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, pages_4k),
237 STATS_DESC_ICOUNTER(VM, pages_2m),
238 STATS_DESC_ICOUNTER(VM, pages_1g),
239 STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
240 STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
241 STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
244 const struct kvm_stats_header kvm_vm_stats_header = {
245 .name_size = KVM_STATS_NAME_SIZE,
246 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
247 .id_offset = sizeof(struct kvm_stats_header),
248 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
249 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
250 sizeof(kvm_vm_stats_desc),
253 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
254 KVM_GENERIC_VCPU_STATS(),
255 STATS_DESC_COUNTER(VCPU, pf_fixed),
256 STATS_DESC_COUNTER(VCPU, pf_guest),
257 STATS_DESC_COUNTER(VCPU, tlb_flush),
258 STATS_DESC_COUNTER(VCPU, invlpg),
259 STATS_DESC_COUNTER(VCPU, exits),
260 STATS_DESC_COUNTER(VCPU, io_exits),
261 STATS_DESC_COUNTER(VCPU, mmio_exits),
262 STATS_DESC_COUNTER(VCPU, signal_exits),
263 STATS_DESC_COUNTER(VCPU, irq_window_exits),
264 STATS_DESC_COUNTER(VCPU, nmi_window_exits),
265 STATS_DESC_COUNTER(VCPU, l1d_flush),
266 STATS_DESC_COUNTER(VCPU, halt_exits),
267 STATS_DESC_COUNTER(VCPU, request_irq_exits),
268 STATS_DESC_COUNTER(VCPU, irq_exits),
269 STATS_DESC_COUNTER(VCPU, host_state_reload),
270 STATS_DESC_COUNTER(VCPU, fpu_reload),
271 STATS_DESC_COUNTER(VCPU, insn_emulation),
272 STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
273 STATS_DESC_COUNTER(VCPU, hypercalls),
274 STATS_DESC_COUNTER(VCPU, irq_injections),
275 STATS_DESC_COUNTER(VCPU, nmi_injections),
276 STATS_DESC_COUNTER(VCPU, req_event),
277 STATS_DESC_COUNTER(VCPU, nested_run),
278 STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
279 STATS_DESC_COUNTER(VCPU, directed_yield_successful),
280 STATS_DESC_ICOUNTER(VCPU, guest_mode)
283 const struct kvm_stats_header kvm_vcpu_stats_header = {
284 .name_size = KVM_STATS_NAME_SIZE,
285 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
286 .id_offset = sizeof(struct kvm_stats_header),
287 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
288 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
289 sizeof(kvm_vcpu_stats_desc),
292 u64 __read_mostly host_xcr0;
293 u64 __read_mostly supported_xcr0;
294 EXPORT_SYMBOL_GPL(supported_xcr0);
296 static struct kmem_cache *x86_fpu_cache;
298 static struct kmem_cache *x86_emulator_cache;
301 * When called, it means the previous get/set msr reached an invalid msr.
302 * Return true if we want to ignore/silent this failed msr access.
304 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
306 const char *op = write ? "wrmsr" : "rdmsr";
309 if (report_ignored_msrs)
310 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
315 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
321 static struct kmem_cache *kvm_alloc_emulator_cache(void)
323 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
324 unsigned int size = sizeof(struct x86_emulate_ctxt);
326 return kmem_cache_create_usercopy("x86_emulator", size,
327 __alignof__(struct x86_emulate_ctxt),
328 SLAB_ACCOUNT, useroffset,
329 size - useroffset, NULL);
332 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
334 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
337 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
338 vcpu->arch.apf.gfns[i] = ~0;
341 static void kvm_on_user_return(struct user_return_notifier *urn)
344 struct kvm_user_return_msrs *msrs
345 = container_of(urn, struct kvm_user_return_msrs, urn);
346 struct kvm_user_return_msr_values *values;
350 * Disabling irqs at this point since the following code could be
351 * interrupted and executed through kvm_arch_hardware_disable()
353 local_irq_save(flags);
354 if (msrs->registered) {
355 msrs->registered = false;
356 user_return_notifier_unregister(urn);
358 local_irq_restore(flags);
359 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
360 values = &msrs->values[slot];
361 if (values->host != values->curr) {
362 wrmsrl(kvm_uret_msrs_list[slot], values->host);
363 values->curr = values->host;
368 static int kvm_probe_user_return_msr(u32 msr)
374 ret = rdmsrl_safe(msr, &val);
377 ret = wrmsrl_safe(msr, val);
383 int kvm_add_user_return_msr(u32 msr)
385 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
387 if (kvm_probe_user_return_msr(msr))
390 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
391 return kvm_nr_uret_msrs++;
393 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
395 int kvm_find_user_return_msr(u32 msr)
399 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
400 if (kvm_uret_msrs_list[i] == msr)
405 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
407 static void kvm_user_return_msr_cpu_online(void)
409 unsigned int cpu = smp_processor_id();
410 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
414 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
415 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
416 msrs->values[i].host = value;
417 msrs->values[i].curr = value;
421 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
423 unsigned int cpu = smp_processor_id();
424 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
427 value = (value & mask) | (msrs->values[slot].host & ~mask);
428 if (value == msrs->values[slot].curr)
430 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
434 msrs->values[slot].curr = value;
435 if (!msrs->registered) {
436 msrs->urn.on_user_return = kvm_on_user_return;
437 user_return_notifier_register(&msrs->urn);
438 msrs->registered = true;
442 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
444 static void drop_user_return_notifiers(void)
446 unsigned int cpu = smp_processor_id();
447 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
449 if (msrs->registered)
450 kvm_on_user_return(&msrs->urn);
453 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
455 return vcpu->arch.apic_base;
457 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
459 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
461 return kvm_apic_mode(kvm_get_apic_base(vcpu));
463 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
465 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
467 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
468 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
469 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
470 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
472 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
474 if (!msr_info->host_initiated) {
475 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
477 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
481 kvm_lapic_set_base(vcpu, msr_info->data);
482 kvm_recalculate_apic_map(vcpu->kvm);
485 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
488 * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
490 * Hardware virtualization extension instructions may fault if a reboot turns
491 * off virtualization while processes are running. Usually after catching the
492 * fault we just panic; during reboot instead the instruction is ignored.
494 noinstr void kvm_spurious_fault(void)
496 /* Fault while not rebooting. We want the trace. */
497 BUG_ON(!kvm_rebooting);
499 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
501 #define EXCPT_BENIGN 0
502 #define EXCPT_CONTRIBUTORY 1
505 static int exception_class(int vector)
515 return EXCPT_CONTRIBUTORY;
522 #define EXCPT_FAULT 0
524 #define EXCPT_ABORT 2
525 #define EXCPT_INTERRUPT 3
527 static int exception_type(int vector)
531 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
532 return EXCPT_INTERRUPT;
536 /* #DB is trap, as instruction watchpoints are handled elsewhere */
537 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
540 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
543 /* Reserved exceptions will result in fault */
547 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
549 unsigned nr = vcpu->arch.exception.nr;
550 bool has_payload = vcpu->arch.exception.has_payload;
551 unsigned long payload = vcpu->arch.exception.payload;
559 * "Certain debug exceptions may clear bit 0-3. The
560 * remaining contents of the DR6 register are never
561 * cleared by the processor".
563 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
565 * In order to reflect the #DB exception payload in guest
566 * dr6, three components need to be considered: active low
567 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
569 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
570 * In the target guest dr6:
571 * FIXED_1 bits should always be set.
572 * Active low bits should be cleared if 1-setting in payload.
573 * Active high bits should be set if 1-setting in payload.
575 * Note, the payload is compatible with the pending debug
576 * exceptions/exit qualification under VMX, that active_low bits
577 * are active high in payload.
578 * So they need to be flipped for DR6.
580 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
581 vcpu->arch.dr6 |= payload;
582 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
585 * The #DB payload is defined as compatible with the 'pending
586 * debug exceptions' field under VMX, not DR6. While bit 12 is
587 * defined in the 'pending debug exceptions' field (enabled
588 * breakpoint), it is reserved and must be zero in DR6.
590 vcpu->arch.dr6 &= ~BIT(12);
593 vcpu->arch.cr2 = payload;
597 vcpu->arch.exception.has_payload = false;
598 vcpu->arch.exception.payload = 0;
600 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
602 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
603 unsigned nr, bool has_error, u32 error_code,
604 bool has_payload, unsigned long payload, bool reinject)
609 kvm_make_request(KVM_REQ_EVENT, vcpu);
611 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
615 * On vmentry, vcpu->arch.exception.pending is only
616 * true if an event injection was blocked by
617 * nested_run_pending. In that case, however,
618 * vcpu_enter_guest requests an immediate exit,
619 * and the guest shouldn't proceed far enough to
622 WARN_ON_ONCE(vcpu->arch.exception.pending);
623 vcpu->arch.exception.injected = true;
624 if (WARN_ON_ONCE(has_payload)) {
626 * A reinjected event has already
627 * delivered its payload.
633 vcpu->arch.exception.pending = true;
634 vcpu->arch.exception.injected = false;
636 vcpu->arch.exception.has_error_code = has_error;
637 vcpu->arch.exception.nr = nr;
638 vcpu->arch.exception.error_code = error_code;
639 vcpu->arch.exception.has_payload = has_payload;
640 vcpu->arch.exception.payload = payload;
641 if (!is_guest_mode(vcpu))
642 kvm_deliver_exception_payload(vcpu);
646 /* to check exception */
647 prev_nr = vcpu->arch.exception.nr;
648 if (prev_nr == DF_VECTOR) {
649 /* triple fault -> shutdown */
650 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
653 class1 = exception_class(prev_nr);
654 class2 = exception_class(nr);
655 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
656 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
658 * Generate double fault per SDM Table 5-5. Set
659 * exception.pending = true so that the double fault
660 * can trigger a nested vmexit.
662 vcpu->arch.exception.pending = true;
663 vcpu->arch.exception.injected = false;
664 vcpu->arch.exception.has_error_code = true;
665 vcpu->arch.exception.nr = DF_VECTOR;
666 vcpu->arch.exception.error_code = 0;
667 vcpu->arch.exception.has_payload = false;
668 vcpu->arch.exception.payload = 0;
670 /* replace previous exception with a new one in a hope
671 that instruction re-execution will regenerate lost
676 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
678 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
680 EXPORT_SYMBOL_GPL(kvm_queue_exception);
682 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
684 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
686 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
688 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
689 unsigned long payload)
691 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
693 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
695 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
696 u32 error_code, unsigned long payload)
698 kvm_multiple_exception(vcpu, nr, true, error_code,
699 true, payload, false);
702 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
705 kvm_inject_gp(vcpu, 0);
707 return kvm_skip_emulated_instruction(vcpu);
711 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
713 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
715 ++vcpu->stat.pf_guest;
716 vcpu->arch.exception.nested_apf =
717 is_guest_mode(vcpu) && fault->async_page_fault;
718 if (vcpu->arch.exception.nested_apf) {
719 vcpu->arch.apf.nested_apf_token = fault->address;
720 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
722 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
726 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
728 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
729 struct x86_exception *fault)
731 struct kvm_mmu *fault_mmu;
732 WARN_ON_ONCE(fault->vector != PF_VECTOR);
734 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
738 * Invalidate the TLB entry for the faulting address, if it exists,
739 * else the access will fault indefinitely (and to emulate hardware).
741 if ((fault->error_code & PFERR_PRESENT_MASK) &&
742 !(fault->error_code & PFERR_RSVD_MASK))
743 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
744 fault_mmu->root_hpa);
746 fault_mmu->inject_page_fault(vcpu, fault);
747 return fault->nested_page_fault;
749 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
751 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
753 atomic_inc(&vcpu->arch.nmi_queued);
754 kvm_make_request(KVM_REQ_NMI, vcpu);
756 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
758 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
760 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
762 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
764 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
766 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
768 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
771 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
772 * a #GP and return false.
774 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
776 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
778 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
781 EXPORT_SYMBOL_GPL(kvm_require_cpl);
783 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
785 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
788 kvm_queue_exception(vcpu, UD_VECTOR);
791 EXPORT_SYMBOL_GPL(kvm_require_dr);
793 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
795 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
799 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
801 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
803 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
807 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
810 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
813 real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(pdpt_gfn),
814 PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
815 if (real_gpa == UNMAPPED_GVA)
818 /* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
819 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
820 cr3 & GENMASK(11, 5), sizeof(pdpte));
824 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
825 if ((pdpte[i] & PT_PRESENT_MASK) &&
826 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
831 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
832 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
833 vcpu->arch.pdptrs_from_userspace = false;
837 EXPORT_SYMBOL_GPL(load_pdptrs);
839 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
841 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
842 kvm_clear_async_pf_completion_queue(vcpu);
843 kvm_async_pf_hash_reset(vcpu);
846 if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
847 kvm_mmu_reset_context(vcpu);
849 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
850 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
851 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
852 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
854 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
856 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
858 unsigned long old_cr0 = kvm_read_cr0(vcpu);
859 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
864 if (cr0 & 0xffffffff00000000UL)
868 cr0 &= ~CR0_RESERVED_BITS;
870 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
873 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
877 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
878 (cr0 & X86_CR0_PG)) {
883 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
888 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
889 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
890 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
893 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
896 static_call(kvm_x86_set_cr0)(vcpu, cr0);
898 kvm_post_set_cr0(vcpu, old_cr0, cr0);
902 EXPORT_SYMBOL_GPL(kvm_set_cr0);
904 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
906 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
908 EXPORT_SYMBOL_GPL(kvm_lmsw);
910 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
912 if (vcpu->arch.guest_state_protected)
915 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
917 if (vcpu->arch.xcr0 != host_xcr0)
918 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
920 if (vcpu->arch.xsaves_enabled &&
921 vcpu->arch.ia32_xss != host_xss)
922 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
925 if (static_cpu_has(X86_FEATURE_PKU) &&
926 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
927 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
928 vcpu->arch.pkru != vcpu->arch.host_pkru)
929 write_pkru(vcpu->arch.pkru);
931 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
933 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
935 if (vcpu->arch.guest_state_protected)
938 if (static_cpu_has(X86_FEATURE_PKU) &&
939 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
940 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
941 vcpu->arch.pkru = rdpkru();
942 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
943 write_pkru(vcpu->arch.host_pkru);
946 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
948 if (vcpu->arch.xcr0 != host_xcr0)
949 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
951 if (vcpu->arch.xsaves_enabled &&
952 vcpu->arch.ia32_xss != host_xss)
953 wrmsrl(MSR_IA32_XSS, host_xss);
957 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
959 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
962 u64 old_xcr0 = vcpu->arch.xcr0;
965 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
966 if (index != XCR_XFEATURE_ENABLED_MASK)
968 if (!(xcr0 & XFEATURE_MASK_FP))
970 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
974 * Do not allow the guest to set bits that we do not support
975 * saving. However, xcr0 bit 0 is always set, even if the
976 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
978 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
979 if (xcr0 & ~valid_bits)
982 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
983 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
986 if (xcr0 & XFEATURE_MASK_AVX512) {
987 if (!(xcr0 & XFEATURE_MASK_YMM))
989 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
992 vcpu->arch.xcr0 = xcr0;
994 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
995 kvm_update_cpuid_runtime(vcpu);
999 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1001 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1002 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1003 kvm_inject_gp(vcpu, 0);
1007 return kvm_skip_emulated_instruction(vcpu);
1009 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1011 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1013 if (cr4 & cr4_reserved_bits)
1016 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1019 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1021 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1023 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1026 * If any role bit is changed, the MMU needs to be reset.
1028 * If CR4.PCIDE is changed 1 -> 0, the guest TLB must be flushed.
1029 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1030 * according to the SDM; however, stale prev_roots could be reused
1031 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1032 * free them all. KVM_REQ_MMU_RELOAD is fit for the both cases; it
1033 * is slow, but changing CR4.PCIDE is a rare case.
1035 * If CR4.PGE is changed, the guest TLB must be flushed.
1037 * Note: resetting MMU is a superset of KVM_REQ_MMU_RELOAD and
1038 * KVM_REQ_MMU_RELOAD is a superset of KVM_REQ_TLB_FLUSH_GUEST, hence
1039 * the usage of "else if".
1041 if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1042 kvm_mmu_reset_context(vcpu);
1043 else if ((cr4 ^ old_cr4) & X86_CR4_PCIDE)
1044 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
1045 else if ((cr4 ^ old_cr4) & X86_CR4_PGE)
1046 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1048 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1050 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1052 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1053 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1056 if (!kvm_is_valid_cr4(vcpu, cr4))
1059 if (is_long_mode(vcpu)) {
1060 if (!(cr4 & X86_CR4_PAE))
1062 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1064 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1065 && ((cr4 ^ old_cr4) & pdptr_bits)
1066 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1067 kvm_read_cr3(vcpu)))
1070 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1071 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1074 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1075 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1079 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1081 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1085 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1087 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1089 struct kvm_mmu *mmu = vcpu->arch.mmu;
1090 unsigned long roots_to_free = 0;
1094 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1095 * this is reachable when running EPT=1 and unrestricted_guest=0, and
1096 * also via the emulator. KVM's TDP page tables are not in the scope of
1097 * the invalidation, but the guest's TLB entries need to be flushed as
1098 * the CPU may have cached entries in its TLB for the target PCID.
1100 if (unlikely(tdp_enabled)) {
1101 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1106 * If neither the current CR3 nor any of the prev_roots use the given
1107 * PCID, then nothing needs to be done here because a resync will
1108 * happen anyway before switching to any other CR3.
1110 if (kvm_get_active_pcid(vcpu) == pcid) {
1111 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1112 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1116 * If PCID is disabled, there is no need to free prev_roots even if the
1117 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1120 if (!kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
1123 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1124 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1125 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1127 kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1130 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1132 bool skip_tlb_flush = false;
1133 unsigned long pcid = 0;
1134 #ifdef CONFIG_X86_64
1135 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1138 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1139 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1140 pcid = cr3 & X86_CR3_PCID_MASK;
1144 /* PDPTRs are always reloaded for PAE paging. */
1145 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1146 goto handle_tlb_flush;
1149 * Do not condition the GPA check on long mode, this helper is used to
1150 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1151 * the current vCPU mode is accurate.
1153 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1156 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1159 if (cr3 != kvm_read_cr3(vcpu))
1160 kvm_mmu_new_pgd(vcpu, cr3);
1162 vcpu->arch.cr3 = cr3;
1163 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1167 * A load of CR3 that flushes the TLB flushes only the current PCID,
1168 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1169 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1170 * and it's impossible to use a non-zero PCID when PCID is disabled,
1171 * i.e. only PCID=0 can be relevant.
1173 if (!skip_tlb_flush)
1174 kvm_invalidate_pcid(vcpu, pcid);
1178 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1180 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1182 if (cr8 & CR8_RESERVED_BITS)
1184 if (lapic_in_kernel(vcpu))
1185 kvm_lapic_set_tpr(vcpu, cr8);
1187 vcpu->arch.cr8 = cr8;
1190 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1192 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1194 if (lapic_in_kernel(vcpu))
1195 return kvm_lapic_get_cr8(vcpu);
1197 return vcpu->arch.cr8;
1199 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1201 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1205 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1206 for (i = 0; i < KVM_NR_DB_REGS; i++)
1207 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1211 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1215 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1216 dr7 = vcpu->arch.guest_debug_dr7;
1218 dr7 = vcpu->arch.dr7;
1219 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1220 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1221 if (dr7 & DR7_BP_EN_MASK)
1222 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1224 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1226 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1228 u64 fixed = DR6_FIXED_1;
1230 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1233 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1234 fixed |= DR6_BUS_LOCK;
1238 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1240 size_t size = ARRAY_SIZE(vcpu->arch.db);
1244 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1245 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1246 vcpu->arch.eff_db[dr] = val;
1250 if (!kvm_dr6_valid(val))
1252 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1256 if (!kvm_dr7_valid(val))
1258 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1259 kvm_update_dr7(vcpu);
1265 EXPORT_SYMBOL_GPL(kvm_set_dr);
1267 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1269 size_t size = ARRAY_SIZE(vcpu->arch.db);
1273 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1277 *val = vcpu->arch.dr6;
1281 *val = vcpu->arch.dr7;
1285 EXPORT_SYMBOL_GPL(kvm_get_dr);
1287 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1289 u32 ecx = kvm_rcx_read(vcpu);
1292 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1293 kvm_inject_gp(vcpu, 0);
1297 kvm_rax_write(vcpu, (u32)data);
1298 kvm_rdx_write(vcpu, data >> 32);
1299 return kvm_skip_emulated_instruction(vcpu);
1301 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1304 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1305 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1307 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1308 * extract the supported MSRs from the related const lists.
1309 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1310 * capabilities of the host cpu. This capabilities test skips MSRs that are
1311 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1312 * may depend on host virtualization features rather than host cpu features.
1315 static const u32 msrs_to_save_all[] = {
1316 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1318 #ifdef CONFIG_X86_64
1319 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1321 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1322 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1324 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1325 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1326 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1327 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1328 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1329 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1330 MSR_IA32_UMWAIT_CONTROL,
1332 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1333 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1334 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1335 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1336 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1337 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1338 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1339 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1340 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1341 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1342 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1343 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1344 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1345 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1346 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1347 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1348 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1349 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1350 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1351 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1352 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1353 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1355 MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1356 MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1357 MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1358 MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1359 MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1360 MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1363 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1364 static unsigned num_msrs_to_save;
1366 static const u32 emulated_msrs_all[] = {
1367 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1368 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1369 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1370 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1371 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1372 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1373 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1375 HV_X64_MSR_VP_INDEX,
1376 HV_X64_MSR_VP_RUNTIME,
1377 HV_X64_MSR_SCONTROL,
1378 HV_X64_MSR_STIMER0_CONFIG,
1379 HV_X64_MSR_VP_ASSIST_PAGE,
1380 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1381 HV_X64_MSR_TSC_EMULATION_STATUS,
1382 HV_X64_MSR_SYNDBG_OPTIONS,
1383 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1384 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1385 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1387 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1388 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1390 MSR_IA32_TSC_ADJUST,
1391 MSR_IA32_TSC_DEADLINE,
1392 MSR_IA32_ARCH_CAPABILITIES,
1393 MSR_IA32_PERF_CAPABILITIES,
1394 MSR_IA32_MISC_ENABLE,
1395 MSR_IA32_MCG_STATUS,
1397 MSR_IA32_MCG_EXT_CTL,
1401 MSR_MISC_FEATURES_ENABLES,
1402 MSR_AMD64_VIRT_SPEC_CTRL,
1403 MSR_AMD64_TSC_RATIO,
1408 * The following list leaves out MSRs whose values are determined
1409 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1410 * We always support the "true" VMX control MSRs, even if the host
1411 * processor does not, so I am putting these registers here rather
1412 * than in msrs_to_save_all.
1415 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1416 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1417 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1418 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1420 MSR_IA32_VMX_CR0_FIXED0,
1421 MSR_IA32_VMX_CR4_FIXED0,
1422 MSR_IA32_VMX_VMCS_ENUM,
1423 MSR_IA32_VMX_PROCBASED_CTLS2,
1424 MSR_IA32_VMX_EPT_VPID_CAP,
1425 MSR_IA32_VMX_VMFUNC,
1428 MSR_KVM_POLL_CONTROL,
1431 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1432 static unsigned num_emulated_msrs;
1435 * List of msr numbers which are used to expose MSR-based features that
1436 * can be used by a hypervisor to validate requested CPU features.
1438 static const u32 msr_based_features_all[] = {
1440 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1441 MSR_IA32_VMX_PINBASED_CTLS,
1442 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1443 MSR_IA32_VMX_PROCBASED_CTLS,
1444 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1445 MSR_IA32_VMX_EXIT_CTLS,
1446 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1447 MSR_IA32_VMX_ENTRY_CTLS,
1449 MSR_IA32_VMX_CR0_FIXED0,
1450 MSR_IA32_VMX_CR0_FIXED1,
1451 MSR_IA32_VMX_CR4_FIXED0,
1452 MSR_IA32_VMX_CR4_FIXED1,
1453 MSR_IA32_VMX_VMCS_ENUM,
1454 MSR_IA32_VMX_PROCBASED_CTLS2,
1455 MSR_IA32_VMX_EPT_VPID_CAP,
1456 MSR_IA32_VMX_VMFUNC,
1460 MSR_IA32_ARCH_CAPABILITIES,
1461 MSR_IA32_PERF_CAPABILITIES,
1464 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1465 static unsigned int num_msr_based_features;
1467 static u64 kvm_get_arch_capabilities(void)
1471 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1472 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1475 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1476 * the nested hypervisor runs with NX huge pages. If it is not,
1477 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1478 * L1 guests, so it need not worry about its own (L2) guests.
1480 data |= ARCH_CAP_PSCHANGE_MC_NO;
1483 * If we're doing cache flushes (either "always" or "cond")
1484 * we will do one whenever the guest does a vmlaunch/vmresume.
1485 * If an outer hypervisor is doing the cache flush for us
1486 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1487 * capability to the guest too, and if EPT is disabled we're not
1488 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1489 * require a nested hypervisor to do a flush of its own.
1491 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1492 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1494 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1495 data |= ARCH_CAP_RDCL_NO;
1496 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1497 data |= ARCH_CAP_SSB_NO;
1498 if (!boot_cpu_has_bug(X86_BUG_MDS))
1499 data |= ARCH_CAP_MDS_NO;
1501 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1503 * If RTM=0 because the kernel has disabled TSX, the host might
1504 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1505 * and therefore knows that there cannot be TAA) but keep
1506 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1507 * and we want to allow migrating those guests to tsx=off hosts.
1509 data &= ~ARCH_CAP_TAA_NO;
1510 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1511 data |= ARCH_CAP_TAA_NO;
1514 * Nothing to do here; we emulate TSX_CTRL if present on the
1515 * host so the guest can choose between disabling TSX or
1516 * using VERW to clear CPU buffers.
1523 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1525 switch (msr->index) {
1526 case MSR_IA32_ARCH_CAPABILITIES:
1527 msr->data = kvm_get_arch_capabilities();
1529 case MSR_IA32_UCODE_REV:
1530 rdmsrl_safe(msr->index, &msr->data);
1533 return static_call(kvm_x86_get_msr_feature)(msr);
1538 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1540 struct kvm_msr_entry msr;
1544 r = kvm_get_msr_feature(&msr);
1546 if (r == KVM_MSR_RET_INVALID) {
1547 /* Unconditionally clear the output for simplicity */
1549 if (kvm_msr_ignored_check(index, 0, false))
1561 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1563 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1566 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1569 if (efer & (EFER_LME | EFER_LMA) &&
1570 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1573 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1579 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1581 if (efer & efer_reserved_bits)
1584 return __kvm_valid_efer(vcpu, efer);
1586 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1588 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1590 u64 old_efer = vcpu->arch.efer;
1591 u64 efer = msr_info->data;
1594 if (efer & efer_reserved_bits)
1597 if (!msr_info->host_initiated) {
1598 if (!__kvm_valid_efer(vcpu, efer))
1601 if (is_paging(vcpu) &&
1602 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1607 efer |= vcpu->arch.efer & EFER_LMA;
1609 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1615 /* Update reserved bits */
1616 if ((efer ^ old_efer) & EFER_NX)
1617 kvm_mmu_reset_context(vcpu);
1622 void kvm_enable_efer_bits(u64 mask)
1624 efer_reserved_bits &= ~mask;
1626 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1628 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1630 struct kvm_x86_msr_filter *msr_filter;
1631 struct msr_bitmap_range *ranges;
1632 struct kvm *kvm = vcpu->kvm;
1637 /* x2APIC MSRs do not support filtering. */
1638 if (index >= 0x800 && index <= 0x8ff)
1641 idx = srcu_read_lock(&kvm->srcu);
1643 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1649 allowed = msr_filter->default_allow;
1650 ranges = msr_filter->ranges;
1652 for (i = 0; i < msr_filter->count; i++) {
1653 u32 start = ranges[i].base;
1654 u32 end = start + ranges[i].nmsrs;
1655 u32 flags = ranges[i].flags;
1656 unsigned long *bitmap = ranges[i].bitmap;
1658 if ((index >= start) && (index < end) && (flags & type)) {
1659 allowed = !!test_bit(index - start, bitmap);
1665 srcu_read_unlock(&kvm->srcu, idx);
1669 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1672 * Write @data into the MSR specified by @index. Select MSR specific fault
1673 * checks are bypassed if @host_initiated is %true.
1674 * Returns 0 on success, non-0 otherwise.
1675 * Assumes vcpu_load() was already called.
1677 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1678 bool host_initiated)
1680 struct msr_data msr;
1682 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1683 return KVM_MSR_RET_FILTERED;
1688 case MSR_KERNEL_GS_BASE:
1691 if (is_noncanonical_address(data, vcpu))
1694 case MSR_IA32_SYSENTER_EIP:
1695 case MSR_IA32_SYSENTER_ESP:
1697 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1698 * non-canonical address is written on Intel but not on
1699 * AMD (which ignores the top 32-bits, because it does
1700 * not implement 64-bit SYSENTER).
1702 * 64-bit code should hence be able to write a non-canonical
1703 * value on AMD. Making the address canonical ensures that
1704 * vmentry does not fail on Intel after writing a non-canonical
1705 * value, and that something deterministic happens if the guest
1706 * invokes 64-bit SYSENTER.
1708 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1711 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1714 if (!host_initiated &&
1715 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1716 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1720 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1721 * incomplete and conflicting architectural behavior. Current
1722 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1723 * reserved and always read as zeros. Enforce Intel's reserved
1724 * bits check if and only if the guest CPU is Intel, and clear
1725 * the bits in all other cases. This ensures cross-vendor
1726 * migration will provide consistent behavior for the guest.
1728 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1737 msr.host_initiated = host_initiated;
1739 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1742 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1743 u32 index, u64 data, bool host_initiated)
1745 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1747 if (ret == KVM_MSR_RET_INVALID)
1748 if (kvm_msr_ignored_check(index, data, true))
1755 * Read the MSR specified by @index into @data. Select MSR specific fault
1756 * checks are bypassed if @host_initiated is %true.
1757 * Returns 0 on success, non-0 otherwise.
1758 * Assumes vcpu_load() was already called.
