvm->arch;
8122 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
8123 pvclock_update_vm_gtod_copy(kvm);
8124 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
8126 kvm_for_each_vcpu(cpu, vcpu, kvm)
8127 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8129 kvm_for_each_vcpu(cpu, vcpu, kvm)
8130 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
8132 mutex_unlock(&kvm_lock);
8136 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8139 struct kvm_vcpu *vcpu;
8140 int i, send_ipi = 0;
8143 * We allow guests to temporarily run on slowing clocks,
8144 * provided we notify them after, or to run on accelerating
8145 * clocks, provided we notify them before. Thus time never
8148 * However, we have a problem. We can't atomically update
8149 * the frequency of a given CPU from this function; it is
8150 * merely a notifier, which can be called from any CPU.
8151 * Changing the TSC frequency at arbitrary points in time
8152 * requires a recomputation of local variables related to
8153 * the TSC for each VCPU. We must flag these local variables
8154 * to be updated and be sure the update takes place with the
8155 * new frequency before any guests proceed.
8157 * Unfortunately, the combination of hotplug CPU and frequency
8158 * change creates an intractable locking scenario; the order
8159 * of when these callouts happen is undefined with respect to
8160 * CPU hotplug, and they can race with each other. As such,
8161 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8162 * undefined; you can actually have a CPU frequency change take
8163 * place in between the computation of X and the setting of the
8164 * variable. To protect against this problem, all updates of
8165 * the per_cpu tsc_khz variable are done in an interrupt
8166 * protected IPI, and all callers wishing to update the value
8167 * must wait for a synchronous IPI to complete (which is trivial
8168 * if the caller is on the CPU already). This establishes the
8169 * necessary total order on variable updates.
8171 * Note that because a guest time update may take place
8172 * anytime after the setting of the VCPU's request bit, the
8173 * correct TSC value must be set before the request. However,
8174 * to ensure the update actually makes it to any guest which
8175 * starts running in hardware virtualization between the set
8176 * and the acquisition of the spinlock, we must also ping the
8177 * CPU after setting the request bit.
8181 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8183 mutex_lock(&kvm_lock);
8184 list_for_each_entry(kvm, &vm_list, vm_list) {
8185 kvm_for_each_vcpu(i, vcpu, kvm) {
8186 if (vcpu->cpu != cpu)
8188 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8189 if (vcpu->cpu != raw_smp_processor_id())
8193 mutex_unlock(&kvm_lock);
8195 if (freq->old < freq->new && send_ipi) {
8197 * We upscale the frequency. Must make the guest
8198 * doesn't see old kvmclock values while running with
8199 * the new frequency, otherwise we risk the guest sees
8200 * time go backwards.
8202 * In case we update the frequency for another cpu
8203 * (which might be in guest context) send an interrupt
8204 * to kick the cpu out of guest context. Next time
8205 * guest context is entered kvmclock will be updated,
8206 * so the guest will not see stale values.
8208 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8212 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8215 struct cpufreq_freqs *freq = data;
8218 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8220 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8223 for_each_cpu(cpu, freq->policy->cpus)
8224 __kvmclock_cpufreq_notifier(freq, cpu);
8229 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8230 .notifier_call = kvmclock_cpufreq_notifier
8233 static int kvmclock_cpu_online(unsigned int cpu)
8235 tsc_khz_changed(NULL);
8239 static void kvm_timer_init(void)
8241 max_tsc_khz = tsc_khz;
8243 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8244 #ifdef CONFIG_CPU_FREQ
8245 struct cpufreq_policy *policy;
8249 policy = cpufreq_cpu_get(cpu);
8251 if (policy->cpuinfo.max_freq)
8252 max_tsc_khz = policy->cpuinfo.max_freq;
8253 cpufreq_cpu_put(policy);
8257 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8258 CPUFREQ_TRANSITION_NOTIFIER);
8261 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8262 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8265 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8266 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8268 int kvm_is_in_guest(void)
8270 return __this_cpu_read(current_vcpu) != NULL;
8273 static int kvm_is_user_mode(void)
8277 if (__this_cpu_read(current_vcpu))
8278 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8280 return user_mode != 0;
8283 static unsigned long kvm_get_guest_ip(void)
8285 unsigned long ip = 0;
8287 if (__this_cpu_read(current_vcpu))
8288 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8293 static void kvm_handle_intel_pt_intr(void)
8295 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8297 kvm_make_request(KVM_REQ_PMI, vcpu);
8298 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8299 (unsigned long *)&vcpu->arch.pmu.global_status);
8302 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8303 .is_in_guest = kvm_is_in_guest,
8304 .is_user_mode = kvm_is_user_mode,
8305 .get_guest_ip = kvm_get_guest_ip,
8306 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8309 #ifdef CONFIG_X86_64
8310 static void pvclock_gtod_update_fn(struct work_struct *work)
8314 struct kvm_vcpu *vcpu;
8317 mutex_lock(&kvm_lock);
8318 list_for_each_entry(kvm, &vm_list, vm_list)
8319 kvm_for_each_vcpu(i, vcpu, kvm)
8320 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8321 atomic_set(&kvm_guest_has_master_clock, 0);
8322 mutex_unlock(&kvm_lock);
8325 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8328 * Indirection to move queue_work() out of the tk_core.seq write held
8329 * region to prevent possible deadlocks against time accessors which
8330 * are invoked with work related locks held.
8332 static void pvclock_irq_work_fn(struct irq_work *w)
8334 queue_work(system_long_wq, &pvclock_gtod_work);
8337 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8340 * Notification about pvclock gtod data update.
8342 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8345 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8346 struct timekeeper *tk = priv;
8348 update_pvclock_gtod(tk);
8351 * Disable master clock if host does not trust, or does not use,
8352 * TSC based clocksource. Delegate queue_work() to irq_work as
8353 * this is invoked with tk_core.seq write held.
8355 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8356 atomic_read(&kvm_guest_has_master_clock) != 0)
8357 irq_work_queue(&pvclock_irq_work);
8361 static struct notifier_block pvclock_gtod_notifier = {
8362 .notifier_call = pvclock_gtod_notify,
8366 int kvm_arch_init(void *opaque)
8368 struct kvm_x86_init_ops *ops = opaque;
8371 if (kvm_x86_ops.hardware_enable) {
8372 printk(KERN_ERR "kvm: already loaded the other module\n");
8377 if (!ops->cpu_has_kvm_support()) {
8378 pr_err_ratelimited("kvm: no hardware support\n");
8382 if (ops->disabled_by_bios()) {
8383 pr_err_ratelimited("kvm: disabled by bios\n");
8389 * KVM explicitly assumes that the guest has an FPU and
8390 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8391 * vCPU's FPU state as a fxregs_state struct.
8393 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8394 printk(KERN_ERR "kvm: inadequate fpu\n");
8400 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8401 __alignof__(struct fpu), SLAB_ACCOUNT,
8403 if (!x86_fpu_cache) {
8404 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8408 x86_emulator_cache = kvm_alloc_emulator_cache();
8409 if (!x86_emulator_cache) {
8410 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8411 goto out_free_x86_fpu_cache;
8414 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8415 if (!user_return_msrs) {
8416 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8417 goto out_free_x86_emulator_cache;
8419 kvm_nr_uret_msrs = 0;
8421 r = kvm_mmu_module_init();
8423 goto out_free_percpu;
8427 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8429 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8430 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8431 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8434 if (pi_inject_timer == -1)
8435 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8436 #ifdef CONFIG_X86_64
8437 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8439 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8440 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8446 free_percpu(user_return_msrs);
8447 out_free_x86_emulator_cache:
8448 kmem_cache_destroy(x86_emulator_cache);
8449 out_free_x86_fpu_cache:
8450 kmem_cache_destroy(x86_fpu_cache);
8455 void kvm_arch_exit(void)
8457 #ifdef CONFIG_X86_64
8458 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8459 clear_hv_tscchange_cb();
8462 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8464 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8465 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8466 CPUFREQ_TRANSITION_NOTIFIER);
8467 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8468 #ifdef CONFIG_X86_64
8469 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8470 irq_work_sync(&pvclock_irq_work);
8471 cancel_work_sync(&pvclock_gtod_work);
8473 kvm_x86_ops.hardware_enable = NULL;
8474 kvm_mmu_module_exit();
8475 free_percpu(user_return_msrs);
8476 kmem_cache_destroy(x86_emulator_cache);
8477 kmem_cache_destroy(x86_fpu_cache);
8478 #ifdef CONFIG_KVM_XEN
8479 static_key_deferred_flush(&kvm_xen_enabled);
8480 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8484 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8486 ++vcpu->stat.halt_exits;
8487 if (lapic_in_kernel(vcpu)) {
8488 vcpu->arch.mp_state = state;
8491 vcpu->run->exit_reason = reason;
8496 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8498 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8500 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8502 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8504 int ret = kvm_skip_emulated_instruction(vcpu);
8506 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8507 * KVM_EXIT_DEBUG here.
8509 return kvm_vcpu_halt(vcpu) && ret;
8511 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8513 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8515 int ret = kvm_skip_emulated_instruction(vcpu);
8517 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8519 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8521 #ifdef CONFIG_X86_64
8522 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8523 unsigned long clock_type)
8525 struct kvm_clock_pairing clock_pairing;
8526 struct timespec64 ts;
8530 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8531 return -KVM_EOPNOTSUPP;
8533 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8534 return -KVM_EOPNOTSUPP;
8536 clock_pairing.sec = ts.tv_sec;
8537 clock_pairing.nsec = ts.tv_nsec;
8538 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8539 clock_pairing.flags = 0;
8540 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8543 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8544 sizeof(struct kvm_clock_pairing)))
8552 * kvm_pv_kick_cpu_op: Kick a vcpu.
8554 * @apicid - apicid of vcpu to be kicked.
