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>
69 #include <linux/kernel_stat.h>
70 #include <asm/fpu/internal.h> /* Ugh! */
71 #include <asm/pvclock.h>
72 #include <asm/div64.h>
73 #include <asm/irq_remapping.h>
74 #include <asm/mshyperv.h>
75 #include <asm/hypervisor.h>
76 #include <asm/tlbflush.h>
77 #include <asm/intel_pt.h>
78 #include <asm/emulate_prefix.h>
80 #include <clocksource/hyperv_timer.h>
82 #define CREATE_TRACE_POINTS
85 #define MAX_IO_MSRS 256
86 #define KVM_MAX_MCE_BANKS 32
87 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
88 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
90 #define emul_to_vcpu(ctxt) \
91 ((struct kvm_vcpu *)(ctxt)->vcpu)
94 * - enable syscall per default because its emulated by KVM
95 * - enable LME and LMA per default on 64 bit KVM
99 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
101 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
104 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
106 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
107 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
109 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
110 static void process_nmi(struct kvm_vcpu *vcpu);
111 static void process_smi(struct kvm_vcpu *vcpu);
112 static void enter_smm(struct kvm_vcpu *vcpu);
113 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
114 static void store_regs(struct kvm_vcpu *vcpu);
115 static int sync_regs(struct kvm_vcpu *vcpu);
117 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
118 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
120 struct kvm_x86_ops kvm_x86_ops __read_mostly;
121 EXPORT_SYMBOL_GPL(kvm_x86_ops);
123 #define KVM_X86_OP(func) \
124 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
125 *(((struct kvm_x86_ops *)0)->func));
126 #define KVM_X86_OP_NULL KVM_X86_OP
127 #include <asm/kvm-x86-ops.h>
128 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
129 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
130 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
132 static bool __read_mostly ignore_msrs = 0;
133 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
135 bool __read_mostly report_ignored_msrs = true;
136 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
137 EXPORT_SYMBOL_GPL(report_ignored_msrs);
139 unsigned int min_timer_period_us = 200;
140 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
142 static bool __read_mostly kvmclock_periodic_sync = true;
143 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
145 bool __read_mostly kvm_has_tsc_control;
146 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
147 u32 __read_mostly kvm_max_guest_tsc_khz;
148 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
149 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
150 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
151 u64 __read_mostly kvm_max_tsc_scaling_ratio;
152 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
153 u64 __read_mostly kvm_default_tsc_scaling_ratio;
154 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
155 bool __read_mostly kvm_has_bus_lock_exit;
156 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
158 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
159 static u32 __read_mostly tsc_tolerance_ppm = 250;
160 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
163 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
164 * adaptive tuning starting from default advancement of 1000ns. '0' disables
165 * advancement entirely. Any other value is used as-is and disables adaptive
166 * tuning, i.e. allows privileged userspace to set an exact advancement time.
168 static int __read_mostly lapic_timer_advance_ns = -1;
169 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
171 static bool __read_mostly vector_hashing = true;
172 module_param(vector_hashing, bool, S_IRUGO);
174 bool __read_mostly enable_vmware_backdoor = false;
175 module_param(enable_vmware_backdoor, bool, S_IRUGO);
176 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
178 static bool __read_mostly force_emulation_prefix = false;
179 module_param(force_emulation_prefix, bool, S_IRUGO);
181 int __read_mostly pi_inject_timer = -1;
182 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
185 * Restoring the host value for MSRs that are only consumed when running in
186 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
187 * returns to userspace, i.e. the kernel can run with the guest's value.
189 #define KVM_MAX_NR_USER_RETURN_MSRS 16
191 struct kvm_user_return_msrs {
192 struct user_return_notifier urn;
194 struct kvm_user_return_msr_values {
197 } values[KVM_MAX_NR_USER_RETURN_MSRS];
200 u32 __read_mostly kvm_nr_uret_msrs;
201 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
202 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
203 static struct kvm_user_return_msrs __percpu *user_return_msrs;
205 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
206 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
207 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
208 | XFEATURE_MASK_PKRU)
210 u64 __read_mostly host_efer;
211 EXPORT_SYMBOL_GPL(host_efer);
213 bool __read_mostly allow_smaller_maxphyaddr = 0;
214 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
216 bool __read_mostly enable_apicv = true;
217 EXPORT_SYMBOL_GPL(enable_apicv);
219 u64 __read_mostly host_xss;
220 EXPORT_SYMBOL_GPL(host_xss);
221 u64 __read_mostly supported_xss;
222 EXPORT_SYMBOL_GPL(supported_xss);
224 struct kvm_stats_debugfs_item debugfs_entries[] = {
225 VCPU_STAT("pf_fixed", pf_fixed),
226 VCPU_STAT("pf_guest", pf_guest),
227 VCPU_STAT("tlb_flush", tlb_flush),
228 VCPU_STAT("invlpg", invlpg),
229 VCPU_STAT("exits", exits),
230 VCPU_STAT("io_exits", io_exits),
231 VCPU_STAT("mmio_exits", mmio_exits),
232 VCPU_STAT("signal_exits", signal_exits),
233 VCPU_STAT("irq_window", irq_window_exits),
234 VCPU_STAT("nmi_window", nmi_window_exits),
235 VCPU_STAT("halt_exits", halt_exits),
236 VCPU_STAT("halt_successful_poll", halt_successful_poll),
237 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
238 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
239 VCPU_STAT("halt_wakeup", halt_wakeup),
240 VCPU_STAT("hypercalls", hypercalls),
241 VCPU_STAT("request_irq", request_irq_exits),
242 VCPU_STAT("irq_exits", irq_exits),
243 VCPU_STAT("host_state_reload", host_state_reload),
244 VCPU_STAT("fpu_reload", fpu_reload),
245 VCPU_STAT("insn_emulation", insn_emulation),
246 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
247 VCPU_STAT("irq_injections", irq_injections),
248 VCPU_STAT("nmi_injections", nmi_injections),
249 VCPU_STAT("req_event", req_event),
250 VCPU_STAT("l1d_flush", l1d_flush),
251 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
252 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
253 VCPU_STAT("nested_run", nested_run),
254 VCPU_STAT("directed_yield_attempted", directed_yield_attempted),
255 VCPU_STAT("directed_yield_successful", directed_yield_successful),
256 VCPU_STAT("guest_mode", guest_mode),
257 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
258 VM_STAT("mmu_pte_write", mmu_pte_write),
259 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
260 VM_STAT("mmu_flooded", mmu_flooded),
261 VM_STAT("mmu_recycled", mmu_recycled),
262 VM_STAT("mmu_cache_miss", mmu_cache_miss),
263 VM_STAT("mmu_unsync", mmu_unsync),
264 VM_STAT("remote_tlb_flush", remote_tlb_flush),
265 VM_STAT("largepages", lpages, .mode = 0444),
266 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
267 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
271 u64 __read_mostly host_xcr0;
272 u64 __read_mostly supported_xcr0;
273 EXPORT_SYMBOL_GPL(supported_xcr0);
275 static struct kmem_cache *x86_fpu_cache;
277 static struct kmem_cache *x86_emulator_cache;
280 * When called, it means the previous get/set msr reached an invalid msr.
281 * Return true if we want to ignore/silent this failed msr access.
283 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
285 const char *op = write ? "wrmsr" : "rdmsr";
288 if (report_ignored_msrs)
289 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
294 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
300 static struct kmem_cache *kvm_alloc_emulator_cache(void)
302 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
303 unsigned int size = sizeof(struct x86_emulate_ctxt);
305 return kmem_cache_create_usercopy("x86_emulator", size,
306 __alignof__(struct x86_emulate_ctxt),
307 SLAB_ACCOUNT, useroffset,
308 size - useroffset, NULL);
311 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
313 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
316 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
317 vcpu->arch.apf.gfns[i] = ~0;
320 static void kvm_on_user_return(struct user_return_notifier *urn)
323 struct kvm_user_return_msrs *msrs
324 = container_of(urn, struct kvm_user_return_msrs, urn);
325 struct kvm_user_return_msr_values *values;
329 * Disabling irqs at this point since the following code could be
330 * interrupted and executed through kvm_arch_hardware_disable()
332 local_irq_save(flags);
333 if (msrs->registered) {
334 msrs->registered = false;
335 user_return_notifier_unregister(urn);
337 local_irq_restore(flags);
338 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
339 values = &msrs->values[slot];
340 if (values->host != values->curr) {
341 wrmsrl(kvm_uret_msrs_list[slot], values->host);
342 values->curr = values->host;
347 static int kvm_probe_user_return_msr(u32 msr)
353 ret = rdmsrl_safe(msr, &val);
356 ret = wrmsrl_safe(msr, val);
362 int kvm_add_user_return_msr(u32 msr)
364 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
366 if (kvm_probe_user_return_msr(msr))
369 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
370 return kvm_nr_uret_msrs++;
372 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
374 int kvm_find_user_return_msr(u32 msr)
378 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
379 if (kvm_uret_msrs_list[i] == msr)
384 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
386 static void kvm_user_return_msr_cpu_online(void)
388 unsigned int cpu = smp_processor_id();
389 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
393 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
394 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
395 msrs->values[i].host = value;
396 msrs->values[i].curr = value;
400 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
402 unsigned int cpu = smp_processor_id();
403 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
406 value = (value & mask) | (msrs->values[slot].host & ~mask);
407 if (value == msrs->values[slot].curr)
409 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
413 msrs->values[slot].curr = value;
414 if (!msrs->registered) {
415 msrs->urn.on_user_return = kvm_on_user_return;
416 user_return_notifier_register(&msrs->urn);
417 msrs->registered = true;
421 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
423 static void drop_user_return_notifiers(void)
425 unsigned int cpu = smp_processor_id();
426 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
428 if (msrs->registered)
429 kvm_on_user_return(&msrs->urn);
432 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
434 return vcpu->arch.apic_base;
436 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
438 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
440 return kvm_apic_mode(kvm_get_apic_base(vcpu));
442 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
444 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
446 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
447 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
448 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
449 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
451 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
453 if (!msr_info->host_initiated) {
454 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
456 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
460 kvm_lapic_set_base(vcpu, msr_info->data);
461 kvm_recalculate_apic_map(vcpu->kvm);
464 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
466 asmlinkage __visible noinstr void kvm_spurious_fault(void)
468 /* Fault while not rebooting. We want the trace. */
469 BUG_ON(!kvm_rebooting);
471 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
473 #define EXCPT_BENIGN 0
474 #define EXCPT_CONTRIBUTORY 1
477 static int exception_class(int vector)
487 return EXCPT_CONTRIBUTORY;
494 #define EXCPT_FAULT 0
496 #define EXCPT_ABORT 2
497 #define EXCPT_INTERRUPT 3
499 static int exception_type(int vector)
503 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
504 return EXCPT_INTERRUPT;
508 /* #DB is trap, as instruction watchpoints are handled elsewhere */
509 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
512 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
515 /* Reserved exceptions will result in fault */
519 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
521 unsigned nr = vcpu->arch.exception.nr;
522 bool has_payload = vcpu->arch.exception.has_payload;
523 unsigned long payload = vcpu->arch.exception.payload;
531 * "Certain debug exceptions may clear bit 0-3. The
532 * remaining contents of the DR6 register are never
533 * cleared by the processor".
535 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
537 * In order to reflect the #DB exception payload in guest
538 * dr6, three components need to be considered: active low
539 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
541 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
542 * In the target guest dr6:
543 * FIXED_1 bits should always be set.
544 * Active low bits should be cleared if 1-setting in payload.
545 * Active high bits should be set if 1-setting in payload.
547 * Note, the payload is compatible with the pending debug
548 * exceptions/exit qualification under VMX, that active_low bits
549 * are active high in payload.
550 * So they need to be flipped for DR6.
552 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
553 vcpu->arch.dr6 |= payload;
554 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
557 * The #DB payload is defined as compatible with the 'pending
558 * debug exceptions' field under VMX, not DR6. While bit 12 is
559 * defined in the 'pending debug exceptions' field (enabled
560 * breakpoint), it is reserved and must be zero in DR6.
562 vcpu->arch.dr6 &= ~BIT(12);
565 vcpu->arch.cr2 = payload;
569 vcpu->arch.exception.has_payload = false;
570 vcpu->arch.exception.payload = 0;
572 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
574 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
575 unsigned nr, bool has_error, u32 error_code,
576 bool has_payload, unsigned long payload, bool reinject)
581 kvm_make_request(KVM_REQ_EVENT, vcpu);
583 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
587 * On vmentry, vcpu->arch.exception.pending is only
588 * true if an event injection was blocked by
589 * nested_run_pending. In that case, however,
590 * vcpu_enter_guest requests an immediate exit,
591 * and the guest shouldn't proceed far enough to
594 WARN_ON_ONCE(vcpu->arch.exception.pending);
595 vcpu->arch.exception.injected = true;
596 if (WARN_ON_ONCE(has_payload)) {
598 * A reinjected event has already
599 * delivered its payload.
605 vcpu->arch.exception.pending = true;
606 vcpu->arch.exception.injected = false;
608 vcpu->arch.exception.has_error_code = has_error;
609 vcpu->arch.exception.nr = nr;
610 vcpu->arch.exception.error_code = error_code;
611 vcpu->arch.exception.has_payload = has_payload;
612 vcpu->arch.exception.payload = payload;
613 if (!is_guest_mode(vcpu))
614 kvm_deliver_exception_payload(vcpu);
618 /* to check exception */
619 prev_nr = vcpu->arch.exception.nr;
620 if (prev_nr == DF_VECTOR) {
621 /* triple fault -> shutdown */
622 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
625 class1 = exception_class(prev_nr);
626 class2 = exception_class(nr);
627 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
628 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
630 * Generate double fault per SDM Table 5-5. Set
631 * exception.pending = true so that the double fault
632 * can trigger a nested vmexit.