1760 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1761 bool host_initiated)
1763 struct msr_data msr;
1766 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1767 return KVM_MSR_RET_FILTERED;
1771 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1774 if (!host_initiated &&
1775 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1776 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1782 msr.host_initiated = host_initiated;
1784 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1790 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1791 u32 index, u64 *data, bool host_initiated)
1793 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1795 if (ret == KVM_MSR_RET_INVALID) {
1796 /* Unconditionally clear *data for simplicity */
1798 if (kvm_msr_ignored_check(index, 0, false))
1805 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1807 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1809 EXPORT_SYMBOL_GPL(kvm_get_msr);
1811 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1813 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1815 EXPORT_SYMBOL_GPL(kvm_set_msr);
1817 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1819 int err = vcpu->run->msr.error;
1821 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1822 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1825 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1828 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1830 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1833 static u64 kvm_msr_reason(int r)
1836 case KVM_MSR_RET_INVALID:
1837 return KVM_MSR_EXIT_REASON_UNKNOWN;
1838 case KVM_MSR_RET_FILTERED:
1839 return KVM_MSR_EXIT_REASON_FILTER;
1841 return KVM_MSR_EXIT_REASON_INVAL;
1845 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1846 u32 exit_reason, u64 data,
1847 int (*completion)(struct kvm_vcpu *vcpu),
1850 u64 msr_reason = kvm_msr_reason(r);
1852 /* Check if the user wanted to know about this MSR fault */
1853 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1856 vcpu->run->exit_reason = exit_reason;
1857 vcpu->run->msr.error = 0;
1858 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1859 vcpu->run->msr.reason = msr_reason;
1860 vcpu->run->msr.index = index;
1861 vcpu->run->msr.data = data;
1862 vcpu->arch.complete_userspace_io = completion;
1867 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1869 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1870 complete_emulated_rdmsr, r);
1873 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1875 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1876 complete_emulated_wrmsr, r);
1879 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1881 u32 ecx = kvm_rcx_read(vcpu);
1885 r = kvm_get_msr(vcpu, ecx, &data);
1887 /* MSR read failed? See if we should ask user space */
1888 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1889 /* Bounce to user space */
1894 trace_kvm_msr_read(ecx, data);
1896 kvm_rax_write(vcpu, data & -1u);
1897 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1899 trace_kvm_msr_read_ex(ecx);
1902 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1904 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1906 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1908 u32 ecx = kvm_rcx_read(vcpu);
1909 u64 data = kvm_read_edx_eax(vcpu);
1912 r = kvm_set_msr(vcpu, ecx, data);
1914 /* MSR write failed? See if we should ask user space */
1915 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1916 /* Bounce to user space */
1919 /* Signal all other negative errors to userspace */
1924 trace_kvm_msr_write(ecx, data);
1926 trace_kvm_msr_write_ex(ecx, data);
1928 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1930 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1932 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1934 return kvm_skip_emulated_instruction(vcpu);
1936 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1938 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1940 /* Treat an INVD instruction as a NOP and just skip it. */
1941 return kvm_emulate_as_nop(vcpu);
1943 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1945 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1947 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1948 return kvm_emulate_as_nop(vcpu);
1950 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1952 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1954 kvm_queue_exception(vcpu, UD_VECTOR);
1957 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1959 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1961 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1962 return kvm_emulate_as_nop(vcpu);
1964 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1966 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1968 xfer_to_guest_mode_prepare();
1969 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1970 xfer_to_guest_mode_work_pending();
1974 * The fast path for frequent and performance sensitive wrmsr emulation,
1975 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1976 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1977 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1978 * other cases which must be called after interrupts are enabled on the host.
1980 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1982 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1985 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1986 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1987 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1988 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1991 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1992 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1993 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1994 trace_kvm_apic_write(APIC_ICR, (u32)data);
2001 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2003 if (!kvm_can_use_hv_timer(vcpu))
2006 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2010 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2012 u32 msr = kvm_rcx_read(vcpu);
2014 fastpath_t ret = EXIT_FASTPATH_NONE;
2017 case APIC_BASE_MSR + (APIC_ICR >> 4):
2018 data = kvm_read_edx_eax(vcpu);
2019 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2020 kvm_skip_emulated_instruction(vcpu);
2021 ret = EXIT_FASTPATH_EXIT_HANDLED;
2024 case MSR_IA32_TSC_DEADLINE:
2025 data = kvm_read_edx_eax(vcpu);
2026 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2027 kvm_skip_emulated_instruction(vcpu);
2028 ret = EXIT_FASTPATH_REENTER_GUEST;
2035 if (ret != EXIT_FASTPATH_NONE)
2036 trace_kvm_msr_write(msr, data);
2040 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2043 * Adapt set_msr() to msr_io()'s calling convention
2045 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2047 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2050 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2052 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2055 #ifdef CONFIG_X86_64
2056 struct pvclock_clock {
2066 struct pvclock_gtod_data {
2069 struct pvclock_clock clock; /* extract of a clocksource struct */
2070 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2076 static struct pvclock_gtod_data pvclock_gtod_data;
2078 static void update_pvclock_gtod(struct timekeeper *tk)
2080 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2082 write_seqcount_begin(&vdata->seq);
2084 /* copy pvclock gtod data */
2085 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2086 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2087 vdata->clock.mask = tk->tkr_mono.mask;
2088 vdata->clock.mult = tk->tkr_mono.mult;
2089 vdata->clock.shift = tk->tkr_mono.shift;
2090 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2091 vdata->clock.offset = tk->tkr_mono.base;
2093 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2094 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2095 vdata->raw_clock.mask = tk->tkr_raw.mask;
2096 vdata->raw_clock.mult = tk->tkr_raw.mult;
2097 vdata->raw_clock.shift = tk->tkr_raw.shift;
2098 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2099 vdata->raw_clock.offset = tk->tkr_raw.base;
2101 vdata->wall_time_sec = tk->xtime_sec;
2103 vdata->offs_boot = tk->offs_boot;
2105 write_seqcount_end(&vdata->seq);
2108 static s64 get_kvmclock_base_ns(void)
2110 /* Count up from boot time, but with the frequency of the raw clock. */
2111 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2114 static s64 get_kvmclock_base_ns(void)
2116 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2117 return ktime_get_boottime_ns();
2121 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2125 struct pvclock_wall_clock wc;
2132 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2137 ++version; /* first time write, random junk */
2141 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2145 * The guest calculates current wall clock time by adding
2146 * system time (updated by kvm_guest_time_update below) to the
2147 * wall clock specified here. We do the reverse here.
2149 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2151 wc.nsec = do_div(wall_nsec, 1000000000);
2152 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2153 wc.version = version;
2155 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2158 wc_sec_hi = wall_nsec >> 32;
2159 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2160 &wc_sec_hi, sizeof(wc_sec_hi));
2164 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2167 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2168 bool old_msr, bool host_initiated)
2170 struct kvm_arch *ka = &vcpu->kvm->arch;
2172 if (vcpu->vcpu_id == 0 && !host_initiated) {
2173 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2174 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2176 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2179 vcpu->arch.time = system_time;
2180 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2182 /* we verify if the enable bit is set... */
2183 vcpu->arch.pv_time_enabled = false;
2184 if (!(system_time & 1))
2187 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2188 &vcpu->arch.pv_time, system_time & ~1ULL,
2189 sizeof(struct pvclock_vcpu_time_info)))
2190 vcpu->arch.pv_time_enabled = true;
2195 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2197 do_shl32_div32(dividend, divisor);
2201 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2202 s8 *pshift, u32 *pmultiplier)
2210 scaled64 = scaled_hz;
2211 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2216 tps32 = (uint32_t)tps64;
2217 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2218 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2226 *pmultiplier = div_frac(scaled64, tps32);
2229 #ifdef CONFIG_X86_64
2230 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2233 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2234 static unsigned long max_tsc_khz;
2236 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2238 u64 v = (u64)khz * (1000000 + ppm);
2243 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2245 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2249 /* Guest TSC same frequency as host TSC? */
2251 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2255 /* TSC scaling supported? */
2256 if (!kvm_has_tsc_control) {
2257 if (user_tsc_khz > tsc_khz) {
2258 vcpu->arch.tsc_catchup = 1;
2259 vcpu->arch.tsc_always_catchup = 1;
2262 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2267 /* TSC scaling required - calculate ratio */
2268 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2269 user_tsc_khz, tsc_khz);
2271 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2272 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2277 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2281 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2283 u32 thresh_lo, thresh_hi;
2284 int use_scaling = 0;
2286 /* tsc_khz can be zero if TSC calibration fails */
2287 if (user_tsc_khz == 0) {
2288 /* set tsc_scaling_ratio to a safe value */
2289 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2293 /* Compute a scale to convert nanoseconds in TSC cycles */
2294 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2295 &vcpu->arch.virtual_tsc_shift,
2296 &vcpu->arch.virtual_tsc_mult);
2297 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2300 * Compute the variation in TSC rate which is acceptable
2301 * within the range of tolerance and decide if the
2302 * rate being applied is within that bounds of the hardware
2303 * rate. If so, no scaling or compensation need be done.
2305 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2306 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2307 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2308 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2311 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2314 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2316 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2317 vcpu->arch.virtual_tsc_mult,
2318 vcpu->arch.virtual_tsc_shift);
2319 tsc += vcpu->arch.this_tsc_write;
2323 static inline int gtod_is_based_on_tsc(int mode)
2325 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2328 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2330 #ifdef CONFIG_X86_64
2332 struct kvm_arch *ka = &vcpu->kvm->arch;
2333 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2335 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2336 atomic_read(&vcpu->kvm->online_vcpus));
2339 * Once the masterclock is enabled, always perform request in
2340 * order to update it.
2342 * In order to enable masterclock, the host clocksource must be TSC
2343 * and the vcpus need to have matched TSCs. When that happens,
2344 * perform request to enable masterclock.
2346 if (ka->use_master_clock ||
2347 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2348 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2350 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2351 atomic_read(&vcpu->kvm->online_vcpus),
2352 ka->use_master_clock, gtod->clock.vclock_mode);
2357 * Multiply tsc by a fixed point number represented by ratio.
2359 * The most significant 64-N bits (mult) of ratio represent the
2360 * integral part of the fixed point number; the remaining N bits
2361 * (frac) represent the fractional part, ie. ratio represents a fixed
2362 * point number (mult + frac * 2^(-N)).
2364 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2366 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2368 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2371 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2375 if (ratio != kvm_default_tsc_scaling_ratio)
2376 _tsc = __scale_tsc(ratio, tsc);
2380 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2382 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2386 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2388 return target_tsc - tsc;
2391 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2393 return vcpu->arch.l1_tsc_offset +
2394 kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2396 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2398 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2402 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2403 nested_offset = l1_offset;
2405 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2406 kvm_tsc_scaling_ratio_frac_bits);
2408 nested_offset += l2_offset;
2409 return nested_offset;
2411 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2413 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2415 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2416 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2417 kvm_tsc_scaling_ratio_frac_bits);
2419 return l1_multiplier;
2421 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2423 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2425 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2426 vcpu->arch.l1_tsc_offset,
2429 vcpu->arch.l1_tsc_offset = l1_offset;
2432 * If we are here because L1 chose not to trap WRMSR to TSC then
2433 * according to the spec this should set L1's TSC (as opposed to
2434 * setting L1's offset for L2).
2436 if (is_guest_mode(vcpu))
2437 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2439 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2440 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2442 vcpu->arch.tsc_offset = l1_offset;
2444 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2447 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2449 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2451 /* Userspace is changing the multiplier while L2 is active */
2452 if (is_guest_mode(vcpu))
2453 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2455 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2457 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2459 if (kvm_has_tsc_control)
2460 static_call(kvm_x86_write_tsc_multiplier)(
2461 vcpu, vcpu->arch.tsc_scaling_ratio);
2464 static inline bool kvm_check_tsc_unstable(void)
2466 #ifdef CONFIG_X86_64
2468 * TSC is marked unstable when we're running on Hyper-V,
2469 * 'TSC page' clocksource is good.
2471 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2474 return check_tsc_unstable();
2478 * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2479 * offset for the vcpu and tracks the TSC matching generation that the vcpu
2482 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2483 u64 ns, bool matched)
2485 struct kvm *kvm = vcpu->kvm;
2487 lockdep_assert_held(&kvm->arch.tsc_write_lock);
2490 * We also track th most recent recorded KHZ, write and time to
2491 * allow the matching interval to be extended at each write.
2493 kvm->arch.last_tsc_nsec = ns;
2494 kvm->arch.last_tsc_write = tsc;
2495 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2496 kvm->arch.last_tsc_offset = offset;
2498 vcpu->arch.last_guest_tsc = tsc;
2500 kvm_vcpu_write_tsc_offset(vcpu, offset);
2504 * We split periods of matched TSC writes into generations.
2505 * For each generation, we track the original measured
2506 * nanosecond time, offset, and write, so if TSCs are in
2507 * sync, we can match exact offset, and if not, we can match
2508 * exact software computation in compute_guest_tsc()
2510 * These values are tracked in kvm->arch.cur_xxx variables.
2512 kvm->arch.cur_tsc_generation++;
2513 kvm->arch.cur_tsc_nsec = ns;
2514 kvm->arch.cur_tsc_write = tsc;
2515 kvm->arch.cur_tsc_offset = offset;
2516 kvm->arch.nr_vcpus_matched_tsc = 0;
2517 } else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2518 kvm->arch.nr_vcpus_matched_tsc++;
2521 /* Keep track of which generation this VCPU has synchronized to */
2522 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2523 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2524 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2526 kvm_track_tsc_matching(vcpu);
2529 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2531 struct kvm *kvm = vcpu->kvm;
2532 u64 offset, ns, elapsed;
2533 unsigned long flags;
2534 bool matched = false;
2535 bool synchronizing = false;
2537 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2538 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2539 ns = get_kvmclock_base_ns();
2540 elapsed = ns - kvm->arch.last_tsc_nsec;
2542 if (vcpu->arch.virtual_tsc_khz) {
2545 * detection of vcpu initialization -- need to sync
2546 * with other vCPUs. This particularly helps to keep
2547 * kvm_clock stable after CPU hotplug
2549 synchronizing = true;
2551 u64 tsc_exp = kvm->arch.last_tsc_write +
2552 nsec_to_cycles(vcpu, elapsed);
2553 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2555 * Special case: TSC write with a small delta (1 second)
2556 * of virtual cycle time against real time is
2557 * interpreted as an attempt to synchronize the CPU.
2559 synchronizing = data < tsc_exp + tsc_hz &&
2560 data + tsc_hz > tsc_exp;
2565 * For a reliable TSC, we can match TSC offsets, and for an unstable
2566 * TSC, we add elapsed time in this computation. We could let the
2567 * compensation code attempt to catch up if we fall behind, but
2568 * it's better to try to match offsets from the beginning.
2570 if (synchronizing &&
2571 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2572 if (!kvm_check_tsc_unstable()) {
2573 offset = kvm->arch.cur_tsc_offset;
2575 u64 delta = nsec_to_cycles(vcpu, elapsed);
2577 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2582 __kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2583 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2586 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2589 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2590 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2593 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2595 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2596 WARN_ON(adjustment < 0);
2597 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2598 vcpu->arch.l1_tsc_scaling_ratio);
2599 adjust_tsc_offset_guest(vcpu, adjustment);
2602 #ifdef CONFIG_X86_64
2604 static u64 read_tsc(void)
2606 u64 ret = (u64)rdtsc_ordered();
2607 u64 last = pvclock_gtod_data.clock.cycle_last;
2609 if (likely(ret >= last))
2613 * GCC likes to generate cmov here, but this branch is extremely
2614 * predictable (it's just a function of time and the likely is
2615 * very likely) and there's a data dependence, so force GCC
2616 * to generate a branch instead. I don't barrier() because
2617 * we don't actually need a barrier, and if this function
2618 * ever gets inlined it will generate worse code.
2624 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2630 switch (clock->vclock_mode) {
2631 case VDSO_CLOCKMODE_HVCLOCK:
2632 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2634 if (tsc_pg_val != U64_MAX) {
2635 /* TSC page valid */
2636 *mode = VDSO_CLOCKMODE_HVCLOCK;
2637 v = (tsc_pg_val - clock->cycle_last) &
2640 /* TSC page invalid */
2641 *mode = VDSO_CLOCKMODE_NONE;
2644 case VDSO_CLOCKMODE_TSC:
2645 *mode = VDSO_CLOCKMODE_TSC;
2646 *tsc_timestamp = read_tsc();
2647 v = (*tsc_timestamp - clock->cycle_last) &
2651 *mode = VDSO_CLOCKMODE_NONE;
2654 if (*mode == VDSO_CLOCKMODE_NONE)
2655 *tsc_timestamp = v = 0;
2657 return v * clock->mult;
2660 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2662 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2668 seq = read_seqcount_begin(>od->seq);
2669 ns = gtod->raw_clock.base_cycles;
2670 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2671 ns >>= gtod->raw_clock.shift;
2672 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2673 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2679 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2681 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2687 seq = read_seqcount_begin(>od->seq);
2688 ts->tv_sec = gtod->wall_time_sec;
2689 ns = gtod->clock.base_cycles;
2690 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2691 ns >>= gtod->clock.shift;
2692 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2694 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2700 /* returns true if host is using TSC based clocksource */
2701 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2703 /* checked again under seqlock below */
2704 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2707 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2711 /* returns true if host is using TSC based clocksource */
2712 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2715 /* checked again under seqlock below */
2716 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2719 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2725 * Assuming a stable TSC across physical CPUS, and a stable TSC
2726 * across virtual CPUs, the following condition is possible.
2727 * Each numbered line represents an event visible to both
2728 * CPUs at the next numbered event.
2730 * "timespecX" represents host monotonic time. "tscX" represents
2733 * VCPU0 on CPU0 | VCPU1 on CPU1
2735 * 1. read timespec0,tsc0
2736 * 2. | timespec1 = timespec0 + N
2738 * 3. transition to guest | transition to guest
2739 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2740 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2741 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2743 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2746 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2748 * - 0 < N - M => M < N
2750 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2751 * always the case (the difference between two distinct xtime instances
2752 * might be smaller then the difference between corresponding TSC reads,
2753 * when updating guest vcpus pvclock areas).
2755 * To avoid that problem, do not allow visibility of distinct
2756 * system_timestamp/tsc_timestamp values simultaneously: use a master
2757 * copy of host monotonic time values. Update that master copy
2760 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2764 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2766 #ifdef CONFIG_X86_64
2767 struct kvm_arch *ka = &kvm->arch;
2769 bool host_tsc_clocksource, vcpus_matched;
2771 lockdep_assert_held(&kvm->arch.tsc_write_lock);
2772 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2773 atomic_read(&kvm->online_vcpus));
2776 * If the host uses TSC clock, then passthrough TSC as stable
2779 host_tsc_clocksource = kvm_get_time_and_clockread(
2780 &ka->master_kernel_ns,
2781 &ka->master_cycle_now);
2783 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2784 && !ka->backwards_tsc_observed
2785 && !ka->boot_vcpu_runs_old_kvmclock;
2787 if (ka->use_master_clock)
2788 atomic_set(&kvm_guest_has_master_clock, 1);
2790 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2791 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2796 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2798 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2801 static void __kvm_start_pvclock_update(struct kvm *kvm)
2803 raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2804 write_seqcount_begin(&kvm->arch.pvclock_sc);
2807 static void kvm_start_pvclock_update(struct kvm *kvm)
2809 kvm_make_mclock_inprogress_request(kvm);
2811 /* no guest entries from this point */
2812 __kvm_start_pvclock_update(kvm);
2815 static void kvm_end_pvclock_update(struct kvm *kvm)
2817 struct kvm_arch *ka = &kvm->arch;
2818 struct kvm_vcpu *vcpu;
2821 write_seqcount_end(&ka->pvclock_sc);
2822 raw_spin_unlock_irq(&ka->tsc_write_lock);
2823 kvm_for_each_vcpu(i, vcpu, kvm)
2824 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2826 /* guest entries allowed */
2827 kvm_for_each_vcpu(i, vcpu, kvm)
2828 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2831 static void kvm_update_masterclock(struct kvm *kvm)
2833 kvm_hv_invalidate_tsc_page(kvm);
2834 kvm_start_pvclock_update(kvm);
2835 pvclock_update_vm_gtod_copy(kvm);
2836 kvm_end_pvclock_update(kvm);
2839 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc. */
2840 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2842 struct kvm_arch *ka = &kvm->arch;
2843 struct pvclock_vcpu_time_info hv_clock;
2845 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2849 if (ka->use_master_clock && __this_cpu_read(cpu_tsc_khz)) {
2850 #ifdef CONFIG_X86_64
2851 struct timespec64 ts;
2853 if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
2854 data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
2855 data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
2858 data->host_tsc = rdtsc();
2860 data->flags |= KVM_CLOCK_TSC_STABLE;
2861 hv_clock.tsc_timestamp = ka->master_cycle_now;
2862 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2863 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2864 &hv_clock.tsc_shift,
2865 &hv_clock.tsc_to_system_mul);
2866 data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
2868 data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
2874 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2876 struct kvm_arch *ka = &kvm->arch;
2880 seq = read_seqcount_begin(&ka->pvclock_sc);
2881 __get_kvmclock(kvm, data);
2882 } while (read_seqcount_retry(&ka->pvclock_sc, seq));
2885 u64 get_kvmclock_ns(struct kvm *kvm)
2887 struct kvm_clock_data data;
2889 get_kvmclock(kvm, &data);
2893 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2894 struct gfn_to_hva_cache *cache,
2895 unsigned int offset)
2897 struct kvm_vcpu_arch *vcpu = &v->arch;
2898 struct pvclock_vcpu_time_info guest_hv_clock;
2900 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2901 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2904 /* This VCPU is paused, but it's legal for a guest to read another
2905 * VCPU's kvmclock, so we really have to follow the specification where
2906 * it says that version is odd if data is being modified, and even after
2909 * Version field updates must be kept separate. This is because
2910 * kvm_write_guest_cached might use a "rep movs" instruction, and
2911 * writes within a string instruction are weakly ordered. So there
2912 * are three writes overall.
2914 * As a small optimization, only write the version field in the first
2915 * and third write. The vcpu->pv_time cache is still valid, because the
2916 * version field is the first in the struct.
2918 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2920 if (guest_hv_clock.version & 1)
2921 ++guest_hv_clock.version; /* first time write, random junk */
2923 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2924 kvm_write_guest_offset_cached(v->kvm, cache,
2925 &vcpu->hv_clock, offset,
2926 sizeof(vcpu->hv_clock.version));
2930 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2931 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2933 if (vcpu->pvclock_set_guest_stopped_request) {
2934 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2935 vcpu->pvclock_set_guest_stopped_request = false;
2938 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2940 kvm_write_guest_offset_cached(v->kvm, cache,
2941 &vcpu->hv_clock, offset,
2942 sizeof(vcpu->hv_clock));
2946 vcpu->hv_clock.version++;
2947 kvm_write_guest_offset_cached(v->kvm, cache,
2948 &vcpu->hv_clock, offset,
2949 sizeof(vcpu->hv_clock.version));
2952 static int kvm_guest_time_update(struct kvm_vcpu *v)
2954 unsigned long flags, tgt_tsc_khz;
2956 struct kvm_vcpu_arch *vcpu = &v->arch;
2957 struct kvm_arch *ka = &v->kvm->arch;
2959 u64 tsc_timestamp, host_tsc;
2961 bool use_master_clock;
2967 * If the host uses TSC clock, then passthrough TSC as stable
2971 seq = read_seqcount_begin(&ka->pvclock_sc);
2972 use_master_clock = ka->use_master_clock;
2973 if (use_master_clock) {
2974 host_tsc = ka->master_cycle_now;
2975 kernel_ns = ka->master_kernel_ns;
2977 } while (read_seqcount_retry(&ka->pvclock_sc, seq));
2979 /* Keep irq disabled to prevent changes to the clock */
2980 local_irq_save(flags);
2981 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2982 if (unlikely(tgt_tsc_khz == 0)) {
2983 local_irq_restore(flags);
2984 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2987 if (!use_master_clock) {
2989 kernel_ns = get_kvmclock_base_ns();
2992 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2995 * We may have to catch up the TSC to match elapsed wall clock
2996 * time for two reasons, even if kvmclock is used.
2997 * 1) CPU could have been running below the maximum TSC rate
2998 * 2) Broken TSC compensation resets the base at each VCPU
2999 * entry to avoid unknown leaps of TSC even when running
3000 * again on the same CPU. This may cause apparent elapsed
3001 * time to disappear, and the guest to stand still or run
3004 if (vcpu->tsc_catchup) {
3005 u64 tsc = compute_guest_tsc(v, kernel_ns);
3006 if (tsc > tsc_timestamp) {
3007 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3008 tsc_timestamp = tsc;
3012 local_irq_restore(flags);
3014 /* With all the info we got, fill in the values */
3016 if (kvm_has_tsc_control)
3017 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
3018 v->arch.l1_tsc_scaling_ratio);
3020 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3021 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3022 &vcpu->hv_clock.tsc_shift,
3023 &vcpu->hv_clock.tsc_to_system_mul);
3024 vcpu->hw_tsc_khz = tgt_tsc_khz;
3027 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3028 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3029 vcpu->last_guest_tsc = tsc_timestamp;
3031 /* If the host uses TSC clocksource, then it is stable */
3033 if (use_master_clock)
3034 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3036 vcpu->hv_clock.flags = pvclock_flags;
3038 if (vcpu->pv_time_enabled)
3039 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
3040 if (vcpu->xen.vcpu_info_set)
3041 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
3042 offsetof(struct compat_vcpu_info, time));
3043 if (vcpu->xen.vcpu_time_info_set)
3044 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
3046 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3051 * kvmclock updates which are isolated to a given vcpu, such as
3052 * vcpu->cpu migration, should not allow system_timestamp from
3053 * the rest of the vcpus to remain static. Otherwise ntp frequency
3054 * correction applies to one vcpu's system_timestamp but not
3057 * So in those cases, request a kvmclock update for all vcpus.
3058 * We need to rate-limit these requests though, as they can
3059 * considerably slow guests that have a large number of vcpus.
3060 * The time for a remote vcpu to update its kvmclock is bound
3061 * by the delay we use to rate-limit the updates.
3064 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3066 static void kvmclock_update_fn(struct work_struct *work)
3069 struct delayed_work *dwork = to_delayed_work(work);
3070 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3071 kvmclock_update_work);
3072 struct kvm *kvm = container_of(ka, struct kvm, arch);
3073 struct kvm_vcpu *vcpu;
3075 kvm_for_each_vcpu(i, vcpu, kvm) {
3076 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3077 kvm_vcpu_kick(vcpu);
3081 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3083 struct kvm *kvm = v->kvm;
3085 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3086 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3087 KVMCLOCK_UPDATE_DELAY);
3090 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3092 static void kvmclock_sync_fn(struct work_struct *work)
3094 struct delayed_work *dwork = to_delayed_work(work);
3095 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3096 kvmclock_sync_work);
3097 struct kvm *kvm = container_of(ka, struct kvm, arch);
3099 if (!kvmclock_periodic_sync)
3102 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3103 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3104 KVMCLOCK_SYNC_PERIOD);
3108 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3110 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3112 /* McStatusWrEn enabled? */
3113 if (guest_cpuid_is_amd_or_hygon(vcpu))
3114 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3119 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3121 u64 mcg_cap = vcpu->arch.mcg_cap;
3122 unsigned bank_num = mcg_cap & 0xff;
3123 u32 msr = msr_info->index;
3124 u64 data = msr_info->data;
3127 case MSR_IA32_MCG_STATUS:
3128 vcpu->arch.mcg_status = data;
3130 case MSR_IA32_MCG_CTL:
3131 if (!(mcg_cap & MCG_CTL_P) &&
3132 (data || !msr_info->host_initiated))
3134 if (data != 0 && data != ~(u64)0)
3136 vcpu->arch.mcg_ctl = data;
3139 if (msr >= MSR_IA32_MC0_CTL &&
3140 msr < MSR_IA32_MCx_CTL(bank_num)) {
3141 u32 offset = array_index_nospec(
3142 msr - MSR_IA32_MC0_CTL,
3143 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3145 /* only 0 or all 1s can be written to IA32_MCi_CTL
3146 * some Linux kernels though clear bit 10 in bank 4 to
3147 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3148 * this to avoid an uncatched #GP in the guest
3150 if ((offset & 0x3) == 0 &&
3151 data != 0 && (data | (1 << 10)) != ~(u64)0)
3155 if (!msr_info->host_initiated &&
3156 (offset & 0x3) == 1 && data != 0) {
3157 if (!can_set_mci_status(vcpu))
3161 vcpu->arch.mce_banks[offset] = data;
3169 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3171 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3173 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3176 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3178 gpa_t gpa = data & ~0x3f;
3180 /* Bits 4:5 are reserved, Should be zero */
3184 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3185 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3188 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3189 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3192 if (!lapic_in_kernel(vcpu))
3193 return data ? 1 : 0;
3195 vcpu->arch.apf.msr_en_val = data;
3197 if (!kvm_pv_async_pf_enabled(vcpu)) {
3198 kvm_clear_async_pf_completion_queue(vcpu);
3199 kvm_async_pf_hash_reset(vcpu);
3203 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3207 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3208 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3210 kvm_async_pf_wakeup_all(vcpu);
3215 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3217 /* Bits 8-63 are reserved */
3221 if (!lapic_in_kernel(vcpu))
3224 vcpu->arch.apf.msr_int_val = data;
3226 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3231 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3233 vcpu->arch.pv_time_enabled = false;
3234 vcpu->arch.time = 0;
3237 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3239 ++vcpu->stat.tlb_flush;
3240 static_call(kvm_x86_tlb_flush_all)(vcpu);
3243 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3245 ++vcpu->stat.tlb_flush;
3249 * A TLB flush on behalf of the guest is equivalent to
3250 * INVPCID(all), toggling CR4.PGE, etc., which requires
3251 * a forced sync of the shadow page tables. Ensure all the
3252 * roots are synced and the guest TLB in hardware is clean.
3254 kvm_mmu_sync_roots(vcpu);
3255 kvm_mmu_sync_prev_roots(vcpu);
3258 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3261 static void record_steal_time(struct kvm_vcpu *vcpu)
3263 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3264 struct kvm_steal_time __user *st;
3265 struct kvm_memslots *slots;
3269 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3270 kvm_xen_runstate_set_running(vcpu);
3274 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3277 if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3280 slots = kvm_memslots(vcpu->kvm);
3282 if (unlikely(slots->generation != ghc->generation ||
3283 kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3284 gfn_t gfn = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3286 /* We rely on the fact that it fits in a single page. */
3287 BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3289 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gfn, sizeof(*st)) ||
3290 kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3294 st = (struct kvm_steal_time __user *)ghc->hva;
3295 if (!user_access_begin(st, sizeof(*st)))
3299 * Doing a TLB flush here, on the guest's behalf, can avoid
3302 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3303 u8 st_preempted = 0;
3306 asm volatile("1: xchgb %0, %2\n"
3309 _ASM_EXTABLE_UA(1b, 2b)
3310 : "+r" (st_preempted),
3312 : "m" (st->preempted));
3318 vcpu->arch.st.preempted = 0;
3320 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3321 st_preempted & KVM_VCPU_FLUSH_TLB);
3322 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3323 kvm_vcpu_flush_tlb_guest(vcpu);
3325 if (!user_access_begin(st, sizeof(*st)))
3328 unsafe_put_user(0, &st->preempted, out);
3329 vcpu->arch.st.preempted = 0;
3332 unsafe_get_user(version, &st->version, out);
3334 version += 1; /* first time write, random junk */
3337 unsafe_put_user(version, &st->version, out);
3341 unsafe_get_user(steal, &st->steal, out);
3342 steal += current->sched_info.run_delay -
3343 vcpu->arch.st.last_steal;
3344 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3345 unsafe_put_user(steal, &st->steal, out);
3348 unsafe_put_user(version, &st->version, out);
3353 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3356 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3359 u32 msr = msr_info->index;
3360 u64 data = msr_info->data;
3362 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3363 return kvm_xen_write_hypercall_page(vcpu, data);
3366 case MSR_AMD64_NB_CFG:
3367 case MSR_IA32_UCODE_WRITE:
3368 case MSR_VM_HSAVE_PA:
3369 case MSR_AMD64_PATCH_LOADER:
3370 case MSR_AMD64_BU_CFG2:
3371 case MSR_AMD64_DC_CFG:
3372 case MSR_F15H_EX_CFG:
3375 case MSR_IA32_UCODE_REV:
3376 if (msr_info->host_initiated)
3377 vcpu->arch.microcode_version = data;
3379 case MSR_IA32_ARCH_CAPABILITIES:
3380 if (!msr_info->host_initiated)
3382 vcpu->arch.arch_capabilities = data;
3384 case MSR_IA32_PERF_CAPABILITIES: {
3385 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3387 if (!msr_info->host_initiated)
3389 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3391 if (data & ~msr_ent.data)
3394 vcpu->arch.perf_capabilities = data;
3399 return set_efer(vcpu, msr_info);
3401 data &= ~(u64)0x40; /* ignore flush filter disable */
3402 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3403 data &= ~(u64)0x8; /* ignore TLB cache disable */
3405 /* Handle McStatusWrEn */
3406 if (data == BIT_ULL(18)) {
3407 vcpu->arch.msr_hwcr = data;
3408 } else if (data != 0) {
3409 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3414 case MSR_FAM10H_MMIO_CONF_BASE:
3416 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3421 case 0x200 ... 0x2ff:
3422 return kvm_mtrr_set_msr(vcpu, msr, data);
3423 case MSR_IA32_APICBASE:
3424 return kvm_set_apic_base(vcpu, msr_info);
3425 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3426 return kvm_x2apic_msr_write(vcpu, msr, data);
3427 case MSR_IA32_TSC_DEADLINE:
3428 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3430 case MSR_IA32_TSC_ADJUST:
3431 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3432 if (!msr_info->host_initiated) {
3433 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3434 adjust_tsc_offset_guest(vcpu, adj);
3435 /* Before back to guest, tsc_timestamp must be adjusted
3436 * as well, otherwise guest's percpu pvclock time could jump.