8556 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8558 struct kvm_lapic_irq lapic_irq;
8560 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8561 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8562 lapic_irq.level = 0;
8563 lapic_irq.dest_id = apicid;
8564 lapic_irq.msi_redir_hint = false;
8566 lapic_irq.delivery_mode = APIC_DM_REMRD;
8567 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8570 bool kvm_apicv_activated(struct kvm *kvm)
8572 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8574 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8576 static void kvm_apicv_init(struct kvm *kvm)
8578 mutex_init(&kvm->arch.apicv_update_lock);
8581 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8582 &kvm->arch.apicv_inhibit_reasons);
8584 set_bit(APICV_INHIBIT_REASON_DISABLE,
8585 &kvm->arch.apicv_inhibit_reasons);
8588 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8590 struct kvm_vcpu *target = NULL;
8591 struct kvm_apic_map *map;
8593 vcpu->stat.directed_yield_attempted++;
8595 if (single_task_running())
8599 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8601 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8602 target = map->phys_map[dest_id]->vcpu;
8606 if (!target || !READ_ONCE(target->ready))
8609 /* Ignore requests to yield to self */
8613 if (kvm_vcpu_yield_to(target) <= 0)
8616 vcpu->stat.directed_yield_successful++;
8622 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8624 u64 ret = vcpu->run->hypercall.ret;
8626 if (!is_64_bit_mode(vcpu))
8628 kvm_rax_write(vcpu, ret);
8629 ++vcpu->stat.hypercalls;
8630 return kvm_skip_emulated_instruction(vcpu);
8633 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8635 unsigned long nr, a0, a1, a2, a3, ret;
8638 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8639 return kvm_xen_hypercall(vcpu);
8641 if (kvm_hv_hypercall_enabled(vcpu))
8642 return kvm_hv_hypercall(vcpu);
8644 nr = kvm_rax_read(vcpu);
8645 a0 = kvm_rbx_read(vcpu);
8646 a1 = kvm_rcx_read(vcpu);
8647 a2 = kvm_rdx_read(vcpu);
8648 a3 = kvm_rsi_read(vcpu);
8650 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8652 op_64_bit = is_64_bit_mode(vcpu);
8661 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8669 case KVM_HC_VAPIC_POLL_IRQ:
8672 case KVM_HC_KICK_CPU:
8673 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8676 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8677 kvm_sched_yield(vcpu, a1);
8680 #ifdef CONFIG_X86_64
8681 case KVM_HC_CLOCK_PAIRING:
8682 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8685 case KVM_HC_SEND_IPI:
8686 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8689 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8691 case KVM_HC_SCHED_YIELD:
8692 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8695 kvm_sched_yield(vcpu, a0);
8698 case KVM_HC_MAP_GPA_RANGE: {
8699 u64 gpa = a0, npages = a1, attrs = a2;
8702 if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
8705 if (!PAGE_ALIGNED(gpa) || !npages ||
8706 gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
8711 vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
8712 vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
8713 vcpu->run->hypercall.args[0] = gpa;
8714 vcpu->run->hypercall.args[1] = npages;
8715 vcpu->run->hypercall.args[2] = attrs;
8716 vcpu->run->hypercall.longmode = op_64_bit;
8717 vcpu->arch.complete_userspace_io = complete_hypercall_exit;
8727 kvm_rax_write(vcpu, ret);
8729 ++vcpu->stat.hypercalls;
8730 return kvm_skip_emulated_instruction(vcpu);
8732 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8734 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8736 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8737 char instruction[3];
8738 unsigned long rip = kvm_rip_read(vcpu);
8740 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8742 return emulator_write_emulated(ctxt, rip, instruction, 3,
8746 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8748 return vcpu->run->request_interrupt_window &&
8749 likely(!pic_in_kernel(vcpu->kvm));
8752 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8754 struct kvm_run *kvm_run = vcpu->run;
8757 * if_flag is obsolete and useless, so do not bother
8758 * setting it for SEV-ES guests. Userspace can just
8759 * use kvm_run->ready_for_interrupt_injection.
8761 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8762 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8764 kvm_run->cr8 = kvm_get_cr8(vcpu);
8765 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8766 kvm_run->ready_for_interrupt_injection =
8767 pic_in_kernel(vcpu->kvm) ||
8768 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8771 kvm_run->flags |= KVM_RUN_X86_SMM;
8774 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8778 if (!kvm_x86_ops.update_cr8_intercept)
8781 if (!lapic_in_kernel(vcpu))
8784 if (vcpu->arch.apicv_active)
8787 if (!vcpu->arch.apic->vapic_addr)
8788 max_irr = kvm_lapic_find_highest_irr(vcpu);
8795 tpr = kvm_lapic_get_cr8(vcpu);
8797 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8801 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8803 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8804 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8808 return kvm_x86_ops.nested_ops->check_events(vcpu);
8811 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8813 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8814 vcpu->arch.exception.error_code = false;
8815 static_call(kvm_x86_queue_exception)(vcpu);
8818 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8821 bool can_inject = true;
8823 /* try to reinject previous events if any */
8825 if (vcpu->arch.exception.injected) {
8826 kvm_inject_exception(vcpu);
8830 * Do not inject an NMI or interrupt if there is a pending
8831 * exception. Exceptions and interrupts are recognized at
8832 * instruction boundaries, i.e. the start of an instruction.
8833 * Trap-like exceptions, e.g. #DB, have higher priority than
8834 * NMIs and interrupts, i.e. traps are recognized before an
8835 * NMI/interrupt that's pending on the same instruction.
8836 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8837 * priority, but are only generated (pended) during instruction
8838 * execution, i.e. a pending fault-like exception means the
8839 * fault occurred on the *previous* instruction and must be
8840 * serviced prior to recognizing any new events in order to
8841 * fully complete the previous instruction.
8843 else if (!vcpu->arch.exception.pending) {
8844 if (vcpu->arch.nmi_injected) {
8845 static_call(kvm_x86_set_nmi)(vcpu);
8847 } else if (vcpu->arch.interrupt.injected) {
8848 static_call(kvm_x86_set_irq)(vcpu);
8853 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8854 vcpu->arch.exception.pending);
8857 * Call check_nested_events() even if we reinjected a previous event
8858 * in order for caller to determine if it should require immediate-exit
8859 * from L2 to L1 due to pending L1 events which require exit
8862 if (is_guest_mode(vcpu)) {
8863 r = kvm_check_nested_events(vcpu);
8868 /* try to inject new event if pending */
8869 if (vcpu->arch.exception.pending) {
8870 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8871 vcpu->arch.exception.has_error_code,
8872 vcpu->arch.exception.error_code);
8874 vcpu->arch.exception.pending = false;
8875 vcpu->arch.exception.injected = true;
8877 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8878 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8881 if (vcpu->arch.exception.nr == DB_VECTOR) {
8882 kvm_deliver_exception_payload(vcpu);
8883 if (vcpu->arch.dr7 & DR7_GD) {
8884 vcpu->arch.dr7 &= ~DR7_GD;
8885 kvm_update_dr7(vcpu);
8889 kvm_inject_exception(vcpu);
8894 * Finally, inject interrupt events. If an event cannot be injected
8895 * due to architectural conditions (e.g. IF=0) a window-open exit
8896 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8897 * and can architecturally be injected, but we cannot do it right now:
8898 * an interrupt could have arrived just now and we have to inject it
8899 * as a vmexit, or there could already an event in the queue, which is
8900 * indicated by can_inject. In that case we request an immediate exit
8901 * in order to make progress and get back here for another iteration.
8902 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8904 if (vcpu->arch.smi_pending) {
8905 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8909 vcpu->arch.smi_pending = false;
8910 ++vcpu->arch.smi_count;
8914 static_call(kvm_x86_enable_smi_window)(vcpu);
8917 if (vcpu->arch.nmi_pending) {
8918 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8922 --vcpu->arch.nmi_pending;
8923 vcpu->arch.nmi_injected = true;
8924 static_call(kvm_x86_set_nmi)(vcpu);
8926 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8928 if (vcpu->arch.nmi_pending)
8929 static_call(kvm_x86_enable_nmi_window)(vcpu);
8932 if (kvm_cpu_has_injectable_intr(vcpu)) {
8933 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8937 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8938 static_call(kvm_x86_set_irq)(vcpu);
8939 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8941 if (kvm_cpu_has_injectable_intr(vcpu))
8942 static_call(kvm_x86_enable_irq_window)(vcpu);
8945 if (is_guest_mode(vcpu) &&
8946 kvm_x86_ops.nested_ops->hv_timer_pending &&
8947 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8948 *req_immediate_exit = true;
8950 WARN_ON(vcpu->arch.exception.pending);
8955 *req_immediate_exit = true;
8961 static void process_nmi(struct kvm_vcpu *vcpu)
8966 * x86 is limited to one NMI running, and one NMI pending after it.
8967 * If an NMI is already in progress, limit further NMIs to just one.
8968 * Otherwise, allow two (and we'll inject the first one immediately).
8970 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8973 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8974 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8975 kvm_make_request(KVM_REQ_EVENT, vcpu);
8978 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8981 flags |= seg->g << 23;
8982 flags |= seg->db << 22;
8983 flags |= seg->l << 21;
8984 flags |= seg->avl << 20;
8985 flags |= seg->present << 15;
8986 flags |= seg->dpl << 13;
8987 flags |= seg->s << 12;
8988 flags |= seg->type << 8;
8992 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8994 struct kvm_segment seg;
8997 kvm_get_segment(vcpu, &seg, n);
8998 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9001 offset = 0x7f84 + n * 12;
9003 offset = 0x7f2c + (n - 3) * 12;
9005 put_smstate(u32, buf, offset + 8, seg.base);
9006 put_smstate(u32, buf, offset + 4, seg.limit);
9007 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9010 #ifdef CONFIG_X86_64
9011 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9013 struct kvm_segment seg;
9017 kvm_get_segment(vcpu, &seg, n);
9018 offset = 0x7e00 + n * 16;
9020 flags = enter_smm_get_segment_flags(&seg) >> 8;
9021 put_smstate(u16, buf, offset, seg.selector);
9022 put_smstate(u16, buf, offset + 2, flags);
9023 put_smstate(u32, buf, offset + 4, seg.limit);
9024 put_smstate(u64, buf, offset + 8, seg.base);
9028 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9031 struct kvm_segment seg;
9035 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9036 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9037 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9038 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9040 for (i = 0; i < 8; i++)
9041 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9043 kvm_get_dr(vcpu, 6, &val);
9044 put_smstate(u32, buf, 0x7fcc, (u32)val);
9045 kvm_get_dr(vcpu, 7, &val);
9046 put_smstate(u32, buf, 0x7fc8, (u32)val);
9048 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9049 put_smstate(u32, buf, 0x7fc4, seg.selector);
9050 put_smstate(u32, buf, 0x7f64, seg.base);
9051 put_smstate(u32, buf, 0x7f60, seg.limit);
9052 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9054 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9055 put_smstate(u32, buf, 0x7fc0, seg.selector);
9056 put_smstate(u32, buf, 0x7f80, seg.base);
9057 put_smstate(u32, buf, 0x7f7c, seg.limit);
9058 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9060 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9061 put_smstate(u32, buf, 0x7f74, dt.address);
9062 put_smstate(u32, buf, 0x7f70, dt.size);
9064 static_call(kvm_x86_get_idt)(vcpu, &dt);
9065 put_smstate(u32, buf, 0x7f58, dt.address);
9066 put_smstate(u32, buf, 0x7f54, dt.size);
9068 for (i = 0; i < 6; i++)
9069 enter_smm_save_seg_32(vcpu, buf, i);
9071 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9074 put_smstate(u32, buf, 0x7efc, 0x00020000);
9075 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9078 #ifdef CONFIG_X86_64
9079 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9082 struct kvm_segment seg;
9086 for (i = 0; i < 16; i++)
9087 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9089 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9090 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9092 kvm_get_dr(vcpu, 6, &val);
9093 put_smstate(u64, buf, 0x7f68, val);
9094 kvm_get_dr(vcpu, 7, &val);
9095 put_smstate(u64, buf, 0x7f60, val);
9097 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9098 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9099 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9101 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9104 put_smstate(u32, buf, 0x7efc, 0x00020064);
9106 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9108 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9109 put_smstate(u16, buf, 0x7e90, seg.selector);
9110 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9111 put_smstate(u32, buf, 0x7e94, seg.limit);
9112 put_smstate(u64, buf, 0x7e98, seg.base);
9114 static_call(kvm_x86_get_idt)(vcpu, &dt);
9115 put_smstate(u32, buf, 0x7e84, dt.size);
9116 put_smstate(u64, buf, 0x7e88, dt.address);
9118 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9119 put_smstate(u16, buf, 0x7e70, seg.selector);
9120 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9121 put_smstate(u32, buf, 0x7e74, seg.limit);
9122 put_smstate(u64, buf, 0x7e78, seg.base);
9124 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9125 put_smstate(u32, buf, 0x7e64, dt.size);
9126 put_smstate(u64, buf, 0x7e68, dt.address);
9128 for (i = 0; i < 6; i++)
9129 enter_smm_save_seg_64(vcpu, buf, i);
9133 static void enter_smm(struct kvm_vcpu *vcpu)
9135 struct kvm_segment cs, ds;
9140 memset(buf, 0, 512);
9141 #ifdef CONFIG_X86_64
9142 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9143 enter_smm_save_state_64(vcpu, buf);
9146 enter_smm_save_state_32(vcpu, buf);
9149 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9150 * state (e.g. leave guest mode) after we've saved the state into the
9151 * SMM state-save area.