634 vcpu->arch.exception.pending = true;
635 vcpu->arch.exception.injected = false;
636 vcpu->arch.exception.has_error_code = true;
637 vcpu->arch.exception.nr = DF_VECTOR;
638 vcpu->arch.exception.error_code = 0;
639 vcpu->arch.exception.has_payload = false;
640 vcpu->arch.exception.payload = 0;
642 /* replace previous exception with a new one in a hope
643 that instruction re-execution will regenerate lost
648 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
650 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
652 EXPORT_SYMBOL_GPL(kvm_queue_exception);
654 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
656 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
658 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
660 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
661 unsigned long payload)
663 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
665 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
667 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
668 u32 error_code, unsigned long payload)
670 kvm_multiple_exception(vcpu, nr, true, error_code,
671 true, payload, false);
674 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
677 kvm_inject_gp(vcpu, 0);
679 return kvm_skip_emulated_instruction(vcpu);
683 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
685 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
687 ++vcpu->stat.pf_guest;
688 vcpu->arch.exception.nested_apf =
689 is_guest_mode(vcpu) && fault->async_page_fault;
690 if (vcpu->arch.exception.nested_apf) {
691 vcpu->arch.apf.nested_apf_token = fault->address;
692 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
694 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
698 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
700 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
701 struct x86_exception *fault)
703 struct kvm_mmu *fault_mmu;
704 WARN_ON_ONCE(fault->vector != PF_VECTOR);
706 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
710 * Invalidate the TLB entry for the faulting address, if it exists,
711 * else the access will fault indefinitely (and to emulate hardware).
713 if ((fault->error_code & PFERR_PRESENT_MASK) &&
714 !(fault->error_code & PFERR_RSVD_MASK))
715 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
716 fault_mmu->root_hpa);
718 fault_mmu->inject_page_fault(vcpu, fault);
719 return fault->nested_page_fault;
721 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
723 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
725 atomic_inc(&vcpu->arch.nmi_queued);
726 kvm_make_request(KVM_REQ_NMI, vcpu);
728 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
730 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
732 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
734 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
736 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
738 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
740 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
743 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
744 * a #GP and return false.
746 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
748 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
750 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
753 EXPORT_SYMBOL_GPL(kvm_require_cpl);
755 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
757 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
760 kvm_queue_exception(vcpu, UD_VECTOR);
763 EXPORT_SYMBOL_GPL(kvm_require_dr);
766 * This function will be used to read from the physical memory of the currently
767 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
768 * can read from guest physical or from the guest's guest physical memory.
770 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
771 gfn_t ngfn, void *data, int offset, int len,
774 struct x86_exception exception;
778 ngpa = gfn_to_gpa(ngfn);
779 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
780 if (real_gfn == UNMAPPED_GVA)
783 real_gfn = gpa_to_gfn(real_gfn);
785 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
787 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
789 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
791 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
795 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
797 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
799 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
800 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
803 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
805 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
806 offset * sizeof(u64), sizeof(pdpte),
807 PFERR_USER_MASK|PFERR_WRITE_MASK);
812 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
813 if ((pdpte[i] & PT_PRESENT_MASK) &&
814 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
821 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
822 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
823 vcpu->arch.pdptrs_from_userspace = false;
829 EXPORT_SYMBOL_GPL(load_pdptrs);
831 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
833 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
835 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
836 kvm_clear_async_pf_completion_queue(vcpu);
837 kvm_async_pf_hash_reset(vcpu);
840 if ((cr0 ^ old_cr0) & update_bits)
841 kvm_mmu_reset_context(vcpu);
843 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
844 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
845 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
846 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
848 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
850 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
852 unsigned long old_cr0 = kvm_read_cr0(vcpu);
853 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
858 if (cr0 & 0xffffffff00000000UL)
862 cr0 &= ~CR0_RESERVED_BITS;
864 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
867 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
871 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
872 (cr0 & X86_CR0_PG)) {
877 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
882 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
883 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
884 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
887 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
890 static_call(kvm_x86_set_cr0)(vcpu, cr0);
892 kvm_post_set_cr0(vcpu, old_cr0, cr0);
896 EXPORT_SYMBOL_GPL(kvm_set_cr0);
898 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
900 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
902 EXPORT_SYMBOL_GPL(kvm_lmsw);
904 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
906 if (vcpu->arch.guest_state_protected)
909 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
911 if (vcpu->arch.xcr0 != host_xcr0)
912 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
914 if (vcpu->arch.xsaves_enabled &&
915 vcpu->arch.ia32_xss != host_xss)
916 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
919 if (static_cpu_has(X86_FEATURE_PKU) &&
920 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
921 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
922 vcpu->arch.pkru != vcpu->arch.host_pkru)
923 __write_pkru(vcpu->arch.pkru);
925 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
927 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
929 if (vcpu->arch.guest_state_protected)
932 if (static_cpu_has(X86_FEATURE_PKU) &&
933 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
934 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
935 vcpu->arch.pkru = rdpkru();
936 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
937 __write_pkru(vcpu->arch.host_pkru);
940 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
942 if (vcpu->arch.xcr0 != host_xcr0)
943 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
945 if (vcpu->arch.xsaves_enabled &&
946 vcpu->arch.ia32_xss != host_xss)
947 wrmsrl(MSR_IA32_XSS, host_xss);
951 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
953 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
956 u64 old_xcr0 = vcpu->arch.xcr0;
959 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
960 if (index != XCR_XFEATURE_ENABLED_MASK)
962 if (!(xcr0 & XFEATURE_MASK_FP))
964 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
968 * Do not allow the guest to set bits that we do not support
969 * saving. However, xcr0 bit 0 is always set, even if the
970 * emulated CPU does not support XSAVE (see fx_init).
972 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
973 if (xcr0 & ~valid_bits)
976 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
977 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
980 if (xcr0 & XFEATURE_MASK_AVX512) {
981 if (!(xcr0 & XFEATURE_MASK_YMM))
983 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
986 vcpu->arch.xcr0 = xcr0;
988 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
989 kvm_update_cpuid_runtime(vcpu);
993 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
995 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
996 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
997 kvm_inject_gp(vcpu, 0);
1001 return kvm_skip_emulated_instruction(vcpu);
1003 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1005 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1007 if (cr4 & cr4_reserved_bits)
1010 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1013 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1015 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1017 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1019 unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1020 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
1022 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1023 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1024 kvm_mmu_reset_context(vcpu);
1026 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1028 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1030 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1031 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1034 if (!kvm_is_valid_cr4(vcpu, cr4))
1037 if (is_long_mode(vcpu)) {
1038 if (!(cr4 & X86_CR4_PAE))
1040 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1042 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1043 && ((cr4 ^ old_cr4) & pdptr_bits)
1044 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1045 kvm_read_cr3(vcpu)))
1048 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1049 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1052 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1053 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1057 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1059 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1063 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1065 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1067 bool skip_tlb_flush = false;
1068 #ifdef CONFIG_X86_64
1069 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1072 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1073 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1077 /* PDPTRs are always reloaded for PAE paging. */
1078 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu)) {
1079 if (!skip_tlb_flush) {
1080 kvm_mmu_sync_roots(vcpu);
1081 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1087 * Do not condition the GPA check on long mode, this helper is used to
1088 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1089 * the current vCPU mode is accurate.
1091 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1094 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1097 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1098 vcpu->arch.cr3 = cr3;
1099 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1103 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1105 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1107 if (cr8 & CR8_RESERVED_BITS)
1109 if (lapic_in_kernel(vcpu))
1110 kvm_lapic_set_tpr(vcpu, cr8);
1112 vcpu->arch.cr8 = cr8;
1115 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1117 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1119 if (lapic_in_kernel(vcpu))
1120 return kvm_lapic_get_cr8(vcpu);
1122 return vcpu->arch.cr8;
1124 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1126 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1130 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1131 for (i = 0; i < KVM_NR_DB_REGS; i++)
1132 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1133 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1137 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1141 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1142 dr7 = vcpu->arch.guest_debug_dr7;
1144 dr7 = vcpu->arch.dr7;
1145 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1146 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1147 if (dr7 & DR7_BP_EN_MASK)
1148 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1150 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1152 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1154 u64 fixed = DR6_FIXED_1;
1156 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1159 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1160 fixed |= DR6_BUS_LOCK;
1164 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1166 size_t size = ARRAY_SIZE(vcpu->arch.db);
1170 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1171 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1172 vcpu->arch.eff_db[dr] = val;
1176 if (!kvm_dr6_valid(val))
1178 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1182 if (!kvm_dr7_valid(val))
1184 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1185 kvm_update_dr7(vcpu);
1191 EXPORT_SYMBOL_GPL(kvm_set_dr);
1193 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1195 size_t size = ARRAY_SIZE(vcpu->arch.db);
1199 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1203 *val = vcpu->arch.dr6;
1207 *val = vcpu->arch.dr7;
1211 EXPORT_SYMBOL_GPL(kvm_get_dr);
1213 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1215 u32 ecx = kvm_rcx_read(vcpu);
1218 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1219 kvm_inject_gp(vcpu, 0);
1223 kvm_rax_write(vcpu, (u32)data);
1224 kvm_rdx_write(vcpu, data >> 32);
1225 return kvm_skip_emulated_instruction(vcpu);
1227 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1230 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1231 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1233 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1234 * extract the supported MSRs from the related const lists.
1235 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1236 * capabilities of the host cpu. This capabilities test skips MSRs that are
1237 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1238 * may depend on host virtualization features rather than host cpu features.
1241 static const u32 msrs_to_save_all[] = {
1242 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1244 #ifdef CONFIG_X86_64
1245 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1247 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1248 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1250 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1251 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1252 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1253 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1254 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1255 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1256 MSR_IA32_UMWAIT_CONTROL,
1258 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1259 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1260 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1261 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1262 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1263 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1264 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1265 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1266 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1267 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1268 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1269 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1270 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1271 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1272 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1273 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1274 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1275 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1276 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1277 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1278 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1279 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1282 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1283 static unsigned num_msrs_to_save;
1285 static const u32 emulated_msrs_all[] = {
1286 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1287 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1288 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1289 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1290 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1291 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1292 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1294 HV_X64_MSR_VP_INDEX,
1295 HV_X64_MSR_VP_RUNTIME,
1296 HV_X64_MSR_SCONTROL,
1297 HV_X64_MSR_STIMER0_CONFIG,
1298 HV_X64_MSR_VP_ASSIST_PAGE,
1299 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1300 HV_X64_MSR_TSC_EMULATION_STATUS,
1301 HV_X64_MSR_SYNDBG_OPTIONS,
1302 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1303 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1304 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1306 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1307 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1309 MSR_IA32_TSC_ADJUST,
1310 MSR_IA32_TSC_DEADLINE,
1311 MSR_IA32_ARCH_CAPABILITIES,
1312 MSR_IA32_PERF_CAPABILITIES,
1313 MSR_IA32_MISC_ENABLE,
1314 MSR_IA32_MCG_STATUS,
1316 MSR_IA32_MCG_EXT_CTL,
1320 MSR_MISC_FEATURES_ENABLES,
1321 MSR_AMD64_VIRT_SPEC_CTRL,
1326 * The following list leaves out MSRs whose values are determined
1327 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1328 * We always support the "true" VMX control MSRs, even if the host
1329 * processor does not, so I am putting these registers here rather
1330 * than in msrs_to_save_all.
1333 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1334 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1335 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1336 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1338 MSR_IA32_VMX_CR0_FIXED0,
1339 MSR_IA32_VMX_CR4_FIXED0,
1340 MSR_IA32_VMX_VMCS_ENUM,
1341 MSR_IA32_VMX_PROCBASED_CTLS2,
1342 MSR_IA32_VMX_EPT_VPID_CAP,
1343 MSR_IA32_VMX_VMFUNC,
1346 MSR_KVM_POLL_CONTROL,
1349 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1350 static unsigned num_emulated_msrs;
1353 * List of msr numbers which are used to expose MSR-based features that
1354 * can be used by a hypervisor to validate requested CPU features.
1356 static const u32 msr_based_features_all[] = {
1358 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1359 MSR_IA32_VMX_PINBASED_CTLS,
1360 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1361 MSR_IA32_VMX_PROCBASED_CTLS,
1362 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1363 MSR_IA32_VMX_EXIT_CTLS,
1364 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1365 MSR_IA32_VMX_ENTRY_CTLS,
1367 MSR_IA32_VMX_CR0_FIXED0,
1368 MSR_IA32_VMX_CR0_FIXED1,
1369 MSR_IA32_VMX_CR4_FIXED0,
1370 MSR_IA32_VMX_CR4_FIXED1,
1371 MSR_IA32_VMX_VMCS_ENUM,
1372 MSR_IA32_VMX_PROCBASED_CTLS2,
1373 MSR_IA32_VMX_EPT_VPID_CAP,
1374 MSR_IA32_VMX_VMFUNC,
1378 MSR_IA32_ARCH_CAPABILITIES,
1379 MSR_IA32_PERF_CAPABILITIES,
1382 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1383 static unsigned int num_msr_based_features;
1385 static u64 kvm_get_arch_capabilities(void)
1389 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1390 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1393 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1394 * the nested hypervisor runs with NX huge pages. If it is not,
1395 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1396 * L1 guests, so it need not worry about its own (L2) guests.
1398 data |= ARCH_CAP_PSCHANGE_MC_NO;
1401 * If we're doing cache flushes (either "always" or "cond")
1402 * we will do one whenever the guest does a vmlaunch/vmresume.
1403 * If an outer hypervisor is doing the cache flush for us
1404 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1405 * capability to the guest too, and if EPT is disabled we're not
1406 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1407 * require a nested hypervisor to do a flush of its own.
1409 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1410 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1412 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1413 data |= ARCH_CAP_RDCL_NO;
1414 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1415 data |= ARCH_CAP_SSB_NO;
1416 if (!boot_cpu_has_bug(X86_BUG_MDS))
1417 data |= ARCH_CAP_MDS_NO;
1419 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1421 * If RTM=0 because the kernel has disabled TSX, the host might
1422 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1423 * and therefore knows that there cannot be TAA) but keep
1424 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1425 * and we want to allow migrating those guests to tsx=off hosts.