3438 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3440 vcpu->arch.ia32_tsc_adjust_msr = data;
3443 case MSR_IA32_MISC_ENABLE:
3444 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3445 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3446 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3448 vcpu->arch.ia32_misc_enable_msr = data;
3449 kvm_update_cpuid_runtime(vcpu);
3451 vcpu->arch.ia32_misc_enable_msr = data;
3454 case MSR_IA32_SMBASE:
3455 if (!msr_info->host_initiated)
3457 vcpu->arch.smbase = data;
3459 case MSR_IA32_POWER_CTL:
3460 vcpu->arch.msr_ia32_power_ctl = data;
3463 if (msr_info->host_initiated) {
3464 kvm_synchronize_tsc(vcpu, data);
3466 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3467 adjust_tsc_offset_guest(vcpu, adj);
3468 vcpu->arch.ia32_tsc_adjust_msr += adj;
3472 if (!msr_info->host_initiated &&
3473 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3476 * KVM supports exposing PT to the guest, but does not support
3477 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3478 * XSAVES/XRSTORS to save/restore PT MSRs.
3480 if (data & ~supported_xss)
3482 vcpu->arch.ia32_xss = data;
3485 if (!msr_info->host_initiated)
3487 vcpu->arch.smi_count = data;
3489 case MSR_KVM_WALL_CLOCK_NEW:
3490 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3493 vcpu->kvm->arch.wall_clock = data;
3494 kvm_write_wall_clock(vcpu->kvm, data, 0);
3496 case MSR_KVM_WALL_CLOCK:
3497 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3500 vcpu->kvm->arch.wall_clock = data;
3501 kvm_write_wall_clock(vcpu->kvm, data, 0);
3503 case MSR_KVM_SYSTEM_TIME_NEW:
3504 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3507 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3509 case MSR_KVM_SYSTEM_TIME:
3510 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3513 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3515 case MSR_KVM_ASYNC_PF_EN:
3516 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3519 if (kvm_pv_enable_async_pf(vcpu, data))
3522 case MSR_KVM_ASYNC_PF_INT:
3523 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3526 if (kvm_pv_enable_async_pf_int(vcpu, data))
3529 case MSR_KVM_ASYNC_PF_ACK:
3530 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3533 vcpu->arch.apf.pageready_pending = false;
3534 kvm_check_async_pf_completion(vcpu);
3537 case MSR_KVM_STEAL_TIME:
3538 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3541 if (unlikely(!sched_info_on()))
3544 if (data & KVM_STEAL_RESERVED_MASK)
3547 vcpu->arch.st.msr_val = data;
3549 if (!(data & KVM_MSR_ENABLED))
3552 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3555 case MSR_KVM_PV_EOI_EN:
3556 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3559 if (kvm_lapic_set_pv_eoi(vcpu, data, sizeof(u8)))
3563 case MSR_KVM_POLL_CONTROL:
3564 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3567 /* only enable bit supported */
3568 if (data & (-1ULL << 1))
3571 vcpu->arch.msr_kvm_poll_control = data;
3574 case MSR_IA32_MCG_CTL:
3575 case MSR_IA32_MCG_STATUS:
3576 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3577 return set_msr_mce(vcpu, msr_info);
3579 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3580 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3583 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3584 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3585 if (kvm_pmu_is_valid_msr(vcpu, msr))
3586 return kvm_pmu_set_msr(vcpu, msr_info);
3588 if (pr || data != 0)
3589 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3590 "0x%x data 0x%llx\n", msr, data);
3592 case MSR_K7_CLK_CTL:
3594 * Ignore all writes to this no longer documented MSR.
3595 * Writes are only relevant for old K7 processors,
3596 * all pre-dating SVM, but a recommended workaround from
3597 * AMD for these chips. It is possible to specify the
3598 * affected processor models on the command line, hence
3599 * the need to ignore the workaround.
3602 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3603 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3604 case HV_X64_MSR_SYNDBG_OPTIONS:
3605 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3606 case HV_X64_MSR_CRASH_CTL:
3607 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3608 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3609 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3610 case HV_X64_MSR_TSC_EMULATION_STATUS:
3611 return kvm_hv_set_msr_common(vcpu, msr, data,
3612 msr_info->host_initiated);
3613 case MSR_IA32_BBL_CR_CTL3:
3614 /* Drop writes to this legacy MSR -- see rdmsr
3615 * counterpart for further detail.
3617 if (report_ignored_msrs)
3618 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3621 case MSR_AMD64_OSVW_ID_LENGTH:
3622 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3624 vcpu->arch.osvw.length = data;
3626 case MSR_AMD64_OSVW_STATUS:
3627 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3629 vcpu->arch.osvw.status = data;
3631 case MSR_PLATFORM_INFO:
3632 if (!msr_info->host_initiated ||
3633 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3634 cpuid_fault_enabled(vcpu)))
3636 vcpu->arch.msr_platform_info = data;
3638 case MSR_MISC_FEATURES_ENABLES:
3639 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3640 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3641 !supports_cpuid_fault(vcpu)))
3643 vcpu->arch.msr_misc_features_enables = data;
3646 if (kvm_pmu_is_valid_msr(vcpu, msr))
3647 return kvm_pmu_set_msr(vcpu, msr_info);
3648 return KVM_MSR_RET_INVALID;
3652 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3654 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3657 u64 mcg_cap = vcpu->arch.mcg_cap;
3658 unsigned bank_num = mcg_cap & 0xff;
3661 case MSR_IA32_P5_MC_ADDR:
3662 case MSR_IA32_P5_MC_TYPE:
3665 case MSR_IA32_MCG_CAP:
3666 data = vcpu->arch.mcg_cap;
3668 case MSR_IA32_MCG_CTL:
3669 if (!(mcg_cap & MCG_CTL_P) && !host)
3671 data = vcpu->arch.mcg_ctl;
3673 case MSR_IA32_MCG_STATUS:
3674 data = vcpu->arch.mcg_status;
3677 if (msr >= MSR_IA32_MC0_CTL &&
3678 msr < MSR_IA32_MCx_CTL(bank_num)) {
3679 u32 offset = array_index_nospec(
3680 msr - MSR_IA32_MC0_CTL,
3681 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3683 data = vcpu->arch.mce_banks[offset];
3692 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3694 switch (msr_info->index) {
3695 case MSR_IA32_PLATFORM_ID:
3696 case MSR_IA32_EBL_CR_POWERON:
3697 case MSR_IA32_LASTBRANCHFROMIP:
3698 case MSR_IA32_LASTBRANCHTOIP:
3699 case MSR_IA32_LASTINTFROMIP:
3700 case MSR_IA32_LASTINTTOIP:
3701 case MSR_AMD64_SYSCFG:
3702 case MSR_K8_TSEG_ADDR:
3703 case MSR_K8_TSEG_MASK:
3704 case MSR_VM_HSAVE_PA:
3705 case MSR_K8_INT_PENDING_MSG:
3706 case MSR_AMD64_NB_CFG:
3707 case MSR_FAM10H_MMIO_CONF_BASE:
3708 case MSR_AMD64_BU_CFG2:
3709 case MSR_IA32_PERF_CTL:
3710 case MSR_AMD64_DC_CFG:
3711 case MSR_F15H_EX_CFG:
3713 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3714 * limit) MSRs. Just return 0, as we do not want to expose the host
3715 * data here. Do not conditionalize this on CPUID, as KVM does not do
3716 * so for existing CPU-specific MSRs.
3718 case MSR_RAPL_POWER_UNIT:
3719 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3720 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3721 case MSR_PKG_ENERGY_STATUS: /* Total package */
3722 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3725 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3726 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3727 return kvm_pmu_get_msr(vcpu, msr_info);
3728 if (!msr_info->host_initiated)
3732 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3733 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3734 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3735 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3736 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3737 return kvm_pmu_get_msr(vcpu, msr_info);
3740 case MSR_IA32_UCODE_REV:
3741 msr_info->data = vcpu->arch.microcode_version;
3743 case MSR_IA32_ARCH_CAPABILITIES:
3744 if (!msr_info->host_initiated &&
3745 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3747 msr_info->data = vcpu->arch.arch_capabilities;
3749 case MSR_IA32_PERF_CAPABILITIES:
3750 if (!msr_info->host_initiated &&
3751 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3753 msr_info->data = vcpu->arch.perf_capabilities;
3755 case MSR_IA32_POWER_CTL:
3756 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3758 case MSR_IA32_TSC: {
3760 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3761 * even when not intercepted. AMD manual doesn't explicitly
3762 * state this but appears to behave the same.
3764 * On userspace reads and writes, however, we unconditionally
3765 * return L1's TSC value to ensure backwards-compatible
3766 * behavior for migration.
3770 if (msr_info->host_initiated) {
3771 offset = vcpu->arch.l1_tsc_offset;
3772 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3774 offset = vcpu->arch.tsc_offset;
3775 ratio = vcpu->arch.tsc_scaling_ratio;
3778 msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3782 case 0x200 ... 0x2ff:
3783 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3784 case 0xcd: /* fsb frequency */
3788 * MSR_EBC_FREQUENCY_ID
3789 * Conservative value valid for even the basic CPU models.
3790 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3791 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3792 * and 266MHz for model 3, or 4. Set Core Clock
3793 * Frequency to System Bus Frequency Ratio to 1 (bits
3794 * 31:24) even though these are only valid for CPU
3795 * models > 2, however guests may end up dividing or
3796 * multiplying by zero otherwise.
3798 case MSR_EBC_FREQUENCY_ID:
3799 msr_info->data = 1 << 24;
3801 case MSR_IA32_APICBASE:
3802 msr_info->data = kvm_get_apic_base(vcpu);
3804 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3805 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3806 case MSR_IA32_TSC_DEADLINE:
3807 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3809 case MSR_IA32_TSC_ADJUST:
3810 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3812 case MSR_IA32_MISC_ENABLE:
3813 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3815 case MSR_IA32_SMBASE:
3816 if (!msr_info->host_initiated)
3818 msr_info->data = vcpu->arch.smbase;
3821 msr_info->data = vcpu->arch.smi_count;
3823 case MSR_IA32_PERF_STATUS:
3824 /* TSC increment by tick */
3825 msr_info->data = 1000ULL;
3826 /* CPU multiplier */
3827 msr_info->data |= (((uint64_t)4ULL) << 40);
3830 msr_info->data = vcpu->arch.efer;
3832 case MSR_KVM_WALL_CLOCK:
3833 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3836 msr_info->data = vcpu->kvm->arch.wall_clock;
3838 case MSR_KVM_WALL_CLOCK_NEW:
3839 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3842 msr_info->data = vcpu->kvm->arch.wall_clock;
3844 case MSR_KVM_SYSTEM_TIME:
3845 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3848 msr_info->data = vcpu->arch.time;
3850 case MSR_KVM_SYSTEM_TIME_NEW:
3851 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3854 msr_info->data = vcpu->arch.time;
3856 case MSR_KVM_ASYNC_PF_EN:
3857 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3860 msr_info->data = vcpu->arch.apf.msr_en_val;
3862 case MSR_KVM_ASYNC_PF_INT:
3863 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3866 msr_info->data = vcpu->arch.apf.msr_int_val;
3868 case MSR_KVM_ASYNC_PF_ACK:
3869 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3874 case MSR_KVM_STEAL_TIME:
3875 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3878 msr_info->data = vcpu->arch.st.msr_val;
3880 case MSR_KVM_PV_EOI_EN:
3881 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3884 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3886 case MSR_KVM_POLL_CONTROL:
3887 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3890 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3892 case MSR_IA32_P5_MC_ADDR:
3893 case MSR_IA32_P5_MC_TYPE:
3894 case MSR_IA32_MCG_CAP:
3895 case MSR_IA32_MCG_CTL:
3896 case MSR_IA32_MCG_STATUS:
3897 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3898 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3899 msr_info->host_initiated);
3901 if (!msr_info->host_initiated &&
3902 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3904 msr_info->data = vcpu->arch.ia32_xss;
3906 case MSR_K7_CLK_CTL:
3908 * Provide expected ramp-up count for K7. All other
3909 * are set to zero, indicating minimum divisors for
3912 * This prevents guest kernels on AMD host with CPU
3913 * type 6, model 8 and higher from exploding due to
3914 * the rdmsr failing.
3916 msr_info->data = 0x20000000;
3918 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3919 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3920 case HV_X64_MSR_SYNDBG_OPTIONS:
3921 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3922 case HV_X64_MSR_CRASH_CTL:
3923 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3924 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3925 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3926 case HV_X64_MSR_TSC_EMULATION_STATUS:
3927 return kvm_hv_get_msr_common(vcpu,
3928 msr_info->index, &msr_info->data,
3929 msr_info->host_initiated);
3930 case MSR_IA32_BBL_CR_CTL3:
3931 /* This legacy MSR exists but isn't fully documented in current
3932 * silicon. It is however accessed by winxp in very narrow
3933 * scenarios where it sets bit #19, itself documented as
3934 * a "reserved" bit. Best effort attempt to source coherent
3935 * read data here should the balance of the register be
3936 * interpreted by the guest:
3938 * L2 cache control register 3: 64GB range, 256KB size,
3939 * enabled, latency 0x1, configured
3941 msr_info->data = 0xbe702111;
3943 case MSR_AMD64_OSVW_ID_LENGTH:
3944 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3946 msr_info->data = vcpu->arch.osvw.length;
3948 case MSR_AMD64_OSVW_STATUS:
3949 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3951 msr_info->data = vcpu->arch.osvw.status;
3953 case MSR_PLATFORM_INFO:
3954 if (!msr_info->host_initiated &&
3955 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3957 msr_info->data = vcpu->arch.msr_platform_info;
3959 case MSR_MISC_FEATURES_ENABLES:
3960 msr_info->data = vcpu->arch.msr_misc_features_enables;
3963 msr_info->data = vcpu->arch.msr_hwcr;
3966 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3967 return kvm_pmu_get_msr(vcpu, msr_info);
3968 return KVM_MSR_RET_INVALID;
3972 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3975 * Read or write a bunch of msrs. All parameters are kernel addresses.
3977 * @return number of msrs set successfully.
3979 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3980 struct kvm_msr_entry *entries,
3981 int (*do_msr)(struct kvm_vcpu *vcpu,
3982 unsigned index, u64 *data))
3986 for (i = 0; i < msrs->nmsrs; ++i)
3987 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3994 * Read or write a bunch of msrs. Parameters are user addresses.
3996 * @return number of msrs set successfully.
3998 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3999 int (*do_msr)(struct kvm_vcpu *vcpu,
4000 unsigned index, u64 *data),
4003 struct kvm_msrs msrs;
4004 struct kvm_msr_entry *entries;
4009 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4013 if (msrs.nmsrs >= MAX_IO_MSRS)
4016 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4017 entries = memdup_user(user_msrs->entries, size);
4018 if (IS_ERR(entries)) {
4019 r = PTR_ERR(entries);
4023 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
4028 if (writeback && copy_to_user(user_msrs->entries, entries, size))
4039 static inline bool kvm_can_mwait_in_guest(void)
4041 return boot_cpu_has(X86_FEATURE_MWAIT) &&
4042 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
4043 boot_cpu_has(X86_FEATURE_ARAT);
4046 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4047 struct kvm_cpuid2 __user *cpuid_arg)
4049 struct kvm_cpuid2 cpuid;
4053 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4056 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4061 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4067 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4072 case KVM_CAP_IRQCHIP:
4074 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4075 case KVM_CAP_SET_TSS_ADDR:
4076 case KVM_CAP_EXT_CPUID:
4077 case KVM_CAP_EXT_EMUL_CPUID:
4078 case KVM_CAP_CLOCKSOURCE:
4080 case KVM_CAP_NOP_IO_DELAY:
4081 case KVM_CAP_MP_STATE:
4082 case KVM_CAP_SYNC_MMU:
4083 case KVM_CAP_USER_NMI:
4084 case KVM_CAP_REINJECT_CONTROL:
4085 case KVM_CAP_IRQ_INJECT_STATUS:
4086 case KVM_CAP_IOEVENTFD:
4087 case KVM_CAP_IOEVENTFD_NO_LENGTH:
4089 case KVM_CAP_PIT_STATE2:
4090 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4091 case KVM_CAP_VCPU_EVENTS:
4092 case KVM_CAP_HYPERV:
4093 case KVM_CAP_HYPERV_VAPIC:
4094 case KVM_CAP_HYPERV_SPIN:
4095 case KVM_CAP_HYPERV_SYNIC:
4096 case KVM_CAP_HYPERV_SYNIC2:
4097 case KVM_CAP_HYPERV_VP_INDEX:
4098 case KVM_CAP_HYPERV_EVENTFD:
4099 case KVM_CAP_HYPERV_TLBFLUSH:
4100 case KVM_CAP_HYPERV_SEND_IPI:
4101 case KVM_CAP_HYPERV_CPUID:
4102 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4103 case KVM_CAP_SYS_HYPERV_CPUID:
4104 case KVM_CAP_PCI_SEGMENT:
4105 case KVM_CAP_DEBUGREGS:
4106 case KVM_CAP_X86_ROBUST_SINGLESTEP:
4108 case KVM_CAP_ASYNC_PF:
4109 case KVM_CAP_ASYNC_PF_INT:
4110 case KVM_CAP_GET_TSC_KHZ:
4111 case KVM_CAP_KVMCLOCK_CTRL:
4112 case KVM_CAP_READONLY_MEM:
4113 case KVM_CAP_HYPERV_TIME:
4114 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4115 case KVM_CAP_TSC_DEADLINE_TIMER:
4116 case KVM_CAP_DISABLE_QUIRKS:
4117 case KVM_CAP_SET_BOOT_CPU_ID:
4118 case KVM_CAP_SPLIT_IRQCHIP:
4119 case KVM_CAP_IMMEDIATE_EXIT:
4120 case KVM_CAP_PMU_EVENT_FILTER:
4121 case KVM_CAP_GET_MSR_FEATURES:
4122 case KVM_CAP_MSR_PLATFORM_INFO:
4123 case KVM_CAP_EXCEPTION_PAYLOAD:
4124 case KVM_CAP_SET_GUEST_DEBUG:
4125 case KVM_CAP_LAST_CPU:
4126 case KVM_CAP_X86_USER_SPACE_MSR:
4127 case KVM_CAP_X86_MSR_FILTER:
4128 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4129 #ifdef CONFIG_X86_SGX_KVM
4130 case KVM_CAP_SGX_ATTRIBUTE:
4132 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4133 case KVM_CAP_SREGS2:
4134 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4135 case KVM_CAP_VCPU_ATTRIBUTES:
4138 case KVM_CAP_EXIT_HYPERCALL:
4139 r = KVM_EXIT_HYPERCALL_VALID_MASK;
4141 case KVM_CAP_SET_GUEST_DEBUG2:
4142 return KVM_GUESTDBG_VALID_MASK;
4143 #ifdef CONFIG_KVM_XEN
4144 case KVM_CAP_XEN_HVM:
4145 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4146 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4147 KVM_XEN_HVM_CONFIG_SHARED_INFO;
4148 if (sched_info_on())
4149 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4152 case KVM_CAP_SYNC_REGS:
4153 r = KVM_SYNC_X86_VALID_FIELDS;
4155 case KVM_CAP_ADJUST_CLOCK:
4156 r = KVM_CLOCK_VALID_FLAGS;
4158 case KVM_CAP_X86_DISABLE_EXITS:
4159 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4160 KVM_X86_DISABLE_EXITS_CSTATE;
4161 if(kvm_can_mwait_in_guest())
4162 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4164 case KVM_CAP_X86_SMM:
4165 /* SMBASE is usually relocated above 1M on modern chipsets,
4166 * and SMM handlers might indeed rely on 4G segment limits,
4167 * so do not report SMM to be available if real mode is
4168 * emulated via vm86 mode. Still, do not go to great lengths
4169 * to avoid userspace's usage of the feature, because it is a
4170 * fringe case that is not enabled except via specific settings
4171 * of the module parameters.
4173 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4176 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4178 case KVM_CAP_NR_VCPUS:
4179 r = KVM_SOFT_MAX_VCPUS;
4181 case KVM_CAP_MAX_VCPUS:
4184 case KVM_CAP_MAX_VCPU_ID:
4185 r = KVM_MAX_VCPU_IDS;
4187 case KVM_CAP_PV_MMU: /* obsolete */
4191 r = KVM_MAX_MCE_BANKS;
4194 r = boot_cpu_has(X86_FEATURE_XSAVE);
4196 case KVM_CAP_TSC_CONTROL:
4197 r = kvm_has_tsc_control;
4199 case KVM_CAP_X2APIC_API:
4200 r = KVM_X2APIC_API_VALID_FLAGS;
4202 case KVM_CAP_NESTED_STATE:
4203 r = kvm_x86_ops.nested_ops->get_state ?
4204 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4206 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4207 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4209 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4210 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4212 case KVM_CAP_SMALLER_MAXPHYADDR:
4213 r = (int) allow_smaller_maxphyaddr;
4215 case KVM_CAP_STEAL_TIME:
4216 r = sched_info_on();
4218 case KVM_CAP_X86_BUS_LOCK_EXIT:
4219 if (kvm_has_bus_lock_exit)
4220 r = KVM_BUS_LOCK_DETECTION_OFF |
4221 KVM_BUS_LOCK_DETECTION_EXIT;
4232 long kvm_arch_dev_ioctl(struct file *filp,
4233 unsigned int ioctl, unsigned long arg)
4235 void __user *argp = (void __user *)arg;
4239 case KVM_GET_MSR_INDEX_LIST: {
4240 struct kvm_msr_list __user *user_msr_list = argp;
4241 struct kvm_msr_list msr_list;
4245 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4248 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4249 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4252 if (n < msr_list.nmsrs)
4255 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4256 num_msrs_to_save * sizeof(u32)))
4258 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4260 num_emulated_msrs * sizeof(u32)))
4265 case KVM_GET_SUPPORTED_CPUID:
4266 case KVM_GET_EMULATED_CPUID: {
4267 struct kvm_cpuid2 __user *cpuid_arg = argp;
4268 struct kvm_cpuid2 cpuid;
4271 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4274 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4280 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4285 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4287 if (copy_to_user(argp, &kvm_mce_cap_supported,
4288 sizeof(kvm_mce_cap_supported)))
4292 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4293 struct kvm_msr_list __user *user_msr_list = argp;
4294 struct kvm_msr_list msr_list;
4298 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4301 msr_list.nmsrs = num_msr_based_features;
4302 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4305 if (n < msr_list.nmsrs)
4308 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4309 num_msr_based_features * sizeof(u32)))
4315 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4317 case KVM_GET_SUPPORTED_HV_CPUID:
4318 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4328 static void wbinvd_ipi(void *garbage)
4333 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4335 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4338 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4340 /* Address WBINVD may be executed by guest */
4341 if (need_emulate_wbinvd(vcpu)) {
4342 if (static_call(kvm_x86_has_wbinvd_exit)())
4343 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4344 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4345 smp_call_function_single(vcpu->cpu,
4346 wbinvd_ipi, NULL, 1);
4349 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4351 /* Save host pkru register if supported */
4352 vcpu->arch.host_pkru = read_pkru();
4354 /* Apply any externally detected TSC adjustments (due to suspend) */
4355 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4356 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4357 vcpu->arch.tsc_offset_adjustment = 0;
4358 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4361 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4362 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4363 rdtsc() - vcpu->arch.last_host_tsc;
4365 mark_tsc_unstable("KVM discovered backwards TSC");
4367 if (kvm_check_tsc_unstable()) {
4368 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4369 vcpu->arch.last_guest_tsc);
4370 kvm_vcpu_write_tsc_offset(vcpu, offset);
4371 vcpu->arch.tsc_catchup = 1;
4374 if (kvm_lapic_hv_timer_in_use(vcpu))
4375 kvm_lapic_restart_hv_timer(vcpu);
4378 * On a host with synchronized TSC, there is no need to update
4379 * kvmclock on vcpu->cpu migration
4381 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4382 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4383 if (vcpu->cpu != cpu)
4384 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4388 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4391 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4393 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4394 struct kvm_steal_time __user *st;
4395 struct kvm_memslots *slots;
4396 static const u8 preempted = KVM_VCPU_PREEMPTED;
4398 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4401 if (vcpu->arch.st.preempted)
4404 /* This happens on process exit */
4405 if (unlikely(current->mm != vcpu->kvm->mm))
4408 slots = kvm_memslots(vcpu->kvm);
4410 if (unlikely(slots->generation != ghc->generation ||
4411 kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4414 st = (struct kvm_steal_time __user *)ghc->hva;
4415 BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4417 if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4418 vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4420 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4423 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4427 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4428 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4431 * Take the srcu lock as memslots will be accessed to check the gfn
4432 * cache generation against the memslots generation.
4434 idx = srcu_read_lock(&vcpu->kvm->srcu);
4435 if (kvm_xen_msr_enabled(vcpu->kvm))
4436 kvm_xen_runstate_set_preempted(vcpu);
4438 kvm_steal_time_set_preempted(vcpu);
4439 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4441 static_call(kvm_x86_vcpu_put)(vcpu);
4442 vcpu->arch.last_host_tsc = rdtsc();
4445 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4446 struct kvm_lapic_state *s)
4448 if (vcpu->arch.apicv_active)
4449 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4451 return kvm_apic_get_state(vcpu, s);
4454 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4455 struct kvm_lapic_state *s)
4459 r = kvm_apic_set_state(vcpu, s);
4462 update_cr8_intercept(vcpu);
4467 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4470 * We can accept userspace's request for interrupt injection
4471 * as long as we have a place to store the interrupt number.
4472 * The actual injection will happen when the CPU is able to
4473 * deliver the interrupt.
4475 if (kvm_cpu_has_extint(vcpu))
4478 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4479 return (!lapic_in_kernel(vcpu) ||
4480 kvm_apic_accept_pic_intr(vcpu));
4483 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4486 * Do not cause an interrupt window exit if an exception
4487 * is pending or an event needs reinjection; userspace
4488 * might want to inject the interrupt manually using KVM_SET_REGS
4489 * or KVM_SET_SREGS. For that to work, we must be at an
4490 * instruction boundary and with no events half-injected.
4492 return (kvm_arch_interrupt_allowed(vcpu) &&
4493 kvm_cpu_accept_dm_intr(vcpu) &&
4494 !kvm_event_needs_reinjection(vcpu) &&
4495 !vcpu->arch.exception.pending);
4498 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4499 struct kvm_interrupt *irq)
4501 if (irq->irq >= KVM_NR_INTERRUPTS)
4504 if (!irqchip_in_kernel(vcpu->kvm)) {
4505 kvm_queue_interrupt(vcpu, irq->irq, false);
4506 kvm_make_request(KVM_REQ_EVENT, vcpu);
4511 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4512 * fail for in-kernel 8259.
4514 if (pic_in_kernel(vcpu->kvm))
4517 if (vcpu->arch.pending_external_vector != -1)
4520 vcpu->arch.pending_external_vector = irq->irq;
4521 kvm_make_request(KVM_REQ_EVENT, vcpu);
4525 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4527 kvm_inject_nmi(vcpu);
4532 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4534 kvm_make_request(KVM_REQ_SMI, vcpu);
4539 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4540 struct kvm_tpr_access_ctl *tac)
4544 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4548 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4552 unsigned bank_num = mcg_cap & 0xff, bank;
4555 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4557 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4560 vcpu->arch.mcg_cap = mcg_cap;
4561 /* Init IA32_MCG_CTL to all 1s */
4562 if (mcg_cap & MCG_CTL_P)
4563 vcpu->arch.mcg_ctl = ~(u64)0;
4564 /* Init IA32_MCi_CTL to all 1s */
4565 for (bank = 0; bank < bank_num; bank++)
4566 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4568 static_call(kvm_x86_setup_mce)(vcpu);
4573 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4574 struct kvm_x86_mce *mce)
4576 u64 mcg_cap = vcpu->arch.mcg_cap;
4577 unsigned bank_num = mcg_cap & 0xff;
4578 u64 *banks = vcpu->arch.mce_banks;
4580 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4583 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4584 * reporting is disabled
4586 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4587 vcpu->arch.mcg_ctl != ~(u64)0)
4589 banks += 4 * mce->bank;
4591 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4592 * reporting is disabled for the bank
4594 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4596 if (mce->status & MCI_STATUS_UC) {
4597 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4598 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4599 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4602 if (banks[1] & MCI_STATUS_VAL)
4603 mce->status |= MCI_STATUS_OVER;
4604 banks[2] = mce->addr;
4605 banks[3] = mce->misc;
4606 vcpu->arch.mcg_status = mce->mcg_status;
4607 banks[1] = mce->status;
4608 kvm_queue_exception(vcpu, MC_VECTOR);
4609 } else if (!(banks[1] & MCI_STATUS_VAL)
4610 || !(banks[1] & MCI_STATUS_UC)) {
4611 if (banks[1] & MCI_STATUS_VAL)
4612 mce->status |= MCI_STATUS_OVER;
4613 banks[2] = mce->addr;
4614 banks[3] = mce->misc;
4615 banks[1] = mce->status;
4617 banks[1] |= MCI_STATUS_OVER;
4621 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4622 struct kvm_vcpu_events *events)
4626 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4630 * In guest mode, payload delivery should be deferred,
4631 * so that the L1 hypervisor can intercept #PF before
4632 * CR2 is modified (or intercept #DB before DR6 is
4633 * modified under nVMX). Unless the per-VM capability,
4634 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4635 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4636 * opportunistically defer the exception payload, deliver it if the
4637 * capability hasn't been requested before processing a
4638 * KVM_GET_VCPU_EVENTS.