9153 static_call(kvm_x86_enter_smm)(vcpu, buf);
9155 kvm_smm_changed(vcpu, true);
9156 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9158 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9159 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9161 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9163 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9164 kvm_rip_write(vcpu, 0x8000);
9166 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9167 static_call(kvm_x86_set_cr0)(vcpu, cr0);
9168 vcpu->arch.cr0 = cr0;
9170 static_call(kvm_x86_set_cr4)(vcpu, 0);
9172 /* Undocumented: IDT limit is set to zero on entry to SMM. */
9173 dt.address = dt.size = 0;
9174 static_call(kvm_x86_set_idt)(vcpu, &dt);
9176 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9178 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9179 cs.base = vcpu->arch.smbase;
9184 cs.limit = ds.limit = 0xffffffff;
9185 cs.type = ds.type = 0x3;
9186 cs.dpl = ds.dpl = 0;
9191 cs.avl = ds.avl = 0;
9192 cs.present = ds.present = 1;
9193 cs.unusable = ds.unusable = 0;
9194 cs.padding = ds.padding = 0;
9196 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9197 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9198 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9199 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9200 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9201 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9203 #ifdef CONFIG_X86_64
9204 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9205 static_call(kvm_x86_set_efer)(vcpu, 0);
9208 kvm_update_cpuid_runtime(vcpu);
9209 kvm_mmu_reset_context(vcpu);
9212 static void process_smi(struct kvm_vcpu *vcpu)
9214 vcpu->arch.smi_pending = true;
9215 kvm_make_request(KVM_REQ_EVENT, vcpu);
9218 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9219 unsigned long *vcpu_bitmap)
9223 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
9225 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
9226 NULL, vcpu_bitmap, cpus);
9228 free_cpumask_var(cpus);
9231 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9233 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9236 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9240 if (!lapic_in_kernel(vcpu))
9243 mutex_lock(&vcpu->kvm->arch.apicv_update_lock);
9245 activate = kvm_apicv_activated(vcpu->kvm);
9246 if (vcpu->arch.apicv_active == activate)
9249 vcpu->arch.apicv_active = activate;
9250 kvm_apic_update_apicv(vcpu);
9251 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9254 * When APICv gets disabled, we may still have injected interrupts
9255 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9256 * still active when the interrupt got accepted. Make sure
9257 * inject_pending_event() is called to check for that.
9259 if (!vcpu->arch.apicv_active)
9260 kvm_make_request(KVM_REQ_EVENT, vcpu);
9263 mutex_unlock(&vcpu->kvm->arch.apicv_update_lock);
9265 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9267 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9269 unsigned long old, new;
9271 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9272 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9275 old = new = kvm->arch.apicv_inhibit_reasons;
9278 __clear_bit(bit, &new);
9280 __set_bit(bit, &new);
9282 if (!!old != !!new) {
9283 trace_kvm_apicv_update_request(activate, bit);
9284 kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9285 kvm->arch.apicv_inhibit_reasons = new;
9287 unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9288 kvm_zap_gfn_range(kvm, gfn, gfn+1);
9291 kvm->arch.apicv_inhibit_reasons = new;
9293 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9295 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9297 mutex_lock(&kvm->arch.apicv_update_lock);
9298 __kvm_request_apicv_update(kvm, activate, bit);
9299 mutex_unlock(&kvm->arch.apicv_update_lock);
9301 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9303 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9305 if (!kvm_apic_present(vcpu))
9308 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9310 if (irqchip_split(vcpu->kvm))
9311 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9313 if (vcpu->arch.apicv_active)
9314 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9315 if (ioapic_in_kernel(vcpu->kvm))
9316 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9319 if (is_guest_mode(vcpu))
9320 vcpu->arch.load_eoi_exitmap_pending = true;
9322 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9325 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9327 u64 eoi_exit_bitmap[4];
9329 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9332 if (to_hv_vcpu(vcpu))
9333 bitmap_or((ulong *)eoi_exit_bitmap,
9334 vcpu->arch.ioapic_handled_vectors,
9335 to_hv_synic(vcpu)->vec_bitmap, 256);
9337 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9340 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9341 unsigned long start, unsigned long end)
9343 unsigned long apic_address;
9346 * The physical address of apic access page is stored in the VMCS.
9347 * Update it when it becomes invalid.
9349 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9350 if (start <= apic_address && apic_address < end)
9351 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9354 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9356 if (!lapic_in_kernel(vcpu))
9359 if (!kvm_x86_ops.set_apic_access_page_addr)
9362 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9365 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9367 smp_send_reschedule(vcpu->cpu);
9369 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9372 * Returns 1 to let vcpu_run() continue the guest execution loop without
9373 * exiting to the userspace. Otherwise, the value will be returned to the
9376 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9380 dm_request_for_irq_injection(vcpu) &&
9381 kvm_cpu_accept_dm_intr(vcpu);
9382 fastpath_t exit_fastpath;
9384 bool req_immediate_exit = false;
9386 /* Forbid vmenter if vcpu dirty ring is soft-full */
9387 if (unlikely(vcpu->kvm->dirty_ring_size &&
9388 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9389 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9390 trace_kvm_dirty_ring_exit(vcpu);
9395 if (kvm_request_pending(vcpu)) {
9396 if (kvm_check_request(KVM_REQ_VM_BUGGED, vcpu)) {
9400 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9401 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9406 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9407 kvm_mmu_unload(vcpu);
9408 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9409 __kvm_migrate_timers(vcpu);
9410 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9411 kvm_gen_update_masterclock(vcpu->kvm);
9412 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9413 kvm_gen_kvmclock_update(vcpu);
9414 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9415 r = kvm_guest_time_update(vcpu);
9419 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9420 kvm_mmu_sync_roots(vcpu);
9421 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9422 kvm_mmu_load_pgd(vcpu);
9423 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9424 kvm_vcpu_flush_tlb_all(vcpu);
9426 /* Flushing all ASIDs flushes the current ASID... */
9427 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9429 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9430 kvm_vcpu_flush_tlb_current(vcpu);
9431 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9432 kvm_vcpu_flush_tlb_guest(vcpu);
9434 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9435 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9439 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9440 if (is_guest_mode(vcpu)) {
9441 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9443 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9444 vcpu->mmio_needed = 0;
9449 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9450 /* Page is swapped out. Do synthetic halt */
9451 vcpu->arch.apf.halted = true;
9455 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9456 record_steal_time(vcpu);
9457 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9459 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9461 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9462 kvm_pmu_handle_event(vcpu);
9463 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9464 kvm_pmu_deliver_pmi(vcpu);
9465 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9466 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9467 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9468 vcpu->arch.ioapic_handled_vectors)) {
9469 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9470 vcpu->run->eoi.vector =
9471 vcpu->arch.pending_ioapic_eoi;
9476 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9477 vcpu_scan_ioapic(vcpu);
9478 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9479 vcpu_load_eoi_exitmap(vcpu);
9480 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9481 kvm_vcpu_reload_apic_access_page(vcpu);
9482 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9483 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9484 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9488 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9489 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9490 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9494 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9495 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9497 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9498 vcpu->run->hyperv = hv_vcpu->exit;
9504 * KVM_REQ_HV_STIMER has to be processed after
9505 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9506 * depend on the guest clock being up-to-date
9508 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9509 kvm_hv_process_stimers(vcpu);
9510 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9511 kvm_vcpu_update_apicv(vcpu);
9512 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9513 kvm_check_async_pf_completion(vcpu);
9514 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9515 static_call(kvm_x86_msr_filter_changed)(vcpu);
9517 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9518 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9521 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9522 kvm_xen_has_interrupt(vcpu)) {
9523 ++vcpu->stat.req_event;
9524 r = kvm_apic_accept_events(vcpu);
9529 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9534 r = inject_pending_event(vcpu, &req_immediate_exit);
9540 static_call(kvm_x86_enable_irq_window)(vcpu);
9542 if (kvm_lapic_enabled(vcpu)) {
9543 update_cr8_intercept(vcpu);
9544 kvm_lapic_sync_to_vapic(vcpu);
9548 r = kvm_mmu_reload(vcpu);
9550 goto cancel_injection;
9555 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9558 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9559 * IPI are then delayed after guest entry, which ensures that they
9560 * result in virtual interrupt delivery.
9562 local_irq_disable();
9563 vcpu->mode = IN_GUEST_MODE;
9565 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9568 * 1) We should set ->mode before checking ->requests. Please see
9569 * the comment in kvm_vcpu_exiting_guest_mode().
9571 * 2) For APICv, we should set ->mode before checking PID.ON. This
9572 * pairs with the memory barrier implicit in pi_test_and_set_on
9573 * (see vmx_deliver_posted_interrupt).
9575 * 3) This also orders the write to mode from any reads to the page
9576 * tables done while the VCPU is running. Please see the comment
9577 * in kvm_flush_remote_tlbs.
9579 smp_mb__after_srcu_read_unlock();
9582 * This handles the case where a posted interrupt was
9583 * notified with kvm_vcpu_kick.
9585 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9586 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9588 if (kvm_vcpu_exit_request(vcpu)) {
9589 vcpu->mode = OUTSIDE_GUEST_MODE;
9593 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9595 goto cancel_injection;
9598 if (req_immediate_exit) {
9599 kvm_make_request(KVM_REQ_EVENT, vcpu);
9600 static_call(kvm_x86_request_immediate_exit)(vcpu);
9603 fpregs_assert_state_consistent();
9604 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9605 switch_fpu_return();
9607 if (unlikely(vcpu->arch.switch_db_regs)) {
9609 set_debugreg(vcpu->arch.eff_db[0], 0);
9610 set_debugreg(vcpu->arch.eff_db[1], 1);
9611 set_debugreg(vcpu->arch.eff_db[2], 2);
9612 set_debugreg(vcpu->arch.eff_db[3], 3);
9613 } else if (unlikely(hw_breakpoint_active())) {
9618 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9619 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9622 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9623 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9627 if (vcpu->arch.apicv_active)
9628 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9632 * Do this here before restoring debug registers on the host. And
9633 * since we do this before handling the vmexit, a DR access vmexit
9634 * can (a) read the correct value of the debug registers, (b) set
9635 * KVM_DEBUGREG_WONT_EXIT again.
9637 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9638 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9639 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9640 kvm_update_dr0123(vcpu);
9641 kvm_update_dr7(vcpu);
9645 * If the guest has used debug registers, at least dr7
9646 * will be disabled while returning to the host.
9647 * If we don't have active breakpoints in the host, we don't
9648 * care about the messed up debug address registers. But if
9649 * we have some of them active, restore the old state.
9651 if (hw_breakpoint_active())
9652 hw_breakpoint_restore();
9654 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9655 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9657 vcpu->mode = OUTSIDE_GUEST_MODE;
9660 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9663 * Consume any pending interrupts, including the possible source of
9664 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9665 * An instruction is required after local_irq_enable() to fully unblock
9666 * interrupts on processors that implement an interrupt shadow, the
9667 * stat.exits increment will do nicely.
9669 kvm_before_interrupt(vcpu);
9672 local_irq_disable();
9673 kvm_after_interrupt(vcpu);
9676 * Wait until after servicing IRQs to account guest time so that any
9677 * ticks that occurred while running the guest are properly accounted
9678 * to the guest. Waiting until IRQs are enabled degrades the accuracy
9679 * of accounting via context tracking, but the loss of accuracy is
9680 * acceptable for all known use cases.