1427 data &= ~ARCH_CAP_TAA_NO;
1428 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1429 data |= ARCH_CAP_TAA_NO;
1432 * Nothing to do here; we emulate TSX_CTRL if present on the
1433 * host so the guest can choose between disabling TSX or
1434 * using VERW to clear CPU buffers.
1441 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1443 switch (msr->index) {
1444 case MSR_IA32_ARCH_CAPABILITIES:
1445 msr->data = kvm_get_arch_capabilities();
1447 case MSR_IA32_UCODE_REV:
1448 rdmsrl_safe(msr->index, &msr->data);
1451 return static_call(kvm_x86_get_msr_feature)(msr);
1456 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1458 struct kvm_msr_entry msr;
1462 r = kvm_get_msr_feature(&msr);
1464 if (r == KVM_MSR_RET_INVALID) {
1465 /* Unconditionally clear the output for simplicity */
1467 if (kvm_msr_ignored_check(index, 0, false))
1479 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1481 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1484 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1487 if (efer & (EFER_LME | EFER_LMA) &&
1488 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1491 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1497 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1499 if (efer & efer_reserved_bits)
1502 return __kvm_valid_efer(vcpu, efer);
1504 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1506 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1508 u64 old_efer = vcpu->arch.efer;
1509 u64 efer = msr_info->data;
1512 if (efer & efer_reserved_bits)
1515 if (!msr_info->host_initiated) {
1516 if (!__kvm_valid_efer(vcpu, efer))
1519 if (is_paging(vcpu) &&
1520 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1525 efer |= vcpu->arch.efer & EFER_LMA;
1527 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1533 /* Update reserved bits */
1534 if ((efer ^ old_efer) & EFER_NX)
1535 kvm_mmu_reset_context(vcpu);
1540 void kvm_enable_efer_bits(u64 mask)
1542 efer_reserved_bits &= ~mask;
1544 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1546 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1548 struct kvm_x86_msr_filter *msr_filter;
1549 struct msr_bitmap_range *ranges;
1550 struct kvm *kvm = vcpu->kvm;
1555 /* x2APIC MSRs do not support filtering. */
1556 if (index >= 0x800 && index <= 0x8ff)
1559 idx = srcu_read_lock(&kvm->srcu);
1561 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1567 allowed = msr_filter->default_allow;
1568 ranges = msr_filter->ranges;
1570 for (i = 0; i < msr_filter->count; i++) {
1571 u32 start = ranges[i].base;
1572 u32 end = start + ranges[i].nmsrs;
1573 u32 flags = ranges[i].flags;
1574 unsigned long *bitmap = ranges[i].bitmap;
1576 if ((index >= start) && (index < end) && (flags & type)) {
1577 allowed = !!test_bit(index - start, bitmap);
1583 srcu_read_unlock(&kvm->srcu, idx);
1587 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1590 * Write @data into the MSR specified by @index. Select MSR specific fault
1591 * checks are bypassed if @host_initiated is %true.
1592 * Returns 0 on success, non-0 otherwise.
1593 * Assumes vcpu_load() was already called.
1595 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1596 bool host_initiated)
1598 struct msr_data msr;
1600 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1601 return KVM_MSR_RET_FILTERED;
1606 case MSR_KERNEL_GS_BASE:
1609 if (is_noncanonical_address(data, vcpu))
1612 case MSR_IA32_SYSENTER_EIP:
1613 case MSR_IA32_SYSENTER_ESP:
1615 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1616 * non-canonical address is written on Intel but not on
1617 * AMD (which ignores the top 32-bits, because it does
1618 * not implement 64-bit SYSENTER).
1620 * 64-bit code should hence be able to write a non-canonical
1621 * value on AMD. Making the address canonical ensures that
1622 * vmentry does not fail on Intel after writing a non-canonical
1623 * value, and that something deterministic happens if the guest
1624 * invokes 64-bit SYSENTER.
1626 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1629 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1632 if (!host_initiated &&
1633 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1634 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1638 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1639 * incomplete and conflicting architectural behavior. Current
1640 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1641 * reserved and always read as zeros. Enforce Intel's reserved
1642 * bits check if and only if the guest CPU is Intel, and clear
1643 * the bits in all other cases. This ensures cross-vendor
1644 * migration will provide consistent behavior for the guest.
1646 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1655 msr.host_initiated = host_initiated;
1657 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1660 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1661 u32 index, u64 data, bool host_initiated)
1663 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1665 if (ret == KVM_MSR_RET_INVALID)
1666 if (kvm_msr_ignored_check(index, data, true))
1673 * Read the MSR specified by @index into @data. Select MSR specific fault
1674 * checks are bypassed if @host_initiated is %true.
1675 * Returns 0 on success, non-0 otherwise.
1676 * Assumes vcpu_load() was already called.
1678 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1679 bool host_initiated)
1681 struct msr_data msr;
1684 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1685 return KVM_MSR_RET_FILTERED;
1689 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1692 if (!host_initiated &&
1693 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1694 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1700 msr.host_initiated = host_initiated;
1702 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1708 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1709 u32 index, u64 *data, bool host_initiated)
1711 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1713 if (ret == KVM_MSR_RET_INVALID) {
1714 /* Unconditionally clear *data for simplicity */
1716 if (kvm_msr_ignored_check(index, 0, false))
1723 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1725 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1727 EXPORT_SYMBOL_GPL(kvm_get_msr);
1729 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1731 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1733 EXPORT_SYMBOL_GPL(kvm_set_msr);
1735 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1737 int err = vcpu->run->msr.error;
1739 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1740 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1743 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1746 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1748 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1751 static u64 kvm_msr_reason(int r)
1754 case KVM_MSR_RET_INVALID:
1755 return KVM_MSR_EXIT_REASON_UNKNOWN;
1756 case KVM_MSR_RET_FILTERED:
1757 return KVM_MSR_EXIT_REASON_FILTER;
1759 return KVM_MSR_EXIT_REASON_INVAL;
1763 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1764 u32 exit_reason, u64 data,
1765 int (*completion)(struct kvm_vcpu *vcpu),
1768 u64 msr_reason = kvm_msr_reason(r);
1770 /* Check if the user wanted to know about this MSR fault */
1771 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1774 vcpu->run->exit_reason = exit_reason;
1775 vcpu->run->msr.error = 0;
1776 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1777 vcpu->run->msr.reason = msr_reason;
1778 vcpu->run->msr.index = index;
1779 vcpu->run->msr.data = data;
1780 vcpu->arch.complete_userspace_io = completion;
1785 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1787 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1788 complete_emulated_rdmsr, r);
1791 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1793 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1794 complete_emulated_wrmsr, r);
1797 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1799 u32 ecx = kvm_rcx_read(vcpu);
1803 r = kvm_get_msr(vcpu, ecx, &data);
1805 /* MSR read failed? See if we should ask user space */
1806 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1807 /* Bounce to user space */
1812 trace_kvm_msr_read(ecx, data);
1814 kvm_rax_write(vcpu, data & -1u);
1815 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1817 trace_kvm_msr_read_ex(ecx);
1820 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1822 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1824 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1826 u32 ecx = kvm_rcx_read(vcpu);
1827 u64 data = kvm_read_edx_eax(vcpu);
1830 r = kvm_set_msr(vcpu, ecx, data);
1832 /* MSR write failed? See if we should ask user space */
1833 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1834 /* Bounce to user space */
1837 /* Signal all other negative errors to userspace */
1842 trace_kvm_msr_write(ecx, data);
1844 trace_kvm_msr_write_ex(ecx, data);
1846 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1848 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1850 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1852 return kvm_skip_emulated_instruction(vcpu);
1854 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1856 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1858 /* Treat an INVD instruction as a NOP and just skip it. */
1859 return kvm_emulate_as_nop(vcpu);
1861 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1863 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1865 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1866 return kvm_emulate_as_nop(vcpu);
1868 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1870 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1872 kvm_queue_exception(vcpu, UD_VECTOR);
1875 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1877 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1879 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1880 return kvm_emulate_as_nop(vcpu);
1882 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1884 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1886 xfer_to_guest_mode_prepare();
1887 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1888 xfer_to_guest_mode_work_pending();
1892 * The fast path for frequent and performance sensitive wrmsr emulation,
1893 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1894 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1895 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1896 * other cases which must be called after interrupts are enabled on the host.
1898 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1900 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1903 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1904 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1905 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1906 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1909 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1910 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1911 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1912 trace_kvm_apic_write(APIC_ICR, (u32)data);
1919 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1921 if (!kvm_can_use_hv_timer(vcpu))
1924 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1928 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1930 u32 msr = kvm_rcx_read(vcpu);
1932 fastpath_t ret = EXIT_FASTPATH_NONE;
1935 case APIC_BASE_MSR + (APIC_ICR >> 4):
1936 data = kvm_read_edx_eax(vcpu);
1937 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1938 kvm_skip_emulated_instruction(vcpu);
1939 ret = EXIT_FASTPATH_EXIT_HANDLED;
1942 case MSR_IA32_TSC_DEADLINE:
1943 data = kvm_read_edx_eax(vcpu);
1944 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1945 kvm_skip_emulated_instruction(vcpu);
1946 ret = EXIT_FASTPATH_REENTER_GUEST;
1953 if (ret != EXIT_FASTPATH_NONE)
1954 trace_kvm_msr_write(msr, data);
1958 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1961 * Adapt set_msr() to msr_io()'s calling convention
1963 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1965 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1968 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1970 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1973 #ifdef CONFIG_X86_64
1974 struct pvclock_clock {
1984 struct pvclock_gtod_data {
1987 struct pvclock_clock clock; /* extract of a clocksource struct */
1988 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1994 static struct pvclock_gtod_data pvclock_gtod_data;
1996 static void update_pvclock_gtod(struct timekeeper *tk)
1998 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2000 write_seqcount_begin(&vdata->seq);
2002 /* copy pvclock gtod data */
2003 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2004 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2005 vdata->clock.mask = tk->tkr_mono.mask;
2006 vdata->clock.mult = tk->tkr_mono.mult;
2007 vdata->clock.shift = tk->tkr_mono.shift;
2008 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2009 vdata->clock.offset = tk->tkr_mono.base;
2011 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2012 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2013 vdata->raw_clock.mask = tk->tkr_raw.mask;
2014 vdata->raw_clock.mult = tk->tkr_raw.mult;
2015 vdata->raw_clock.shift = tk->tkr_raw.shift;
2016 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2017 vdata->raw_clock.offset = tk->tkr_raw.base;
2019 vdata->wall_time_sec = tk->xtime_sec;
2021 vdata->offs_boot = tk->offs_boot;
2023 write_seqcount_end(&vdata->seq);
2026 static s64 get_kvmclock_base_ns(void)
2028 /* Count up from boot time, but with the frequency of the raw clock. */
2029 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2032 static s64 get_kvmclock_base_ns(void)
2034 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2035 return ktime_get_boottime_ns();
2039 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2043 struct pvclock_wall_clock wc;
2050 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2055 ++version; /* first time write, random junk */
2059 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2063 * The guest calculates current wall clock time by adding
2064 * system time (updated by kvm_guest_time_update below) to the
2065 * wall clock specified here. We do the reverse here.
2067 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2069 wc.nsec = do_div(wall_nsec, 1000000000);
2070 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2071 wc.version = version;
2073 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2076 wc_sec_hi = wall_nsec >> 32;
2077 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2078 &wc_sec_hi, sizeof(wc_sec_hi));
2082 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2085 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2086 bool old_msr, bool host_initiated)
2088 struct kvm_arch *ka = &vcpu->kvm->arch;
2090 if (vcpu->vcpu_id == 0 && !host_initiated) {
2091 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2092 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2094 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2097 vcpu->arch.time = system_time;
2098 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2100 /* we verify if the enable bit is set... */
2101 vcpu->arch.pv_time_enabled = false;
2102 if (!(system_time & 1))
2105 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2106 &vcpu->arch.pv_time, system_time & ~1ULL,
2107 sizeof(struct pvclock_vcpu_time_info)))
2108 vcpu->arch.pv_time_enabled = true;
2113 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2115 do_shl32_div32(dividend, divisor);
2119 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2120 s8 *pshift, u32 *pmultiplier)
2128 scaled64 = scaled_hz;
2129 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2134 tps32 = (uint32_t)tps64;
2135 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2136 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2144 *pmultiplier = div_frac(scaled64, tps32);
2147 #ifdef CONFIG_X86_64
2148 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2151 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2152 static unsigned long max_tsc_khz;
2154 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2156 u64 v = (u64)khz * (1000000 + ppm);
2161 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2163 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2167 /* Guest TSC same frequency as host TSC? */
2169 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2173 /* TSC scaling supported? */
2174 if (!kvm_has_tsc_control) {
2175 if (user_tsc_khz > tsc_khz) {
2176 vcpu->arch.tsc_catchup = 1;
2177 vcpu->arch.tsc_always_catchup = 1;
2180 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2185 /* TSC scaling required - calculate ratio */
2186 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2187 user_tsc_khz, tsc_khz);
2189 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2190 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2195 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2199 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2201 u32 thresh_lo, thresh_hi;
2202 int use_scaling = 0;
2204 /* tsc_khz can be zero if TSC calibration fails */
2205 if (user_tsc_khz == 0) {
2206 /* set tsc_scaling_ratio to a safe value */
2207 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2211 /* Compute a scale to convert nanoseconds in TSC cycles */
2212 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2213 &vcpu->arch.virtual_tsc_shift,
2214 &vcpu->arch.virtual_tsc_mult);
2215 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2218 * Compute the variation in TSC rate which is acceptable
2219 * within the range of tolerance and decide if the
2220 * rate being applied is within that bounds of the hardware
2221 * rate. If so, no scaling or compensation need be done.