4640 if (!vcpu->kvm->arch.exception_payload_enabled &&
4641 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4642 kvm_deliver_exception_payload(vcpu);
4645 * The API doesn't provide the instruction length for software
4646 * exceptions, so don't report them. As long as the guest RIP
4647 * isn't advanced, we should expect to encounter the exception
4650 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4651 events->exception.injected = 0;
4652 events->exception.pending = 0;
4654 events->exception.injected = vcpu->arch.exception.injected;
4655 events->exception.pending = vcpu->arch.exception.pending;
4657 * For ABI compatibility, deliberately conflate
4658 * pending and injected exceptions when
4659 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4661 if (!vcpu->kvm->arch.exception_payload_enabled)
4662 events->exception.injected |=
4663 vcpu->arch.exception.pending;
4665 events->exception.nr = vcpu->arch.exception.nr;
4666 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4667 events->exception.error_code = vcpu->arch.exception.error_code;
4668 events->exception_has_payload = vcpu->arch.exception.has_payload;
4669 events->exception_payload = vcpu->arch.exception.payload;
4671 events->interrupt.injected =
4672 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4673 events->interrupt.nr = vcpu->arch.interrupt.nr;
4674 events->interrupt.soft = 0;
4675 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4677 events->nmi.injected = vcpu->arch.nmi_injected;
4678 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4679 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4680 events->nmi.pad = 0;
4682 events->sipi_vector = 0; /* never valid when reporting to user space */
4684 events->smi.smm = is_smm(vcpu);
4685 events->smi.pending = vcpu->arch.smi_pending;
4686 events->smi.smm_inside_nmi =
4687 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4688 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4690 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4691 | KVM_VCPUEVENT_VALID_SHADOW
4692 | KVM_VCPUEVENT_VALID_SMM);
4693 if (vcpu->kvm->arch.exception_payload_enabled)
4694 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4696 memset(&events->reserved, 0, sizeof(events->reserved));
4699 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4701 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4702 struct kvm_vcpu_events *events)
4704 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4705 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4706 | KVM_VCPUEVENT_VALID_SHADOW
4707 | KVM_VCPUEVENT_VALID_SMM
4708 | KVM_VCPUEVENT_VALID_PAYLOAD))
4711 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4712 if (!vcpu->kvm->arch.exception_payload_enabled)
4714 if (events->exception.pending)
4715 events->exception.injected = 0;
4717 events->exception_has_payload = 0;
4719 events->exception.pending = 0;
4720 events->exception_has_payload = 0;
4723 if ((events->exception.injected || events->exception.pending) &&
4724 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4727 /* INITs are latched while in SMM */
4728 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4729 (events->smi.smm || events->smi.pending) &&
4730 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4734 vcpu->arch.exception.injected = events->exception.injected;
4735 vcpu->arch.exception.pending = events->exception.pending;
4736 vcpu->arch.exception.nr = events->exception.nr;
4737 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4738 vcpu->arch.exception.error_code = events->exception.error_code;
4739 vcpu->arch.exception.has_payload = events->exception_has_payload;
4740 vcpu->arch.exception.payload = events->exception_payload;
4742 vcpu->arch.interrupt.injected = events->interrupt.injected;
4743 vcpu->arch.interrupt.nr = events->interrupt.nr;
4744 vcpu->arch.interrupt.soft = events->interrupt.soft;
4745 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4746 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4747 events->interrupt.shadow);
4749 vcpu->arch.nmi_injected = events->nmi.injected;
4750 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4751 vcpu->arch.nmi_pending = events->nmi.pending;
4752 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4754 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4755 lapic_in_kernel(vcpu))
4756 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4758 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4759 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4760 kvm_smm_changed(vcpu, events->smi.smm);
4762 vcpu->arch.smi_pending = events->smi.pending;
4764 if (events->smi.smm) {
4765 if (events->smi.smm_inside_nmi)
4766 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4768 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4771 if (lapic_in_kernel(vcpu)) {
4772 if (events->smi.latched_init)
4773 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4775 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4779 kvm_make_request(KVM_REQ_EVENT, vcpu);
4784 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4785 struct kvm_debugregs *dbgregs)
4789 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4790 kvm_get_dr(vcpu, 6, &val);
4792 dbgregs->dr7 = vcpu->arch.dr7;
4794 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4797 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4798 struct kvm_debugregs *dbgregs)
4803 if (!kvm_dr6_valid(dbgregs->dr6))
4805 if (!kvm_dr7_valid(dbgregs->dr7))
4808 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4809 kvm_update_dr0123(vcpu);
4810 vcpu->arch.dr6 = dbgregs->dr6;
4811 vcpu->arch.dr7 = dbgregs->dr7;
4812 kvm_update_dr7(vcpu);
4817 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4819 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4821 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4822 u64 xstate_bv = xsave->header.xfeatures;
4826 * Copy legacy XSAVE area, to avoid complications with CPUID
4827 * leaves 0 and 1 in the loop below.
4829 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4832 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4833 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4836 * Copy each region from the possibly compacted offset to the
4837 * non-compacted offset.
4839 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4841 u32 size, offset, ecx, edx;
4842 u64 xfeature_mask = valid & -valid;
4843 int xfeature_nr = fls64(xfeature_mask) - 1;
4846 cpuid_count(XSTATE_CPUID, xfeature_nr,
4847 &size, &offset, &ecx, &edx);
4849 if (xfeature_nr == XFEATURE_PKRU) {
4850 memcpy(dest + offset, &vcpu->arch.pkru,
4851 sizeof(vcpu->arch.pkru));
4853 src = get_xsave_addr(xsave, xfeature_nr);
4855 memcpy(dest + offset, src, size);
4858 valid -= xfeature_mask;
4862 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4864 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4865 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4869 * Copy legacy XSAVE area, to avoid complications with CPUID
4870 * leaves 0 and 1 in the loop below.
4872 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4874 /* Set XSTATE_BV and possibly XCOMP_BV. */
4875 xsave->header.xfeatures = xstate_bv;
4876 if (boot_cpu_has(X86_FEATURE_XSAVES))
4877 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4880 * Copy each region from the non-compacted offset to the
4881 * possibly compacted offset.
4883 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4885 u32 size, offset, ecx, edx;
4886 u64 xfeature_mask = valid & -valid;
4887 int xfeature_nr = fls64(xfeature_mask) - 1;
4889 cpuid_count(XSTATE_CPUID, xfeature_nr,
4890 &size, &offset, &ecx, &edx);
4892 if (xfeature_nr == XFEATURE_PKRU) {
4893 memcpy(&vcpu->arch.pkru, src + offset,
4894 sizeof(vcpu->arch.pkru));
4896 void *dest = get_xsave_addr(xsave, xfeature_nr);
4899 memcpy(dest, src + offset, size);
4902 valid -= xfeature_mask;
4906 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4907 struct kvm_xsave *guest_xsave)
4909 if (!vcpu->arch.guest_fpu)
4912 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4913 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4914 fill_xsave((u8 *) guest_xsave->region, vcpu);
4916 memcpy(guest_xsave->region,
4917 &vcpu->arch.guest_fpu->state.fxsave,
4918 sizeof(struct fxregs_state));
4919 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4920 XFEATURE_MASK_FPSSE;
4924 #define XSAVE_MXCSR_OFFSET 24
4926 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4927 struct kvm_xsave *guest_xsave)
4932 if (!vcpu->arch.guest_fpu)
4935 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4936 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4938 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4940 * Here we allow setting states that are not present in
4941 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4942 * with old userspace.
4944 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4946 load_xsave(vcpu, (u8 *)guest_xsave->region);
4948 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4949 mxcsr & ~mxcsr_feature_mask)
4951 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4952 guest_xsave->region, sizeof(struct fxregs_state));
4957 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4958 struct kvm_xcrs *guest_xcrs)
4960 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4961 guest_xcrs->nr_xcrs = 0;
4965 guest_xcrs->nr_xcrs = 1;
4966 guest_xcrs->flags = 0;
4967 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4968 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4971 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4972 struct kvm_xcrs *guest_xcrs)
4976 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4979 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4982 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4983 /* Only support XCR0 currently */
4984 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4985 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4986 guest_xcrs->xcrs[i].value);
4995 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4996 * stopped by the hypervisor. This function will be called from the host only.
4997 * EINVAL is returned when the host attempts to set the flag for a guest that
4998 * does not support pv clocks.
5000 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
5002 if (!vcpu->arch.pv_time_enabled)
5004 vcpu->arch.pvclock_set_guest_stopped_request = true;
5005 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5009 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
5010 struct kvm_device_attr *attr)
5014 switch (attr->attr) {
5015 case KVM_VCPU_TSC_OFFSET:
5025 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
5026 struct kvm_device_attr *attr)
5028 u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
5031 if ((u64)(unsigned long)uaddr != attr->addr)
5034 switch (attr->attr) {
5035 case KVM_VCPU_TSC_OFFSET:
5037 if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
5048 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
5049 struct kvm_device_attr *attr)
5051 u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
5052 struct kvm *kvm = vcpu->kvm;
5055 if ((u64)(unsigned long)uaddr != attr->addr)
5058 switch (attr->attr) {
5059 case KVM_VCPU_TSC_OFFSET: {
5060 u64 offset, tsc, ns;
5061 unsigned long flags;
5065 if (get_user(offset, uaddr))
5068 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
5070 matched = (vcpu->arch.virtual_tsc_khz &&
5071 kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
5072 kvm->arch.last_tsc_offset == offset);
5074 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
5075 ns = get_kvmclock_base_ns();
5077 __kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
5078 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
5090 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
5094 struct kvm_device_attr attr;
5097 if (copy_from_user(&attr, argp, sizeof(attr)))
5100 if (attr.group != KVM_VCPU_TSC_CTRL)
5104 case KVM_HAS_DEVICE_ATTR:
5105 r = kvm_arch_tsc_has_attr(vcpu, &attr);
5107 case KVM_GET_DEVICE_ATTR:
5108 r = kvm_arch_tsc_get_attr(vcpu, &attr);
5110 case KVM_SET_DEVICE_ATTR:
5111 r = kvm_arch_tsc_set_attr(vcpu, &attr);
5118 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5119 struct kvm_enable_cap *cap)
5122 uint16_t vmcs_version;
5123 void __user *user_ptr;
5129 case KVM_CAP_HYPERV_SYNIC2:
5134 case KVM_CAP_HYPERV_SYNIC:
5135 if (!irqchip_in_kernel(vcpu->kvm))
5137 return kvm_hv_activate_synic(vcpu, cap->cap ==
5138 KVM_CAP_HYPERV_SYNIC2);
5139 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5140 if (!kvm_x86_ops.nested_ops->enable_evmcs)
5142 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5144 user_ptr = (void __user *)(uintptr_t)cap->args[0];
5145 if (copy_to_user(user_ptr, &vmcs_version,
5146 sizeof(vmcs_version)))
5150 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5151 if (!kvm_x86_ops.enable_direct_tlbflush)
5154 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
5156 case KVM_CAP_HYPERV_ENFORCE_CPUID:
5157 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5159 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5160 vcpu->arch.pv_cpuid.enforce = cap->args[0];
5161 if (vcpu->arch.pv_cpuid.enforce)
5162 kvm_update_pv_runtime(vcpu);
5170 long kvm_arch_vcpu_ioctl(struct file *filp,
5171 unsigned int ioctl, unsigned long arg)
5173 struct kvm_vcpu *vcpu = filp->private_data;
5174 void __user *argp = (void __user *)arg;
5177 struct kvm_sregs2 *sregs2;
5178 struct kvm_lapic_state *lapic;
5179 struct kvm_xsave *xsave;
5180 struct kvm_xcrs *xcrs;
5188 case KVM_GET_LAPIC: {
5190 if (!lapic_in_kernel(vcpu))
5192 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5193 GFP_KERNEL_ACCOUNT);
5198 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5202 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5207 case KVM_SET_LAPIC: {
5209 if (!lapic_in_kernel(vcpu))
5211 u.lapic = memdup_user(argp, sizeof(*u.lapic));
5212 if (IS_ERR(u.lapic)) {
5213 r = PTR_ERR(u.lapic);
5217 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5220 case KVM_INTERRUPT: {
5221 struct kvm_interrupt irq;
5224 if (copy_from_user(&irq, argp, sizeof(irq)))
5226 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5230 r = kvm_vcpu_ioctl_nmi(vcpu);
5234 r = kvm_vcpu_ioctl_smi(vcpu);
5237 case KVM_SET_CPUID: {
5238 struct kvm_cpuid __user *cpuid_arg = argp;
5239 struct kvm_cpuid cpuid;
5242 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5244 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5247 case KVM_SET_CPUID2: {
5248 struct kvm_cpuid2 __user *cpuid_arg = argp;
5249 struct kvm_cpuid2 cpuid;
5252 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5254 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5255 cpuid_arg->entries);
5258 case KVM_GET_CPUID2: {
5259 struct kvm_cpuid2 __user *cpuid_arg = argp;
5260 struct kvm_cpuid2 cpuid;
5263 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5265 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5266 cpuid_arg->entries);
5270 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5275 case KVM_GET_MSRS: {
5276 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5277 r = msr_io(vcpu, argp, do_get_msr, 1);
5278 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5281 case KVM_SET_MSRS: {
5282 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5283 r = msr_io(vcpu, argp, do_set_msr, 0);
5284 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5287 case KVM_TPR_ACCESS_REPORTING: {
5288 struct kvm_tpr_access_ctl tac;
5291 if (copy_from_user(&tac, argp, sizeof(tac)))
5293 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5297 if (copy_to_user(argp, &tac, sizeof(tac)))
5302 case KVM_SET_VAPIC_ADDR: {
5303 struct kvm_vapic_addr va;
5307 if (!lapic_in_kernel(vcpu))
5310 if (copy_from_user(&va, argp, sizeof(va)))
5312 idx = srcu_read_lock(&vcpu->kvm->srcu);
5313 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5314 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5317 case KVM_X86_SETUP_MCE: {
5321 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5323 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5326 case KVM_X86_SET_MCE: {
5327 struct kvm_x86_mce mce;
5330 if (copy_from_user(&mce, argp, sizeof(mce)))
5332 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5335 case KVM_GET_VCPU_EVENTS: {
5336 struct kvm_vcpu_events events;
5338 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5341 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5346 case KVM_SET_VCPU_EVENTS: {
5347 struct kvm_vcpu_events events;
5350 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5353 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5356 case KVM_GET_DEBUGREGS: {
5357 struct kvm_debugregs dbgregs;
5359 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5362 if (copy_to_user(argp, &dbgregs,
5363 sizeof(struct kvm_debugregs)))
5368 case KVM_SET_DEBUGREGS: {
5369 struct kvm_debugregs dbgregs;
5372 if (copy_from_user(&dbgregs, argp,
5373 sizeof(struct kvm_debugregs)))
5376 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5379 case KVM_GET_XSAVE: {
5380 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5385 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5388 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5393 case KVM_SET_XSAVE: {
5394 u.xsave = memdup_user(argp, sizeof(*u.xsave));
5395 if (IS_ERR(u.xsave)) {
5396 r = PTR_ERR(u.xsave);
5400 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5403 case KVM_GET_XCRS: {
5404 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5409 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5412 if (copy_to_user(argp, u.xcrs,
5413 sizeof(struct kvm_xcrs)))
5418 case KVM_SET_XCRS: {
5419 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5420 if (IS_ERR(u.xcrs)) {
5421 r = PTR_ERR(u.xcrs);
5425 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5428 case KVM_SET_TSC_KHZ: {
5432 user_tsc_khz = (u32)arg;
5434 if (kvm_has_tsc_control &&
5435 user_tsc_khz >= kvm_max_guest_tsc_khz)
5438 if (user_tsc_khz == 0)
5439 user_tsc_khz = tsc_khz;
5441 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5446 case KVM_GET_TSC_KHZ: {
5447 r = vcpu->arch.virtual_tsc_khz;
5450 case KVM_KVMCLOCK_CTRL: {
5451 r = kvm_set_guest_paused(vcpu);
5454 case KVM_ENABLE_CAP: {
5455 struct kvm_enable_cap cap;
5458 if (copy_from_user(&cap, argp, sizeof(cap)))
5460 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5463 case KVM_GET_NESTED_STATE: {
5464 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5468 if (!kvm_x86_ops.nested_ops->get_state)
5471 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5473 if (get_user(user_data_size, &user_kvm_nested_state->size))
5476 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5481 if (r > user_data_size) {
5482 if (put_user(r, &user_kvm_nested_state->size))
5492 case KVM_SET_NESTED_STATE: {
5493 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5494 struct kvm_nested_state kvm_state;
5498 if (!kvm_x86_ops.nested_ops->set_state)
5502 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5506 if (kvm_state.size < sizeof(kvm_state))
5509 if (kvm_state.flags &
5510 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5511 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5512 | KVM_STATE_NESTED_GIF_SET))
5515 /* nested_run_pending implies guest_mode. */
5516 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5517 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5520 idx = srcu_read_lock(&vcpu->kvm->srcu);
5521 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5522 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5525 case KVM_GET_SUPPORTED_HV_CPUID:
5526 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5528 #ifdef CONFIG_KVM_XEN
5529 case KVM_XEN_VCPU_GET_ATTR: {
5530 struct kvm_xen_vcpu_attr xva;
5533 if (copy_from_user(&xva, argp, sizeof(xva)))
5535 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5536 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5540 case KVM_XEN_VCPU_SET_ATTR: {
5541 struct kvm_xen_vcpu_attr xva;
5544 if (copy_from_user(&xva, argp, sizeof(xva)))
5546 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5550 case KVM_GET_SREGS2: {
5551 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5555 __get_sregs2(vcpu, u.sregs2);
5557 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5562 case KVM_SET_SREGS2: {
5563 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5564 if (IS_ERR(u.sregs2)) {
5565 r = PTR_ERR(u.sregs2);
5569 r = __set_sregs2(vcpu, u.sregs2);
5572 case KVM_HAS_DEVICE_ATTR:
5573 case KVM_GET_DEVICE_ATTR:
5574 case KVM_SET_DEVICE_ATTR:
5575 r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
5587 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5589 return VM_FAULT_SIGBUS;
5592 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5596 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5598 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5602 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5605 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5608 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5609 unsigned long kvm_nr_mmu_pages)
5611 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5614 mutex_lock(&kvm->slots_lock);
5616 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5617 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5619 mutex_unlock(&kvm->slots_lock);
5623 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5625 return kvm->arch.n_max_mmu_pages;
5628 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5630 struct kvm_pic *pic = kvm->arch.vpic;
5634 switch (chip->chip_id) {
5635 case KVM_IRQCHIP_PIC_MASTER:
5636 memcpy(&chip->chip.pic, &pic->pics[0],
5637 sizeof(struct kvm_pic_state));
5639 case KVM_IRQCHIP_PIC_SLAVE:
5640 memcpy(&chip->chip.pic, &pic->pics[1],
5641 sizeof(struct kvm_pic_state));
5643 case KVM_IRQCHIP_IOAPIC:
5644 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5653 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5655 struct kvm_pic *pic = kvm->arch.vpic;
5659 switch (chip->chip_id) {
5660 case KVM_IRQCHIP_PIC_MASTER:
5661 spin_lock(&pic->lock);
5662 memcpy(&pic->pics[0], &chip->chip.pic,
5663 sizeof(struct kvm_pic_state));
5664 spin_unlock(&pic->lock);
5666 case KVM_IRQCHIP_PIC_SLAVE:
5667 spin_lock(&pic->lock);
5668 memcpy(&pic->pics[1], &chip->chip.pic,
5669 sizeof(struct kvm_pic_state));
5670 spin_unlock(&pic->lock);
5672 case KVM_IRQCHIP_IOAPIC:
5673 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5679 kvm_pic_update_irq(pic);
5683 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5685 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5687 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5689 mutex_lock(&kps->lock);
5690 memcpy(ps, &kps->channels, sizeof(*ps));
5691 mutex_unlock(&kps->lock);
5695 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5698 struct kvm_pit *pit = kvm->arch.vpit;
5700 mutex_lock(&pit->pit_state.lock);
5701 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5702 for (i = 0; i < 3; i++)
5703 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5704 mutex_unlock(&pit->pit_state.lock);
5708 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5710 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5711 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5712 sizeof(ps->channels));
5713 ps->flags = kvm->arch.vpit->pit_state.flags;
5714 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5715 memset(&ps->reserved, 0, sizeof(ps->reserved));
5719 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5723 u32 prev_legacy, cur_legacy;
5724 struct kvm_pit *pit = kvm->arch.vpit;
5726 mutex_lock(&pit->pit_state.lock);
5727 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5728 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5729 if (!prev_legacy && cur_legacy)
5731 memcpy(&pit->pit_state.channels, &ps->channels,
5732 sizeof(pit->pit_state.channels));
5733 pit->pit_state.flags = ps->flags;
5734 for (i = 0; i < 3; i++)
5735 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5737 mutex_unlock(&pit->pit_state.lock);
5741 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5742 struct kvm_reinject_control *control)
5744 struct kvm_pit *pit = kvm->arch.vpit;
5746 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5747 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5748 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5750 mutex_lock(&pit->pit_state.lock);
5751 kvm_pit_set_reinject(pit, control->pit_reinject);
5752 mutex_unlock(&pit->pit_state.lock);
5757 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5761 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5762 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5763 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5766 struct kvm_vcpu *vcpu;
5769 kvm_for_each_vcpu(i, vcpu, kvm)
5770 kvm_vcpu_kick(vcpu);
5773 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5776 if (!irqchip_in_kernel(kvm))
5779 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5780 irq_event->irq, irq_event->level,
5785 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5786 struct kvm_enable_cap *cap)
5794 case KVM_CAP_DISABLE_QUIRKS:
5795 kvm->arch.disabled_quirks = cap->args[0];
5798 case KVM_CAP_SPLIT_IRQCHIP: {
5799 mutex_lock(&kvm->lock);
5801 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5802 goto split_irqchip_unlock;
5804 if (irqchip_in_kernel(kvm))
5805 goto split_irqchip_unlock;
5806 if (kvm->created_vcpus)
5807 goto split_irqchip_unlock;
5808 r = kvm_setup_empty_irq_routing(kvm);
5810 goto split_irqchip_unlock;
5811 /* Pairs with irqchip_in_kernel. */
5813 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5814 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5816 split_irqchip_unlock:
5817 mutex_unlock(&kvm->lock);
5820 case KVM_CAP_X2APIC_API:
5822 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5825 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5826 kvm->arch.x2apic_format = true;
5827 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5828 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5832 case KVM_CAP_X86_DISABLE_EXITS:
5834 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5837 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5838 kvm_can_mwait_in_guest())
5839 kvm->arch.mwait_in_guest = true;
5840 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5841 kvm->arch.hlt_in_guest = true;
5842 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5843 kvm->arch.pause_in_guest = true;
5844 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5845 kvm->arch.cstate_in_guest = true;
5848 case KVM_CAP_MSR_PLATFORM_INFO:
5849 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5852 case KVM_CAP_EXCEPTION_PAYLOAD:
5853 kvm->arch.exception_payload_enabled = cap->args[0];
5856 case KVM_CAP_X86_USER_SPACE_MSR:
5857 kvm->arch.user_space_msr_mask = cap->args[0];
5860 case KVM_CAP_X86_BUS_LOCK_EXIT:
5862 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5865 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5866 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5869 if (kvm_has_bus_lock_exit &&
5870 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5871 kvm->arch.bus_lock_detection_enabled = true;
5874 #ifdef CONFIG_X86_SGX_KVM
5875 case KVM_CAP_SGX_ATTRIBUTE: {
5876 unsigned long allowed_attributes = 0;
5878 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5882 /* KVM only supports the PROVISIONKEY privileged attribute. */
5883 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5884 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5885 kvm->arch.sgx_provisioning_allowed = true;
5891 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5893 if (kvm_x86_ops.vm_copy_enc_context_from)
5894 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5896 case KVM_CAP_EXIT_HYPERCALL:
5897 if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5901 kvm->arch.hypercall_exit_enabled = cap->args[0];
5904 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5906 if (cap->args[0] & ~1)
5908 kvm->arch.exit_on_emulation_error = cap->args[0];
5918 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5920 struct kvm_x86_msr_filter *msr_filter;
5922 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5926 msr_filter->default_allow = default_allow;
5930 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5937 for (i = 0; i < msr_filter->count; i++)
5938 kfree(msr_filter->ranges[i].bitmap);
5943 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5944 struct kvm_msr_filter_range *user_range)
5946 unsigned long *bitmap = NULL;
5949 if (!user_range->nmsrs)
5952 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5955 if (!user_range->flags)
5958 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5959 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5962 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5964 return PTR_ERR(bitmap);
5966 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5967 .flags = user_range->flags,
5968 .base = user_range->base,
5969 .nmsrs = user_range->nmsrs,
5973 msr_filter->count++;
5977 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5979 struct kvm_msr_filter __user *user_msr_filter = argp;
5980 struct kvm_x86_msr_filter *new_filter, *old_filter;
5981 struct kvm_msr_filter filter;
5987 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5990 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5991 empty &= !filter.ranges[i].nmsrs;
5993 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5994 if (empty && !default_allow)
5997 new_filter = kvm_alloc_msr_filter(default_allow);
6001 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
6002 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
6004 kvm_free_msr_filter(new_filter);
6009 mutex_lock(&kvm->lock);
6011 /* The per-VM filter is protected by kvm->lock... */
6012 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
6014 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
6015 synchronize_srcu(&kvm->srcu);
6017 kvm_free_msr_filter(old_filter);
6019 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
6020 mutex_unlock(&kvm->lock);
6025 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
6026 static int kvm_arch_suspend_notifier(struct kvm *kvm)
6028 struct kvm_vcpu *vcpu;
6031 mutex_lock(&kvm->lock);
6032 kvm_for_each_vcpu(i, vcpu, kvm) {
6033 if (!vcpu->arch.pv_time_enabled)
6036 ret = kvm_set_guest_paused(vcpu);
6038 kvm_err("Failed to pause guest VCPU%d: %d\n",
6039 vcpu->vcpu_id, ret);
6043 mutex_unlock(&kvm->lock);
6045 return ret ? NOTIFY_BAD : NOTIFY_DONE;
6048 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
6051 case PM_HIBERNATION_PREPARE:
6052 case PM_SUSPEND_PREPARE:
6053 return kvm_arch_suspend_notifier(kvm);
6058 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
6060 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
6062 struct kvm_clock_data data = { 0 };
6064 get_kvmclock(kvm, &data);
6065 if (copy_to_user(argp, &data, sizeof(data)))
6071 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
6073 struct kvm_arch *ka = &kvm->arch;
6074 struct kvm_clock_data data;
6077 if (copy_from_user(&data, argp, sizeof(data)))
6081 * Only KVM_CLOCK_REALTIME is used, but allow passing the
6082 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
6084 if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
6087 kvm_hv_invalidate_tsc_page(kvm);
6088 kvm_start_pvclock_update(kvm);
6089 pvclock_update_vm_gtod_copy(kvm);
6092 * This pairs with kvm_guest_time_update(): when masterclock is
6093 * in use, we use master_kernel_ns + kvmclock_offset to set
6094 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6095 * is slightly ahead) here we risk going negative on unsigned
6096 * 'system_time' when 'data.clock' is very small.
6098 if (data.flags & KVM_CLOCK_REALTIME) {
6099 u64 now_real_ns = ktime_get_real_ns();
6102 * Avoid stepping the kvmclock backwards.
6104 if (now_real_ns > data.realtime)
6105 data.clock += now_real_ns - data.realtime;
6108 if (ka->use_master_clock)
6109 now_raw_ns = ka->master_kernel_ns;
6111 now_raw_ns = get_kvmclock_base_ns();
6112 ka->kvmclock_offset = data.clock - now_raw_ns;
6113 kvm_end_pvclock_update(kvm);
6117 long kvm_arch_vm_ioctl(struct file *filp,
6118 unsigned int ioctl, unsigned long arg)
6120 struct kvm *kvm = filp->private_data;
6121 void __user *argp = (void __user *)arg;
6124 * This union makes it completely explicit to gcc-3.x
6125 * that these two variables' stack usage should be
6126 * combined, not added together.
6129 struct kvm_pit_state ps;
6130 struct kvm_pit_state2 ps2;
6131 struct kvm_pit_config pit_config;
6135 case KVM_SET_TSS_ADDR:
6136 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6138 case KVM_SET_IDENTITY_MAP_ADDR: {
6141 mutex_lock(&kvm->lock);
6143 if (kvm->created_vcpus)
6144 goto set_identity_unlock;
6146 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6147 goto set_identity_unlock;
6148 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6149 set_identity_unlock:
6150 mutex_unlock(&kvm->lock);
6153 case KVM_SET_NR_MMU_PAGES:
6154 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6156 case KVM_GET_NR_MMU_PAGES:
6157 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
6159 case KVM_CREATE_IRQCHIP: {
6160 mutex_lock(&kvm->lock);
6163 if (irqchip_in_kernel(kvm))
6164 goto create_irqchip_unlock;
6167 if (kvm->created_vcpus)
6168 goto create_irqchip_unlock;
6170 r = kvm_pic_init(kvm);
6172 goto create_irqchip_unlock;
6174 r = kvm_ioapic_init(kvm);
6176 kvm_pic_destroy(kvm);
6177 goto create_irqchip_unlock;
6180 r = kvm_setup_default_irq_routing(kvm);
6182 kvm_ioapic_destroy(kvm);
6183 kvm_pic_destroy(kvm);
6184 goto create_irqchip_unlock;
6186 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6188 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6189 create_irqchip_unlock:
6190 mutex_unlock(&kvm->lock);
6193 case KVM_CREATE_PIT:
6194 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6196 case KVM_CREATE_PIT2:
6198 if (copy_from_user(&u.pit_config, argp,
6199 sizeof(struct kvm_pit_config)))
6202 mutex_lock(&kvm->lock);
6205 goto create_pit_unlock;
6207 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6211 mutex_unlock(&kvm->lock);
6213 case KVM_GET_IRQCHIP: {
6214 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6215 struct kvm_irqchip *chip;
6217 chip = memdup_user(argp, sizeof(*chip));
6224 if (!irqchip_kernel(kvm))
6225 goto get_irqchip_out;
6226 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6228 goto get_irqchip_out;
6230 if (copy_to_user(argp, chip, sizeof(*chip)))
6231 goto get_irqchip_out;
6237 case KVM_SET_IRQCHIP: {
6238 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6239 struct kvm_irqchip *chip;
6241 chip = memdup_user(argp, sizeof(*chip));
6248 if (!irqchip_kernel(kvm))
6249 goto set_irqchip_out;
6250 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6257 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6260 if (!kvm->arch.vpit)
6262 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6266 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6273 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6275 mutex_lock(&kvm->lock);
6277 if (!kvm->arch.vpit)
6279 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6281 mutex_unlock(&kvm->lock);
6284 case KVM_GET_PIT2: {
6286 if (!kvm->arch.vpit)
6288 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6292 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6297 case KVM_SET_PIT2: {
6299 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6301 mutex_lock(&kvm->lock);
6303 if (!kvm->arch.vpit)
6305 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6307 mutex_unlock(&kvm->lock);
6310 case KVM_REINJECT_CONTROL: {
6311 struct kvm_reinject_control control;
6313 if (copy_from_user(&control, argp, sizeof(control)))
6316 if (!kvm->arch.vpit)
6318 r = kvm_vm_ioctl_reinject(kvm, &control);
6321 case KVM_SET_BOOT_CPU_ID:
6323 mutex_lock(&kvm->lock);
6324 if (kvm->created_vcpus)
6327 kvm->arch.bsp_vcpu_id = arg;
6328 mutex_unlock(&kvm->lock);
6330 #ifdef CONFIG_KVM_XEN
6331 case KVM_XEN_HVM_CONFIG: {
6332 struct kvm_xen_hvm_config xhc;
6334 if (copy_from_user(&xhc, argp, sizeof(xhc)))
6336 r = kvm_xen_hvm_config(kvm, &xhc);
6339 case KVM_XEN_HVM_GET_ATTR: {
6340 struct kvm_xen_hvm_attr xha;
6343 if (copy_from_user(&xha, argp, sizeof(xha)))
6345 r = kvm_xen_hvm_get_attr(kvm, &xha);
6346 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6350 case KVM_XEN_HVM_SET_ATTR: {
6351 struct kvm_xen_hvm_attr xha;
6354 if (copy_from_user(&xha, argp, sizeof(xha)))
6356 r = kvm_xen_hvm_set_attr(kvm, &xha);
6361 r = kvm_vm_ioctl_set_clock(kvm, argp);
6364 r = kvm_vm_ioctl_get_clock(kvm, argp);
6366 case KVM_MEMORY_ENCRYPT_OP: {
6368 if (kvm_x86_ops.mem_enc_op)
6369 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6372 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6373 struct kvm_enc_region region;
6376 if (copy_from_user(®ion, argp, sizeof(region)))
6380 if (kvm_x86_ops.mem_enc_reg_region)
6381 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
6384 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6385 struct kvm_enc_region region;
6388 if (copy_from_user(®ion, argp, sizeof(region)))
6392 if (kvm_x86_ops.mem_enc_unreg_region)
6393 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
6396 case KVM_HYPERV_EVENTFD: {
6397 struct kvm_hyperv_eventfd hvevfd;
6400 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6402 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6405 case KVM_SET_PMU_EVENT_FILTER:
6406 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6408 case KVM_X86_SET_MSR_FILTER:
6409 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6418 static void kvm_init_msr_list(void)
6420 struct x86_pmu_capability x86_pmu;
6424 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6425 "Please update the fixed PMCs in msrs_to_saved_all[]");
6427 perf_get_x86_pmu_capability(&x86_pmu);
6429 num_msrs_to_save = 0;
6430 num_emulated_msrs = 0;
6431 num_msr_based_features = 0;
6433 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6434 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6438 * Even MSRs that are valid in the host may not be exposed
6439 * to the guests in some cases.