9682 vtime_account_guest_exit();
9684 if (lapic_in_kernel(vcpu)) {
9685 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9686 if (delta != S64_MIN) {
9687 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9688 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9695 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9698 * Profile KVM exit RIPs:
9700 if (unlikely(prof_on == KVM_PROFILING)) {
9701 unsigned long rip = kvm_rip_read(vcpu);
9702 profile_hit(KVM_PROFILING, (void *)rip);
9705 if (unlikely(vcpu->arch.tsc_always_catchup))
9706 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9708 if (vcpu->arch.apic_attention)
9709 kvm_lapic_sync_from_vapic(vcpu);
9711 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9715 if (req_immediate_exit)
9716 kvm_make_request(KVM_REQ_EVENT, vcpu);
9717 static_call(kvm_x86_cancel_injection)(vcpu);
9718 if (unlikely(vcpu->arch.apic_attention))
9719 kvm_lapic_sync_from_vapic(vcpu);
9724 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9726 if (!kvm_arch_vcpu_runnable(vcpu) &&
9727 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9728 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9729 kvm_vcpu_block(vcpu);
9730 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9732 if (kvm_x86_ops.post_block)
9733 static_call(kvm_x86_post_block)(vcpu);
9735 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9739 if (kvm_apic_accept_events(vcpu) < 0)
9741 switch(vcpu->arch.mp_state) {
9742 case KVM_MP_STATE_HALTED:
9743 case KVM_MP_STATE_AP_RESET_HOLD:
9744 vcpu->arch.pv.pv_unhalted = false;
9745 vcpu->arch.mp_state =
9746 KVM_MP_STATE_RUNNABLE;
9748 case KVM_MP_STATE_RUNNABLE:
9749 vcpu->arch.apf.halted = false;
9751 case KVM_MP_STATE_INIT_RECEIVED:
9759 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9761 if (is_guest_mode(vcpu))
9762 kvm_check_nested_events(vcpu);
9764 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9765 !vcpu->arch.apf.halted);
9768 static int vcpu_run(struct kvm_vcpu *vcpu)
9771 struct kvm *kvm = vcpu->kvm;
9773 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9774 vcpu->arch.l1tf_flush_l1d = true;
9777 if (kvm_vcpu_running(vcpu)) {
9778 r = vcpu_enter_guest(vcpu);
9780 r = vcpu_block(kvm, vcpu);
9786 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9787 if (kvm_cpu_has_pending_timer(vcpu))
9788 kvm_inject_pending_timer_irqs(vcpu);
9790 if (dm_request_for_irq_injection(vcpu) &&
9791 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9793 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9794 ++vcpu->stat.request_irq_exits;
9798 if (__xfer_to_guest_mode_work_pending()) {
9799 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9800 r = xfer_to_guest_mode_handle_work(vcpu);
9803 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9807 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9812 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9816 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9817 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9818 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9822 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9824 BUG_ON(!vcpu->arch.pio.count);
9826 return complete_emulated_io(vcpu);
9830 * Implements the following, as a state machine:
9834 * for each mmio piece in the fragment
9842 * for each mmio piece in the fragment
9847 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9849 struct kvm_run *run = vcpu->run;
9850 struct kvm_mmio_fragment *frag;
9853 BUG_ON(!vcpu->mmio_needed);
9855 /* Complete previous fragment */
9856 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9857 len = min(8u, frag->len);
9858 if (!vcpu->mmio_is_write)
9859 memcpy(frag->data, run->mmio.data, len);
9861 if (frag->len <= 8) {
9862 /* Switch to the next fragment. */
9864 vcpu->mmio_cur_fragment++;
9866 /* Go forward to the next mmio piece. */
9872 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9873 vcpu->mmio_needed = 0;
9875 /* FIXME: return into emulator if single-stepping. */
9876 if (vcpu->mmio_is_write)
9878 vcpu->mmio_read_completed = 1;
9879 return complete_emulated_io(vcpu);
9882 run->exit_reason = KVM_EXIT_MMIO;
9883 run->mmio.phys_addr = frag->gpa;
9884 if (vcpu->mmio_is_write)
9885 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9886 run->mmio.len = min(8u, frag->len);
9887 run->mmio.is_write = vcpu->mmio_is_write;
9888 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9892 static void kvm_save_current_fpu(struct fpu *fpu)
9895 * If the target FPU state is not resident in the CPU registers, just
9896 * memcpy() from current, else save CPU state directly to the target.
9898 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9899 memcpy(&fpu->state, ¤t->thread.fpu.state,
9900 fpu_kernel_xstate_size);
9902 save_fpregs_to_fpstate(fpu);
9905 /* Swap (qemu) user FPU context for the guest FPU context. */
9906 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9910 kvm_save_current_fpu(vcpu->arch.user_fpu);
9913 * Guests with protected state can't have it set by the hypervisor,
9914 * so skip trying to set it.
9916 if (vcpu->arch.guest_fpu)
9917 /* PKRU is separately restored in kvm_x86_ops.run. */
9918 __restore_fpregs_from_fpstate(&vcpu->arch.guest_fpu->state,
9919 ~XFEATURE_MASK_PKRU);
9921 fpregs_mark_activate();
9927 /* When vcpu_run ends, restore user space FPU context. */
9928 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9933 * Guests with protected state can't have it read by the hypervisor,
9934 * so skip trying to save it.
9936 if (vcpu->arch.guest_fpu)
9937 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9939 restore_fpregs_from_fpstate(&vcpu->arch.user_fpu->state);
9941 fpregs_mark_activate();
9944 ++vcpu->stat.fpu_reload;
9948 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9950 struct kvm_run *kvm_run = vcpu->run;
9954 kvm_sigset_activate(vcpu);
9956 kvm_load_guest_fpu(vcpu);
9958 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9959 if (kvm_run->immediate_exit) {
9963 kvm_vcpu_block(vcpu);
9964 if (kvm_apic_accept_events(vcpu) < 0) {
9968 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9970 if (signal_pending(current)) {
9972 kvm_run->exit_reason = KVM_EXIT_INTR;
9973 ++vcpu->stat.signal_exits;
9978 if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
9979 (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
9984 if (kvm_run->kvm_dirty_regs) {
9985 r = sync_regs(vcpu);
9990 /* re-sync apic's tpr */
9991 if (!lapic_in_kernel(vcpu)) {
9992 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9998 if (unlikely(vcpu->arch.complete_userspace_io)) {
9999 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10000 vcpu->arch.complete_userspace_io = NULL;
10005 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10007 if (kvm_run->immediate_exit)
10010 r = vcpu_run(vcpu);
10013 kvm_put_guest_fpu(vcpu);
10014 if (kvm_run->kvm_valid_regs)
10016 post_kvm_run_save(vcpu);
10017 kvm_sigset_deactivate(vcpu);
10023 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10025 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10027 * We are here if userspace calls get_regs() in the middle of
10028 * instruction emulation. Registers state needs to be copied
10029 * back from emulation context to vcpu. Userspace shouldn't do
10030 * that usually, but some bad designed PV devices (vmware
10031 * backdoor interface) need this to work
10033 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10034 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10036 regs->rax = kvm_rax_read(vcpu);
10037 regs->rbx = kvm_rbx_read(vcpu);
10038 regs->rcx = kvm_rcx_read(vcpu);
10039 regs->rdx = kvm_rdx_read(vcpu);
10040 regs->rsi = kvm_rsi_read(vcpu);
10041 regs->rdi = kvm_rdi_read(vcpu);
10042 regs->rsp = kvm_rsp_read(vcpu);
10043 regs->rbp = kvm_rbp_read(vcpu);
10044 #ifdef CONFIG_X86_64
10045 regs->r8 = kvm_r8_read(vcpu);
10046 regs->r9 = kvm_r9_read(vcpu);
10047 regs->r10 = kvm_r10_read(vcpu);
10048 regs->r11 = kvm_r11_read(vcpu);
10049 regs->r12 = kvm_r12_read(vcpu);
10050 regs->r13 = kvm_r13_read(vcpu);
10051 regs->r14 = kvm_r14_read(vcpu);
10052 regs->r15 = kvm_r15_read(vcpu);
10055 regs->rip = kvm_rip_read(vcpu);
10056 regs->rflags = kvm_get_rflags(vcpu);
10059 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10062 __get_regs(vcpu, regs);
10067 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10069 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10070 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10072 kvm_rax_write(vcpu, regs->rax);
10073 kvm_rbx_write(vcpu, regs->rbx);
10074 kvm_rcx_write(vcpu, regs->rcx);
10075 kvm_rdx_write(vcpu, regs->rdx);
10076 kvm_rsi_write(vcpu, regs->rsi);
10077 kvm_rdi_write(vcpu, regs->rdi);
10078 kvm_rsp_write(vcpu, regs->rsp);
10079 kvm_rbp_write(vcpu, regs->rbp);
10080 #ifdef CONFIG_X86_64
10081 kvm_r8_write(vcpu, regs->r8);
10082 kvm_r9_write(vcpu, regs->r9);
10083 kvm_r10_write(vcpu, regs->r10);
10084 kvm_r11_write(vcpu, regs->r11);
10085 kvm_r12_write(vcpu, regs->r12);
10086 kvm_r13_write(vcpu, regs->r13);
10087 kvm_r14_write(vcpu, regs->r14);
10088 kvm_r15_write(vcpu, regs->r15);
10091 kvm_rip_write(vcpu, regs->rip);
10092 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10094 vcpu->arch.exception.pending = false;
10096 kvm_make_request(KVM_REQ_EVENT, vcpu);
10099 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10102 __set_regs(vcpu, regs);
10107 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10109 struct kvm_segment cs;
10111 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10115 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10117 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10119 struct desc_ptr dt;
10121 if (vcpu->arch.guest_state_protected)
10122 goto skip_protected_regs;
10124 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10125 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10126 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10127 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10128 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10129 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10131 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10132 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10134 static_call(kvm_x86_get_idt)(vcpu, &dt);
10135 sregs->idt.limit = dt.size;
10136 sregs->idt.base = dt.address;
10137 static_call(kvm_x86_get_gdt)(vcpu, &dt);
10138 sregs->gdt.limit = dt.size;
10139 sregs->gdt.base = dt.address;
10141 sregs->cr2 = vcpu->arch.cr2;
10142 sregs->cr3 = kvm_read_cr3(vcpu);
10144 skip_protected_regs:
10145 sregs->cr0 = kvm_read_cr0(vcpu);
10146 sregs->cr4 = kvm_read_cr4(vcpu);
10147 sregs->cr8 = kvm_get_cr8(vcpu);
10148 sregs->efer = vcpu->arch.efer;
10149 sregs->apic_base = kvm_get_apic_base(vcpu);
10152 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10154 __get_sregs_common(vcpu, sregs);
10156 if (vcpu->arch.guest_state_protected)
10159 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10160 set_bit(vcpu->arch.interrupt.nr,
10161 (unsigned long *)sregs->interrupt_bitmap);
10164 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10168 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10170 if (vcpu->arch.guest_state_protected)
10173 if (is_pae_paging(vcpu)) {
10174 for (i = 0 ; i < 4 ; i++)
10175 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10176 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10180 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10181 struct kvm_sregs *sregs)
10184 __get_sregs(vcpu, sregs);
10189 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10190 struct kvm_mp_state *mp_state)
10195 if (kvm_mpx_supported())
10196 kvm_load_guest_fpu(vcpu);
10198 r = kvm_apic_accept_events(vcpu);
10203 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10204 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10205 vcpu->arch.pv.pv_unhalted)
10206 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10208 mp_state->mp_state = vcpu->arch.mp_state;
10211 if (kvm_mpx_supported())
10212 kvm_put_guest_fpu(vcpu);
10217 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10218 struct kvm_mp_state *mp_state)
10224 if (!lapic_in_kernel(vcpu) &&
10225 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10229 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10230 * INIT state; latched init should be reported using
10231 * KVM_SET_VCPU_EVENTS, so reject it here.