2223 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2224 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2225 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2226 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2229 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2232 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2234 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2235 vcpu->arch.virtual_tsc_mult,
2236 vcpu->arch.virtual_tsc_shift);
2237 tsc += vcpu->arch.this_tsc_write;
2241 static inline int gtod_is_based_on_tsc(int mode)
2243 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2246 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2248 #ifdef CONFIG_X86_64
2250 struct kvm_arch *ka = &vcpu->kvm->arch;
2251 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2253 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2254 atomic_read(&vcpu->kvm->online_vcpus));
2257 * Once the masterclock is enabled, always perform request in
2258 * order to update it.
2260 * In order to enable masterclock, the host clocksource must be TSC
2261 * and the vcpus need to have matched TSCs. When that happens,
2262 * perform request to enable masterclock.
2264 if (ka->use_master_clock ||
2265 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2266 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2268 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2269 atomic_read(&vcpu->kvm->online_vcpus),
2270 ka->use_master_clock, gtod->clock.vclock_mode);
2275 * Multiply tsc by a fixed point number represented by ratio.
2277 * The most significant 64-N bits (mult) of ratio represent the
2278 * integral part of the fixed point number; the remaining N bits
2279 * (frac) represent the fractional part, ie. ratio represents a fixed
2280 * point number (mult + frac * 2^(-N)).
2282 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2284 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2286 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2289 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2293 if (ratio != kvm_default_tsc_scaling_ratio)
2294 _tsc = __scale_tsc(ratio, tsc);
2298 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2300 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2304 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2306 return target_tsc - tsc;
2309 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2311 return vcpu->arch.l1_tsc_offset +
2312 kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2314 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2316 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2320 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2321 nested_offset = l1_offset;
2323 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2324 kvm_tsc_scaling_ratio_frac_bits);
2326 nested_offset += l2_offset;
2327 return nested_offset;
2329 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2331 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2333 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2334 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2335 kvm_tsc_scaling_ratio_frac_bits);
2337 return l1_multiplier;
2339 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2341 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2343 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2344 vcpu->arch.l1_tsc_offset,
2347 vcpu->arch.l1_tsc_offset = l1_offset;
2350 * If we are here because L1 chose not to trap WRMSR to TSC then
2351 * according to the spec this should set L1's TSC (as opposed to
2352 * setting L1's offset for L2).
2354 if (is_guest_mode(vcpu))
2355 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2357 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2358 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2360 vcpu->arch.tsc_offset = l1_offset;
2362 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2365 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2367 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2369 /* Userspace is changing the multiplier while L2 is active */
2370 if (is_guest_mode(vcpu))
2371 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2373 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2375 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2377 if (kvm_has_tsc_control)
2378 static_call(kvm_x86_write_tsc_multiplier)(
2379 vcpu, vcpu->arch.tsc_scaling_ratio);
2382 static inline bool kvm_check_tsc_unstable(void)
2384 #ifdef CONFIG_X86_64
2386 * TSC is marked unstable when we're running on Hyper-V,
2387 * 'TSC page' clocksource is good.
2389 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2392 return check_tsc_unstable();
2395 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2397 struct kvm *kvm = vcpu->kvm;
2398 u64 offset, ns, elapsed;
2399 unsigned long flags;
2401 bool already_matched;
2402 bool synchronizing = false;
2404 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2405 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2406 ns = get_kvmclock_base_ns();
2407 elapsed = ns - kvm->arch.last_tsc_nsec;
2409 if (vcpu->arch.virtual_tsc_khz) {
2412 * detection of vcpu initialization -- need to sync
2413 * with other vCPUs. This particularly helps to keep
2414 * kvm_clock stable after CPU hotplug
2416 synchronizing = true;
2418 u64 tsc_exp = kvm->arch.last_tsc_write +
2419 nsec_to_cycles(vcpu, elapsed);
2420 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2422 * Special case: TSC write with a small delta (1 second)
2423 * of virtual cycle time against real time is
2424 * interpreted as an attempt to synchronize the CPU.
2426 synchronizing = data < tsc_exp + tsc_hz &&
2427 data + tsc_hz > tsc_exp;
2432 * For a reliable TSC, we can match TSC offsets, and for an unstable
2433 * TSC, we add elapsed time in this computation. We could let the
2434 * compensation code attempt to catch up if we fall behind, but
2435 * it's better to try to match offsets from the beginning.
2437 if (synchronizing &&
2438 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2439 if (!kvm_check_tsc_unstable()) {
2440 offset = kvm->arch.cur_tsc_offset;
2442 u64 delta = nsec_to_cycles(vcpu, elapsed);
2444 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2447 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2450 * We split periods of matched TSC writes into generations.
2451 * For each generation, we track the original measured
2452 * nanosecond time, offset, and write, so if TSCs are in
2453 * sync, we can match exact offset, and if not, we can match
2454 * exact software computation in compute_guest_tsc()
2456 * These values are tracked in kvm->arch.cur_xxx variables.
2458 kvm->arch.cur_tsc_generation++;
2459 kvm->arch.cur_tsc_nsec = ns;
2460 kvm->arch.cur_tsc_write = data;
2461 kvm->arch.cur_tsc_offset = offset;
2466 * We also track th most recent recorded KHZ, write and time to
2467 * allow the matching interval to be extended at each write.
2469 kvm->arch.last_tsc_nsec = ns;
2470 kvm->arch.last_tsc_write = data;
2471 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2473 vcpu->arch.last_guest_tsc = data;
2475 /* Keep track of which generation this VCPU has synchronized to */
2476 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2477 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2478 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2480 kvm_vcpu_write_tsc_offset(vcpu, offset);
2481 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2483 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2485 kvm->arch.nr_vcpus_matched_tsc = 0;
2486 } else if (!already_matched) {
2487 kvm->arch.nr_vcpus_matched_tsc++;
2490 kvm_track_tsc_matching(vcpu);
2491 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2494 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2497 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2498 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2501 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2503 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2504 WARN_ON(adjustment < 0);
2505 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2506 vcpu->arch.l1_tsc_scaling_ratio);
2507 adjust_tsc_offset_guest(vcpu, adjustment);
2510 #ifdef CONFIG_X86_64
2512 static u64 read_tsc(void)
2514 u64 ret = (u64)rdtsc_ordered();
2515 u64 last = pvclock_gtod_data.clock.cycle_last;
2517 if (likely(ret >= last))
2521 * GCC likes to generate cmov here, but this branch is extremely
2522 * predictable (it's just a function of time and the likely is
2523 * very likely) and there's a data dependence, so force GCC
2524 * to generate a branch instead. I don't barrier() because
2525 * we don't actually need a barrier, and if this function
2526 * ever gets inlined it will generate worse code.
2532 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2538 switch (clock->vclock_mode) {
2539 case VDSO_CLOCKMODE_HVCLOCK:
2540 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2542 if (tsc_pg_val != U64_MAX) {
2543 /* TSC page valid */
2544 *mode = VDSO_CLOCKMODE_HVCLOCK;
2545 v = (tsc_pg_val - clock->cycle_last) &
2548 /* TSC page invalid */
2549 *mode = VDSO_CLOCKMODE_NONE;
2552 case VDSO_CLOCKMODE_TSC:
2553 *mode = VDSO_CLOCKMODE_TSC;
2554 *tsc_timestamp = read_tsc();
2555 v = (*tsc_timestamp - clock->cycle_last) &
2559 *mode = VDSO_CLOCKMODE_NONE;
2562 if (*mode == VDSO_CLOCKMODE_NONE)
2563 *tsc_timestamp = v = 0;
2565 return v * clock->mult;
2568 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2570 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2576 seq = read_seqcount_begin(>od->seq);
2577 ns = gtod->raw_clock.base_cycles;
2578 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2579 ns >>= gtod->raw_clock.shift;
2580 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2581 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2587 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2589 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2595 seq = read_seqcount_begin(>od->seq);
2596 ts->tv_sec = gtod->wall_time_sec;
2597 ns = gtod->clock.base_cycles;
2598 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2599 ns >>= gtod->clock.shift;
2600 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2602 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2608 /* returns true if host is using TSC based clocksource */
2609 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2611 /* checked again under seqlock below */
2612 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2615 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2619 /* returns true if host is using TSC based clocksource */
2620 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2623 /* checked again under seqlock below */
2624 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2627 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2633 * Assuming a stable TSC across physical CPUS, and a stable TSC
2634 * across virtual CPUs, the following condition is possible.
2635 * Each numbered line represents an event visible to both
2636 * CPUs at the next numbered event.
2638 * "timespecX" represents host monotonic time. "tscX" represents
2641 * VCPU0 on CPU0 | VCPU1 on CPU1
2643 * 1. read timespec0,tsc0
2644 * 2. | timespec1 = timespec0 + N
2646 * 3. transition to guest | transition to guest
2647 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2648 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2649 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2651 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2654 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2656 * - 0 < N - M => M < N
2658 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2659 * always the case (the difference between two distinct xtime instances
2660 * might be smaller then the difference between corresponding TSC reads,
2661 * when updating guest vcpus pvclock areas).
2663 * To avoid that problem, do not allow visibility of distinct
2664 * system_timestamp/tsc_timestamp values simultaneously: use a master
2665 * copy of host monotonic time values. Update that master copy
2668 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2672 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2674 #ifdef CONFIG_X86_64
2675 struct kvm_arch *ka = &kvm->arch;
2677 bool host_tsc_clocksource, vcpus_matched;
2679 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2680 atomic_read(&kvm->online_vcpus));
2683 * If the host uses TSC clock, then passthrough TSC as stable
2686 host_tsc_clocksource = kvm_get_time_and_clockread(
2687 &ka->master_kernel_ns,
2688 &ka->master_cycle_now);
2690 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2691 && !ka->backwards_tsc_observed
2692 && !ka->boot_vcpu_runs_old_kvmclock;
2694 if (ka->use_master_clock)
2695 atomic_set(&kvm_guest_has_master_clock, 1);
2697 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2698 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2703 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2705 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2708 static void kvm_gen_update_masterclock(struct kvm *kvm)
2710 #ifdef CONFIG_X86_64
2712 struct kvm_vcpu *vcpu;
2713 struct kvm_arch *ka = &kvm->arch;
2714 unsigned long flags;
2716 kvm_hv_invalidate_tsc_page(kvm);
2718 kvm_make_mclock_inprogress_request(kvm);
2720 /* no guest entries from this point */
2721 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2722 pvclock_update_vm_gtod_copy(kvm);
2723 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2725 kvm_for_each_vcpu(i, vcpu, kvm)
2726 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2728 /* guest entries allowed */
2729 kvm_for_each_vcpu(i, vcpu, kvm)
2730 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2734 u64 get_kvmclock_ns(struct kvm *kvm)
2736 struct kvm_arch *ka = &kvm->arch;
2737 struct pvclock_vcpu_time_info hv_clock;
2738 unsigned long flags;
2741 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2742 if (!ka->use_master_clock) {
2743 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2744 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2747 hv_clock.tsc_timestamp = ka->master_cycle_now;
2748 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2749 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2751 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2754 if (__this_cpu_read(cpu_tsc_khz)) {
2755 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2756 &hv_clock.tsc_shift,
2757 &hv_clock.tsc_to_system_mul);
2758 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2760 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2767 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2768 struct gfn_to_hva_cache *cache,
2769 unsigned int offset)
2771 struct kvm_vcpu_arch *vcpu = &v->arch;
2772 struct pvclock_vcpu_time_info guest_hv_clock;
2774 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2775 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2778 /* This VCPU is paused, but it's legal for a guest to read another
2779 * VCPU's kvmclock, so we really have to follow the specification where
2780 * it says that version is odd if data is being modified, and even after
2783 * Version field updates must be kept separate. This is because
2784 * kvm_write_guest_cached might use a "rep movs" instruction, and
2785 * writes within a string instruction are weakly ordered. So there
2786 * are three writes overall.
2788 * As a small optimization, only write the version field in the first
2789 * and third write. The vcpu->pv_time cache is still valid, because the
2790 * version field is the first in the struct.
2792 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2794 if (guest_hv_clock.version & 1)
2795 ++guest_hv_clock.version; /* first time write, random junk */
2797 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2798 kvm_write_guest_offset_cached(v->kvm, cache,
2799 &vcpu->hv_clock, offset,
2800 sizeof(vcpu->hv_clock.version));
2804 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2805 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2807 if (vcpu->pvclock_set_guest_stopped_request) {
2808 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2809 vcpu->pvclock_set_guest_stopped_request = false;
2812 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2814 kvm_write_guest_offset_cached(v->kvm, cache,
2815 &vcpu->hv_clock, offset,
2816 sizeof(vcpu->hv_clock));
2820 vcpu->hv_clock.version++;
2821 kvm_write_guest_offset_cached(v->kvm, cache,
2822 &vcpu->hv_clock, offset,
2823 sizeof(vcpu->hv_clock.version));
2826 static int kvm_guest_time_update(struct kvm_vcpu *v)
2828 unsigned long flags, tgt_tsc_khz;
2829 struct kvm_vcpu_arch *vcpu = &v->arch;
2830 struct kvm_arch *ka = &v->kvm->arch;
2832 u64 tsc_timestamp, host_tsc;
2834 bool use_master_clock;
2840 * If the host uses TSC clock, then passthrough TSC as stable
2843 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2844 use_master_clock = ka->use_master_clock;
2845 if (use_master_clock) {
2846 host_tsc = ka->master_cycle_now;
2847 kernel_ns = ka->master_kernel_ns;
2849 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2851 /* Keep irq disabled to prevent changes to the clock */
2852 local_irq_save(flags);
2853 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2854 if (unlikely(tgt_tsc_khz == 0)) {
2855 local_irq_restore(flags);
2856 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2859 if (!use_master_clock) {
2861 kernel_ns = get_kvmclock_base_ns();
2864 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2867 * We may have to catch up the TSC to match elapsed wall clock
2868 * time for two reasons, even if kvmclock is used.