6441 switch (msrs_to_save_all[i]) {
6442 case MSR_IA32_BNDCFGS:
6443 if (!kvm_mpx_supported())
6447 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6448 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6451 case MSR_IA32_UMWAIT_CONTROL:
6452 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6455 case MSR_IA32_RTIT_CTL:
6456 case MSR_IA32_RTIT_STATUS:
6457 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6460 case MSR_IA32_RTIT_CR3_MATCH:
6461 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6462 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6465 case MSR_IA32_RTIT_OUTPUT_BASE:
6466 case MSR_IA32_RTIT_OUTPUT_MASK:
6467 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6468 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6469 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6472 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6473 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6474 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6475 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6478 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6479 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6480 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6483 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6484 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6485 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6492 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6495 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6496 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6499 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6502 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6503 struct kvm_msr_entry msr;
6505 msr.index = msr_based_features_all[i];
6506 if (kvm_get_msr_feature(&msr))
6509 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6513 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6521 if (!(lapic_in_kernel(vcpu) &&
6522 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6523 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6534 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6541 if (!(lapic_in_kernel(vcpu) &&
6542 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6544 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6546 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6556 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6557 struct kvm_segment *var, int seg)
6559 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6562 void kvm_get_segment(struct kvm_vcpu *vcpu,
6563 struct kvm_segment *var, int seg)
6565 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6568 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6569 struct x86_exception *exception)
6573 BUG_ON(!mmu_is_nested(vcpu));
6575 /* NPT walks are always user-walks */
6576 access |= PFERR_USER_MASK;
6577 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6582 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6583 struct x86_exception *exception)
6585 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6586 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6588 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6590 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6591 struct x86_exception *exception)
6593 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6594 access |= PFERR_FETCH_MASK;
6595 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6598 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6599 struct x86_exception *exception)
6601 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6602 access |= PFERR_WRITE_MASK;
6603 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6605 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6607 /* uses this to access any guest's mapped memory without checking CPL */
6608 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6609 struct x86_exception *exception)
6611 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6614 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6615 struct kvm_vcpu *vcpu, u32 access,
6616 struct x86_exception *exception)
6619 int r = X86EMUL_CONTINUE;
6622 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6624 unsigned offset = addr & (PAGE_SIZE-1);
6625 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6628 if (gpa == UNMAPPED_GVA)
6629 return X86EMUL_PROPAGATE_FAULT;
6630 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6633 r = X86EMUL_IO_NEEDED;
6645 /* used for instruction fetching */
6646 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6647 gva_t addr, void *val, unsigned int bytes,
6648 struct x86_exception *exception)
6650 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6651 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6655 /* Inline kvm_read_guest_virt_helper for speed. */
6656 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6658 if (unlikely(gpa == UNMAPPED_GVA))
6659 return X86EMUL_PROPAGATE_FAULT;
6661 offset = addr & (PAGE_SIZE-1);
6662 if (WARN_ON(offset + bytes > PAGE_SIZE))
6663 bytes = (unsigned)PAGE_SIZE - offset;
6664 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6666 if (unlikely(ret < 0))
6667 return X86EMUL_IO_NEEDED;
6669 return X86EMUL_CONTINUE;
6672 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6673 gva_t addr, void *val, unsigned int bytes,
6674 struct x86_exception *exception)
6676 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6679 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6680 * is returned, but our callers are not ready for that and they blindly
6681 * call kvm_inject_page_fault. Ensure that they at least do not leak
6682 * uninitialized kernel stack memory into cr2 and error code.
6684 memset(exception, 0, sizeof(*exception));
6685 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6688 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6690 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6691 gva_t addr, void *val, unsigned int bytes,
6692 struct x86_exception *exception, bool system)
6694 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6697 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6698 access |= PFERR_USER_MASK;
6700 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6703 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6704 unsigned long addr, void *val, unsigned int bytes)
6706 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6707 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6709 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6712 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6713 struct kvm_vcpu *vcpu, u32 access,
6714 struct x86_exception *exception)
6717 int r = X86EMUL_CONTINUE;
6720 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6723 unsigned offset = addr & (PAGE_SIZE-1);
6724 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6727 if (gpa == UNMAPPED_GVA)
6728 return X86EMUL_PROPAGATE_FAULT;
6729 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6731 r = X86EMUL_IO_NEEDED;
6743 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6744 unsigned int bytes, struct x86_exception *exception,
6747 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6748 u32 access = PFERR_WRITE_MASK;
6750 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6751 access |= PFERR_USER_MASK;
6753 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6757 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6758 unsigned int bytes, struct x86_exception *exception)
6760 /* kvm_write_guest_virt_system can pull in tons of pages. */
6761 vcpu->arch.l1tf_flush_l1d = true;
6763 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6764 PFERR_WRITE_MASK, exception);
6766 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6768 int handle_ud(struct kvm_vcpu *vcpu)
6770 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6771 int emul_type = EMULTYPE_TRAP_UD;
6772 char sig[5]; /* ud2; .ascii "kvm" */
6773 struct x86_exception e;
6775 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6778 if (force_emulation_prefix &&
6779 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6780 sig, sizeof(sig), &e) == 0 &&
6781 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6782 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6783 emul_type = EMULTYPE_TRAP_UD_FORCED;
6786 return kvm_emulate_instruction(vcpu, emul_type);
6788 EXPORT_SYMBOL_GPL(handle_ud);
6790 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6791 gpa_t gpa, bool write)
6793 /* For APIC access vmexit */
6794 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6797 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6798 trace_vcpu_match_mmio(gva, gpa, write, true);
6805 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6806 gpa_t *gpa, struct x86_exception *exception,
6809 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6810 | (write ? PFERR_WRITE_MASK : 0);
6813 * currently PKRU is only applied to ept enabled guest so
6814 * there is no pkey in EPT page table for L1 guest or EPT
6815 * shadow page table for L2 guest.
6817 if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6818 !permission_fault(vcpu, vcpu->arch.walk_mmu,
6819 vcpu->arch.mmio_access, 0, access))) {
6820 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6821 (gva & (PAGE_SIZE - 1));
6822 trace_vcpu_match_mmio(gva, *gpa, write, false);
6826 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6828 if (*gpa == UNMAPPED_GVA)
6831 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6834 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6835 const void *val, int bytes)
6839 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6842 kvm_page_track_write(vcpu, gpa, val, bytes);
6846 struct read_write_emulator_ops {
6847 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6849 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6850 void *val, int bytes);
6851 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6852 int bytes, void *val);
6853 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6854 void *val, int bytes);
6858 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6860 if (vcpu->mmio_read_completed) {
6861 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6862 vcpu->mmio_fragments[0].gpa, val);
6863 vcpu->mmio_read_completed = 0;
6870 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6871 void *val, int bytes)
6873 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6876 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6877 void *val, int bytes)
6879 return emulator_write_phys(vcpu, gpa, val, bytes);
6882 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6884 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6885 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6888 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6889 void *val, int bytes)
6891 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6892 return X86EMUL_IO_NEEDED;
6895 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6896 void *val, int bytes)
6898 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6900 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6901 return X86EMUL_CONTINUE;
6904 static const struct read_write_emulator_ops read_emultor = {
6905 .read_write_prepare = read_prepare,
6906 .read_write_emulate = read_emulate,
6907 .read_write_mmio = vcpu_mmio_read,
6908 .read_write_exit_mmio = read_exit_mmio,
6911 static const struct read_write_emulator_ops write_emultor = {
6912 .read_write_emulate = write_emulate,
6913 .read_write_mmio = write_mmio,
6914 .read_write_exit_mmio = write_exit_mmio,
6918 static int emulator_read_write_onepage(unsigned long addr, void *val,
6920 struct x86_exception *exception,
6921 struct kvm_vcpu *vcpu,
6922 const struct read_write_emulator_ops *ops)
6926 bool write = ops->write;
6927 struct kvm_mmio_fragment *frag;
6928 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6931 * If the exit was due to a NPF we may already have a GPA.
6932 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6933 * Note, this cannot be used on string operations since string
6934 * operation using rep will only have the initial GPA from the NPF
6937 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6938 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6939 gpa = ctxt->gpa_val;
6940 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6942 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6944 return X86EMUL_PROPAGATE_FAULT;
6947 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6948 return X86EMUL_CONTINUE;
6951 * Is this MMIO handled locally?
6953 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6954 if (handled == bytes)
6955 return X86EMUL_CONTINUE;
6961 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6962 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6966 return X86EMUL_CONTINUE;
6969 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6971 void *val, unsigned int bytes,
6972 struct x86_exception *exception,
6973 const struct read_write_emulator_ops *ops)
6975 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6979 if (ops->read_write_prepare &&
6980 ops->read_write_prepare(vcpu, val, bytes))
6981 return X86EMUL_CONTINUE;
6983 vcpu->mmio_nr_fragments = 0;
6985 /* Crossing a page boundary? */
6986 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6989 now = -addr & ~PAGE_MASK;
6990 rc = emulator_read_write_onepage(addr, val, now, exception,
6993 if (rc != X86EMUL_CONTINUE)
6996 if (ctxt->mode != X86EMUL_MODE_PROT64)
7002 rc = emulator_read_write_onepage(addr, val, bytes, exception,
7004 if (rc != X86EMUL_CONTINUE)
7007 if (!vcpu->mmio_nr_fragments)
7010 gpa = vcpu->mmio_fragments[0].gpa;
7012 vcpu->mmio_needed = 1;
7013 vcpu->mmio_cur_fragment = 0;
7015 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
7016 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
7017 vcpu->run->exit_reason = KVM_EXIT_MMIO;
7018 vcpu->run->mmio.phys_addr = gpa;
7020 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
7023 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
7027 struct x86_exception *exception)
7029 return emulator_read_write(ctxt, addr, val, bytes,
7030 exception, &read_emultor);
7033 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
7037 struct x86_exception *exception)
7039 return emulator_read_write(ctxt, addr, (void *)val, bytes,
7040 exception, &write_emultor);
7043 #define CMPXCHG_TYPE(t, ptr, old, new) \
7044 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
7046 #ifdef CONFIG_X86_64
7047 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
7049 # define CMPXCHG64(ptr, old, new) \
7050 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
7053 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
7058 struct x86_exception *exception)
7060 struct kvm_host_map map;
7061 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7067 /* guests cmpxchg8b have to be emulated atomically */
7068 if (bytes > 8 || (bytes & (bytes - 1)))
7071 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
7073 if (gpa == UNMAPPED_GVA ||
7074 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7078 * Emulate the atomic as a straight write to avoid #AC if SLD is
7079 * enabled in the host and the access splits a cache line.
7081 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
7082 page_line_mask = ~(cache_line_size() - 1);
7084 page_line_mask = PAGE_MASK;
7086 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
7089 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
7092 kaddr = map.hva + offset_in_page(gpa);
7096 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
7099 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
7102 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
7105 exchanged = CMPXCHG64(kaddr, old, new);
7111 kvm_vcpu_unmap(vcpu, &map, true);
7114 return X86EMUL_CMPXCHG_FAILED;
7116 kvm_page_track_write(vcpu, gpa, new, bytes);
7118 return X86EMUL_CONTINUE;
7121 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
7123 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7126 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
7130 for (i = 0; i < vcpu->arch.pio.count; i++) {
7131 if (vcpu->arch.pio.in)
7132 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
7133 vcpu->arch.pio.size, pd);
7135 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
7136 vcpu->arch.pio.port, vcpu->arch.pio.size,
7140 pd += vcpu->arch.pio.size;
7145 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7146 unsigned short port, void *val,
7147 unsigned int count, bool in)
7149 vcpu->arch.pio.port = port;
7150 vcpu->arch.pio.in = in;
7151 vcpu->arch.pio.count = count;
7152 vcpu->arch.pio.size = size;
7154 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7155 vcpu->arch.pio.count = 0;
7159 vcpu->run->exit_reason = KVM_EXIT_IO;
7160 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7161 vcpu->run->io.size = size;
7162 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7163 vcpu->run->io.count = count;
7164 vcpu->run->io.port = port;
7169 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7170 unsigned short port, void *val, unsigned int count)
7174 if (vcpu->arch.pio.count)
7177 memset(vcpu->arch.pio_data, 0, size * count);
7179 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
7182 memcpy(val, vcpu->arch.pio_data, size * count);
7183 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
7184 vcpu->arch.pio.count = 0;
7191 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7192 int size, unsigned short port, void *val,
7195 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
7199 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7200 unsigned short port, const void *val,
7203 memcpy(vcpu->arch.pio_data, val, size * count);
7204 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
7205 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
7208 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7209 int size, unsigned short port,
7210 const void *val, unsigned int count)
7212 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7215 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7217 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7220 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7222 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7225 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7227 if (!need_emulate_wbinvd(vcpu))
7228 return X86EMUL_CONTINUE;
7230 if (static_call(kvm_x86_has_wbinvd_exit)()) {
7231 int cpu = get_cpu();
7233 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7234 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7235 wbinvd_ipi, NULL, 1);
7237 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7240 return X86EMUL_CONTINUE;
7243 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7245 kvm_emulate_wbinvd_noskip(vcpu);
7246 return kvm_skip_emulated_instruction(vcpu);
7248 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7252 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7254 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7257 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7258 unsigned long *dest)
7260 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7263 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7264 unsigned long value)
7267 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7270 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7272 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7275 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7277 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7278 unsigned long value;
7282 value = kvm_read_cr0(vcpu);
7285 value = vcpu->arch.cr2;
7288 value = kvm_read_cr3(vcpu);
7291 value = kvm_read_cr4(vcpu);
7294 value = kvm_get_cr8(vcpu);
7297 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7304 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7306 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7311 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7314 vcpu->arch.cr2 = val;
7317 res = kvm_set_cr3(vcpu, val);
7320 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7323 res = kvm_set_cr8(vcpu, val);
7326 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7333 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7335 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7338 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7340 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7343 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7345 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7348 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7350 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7353 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7355 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7358 static unsigned long emulator_get_cached_segment_base(
7359 struct x86_emulate_ctxt *ctxt, int seg)
7361 return get_segment_base(emul_to_vcpu(ctxt), seg);
7364 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7365 struct desc_struct *desc, u32 *base3,
7368 struct kvm_segment var;
7370 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7371 *selector = var.selector;
7374 memset(desc, 0, sizeof(*desc));
7382 set_desc_limit(desc, var.limit);
7383 set_desc_base(desc, (unsigned long)var.base);
7384 #ifdef CONFIG_X86_64
7386 *base3 = var.base >> 32;
7388 desc->type = var.type;
7390 desc->dpl = var.dpl;
7391 desc->p = var.present;
7392 desc->avl = var.avl;
7400 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7401 struct desc_struct *desc, u32 base3,
7404 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7405 struct kvm_segment var;
7407 var.selector = selector;
7408 var.base = get_desc_base(desc);
7409 #ifdef CONFIG_X86_64
7410 var.base |= ((u64)base3) << 32;
7412 var.limit = get_desc_limit(desc);
7414 var.limit = (var.limit << 12) | 0xfff;
7415 var.type = desc->type;
7416 var.dpl = desc->dpl;
7421 var.avl = desc->avl;
7422 var.present = desc->p;
7423 var.unusable = !var.present;
7426 kvm_set_segment(vcpu, &var, seg);
7430 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7431 u32 msr_index, u64 *pdata)
7433 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7436 r = kvm_get_msr(vcpu, msr_index, pdata);
7438 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7439 /* Bounce to user space */
7440 return X86EMUL_IO_NEEDED;
7446 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7447 u32 msr_index, u64 data)
7449 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7452 r = kvm_set_msr(vcpu, msr_index, data);
7454 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7455 /* Bounce to user space */
7456 return X86EMUL_IO_NEEDED;
7462 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7464 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7466 return vcpu->arch.smbase;
7469 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7471 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7473 vcpu->arch.smbase = smbase;
7476 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7479 if (kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc))
7484 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7485 u32 pmc, u64 *pdata)
7487 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7490 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7492 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7495 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7496 struct x86_instruction_info *info,
7497 enum x86_intercept_stage stage)
7499 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7503 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7504 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7507 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7510 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7512 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7515 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7517 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7520 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7522 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7525 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7527 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7530 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7532 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7535 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7537 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7540 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7542 return emul_to_vcpu(ctxt)->arch.hflags;
7545 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7547 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7549 kvm_smm_changed(vcpu, false);
7552 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7553 const char *smstate)
7555 return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7558 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7560 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7563 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7565 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7568 static const struct x86_emulate_ops emulate_ops = {
7569 .read_gpr = emulator_read_gpr,
7570 .write_gpr = emulator_write_gpr,
7571 .read_std = emulator_read_std,
7572 .write_std = emulator_write_std,
7573 .read_phys = kvm_read_guest_phys_system,
7574 .fetch = kvm_fetch_guest_virt,
7575 .read_emulated = emulator_read_emulated,
7576 .write_emulated = emulator_write_emulated,
7577 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7578 .invlpg = emulator_invlpg,
7579 .pio_in_emulated = emulator_pio_in_emulated,
7580 .pio_out_emulated = emulator_pio_out_emulated,
7581 .get_segment = emulator_get_segment,
7582 .set_segment = emulator_set_segment,
7583 .get_cached_segment_base = emulator_get_cached_segment_base,
7584 .get_gdt = emulator_get_gdt,
7585 .get_idt = emulator_get_idt,
7586 .set_gdt = emulator_set_gdt,
7587 .set_idt = emulator_set_idt,
7588 .get_cr = emulator_get_cr,
7589 .set_cr = emulator_set_cr,
7590 .cpl = emulator_get_cpl,
7591 .get_dr = emulator_get_dr,
7592 .set_dr = emulator_set_dr,
7593 .get_smbase = emulator_get_smbase,
7594 .set_smbase = emulator_set_smbase,
7595 .set_msr = emulator_set_msr,
7596 .get_msr = emulator_get_msr,
7597 .check_pmc = emulator_check_pmc,
7598 .read_pmc = emulator_read_pmc,
7599 .halt = emulator_halt,
7600 .wbinvd = emulator_wbinvd,
7601 .fix_hypercall = emulator_fix_hypercall,
7602 .intercept = emulator_intercept,
7603 .get_cpuid = emulator_get_cpuid,
7604 .guest_has_long_mode = emulator_guest_has_long_mode,
7605 .guest_has_movbe = emulator_guest_has_movbe,
7606 .guest_has_fxsr = emulator_guest_has_fxsr,
7607 .set_nmi_mask = emulator_set_nmi_mask,
7608 .get_hflags = emulator_get_hflags,
7609 .exiting_smm = emulator_exiting_smm,
7610 .leave_smm = emulator_leave_smm,
7611 .triple_fault = emulator_triple_fault,
7612 .set_xcr = emulator_set_xcr,
7615 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7617 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7619 * an sti; sti; sequence only disable interrupts for the first
7620 * instruction. So, if the last instruction, be it emulated or
7621 * not, left the system with the INT_STI flag enabled, it
7622 * means that the last instruction is an sti. We should not
7623 * leave the flag on in this case. The same goes for mov ss
7625 if (int_shadow & mask)
7627 if (unlikely(int_shadow || mask)) {
7628 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7630 kvm_make_request(KVM_REQ_EVENT, vcpu);
7634 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7636 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7637 if (ctxt->exception.vector == PF_VECTOR)
7638 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7640 if (ctxt->exception.error_code_valid)
7641 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7642 ctxt->exception.error_code);
7644 kvm_queue_exception(vcpu, ctxt->exception.vector);
7648 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7650 struct x86_emulate_ctxt *ctxt;
7652 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7654 pr_err("kvm: failed to allocate vcpu's emulator\n");
7659 ctxt->ops = &emulate_ops;
7660 vcpu->arch.emulate_ctxt = ctxt;
7665 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7667 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7670 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7672 ctxt->gpa_available = false;
7673 ctxt->eflags = kvm_get_rflags(vcpu);
7674 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7676 ctxt->eip = kvm_rip_read(vcpu);
7677 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7678 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7679 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7680 cs_db ? X86EMUL_MODE_PROT32 :
7681 X86EMUL_MODE_PROT16;
7682 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7683 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7684 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7686 ctxt->interruptibility = 0;
7687 ctxt->have_exception = false;
7688 ctxt->exception.vector = -1;
7689 ctxt->perm_ok = false;
7691 init_decode_cache(ctxt);
7692 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7695 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7697 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7700 init_emulate_ctxt(vcpu);
7704 ctxt->_eip = ctxt->eip + inc_eip;
7705 ret = emulate_int_real(ctxt, irq);
7707 if (ret != X86EMUL_CONTINUE) {
7708 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7710 ctxt->eip = ctxt->_eip;
7711 kvm_rip_write(vcpu, ctxt->eip);
7712 kvm_set_rflags(vcpu, ctxt->eflags);
7715 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7717 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7718 u8 ndata, u8 *insn_bytes, u8 insn_size)
7720 struct kvm_run *run = vcpu->run;
7725 * Zero the whole array used to retrieve the exit info, as casting to
7726 * u32 for select entries will leave some chunks uninitialized.
7728 memset(&info, 0, sizeof(info));
7730 static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
7731 &info[2], (u32 *)&info[3],
7734 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7735 run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7738 * There's currently space for 13 entries, but 5 are used for the exit
7739 * reason and info. Restrict to 4 to reduce the maintenance burden
7740 * when expanding kvm_run.emulation_failure in the future.
7742 if (WARN_ON_ONCE(ndata > 4))
7745 /* Always include the flags as a 'data' entry. */
7747 run->emulation_failure.flags = 0;
7750 BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
7751 sizeof(run->emulation_failure.insn_bytes) != 16));
7753 run->emulation_failure.flags |=
7754 KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7755 run->emulation_failure.insn_size = insn_size;
7756 memset(run->emulation_failure.insn_bytes, 0x90,
7757 sizeof(run->emulation_failure.insn_bytes));
7758 memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
7761 memcpy(&run->internal.data[info_start], info, sizeof(info));
7762 memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
7763 ndata * sizeof(data[0]));
7765 run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
7768 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
7770 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7772 prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
7773 ctxt->fetch.end - ctxt->fetch.data);
7776 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7779 prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
7781 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
7783 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7785 __kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
7787 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
7789 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7791 struct kvm *kvm = vcpu->kvm;
7793 ++vcpu->stat.insn_emulation_fail;
7794 trace_kvm_emulate_insn_failed(vcpu);
7796 if (emulation_type & EMULTYPE_VMWARE_GP) {
7797 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7801 if (kvm->arch.exit_on_emulation_error ||
7802 (emulation_type & EMULTYPE_SKIP)) {
7803 prepare_emulation_ctxt_failure_exit(vcpu);
7807 kvm_queue_exception(vcpu, UD_VECTOR);
7809 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7810 prepare_emulation_ctxt_failure_exit(vcpu);
7817 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7818 bool write_fault_to_shadow_pgtable,
7821 gpa_t gpa = cr2_or_gpa;
7824 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7827 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7828 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7831 if (!vcpu->arch.mmu->direct_map) {
7833 * Write permission should be allowed since only
7834 * write access need to be emulated.
7836 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7839 * If the mapping is invalid in guest, let cpu retry
7840 * it to generate fault.
7842 if (gpa == UNMAPPED_GVA)
7847 * Do not retry the unhandleable instruction if it faults on the
7848 * readonly host memory, otherwise it will goto a infinite loop:
7849 * retry instruction -> write #PF -> emulation fail -> retry
7850 * instruction -> ...
7852 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7855 * If the instruction failed on the error pfn, it can not be fixed,
7856 * report the error to userspace.
7858 if (is_error_noslot_pfn(pfn))
7861 kvm_release_pfn_clean(pfn);
7863 /* The instructions are well-emulated on direct mmu. */
7864 if (vcpu->arch.mmu->direct_map) {
7865 unsigned int indirect_shadow_pages;
7867 write_lock(&vcpu->kvm->mmu_lock);
7868 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7869 write_unlock(&vcpu->kvm->mmu_lock);
7871 if (indirect_shadow_pages)
7872 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7878 * if emulation was due to access to shadowed page table
7879 * and it failed try to unshadow page and re-enter the
7880 * guest to let CPU execute the instruction.
7882 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7885 * If the access faults on its page table, it can not
7886 * be fixed by unprotecting shadow page and it should
7887 * be reported to userspace.
7889 return !write_fault_to_shadow_pgtable;
7892 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7893 gpa_t cr2_or_gpa, int emulation_type)
7895 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7896 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7898 last_retry_eip = vcpu->arch.last_retry_eip;
7899 last_retry_addr = vcpu->arch.last_retry_addr;
7902 * If the emulation is caused by #PF and it is non-page_table
7903 * writing instruction, it means the VM-EXIT is caused by shadow
7904 * page protected, we can zap the shadow page and retry this
7905 * instruction directly.
7907 * Note: if the guest uses a non-page-table modifying instruction
7908 * on the PDE that points to the instruction, then we will unmap
7909 * the instruction and go to an infinite loop. So, we cache the
7910 * last retried eip and the last fault address, if we meet the eip
7911 * and the address again, we can break out of the potential infinite
7914 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7916 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7919 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7920 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7923 if (x86_page_table_writing_insn(ctxt))
7926 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7929 vcpu->arch.last_retry_eip = ctxt->eip;
7930 vcpu->arch.last_retry_addr = cr2_or_gpa;
7932 if (!vcpu->arch.mmu->direct_map)
7933 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7935 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7940 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7941 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7943 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7945 trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7948 vcpu->arch.hflags |= HF_SMM_MASK;
7950 vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7952 /* Process a latched INIT or SMI, if any. */
7953 kvm_make_request(KVM_REQ_EVENT, vcpu);
7956 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
7957 * on SMM exit we still need to reload them from
7960 vcpu->arch.pdptrs_from_userspace = false;
7963 kvm_mmu_reset_context(vcpu);
7966 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7975 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7976 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7981 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7983 struct kvm_run *kvm_run = vcpu->run;
7985 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7986 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7987 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7988 kvm_run->debug.arch.exception = DB_VECTOR;
7989 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7992 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7996 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7998 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8001 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
8006 * rflags is the old, "raw" value of the flags. The new value has
8007 * not been saved yet.
8009 * This is correct even for TF set by the guest, because "the
8010 * processor will not generate this exception after the instruction
8011 * that sets the TF flag".
8013 if (unlikely(rflags & X86_EFLAGS_TF))
8014 r = kvm_vcpu_do_singlestep(vcpu);
8017 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
8019 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
8021 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
8022 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
8023 struct kvm_run *kvm_run = vcpu->run;
8024 unsigned long eip = kvm_get_linear_rip(vcpu);
8025 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8026 vcpu->arch.guest_debug_dr7,
8030 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
8031 kvm_run->debug.arch.pc = eip;
8032 kvm_run->debug.arch.exception = DB_VECTOR;
8033 kvm_run->exit_reason = KVM_EXIT_DEBUG;
8039 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
8040 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
8041 unsigned long eip = kvm_get_linear_rip(vcpu);
8042 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8047 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
8056 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
8058 switch (ctxt->opcode_len) {
8065 case 0xe6: /* OUT */
8069 case 0x6c: /* INS */
8071 case 0x6e: /* OUTS */
8078 case 0x33: /* RDPMC */
8088 * Decode to be emulated instruction. Return EMULATION_OK if success.
8090 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8091 void *insn, int insn_len)
8093 int r = EMULATION_OK;
8094 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8096 init_emulate_ctxt(vcpu);
8099 * We will reenter on the same instruction since we do not set
8100 * complete_userspace_io. This does not handle watchpoints yet,
8101 * those would be handled in the emulate_ops.
8103 if (!(emulation_type & EMULTYPE_SKIP) &&
8104 kvm_vcpu_check_breakpoint(vcpu, &r))
8107 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8109 trace_kvm_emulate_insn_start(vcpu);
8110 ++vcpu->stat.insn_emulation;
8114 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8116 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8117 int emulation_type, void *insn, int insn_len)
8120 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8121 bool writeback = true;
8122 bool write_fault_to_spt;
8124 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
8127 vcpu->arch.l1tf_flush_l1d = true;
8130 * Clear write_fault_to_shadow_pgtable here to ensure it is
8133 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
8134 vcpu->arch.write_fault_to_shadow_pgtable = false;
8136 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8137 kvm_clear_exception_queue(vcpu);
8139 r = x86_decode_emulated_instruction(vcpu, emulation_type,
8141 if (r != EMULATION_OK) {
8142 if ((emulation_type & EMULTYPE_TRAP_UD) ||
8143 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8144 kvm_queue_exception(vcpu, UD_VECTOR);
8147 if (reexecute_instruction(vcpu, cr2_or_gpa,
8151 if (ctxt->have_exception) {
8153 * #UD should result in just EMULATION_FAILED, and trap-like
8154 * exception should not be encountered during decode.
8156 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8157 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8158 inject_emulated_exception(vcpu);
8161 return handle_emulation_failure(vcpu, emulation_type);
8165 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8166 !is_vmware_backdoor_opcode(ctxt)) {
8167 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8172 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
8173 * for kvm_skip_emulated_instruction(). The caller is responsible for
8174 * updating interruptibility state and injecting single-step #DBs.
8176 if (emulation_type & EMULTYPE_SKIP) {
8177 kvm_rip_write(vcpu, ctxt->_eip);
8178 if (ctxt->eflags & X86_EFLAGS_RF)
8179 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8183 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8186 /* this is needed for vmware backdoor interface to work since it
8187 changes registers values during IO operation */
8188 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8189 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8190 emulator_invalidate_register_cache(ctxt);
8194 if (emulation_type & EMULTYPE_PF) {
8195 /* Save the faulting GPA (cr2) in the address field */
8196 ctxt->exception.address = cr2_or_gpa;
8198 /* With shadow page tables, cr2 contains a GVA or nGPA. */
8199 if (vcpu->arch.mmu->direct_map) {
8200 ctxt->gpa_available = true;
8201 ctxt->gpa_val = cr2_or_gpa;
8204 /* Sanitize the address out of an abundance of paranoia. */
8205 ctxt->exception.address = 0;
8208 r = x86_emulate_insn(ctxt);
8210 if (r == EMULATION_INTERCEPTED)
8213 if (r == EMULATION_FAILED) {
8214 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8218 return handle_emulation_failure(vcpu, emulation_type);
8221 if (ctxt->have_exception) {
8223 if (inject_emulated_exception(vcpu))
8225 } else if (vcpu->arch.pio.count) {
8226 if (!vcpu->arch.pio.in) {
8227 /* FIXME: return into emulator if single-stepping. */
8228 vcpu->arch.pio.count = 0;
8231 vcpu->arch.complete_userspace_io = complete_emulated_pio;
8234 } else if (vcpu->mmio_needed) {
8235 ++vcpu->stat.mmio_exits;
8237 if (!vcpu->mmio_is_write)
8240 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8241 } else if (r == EMULATION_RESTART)
8247 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8248 toggle_interruptibility(vcpu, ctxt->interruptibility);
8249 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8250 if (!ctxt->have_exception ||
8251 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8252 kvm_rip_write(vcpu, ctxt->eip);
8253 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8254 r = kvm_vcpu_do_singlestep(vcpu);
8255 if (kvm_x86_ops.update_emulated_instruction)
8256 static_call(kvm_x86_update_emulated_instruction)(vcpu);
8257 __kvm_set_rflags(vcpu, ctxt->eflags);
8261 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8262 * do nothing, and it will be requested again as soon as
8263 * the shadow expires. But we still need to check here,
8264 * because POPF has no interrupt shadow.