10233 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10234 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10235 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10238 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10239 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10240 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10242 vcpu->arch.mp_state = mp_state->mp_state;
10243 kvm_make_request(KVM_REQ_EVENT, vcpu);
10251 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10252 int reason, bool has_error_code, u32 error_code)
10254 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10257 init_emulate_ctxt(vcpu);
10259 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10260 has_error_code, error_code);
10262 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10263 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10264 vcpu->run->internal.ndata = 0;
10268 kvm_rip_write(vcpu, ctxt->eip);
10269 kvm_set_rflags(vcpu, ctxt->eflags);
10272 EXPORT_SYMBOL_GPL(kvm_task_switch);
10274 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10276 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10278 * When EFER.LME and CR0.PG are set, the processor is in
10279 * 64-bit mode (though maybe in a 32-bit code segment).
10280 * CR4.PAE and EFER.LMA must be set.
10282 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10284 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10288 * Not in 64-bit mode: EFER.LMA is clear and the code
10289 * segment cannot be 64-bit.
10291 if (sregs->efer & EFER_LMA || sregs->cs.l)
10295 return kvm_is_valid_cr4(vcpu, sregs->cr4);
10298 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10299 int *mmu_reset_needed, bool update_pdptrs)
10301 struct msr_data apic_base_msr;
10303 struct desc_ptr dt;
10305 if (!kvm_is_valid_sregs(vcpu, sregs))
10308 apic_base_msr.data = sregs->apic_base;
10309 apic_base_msr.host_initiated = true;
10310 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10313 if (vcpu->arch.guest_state_protected)
10316 dt.size = sregs->idt.limit;
10317 dt.address = sregs->idt.base;
10318 static_call(kvm_x86_set_idt)(vcpu, &dt);
10319 dt.size = sregs->gdt.limit;
10320 dt.address = sregs->gdt.base;
10321 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10323 vcpu->arch.cr2 = sregs->cr2;
10324 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10325 vcpu->arch.cr3 = sregs->cr3;
10326 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10328 kvm_set_cr8(vcpu, sregs->cr8);
10330 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10331 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10333 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10334 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10335 vcpu->arch.cr0 = sregs->cr0;
10337 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10338 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10340 if (update_pdptrs) {
10341 idx = srcu_read_lock(&vcpu->kvm->srcu);
10342 if (is_pae_paging(vcpu)) {
10343 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10344 *mmu_reset_needed = 1;
10346 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10349 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10350 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10351 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10352 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10353 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10354 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10356 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10357 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10359 update_cr8_intercept(vcpu);
10361 /* Older userspace won't unhalt the vcpu on reset. */
10362 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10363 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10364 !is_protmode(vcpu))
10365 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10370 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10372 int pending_vec, max_bits;
10373 int mmu_reset_needed = 0;
10374 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10379 if (mmu_reset_needed)
10380 kvm_mmu_reset_context(vcpu);
10382 max_bits = KVM_NR_INTERRUPTS;
10383 pending_vec = find_first_bit(
10384 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10386 if (pending_vec < max_bits) {
10387 kvm_queue_interrupt(vcpu, pending_vec, false);
10388 pr_debug("Set back pending irq %d\n", pending_vec);
10389 kvm_make_request(KVM_REQ_EVENT, vcpu);
10394 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10396 int mmu_reset_needed = 0;
10397 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10398 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10399 !(sregs2->efer & EFER_LMA);
10402 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10405 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10408 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10409 &mmu_reset_needed, !valid_pdptrs);
10413 if (valid_pdptrs) {
10414 for (i = 0; i < 4 ; i++)
10415 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10417 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10418 mmu_reset_needed = 1;
10419 vcpu->arch.pdptrs_from_userspace = true;
10421 if (mmu_reset_needed)
10422 kvm_mmu_reset_context(vcpu);
10426 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10427 struct kvm_sregs *sregs)
10432 ret = __set_sregs(vcpu, sregs);
10437 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10438 struct kvm_guest_debug *dbg)
10440 unsigned long rflags;
10443 if (vcpu->arch.guest_state_protected)
10448 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10450 if (vcpu->arch.exception.pending)
10452 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10453 kvm_queue_exception(vcpu, DB_VECTOR);
10455 kvm_queue_exception(vcpu, BP_VECTOR);
10459 * Read rflags as long as potentially injected trace flags are still
10462 rflags = kvm_get_rflags(vcpu);
10464 vcpu->guest_debug = dbg->control;
10465 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10466 vcpu->guest_debug = 0;
10468 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10469 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10470 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10471 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10473 for (i = 0; i < KVM_NR_DB_REGS; i++)
10474 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10476 kvm_update_dr7(vcpu);
10478 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10479 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10482 * Trigger an rflags update that will inject or remove the trace
10485 kvm_set_rflags(vcpu, rflags);
10487 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10497 * Translate a guest virtual address to a guest physical address.
10499 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10500 struct kvm_translation *tr)
10502 unsigned long vaddr = tr->linear_address;
10508 idx = srcu_read_lock(&vcpu->kvm->srcu);
10509 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10510 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10511 tr->physical_address = gpa;
10512 tr->valid = gpa != UNMAPPED_GVA;
10520 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10522 struct fxregs_state *fxsave;
10524 if (!vcpu->arch.guest_fpu)
10529 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10530 memcpy(fpu->fpr, fxsave->st_space, 128);
10531 fpu->fcw = fxsave->cwd;
10532 fpu->fsw = fxsave->swd;
10533 fpu->ftwx = fxsave->twd;
10534 fpu->last_opcode = fxsave->fop;
10535 fpu->last_ip = fxsave->rip;
10536 fpu->last_dp = fxsave->rdp;
10537 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10543 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10545 struct fxregs_state *fxsave;
10547 if (!vcpu->arch.guest_fpu)
10552 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10554 memcpy(fxsave->st_space, fpu->fpr, 128);
10555 fxsave->cwd = fpu->fcw;
10556 fxsave->swd = fpu->fsw;
10557 fxsave->twd = fpu->ftwx;
10558 fxsave->fop = fpu->last_opcode;
10559 fxsave->rip = fpu->last_ip;
10560 fxsave->rdp = fpu->last_dp;
10561 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10567 static void store_regs(struct kvm_vcpu *vcpu)
10569 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10571 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10572 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10574 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10575 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10577 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10578 kvm_vcpu_ioctl_x86_get_vcpu_events(
10579 vcpu, &vcpu->run->s.regs.events);
10582 static int sync_regs(struct kvm_vcpu *vcpu)
10584 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10585 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10586 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10588 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10589 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10591 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10593 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10594 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10595 vcpu, &vcpu->run->s.regs.events))
10597 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10603 static void fx_init(struct kvm_vcpu *vcpu)
10605 if (!vcpu->arch.guest_fpu)
10608 fpstate_init(&vcpu->arch.guest_fpu->state);
10609 if (boot_cpu_has(X86_FEATURE_XSAVES))
10610 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10611 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10614 * Ensure guest xcr0 is valid for loading
10616 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10618 vcpu->arch.cr0 |= X86_CR0_ET;
10621 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10623 if (vcpu->arch.guest_fpu) {
10624 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10625 vcpu->arch.guest_fpu = NULL;
10628 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10630 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10632 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10633 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10634 "guest TSC will not be reliable\n");
10639 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10644 vcpu->arch.last_vmentry_cpu = -1;
10646 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10647 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10649 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10651 r = kvm_mmu_create(vcpu);
10655 if (irqchip_in_kernel(vcpu->kvm)) {
10656 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10658 goto fail_mmu_destroy;
10659 if (kvm_apicv_activated(vcpu->kvm))
10660 vcpu->arch.apicv_active = true;
10662 static_branch_inc(&kvm_has_noapic_vcpu);
10666 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10668 goto fail_free_lapic;
10669 vcpu->arch.pio_data = page_address(page);
10671 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10672 GFP_KERNEL_ACCOUNT);
10673 if (!vcpu->arch.mce_banks)
10674 goto fail_free_pio_data;
10675 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10677 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10678 GFP_KERNEL_ACCOUNT))
10679 goto fail_free_mce_banks;
10681 if (!alloc_emulate_ctxt(vcpu))
10682 goto free_wbinvd_dirty_mask;
10684 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10685 GFP_KERNEL_ACCOUNT);
10686 if (!vcpu->arch.user_fpu) {
10687 pr_err("kvm: failed to allocate userspace's fpu\n");
10688 goto free_emulate_ctxt;
10691 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10692 GFP_KERNEL_ACCOUNT);
10693 if (!vcpu->arch.guest_fpu) {
10694 pr_err("kvm: failed to allocate vcpu's fpu\n");
10695 goto free_user_fpu;
10699 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10700 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10702 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10704 kvm_async_pf_hash_reset(vcpu);
10705 kvm_pmu_init(vcpu);
10707 vcpu->arch.pending_external_vector = -1;
10708 vcpu->arch.preempted_in_kernel = false;
10710 #if IS_ENABLED(CONFIG_HYPERV)
10711 vcpu->arch.hv_root_tdp = INVALID_PAGE;
10714 r = static_call(kvm_x86_vcpu_create)(vcpu);
10716 goto free_guest_fpu;
10718 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10719 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10720 kvm_vcpu_mtrr_init(vcpu);
10722 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10723 kvm_vcpu_reset(vcpu, false);
10724 kvm_init_mmu(vcpu);
10729 kvm_free_guest_fpu(vcpu);
10731 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10733 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10734 free_wbinvd_dirty_mask:
10735 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10736 fail_free_mce_banks:
10737 kfree(vcpu->arch.mce_banks);
10738 fail_free_pio_data:
10739 free_page((unsigned long)vcpu->arch.pio_data);
10741 kvm_free_lapic(vcpu);
10743 kvm_mmu_destroy(vcpu);
10747 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10749 struct kvm *kvm = vcpu->kvm;
10751 if (mutex_lock_killable(&vcpu->mutex))
10754 kvm_synchronize_tsc(vcpu, 0);
10757 /* poll control enabled by default */
10758 vcpu->arch.msr_kvm_poll_control = 1;
10760 mutex_unlock(&vcpu->mutex);
10762 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10763 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10764 KVMCLOCK_SYNC_PERIOD);
10767 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10769 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10772 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10774 kvmclock_reset(vcpu);
10776 static_call(kvm_x86_vcpu_free)(vcpu);
10778 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10779 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10780 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10781 kvm_free_guest_fpu(vcpu);
10783 kvm_hv_vcpu_uninit(vcpu);
10784 kvm_pmu_destroy(vcpu);
10785 kfree(vcpu->arch.mce_banks);
10786 kvm_free_lapic(vcpu);
10787 idx = srcu_read_lock(&vcpu->kvm->srcu);
10788 kvm_mmu_destroy(vcpu);
10789 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10790 free_page((unsigned long)vcpu->arch.pio_data);
10791 kvfree(vcpu->arch.cpuid_entries);
10792 if (!lapic_in_kernel(vcpu))
10793 static_branch_dec(&kvm_has_noapic_vcpu);
10796 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10798 unsigned long old_cr0 = kvm_read_cr0(vcpu);
10799 unsigned long new_cr0;
10802 kvm_lapic_reset(vcpu, init_event);
10804 vcpu->arch.hflags = 0;
10806 vcpu->arch.smi_pending = 0;
10807 vcpu->arch.smi_count = 0;
10808 atomic_set(&vcpu->arch.nmi_queued, 0);
10809 vcpu->arch.nmi_pending = 0;
10810 vcpu->arch.nmi_injected = false;
10811 kvm_clear_interrupt_queue(vcpu);
10812 kvm_clear_exception_queue(vcpu);
10814 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10815 kvm_update_dr0123(vcpu);
10816 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10817 vcpu->arch.dr7 = DR7_FIXED_1;
10818 kvm_update_dr7(vcpu);
10820 vcpu->arch.cr2 = 0;
10822 kvm_make_request(KVM_REQ_EVENT, vcpu);
10823 vcpu->arch.apf.msr_en_val = 0;
10824 vcpu->arch.apf.msr_int_val = 0;
10825 vcpu->arch.st.msr_val = 0;
10827 kvmclock_reset(vcpu);
10829 kvm_clear_async_pf_completion_queue(vcpu);
10830 kvm_async_pf_hash_reset(vcpu);
10831 vcpu->arch.apf.halted = false;
10833 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10834 void *mpx_state_buffer;
10837 * To avoid have the INIT path from kvm_apic_has_events() that be
10838 * called with loaded FPU and does not let userspace fix the state.