2869 * 1) CPU could have been running below the maximum TSC rate
2870 * 2) Broken TSC compensation resets the base at each VCPU
2871 * entry to avoid unknown leaps of TSC even when running
2872 * again on the same CPU. This may cause apparent elapsed
2873 * time to disappear, and the guest to stand still or run
2876 if (vcpu->tsc_catchup) {
2877 u64 tsc = compute_guest_tsc(v, kernel_ns);
2878 if (tsc > tsc_timestamp) {
2879 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2880 tsc_timestamp = tsc;
2884 local_irq_restore(flags);
2886 /* With all the info we got, fill in the values */
2888 if (kvm_has_tsc_control)
2889 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
2890 v->arch.l1_tsc_scaling_ratio);
2892 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2893 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2894 &vcpu->hv_clock.tsc_shift,
2895 &vcpu->hv_clock.tsc_to_system_mul);
2896 vcpu->hw_tsc_khz = tgt_tsc_khz;
2899 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2900 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2901 vcpu->last_guest_tsc = tsc_timestamp;
2903 /* If the host uses TSC clocksource, then it is stable */
2905 if (use_master_clock)
2906 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2908 vcpu->hv_clock.flags = pvclock_flags;
2910 if (vcpu->pv_time_enabled)
2911 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2912 if (vcpu->xen.vcpu_info_set)
2913 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2914 offsetof(struct compat_vcpu_info, time));
2915 if (vcpu->xen.vcpu_time_info_set)
2916 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2917 if (v == kvm_get_vcpu(v->kvm, 0))
2918 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2923 * kvmclock updates which are isolated to a given vcpu, such as
2924 * vcpu->cpu migration, should not allow system_timestamp from
2925 * the rest of the vcpus to remain static. Otherwise ntp frequency
2926 * correction applies to one vcpu's system_timestamp but not
2929 * So in those cases, request a kvmclock update for all vcpus.
2930 * We need to rate-limit these requests though, as they can
2931 * considerably slow guests that have a large number of vcpus.
2932 * The time for a remote vcpu to update its kvmclock is bound
2933 * by the delay we use to rate-limit the updates.
2936 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2938 static void kvmclock_update_fn(struct work_struct *work)
2941 struct delayed_work *dwork = to_delayed_work(work);
2942 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2943 kvmclock_update_work);
2944 struct kvm *kvm = container_of(ka, struct kvm, arch);
2945 struct kvm_vcpu *vcpu;
2947 kvm_for_each_vcpu(i, vcpu, kvm) {
2948 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2949 kvm_vcpu_kick(vcpu);
2953 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2955 struct kvm *kvm = v->kvm;
2957 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2958 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2959 KVMCLOCK_UPDATE_DELAY);
2962 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2964 static void kvmclock_sync_fn(struct work_struct *work)
2966 struct delayed_work *dwork = to_delayed_work(work);
2967 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2968 kvmclock_sync_work);
2969 struct kvm *kvm = container_of(ka, struct kvm, arch);
2971 if (!kvmclock_periodic_sync)
2974 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2975 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2976 KVMCLOCK_SYNC_PERIOD);
2980 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2982 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2984 /* McStatusWrEn enabled? */
2985 if (guest_cpuid_is_amd_or_hygon(vcpu))
2986 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2991 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2993 u64 mcg_cap = vcpu->arch.mcg_cap;
2994 unsigned bank_num = mcg_cap & 0xff;
2995 u32 msr = msr_info->index;
2996 u64 data = msr_info->data;
2999 case MSR_IA32_MCG_STATUS:
3000 vcpu->arch.mcg_status = data;
3002 case MSR_IA32_MCG_CTL:
3003 if (!(mcg_cap & MCG_CTL_P) &&
3004 (data || !msr_info->host_initiated))
3006 if (data != 0 && data != ~(u64)0)
3008 vcpu->arch.mcg_ctl = data;
3011 if (msr >= MSR_IA32_MC0_CTL &&
3012 msr < MSR_IA32_MCx_CTL(bank_num)) {
3013 u32 offset = array_index_nospec(
3014 msr - MSR_IA32_MC0_CTL,
3015 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3017 /* only 0 or all 1s can be written to IA32_MCi_CTL
3018 * some Linux kernels though clear bit 10 in bank 4 to
3019 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3020 * this to avoid an uncatched #GP in the guest
3022 if ((offset & 0x3) == 0 &&
3023 data != 0 && (data | (1 << 10)) != ~(u64)0)
3027 if (!msr_info->host_initiated &&
3028 (offset & 0x3) == 1 && data != 0) {
3029 if (!can_set_mci_status(vcpu))
3033 vcpu->arch.mce_banks[offset] = data;
3041 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3043 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3045 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3048 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3050 gpa_t gpa = data & ~0x3f;
3052 /* Bits 4:5 are reserved, Should be zero */
3056 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3057 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3060 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3061 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3064 if (!lapic_in_kernel(vcpu))
3065 return data ? 1 : 0;
3067 vcpu->arch.apf.msr_en_val = data;
3069 if (!kvm_pv_async_pf_enabled(vcpu)) {
3070 kvm_clear_async_pf_completion_queue(vcpu);
3071 kvm_async_pf_hash_reset(vcpu);
3075 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3079 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3080 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3082 kvm_async_pf_wakeup_all(vcpu);
3087 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3089 /* Bits 8-63 are reserved */
3093 if (!lapic_in_kernel(vcpu))
3096 vcpu->arch.apf.msr_int_val = data;
3098 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3103 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3105 vcpu->arch.pv_time_enabled = false;
3106 vcpu->arch.time = 0;
3109 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3111 ++vcpu->stat.tlb_flush;
3112 static_call(kvm_x86_tlb_flush_all)(vcpu);
3115 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3117 ++vcpu->stat.tlb_flush;
3121 * A TLB flush on behalf of the guest is equivalent to
3122 * INVPCID(all), toggling CR4.PGE, etc., which requires
3123 * a forced sync of the shadow page tables. Unload the
3124 * entire MMU here and the subsequent load will sync the
3125 * shadow page tables, and also flush the TLB.
3127 kvm_mmu_unload(vcpu);
3131 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3134 static void record_steal_time(struct kvm_vcpu *vcpu)
3136 struct kvm_host_map map;
3137 struct kvm_steal_time *st;
3139 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3140 kvm_xen_runstate_set_running(vcpu);
3144 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3147 /* -EAGAIN is returned in atomic context so we can just return. */
3148 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3149 &map, &vcpu->arch.st.cache, false))
3153 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3156 * Doing a TLB flush here, on the guest's behalf, can avoid
3159 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3160 u8 st_preempted = xchg(&st->preempted, 0);
3162 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3163 st_preempted & KVM_VCPU_FLUSH_TLB);
3164 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3165 kvm_vcpu_flush_tlb_guest(vcpu);
3170 vcpu->arch.st.preempted = 0;
3172 if (st->version & 1)
3173 st->version += 1; /* first time write, random junk */
3179 st->steal += current->sched_info.run_delay -
3180 vcpu->arch.st.last_steal;
3181 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3187 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3190 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3193 u32 msr = msr_info->index;
3194 u64 data = msr_info->data;
3196 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3197 return kvm_xen_write_hypercall_page(vcpu, data);
3200 case MSR_AMD64_NB_CFG:
3201 case MSR_IA32_UCODE_WRITE:
3202 case MSR_VM_HSAVE_PA:
3203 case MSR_AMD64_PATCH_LOADER:
3204 case MSR_AMD64_BU_CFG2:
3205 case MSR_AMD64_DC_CFG:
3206 case MSR_F15H_EX_CFG:
3209 case MSR_IA32_UCODE_REV:
3210 if (msr_info->host_initiated)
3211 vcpu->arch.microcode_version = data;
3213 case MSR_IA32_ARCH_CAPABILITIES:
3214 if (!msr_info->host_initiated)
3216 vcpu->arch.arch_capabilities = data;
3218 case MSR_IA32_PERF_CAPABILITIES: {
3219 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3221 if (!msr_info->host_initiated)
3223 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3225 if (data & ~msr_ent.data)
3228 vcpu->arch.perf_capabilities = data;
3233 return set_efer(vcpu, msr_info);
3235 data &= ~(u64)0x40; /* ignore flush filter disable */
3236 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3237 data &= ~(u64)0x8; /* ignore TLB cache disable */
3239 /* Handle McStatusWrEn */
3240 if (data == BIT_ULL(18)) {
3241 vcpu->arch.msr_hwcr = data;
3242 } else if (data != 0) {
3243 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3248 case MSR_FAM10H_MMIO_CONF_BASE:
3250 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3255 case 0x200 ... 0x2ff:
3256 return kvm_mtrr_set_msr(vcpu, msr, data);
3257 case MSR_IA32_APICBASE:
3258 return kvm_set_apic_base(vcpu, msr_info);
3259 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3260 return kvm_x2apic_msr_write(vcpu, msr, data);
3261 case MSR_IA32_TSC_DEADLINE:
3262 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3264 case MSR_IA32_TSC_ADJUST:
3265 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3266 if (!msr_info->host_initiated) {
3267 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3268 adjust_tsc_offset_guest(vcpu, adj);
3270 vcpu->arch.ia32_tsc_adjust_msr = data;
3273 case MSR_IA32_MISC_ENABLE:
3274 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3275 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3276 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3278 vcpu->arch.ia32_misc_enable_msr = data;
3279 kvm_update_cpuid_runtime(vcpu);
3281 vcpu->arch.ia32_misc_enable_msr = data;
3284 case MSR_IA32_SMBASE:
3285 if (!msr_info->host_initiated)
3287 vcpu->arch.smbase = data;
3289 case MSR_IA32_POWER_CTL:
3290 vcpu->arch.msr_ia32_power_ctl = data;
3293 if (msr_info->host_initiated) {
3294 kvm_synchronize_tsc(vcpu, data);
3296 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3297 adjust_tsc_offset_guest(vcpu, adj);
3298 vcpu->arch.ia32_tsc_adjust_msr += adj;
3302 if (!msr_info->host_initiated &&
3303 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3306 * KVM supports exposing PT to the guest, but does not support
3307 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3308 * XSAVES/XRSTORS to save/restore PT MSRs.
3310 if (data & ~supported_xss)
3312 vcpu->arch.ia32_xss = data;
3315 if (!msr_info->host_initiated)
3317 vcpu->arch.smi_count = data;
3319 case MSR_KVM_WALL_CLOCK_NEW:
3320 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3323 vcpu->kvm->arch.wall_clock = data;
3324 kvm_write_wall_clock(vcpu->kvm, data, 0);
3326 case MSR_KVM_WALL_CLOCK:
3327 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3330 vcpu->kvm->arch.wall_clock = data;
3331 kvm_write_wall_clock(vcpu->kvm, data, 0);
3333 case MSR_KVM_SYSTEM_TIME_NEW:
3334 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3337 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3339 case MSR_KVM_SYSTEM_TIME:
3340 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3343 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3345 case MSR_KVM_ASYNC_PF_EN:
3346 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3349 if (kvm_pv_enable_async_pf(vcpu, data))
3352 case MSR_KVM_ASYNC_PF_INT:
3353 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3356 if (kvm_pv_enable_async_pf_int(vcpu, data))
3359 case MSR_KVM_ASYNC_PF_ACK:
3360 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3363 vcpu->arch.apf.pageready_pending = false;
3364 kvm_check_async_pf_completion(vcpu);
3367 case MSR_KVM_STEAL_TIME:
3368 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3371 if (unlikely(!sched_info_on()))
3374 if (data & KVM_STEAL_RESERVED_MASK)
3377 vcpu->arch.st.msr_val = data;
3379 if (!(data & KVM_MSR_ENABLED))
3382 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3385 case MSR_KVM_PV_EOI_EN:
3386 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3389 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3393 case MSR_KVM_POLL_CONTROL:
3394 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3397 /* only enable bit supported */
3398 if (data & (-1ULL << 1))
3401 vcpu->arch.msr_kvm_poll_control = data;
3404 case MSR_IA32_MCG_CTL:
3405 case MSR_IA32_MCG_STATUS:
3406 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3407 return set_msr_mce(vcpu, msr_info);
3409 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3410 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3413 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3414 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3415 if (kvm_pmu_is_valid_msr(vcpu, msr))
3416 return kvm_pmu_set_msr(vcpu, msr_info);
3418 if (pr || data != 0)
3419 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3420 "0x%x data 0x%llx\n", msr, data);
3422 case MSR_K7_CLK_CTL:
3424 * Ignore all writes to this no longer documented MSR.
3425 * Writes are only relevant for old K7 processors,
3426 * all pre-dating SVM, but a recommended workaround from
3427 * AMD for these chips. It is possible to specify the
3428 * affected processor models on the command line, hence
3429 * the need to ignore the workaround.
3432 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3433 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3434 case HV_X64_MSR_SYNDBG_OPTIONS:
3435 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3436 case HV_X64_MSR_CRASH_CTL:
3437 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3438 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3439 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3440 case HV_X64_MSR_TSC_EMULATION_STATUS:
3441 return kvm_hv_set_msr_common(vcpu, msr, data,
3442 msr_info->host_initiated);
3443 case MSR_IA32_BBL_CR_CTL3:
3444 /* Drop writes to this legacy MSR -- see rdmsr
3445 * counterpart for further detail.