8266 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8267 kvm_make_request(KVM_REQ_EVENT, vcpu);
8269 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8274 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8276 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8278 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8280 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8281 void *insn, int insn_len)
8283 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8285 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8287 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8289 vcpu->arch.pio.count = 0;
8293 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8295 vcpu->arch.pio.count = 0;
8297 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8300 return kvm_skip_emulated_instruction(vcpu);
8303 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8304 unsigned short port)
8306 unsigned long val = kvm_rax_read(vcpu);
8307 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8313 * Workaround userspace that relies on old KVM behavior of %rip being
8314 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8317 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8318 vcpu->arch.complete_userspace_io =
8319 complete_fast_pio_out_port_0x7e;
8320 kvm_skip_emulated_instruction(vcpu);
8322 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8323 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8328 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8332 /* We should only ever be called with arch.pio.count equal to 1 */
8333 BUG_ON(vcpu->arch.pio.count != 1);
8335 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8336 vcpu->arch.pio.count = 0;
8340 /* For size less than 4 we merge, else we zero extend */
8341 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8344 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8345 * the copy and tracing
8347 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8348 kvm_rax_write(vcpu, val);
8350 return kvm_skip_emulated_instruction(vcpu);
8353 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8354 unsigned short port)
8359 /* For size less than 4 we merge, else we zero extend */
8360 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8362 ret = emulator_pio_in(vcpu, size, port, &val, 1);
8364 kvm_rax_write(vcpu, val);
8368 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8369 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8374 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8379 ret = kvm_fast_pio_in(vcpu, size, port);
8381 ret = kvm_fast_pio_out(vcpu, size, port);
8382 return ret && kvm_skip_emulated_instruction(vcpu);
8384 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8386 static int kvmclock_cpu_down_prep(unsigned int cpu)
8388 __this_cpu_write(cpu_tsc_khz, 0);
8392 static void tsc_khz_changed(void *data)
8394 struct cpufreq_freqs *freq = data;
8395 unsigned long khz = 0;
8399 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8400 khz = cpufreq_quick_get(raw_smp_processor_id());
8403 __this_cpu_write(cpu_tsc_khz, khz);
8406 #ifdef CONFIG_X86_64
8407 static void kvm_hyperv_tsc_notifier(void)
8412 mutex_lock(&kvm_lock);
8413 list_for_each_entry(kvm, &vm_list, vm_list)
8414 kvm_make_mclock_inprogress_request(kvm);
8416 /* no guest entries from this point */
8417 hyperv_stop_tsc_emulation();
8419 /* TSC frequency always matches when on Hyper-V */
8420 for_each_present_cpu(cpu)
8421 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8422 kvm_max_guest_tsc_khz = tsc_khz;
8424 list_for_each_entry(kvm, &vm_list, vm_list) {
8425 __kvm_start_pvclock_update(kvm);
8426 pvclock_update_vm_gtod_copy(kvm);
8427 kvm_end_pvclock_update(kvm);
8430 mutex_unlock(&kvm_lock);
8434 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8437 struct kvm_vcpu *vcpu;
8438 int i, send_ipi = 0;
8441 * We allow guests to temporarily run on slowing clocks,
8442 * provided we notify them after, or to run on accelerating
8443 * clocks, provided we notify them before. Thus time never
8446 * However, we have a problem. We can't atomically update
8447 * the frequency of a given CPU from this function; it is
8448 * merely a notifier, which can be called from any CPU.
8449 * Changing the TSC frequency at arbitrary points in time
8450 * requires a recomputation of local variables related to
8451 * the TSC for each VCPU. We must flag these local variables
8452 * to be updated and be sure the update takes place with the
8453 * new frequency before any guests proceed.
8455 * Unfortunately, the combination of hotplug CPU and frequency
8456 * change creates an intractable locking scenario; the order
8457 * of when these callouts happen is undefined with respect to
8458 * CPU hotplug, and they can race with each other. As such,
8459 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8460 * undefined; you can actually have a CPU frequency change take
8461 * place in between the computation of X and the setting of the
8462 * variable. To protect against this problem, all updates of
8463 * the per_cpu tsc_khz variable are done in an interrupt
8464 * protected IPI, and all callers wishing to update the value
8465 * must wait for a synchronous IPI to complete (which is trivial
8466 * if the caller is on the CPU already). This establishes the
8467 * necessary total order on variable updates.
8469 * Note that because a guest time update may take place
8470 * anytime after the setting of the VCPU's request bit, the
8471 * correct TSC value must be set before the request. However,
8472 * to ensure the update actually makes it to any guest which
8473 * starts running in hardware virtualization between the set
8474 * and the acquisition of the spinlock, we must also ping the
8475 * CPU after setting the request bit.
8479 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8481 mutex_lock(&kvm_lock);
8482 list_for_each_entry(kvm, &vm_list, vm_list) {
8483 kvm_for_each_vcpu(i, vcpu, kvm) {
8484 if (vcpu->cpu != cpu)
8486 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8487 if (vcpu->cpu != raw_smp_processor_id())
8491 mutex_unlock(&kvm_lock);
8493 if (freq->old < freq->new && send_ipi) {
8495 * We upscale the frequency. Must make the guest
8496 * doesn't see old kvmclock values while running with
8497 * the new frequency, otherwise we risk the guest sees
8498 * time go backwards.
8500 * In case we update the frequency for another cpu
8501 * (which might be in guest context) send an interrupt
8502 * to kick the cpu out of guest context. Next time
8503 * guest context is entered kvmclock will be updated,
8504 * so the guest will not see stale values.
8506 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8510 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8513 struct cpufreq_freqs *freq = data;
8516 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8518 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8521 for_each_cpu(cpu, freq->policy->cpus)
8522 __kvmclock_cpufreq_notifier(freq, cpu);
8527 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8528 .notifier_call = kvmclock_cpufreq_notifier
8531 static int kvmclock_cpu_online(unsigned int cpu)
8533 tsc_khz_changed(NULL);
8537 static void kvm_timer_init(void)
8539 max_tsc_khz = tsc_khz;
8541 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8542 #ifdef CONFIG_CPU_FREQ
8543 struct cpufreq_policy *policy;
8547 policy = cpufreq_cpu_get(cpu);
8549 if (policy->cpuinfo.max_freq)
8550 max_tsc_khz = policy->cpuinfo.max_freq;
8551 cpufreq_cpu_put(policy);
8555 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8556 CPUFREQ_TRANSITION_NOTIFIER);
8559 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8560 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8563 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8564 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8566 int kvm_is_in_guest(void)
8568 return __this_cpu_read(current_vcpu) != NULL;
8571 static int kvm_is_user_mode(void)
8575 if (__this_cpu_read(current_vcpu))
8576 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8578 return user_mode != 0;
8581 static unsigned long kvm_get_guest_ip(void)
8583 unsigned long ip = 0;
8585 if (__this_cpu_read(current_vcpu))
8586 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8591 static void kvm_handle_intel_pt_intr(void)
8593 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8595 kvm_make_request(KVM_REQ_PMI, vcpu);
8596 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8597 (unsigned long *)&vcpu->arch.pmu.global_status);
8600 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8601 .is_in_guest = kvm_is_in_guest,
8602 .is_user_mode = kvm_is_user_mode,
8603 .get_guest_ip = kvm_get_guest_ip,
8604 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8607 #ifdef CONFIG_X86_64
8608 static void pvclock_gtod_update_fn(struct work_struct *work)
8612 struct kvm_vcpu *vcpu;
8615 mutex_lock(&kvm_lock);
8616 list_for_each_entry(kvm, &vm_list, vm_list)
8617 kvm_for_each_vcpu(i, vcpu, kvm)
8618 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8619 atomic_set(&kvm_guest_has_master_clock, 0);
8620 mutex_unlock(&kvm_lock);
8623 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8626 * Indirection to move queue_work() out of the tk_core.seq write held
8627 * region to prevent possible deadlocks against time accessors which
8628 * are invoked with work related locks held.
8630 static void pvclock_irq_work_fn(struct irq_work *w)
8632 queue_work(system_long_wq, &pvclock_gtod_work);
8635 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8638 * Notification about pvclock gtod data update.
8640 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8643 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8644 struct timekeeper *tk = priv;
8646 update_pvclock_gtod(tk);
8649 * Disable master clock if host does not trust, or does not use,
8650 * TSC based clocksource. Delegate queue_work() to irq_work as
8651 * this is invoked with tk_core.seq write held.
8653 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8654 atomic_read(&kvm_guest_has_master_clock) != 0)
8655 irq_work_queue(&pvclock_irq_work);
8659 static struct notifier_block pvclock_gtod_notifier = {
8660 .notifier_call = pvclock_gtod_notify,
8664 int kvm_arch_init(void *opaque)
8666 struct kvm_x86_init_ops *ops = opaque;
8669 if (kvm_x86_ops.hardware_enable) {
8670 pr_err("kvm: already loaded vendor module '%s'\n", kvm_x86_ops.name);
8675 if (!ops->cpu_has_kvm_support()) {
8676 pr_err_ratelimited("kvm: no hardware support for '%s'\n",
8677 ops->runtime_ops->name);
8681 if (ops->disabled_by_bios()) {
8682 pr_err_ratelimited("kvm: support for '%s' disabled by bios\n",
8683 ops->runtime_ops->name);
8689 * KVM explicitly assumes that the guest has an FPU and
8690 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8691 * vCPU's FPU state as a fxregs_state struct.
8693 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8694 printk(KERN_ERR "kvm: inadequate fpu\n");
8700 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8701 __alignof__(struct fpu), SLAB_ACCOUNT,
8703 if (!x86_fpu_cache) {
8704 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8708 x86_emulator_cache = kvm_alloc_emulator_cache();
8709 if (!x86_emulator_cache) {
8710 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8711 goto out_free_x86_fpu_cache;
8714 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8715 if (!user_return_msrs) {
8716 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8717 goto out_free_x86_emulator_cache;
8719 kvm_nr_uret_msrs = 0;
8721 r = kvm_mmu_module_init();
8723 goto out_free_percpu;
8727 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8729 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8730 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8731 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8734 if (pi_inject_timer == -1)
8735 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8736 #ifdef CONFIG_X86_64
8737 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8739 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8740 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8746 free_percpu(user_return_msrs);
8747 out_free_x86_emulator_cache:
8748 kmem_cache_destroy(x86_emulator_cache);
8749 out_free_x86_fpu_cache:
8750 kmem_cache_destroy(x86_fpu_cache);
8755 void kvm_arch_exit(void)
8757 #ifdef CONFIG_X86_64
8758 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8759 clear_hv_tscchange_cb();
8762 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8764 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8765 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8766 CPUFREQ_TRANSITION_NOTIFIER);
8767 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8768 #ifdef CONFIG_X86_64
8769 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8770 irq_work_sync(&pvclock_irq_work);
8771 cancel_work_sync(&pvclock_gtod_work);
8773 kvm_x86_ops.hardware_enable = NULL;
8774 kvm_mmu_module_exit();
8775 free_percpu(user_return_msrs);
8776 kmem_cache_destroy(x86_emulator_cache);
8777 kmem_cache_destroy(x86_fpu_cache);
8778 #ifdef CONFIG_KVM_XEN
8779 static_key_deferred_flush(&kvm_xen_enabled);
8780 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8784 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8786 ++vcpu->stat.halt_exits;
8787 if (lapic_in_kernel(vcpu)) {
8788 vcpu->arch.mp_state = state;
8791 vcpu->run->exit_reason = reason;
8796 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8798 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8800 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8802 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8804 int ret = kvm_skip_emulated_instruction(vcpu);
8806 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8807 * KVM_EXIT_DEBUG here.
8809 return kvm_vcpu_halt(vcpu) && ret;
8811 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8813 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8815 int ret = kvm_skip_emulated_instruction(vcpu);
8817 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8819 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8821 #ifdef CONFIG_X86_64
8822 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8823 unsigned long clock_type)
8825 struct kvm_clock_pairing clock_pairing;
8826 struct timespec64 ts;
8830 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8831 return -KVM_EOPNOTSUPP;
8833 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8834 return -KVM_EOPNOTSUPP;
8836 clock_pairing.sec = ts.tv_sec;
8837 clock_pairing.nsec = ts.tv_nsec;
8838 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8839 clock_pairing.flags = 0;
8840 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8843 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8844 sizeof(struct kvm_clock_pairing)))
8852 * kvm_pv_kick_cpu_op: Kick a vcpu.
8854 * @apicid - apicid of vcpu to be kicked.
8856 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8858 struct kvm_lapic_irq lapic_irq;
8860 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8861 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8862 lapic_irq.level = 0;
8863 lapic_irq.dest_id = apicid;
8864 lapic_irq.msi_redir_hint = false;
8866 lapic_irq.delivery_mode = APIC_DM_REMRD;
8867 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8870 bool kvm_apicv_activated(struct kvm *kvm)
8872 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8874 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8876 static void kvm_apicv_init(struct kvm *kvm)
8878 init_rwsem(&kvm->arch.apicv_update_lock);
8881 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8882 &kvm->arch.apicv_inhibit_reasons);
8884 set_bit(APICV_INHIBIT_REASON_DISABLE,
8885 &kvm->arch.apicv_inhibit_reasons);
8888 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8890 struct kvm_vcpu *target = NULL;
8891 struct kvm_apic_map *map;
8893 vcpu->stat.directed_yield_attempted++;
8895 if (single_task_running())
8899 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8901 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8902 target = map->phys_map[dest_id]->vcpu;
8906 if (!target || !READ_ONCE(target->ready))
8909 /* Ignore requests to yield to self */
8913 if (kvm_vcpu_yield_to(target) <= 0)
8916 vcpu->stat.directed_yield_successful++;
8922 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8924 u64 ret = vcpu->run->hypercall.ret;
8926 if (!is_64_bit_mode(vcpu))
8928 kvm_rax_write(vcpu, ret);
8929 ++vcpu->stat.hypercalls;
8930 return kvm_skip_emulated_instruction(vcpu);
8933 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8935 unsigned long nr, a0, a1, a2, a3, ret;
8938 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8939 return kvm_xen_hypercall(vcpu);
8941 if (kvm_hv_hypercall_enabled(vcpu))
8942 return kvm_hv_hypercall(vcpu);
8944 nr = kvm_rax_read(vcpu);
8945 a0 = kvm_rbx_read(vcpu);
8946 a1 = kvm_rcx_read(vcpu);
8947 a2 = kvm_rdx_read(vcpu);
8948 a3 = kvm_rsi_read(vcpu);
8950 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8952 op_64_bit = is_64_bit_mode(vcpu);
8961 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8969 case KVM_HC_VAPIC_POLL_IRQ:
8972 case KVM_HC_KICK_CPU:
8973 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8976 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8977 kvm_sched_yield(vcpu, a1);
8980 #ifdef CONFIG_X86_64
8981 case KVM_HC_CLOCK_PAIRING:
8982 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8985 case KVM_HC_SEND_IPI:
8986 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8989 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8991 case KVM_HC_SCHED_YIELD:
8992 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8995 kvm_sched_yield(vcpu, a0);
8998 case KVM_HC_MAP_GPA_RANGE: {
8999 u64 gpa = a0, npages = a1, attrs = a2;
9002 if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
9005 if (!PAGE_ALIGNED(gpa) || !npages ||
9006 gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
9011 vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
9012 vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
9013 vcpu->run->hypercall.args[0] = gpa;
9014 vcpu->run->hypercall.args[1] = npages;
9015 vcpu->run->hypercall.args[2] = attrs;
9016 vcpu->run->hypercall.longmode = op_64_bit;
9017 vcpu->arch.complete_userspace_io = complete_hypercall_exit;
9027 kvm_rax_write(vcpu, ret);
9029 ++vcpu->stat.hypercalls;
9030 return kvm_skip_emulated_instruction(vcpu);
9032 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
9034 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
9036 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
9037 char instruction[3];
9038 unsigned long rip = kvm_rip_read(vcpu);
9040 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
9042 return emulator_write_emulated(ctxt, rip, instruction, 3,
9046 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
9048 return vcpu->run->request_interrupt_window &&
9049 likely(!pic_in_kernel(vcpu->kvm));
9052 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
9054 struct kvm_run *kvm_run = vcpu->run;
9057 * if_flag is obsolete and useless, so do not bother
9058 * setting it for SEV-ES guests. Userspace can just
9059 * use kvm_run->ready_for_interrupt_injection.
9061 kvm_run->if_flag = !vcpu->arch.guest_state_protected
9062 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
9064 kvm_run->cr8 = kvm_get_cr8(vcpu);
9065 kvm_run->apic_base = kvm_get_apic_base(vcpu);
9066 kvm_run->ready_for_interrupt_injection =
9067 pic_in_kernel(vcpu->kvm) ||
9068 kvm_vcpu_ready_for_interrupt_injection(vcpu);
9071 kvm_run->flags |= KVM_RUN_X86_SMM;
9074 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
9078 if (!kvm_x86_ops.update_cr8_intercept)
9081 if (!lapic_in_kernel(vcpu))
9084 if (vcpu->arch.apicv_active)
9087 if (!vcpu->arch.apic->vapic_addr)
9088 max_irr = kvm_lapic_find_highest_irr(vcpu);
9095 tpr = kvm_lapic_get_cr8(vcpu);
9097 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9101 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9103 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9104 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9108 return kvm_x86_ops.nested_ops->check_events(vcpu);
9111 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9113 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
9114 vcpu->arch.exception.error_code = false;
9115 static_call(kvm_x86_queue_exception)(vcpu);
9118 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
9121 bool can_inject = true;
9123 /* try to reinject previous events if any */
9125 if (vcpu->arch.exception.injected) {
9126 kvm_inject_exception(vcpu);
9130 * Do not inject an NMI or interrupt if there is a pending
9131 * exception. Exceptions and interrupts are recognized at
9132 * instruction boundaries, i.e. the start of an instruction.
9133 * Trap-like exceptions, e.g. #DB, have higher priority than
9134 * NMIs and interrupts, i.e. traps are recognized before an
9135 * NMI/interrupt that's pending on the same instruction.
9136 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
9137 * priority, but are only generated (pended) during instruction
9138 * execution, i.e. a pending fault-like exception means the
9139 * fault occurred on the *previous* instruction and must be
9140 * serviced prior to recognizing any new events in order to
9141 * fully complete the previous instruction.
9143 else if (!vcpu->arch.exception.pending) {
9144 if (vcpu->arch.nmi_injected) {
9145 static_call(kvm_x86_set_nmi)(vcpu);
9147 } else if (vcpu->arch.interrupt.injected) {
9148 static_call(kvm_x86_set_irq)(vcpu);
9153 WARN_ON_ONCE(vcpu->arch.exception.injected &&
9154 vcpu->arch.exception.pending);
9157 * Call check_nested_events() even if we reinjected a previous event
9158 * in order for caller to determine if it should require immediate-exit
9159 * from L2 to L1 due to pending L1 events which require exit
9162 if (is_guest_mode(vcpu)) {
9163 r = kvm_check_nested_events(vcpu);
9168 /* try to inject new event if pending */
9169 if (vcpu->arch.exception.pending) {
9170 trace_kvm_inj_exception(vcpu->arch.exception.nr,
9171 vcpu->arch.exception.has_error_code,
9172 vcpu->arch.exception.error_code);
9174 vcpu->arch.exception.pending = false;
9175 vcpu->arch.exception.injected = true;
9177 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
9178 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
9181 if (vcpu->arch.exception.nr == DB_VECTOR) {
9182 kvm_deliver_exception_payload(vcpu);
9183 if (vcpu->arch.dr7 & DR7_GD) {
9184 vcpu->arch.dr7 &= ~DR7_GD;
9185 kvm_update_dr7(vcpu);
9189 kvm_inject_exception(vcpu);
9193 /* Don't inject interrupts if the user asked to avoid doing so */
9194 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
9198 * Finally, inject interrupt events. If an event cannot be injected
9199 * due to architectural conditions (e.g. IF=0) a window-open exit
9200 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
9201 * and can architecturally be injected, but we cannot do it right now:
9202 * an interrupt could have arrived just now and we have to inject it
9203 * as a vmexit, or there could already an event in the queue, which is
9204 * indicated by can_inject. In that case we request an immediate exit
9205 * in order to make progress and get back here for another iteration.
9206 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
9208 if (vcpu->arch.smi_pending) {
9209 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
9213 vcpu->arch.smi_pending = false;
9214 ++vcpu->arch.smi_count;
9218 static_call(kvm_x86_enable_smi_window)(vcpu);
9221 if (vcpu->arch.nmi_pending) {
9222 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
9226 --vcpu->arch.nmi_pending;
9227 vcpu->arch.nmi_injected = true;
9228 static_call(kvm_x86_set_nmi)(vcpu);
9230 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
9232 if (vcpu->arch.nmi_pending)
9233 static_call(kvm_x86_enable_nmi_window)(vcpu);
9236 if (kvm_cpu_has_injectable_intr(vcpu)) {
9237 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
9241 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
9242 static_call(kvm_x86_set_irq)(vcpu);
9243 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
9245 if (kvm_cpu_has_injectable_intr(vcpu))
9246 static_call(kvm_x86_enable_irq_window)(vcpu);
9249 if (is_guest_mode(vcpu) &&
9250 kvm_x86_ops.nested_ops->hv_timer_pending &&
9251 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
9252 *req_immediate_exit = true;
9254 WARN_ON(vcpu->arch.exception.pending);
9259 *req_immediate_exit = true;
9265 static void process_nmi(struct kvm_vcpu *vcpu)
9270 * x86 is limited to one NMI running, and one NMI pending after it.
9271 * If an NMI is already in progress, limit further NMIs to just one.
9272 * Otherwise, allow two (and we'll inject the first one immediately).
9274 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
9277 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
9278 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
9279 kvm_make_request(KVM_REQ_EVENT, vcpu);
9282 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
9285 flags |= seg->g << 23;
9286 flags |= seg->db << 22;
9287 flags |= seg->l << 21;
9288 flags |= seg->avl << 20;
9289 flags |= seg->present << 15;
9290 flags |= seg->dpl << 13;
9291 flags |= seg->s << 12;
9292 flags |= seg->type << 8;
9296 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
9298 struct kvm_segment seg;
9301 kvm_get_segment(vcpu, &seg, n);
9302 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9305 offset = 0x7f84 + n * 12;
9307 offset = 0x7f2c + (n - 3) * 12;
9309 put_smstate(u32, buf, offset + 8, seg.base);
9310 put_smstate(u32, buf, offset + 4, seg.limit);
9311 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9314 #ifdef CONFIG_X86_64
9315 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9317 struct kvm_segment seg;
9321 kvm_get_segment(vcpu, &seg, n);
9322 offset = 0x7e00 + n * 16;
9324 flags = enter_smm_get_segment_flags(&seg) >> 8;
9325 put_smstate(u16, buf, offset, seg.selector);
9326 put_smstate(u16, buf, offset + 2, flags);
9327 put_smstate(u32, buf, offset + 4, seg.limit);
9328 put_smstate(u64, buf, offset + 8, seg.base);
9332 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9335 struct kvm_segment seg;
9339 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9340 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9341 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9342 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9344 for (i = 0; i < 8; i++)
9345 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9347 kvm_get_dr(vcpu, 6, &val);
9348 put_smstate(u32, buf, 0x7fcc, (u32)val);
9349 kvm_get_dr(vcpu, 7, &val);
9350 put_smstate(u32, buf, 0x7fc8, (u32)val);
9352 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9353 put_smstate(u32, buf, 0x7fc4, seg.selector);
9354 put_smstate(u32, buf, 0x7f64, seg.base);
9355 put_smstate(u32, buf, 0x7f60, seg.limit);
9356 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9358 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9359 put_smstate(u32, buf, 0x7fc0, seg.selector);
9360 put_smstate(u32, buf, 0x7f80, seg.base);
9361 put_smstate(u32, buf, 0x7f7c, seg.limit);
9362 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9364 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9365 put_smstate(u32, buf, 0x7f74, dt.address);
9366 put_smstate(u32, buf, 0x7f70, dt.size);
9368 static_call(kvm_x86_get_idt)(vcpu, &dt);
9369 put_smstate(u32, buf, 0x7f58, dt.address);
9370 put_smstate(u32, buf, 0x7f54, dt.size);
9372 for (i = 0; i < 6; i++)
9373 enter_smm_save_seg_32(vcpu, buf, i);
9375 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9378 put_smstate(u32, buf, 0x7efc, 0x00020000);
9379 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9382 #ifdef CONFIG_X86_64
9383 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9386 struct kvm_segment seg;
9390 for (i = 0; i < 16; i++)
9391 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9393 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9394 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9396 kvm_get_dr(vcpu, 6, &val);
9397 put_smstate(u64, buf, 0x7f68, val);
9398 kvm_get_dr(vcpu, 7, &val);
9399 put_smstate(u64, buf, 0x7f60, val);
9401 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9402 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9403 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9405 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9408 put_smstate(u32, buf, 0x7efc, 0x00020064);
9410 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9412 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9413 put_smstate(u16, buf, 0x7e90, seg.selector);
9414 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9415 put_smstate(u32, buf, 0x7e94, seg.limit);
9416 put_smstate(u64, buf, 0x7e98, seg.base);
9418 static_call(kvm_x86_get_idt)(vcpu, &dt);
9419 put_smstate(u32, buf, 0x7e84, dt.size);
9420 put_smstate(u64, buf, 0x7e88, dt.address);
9422 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9423 put_smstate(u16, buf, 0x7e70, seg.selector);
9424 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9425 put_smstate(u32, buf, 0x7e74, seg.limit);
9426 put_smstate(u64, buf, 0x7e78, seg.base);
9428 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9429 put_smstate(u32, buf, 0x7e64, dt.size);
9430 put_smstate(u64, buf, 0x7e68, dt.address);
9432 for (i = 0; i < 6; i++)
9433 enter_smm_save_seg_64(vcpu, buf, i);
9437 static void enter_smm(struct kvm_vcpu *vcpu)
9439 struct kvm_segment cs, ds;
9444 memset(buf, 0, 512);
9445 #ifdef CONFIG_X86_64
9446 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9447 enter_smm_save_state_64(vcpu, buf);
9450 enter_smm_save_state_32(vcpu, buf);
9453 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9454 * state (e.g. leave guest mode) after we've saved the state into the
9455 * SMM state-save area.
9457 static_call(kvm_x86_enter_smm)(vcpu, buf);
9459 kvm_smm_changed(vcpu, true);
9460 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9462 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9463 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9465 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9467 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9468 kvm_rip_write(vcpu, 0x8000);
9470 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9471 static_call(kvm_x86_set_cr0)(vcpu, cr0);
9472 vcpu->arch.cr0 = cr0;
9474 static_call(kvm_x86_set_cr4)(vcpu, 0);
9476 /* Undocumented: IDT limit is set to zero on entry to SMM. */
9477 dt.address = dt.size = 0;
9478 static_call(kvm_x86_set_idt)(vcpu, &dt);
9480 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9482 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9483 cs.base = vcpu->arch.smbase;
9488 cs.limit = ds.limit = 0xffffffff;
9489 cs.type = ds.type = 0x3;
9490 cs.dpl = ds.dpl = 0;
9495 cs.avl = ds.avl = 0;
9496 cs.present = ds.present = 1;
9497 cs.unusable = ds.unusable = 0;
9498 cs.padding = ds.padding = 0;
9500 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9501 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9502 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9503 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9504 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9505 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9507 #ifdef CONFIG_X86_64
9508 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9509 static_call(kvm_x86_set_efer)(vcpu, 0);
9512 kvm_update_cpuid_runtime(vcpu);
9513 kvm_mmu_reset_context(vcpu);
9516 static void process_smi(struct kvm_vcpu *vcpu)
9518 vcpu->arch.smi_pending = true;
9519 kvm_make_request(KVM_REQ_EVENT, vcpu);
9522 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9523 unsigned long *vcpu_bitmap)
9525 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
9528 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9530 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9533 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9537 if (!lapic_in_kernel(vcpu))
9540 down_read(&vcpu->kvm->arch.apicv_update_lock);
9542 activate = kvm_apicv_activated(vcpu->kvm);
9543 if (vcpu->arch.apicv_active == activate)
9546 vcpu->arch.apicv_active = activate;
9547 kvm_apic_update_apicv(vcpu);
9548 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9551 * When APICv gets disabled, we may still have injected interrupts
9552 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9553 * still active when the interrupt got accepted. Make sure
9554 * inject_pending_event() is called to check for that.
9556 if (!vcpu->arch.apicv_active)
9557 kvm_make_request(KVM_REQ_EVENT, vcpu);
9560 up_read(&vcpu->kvm->arch.apicv_update_lock);
9562 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9564 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9566 unsigned long old, new;
9568 lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
9570 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9571 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9574 old = new = kvm->arch.apicv_inhibit_reasons;
9577 __clear_bit(bit, &new);
9579 __set_bit(bit, &new);
9581 if (!!old != !!new) {
9582 trace_kvm_apicv_update_request(activate, bit);
9584 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
9585 * false positives in the sanity check WARN in svm_vcpu_run().
9586 * This task will wait for all vCPUs to ack the kick IRQ before
9587 * updating apicv_inhibit_reasons, and all other vCPUs will
9588 * block on acquiring apicv_update_lock so that vCPUs can't
9589 * redo svm_vcpu_run() without seeing the new inhibit state.
9591 * Note, holding apicv_update_lock and taking it in the read
9592 * side (handling the request) also prevents other vCPUs from
9593 * servicing the request with a stale apicv_inhibit_reasons.
9595 kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9596 kvm->arch.apicv_inhibit_reasons = new;
9598 unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9599 kvm_zap_gfn_range(kvm, gfn, gfn+1);
9602 kvm->arch.apicv_inhibit_reasons = new;
9604 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9606 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9608 down_write(&kvm->arch.apicv_update_lock);
9609 __kvm_request_apicv_update(kvm, activate, bit);
9610 up_write(&kvm->arch.apicv_update_lock);
9612 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9614 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9616 if (!kvm_apic_present(vcpu))
9619 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9621 if (irqchip_split(vcpu->kvm))
9622 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9624 if (vcpu->arch.apicv_active)
9625 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9626 if (ioapic_in_kernel(vcpu->kvm))
9627 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9630 if (is_guest_mode(vcpu))
9631 vcpu->arch.load_eoi_exitmap_pending = true;
9633 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9636 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9638 u64 eoi_exit_bitmap[4];
9640 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9643 if (to_hv_vcpu(vcpu))
9644 bitmap_or((ulong *)eoi_exit_bitmap,
9645 vcpu->arch.ioapic_handled_vectors,
9646 to_hv_synic(vcpu)->vec_bitmap, 256);
9648 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9651 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9652 unsigned long start, unsigned long end)
9654 unsigned long apic_address;
9657 * The physical address of apic access page is stored in the VMCS.
9658 * Update it when it becomes invalid.