10841 kvm_put_guest_fpu(vcpu);
10842 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10844 if (mpx_state_buffer)
10845 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
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_bndcsr));
10851 kvm_load_guest_fpu(vcpu);
10855 kvm_pmu_reset(vcpu);
10856 vcpu->arch.smbase = 0x30000;
10858 vcpu->arch.msr_misc_features_enables = 0;
10860 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10863 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10864 vcpu->arch.regs_avail = ~0;
10865 vcpu->arch.regs_dirty = ~0;
10868 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
10869 * if no CPUID match is found. Note, it's impossible to get a match at
10870 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
10871 * i.e. it'simpossible for kvm_cpuid() to find a valid entry on RESET.
10872 * But, go through the motions in case that's ever remedied.
10875 if (!kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true))
10877 kvm_rdx_write(vcpu, eax);
10879 vcpu->arch.ia32_xss = 0;
10881 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10883 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
10884 kvm_rip_write(vcpu, 0xfff0);
10887 * CR0.CD/NW are set on RESET, preserved on INIT. Note, some versions
10888 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
10889 * (or qualify) that with a footnote stating that CD/NW are preserved.
10891 new_cr0 = X86_CR0_ET;
10893 new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
10895 new_cr0 |= X86_CR0_NW | X86_CR0_CD;
10897 static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
10898 static_call(kvm_x86_set_cr4)(vcpu, 0);
10899 static_call(kvm_x86_set_efer)(vcpu, 0);
10900 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10903 * Reset the MMU context if paging was enabled prior to INIT (which is
10904 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET). Unlike the
10905 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
10906 * checked because it is unconditionally cleared on INIT and all other
10907 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
10908 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
10910 if (old_cr0 & X86_CR0_PG)
10911 kvm_mmu_reset_context(vcpu);
10914 * Intel's SDM states that all TLB entries are flushed on INIT. AMD's
10915 * APM states the TLBs are untouched by INIT, but it also states that
10916 * the TLBs are flushed on "External initialization of the processor."
10917 * Flush the guest TLB regardless of vendor, there is no meaningful
10918 * benefit in relying on the guest to flush the TLB immediately after
10919 * INIT. A spurious TLB flush is benign and likely negligible from a
10920 * performance perspective.
10923 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
10925 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
10927 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10929 struct kvm_segment cs;
10931 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10932 cs.selector = vector << 8;
10933 cs.base = vector << 12;
10934 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10935 kvm_rip_write(vcpu, 0);
10937 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10939 int kvm_arch_hardware_enable(void)
10942 struct kvm_vcpu *vcpu;
10947 bool stable, backwards_tsc = false;
10949 kvm_user_return_msr_cpu_online();
10950 ret = static_call(kvm_x86_hardware_enable)();
10954 local_tsc = rdtsc();
10955 stable = !kvm_check_tsc_unstable();
10956 list_for_each_entry(kvm, &vm_list, vm_list) {
10957 kvm_for_each_vcpu(i, vcpu, kvm) {
10958 if (!stable && vcpu->cpu == smp_processor_id())
10959 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10960 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10961 backwards_tsc = true;
10962 if (vcpu->arch.last_host_tsc > max_tsc)
10963 max_tsc = vcpu->arch.last_host_tsc;
10969 * Sometimes, even reliable TSCs go backwards. This happens on
10970 * platforms that reset TSC during suspend or hibernate actions, but
10971 * maintain synchronization. We must compensate. Fortunately, we can
10972 * detect that condition here, which happens early in CPU bringup,
10973 * before any KVM threads can be running. Unfortunately, we can't
10974 * bring the TSCs fully up to date with real time, as we aren't yet far
10975 * enough into CPU bringup that we know how much real time has actually
10976 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10977 * variables that haven't been updated yet.
10979 * So we simply find the maximum observed TSC above, then record the
10980 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10981 * the adjustment will be applied. Note that we accumulate
10982 * adjustments, in case multiple suspend cycles happen before some VCPU
10983 * gets a chance to run again. In the event that no KVM threads get a
10984 * chance to run, we will miss the entire elapsed period, as we'll have
10985 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10986 * loose cycle time. This isn't too big a deal, since the loss will be
10987 * uniform across all VCPUs (not to mention the scenario is extremely
10988 * unlikely). It is possible that a second hibernate recovery happens
10989 * much faster than a first, causing the observed TSC here to be
10990 * smaller; this would require additional padding adjustment, which is
10991 * why we set last_host_tsc to the local tsc observed here.
10993 * N.B. - this code below runs only on platforms with reliable TSC,
10994 * as that is the only way backwards_tsc is set above. Also note
10995 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10996 * have the same delta_cyc adjustment applied if backwards_tsc
10997 * is detected. Note further, this adjustment is only done once,
10998 * as we reset last_host_tsc on all VCPUs to stop this from being
10999 * called multiple times (one for each physical CPU bringup).
11001 * Platforms with unreliable TSCs don't have to deal with this, they
11002 * will be compensated by the logic in vcpu_load, which sets the TSC to
11003 * catchup mode. This will catchup all VCPUs to real time, but cannot
11004 * guarantee that they stay in perfect synchronization.
11006 if (backwards_tsc) {
11007 u64 delta_cyc = max_tsc - local_tsc;
11008 list_for_each_entry(kvm, &vm_list, vm_list) {
11009 kvm->arch.backwards_tsc_observed = true;
11010 kvm_for_each_vcpu(i, vcpu, kvm) {
11011 vcpu->arch.tsc_offset_adjustment += delta_cyc;
11012 vcpu->arch.last_host_tsc = local_tsc;
11013 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11017 * We have to disable TSC offset matching.. if you were
11018 * booting a VM while issuing an S4 host suspend....
11019 * you may have some problem. Solving this issue is
11020 * left as an exercise to the reader.
11022 kvm->arch.last_tsc_nsec = 0;
11023 kvm->arch.last_tsc_write = 0;
11030 void kvm_arch_hardware_disable(void)
11032 static_call(kvm_x86_hardware_disable)();
11033 drop_user_return_notifiers();
11036 int kvm_arch_hardware_setup(void *opaque)
11038 struct kvm_x86_init_ops *ops = opaque;
11041 rdmsrl_safe(MSR_EFER, &host_efer);
11043 if (boot_cpu_has(X86_FEATURE_XSAVES))
11044 rdmsrl(MSR_IA32_XSS, host_xss);
11046 r = ops->hardware_setup();
11050 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11051 kvm_ops_static_call_update();
11053 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11056 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11057 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11058 #undef __kvm_cpu_cap_has
11060 if (kvm_has_tsc_control) {
11062 * Make sure the user can only configure tsc_khz values that
11063 * fit into a signed integer.
11064 * A min value is not calculated because it will always
11065 * be 1 on all machines.
11067 u64 max = min(0x7fffffffULL,
11068 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11069 kvm_max_guest_tsc_khz = max;
11071 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11074 kvm_init_msr_list();
11078 void kvm_arch_hardware_unsetup(void)
11080 static_call(kvm_x86_hardware_unsetup)();
11083 int kvm_arch_check_processor_compat(void *opaque)
11085 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11086 struct kvm_x86_init_ops *ops = opaque;
11088 WARN_ON(!irqs_disabled());
11090 if (__cr4_reserved_bits(cpu_has, c) !=
11091 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11094 return ops->check_processor_compatibility();
11097 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11099 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11101 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11103 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11105 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11108 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11109 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11111 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11113 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11115 vcpu->arch.l1tf_flush_l1d = true;
11116 if (pmu->version && unlikely(pmu->event_count)) {
11117 pmu->need_cleanup = true;
11118 kvm_make_request(KVM_REQ_PMU, vcpu);
11120 static_call(kvm_x86_sched_in)(vcpu, cpu);
11123 void kvm_arch_free_vm(struct kvm *kvm)
11125 kfree(to_kvm_hv(kvm)->hv_pa_pg);
11130 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11135 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11136 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11137 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11138 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11139 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11140 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11142 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11143 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11144 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11145 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11146 &kvm->arch.irq_sources_bitmap);
11148 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11149 mutex_init(&kvm->arch.apic_map_lock);
11150 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
11152 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11153 pvclock_update_vm_gtod_copy(kvm);
11155 kvm->arch.guest_can_read_msr_platform_info = true;
11157 #if IS_ENABLED(CONFIG_HYPERV)
11158 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11159 kvm->arch.hv_root_tdp = INVALID_PAGE;
11162 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11163 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11165 kvm_apicv_init(kvm);
11166 kvm_hv_init_vm(kvm);
11167 kvm_page_track_init(kvm);
11168 kvm_mmu_init_vm(kvm);
11169 kvm_xen_init_vm(kvm);
11171 return static_call(kvm_x86_vm_init)(kvm);
11174 int kvm_arch_post_init_vm(struct kvm *kvm)
11176 return kvm_mmu_post_init_vm(kvm);
11179 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11182 kvm_mmu_unload(vcpu);
11186 static void kvm_free_vcpus(struct kvm *kvm)
11189 struct kvm_vcpu *vcpu;
11192 * Unpin any mmu pages first.
11194 kvm_for_each_vcpu(i, vcpu, kvm) {
11195 kvm_clear_async_pf_completion_queue(vcpu);
11196 kvm_unload_vcpu_mmu(vcpu);
11198 kvm_for_each_vcpu(i, vcpu, kvm)
11199 kvm_vcpu_destroy(vcpu);
11201 mutex_lock(&kvm->lock);
11202 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11203 kvm->vcpus[i] = NULL;
11205 atomic_set(&kvm->online_vcpus, 0);
11206 mutex_unlock(&kvm->lock);
11209 void kvm_arch_sync_events(struct kvm *kvm)
11211 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11212 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11216 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
11219 * __x86_set_memory_region: Setup KVM internal memory slot
11221 * @kvm: the kvm pointer to the VM.
11222 * @id: the slot ID to setup.
11223 * @gpa: the GPA to install the slot (unused when @size == 0).