3447 if (report_ignored_msrs)
3448 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3451 case MSR_AMD64_OSVW_ID_LENGTH:
3452 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3454 vcpu->arch.osvw.length = data;
3456 case MSR_AMD64_OSVW_STATUS:
3457 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3459 vcpu->arch.osvw.status = data;
3461 case MSR_PLATFORM_INFO:
3462 if (!msr_info->host_initiated ||
3463 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3464 cpuid_fault_enabled(vcpu)))
3466 vcpu->arch.msr_platform_info = data;
3468 case MSR_MISC_FEATURES_ENABLES:
3469 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3470 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3471 !supports_cpuid_fault(vcpu)))
3473 vcpu->arch.msr_misc_features_enables = data;
3476 if (kvm_pmu_is_valid_msr(vcpu, msr))
3477 return kvm_pmu_set_msr(vcpu, msr_info);
3478 return KVM_MSR_RET_INVALID;
3482 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3484 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3487 u64 mcg_cap = vcpu->arch.mcg_cap;
3488 unsigned bank_num = mcg_cap & 0xff;
3491 case MSR_IA32_P5_MC_ADDR:
3492 case MSR_IA32_P5_MC_TYPE:
3495 case MSR_IA32_MCG_CAP:
3496 data = vcpu->arch.mcg_cap;
3498 case MSR_IA32_MCG_CTL:
3499 if (!(mcg_cap & MCG_CTL_P) && !host)
3501 data = vcpu->arch.mcg_ctl;
3503 case MSR_IA32_MCG_STATUS:
3504 data = vcpu->arch.mcg_status;
3507 if (msr >= MSR_IA32_MC0_CTL &&
3508 msr < MSR_IA32_MCx_CTL(bank_num)) {
3509 u32 offset = array_index_nospec(
3510 msr - MSR_IA32_MC0_CTL,
3511 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3513 data = vcpu->arch.mce_banks[offset];
3522 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3524 switch (msr_info->index) {
3525 case MSR_IA32_PLATFORM_ID:
3526 case MSR_IA32_EBL_CR_POWERON:
3527 case MSR_IA32_LASTBRANCHFROMIP:
3528 case MSR_IA32_LASTBRANCHTOIP:
3529 case MSR_IA32_LASTINTFROMIP:
3530 case MSR_IA32_LASTINTTOIP:
3532 case MSR_K8_TSEG_ADDR:
3533 case MSR_K8_TSEG_MASK:
3534 case MSR_VM_HSAVE_PA:
3535 case MSR_K8_INT_PENDING_MSG:
3536 case MSR_AMD64_NB_CFG:
3537 case MSR_FAM10H_MMIO_CONF_BASE:
3538 case MSR_AMD64_BU_CFG2:
3539 case MSR_IA32_PERF_CTL:
3540 case MSR_AMD64_DC_CFG:
3541 case MSR_F15H_EX_CFG:
3543 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3544 * limit) MSRs. Just return 0, as we do not want to expose the host
3545 * data here. Do not conditionalize this on CPUID, as KVM does not do
3546 * so for existing CPU-specific MSRs.
3548 case MSR_RAPL_POWER_UNIT:
3549 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3550 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3551 case MSR_PKG_ENERGY_STATUS: /* Total package */
3552 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3555 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3556 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3557 return kvm_pmu_get_msr(vcpu, msr_info);
3558 if (!msr_info->host_initiated)
3562 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3563 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3564 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3565 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3566 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3567 return kvm_pmu_get_msr(vcpu, msr_info);
3570 case MSR_IA32_UCODE_REV:
3571 msr_info->data = vcpu->arch.microcode_version;
3573 case MSR_IA32_ARCH_CAPABILITIES:
3574 if (!msr_info->host_initiated &&
3575 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3577 msr_info->data = vcpu->arch.arch_capabilities;
3579 case MSR_IA32_PERF_CAPABILITIES:
3580 if (!msr_info->host_initiated &&
3581 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3583 msr_info->data = vcpu->arch.perf_capabilities;
3585 case MSR_IA32_POWER_CTL:
3586 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3588 case MSR_IA32_TSC: {
3590 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3591 * even when not intercepted. AMD manual doesn't explicitly
3592 * state this but appears to behave the same.
3594 * On userspace reads and writes, however, we unconditionally
3595 * return L1's TSC value to ensure backwards-compatible
3596 * behavior for migration.
3600 if (msr_info->host_initiated) {
3601 offset = vcpu->arch.l1_tsc_offset;
3602 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3604 offset = vcpu->arch.tsc_offset;
3605 ratio = vcpu->arch.tsc_scaling_ratio;
3608 msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3612 case 0x200 ... 0x2ff:
3613 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3614 case 0xcd: /* fsb frequency */
3618 * MSR_EBC_FREQUENCY_ID
3619 * Conservative value valid for even the basic CPU models.
3620 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3621 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3622 * and 266MHz for model 3, or 4. Set Core Clock
3623 * Frequency to System Bus Frequency Ratio to 1 (bits
3624 * 31:24) even though these are only valid for CPU
3625 * models > 2, however guests may end up dividing or
3626 * multiplying by zero otherwise.
3628 case MSR_EBC_FREQUENCY_ID:
3629 msr_info->data = 1 << 24;
3631 case MSR_IA32_APICBASE:
3632 msr_info->data = kvm_get_apic_base(vcpu);
3634 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3635 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3636 case MSR_IA32_TSC_DEADLINE:
3637 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3639 case MSR_IA32_TSC_ADJUST:
3640 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3642 case MSR_IA32_MISC_ENABLE:
3643 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3645 case MSR_IA32_SMBASE:
3646 if (!msr_info->host_initiated)
3648 msr_info->data = vcpu->arch.smbase;
3651 msr_info->data = vcpu->arch.smi_count;
3653 case MSR_IA32_PERF_STATUS:
3654 /* TSC increment by tick */
3655 msr_info->data = 1000ULL;
3656 /* CPU multiplier */
3657 msr_info->data |= (((uint64_t)4ULL) << 40);
3660 msr_info->data = vcpu->arch.efer;
3662 case MSR_KVM_WALL_CLOCK:
3663 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3666 msr_info->data = vcpu->kvm->arch.wall_clock;
3668 case MSR_KVM_WALL_CLOCK_NEW:
3669 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3672 msr_info->data = vcpu->kvm->arch.wall_clock;
3674 case MSR_KVM_SYSTEM_TIME:
3675 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3678 msr_info->data = vcpu->arch.time;
3680 case MSR_KVM_SYSTEM_TIME_NEW:
3681 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3684 msr_info->data = vcpu->arch.time;
3686 case MSR_KVM_ASYNC_PF_EN:
3687 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3690 msr_info->data = vcpu->arch.apf.msr_en_val;
3692 case MSR_KVM_ASYNC_PF_INT:
3693 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3696 msr_info->data = vcpu->arch.apf.msr_int_val;
3698 case MSR_KVM_ASYNC_PF_ACK:
3699 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3704 case MSR_KVM_STEAL_TIME:
3705 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3708 msr_info->data = vcpu->arch.st.msr_val;
3710 case MSR_KVM_PV_EOI_EN:
3711 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3714 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3716 case MSR_KVM_POLL_CONTROL:
3717 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3720 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3722 case MSR_IA32_P5_MC_ADDR:
3723 case MSR_IA32_P5_MC_TYPE:
3724 case MSR_IA32_MCG_CAP:
3725 case MSR_IA32_MCG_CTL:
3726 case MSR_IA32_MCG_STATUS:
3727 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3728 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3729 msr_info->host_initiated);
3731 if (!msr_info->host_initiated &&
3732 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3734 msr_info->data = vcpu->arch.ia32_xss;
3736 case MSR_K7_CLK_CTL:
3738 * Provide expected ramp-up count for K7. All other
3739 * are set to zero, indicating minimum divisors for
3742 * This prevents guest kernels on AMD host with CPU
3743 * type 6, model 8 and higher from exploding due to
3744 * the rdmsr failing.
3746 msr_info->data = 0x20000000;
3748 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3749 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3750 case HV_X64_MSR_SYNDBG_OPTIONS:
3751 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3752 case HV_X64_MSR_CRASH_CTL:
3753 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3754 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3755 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3756 case HV_X64_MSR_TSC_EMULATION_STATUS:
3757 return kvm_hv_get_msr_common(vcpu,
3758 msr_info->index, &msr_info->data,
3759 msr_info->host_initiated);
3760 case MSR_IA32_BBL_CR_CTL3:
3761 /* This legacy MSR exists but isn't fully documented in current
3762 * silicon. It is however accessed by winxp in very narrow
3763 * scenarios where it sets bit #19, itself documented as
3764 * a "reserved" bit. Best effort attempt to source coherent
3765 * read data here should the balance of the register be
3766 * interpreted by the guest:
3768 * L2 cache control register 3: 64GB range, 256KB size,
3769 * enabled, latency 0x1, configured
3771 msr_info->data = 0xbe702111;
3773 case MSR_AMD64_OSVW_ID_LENGTH:
3774 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3776 msr_info->data = vcpu->arch.osvw.length;
3778 case MSR_AMD64_OSVW_STATUS:
3779 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3781 msr_info->data = vcpu->arch.osvw.status;
3783 case MSR_PLATFORM_INFO:
3784 if (!msr_info->host_initiated &&
3785 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3787 msr_info->data = vcpu->arch.msr_platform_info;
3789 case MSR_MISC_FEATURES_ENABLES:
3790 msr_info->data = vcpu->arch.msr_misc_features_enables;
3793 msr_info->data = vcpu->arch.msr_hwcr;
3796 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3797 return kvm_pmu_get_msr(vcpu, msr_info);
3798 return KVM_MSR_RET_INVALID;
3802 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3805 * Read or write a bunch of msrs. All parameters are kernel addresses.
3807 * @return number of msrs set successfully.
3809 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3810 struct kvm_msr_entry *entries,
3811 int (*do_msr)(struct kvm_vcpu *vcpu,
3812 unsigned index, u64 *data))
3816 for (i = 0; i < msrs->nmsrs; ++i)
3817 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3824 * Read or write a bunch of msrs. Parameters are user addresses.
3826 * @return number of msrs set successfully.
3828 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3829 int (*do_msr)(struct kvm_vcpu *vcpu,
3830 unsigned index, u64 *data),
3833 struct kvm_msrs msrs;
3834 struct kvm_msr_entry *entries;
3839 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3843 if (msrs.nmsrs >= MAX_IO_MSRS)
3846 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3847 entries = memdup_user(user_msrs->entries, size);
3848 if (IS_ERR(entries)) {
3849 r = PTR_ERR(entries);
3853 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3858 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3869 static inline bool kvm_can_mwait_in_guest(void)
3871 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3872 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3873 boot_cpu_has(X86_FEATURE_ARAT);
3876 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3877 struct kvm_cpuid2 __user *cpuid_arg)
3879 struct kvm_cpuid2 cpuid;
3883 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3886 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3891 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3897 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3902 case KVM_CAP_IRQCHIP:
3904 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3905 case KVM_CAP_SET_TSS_ADDR:
3906 case KVM_CAP_EXT_CPUID:
3907 case KVM_CAP_EXT_EMUL_CPUID:
3908 case KVM_CAP_CLOCKSOURCE:
3910 case KVM_CAP_NOP_IO_DELAY:
3911 case KVM_CAP_MP_STATE:
3912 case KVM_CAP_SYNC_MMU:
3913 case KVM_CAP_USER_NMI:
3914 case KVM_CAP_REINJECT_CONTROL:
3915 case KVM_CAP_IRQ_INJECT_STATUS:
3916 case KVM_CAP_IOEVENTFD:
3917 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3919 case KVM_CAP_PIT_STATE2:
3920 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3921 case KVM_CAP_VCPU_EVENTS:
3922 case KVM_CAP_HYPERV:
3923 case KVM_CAP_HYPERV_VAPIC:
3924 case KVM_CAP_HYPERV_SPIN:
3925 case KVM_CAP_HYPERV_SYNIC:
3926 case KVM_CAP_HYPERV_SYNIC2:
3927 case KVM_CAP_HYPERV_VP_INDEX:
3928 case KVM_CAP_HYPERV_EVENTFD:
3929 case KVM_CAP_HYPERV_TLBFLUSH:
3930 case KVM_CAP_HYPERV_SEND_IPI:
3931 case KVM_CAP_HYPERV_CPUID:
3932 case KVM_CAP_HYPERV_ENFORCE_CPUID:
3933 case KVM_CAP_SYS_HYPERV_CPUID:
3934 case KVM_CAP_PCI_SEGMENT:
3935 case KVM_CAP_DEBUGREGS:
3936 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3938 case KVM_CAP_ASYNC_PF:
3939 case KVM_CAP_ASYNC_PF_INT:
3940 case KVM_CAP_GET_TSC_KHZ:
3941 case KVM_CAP_KVMCLOCK_CTRL:
3942 case KVM_CAP_READONLY_MEM:
3943 case KVM_CAP_HYPERV_TIME:
3944 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3945 case KVM_CAP_TSC_DEADLINE_TIMER:
3946 case KVM_CAP_DISABLE_QUIRKS:
3947 case KVM_CAP_SET_BOOT_CPU_ID:
3948 case KVM_CAP_SPLIT_IRQCHIP:
3949 case KVM_CAP_IMMEDIATE_EXIT:
3950 case KVM_CAP_PMU_EVENT_FILTER:
3951 case KVM_CAP_GET_MSR_FEATURES:
3952 case KVM_CAP_MSR_PLATFORM_INFO:
3953 case KVM_CAP_EXCEPTION_PAYLOAD:
3954 case KVM_CAP_SET_GUEST_DEBUG:
3955 case KVM_CAP_LAST_CPU:
3956 case KVM_CAP_X86_USER_SPACE_MSR:
3957 case KVM_CAP_X86_MSR_FILTER:
3958 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3959 #ifdef CONFIG_X86_SGX_KVM
3960 case KVM_CAP_SGX_ATTRIBUTE:
3962 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
3963 case KVM_CAP_SREGS2:
3966 case KVM_CAP_SET_GUEST_DEBUG2:
3967 return KVM_GUESTDBG_VALID_MASK;
3968 #ifdef CONFIG_KVM_XEN
3969 case KVM_CAP_XEN_HVM:
3970 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
3971 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
3972 KVM_XEN_HVM_CONFIG_SHARED_INFO;
3973 if (sched_info_on())
3974 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
3977 case KVM_CAP_SYNC_REGS:
3978 r = KVM_SYNC_X86_VALID_FIELDS;
3980 case KVM_CAP_ADJUST_CLOCK:
3981 r = KVM_CLOCK_TSC_STABLE;
3983 case KVM_CAP_X86_DISABLE_EXITS:
3984 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3985 KVM_X86_DISABLE_EXITS_CSTATE;
3986 if(kvm_can_mwait_in_guest())
3987 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3989 case KVM_CAP_X86_SMM:
3990 /* SMBASE is usually relocated above 1M on modern chipsets,
3991 * and SMM handlers might indeed rely on 4G segment limits,
3992 * so do not report SMM to be available if real mode is
3993 * emulated via vm86 mode. Still, do not go to great lengths
3994 * to avoid userspace's usage of the feature, because it is a
3995 * fringe case that is not enabled except via specific settings
3996 * of the module parameters.