9660 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9661 if (start <= apic_address && apic_address < end)
9662 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9665 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9667 if (!lapic_in_kernel(vcpu))
9670 if (!kvm_x86_ops.set_apic_access_page_addr)
9673 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9676 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9678 smp_send_reschedule(vcpu->cpu);
9680 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9683 * Returns 1 to let vcpu_run() continue the guest execution loop without
9684 * exiting to the userspace. Otherwise, the value will be returned to the
9687 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9691 dm_request_for_irq_injection(vcpu) &&
9692 kvm_cpu_accept_dm_intr(vcpu);
9693 fastpath_t exit_fastpath;
9695 bool req_immediate_exit = false;
9697 /* Forbid vmenter if vcpu dirty ring is soft-full */
9698 if (unlikely(vcpu->kvm->dirty_ring_size &&
9699 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9700 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9701 trace_kvm_dirty_ring_exit(vcpu);
9706 if (kvm_request_pending(vcpu)) {
9707 if (kvm_check_request(KVM_REQ_VM_BUGGED, vcpu)) {
9711 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9712 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9717 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9718 kvm_mmu_unload(vcpu);
9719 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9720 __kvm_migrate_timers(vcpu);
9721 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9722 kvm_update_masterclock(vcpu->kvm);
9723 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9724 kvm_gen_kvmclock_update(vcpu);
9725 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9726 r = kvm_guest_time_update(vcpu);
9730 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9731 kvm_mmu_sync_roots(vcpu);
9732 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9733 kvm_mmu_load_pgd(vcpu);
9734 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9735 kvm_vcpu_flush_tlb_all(vcpu);
9737 /* Flushing all ASIDs flushes the current ASID... */
9738 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9740 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9741 kvm_vcpu_flush_tlb_current(vcpu);
9742 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9743 kvm_vcpu_flush_tlb_guest(vcpu);
9745 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9746 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9750 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9751 if (is_guest_mode(vcpu)) {
9752 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9754 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9755 vcpu->mmio_needed = 0;
9760 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9761 /* Page is swapped out. Do synthetic halt */
9762 vcpu->arch.apf.halted = true;
9766 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9767 record_steal_time(vcpu);
9768 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9770 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9772 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9773 kvm_pmu_handle_event(vcpu);
9774 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9775 kvm_pmu_deliver_pmi(vcpu);
9776 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9777 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9778 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9779 vcpu->arch.ioapic_handled_vectors)) {
9780 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9781 vcpu->run->eoi.vector =
9782 vcpu->arch.pending_ioapic_eoi;
9787 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9788 vcpu_scan_ioapic(vcpu);
9789 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9790 vcpu_load_eoi_exitmap(vcpu);
9791 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9792 kvm_vcpu_reload_apic_access_page(vcpu);
9793 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9794 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9795 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9799 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9800 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9801 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9805 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9806 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9808 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9809 vcpu->run->hyperv = hv_vcpu->exit;
9815 * KVM_REQ_HV_STIMER has to be processed after
9816 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9817 * depend on the guest clock being up-to-date
9819 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9820 kvm_hv_process_stimers(vcpu);
9821 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9822 kvm_vcpu_update_apicv(vcpu);
9823 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9824 kvm_check_async_pf_completion(vcpu);
9825 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9826 static_call(kvm_x86_msr_filter_changed)(vcpu);
9828 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9829 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9832 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9833 kvm_xen_has_interrupt(vcpu)) {
9834 ++vcpu->stat.req_event;
9835 r = kvm_apic_accept_events(vcpu);
9840 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9845 r = inject_pending_event(vcpu, &req_immediate_exit);
9851 static_call(kvm_x86_enable_irq_window)(vcpu);
9853 if (kvm_lapic_enabled(vcpu)) {
9854 update_cr8_intercept(vcpu);
9855 kvm_lapic_sync_to_vapic(vcpu);
9859 r = kvm_mmu_reload(vcpu);
9861 goto cancel_injection;
9866 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9869 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9870 * IPI are then delayed after guest entry, which ensures that they
9871 * result in virtual interrupt delivery.
9873 local_irq_disable();
9874 vcpu->mode = IN_GUEST_MODE;
9876 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9879 * 1) We should set ->mode before checking ->requests. Please see
9880 * the comment in kvm_vcpu_exiting_guest_mode().
9882 * 2) For APICv, we should set ->mode before checking PID.ON. This
9883 * pairs with the memory barrier implicit in pi_test_and_set_on
9884 * (see vmx_deliver_posted_interrupt).
9886 * 3) This also orders the write to mode from any reads to the page
9887 * tables done while the VCPU is running. Please see the comment
9888 * in kvm_flush_remote_tlbs.
9890 smp_mb__after_srcu_read_unlock();
9893 * This handles the case where a posted interrupt was
9894 * notified with kvm_vcpu_kick.
9896 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9897 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9899 if (kvm_vcpu_exit_request(vcpu)) {
9900 vcpu->mode = OUTSIDE_GUEST_MODE;
9904 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9906 goto cancel_injection;
9909 if (req_immediate_exit) {
9910 kvm_make_request(KVM_REQ_EVENT, vcpu);
9911 static_call(kvm_x86_request_immediate_exit)(vcpu);
9914 fpregs_assert_state_consistent();
9915 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9916 switch_fpu_return();
9918 if (unlikely(vcpu->arch.switch_db_regs)) {
9920 set_debugreg(vcpu->arch.eff_db[0], 0);
9921 set_debugreg(vcpu->arch.eff_db[1], 1);
9922 set_debugreg(vcpu->arch.eff_db[2], 2);
9923 set_debugreg(vcpu->arch.eff_db[3], 3);
9924 } else if (unlikely(hw_breakpoint_active())) {
9930 * Assert that vCPU vs. VM APICv state is consistent. An APICv
9931 * update must kick and wait for all vCPUs before toggling the
9932 * per-VM state, and responsing vCPUs must wait for the update
9933 * to complete before servicing KVM_REQ_APICV_UPDATE.
9935 WARN_ON_ONCE(kvm_apicv_activated(vcpu->kvm) != kvm_vcpu_apicv_active(vcpu));
9937 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9938 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9941 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9942 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9946 if (vcpu->arch.apicv_active)
9947 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9951 * Do this here before restoring debug registers on the host. And
9952 * since we do this before handling the vmexit, a DR access vmexit
9953 * can (a) read the correct value of the debug registers, (b) set
9954 * KVM_DEBUGREG_WONT_EXIT again.
9956 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9957 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9958 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9959 kvm_update_dr0123(vcpu);
9960 kvm_update_dr7(vcpu);
9964 * If the guest has used debug registers, at least dr7
9965 * will be disabled while returning to the host.
9966 * If we don't have active breakpoints in the host, we don't
9967 * care about the messed up debug address registers. But if
9968 * we have some of them active, restore the old state.
9970 if (hw_breakpoint_active())
9971 hw_breakpoint_restore();
9973 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9974 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9976 vcpu->mode = OUTSIDE_GUEST_MODE;
9979 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9982 * Consume any pending interrupts, including the possible source of
9983 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9984 * An instruction is required after local_irq_enable() to fully unblock
9985 * interrupts on processors that implement an interrupt shadow, the
9986 * stat.exits increment will do nicely.
9988 kvm_before_interrupt(vcpu);
9991 local_irq_disable();
9992 kvm_after_interrupt(vcpu);
9995 * Wait until after servicing IRQs to account guest time so that any
9996 * ticks that occurred while running the guest are properly accounted
9997 * to the guest. Waiting until IRQs are enabled degrades the accuracy
9998 * of accounting via context tracking, but the loss of accuracy is
9999 * acceptable for all known use cases.
10001 vtime_account_guest_exit();
10003 if (lapic_in_kernel(vcpu)) {
10004 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
10005 if (delta != S64_MIN) {
10006 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
10007 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
10011 local_irq_enable();
10014 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
10017 * Profile KVM exit RIPs:
10019 if (unlikely(prof_on == KVM_PROFILING)) {
10020 unsigned long rip = kvm_rip_read(vcpu);
10021 profile_hit(KVM_PROFILING, (void *)rip);
10024 if (unlikely(vcpu->arch.tsc_always_catchup))
10025 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10027 if (vcpu->arch.apic_attention)
10028 kvm_lapic_sync_from_vapic(vcpu);
10030 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
10034 if (req_immediate_exit)
10035 kvm_make_request(KVM_REQ_EVENT, vcpu);
10036 static_call(kvm_x86_cancel_injection)(vcpu);
10037 if (unlikely(vcpu->arch.apic_attention))
10038 kvm_lapic_sync_from_vapic(vcpu);
10043 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
10045 if (!kvm_arch_vcpu_runnable(vcpu) &&
10046 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
10047 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10048 kvm_vcpu_block(vcpu);
10049 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10051 if (kvm_x86_ops.post_block)
10052 static_call(kvm_x86_post_block)(vcpu);
10054 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
10058 if (kvm_apic_accept_events(vcpu) < 0)
10060 switch(vcpu->arch.mp_state) {
10061 case KVM_MP_STATE_HALTED:
10062 case KVM_MP_STATE_AP_RESET_HOLD:
10063 vcpu->arch.pv.pv_unhalted = false;
10064 vcpu->arch.mp_state =
10065 KVM_MP_STATE_RUNNABLE;
10067 case KVM_MP_STATE_RUNNABLE:
10068 vcpu->arch.apf.halted = false;
10070 case KVM_MP_STATE_INIT_RECEIVED:
10078 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
10080 if (is_guest_mode(vcpu))
10081 kvm_check_nested_events(vcpu);
10083 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
10084 !vcpu->arch.apf.halted);
10087 static int vcpu_run(struct kvm_vcpu *vcpu)
10090 struct kvm *kvm = vcpu->kvm;
10092 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10093 vcpu->arch.l1tf_flush_l1d = true;
10096 if (kvm_vcpu_running(vcpu)) {
10097 r = vcpu_enter_guest(vcpu);
10099 r = vcpu_block(kvm, vcpu);
10105 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10106 if (kvm_cpu_has_pending_timer(vcpu))
10107 kvm_inject_pending_timer_irqs(vcpu);
10109 if (dm_request_for_irq_injection(vcpu) &&
10110 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10112 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10113 ++vcpu->stat.request_irq_exits;
10117 if (__xfer_to_guest_mode_work_pending()) {
10118 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10119 r = xfer_to_guest_mode_handle_work(vcpu);
10122 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10126 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10131 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
10135 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
10136 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
10137 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
10141 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
10143 BUG_ON(!vcpu->arch.pio.count);
10145 return complete_emulated_io(vcpu);
10149 * Implements the following, as a state machine:
10152 * for each fragment
10153 * for each mmio piece in the fragment
10160 * for each fragment
10161 * for each mmio piece in the fragment
10166 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10168 struct kvm_run *run = vcpu->run;
10169 struct kvm_mmio_fragment *frag;
10172 BUG_ON(!vcpu->mmio_needed);
10174 /* Complete previous fragment */
10175 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10176 len = min(8u, frag->len);
10177 if (!vcpu->mmio_is_write)
10178 memcpy(frag->data, run->mmio.data, len);
10180 if (frag->len <= 8) {
10181 /* Switch to the next fragment. */
10183 vcpu->mmio_cur_fragment++;
10185 /* Go forward to the next mmio piece. */
10191 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10192 vcpu->mmio_needed = 0;
10194 /* FIXME: return into emulator if single-stepping. */
10195 if (vcpu->mmio_is_write)
10197 vcpu->mmio_read_completed = 1;
10198 return complete_emulated_io(vcpu);
10201 run->exit_reason = KVM_EXIT_MMIO;
10202 run->mmio.phys_addr = frag->gpa;
10203 if (vcpu->mmio_is_write)
10204 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10205 run->mmio.len = min(8u, frag->len);
10206 run->mmio.is_write = vcpu->mmio_is_write;
10207 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10211 static void kvm_save_current_fpu(struct fpu *fpu)
10214 * If the target FPU state is not resident in the CPU registers, just
10215 * memcpy() from current, else save CPU state directly to the target.
10217 if (test_thread_flag(TIF_NEED_FPU_LOAD))
10218 memcpy(&fpu->state, ¤t->thread.fpu.state,
10219 fpu_kernel_xstate_size);
10221 save_fpregs_to_fpstate(fpu);
10224 /* Swap (qemu) user FPU context for the guest FPU context. */
10225 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10229 kvm_save_current_fpu(vcpu->arch.user_fpu);
10232 * Guests with protected state can't have it set by the hypervisor,
10233 * so skip trying to set it.
10235 if (vcpu->arch.guest_fpu)
10236 /* PKRU is separately restored in kvm_x86_ops.run. */
10237 __restore_fpregs_from_fpstate(&vcpu->arch.guest_fpu->state,
10238 ~XFEATURE_MASK_PKRU);
10240 fpregs_mark_activate();
10246 /* When vcpu_run ends, restore user space FPU context. */
10247 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10252 * Guests with protected state can't have it read by the hypervisor,
10253 * so skip trying to save it.
10255 if (vcpu->arch.guest_fpu)
10256 kvm_save_current_fpu(vcpu->arch.guest_fpu);
10258 restore_fpregs_from_fpstate(&vcpu->arch.user_fpu->state);
10260 fpregs_mark_activate();
10263 ++vcpu->stat.fpu_reload;
10267 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10269 struct kvm_run *kvm_run = vcpu->run;
10273 kvm_sigset_activate(vcpu);
10274 kvm_run->flags = 0;
10275 kvm_load_guest_fpu(vcpu);
10277 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10278 if (kvm_run->immediate_exit) {
10282 kvm_vcpu_block(vcpu);
10283 if (kvm_apic_accept_events(vcpu) < 0) {
10287 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
10289 if (signal_pending(current)) {
10291 kvm_run->exit_reason = KVM_EXIT_INTR;
10292 ++vcpu->stat.signal_exits;
10297 if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10298 (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10303 if (kvm_run->kvm_dirty_regs) {
10304 r = sync_regs(vcpu);
10309 /* re-sync apic's tpr */
10310 if (!lapic_in_kernel(vcpu)) {
10311 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10317 if (unlikely(vcpu->arch.complete_userspace_io)) {
10318 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10319 vcpu->arch.complete_userspace_io = NULL;
10324 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10326 if (kvm_run->immediate_exit)
10329 r = vcpu_run(vcpu);
10332 kvm_put_guest_fpu(vcpu);
10333 if (kvm_run->kvm_valid_regs)
10335 post_kvm_run_save(vcpu);
10336 kvm_sigset_deactivate(vcpu);
10342 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10344 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10346 * We are here if userspace calls get_regs() in the middle of
10347 * instruction emulation. Registers state needs to be copied
10348 * back from emulation context to vcpu. Userspace shouldn't do
10349 * that usually, but some bad designed PV devices (vmware
10350 * backdoor interface) need this to work
10352 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10353 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10355 regs->rax = kvm_rax_read(vcpu);
10356 regs->rbx = kvm_rbx_read(vcpu);
10357 regs->rcx = kvm_rcx_read(vcpu);
10358 regs->rdx = kvm_rdx_read(vcpu);
10359 regs->rsi = kvm_rsi_read(vcpu);
10360 regs->rdi = kvm_rdi_read(vcpu);
10361 regs->rsp = kvm_rsp_read(vcpu);
10362 regs->rbp = kvm_rbp_read(vcpu);
10363 #ifdef CONFIG_X86_64
10364 regs->r8 = kvm_r8_read(vcpu);
10365 regs->r9 = kvm_r9_read(vcpu);
10366 regs->r10 = kvm_r10_read(vcpu);
10367 regs->r11 = kvm_r11_read(vcpu);
10368 regs->r12 = kvm_r12_read(vcpu);
10369 regs->r13 = kvm_r13_read(vcpu);
10370 regs->r14 = kvm_r14_read(vcpu);
10371 regs->r15 = kvm_r15_read(vcpu);
10374 regs->rip = kvm_rip_read(vcpu);
10375 regs->rflags = kvm_get_rflags(vcpu);
10378 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10381 __get_regs(vcpu, regs);
10386 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10388 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10389 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10391 kvm_rax_write(vcpu, regs->rax);
10392 kvm_rbx_write(vcpu, regs->rbx);
10393 kvm_rcx_write(vcpu, regs->rcx);
10394 kvm_rdx_write(vcpu, regs->rdx);
10395 kvm_rsi_write(vcpu, regs->rsi);
10396 kvm_rdi_write(vcpu, regs->rdi);
10397 kvm_rsp_write(vcpu, regs->rsp);
10398 kvm_rbp_write(vcpu, regs->rbp);
10399 #ifdef CONFIG_X86_64
10400 kvm_r8_write(vcpu, regs->r8);
10401 kvm_r9_write(vcpu, regs->r9);
10402 kvm_r10_write(vcpu, regs->r10);
10403 kvm_r11_write(vcpu, regs->r11);
10404 kvm_r12_write(vcpu, regs->r12);
10405 kvm_r13_write(vcpu, regs->r13);
10406 kvm_r14_write(vcpu, regs->r14);
10407 kvm_r15_write(vcpu, regs->r15);
10410 kvm_rip_write(vcpu, regs->rip);
10411 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10413 vcpu->arch.exception.pending = false;
10415 kvm_make_request(KVM_REQ_EVENT, vcpu);
10418 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10421 __set_regs(vcpu, regs);
10426 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10428 struct kvm_segment cs;
10430 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10434 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10436 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10438 struct desc_ptr dt;
10440 if (vcpu->arch.guest_state_protected)
10441 goto skip_protected_regs;
10443 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10444 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10445 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10446 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10447 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10448 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10450 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10451 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10453 static_call(kvm_x86_get_idt)(vcpu, &dt);
10454 sregs->idt.limit = dt.size;
10455 sregs->idt.base = dt.address;
10456 static_call(kvm_x86_get_gdt)(vcpu, &dt);
10457 sregs->gdt.limit = dt.size;
10458 sregs->gdt.base = dt.address;
10460 sregs->cr2 = vcpu->arch.cr2;
10461 sregs->cr3 = kvm_read_cr3(vcpu);
10463 skip_protected_regs:
10464 sregs->cr0 = kvm_read_cr0(vcpu);
10465 sregs->cr4 = kvm_read_cr4(vcpu);
10466 sregs->cr8 = kvm_get_cr8(vcpu);
10467 sregs->efer = vcpu->arch.efer;
10468 sregs->apic_base = kvm_get_apic_base(vcpu);
10471 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10473 __get_sregs_common(vcpu, sregs);
10475 if (vcpu->arch.guest_state_protected)
10478 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10479 set_bit(vcpu->arch.interrupt.nr,
10480 (unsigned long *)sregs->interrupt_bitmap);
10483 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10487 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10489 if (vcpu->arch.guest_state_protected)
10492 if (is_pae_paging(vcpu)) {
10493 for (i = 0 ; i < 4 ; i++)
10494 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10495 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10499 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10500 struct kvm_sregs *sregs)
10503 __get_sregs(vcpu, sregs);
10508 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10509 struct kvm_mp_state *mp_state)
10514 if (kvm_mpx_supported())
10515 kvm_load_guest_fpu(vcpu);
10517 r = kvm_apic_accept_events(vcpu);
10522 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10523 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10524 vcpu->arch.pv.pv_unhalted)
10525 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10527 mp_state->mp_state = vcpu->arch.mp_state;
10530 if (kvm_mpx_supported())
10531 kvm_put_guest_fpu(vcpu);
10536 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10537 struct kvm_mp_state *mp_state)
10543 if (!lapic_in_kernel(vcpu) &&
10544 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10548 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10549 * INIT state; latched init should be reported using
10550 * KVM_SET_VCPU_EVENTS, so reject it here.
10552 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10553 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10554 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10557 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10558 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10559 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10561 vcpu->arch.mp_state = mp_state->mp_state;
10562 kvm_make_request(KVM_REQ_EVENT, vcpu);
10570 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10571 int reason, bool has_error_code, u32 error_code)
10573 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10576 init_emulate_ctxt(vcpu);
10578 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10579 has_error_code, error_code);
10581 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10582 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10583 vcpu->run->internal.ndata = 0;
10587 kvm_rip_write(vcpu, ctxt->eip);
10588 kvm_set_rflags(vcpu, ctxt->eflags);
10591 EXPORT_SYMBOL_GPL(kvm_task_switch);
10593 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10595 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10597 * When EFER.LME and CR0.PG are set, the processor is in
10598 * 64-bit mode (though maybe in a 32-bit code segment).
10599 * CR4.PAE and EFER.LMA must be set.
10601 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10603 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10607 * Not in 64-bit mode: EFER.LMA is clear and the code
10608 * segment cannot be 64-bit.
10610 if (sregs->efer & EFER_LMA || sregs->cs.l)
10614 return kvm_is_valid_cr4(vcpu, sregs->cr4);
10617 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10618 int *mmu_reset_needed, bool update_pdptrs)
10620 struct msr_data apic_base_msr;
10622 struct desc_ptr dt;
10624 if (!kvm_is_valid_sregs(vcpu, sregs))
10627 apic_base_msr.data = sregs->apic_base;
10628 apic_base_msr.host_initiated = true;
10629 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10632 if (vcpu->arch.guest_state_protected)
10635 dt.size = sregs->idt.limit;
10636 dt.address = sregs->idt.base;
10637 static_call(kvm_x86_set_idt)(vcpu, &dt);
10638 dt.size = sregs->gdt.limit;
10639 dt.address = sregs->gdt.base;
10640 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10642 vcpu->arch.cr2 = sregs->cr2;
10643 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10644 vcpu->arch.cr3 = sregs->cr3;
10645 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10647 kvm_set_cr8(vcpu, sregs->cr8);
10649 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10650 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10652 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10653 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10654 vcpu->arch.cr0 = sregs->cr0;
10656 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10657 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10659 if (update_pdptrs) {
10660 idx = srcu_read_lock(&vcpu->kvm->srcu);
10661 if (is_pae_paging(vcpu)) {
10662 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10663 *mmu_reset_needed = 1;
10665 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10668 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10669 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10670 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10671 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10672 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10673 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10675 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10676 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10678 update_cr8_intercept(vcpu);
10680 /* Older userspace won't unhalt the vcpu on reset. */
10681 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10682 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10683 !is_protmode(vcpu))
10684 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10689 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10691 int pending_vec, max_bits;
10692 int mmu_reset_needed = 0;
10693 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10698 if (mmu_reset_needed)
10699 kvm_mmu_reset_context(vcpu);
10701 max_bits = KVM_NR_INTERRUPTS;
10702 pending_vec = find_first_bit(
10703 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10705 if (pending_vec < max_bits) {
10706 kvm_queue_interrupt(vcpu, pending_vec, false);
10707 pr_debug("Set back pending irq %d\n", pending_vec);
10708 kvm_make_request(KVM_REQ_EVENT, vcpu);
10713 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10715 int mmu_reset_needed = 0;
10716 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10717 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10718 !(sregs2->efer & EFER_LMA);
10721 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10724 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10727 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10728 &mmu_reset_needed, !valid_pdptrs);
10732 if (valid_pdptrs) {
10733 for (i = 0; i < 4 ; i++)
10734 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10736 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10737 mmu_reset_needed = 1;
10738 vcpu->arch.pdptrs_from_userspace = true;
10740 if (mmu_reset_needed)
10741 kvm_mmu_reset_context(vcpu);
10745 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10746 struct kvm_sregs *sregs)
10751 ret = __set_sregs(vcpu, sregs);
10756 static void kvm_arch_vcpu_guestdbg_update_apicv_inhibit(struct kvm *kvm)
10758 bool inhibit = false;
10759 struct kvm_vcpu *vcpu;
10762 down_write(&kvm->arch.apicv_update_lock);
10764 kvm_for_each_vcpu(i, vcpu, kvm) {
10765 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ) {
10770 __kvm_request_apicv_update(kvm, !inhibit, APICV_INHIBIT_REASON_BLOCKIRQ);
10771 up_write(&kvm->arch.apicv_update_lock);
10774 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10775 struct kvm_guest_debug *dbg)
10777 unsigned long rflags;
10780 if (vcpu->arch.guest_state_protected)
10785 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10787 if (vcpu->arch.exception.pending)
10789 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10790 kvm_queue_exception(vcpu, DB_VECTOR);
10792 kvm_queue_exception(vcpu, BP_VECTOR);
10796 * Read rflags as long as potentially injected trace flags are still
10799 rflags = kvm_get_rflags(vcpu);
10801 vcpu->guest_debug = dbg->control;
10802 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10803 vcpu->guest_debug = 0;
10805 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10806 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10807 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10808 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10810 for (i = 0; i < KVM_NR_DB_REGS; i++)
10811 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10813 kvm_update_dr7(vcpu);
10815 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10816 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10819 * Trigger an rflags update that will inject or remove the trace
10822 kvm_set_rflags(vcpu, rflags);
10824 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10826 kvm_arch_vcpu_guestdbg_update_apicv_inhibit(vcpu->kvm);
10836 * Translate a guest virtual address to a guest physical address.
10838 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10839 struct kvm_translation *tr)
10841 unsigned long vaddr = tr->linear_address;
10847 idx = srcu_read_lock(&vcpu->kvm->srcu);
10848 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10849 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10850 tr->physical_address = gpa;
10851 tr->valid = gpa != UNMAPPED_GVA;
10859 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10861 struct fxregs_state *fxsave;
10863 if (!vcpu->arch.guest_fpu)
10868 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10869 memcpy(fpu->fpr, fxsave->st_space, 128);
10870 fpu->fcw = fxsave->cwd;
10871 fpu->fsw = fxsave->swd;
10872 fpu->ftwx = fxsave->twd;
10873 fpu->last_opcode = fxsave->fop;
10874 fpu->last_ip = fxsave->rip;
10875 fpu->last_dp = fxsave->rdp;
10876 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10882 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10884 struct fxregs_state *fxsave;
10886 if (!vcpu->arch.guest_fpu)
10891 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10893 memcpy(fxsave->st_space, fpu->fpr, 128);
10894 fxsave->cwd = fpu->fcw;
10895 fxsave->swd = fpu->fsw;
10896 fxsave->twd = fpu->ftwx;
10897 fxsave->fop = fpu->last_opcode;
10898 fxsave->rip = fpu->last_ip;
10899 fxsave->rdp = fpu->last_dp;
10900 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10906 static void store_regs(struct kvm_vcpu *vcpu)
10908 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10910 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10911 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10913 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10914 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10916 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10917 kvm_vcpu_ioctl_x86_get_vcpu_events(
10918 vcpu, &vcpu->run->s.regs.events);
10921 static int sync_regs(struct kvm_vcpu *vcpu)
10923 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10924 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10925 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10927 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10928 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10930 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10932 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10933 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10934 vcpu, &vcpu->run->s.regs.events))
10936 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10942 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10944 if (vcpu->arch.guest_fpu) {
10945 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10946 vcpu->arch.guest_fpu = NULL;
10949 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10951 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10953 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10954 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10955 "guest TSC will not be reliable\n");
10960 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10965 vcpu->arch.last_vmentry_cpu = -1;
10966 vcpu->arch.regs_avail = ~0;
10967 vcpu->arch.regs_dirty = ~0;
10969 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10970 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10972 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10974 r = kvm_mmu_create(vcpu);
10978 if (irqchip_in_kernel(vcpu->kvm)) {
10979 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10981 goto fail_mmu_destroy;
10982 if (kvm_apicv_activated(vcpu->kvm))
10983 vcpu->arch.apicv_active = true;
10985 static_branch_inc(&kvm_has_noapic_vcpu);
10989 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10991 goto fail_free_lapic;
10992 vcpu->arch.pio_data = page_address(page);
10994 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10995 GFP_KERNEL_ACCOUNT);
10996 if (!vcpu->arch.mce_banks)
10997 goto fail_free_pio_data;
10998 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
11000 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
11001 GFP_KERNEL_ACCOUNT))
11002 goto fail_free_mce_banks;
11004 if (!alloc_emulate_ctxt(vcpu))
11005 goto free_wbinvd_dirty_mask;
11007 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
11008 GFP_KERNEL_ACCOUNT);
11009 if (!vcpu->arch.user_fpu) {
11010 pr_err("kvm: failed to allocate userspace's fpu\n");
11011 goto free_emulate_ctxt;
11014 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
11015 GFP_KERNEL_ACCOUNT);
11016 if (!vcpu->arch.guest_fpu) {
11017 pr_err("kvm: failed to allocate vcpu's fpu\n");
11018 goto free_user_fpu;
11020 fpstate_init(&vcpu->arch.guest_fpu->state);
11021 if (boot_cpu_has(X86_FEATURE_XSAVES))
11022 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
11023 host_xcr0 | XSTATE_COMPACTION_ENABLED;
11025 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
11026 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
11028 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
11030 kvm_async_pf_hash_reset(vcpu);
11031 kvm_pmu_init(vcpu);
11033 vcpu->arch.pending_external_vector = -1;
11034 vcpu->arch.preempted_in_kernel = false;
11036 #if IS_ENABLED(CONFIG_HYPERV)
11037 vcpu->arch.hv_root_tdp = INVALID_PAGE;
11040 r = static_call(kvm_x86_vcpu_create)(vcpu);
11042 goto free_guest_fpu;
11044 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
11045 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
11046 kvm_vcpu_mtrr_init(vcpu);
11048 kvm_set_tsc_khz(vcpu, max_tsc_khz);
11049 kvm_vcpu_reset(vcpu, false);
11050 kvm_init_mmu(vcpu);
11055 kvm_free_guest_fpu(vcpu);
11057 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
11059 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11060 free_wbinvd_dirty_mask:
11061 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11062 fail_free_mce_banks:
11063 kfree(vcpu->arch.mce_banks);
11064 fail_free_pio_data:
11065 free_page((unsigned long)vcpu->arch.pio_data);
11067 kvm_free_lapic(vcpu);
11069 kvm_mmu_destroy(vcpu);
11073 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
11075 struct kvm *kvm = vcpu->kvm;
11077 if (mutex_lock_killable(&vcpu->mutex))
11080 kvm_synchronize_tsc(vcpu, 0);
11083 /* poll control enabled by default */
11084 vcpu->arch.msr_kvm_poll_control = 1;
11086 mutex_unlock(&vcpu->mutex);
11088 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
11089 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
11090 KVMCLOCK_SYNC_PERIOD);
11093 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
11097 kvmclock_reset(vcpu);
11099 static_call(kvm_x86_vcpu_free)(vcpu);
11101 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11102 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11103 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
11104 kvm_free_guest_fpu(vcpu);
11106 kvm_hv_vcpu_uninit(vcpu);
11107 kvm_pmu_destroy(vcpu);
11108 kfree(vcpu->arch.mce_banks);
11109 kvm_free_lapic(vcpu);
11110 idx = srcu_read_lock(&vcpu->kvm->srcu);
11111 kvm_mmu_destroy(vcpu);
11112 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11113 free_page((unsigned long)vcpu->arch.pio_data);
11114 kvfree(vcpu->arch.cpuid_entries);
11115 if (!lapic_in_kernel(vcpu))
11116 static_branch_dec(&kvm_has_noapic_vcpu);
11119 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
11121 struct kvm_cpuid_entry2 *cpuid_0x1;
11122 unsigned long old_cr0 = kvm_read_cr0(vcpu);
11123 unsigned long new_cr0;
11126 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
11127 * to handle side effects. RESET emulation hits those flows and relies
11128 * on emulated/virtualized registers, including those that are loaded
11129 * into hardware, to be zeroed at vCPU creation. Use CRs as a sentinel
11130 * to detect improper or missing initialization.
11132 WARN_ON_ONCE(!init_event &&
11133 (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
11135 kvm_lapic_reset(vcpu, init_event);
11137 vcpu->arch.hflags = 0;
11139 vcpu->arch.smi_pending = 0;
11140 vcpu->arch.smi_count = 0;
11141 atomic_set(&vcpu->arch.nmi_queued, 0);
11142 vcpu->arch.nmi_pending = 0;
11143 vcpu->arch.nmi_injected = false;
11144 kvm_clear_interrupt_queue(vcpu);
11145 kvm_clear_exception_queue(vcpu);
11147 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
11148 kvm_update_dr0123(vcpu);
11149 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
11150 vcpu->arch.dr7 = DR7_FIXED_1;
11151 kvm_update_dr7(vcpu);
11153 vcpu->arch.cr2 = 0;
11155 kvm_make_request(KVM_REQ_EVENT, vcpu);
11156 vcpu->arch.apf.msr_en_val = 0;
11157 vcpu->arch.apf.msr_int_val = 0;
11158 vcpu->arch.st.msr_val = 0;
11160 kvmclock_reset(vcpu);
11162 kvm_clear_async_pf_completion_queue(vcpu);
11163 kvm_async_pf_hash_reset(vcpu);
11164 vcpu->arch.apf.halted = false;
11166 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
11167 void *mpx_state_buffer;
11170 * To avoid have the INIT path from kvm_apic_has_events() that be
11171 * called with loaded FPU and does not let userspace fix the state.
11174 kvm_put_guest_fpu(vcpu);
11175 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
11177 if (mpx_state_buffer)
11178 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
11179 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
11181 if (mpx_state_buffer)
11182 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
11184 kvm_load_guest_fpu(vcpu);
11188 kvm_pmu_reset(vcpu);
11189 vcpu->arch.smbase = 0x30000;
11191 vcpu->arch.msr_misc_features_enables = 0;
11193 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
11196 /* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
11197 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
11198 kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
11201 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
11202 * if no CPUID match is found. Note, it's impossible to get a match at
11203 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
11204 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
11205 * on RESET. But, go through the motions in case that's ever remedied.