11224 * @size: the size of the slot. Set to zero to uninstall a slot.
11226 * This function helps to setup a KVM internal memory slot. Specify
11227 * @size > 0 to install a new slot, while @size == 0 to uninstall a
11228 * slot. The return code can be one of the following:
11230 * HVA: on success (uninstall will return a bogus HVA)
11233 * The caller should always use IS_ERR() to check the return value
11234 * before use. Note, the KVM internal memory slots are guaranteed to
11235 * remain valid and unchanged until the VM is destroyed, i.e., the
11236 * GPA->HVA translation will not change. However, the HVA is a user
11237 * address, i.e. its accessibility is not guaranteed, and must be
11238 * accessed via __copy_{to,from}_user().
11240 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11244 unsigned long hva, old_npages;
11245 struct kvm_memslots *slots = kvm_memslots(kvm);
11246 struct kvm_memory_slot *slot;
11248 /* Called with kvm->slots_lock held. */
11249 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11250 return ERR_PTR_USR(-EINVAL);
11252 slot = id_to_memslot(slots, id);
11254 if (slot && slot->npages)
11255 return ERR_PTR_USR(-EEXIST);
11258 * MAP_SHARED to prevent internal slot pages from being moved
11261 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11262 MAP_SHARED | MAP_ANONYMOUS, 0);
11263 if (IS_ERR((void *)hva))
11264 return (void __user *)hva;
11266 if (!slot || !slot->npages)
11269 old_npages = slot->npages;
11270 hva = slot->userspace_addr;
11273 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11274 struct kvm_userspace_memory_region m;
11276 m.slot = id | (i << 16);
11278 m.guest_phys_addr = gpa;
11279 m.userspace_addr = hva;
11280 m.memory_size = size;
11281 r = __kvm_set_memory_region(kvm, &m);
11283 return ERR_PTR_USR(r);
11287 vm_munmap(hva, old_npages * PAGE_SIZE);
11289 return (void __user *)hva;
11291 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11293 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11295 kvm_mmu_pre_destroy_vm(kvm);
11298 void kvm_arch_destroy_vm(struct kvm *kvm)
11300 if (current->mm == kvm->mm) {
11302 * Free memory regions allocated on behalf of userspace,
11303 * unless the the memory map has changed due to process exit
11306 mutex_lock(&kvm->slots_lock);
11307 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11309 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11311 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11312 mutex_unlock(&kvm->slots_lock);
11314 static_call_cond(kvm_x86_vm_destroy)(kvm);
11315 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11316 kvm_pic_destroy(kvm);
11317 kvm_ioapic_destroy(kvm);
11318 kvm_free_vcpus(kvm);
11319 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11320 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11321 kvm_mmu_uninit_vm(kvm);
11322 kvm_page_track_cleanup(kvm);
11323 kvm_xen_destroy_vm(kvm);
11324 kvm_hv_destroy_vm(kvm);
11327 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11331 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11332 kvfree(slot->arch.rmap[i]);
11333 slot->arch.rmap[i] = NULL;
11337 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11341 memslot_rmap_free(slot);
11343 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11344 kvfree(slot->arch.lpage_info[i - 1]);
11345 slot->arch.lpage_info[i - 1] = NULL;
11348 kvm_page_track_free_memslot(slot);
11351 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
11352 unsigned long npages)
11354 const int sz = sizeof(*slot->arch.rmap[0]);
11357 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11359 int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11361 WARN_ON(slot->arch.rmap[i]);
11363 slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11364 if (!slot->arch.rmap[i]) {
11365 memslot_rmap_free(slot);
11373 int alloc_all_memslots_rmaps(struct kvm *kvm)
11375 struct kvm_memslots *slots;
11376 struct kvm_memory_slot *slot;
11380 * Check if memslots alreday have rmaps early before acquiring
11381 * the slots_arch_lock below.
11383 if (kvm_memslots_have_rmaps(kvm))
11386 mutex_lock(&kvm->slots_arch_lock);
11389 * Read memslots_have_rmaps again, under the slots arch lock,
11390 * before allocating the rmaps
11392 if (kvm_memslots_have_rmaps(kvm)) {
11393 mutex_unlock(&kvm->slots_arch_lock);
11397 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11398 slots = __kvm_memslots(kvm, i);
11399 kvm_for_each_memslot(slot, slots) {
11400 r = memslot_rmap_alloc(slot, slot->npages);
11402 mutex_unlock(&kvm->slots_arch_lock);
11409 * Ensure that memslots_have_rmaps becomes true strictly after
11410 * all the rmap pointers are set.
11412 smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11413 mutex_unlock(&kvm->slots_arch_lock);
11417 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11418 struct kvm_memory_slot *slot,
11419 unsigned long npages)
11424 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
11425 * old arrays will be freed by __kvm_set_memory_region() if installing
11426 * the new memslot is successful.
11428 memset(&slot->arch, 0, sizeof(slot->arch));
11430 if (kvm_memslots_have_rmaps(kvm)) {
11431 r = memslot_rmap_alloc(slot, npages);
11436 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11437 struct kvm_lpage_info *linfo;
11438 unsigned long ugfn;
11442 lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11444 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11448 slot->arch.lpage_info[i - 1] = linfo;
11450 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11451 linfo[0].disallow_lpage = 1;
11452 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11453 linfo[lpages - 1].disallow_lpage = 1;
11454 ugfn = slot->userspace_addr >> PAGE_SHIFT;
11456 * If the gfn and userspace address are not aligned wrt each
11457 * other, disable large page support for this slot.
11459 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11462 for (j = 0; j < lpages; ++j)
11463 linfo[j].disallow_lpage = 1;
11467 if (kvm_page_track_create_memslot(slot, npages))
11473 memslot_rmap_free(slot);
11475 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11476 kvfree(slot->arch.lpage_info[i - 1]);
11477 slot->arch.lpage_info[i - 1] = NULL;
11482 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11484 struct kvm_vcpu *vcpu;
11488 * memslots->generation has been incremented.
11489 * mmio generation may have reached its maximum value.
11491 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11493 /* Force re-initialization of steal_time cache */
11494 kvm_for_each_vcpu(i, vcpu, kvm)
11495 kvm_vcpu_kick(vcpu);
11498 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11499 struct kvm_memory_slot *memslot,
11500 const struct kvm_userspace_memory_region *mem,
11501 enum kvm_mr_change change)
11503 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11504 return kvm_alloc_memslot_metadata(kvm, memslot,
11505 mem->memory_size >> PAGE_SHIFT);
11510 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11512 struct kvm_arch *ka = &kvm->arch;
11514 if (!kvm_x86_ops.cpu_dirty_log_size)
11517 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11518 (!enable && --ka->cpu_dirty_logging_count == 0))
11519 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11521 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11524 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11525 struct kvm_memory_slot *old,
11526 const struct kvm_memory_slot *new,
11527 enum kvm_mr_change change)
11529 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11532 * Update CPU dirty logging if dirty logging is being toggled. This
11533 * applies to all operations.
11535 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11536 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11539 * Nothing more to do for RO slots (which can't be dirtied and can't be
11540 * made writable) or CREATE/MOVE/DELETE of a slot.
11542 * For a memslot with dirty logging disabled:
11543 * CREATE: No dirty mappings will already exist.
11544 * MOVE/DELETE: The old mappings will already have been cleaned up by
11545 * kvm_arch_flush_shadow_memslot()
11547 * For a memslot with dirty logging enabled:
11548 * CREATE: No shadow pages exist, thus nothing to write-protect
11549 * and no dirty bits to clear.
11550 * MOVE/DELETE: The old mappings will already have been cleaned up by
11551 * kvm_arch_flush_shadow_memslot().
11553 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11557 * READONLY and non-flags changes were filtered out above, and the only
11558 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11559 * logging isn't being toggled on or off.
11561 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11564 if (!log_dirty_pages) {
11566 * Dirty logging tracks sptes in 4k granularity, meaning that
11567 * large sptes have to be split. If live migration succeeds,
11568 * the guest in the source machine will be destroyed and large
11569 * sptes will be created in the destination. However, if the
11570 * guest continues to run in the source machine (for example if
11571 * live migration fails), small sptes will remain around and
11572 * cause bad performance.
11574 * Scan sptes if dirty logging has been stopped, dropping those
11575 * which can be collapsed into a single large-page spte. Later
11576 * page faults will create the large-page sptes.
11578 kvm_mmu_zap_collapsible_sptes(kvm, new);
11581 * Initially-all-set does not require write protecting any page,
11582 * because they're all assumed to be dirty.
11584 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11587 if (kvm_x86_ops.cpu_dirty_log_size) {
11588 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11589 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11591 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11596 void kvm_arch_commit_memory_region(struct kvm *kvm,
11597 const struct kvm_userspace_memory_region *mem,
11598 struct kvm_memory_slot *old,
11599 const struct kvm_memory_slot *new,
11600 enum kvm_mr_change change)
11602 if (!kvm->arch.n_requested_mmu_pages)
11603 kvm_mmu_change_mmu_pages(kvm,
11604 kvm_mmu_calculate_default_mmu_pages(kvm));
11606 kvm_mmu_slot_apply_flags(kvm, old, new, change);
11608 /* Free the arrays associated with the old memslot. */
11609 if (change == KVM_MR_MOVE)
11610 kvm_arch_free_memslot(kvm, old);
11613 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11615 kvm_mmu_zap_all(kvm);
11618 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11619 struct kvm_memory_slot *slot)
11621 kvm_page_track_flush_slot(kvm, slot);
11624 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11626 return (is_guest_mode(vcpu) &&
11627 kvm_x86_ops.guest_apic_has_interrupt &&
11628 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11631 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11633 if (!list_empty_careful(&vcpu->async_pf.done))
11636 if (kvm_apic_has_events(vcpu))
11639 if (vcpu->arch.pv.pv_unhalted)
11642 if (vcpu->arch.exception.pending)
11645 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11646 (vcpu->arch.nmi_pending &&
11647 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11650 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11651 (vcpu->arch.smi_pending &&
11652 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11655 if (kvm_arch_interrupt_allowed(vcpu) &&
11656 (kvm_cpu_has_interrupt(vcpu) ||
11657 kvm_guest_apic_has_interrupt(vcpu)))
11660 if (kvm_hv_has_stimer_pending(vcpu))
11663 if (is_guest_mode(vcpu) &&
11664 kvm_x86_ops.nested_ops->hv_timer_pending &&
11665 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11671 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11673 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11676 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11678 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11684 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11686 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11689 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11690 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11691 kvm_test_request(KVM_REQ_EVENT, vcpu))
11694 return kvm_arch_dy_has_pending_interrupt(vcpu);
11697 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11699 if (vcpu->arch.guest_state_protected)
11702 return vcpu->arch.preempted_in_kernel;
11705 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11707 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11710 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11712 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11715 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11717 /* Can't read the RIP when guest state is protected, just return 0 */
11718 if (vcpu->arch.guest_state_protected)
11721 if (is_64_bit_mode(vcpu))
11722 return kvm_rip_read(vcpu);
11723 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11724 kvm_rip_read(vcpu));
11726 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11728 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11730 return kvm_get_linear_rip(vcpu) == linear_rip;
11732 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11734 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11736 unsigned long rflags;
11738 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11739 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11740 rflags &= ~X86_EFLAGS_TF;
11743 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11745 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11747 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11748 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11749 rflags |= X86_EFLAGS_TF;
11750 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11753 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11755 __kvm_set_rflags(vcpu, rflags);
11756 kvm_make_request(KVM_REQ_EVENT, vcpu);
11758 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11760 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11764 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11768 r = kvm_mmu_reload(vcpu);
11772 if (!vcpu->arch.mmu->direct_map &&
11773 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11776 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11779 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11781 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11783 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11786 static inline u32 kvm_async_pf_next_probe(u32 key)
11788 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11791 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11793 u32 key = kvm_async_pf_hash_fn(gfn);
11795 while (vcpu->arch.apf.gfns[key] != ~0)
11796 key = kvm_async_pf_next_probe(key);
11798 vcpu->arch.apf.gfns[key] = gfn;
11801 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11804 u32 key = kvm_async_pf_hash_fn(gfn);
11806 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11807 (vcpu->arch.apf.gfns[key] != gfn &&
11808 vcpu->arch.apf.gfns[key] != ~0); i++)
11809 key = kvm_async_pf_next_probe(key);
11814 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11816 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11819 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11823 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11825 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11829 vcpu->arch.apf.gfns[i] = ~0;
11831 j = kvm_async_pf_next_probe(j);
11832 if (vcpu->arch.apf.gfns[j] == ~0)
11834 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11836 * k lies cyclically in ]i,j]
11838 * |....j i.k.| or |.k..j i...|
11840 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11841 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11846 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11848 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11850 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11854 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11856 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11858 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11859 &token, offset, sizeof(token));
11862 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11864 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11867 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11868 &val, offset, sizeof(val)))
11874 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11876 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11879 if (!kvm_pv_async_pf_enabled(vcpu) ||
11880 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11886 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11888 if (unlikely(!lapic_in_kernel(vcpu) ||
11889 kvm_event_needs_reinjection(vcpu) ||
11890 vcpu->arch.exception.pending))
11893 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11897 * If interrupts are off we cannot even use an artificial
11900 return kvm_arch_interrupt_allowed(vcpu);
11903 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11904 struct kvm_async_pf *work)
11906 struct x86_exception fault;
11908 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11909 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11911 if (kvm_can_deliver_async_pf(vcpu) &&
11912 !apf_put_user_notpresent(vcpu)) {
11913 fault.vector = PF_VECTOR;
11914 fault.error_code_valid = true;
11915 fault.error_code = 0;
11916 fault.nested_page_fault = false;
11917 fault.address = work->arch.token;
11918 fault.async_page_fault = true;
11919 kvm_inject_page_fault(vcpu, &fault);
11923 * It is not possible to deliver a paravirtualized asynchronous
11924 * page fault, but putting the guest in an artificial halt state
11925 * can be beneficial nevertheless: if an interrupt arrives, we
11926 * can deliver it timely and perhaps the guest will schedule
11927 * another process. When the instruction that triggered a page
11928 * fault is retried, hopefully the page will be ready in the host.