3998 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4001 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4003 case KVM_CAP_NR_VCPUS:
4004 r = KVM_SOFT_MAX_VCPUS;
4006 case KVM_CAP_MAX_VCPUS:
4009 case KVM_CAP_MAX_VCPU_ID:
4010 r = KVM_MAX_VCPU_ID;
4012 case KVM_CAP_PV_MMU: /* obsolete */
4016 r = KVM_MAX_MCE_BANKS;
4019 r = boot_cpu_has(X86_FEATURE_XSAVE);
4021 case KVM_CAP_TSC_CONTROL:
4022 r = kvm_has_tsc_control;
4024 case KVM_CAP_X2APIC_API:
4025 r = KVM_X2APIC_API_VALID_FLAGS;
4027 case KVM_CAP_NESTED_STATE:
4028 r = kvm_x86_ops.nested_ops->get_state ?
4029 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4031 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4032 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4034 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4035 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4037 case KVM_CAP_SMALLER_MAXPHYADDR:
4038 r = (int) allow_smaller_maxphyaddr;
4040 case KVM_CAP_STEAL_TIME:
4041 r = sched_info_on();
4043 case KVM_CAP_X86_BUS_LOCK_EXIT:
4044 if (kvm_has_bus_lock_exit)
4045 r = KVM_BUS_LOCK_DETECTION_OFF |
4046 KVM_BUS_LOCK_DETECTION_EXIT;
4057 long kvm_arch_dev_ioctl(struct file *filp,
4058 unsigned int ioctl, unsigned long arg)
4060 void __user *argp = (void __user *)arg;
4064 case KVM_GET_MSR_INDEX_LIST: {
4065 struct kvm_msr_list __user *user_msr_list = argp;
4066 struct kvm_msr_list msr_list;
4070 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4073 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4074 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4077 if (n < msr_list.nmsrs)
4080 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4081 num_msrs_to_save * sizeof(u32)))
4083 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4085 num_emulated_msrs * sizeof(u32)))
4090 case KVM_GET_SUPPORTED_CPUID:
4091 case KVM_GET_EMULATED_CPUID: {
4092 struct kvm_cpuid2 __user *cpuid_arg = argp;
4093 struct kvm_cpuid2 cpuid;
4096 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4099 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4105 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4110 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4112 if (copy_to_user(argp, &kvm_mce_cap_supported,
4113 sizeof(kvm_mce_cap_supported)))
4117 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4118 struct kvm_msr_list __user *user_msr_list = argp;
4119 struct kvm_msr_list msr_list;
4123 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4126 msr_list.nmsrs = num_msr_based_features;
4127 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4130 if (n < msr_list.nmsrs)
4133 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4134 num_msr_based_features * sizeof(u32)))
4140 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4142 case KVM_GET_SUPPORTED_HV_CPUID:
4143 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4153 static void wbinvd_ipi(void *garbage)
4158 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4160 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4163 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4165 /* Address WBINVD may be executed by guest */
4166 if (need_emulate_wbinvd(vcpu)) {
4167 if (static_call(kvm_x86_has_wbinvd_exit)())
4168 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4169 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4170 smp_call_function_single(vcpu->cpu,
4171 wbinvd_ipi, NULL, 1);
4174 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4176 /* Save host pkru register if supported */
4177 vcpu->arch.host_pkru = read_pkru();
4179 /* Apply any externally detected TSC adjustments (due to suspend) */
4180 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4181 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4182 vcpu->arch.tsc_offset_adjustment = 0;
4183 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4186 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4187 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4188 rdtsc() - vcpu->arch.last_host_tsc;
4190 mark_tsc_unstable("KVM discovered backwards TSC");
4192 if (kvm_check_tsc_unstable()) {
4193 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4194 vcpu->arch.last_guest_tsc);
4195 kvm_vcpu_write_tsc_offset(vcpu, offset);
4196 vcpu->arch.tsc_catchup = 1;
4199 if (kvm_lapic_hv_timer_in_use(vcpu))
4200 kvm_lapic_restart_hv_timer(vcpu);
4203 * On a host with synchronized TSC, there is no need to update
4204 * kvmclock on vcpu->cpu migration
4206 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4207 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4208 if (vcpu->cpu != cpu)
4209 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4213 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4216 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4218 struct kvm_host_map map;
4219 struct kvm_steal_time *st;
4221 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4224 if (vcpu->arch.st.preempted)
4227 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4228 &vcpu->arch.st.cache, true))
4232 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4234 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4236 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4239 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4243 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4244 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4247 * Take the srcu lock as memslots will be accessed to check the gfn
4248 * cache generation against the memslots generation.
4250 idx = srcu_read_lock(&vcpu->kvm->srcu);
4251 if (kvm_xen_msr_enabled(vcpu->kvm))
4252 kvm_xen_runstate_set_preempted(vcpu);
4254 kvm_steal_time_set_preempted(vcpu);
4255 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4257 static_call(kvm_x86_vcpu_put)(vcpu);
4258 vcpu->arch.last_host_tsc = rdtsc();
4260 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4261 * on every vmexit, but if not, we might have a stale dr6 from the
4262 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4267 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4268 struct kvm_lapic_state *s)
4270 if (vcpu->arch.apicv_active)
4271 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4273 return kvm_apic_get_state(vcpu, s);
4276 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4277 struct kvm_lapic_state *s)
4281 r = kvm_apic_set_state(vcpu, s);
4284 update_cr8_intercept(vcpu);
4289 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4292 * We can accept userspace's request for interrupt injection
4293 * as long as we have a place to store the interrupt number.
4294 * The actual injection will happen when the CPU is able to
4295 * deliver the interrupt.
4297 if (kvm_cpu_has_extint(vcpu))
4300 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4301 return (!lapic_in_kernel(vcpu) ||
4302 kvm_apic_accept_pic_intr(vcpu));
4305 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4307 return kvm_arch_interrupt_allowed(vcpu) &&
4308 kvm_cpu_accept_dm_intr(vcpu);
4311 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4312 struct kvm_interrupt *irq)
4314 if (irq->irq >= KVM_NR_INTERRUPTS)
4317 if (!irqchip_in_kernel(vcpu->kvm)) {
4318 kvm_queue_interrupt(vcpu, irq->irq, false);
4319 kvm_make_request(KVM_REQ_EVENT, vcpu);
4324 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4325 * fail for in-kernel 8259.
4327 if (pic_in_kernel(vcpu->kvm))
4330 if (vcpu->arch.pending_external_vector != -1)
4333 vcpu->arch.pending_external_vector = irq->irq;
4334 kvm_make_request(KVM_REQ_EVENT, vcpu);
4338 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4340 kvm_inject_nmi(vcpu);
4345 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4347 kvm_make_request(KVM_REQ_SMI, vcpu);
4352 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4353 struct kvm_tpr_access_ctl *tac)
4357 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4361 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4365 unsigned bank_num = mcg_cap & 0xff, bank;
4368 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4370 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4373 vcpu->arch.mcg_cap = mcg_cap;
4374 /* Init IA32_MCG_CTL to all 1s */
4375 if (mcg_cap & MCG_CTL_P)
4376 vcpu->arch.mcg_ctl = ~(u64)0;
4377 /* Init IA32_MCi_CTL to all 1s */
4378 for (bank = 0; bank < bank_num; bank++)
4379 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4381 static_call(kvm_x86_setup_mce)(vcpu);
4386 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4387 struct kvm_x86_mce *mce)
4389 u64 mcg_cap = vcpu->arch.mcg_cap;
4390 unsigned bank_num = mcg_cap & 0xff;
4391 u64 *banks = vcpu->arch.mce_banks;
4393 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4396 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4397 * reporting is disabled
4399 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4400 vcpu->arch.mcg_ctl != ~(u64)0)
4402 banks += 4 * mce->bank;
4404 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4405 * reporting is disabled for the bank
4407 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4409 if (mce->status & MCI_STATUS_UC) {
4410 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4411 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4412 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4415 if (banks[1] & MCI_STATUS_VAL)
4416 mce->status |= MCI_STATUS_OVER;
4417 banks[2] = mce->addr;
4418 banks[3] = mce->misc;
4419 vcpu->arch.mcg_status = mce->mcg_status;
4420 banks[1] = mce->status;
4421 kvm_queue_exception(vcpu, MC_VECTOR);
4422 } else if (!(banks[1] & MCI_STATUS_VAL)
4423 || !(banks[1] & MCI_STATUS_UC)) {
4424 if (banks[1] & MCI_STATUS_VAL)
4425 mce->status |= MCI_STATUS_OVER;
4426 banks[2] = mce->addr;
4427 banks[3] = mce->misc;
4428 banks[1] = mce->status;
4430 banks[1] |= MCI_STATUS_OVER;
4434 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4435 struct kvm_vcpu_events *events)
4439 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4443 * In guest mode, payload delivery should be deferred,
4444 * so that the L1 hypervisor can intercept #PF before
4445 * CR2 is modified (or intercept #DB before DR6 is
4446 * modified under nVMX). Unless the per-VM capability,
4447 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4448 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4449 * opportunistically defer the exception payload, deliver it if the
4450 * capability hasn't been requested before processing a
4451 * KVM_GET_VCPU_EVENTS.
4453 if (!vcpu->kvm->arch.exception_payload_enabled &&
4454 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4455 kvm_deliver_exception_payload(vcpu);
4458 * The API doesn't provide the instruction length for software
4459 * exceptions, so don't report them. As long as the guest RIP
4460 * isn't advanced, we should expect to encounter the exception
4463 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4464 events->exception.injected = 0;
4465 events->exception.pending = 0;
4467 events->exception.injected = vcpu->arch.exception.injected;
4468 events->exception.pending = vcpu->arch.exception.pending;
4470 * For ABI compatibility, deliberately conflate
4471 * pending and injected exceptions when
4472 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4474 if (!vcpu->kvm->arch.exception_payload_enabled)
4475 events->exception.injected |=
4476 vcpu->arch.exception.pending;
4478 events->exception.nr = vcpu->arch.exception.nr;
4479 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4480 events->exception.error_code = vcpu->arch.exception.error_code;
4481 events->exception_has_payload = vcpu->arch.exception.has_payload;
4482 events->exception_payload = vcpu->arch.exception.payload;
4484 events->interrupt.injected =
4485 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4486 events->interrupt.nr = vcpu->arch.interrupt.nr;
4487 events->interrupt.soft = 0;
4488 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4490 events->nmi.injected = vcpu->arch.nmi_injected;
4491 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4492 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4493 events->nmi.pad = 0;
4495 events->sipi_vector = 0; /* never valid when reporting to user space */
4497 events->smi.smm = is_smm(vcpu);
4498 events->smi.pending = vcpu->arch.smi_pending;
4499 events->smi.smm_inside_nmi =
4500 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4501 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4503 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4504 | KVM_VCPUEVENT_VALID_SHADOW
4505 | KVM_VCPUEVENT_VALID_SMM);
4506 if (vcpu->kvm->arch.exception_payload_enabled)
4507 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4509 memset(&events->reserved, 0, sizeof(events->reserved));
4512 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4514 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4515 struct kvm_vcpu_events *events)
4517 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4518 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4519 | KVM_VCPUEVENT_VALID_SHADOW
4520 | KVM_VCPUEVENT_VALID_SMM
4521 | KVM_VCPUEVENT_VALID_PAYLOAD))
4524 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4525 if (!vcpu->kvm->arch.exception_payload_enabled)
4527 if (events->exception.pending)
4528 events->exception.injected = 0;
4530 events->exception_has_payload = 0;
4532 events->exception.pending = 0;
4533 events->exception_has_payload = 0;
4536 if ((events->exception.injected || events->exception.pending) &&
4537 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4540 /* INITs are latched while in SMM */
4541 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4542 (events->smi.smm || events->smi.pending) &&
4543 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4547 vcpu->arch.exception.injected = events->exception.injected;
4548 vcpu->arch.exception.pending = events->exception.pending;
4549 vcpu->arch.exception.nr = events->exception.nr;
4550 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4551 vcpu->arch.exception.error_code = events->exception.error_code;
4552 vcpu->arch.exception.has_payload = events->exception_has_payload;
4553 vcpu->arch.exception.payload = events->exception_payload;
4555 vcpu->arch.interrupt.injected = events->interrupt.injected;
4556 vcpu->arch.interrupt.nr = events->interrupt.nr;
4557 vcpu->arch.interrupt.soft = events->interrupt.soft;
4558 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4559 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4560 events->interrupt.shadow);
4562 vcpu->arch.nmi_injected = events->nmi.injected;
4563 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4564 vcpu->arch.nmi_pending = events->nmi.pending;
4565 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4567 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4568 lapic_in_kernel(vcpu))
4569 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4571 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4572 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4573 kvm_smm_changed(vcpu, events->smi.smm);
4575 vcpu->arch.smi_pending = events->smi.pending;
4577 if (events->smi.smm) {
4578 if (events->smi.smm_inside_nmi)
4579 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4581 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4584 if (lapic_in_kernel(vcpu)) {
4585 if (events->smi.latched_init)
4586 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4588 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4592 kvm_make_request(KVM_REQ_EVENT, vcpu);
4597 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4598 struct kvm_debugregs *dbgregs)
4602 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4603 kvm_get_dr(vcpu, 6, &val);
4605 dbgregs->dr7 = vcpu->arch.dr7;
4607 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4610 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4611 struct kvm_debugregs *dbgregs)
4616 if (!kvm_dr6_valid(dbgregs->dr6))
4618 if (!kvm_dr7_valid(dbgregs->dr7))
4621 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4622 kvm_update_dr0123(vcpu);
4623 vcpu->arch.dr6 = dbgregs->dr6;
4624 vcpu->arch.dr7 = dbgregs->dr7;
4625 kvm_update_dr7(vcpu);
4630 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4632 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4634 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4635 u64 xstate_bv = xsave->header.xfeatures;
4639 * Copy legacy XSAVE area, to avoid complications with CPUID
4640 * leaves 0 and 1 in the loop below.