11207 cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
11208 kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
11210 vcpu->arch.ia32_xss = 0;
11212 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
11214 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
11215 kvm_rip_write(vcpu, 0xfff0);
11217 vcpu->arch.cr3 = 0;
11218 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11221 * CR0.CD/NW are set on RESET, preserved on INIT. Note, some versions
11222 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
11223 * (or qualify) that with a footnote stating that CD/NW are preserved.
11225 new_cr0 = X86_CR0_ET;
11227 new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
11229 new_cr0 |= X86_CR0_NW | X86_CR0_CD;
11231 static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
11232 static_call(kvm_x86_set_cr4)(vcpu, 0);
11233 static_call(kvm_x86_set_efer)(vcpu, 0);
11234 static_call(kvm_x86_update_exception_bitmap)(vcpu);
11237 * Reset the MMU context if paging was enabled prior to INIT (which is
11238 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET). Unlike the
11239 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
11240 * checked because it is unconditionally cleared on INIT and all other
11241 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
11242 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
11244 if (old_cr0 & X86_CR0_PG)
11245 kvm_mmu_reset_context(vcpu);
11248 * Intel's SDM states that all TLB entries are flushed on INIT. AMD's
11249 * APM states the TLBs are untouched by INIT, but it also states that
11250 * the TLBs are flushed on "External initialization of the processor."
11251 * Flush the guest TLB regardless of vendor, there is no meaningful
11252 * benefit in relying on the guest to flush the TLB immediately after
11253 * INIT. A spurious TLB flush is benign and likely negligible from a
11254 * performance perspective.
11257 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11259 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11261 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11263 struct kvm_segment cs;
11265 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
11266 cs.selector = vector << 8;
11267 cs.base = vector << 12;
11268 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
11269 kvm_rip_write(vcpu, 0);
11271 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
11273 int kvm_arch_hardware_enable(void)
11276 struct kvm_vcpu *vcpu;
11281 bool stable, backwards_tsc = false;
11283 kvm_user_return_msr_cpu_online();
11284 ret = static_call(kvm_x86_hardware_enable)();
11288 local_tsc = rdtsc();
11289 stable = !kvm_check_tsc_unstable();
11290 list_for_each_entry(kvm, &vm_list, vm_list) {
11291 kvm_for_each_vcpu(i, vcpu, kvm) {
11292 if (!stable && vcpu->cpu == smp_processor_id())
11293 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
11294 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
11295 backwards_tsc = true;
11296 if (vcpu->arch.last_host_tsc > max_tsc)
11297 max_tsc = vcpu->arch.last_host_tsc;
11303 * Sometimes, even reliable TSCs go backwards. This happens on
11304 * platforms that reset TSC during suspend or hibernate actions, but
11305 * maintain synchronization. We must compensate. Fortunately, we can
11306 * detect that condition here, which happens early in CPU bringup,
11307 * before any KVM threads can be running. Unfortunately, we can't
11308 * bring the TSCs fully up to date with real time, as we aren't yet far
11309 * enough into CPU bringup that we know how much real time has actually
11310 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
11311 * variables that haven't been updated yet.
11313 * So we simply find the maximum observed TSC above, then record the
11314 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
11315 * the adjustment will be applied. Note that we accumulate
11316 * adjustments, in case multiple suspend cycles happen before some VCPU
11317 * gets a chance to run again. In the event that no KVM threads get a
11318 * chance to run, we will miss the entire elapsed period, as we'll have
11319 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
11320 * loose cycle time. This isn't too big a deal, since the loss will be
11321 * uniform across all VCPUs (not to mention the scenario is extremely
11322 * unlikely). It is possible that a second hibernate recovery happens
11323 * much faster than a first, causing the observed TSC here to be
11324 * smaller; this would require additional padding adjustment, which is
11325 * why we set last_host_tsc to the local tsc observed here.
11327 * N.B. - this code below runs only on platforms with reliable TSC,
11328 * as that is the only way backwards_tsc is set above. Also note
11329 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
11330 * have the same delta_cyc adjustment applied if backwards_tsc
11331 * is detected. Note further, this adjustment is only done once,
11332 * as we reset last_host_tsc on all VCPUs to stop this from being
11333 * called multiple times (one for each physical CPU bringup).
11335 * Platforms with unreliable TSCs don't have to deal with this, they
11336 * will be compensated by the logic in vcpu_load, which sets the TSC to
11337 * catchup mode. This will catchup all VCPUs to real time, but cannot
11338 * guarantee that they stay in perfect synchronization.
11340 if (backwards_tsc) {
11341 u64 delta_cyc = max_tsc - local_tsc;
11342 list_for_each_entry(kvm, &vm_list, vm_list) {
11343 kvm->arch.backwards_tsc_observed = true;
11344 kvm_for_each_vcpu(i, vcpu, kvm) {
11345 vcpu->arch.tsc_offset_adjustment += delta_cyc;
11346 vcpu->arch.last_host_tsc = local_tsc;
11347 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11351 * We have to disable TSC offset matching.. if you were
11352 * booting a VM while issuing an S4 host suspend....
11353 * you may have some problem. Solving this issue is
11354 * left as an exercise to the reader.
11356 kvm->arch.last_tsc_nsec = 0;
11357 kvm->arch.last_tsc_write = 0;
11364 void kvm_arch_hardware_disable(void)
11366 static_call(kvm_x86_hardware_disable)();
11367 drop_user_return_notifiers();
11370 int kvm_arch_hardware_setup(void *opaque)
11372 struct kvm_x86_init_ops *ops = opaque;
11375 rdmsrl_safe(MSR_EFER, &host_efer);
11377 if (boot_cpu_has(X86_FEATURE_XSAVES))
11378 rdmsrl(MSR_IA32_XSS, host_xss);
11380 r = ops->hardware_setup();
11384 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11385 kvm_ops_static_call_update();
11387 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11390 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11391 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11392 #undef __kvm_cpu_cap_has
11394 if (kvm_has_tsc_control) {
11396 * Make sure the user can only configure tsc_khz values that
11397 * fit into a signed integer.
11398 * A min value is not calculated because it will always
11399 * be 1 on all machines.
11401 u64 max = min(0x7fffffffULL,
11402 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11403 kvm_max_guest_tsc_khz = max;
11405 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11408 kvm_init_msr_list();
11412 void kvm_arch_hardware_unsetup(void)
11414 static_call(kvm_x86_hardware_unsetup)();
11417 int kvm_arch_check_processor_compat(void *opaque)
11419 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11420 struct kvm_x86_init_ops *ops = opaque;
11422 WARN_ON(!irqs_disabled());
11424 if (__cr4_reserved_bits(cpu_has, c) !=
11425 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11428 return ops->check_processor_compatibility();
11431 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11433 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11435 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11437 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11439 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11442 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11443 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11445 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11447 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11449 vcpu->arch.l1tf_flush_l1d = true;
11450 if (pmu->version && unlikely(pmu->event_count)) {
11451 pmu->need_cleanup = true;
11452 kvm_make_request(KVM_REQ_PMU, vcpu);
11454 static_call(kvm_x86_sched_in)(vcpu, cpu);
11457 void kvm_arch_free_vm(struct kvm *kvm)
11459 kfree(to_kvm_hv(kvm)->hv_pa_pg);
11460 __kvm_arch_free_vm(kvm);
11464 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11467 unsigned long flags;
11472 ret = kvm_page_track_init(kvm);
11476 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11477 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11478 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11479 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11480 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11481 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11483 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11484 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11485 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11486 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11487 &kvm->arch.irq_sources_bitmap);
11489 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11490 mutex_init(&kvm->arch.apic_map_lock);
11491 seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
11492 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11494 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
11495 pvclock_update_vm_gtod_copy(kvm);
11496 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
11498 kvm->arch.guest_can_read_msr_platform_info = true;
11500 #if IS_ENABLED(CONFIG_HYPERV)
11501 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11502 kvm->arch.hv_root_tdp = INVALID_PAGE;
11505 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11506 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11508 kvm_apicv_init(kvm);
11509 kvm_hv_init_vm(kvm);
11510 kvm_mmu_init_vm(kvm);
11511 kvm_xen_init_vm(kvm);
11513 return static_call(kvm_x86_vm_init)(kvm);
11516 int kvm_arch_post_init_vm(struct kvm *kvm)
11518 return kvm_mmu_post_init_vm(kvm);
11521 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11524 kvm_mmu_unload(vcpu);
11528 static void kvm_free_vcpus(struct kvm *kvm)
11531 struct kvm_vcpu *vcpu;
11534 * Unpin any mmu pages first.
11536 kvm_for_each_vcpu(i, vcpu, kvm) {
11537 kvm_clear_async_pf_completion_queue(vcpu);
11538 kvm_unload_vcpu_mmu(vcpu);
11540 kvm_for_each_vcpu(i, vcpu, kvm)
11541 kvm_vcpu_destroy(vcpu);
11543 mutex_lock(&kvm->lock);
11544 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11545 kvm->vcpus[i] = NULL;
11547 atomic_set(&kvm->online_vcpus, 0);
11548 mutex_unlock(&kvm->lock);
11551 void kvm_arch_sync_events(struct kvm *kvm)
11553 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11554 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11558 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
11561 * __x86_set_memory_region: Setup KVM internal memory slot
11563 * @kvm: the kvm pointer to the VM.
11564 * @id: the slot ID to setup.
11565 * @gpa: the GPA to install the slot (unused when @size == 0).
11566 * @size: the size of the slot. Set to zero to uninstall a slot.
11568 * This function helps to setup a KVM internal memory slot. Specify
11569 * @size > 0 to install a new slot, while @size == 0 to uninstall a
11570 * slot. The return code can be one of the following:
11572 * HVA: on success (uninstall will return a bogus HVA)
11575 * The caller should always use IS_ERR() to check the return value
11576 * before use. Note, the KVM internal memory slots are guaranteed to
11577 * remain valid and unchanged until the VM is destroyed, i.e., the
11578 * GPA->HVA translation will not change. However, the HVA is a user
11579 * address, i.e. its accessibility is not guaranteed, and must be
11580 * accessed via __copy_{to,from}_user().
11582 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11586 unsigned long hva, old_npages;
11587 struct kvm_memslots *slots = kvm_memslots(kvm);
11588 struct kvm_memory_slot *slot;
11590 /* Called with kvm->slots_lock held. */
11591 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11592 return ERR_PTR_USR(-EINVAL);
11594 slot = id_to_memslot(slots, id);
11596 if (slot && slot->npages)
11597 return ERR_PTR_USR(-EEXIST);
11600 * MAP_SHARED to prevent internal slot pages from being moved
11603 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11604 MAP_SHARED | MAP_ANONYMOUS, 0);
11605 if (IS_ERR((void *)hva))
11606 return (void __user *)hva;
11608 if (!slot || !slot->npages)
11611 old_npages = slot->npages;
11612 hva = slot->userspace_addr;
11615 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11616 struct kvm_userspace_memory_region m;
11618 m.slot = id | (i << 16);
11620 m.guest_phys_addr = gpa;
11621 m.userspace_addr = hva;
11622 m.memory_size = size;
11623 r = __kvm_set_memory_region(kvm, &m);
11625 return ERR_PTR_USR(r);
11629 vm_munmap(hva, old_npages * PAGE_SIZE);
11631 return (void __user *)hva;
11633 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11635 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11637 kvm_mmu_pre_destroy_vm(kvm);
11640 void kvm_arch_destroy_vm(struct kvm *kvm)
11642 if (current->mm == kvm->mm) {
11644 * Free memory regions allocated on behalf of userspace,
11645 * unless the the memory map has changed due to process exit
11648 mutex_lock(&kvm->slots_lock);
11649 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11651 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11653 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11654 mutex_unlock(&kvm->slots_lock);
11656 static_call_cond(kvm_x86_vm_destroy)(kvm);
11657 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11658 kvm_pic_destroy(kvm);
11659 kvm_ioapic_destroy(kvm);
11660 kvm_free_vcpus(kvm);
11661 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11662 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11663 kvm_mmu_uninit_vm(kvm);
11664 kvm_page_track_cleanup(kvm);
11665 kvm_xen_destroy_vm(kvm);
11666 kvm_hv_destroy_vm(kvm);
11669 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11673 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11674 kvfree(slot->arch.rmap[i]);
11675 slot->arch.rmap[i] = NULL;
11679 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11683 memslot_rmap_free(slot);
11685 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11686 kvfree(slot->arch.lpage_info[i - 1]);
11687 slot->arch.lpage_info[i - 1] = NULL;
11690 kvm_page_track_free_memslot(slot);
11693 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
11695 const int sz = sizeof(*slot->arch.rmap[0]);
11698 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11700 int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11702 if (slot->arch.rmap[i])
11705 slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11706 if (!slot->arch.rmap[i]) {
11707 memslot_rmap_free(slot);
11715 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11716 struct kvm_memory_slot *slot,
11717 unsigned long npages)
11722 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
11723 * old arrays will be freed by __kvm_set_memory_region() if installing
11724 * the new memslot is successful.
11726 memset(&slot->arch, 0, sizeof(slot->arch));
11728 if (kvm_memslots_have_rmaps(kvm)) {
11729 r = memslot_rmap_alloc(slot, npages);
11734 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11735 struct kvm_lpage_info *linfo;
11736 unsigned long ugfn;
11740 lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11742 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11746 slot->arch.lpage_info[i - 1] = linfo;
11748 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11749 linfo[0].disallow_lpage = 1;
11750 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11751 linfo[lpages - 1].disallow_lpage = 1;
11752 ugfn = slot->userspace_addr >> PAGE_SHIFT;
11754 * If the gfn and userspace address are not aligned wrt each
11755 * other, disable large page support for this slot.
11757 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11760 for (j = 0; j < lpages; ++j)
11761 linfo[j].disallow_lpage = 1;
11765 if (kvm_page_track_create_memslot(kvm, slot, npages))
11771 memslot_rmap_free(slot);
11773 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11774 kvfree(slot->arch.lpage_info[i - 1]);
11775 slot->arch.lpage_info[i - 1] = NULL;
11780 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11782 struct kvm_vcpu *vcpu;
11786 * memslots->generation has been incremented.
11787 * mmio generation may have reached its maximum value.
11789 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11791 /* Force re-initialization of steal_time cache */
11792 kvm_for_each_vcpu(i, vcpu, kvm)
11793 kvm_vcpu_kick(vcpu);
11796 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11797 struct kvm_memory_slot *memslot,
11798 const struct kvm_userspace_memory_region *mem,
11799 enum kvm_mr_change change)
11801 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11802 return kvm_alloc_memslot_metadata(kvm, memslot,
11803 mem->memory_size >> PAGE_SHIFT);
11808 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11810 struct kvm_arch *ka = &kvm->arch;
11812 if (!kvm_x86_ops.cpu_dirty_log_size)
11815 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11816 (!enable && --ka->cpu_dirty_logging_count == 0))
11817 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11819 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11822 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11823 struct kvm_memory_slot *old,
11824 const struct kvm_memory_slot *new,
11825 enum kvm_mr_change change)
11827 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11830 * Update CPU dirty logging if dirty logging is being toggled. This
11831 * applies to all operations.
11833 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11834 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11837 * Nothing more to do for RO slots (which can't be dirtied and can't be
11838 * made writable) or CREATE/MOVE/DELETE of a slot.
11840 * For a memslot with dirty logging disabled:
11841 * CREATE: No dirty mappings will already exist.
11842 * MOVE/DELETE: The old mappings will already have been cleaned up by
11843 * kvm_arch_flush_shadow_memslot()
11845 * For a memslot with dirty logging enabled:
11846 * CREATE: No shadow pages exist, thus nothing to write-protect
11847 * and no dirty bits to clear.
11848 * MOVE/DELETE: The old mappings will already have been cleaned up by
11849 * kvm_arch_flush_shadow_memslot().
11851 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11855 * READONLY and non-flags changes were filtered out above, and the only
11856 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11857 * logging isn't being toggled on or off.
11859 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11862 if (!log_dirty_pages) {
11864 * Dirty logging tracks sptes in 4k granularity, meaning that
11865 * large sptes have to be split. If live migration succeeds,
11866 * the guest in the source machine will be destroyed and large
11867 * sptes will be created in the destination. However, if the
11868 * guest continues to run in the source machine (for example if
11869 * live migration fails), small sptes will remain around and
11870 * cause bad performance.
11872 * Scan sptes if dirty logging has been stopped, dropping those
11873 * which can be collapsed into a single large-page spte. Later
11874 * page faults will create the large-page sptes.
11876 kvm_mmu_zap_collapsible_sptes(kvm, new);
11879 * Initially-all-set does not require write protecting any page,
11880 * because they're all assumed to be dirty.
11882 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11885 if (kvm_x86_ops.cpu_dirty_log_size) {
11886 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11887 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11889 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11894 void kvm_arch_commit_memory_region(struct kvm *kvm,
11895 const struct kvm_userspace_memory_region *mem,
11896 struct kvm_memory_slot *old,
11897 const struct kvm_memory_slot *new,
11898 enum kvm_mr_change change)
11900 if (!kvm->arch.n_requested_mmu_pages)
11901 kvm_mmu_change_mmu_pages(kvm,
11902 kvm_mmu_calculate_default_mmu_pages(kvm));
11904 kvm_mmu_slot_apply_flags(kvm, old, new, change);
11906 /* Free the arrays associated with the old memslot. */
11907 if (change == KVM_MR_MOVE)
11908 kvm_arch_free_memslot(kvm, old);
11911 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11913 kvm_mmu_zap_all(kvm);
11916 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11917 struct kvm_memory_slot *slot)
11919 kvm_page_track_flush_slot(kvm, slot);
11922 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11924 return (is_guest_mode(vcpu) &&
11925 kvm_x86_ops.guest_apic_has_interrupt &&
11926 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11929 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11931 if (!list_empty_careful(&vcpu->async_pf.done))
11934 if (kvm_apic_has_events(vcpu))
11937 if (vcpu->arch.pv.pv_unhalted)
11940 if (vcpu->arch.exception.pending)
11943 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11944 (vcpu->arch.nmi_pending &&
11945 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11948 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11949 (vcpu->arch.smi_pending &&
11950 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11953 if (kvm_arch_interrupt_allowed(vcpu) &&
11954 (kvm_cpu_has_interrupt(vcpu) ||
11955 kvm_guest_apic_has_interrupt(vcpu)))
11958 if (kvm_hv_has_stimer_pending(vcpu))
11961 if (is_guest_mode(vcpu) &&
11962 kvm_x86_ops.nested_ops->hv_timer_pending &&
11963 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11969 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11971 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11974 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11976 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11982 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11984 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11987 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11988 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11989 kvm_test_request(KVM_REQ_EVENT, vcpu))
11992 return kvm_arch_dy_has_pending_interrupt(vcpu);
11995 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11997 if (vcpu->arch.guest_state_protected)
12000 return vcpu->arch.preempted_in_kernel;
12003 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
12005 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
12008 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
12010 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
12013 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
12015 /* Can't read the RIP when guest state is protected, just return 0 */
12016 if (vcpu->arch.guest_state_protected)
12019 if (is_64_bit_mode(vcpu))
12020 return kvm_rip_read(vcpu);
12021 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
12022 kvm_rip_read(vcpu));
12024 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
12026 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
12028 return kvm_get_linear_rip(vcpu) == linear_rip;
12030 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
12032 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
12034 unsigned long rflags;
12036 rflags = static_call(kvm_x86_get_rflags)(vcpu);
12037 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
12038 rflags &= ~X86_EFLAGS_TF;
12041 EXPORT_SYMBOL_GPL(kvm_get_rflags);
12043 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12045 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
12046 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
12047 rflags |= X86_EFLAGS_TF;
12048 static_call(kvm_x86_set_rflags)(vcpu, rflags);
12051 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12053 __kvm_set_rflags(vcpu, rflags);
12054 kvm_make_request(KVM_REQ_EVENT, vcpu);
12056 EXPORT_SYMBOL_GPL(kvm_set_rflags);
12058 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
12062 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
12066 r = kvm_mmu_reload(vcpu);
12070 if (!vcpu->arch.mmu->direct_map &&
12071 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
12074 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
12077 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
12079 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
12081 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
12084 static inline u32 kvm_async_pf_next_probe(u32 key)
12086 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
12089 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12091 u32 key = kvm_async_pf_hash_fn(gfn);
12093 while (vcpu->arch.apf.gfns[key] != ~0)
12094 key = kvm_async_pf_next_probe(key);
12096 vcpu->arch.apf.gfns[key] = gfn;
12099 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
12102 u32 key = kvm_async_pf_hash_fn(gfn);
12104 for (i = 0; i < ASYNC_PF_PER_VCPU &&
12105 (vcpu->arch.apf.gfns[key] != gfn &&
12106 vcpu->arch.apf.gfns[key] != ~0); i++)
12107 key = kvm_async_pf_next_probe(key);
12112 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12114 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
12117 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12121 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
12123 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
12127 vcpu->arch.apf.gfns[i] = ~0;
12129 j = kvm_async_pf_next_probe(j);
12130 if (vcpu->arch.apf.gfns[j] == ~0)
12132 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
12134 * k lies cyclically in ]i,j]
12136 * |....j i.k.| or |.k..j i...|
12138 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
12139 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
12144 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
12146 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
12148 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
12152 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
12154 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12156 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12157 &token, offset, sizeof(token));
12160 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
12162 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12165 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12166 &val, offset, sizeof(val)))
12172 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
12174 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
12177 if (!kvm_pv_async_pf_enabled(vcpu) ||
12178 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
12184 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
12186 if (unlikely(!lapic_in_kernel(vcpu) ||
12187 kvm_event_needs_reinjection(vcpu) ||
12188 vcpu->arch.exception.pending))
12191 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
12195 * If interrupts are off we cannot even use an artificial
12198 return kvm_arch_interrupt_allowed(vcpu);
12201 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
12202 struct kvm_async_pf *work)
12204 struct x86_exception fault;
12206 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
12207 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
12209 if (kvm_can_deliver_async_pf(vcpu) &&
12210 !apf_put_user_notpresent(vcpu)) {
12211 fault.vector = PF_VECTOR;
12212 fault.error_code_valid = true;
12213 fault.error_code = 0;
12214 fault.nested_page_fault = false;
12215 fault.address = work->arch.token;
12216 fault.async_page_fault = true;
12217 kvm_inject_page_fault(vcpu, &fault);
12221 * It is not possible to deliver a paravirtualized asynchronous
12222 * page fault, but putting the guest in an artificial halt state
12223 * can be beneficial nevertheless: if an interrupt arrives, we
12224 * can deliver it timely and perhaps the guest will schedule
12225 * another process. When the instruction that triggered a page
12226 * fault is retried, hopefully the page will be ready in the host.
12228 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
12233 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
12234 struct kvm_async_pf *work)
12236 struct kvm_lapic_irq irq = {
12237 .delivery_mode = APIC_DM_FIXED,
12238 .vector = vcpu->arch.apf.vec
12241 if (work->wakeup_all)
12242 work->arch.token = ~0; /* broadcast wakeup */
12244 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12245 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12247 if ((work->wakeup_all || work->notpresent_injected) &&
12248 kvm_pv_async_pf_enabled(vcpu) &&
12249 !apf_put_user_ready(vcpu, work->arch.token)) {
12250 vcpu->arch.apf.pageready_pending = true;
12251 kvm_apic_set_irq(vcpu, &irq, NULL);
12254 vcpu->arch.apf.halted = false;
12255 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
12258 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
12260 kvm_make_request(KVM_REQ_APF_READY, vcpu);
12261 if (!vcpu->arch.apf.pageready_pending)
12262 kvm_vcpu_kick(vcpu);
12265 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
12267 if (!kvm_pv_async_pf_enabled(vcpu))
12270 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
12273 void kvm_arch_start_assignment(struct kvm *kvm)
12275 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
12276 static_call_cond(kvm_x86_start_assignment)(kvm);
12278 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
12280 void kvm_arch_end_assignment(struct kvm *kvm)
12282 atomic_dec(&kvm->arch.assigned_device_count);
12284 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
12286 bool kvm_arch_has_assigned_device(struct kvm *kvm)
12288 return atomic_read(&kvm->arch.assigned_device_count);
12290 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
12292 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12294 atomic_inc(&kvm->arch.noncoherent_dma_count);
12296 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12298 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12300 atomic_dec(&kvm->arch.noncoherent_dma_count);
12302 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12304 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12306 return atomic_read(&kvm->arch.noncoherent_dma_count);
12308 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12310 bool kvm_arch_has_irq_bypass(void)
12315 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12316 struct irq_bypass_producer *prod)
12318 struct kvm_kernel_irqfd *irqfd =
12319 container_of(cons, struct kvm_kernel_irqfd, consumer);
12322 irqfd->producer = prod;
12323 kvm_arch_start_assignment(irqfd->kvm);
12324 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
12325 prod->irq, irqfd->gsi, 1);
12328 kvm_arch_end_assignment(irqfd->kvm);
12333 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12334 struct irq_bypass_producer *prod)
12337 struct kvm_kernel_irqfd *irqfd =
12338 container_of(cons, struct kvm_kernel_irqfd, consumer);
12340 WARN_ON(irqfd->producer != prod);
12341 irqfd->producer = NULL;
12344 * When producer of consumer is unregistered, we change back to
12345 * remapped mode, so we can re-use the current implementation
12346 * when the irq is masked/disabled or the consumer side (KVM
12347 * int this case doesn't want to receive the interrupts.
12349 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12351 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12352 " fails: %d\n", irqfd->consumer.token, ret);
12354 kvm_arch_end_assignment(irqfd->kvm);
12357 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12358 uint32_t guest_irq, bool set)
12360 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12363 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
12364 struct kvm_kernel_irq_routing_entry *new)
12366 if (new->type != KVM_IRQ_ROUTING_MSI)
12369 return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
12372 bool kvm_vector_hashing_enabled(void)
12374 return vector_hashing;
12377 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12379 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12381 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12384 int kvm_spec_ctrl_test_value(u64 value)
12387 * test that setting IA32_SPEC_CTRL to given value
12388 * is allowed by the host processor
12392 unsigned long flags;
12395 local_irq_save(flags);
12397 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12399 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12402 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12404 local_irq_restore(flags);
12408 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12410 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12412 struct x86_exception fault;
12413 u32 access = error_code &
12414 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12416 if (!(error_code & PFERR_PRESENT_MASK) ||
12417 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12419 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12420 * tables probably do not match the TLB. Just proceed
12421 * with the error code that the processor gave.
12423 fault.vector = PF_VECTOR;
12424 fault.error_code_valid = true;
12425 fault.error_code = error_code;
12426 fault.nested_page_fault = false;
12427 fault.address = gva;
12429 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12431 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12434 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12435 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12436 * indicates whether exit to userspace is needed.
12438 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12439 struct x86_exception *e)
12441 if (r == X86EMUL_PROPAGATE_FAULT) {
12442 kvm_inject_emulated_page_fault(vcpu, e);
12447 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12448 * while handling a VMX instruction KVM could've handled the request
12449 * correctly by exiting to userspace and performing I/O but there
12450 * doesn't seem to be a real use-case behind such requests, just return
12451 * KVM_EXIT_INTERNAL_ERROR for now.
12453 kvm_prepare_emulation_failure_exit(vcpu);
12457 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12459 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12462 struct x86_exception e;
12469 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12470 if (r != X86EMUL_CONTINUE)
12471 return kvm_handle_memory_failure(vcpu, r, &e);
12473 if (operand.pcid >> 12 != 0) {
12474 kvm_inject_gp(vcpu, 0);
12478 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12481 case INVPCID_TYPE_INDIV_ADDR:
12482 if ((!pcid_enabled && (operand.pcid != 0)) ||
12483 is_noncanonical_address(operand.gla, vcpu)) {
12484 kvm_inject_gp(vcpu, 0);
12487 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12488 return kvm_skip_emulated_instruction(vcpu);
12490 case INVPCID_TYPE_SINGLE_CTXT:
12491 if (!pcid_enabled && (operand.pcid != 0)) {
12492 kvm_inject_gp(vcpu, 0);
12496 kvm_invalidate_pcid(vcpu, operand.pcid);
12497 return kvm_skip_emulated_instruction(vcpu);
12499 case INVPCID_TYPE_ALL_NON_GLOBAL:
12501 * Currently, KVM doesn't mark global entries in the shadow
12502 * page tables, so a non-global flush just degenerates to a
12503 * global flush. If needed, we could optimize this later by
12504 * keeping track of global entries in shadow page tables.
12508 case INVPCID_TYPE_ALL_INCL_GLOBAL:
12509 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12510 return kvm_skip_emulated_instruction(vcpu);
12513 BUG(); /* We have already checked above that type <= 3 */
12516 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12518 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12520 struct kvm_run *run = vcpu->run;
12521 struct kvm_mmio_fragment *frag;
12524 BUG_ON(!vcpu->mmio_needed);
12526 /* Complete previous fragment */
12527 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12528 len = min(8u, frag->len);
12529 if (!vcpu->mmio_is_write)
12530 memcpy(frag->data, run->mmio.data, len);
12532 if (frag->len <= 8) {
12533 /* Switch to the next fragment. */
12535 vcpu->mmio_cur_fragment++;
12537 /* Go forward to the next mmio piece. */
12543 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12544 vcpu->mmio_needed = 0;
12546 // VMG change, at this point, we're always done
12547 // RIP has already been advanced
12551 // More MMIO is needed
12552 run->mmio.phys_addr = frag->gpa;
12553 run->mmio.len = min(8u, frag->len);
12554 run->mmio.is_write = vcpu->mmio_is_write;
12555 if (run->mmio.is_write)
12556 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12557 run->exit_reason = KVM_EXIT_MMIO;
12559 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12564 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12568 struct kvm_mmio_fragment *frag;
12573 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12574 if (handled == bytes)
12581 /*TODO: Check if need to increment number of frags */
12582 frag = vcpu->mmio_fragments;
12583 vcpu->mmio_nr_fragments = 1;
12588 vcpu->mmio_needed = 1;
12589 vcpu->mmio_cur_fragment = 0;
12591 vcpu->run->mmio.phys_addr = gpa;
12592 vcpu->run->mmio.len = min(8u, frag->len);
12593 vcpu->run->mmio.is_write = 1;
12594 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12595 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12597 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12601 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12603 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12607 struct kvm_mmio_fragment *frag;
12612 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12613 if (handled == bytes)
12620 /*TODO: Check if need to increment number of frags */
12621 frag = vcpu->mmio_fragments;
12622 vcpu->mmio_nr_fragments = 1;
12627 vcpu->mmio_needed = 1;
12628 vcpu->mmio_cur_fragment = 0;
12630 vcpu->run->mmio.phys_addr = gpa;
12631 vcpu->run->mmio.len = min(8u, frag->len);
12632 vcpu->run->mmio.is_write = 0;
12633 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12635 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12639 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12641 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12643 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
12644 vcpu->arch.pio.count * vcpu->arch.pio.size);
12645 vcpu->arch.pio.count = 0;
12650 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12651 unsigned int port, void *data, unsigned int count)
12655 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
12660 vcpu->arch.pio.count = 0;
12665 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12666 unsigned int port, void *data, unsigned int count)
12670 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
12673 vcpu->arch.pio.count = 0;
12675 vcpu->arch.guest_ins_data = data;
12676 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12682 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12683 unsigned int port, void *data, unsigned int count,
12686 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
12687 : kvm_sev_es_outs(vcpu, size, port, data, count);
12689 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12691 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12692 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12693 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12694 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12695 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12696 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12697 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12698 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12699 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12700 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12701 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12702 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12703 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12704 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12705 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12706 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12707 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12708 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12709 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12710 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12711 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12712 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12713 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12714 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12715 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12716 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12717 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);