11930 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11935 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11936 struct kvm_async_pf *work)
11938 struct kvm_lapic_irq irq = {
11939 .delivery_mode = APIC_DM_FIXED,
11940 .vector = vcpu->arch.apf.vec
11943 if (work->wakeup_all)
11944 work->arch.token = ~0; /* broadcast wakeup */
11946 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11947 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11949 if ((work->wakeup_all || work->notpresent_injected) &&
11950 kvm_pv_async_pf_enabled(vcpu) &&
11951 !apf_put_user_ready(vcpu, work->arch.token)) {
11952 vcpu->arch.apf.pageready_pending = true;
11953 kvm_apic_set_irq(vcpu, &irq, NULL);
11956 vcpu->arch.apf.halted = false;
11957 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11960 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11962 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11963 if (!vcpu->arch.apf.pageready_pending)
11964 kvm_vcpu_kick(vcpu);
11967 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11969 if (!kvm_pv_async_pf_enabled(vcpu))
11972 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11975 void kvm_arch_start_assignment(struct kvm *kvm)
11977 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
11978 static_call_cond(kvm_x86_start_assignment)(kvm);
11980 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11982 void kvm_arch_end_assignment(struct kvm *kvm)
11984 atomic_dec(&kvm->arch.assigned_device_count);
11986 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11988 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11990 return atomic_read(&kvm->arch.assigned_device_count);
11992 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11994 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11996 atomic_inc(&kvm->arch.noncoherent_dma_count);
11998 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12000 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12002 atomic_dec(&kvm->arch.noncoherent_dma_count);
12004 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12006 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12008 return atomic_read(&kvm->arch.noncoherent_dma_count);
12010 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12012 bool kvm_arch_has_irq_bypass(void)
12017 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12018 struct irq_bypass_producer *prod)
12020 struct kvm_kernel_irqfd *irqfd =
12021 container_of(cons, struct kvm_kernel_irqfd, consumer);
12024 irqfd->producer = prod;
12025 kvm_arch_start_assignment(irqfd->kvm);
12026 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
12027 prod->irq, irqfd->gsi, 1);
12030 kvm_arch_end_assignment(irqfd->kvm);
12035 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12036 struct irq_bypass_producer *prod)
12039 struct kvm_kernel_irqfd *irqfd =
12040 container_of(cons, struct kvm_kernel_irqfd, consumer);
12042 WARN_ON(irqfd->producer != prod);
12043 irqfd->producer = NULL;
12046 * When producer of consumer is unregistered, we change back to
12047 * remapped mode, so we can re-use the current implementation
12048 * when the irq is masked/disabled or the consumer side (KVM
12049 * int this case doesn't want to receive the interrupts.
12051 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12053 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12054 " fails: %d\n", irqfd->consumer.token, ret);
12056 kvm_arch_end_assignment(irqfd->kvm);
12059 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12060 uint32_t guest_irq, bool set)
12062 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12065 bool kvm_vector_hashing_enabled(void)
12067 return vector_hashing;
12070 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12072 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12074 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12077 int kvm_spec_ctrl_test_value(u64 value)
12080 * test that setting IA32_SPEC_CTRL to given value
12081 * is allowed by the host processor
12085 unsigned long flags;
12088 local_irq_save(flags);
12090 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12092 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12095 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12097 local_irq_restore(flags);
12101 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12103 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12105 struct x86_exception fault;
12106 u32 access = error_code &
12107 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12109 if (!(error_code & PFERR_PRESENT_MASK) ||
12110 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12112 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12113 * tables probably do not match the TLB. Just proceed
12114 * with the error code that the processor gave.
12116 fault.vector = PF_VECTOR;
12117 fault.error_code_valid = true;
12118 fault.error_code = error_code;
12119 fault.nested_page_fault = false;
12120 fault.address = gva;
12122 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12124 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12127 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12128 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12129 * indicates whether exit to userspace is needed.
12131 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12132 struct x86_exception *e)
12134 if (r == X86EMUL_PROPAGATE_FAULT) {
12135 kvm_inject_emulated_page_fault(vcpu, e);
12140 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12141 * while handling a VMX instruction KVM could've handled the request
12142 * correctly by exiting to userspace and performing I/O but there
12143 * doesn't seem to be a real use-case behind such requests, just return
12144 * KVM_EXIT_INTERNAL_ERROR for now.
12146 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
12147 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
12148 vcpu->run->internal.ndata = 0;
12152 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12154 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12157 struct x86_exception e;
12164 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12165 if (r != X86EMUL_CONTINUE)
12166 return kvm_handle_memory_failure(vcpu, r, &e);
12168 if (operand.pcid >> 12 != 0) {
12169 kvm_inject_gp(vcpu, 0);
12173 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12176 case INVPCID_TYPE_INDIV_ADDR:
12177 if ((!pcid_enabled && (operand.pcid != 0)) ||
12178 is_noncanonical_address(operand.gla, vcpu)) {
12179 kvm_inject_gp(vcpu, 0);
12182 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12183 return kvm_skip_emulated_instruction(vcpu);
12185 case INVPCID_TYPE_SINGLE_CTXT:
12186 if (!pcid_enabled && (operand.pcid != 0)) {
12187 kvm_inject_gp(vcpu, 0);
12191 kvm_invalidate_pcid(vcpu, operand.pcid);
12192 return kvm_skip_emulated_instruction(vcpu);
12194 case INVPCID_TYPE_ALL_NON_GLOBAL:
12196 * Currently, KVM doesn't mark global entries in the shadow
12197 * page tables, so a non-global flush just degenerates to a
12198 * global flush. If needed, we could optimize this later by
12199 * keeping track of global entries in shadow page tables.
12203 case INVPCID_TYPE_ALL_INCL_GLOBAL:
12204 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12205 return kvm_skip_emulated_instruction(vcpu);
12208 BUG(); /* We have already checked above that type <= 3 */
12211 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12213 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12215 struct kvm_run *run = vcpu->run;
12216 struct kvm_mmio_fragment *frag;
12219 BUG_ON(!vcpu->mmio_needed);
12221 /* Complete previous fragment */
12222 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12223 len = min(8u, frag->len);
12224 if (!vcpu->mmio_is_write)
12225 memcpy(frag->data, run->mmio.data, len);
12227 if (frag->len <= 8) {
12228 /* Switch to the next fragment. */
12230 vcpu->mmio_cur_fragment++;
12232 /* Go forward to the next mmio piece. */
12238 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12239 vcpu->mmio_needed = 0;
12241 // VMG change, at this point, we're always done
12242 // RIP has already been advanced
12246 // More MMIO is needed
12247 run->mmio.phys_addr = frag->gpa;
12248 run->mmio.len = min(8u, frag->len);
12249 run->mmio.is_write = vcpu->mmio_is_write;
12250 if (run->mmio.is_write)
12251 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12252 run->exit_reason = KVM_EXIT_MMIO;
12254 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12259 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12263 struct kvm_mmio_fragment *frag;
12268 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12269 if (handled == bytes)
12276 /*TODO: Check if need to increment number of frags */
12277 frag = vcpu->mmio_fragments;
12278 vcpu->mmio_nr_fragments = 1;
12283 vcpu->mmio_needed = 1;
12284 vcpu->mmio_cur_fragment = 0;
12286 vcpu->run->mmio.phys_addr = gpa;
12287 vcpu->run->mmio.len = min(8u, frag->len);
12288 vcpu->run->mmio.is_write = 1;
12289 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12290 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12292 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12296 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12298 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12302 struct kvm_mmio_fragment *frag;
12307 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12308 if (handled == bytes)
12315 /*TODO: Check if need to increment number of frags */
12316 frag = vcpu->mmio_fragments;
12317 vcpu->mmio_nr_fragments = 1;
12322 vcpu->mmio_needed = 1;
12323 vcpu->mmio_cur_fragment = 0;
12325 vcpu->run->mmio.phys_addr = gpa;
12326 vcpu->run->mmio.len = min(8u, frag->len);
12327 vcpu->run->mmio.is_write = 0;
12328 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12330 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12334 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12336 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12338 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
12339 vcpu->arch.pio.count * vcpu->arch.pio.size);
12340 vcpu->arch.pio.count = 0;
12345 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12346 unsigned int port, void *data, unsigned int count)
12350 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
12355 vcpu->arch.pio.count = 0;
12360 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12361 unsigned int port, void *data, unsigned int count)
12365 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
12368 vcpu->arch.pio.count = 0;
12370 vcpu->arch.guest_ins_data = data;
12371 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12377 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12378 unsigned int port, void *data, unsigned int count,
12381 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
12382 : kvm_sev_es_outs(vcpu, size, port, data, count);
12384 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12386 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12387 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12388 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12389 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12390 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12391 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12392 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12393 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12394 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12395 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12396 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12397 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12398 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12399 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12400 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12401 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12402 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12403 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12404 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12405 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12406 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12407 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12408 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12409 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12410 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12411 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12412 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);