4642 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4645 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4646 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4649 * Copy each region from the possibly compacted offset to the
4650 * non-compacted offset.
4652 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4654 u64 xfeature_mask = valid & -valid;
4655 int xfeature_nr = fls64(xfeature_mask) - 1;
4656 void *src = get_xsave_addr(xsave, xfeature_nr);
4659 u32 size, offset, ecx, edx;
4660 cpuid_count(XSTATE_CPUID, xfeature_nr,
4661 &size, &offset, &ecx, &edx);
4662 if (xfeature_nr == XFEATURE_PKRU)
4663 memcpy(dest + offset, &vcpu->arch.pkru,
4664 sizeof(vcpu->arch.pkru));
4666 memcpy(dest + offset, src, size);
4670 valid -= xfeature_mask;
4674 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4676 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4677 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4681 * Copy legacy XSAVE area, to avoid complications with CPUID
4682 * leaves 0 and 1 in the loop below.
4684 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4686 /* Set XSTATE_BV and possibly XCOMP_BV. */
4687 xsave->header.xfeatures = xstate_bv;
4688 if (boot_cpu_has(X86_FEATURE_XSAVES))
4689 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4692 * Copy each region from the non-compacted offset to the
4693 * possibly compacted offset.
4695 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4697 u64 xfeature_mask = valid & -valid;
4698 int xfeature_nr = fls64(xfeature_mask) - 1;
4699 void *dest = get_xsave_addr(xsave, xfeature_nr);
4702 u32 size, offset, ecx, edx;
4703 cpuid_count(XSTATE_CPUID, xfeature_nr,
4704 &size, &offset, &ecx, &edx);
4705 if (xfeature_nr == XFEATURE_PKRU)
4706 memcpy(&vcpu->arch.pkru, src + offset,
4707 sizeof(vcpu->arch.pkru));
4709 memcpy(dest, src + offset, size);
4712 valid -= xfeature_mask;
4716 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4717 struct kvm_xsave *guest_xsave)
4719 if (!vcpu->arch.guest_fpu)
4722 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4723 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4724 fill_xsave((u8 *) guest_xsave->region, vcpu);
4726 memcpy(guest_xsave->region,
4727 &vcpu->arch.guest_fpu->state.fxsave,
4728 sizeof(struct fxregs_state));
4729 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4730 XFEATURE_MASK_FPSSE;
4734 #define XSAVE_MXCSR_OFFSET 24
4736 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4737 struct kvm_xsave *guest_xsave)
4742 if (!vcpu->arch.guest_fpu)
4745 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4746 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4748 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4750 * Here we allow setting states that are not present in
4751 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4752 * with old userspace.
4754 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4756 load_xsave(vcpu, (u8 *)guest_xsave->region);
4758 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4759 mxcsr & ~mxcsr_feature_mask)
4761 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4762 guest_xsave->region, sizeof(struct fxregs_state));
4767 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4768 struct kvm_xcrs *guest_xcrs)
4770 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4771 guest_xcrs->nr_xcrs = 0;
4775 guest_xcrs->nr_xcrs = 1;
4776 guest_xcrs->flags = 0;
4777 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4778 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4781 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4782 struct kvm_xcrs *guest_xcrs)
4786 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4789 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4792 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4793 /* Only support XCR0 currently */
4794 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4795 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4796 guest_xcrs->xcrs[i].value);
4805 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4806 * stopped by the hypervisor. This function will be called from the host only.
4807 * EINVAL is returned when the host attempts to set the flag for a guest that
4808 * does not support pv clocks.
4810 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4812 if (!vcpu->arch.pv_time_enabled)
4814 vcpu->arch.pvclock_set_guest_stopped_request = true;
4815 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4819 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4820 struct kvm_enable_cap *cap)
4823 uint16_t vmcs_version;
4824 void __user *user_ptr;
4830 case KVM_CAP_HYPERV_SYNIC2:
4835 case KVM_CAP_HYPERV_SYNIC:
4836 if (!irqchip_in_kernel(vcpu->kvm))
4838 return kvm_hv_activate_synic(vcpu, cap->cap ==
4839 KVM_CAP_HYPERV_SYNIC2);
4840 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4841 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4843 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4845 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4846 if (copy_to_user(user_ptr, &vmcs_version,
4847 sizeof(vmcs_version)))
4851 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4852 if (!kvm_x86_ops.enable_direct_tlbflush)
4855 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4857 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4858 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
4860 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4861 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4862 if (vcpu->arch.pv_cpuid.enforce)
4863 kvm_update_pv_runtime(vcpu);
4871 long kvm_arch_vcpu_ioctl(struct file *filp,
4872 unsigned int ioctl, unsigned long arg)
4874 struct kvm_vcpu *vcpu = filp->private_data;
4875 void __user *argp = (void __user *)arg;
4878 struct kvm_sregs2 *sregs2;
4879 struct kvm_lapic_state *lapic;
4880 struct kvm_xsave *xsave;
4881 struct kvm_xcrs *xcrs;
4889 case KVM_GET_LAPIC: {
4891 if (!lapic_in_kernel(vcpu))
4893 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4894 GFP_KERNEL_ACCOUNT);
4899 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4903 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4908 case KVM_SET_LAPIC: {
4910 if (!lapic_in_kernel(vcpu))
4912 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4913 if (IS_ERR(u.lapic)) {
4914 r = PTR_ERR(u.lapic);
4918 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4921 case KVM_INTERRUPT: {
4922 struct kvm_interrupt irq;
4925 if (copy_from_user(&irq, argp, sizeof(irq)))
4927 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4931 r = kvm_vcpu_ioctl_nmi(vcpu);
4935 r = kvm_vcpu_ioctl_smi(vcpu);
4938 case KVM_SET_CPUID: {
4939 struct kvm_cpuid __user *cpuid_arg = argp;
4940 struct kvm_cpuid cpuid;
4943 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4945 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4948 case KVM_SET_CPUID2: {
4949 struct kvm_cpuid2 __user *cpuid_arg = argp;
4950 struct kvm_cpuid2 cpuid;
4953 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4955 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4956 cpuid_arg->entries);
4959 case KVM_GET_CPUID2: {
4960 struct kvm_cpuid2 __user *cpuid_arg = argp;
4961 struct kvm_cpuid2 cpuid;
4964 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4966 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4967 cpuid_arg->entries);
4971 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4976 case KVM_GET_MSRS: {
4977 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4978 r = msr_io(vcpu, argp, do_get_msr, 1);
4979 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4982 case KVM_SET_MSRS: {
4983 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4984 r = msr_io(vcpu, argp, do_set_msr, 0);
4985 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4988 case KVM_TPR_ACCESS_REPORTING: {
4989 struct kvm_tpr_access_ctl tac;
4992 if (copy_from_user(&tac, argp, sizeof(tac)))
4994 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4998 if (copy_to_user(argp, &tac, sizeof(tac)))
5003 case KVM_SET_VAPIC_ADDR: {
5004 struct kvm_vapic_addr va;
5008 if (!lapic_in_kernel(vcpu))
5011 if (copy_from_user(&va, argp, sizeof(va)))
5013 idx = srcu_read_lock(&vcpu->kvm->srcu);
5014 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5015 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5018 case KVM_X86_SETUP_MCE: {
5022 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5024 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5027 case KVM_X86_SET_MCE: {
5028 struct kvm_x86_mce mce;
5031 if (copy_from_user(&mce, argp, sizeof(mce)))
5033 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5036 case KVM_GET_VCPU_EVENTS: {
5037 struct kvm_vcpu_events events;
5039 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5042 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5047 case KVM_SET_VCPU_EVENTS: {
5048 struct kvm_vcpu_events events;
5051 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5054 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5057 case KVM_GET_DEBUGREGS: {
5058 struct kvm_debugregs dbgregs;
5060 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5063 if (copy_to_user(argp, &dbgregs,
5064 sizeof(struct kvm_debugregs)))
5069 case KVM_SET_DEBUGREGS: {
5070 struct kvm_debugregs dbgregs;
5073 if (copy_from_user(&dbgregs, argp,
5074 sizeof(struct kvm_debugregs)))
5077 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5080 case KVM_GET_XSAVE: {
5081 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5086 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5089 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5094 case KVM_SET_XSAVE: {
5095 u.xsave = memdup_user(argp, sizeof(*u.xsave));
5096 if (IS_ERR(u.xsave)) {
5097 r = PTR_ERR(u.xsave);
5101 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5104 case KVM_GET_XCRS: {
5105 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5110 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5113 if (copy_to_user(argp, u.xcrs,
5114 sizeof(struct kvm_xcrs)))
5119 case KVM_SET_XCRS: {
5120 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5121 if (IS_ERR(u.xcrs)) {
5122 r = PTR_ERR(u.xcrs);
5126 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5129 case KVM_SET_TSC_KHZ: {
5133 user_tsc_khz = (u32)arg;
5135 if (kvm_has_tsc_control &&
5136 user_tsc_khz >= kvm_max_guest_tsc_khz)
5139 if (user_tsc_khz == 0)
5140 user_tsc_khz = tsc_khz;
5142 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5147 case KVM_GET_TSC_KHZ: {
5148 r = vcpu->arch.virtual_tsc_khz;
5151 case KVM_KVMCLOCK_CTRL: {
5152 r = kvm_set_guest_paused(vcpu);
5155 case KVM_ENABLE_CAP: {
5156 struct kvm_enable_cap cap;
5159 if (copy_from_user(&cap, argp, sizeof(cap)))
5161 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5164 case KVM_GET_NESTED_STATE: {
5165 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5169 if (!kvm_x86_ops.nested_ops->get_state)
5172 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5174 if (get_user(user_data_size, &user_kvm_nested_state->size))
5177 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5182 if (r > user_data_size) {
5183 if (put_user(r, &user_kvm_nested_state->size))
5193 case KVM_SET_NESTED_STATE: {
5194 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5195 struct kvm_nested_state kvm_state;
5199 if (!kvm_x86_ops.nested_ops->set_state)
5203 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5207 if (kvm_state.size < sizeof(kvm_state))
5210 if (kvm_state.flags &
5211 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5212 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5213 | KVM_STATE_NESTED_GIF_SET))
5216 /* nested_run_pending implies guest_mode. */
5217 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5218 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5221 idx = srcu_read_lock(&vcpu->kvm->srcu);
5222 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5223 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5226 case KVM_GET_SUPPORTED_HV_CPUID:
5227 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5229 #ifdef CONFIG_KVM_XEN
5230 case KVM_XEN_VCPU_GET_ATTR: {
5231 struct kvm_xen_vcpu_attr xva;
5234 if (copy_from_user(&xva, argp, sizeof(xva)))
5236 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5237 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5241 case KVM_XEN_VCPU_SET_ATTR: {
5242 struct kvm_xen_vcpu_attr xva;
5245 if (copy_from_user(&xva, argp, sizeof(xva)))
5247 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5251 case KVM_GET_SREGS2: {
5252 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5256 __get_sregs2(vcpu, u.sregs2);
5258 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5263 case KVM_SET_SREGS2: {
5264 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5265 if (IS_ERR(u.sregs2)) {
5266 r = PTR_ERR(u.sregs2);
5270 r = __set_sregs2(vcpu, u.sregs2);
5283 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5285 return VM_FAULT_SIGBUS;
5288 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5292 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5294 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5298 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5301 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5304 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5305 unsigned long kvm_nr_mmu_pages)
5307 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5310 mutex_lock(&kvm->slots_lock);
5312 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5313 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5315 mutex_unlock(&kvm->slots_lock);
5319 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5321 return kvm->arch.n_max_mmu_pages;
5324 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5326 struct kvm_pic *pic = kvm->arch.vpic;
5330 switch (chip->chip_id) {
5331 case KVM_IRQCHIP_PIC_MASTER:
5332 memcpy(&chip->chip.pic, &pic->pics[0],
5333 sizeof(struct kvm_pic_state));
5335 case KVM_IRQCHIP_PIC_SLAVE:
5336 memcpy(&chip->chip.pic, &pic->pics[1],
5337 sizeof(struct kvm_pic_state));
5339 case KVM_IRQCHIP_IOAPIC:
5340 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5349 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5351 struct kvm_pic *pic = kvm->arch.vpic;
5355 switch (chip->chip_id) {
5356 case KVM_IRQCHIP_PIC_MASTER:
5357 spin_lock(&pic->lock);
5358 memcpy(&pic->pics[0], &chip->chip.pic,
5359 sizeof(struct kvm_pic_state));
5360 spin_unlock(&pic->lock);
5362 case KVM_IRQCHIP_PIC_SLAVE:
5363 spin_lock(&pic->lock);
5364 memcpy(&pic->pics[1], &chip->chip.pic,
5365 sizeof(struct kvm_pic_state));
5366 spin_unlock(&pic->lock);
5368 case KVM_IRQCHIP_IOAPIC:
5369 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5375 kvm_pic_update_irq(pic);
5379 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5381 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5383 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5385 mutex_lock(&kps->lock);
5386 memcpy(ps, &kps->channels, sizeof(*ps));
5387 mutex_unlock(&kps->lock);
5391 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5394 struct kvm_pit *pit = kvm->arch.vpit;
5396 mutex_lock(&pit->pit_state.lock);
5397 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5398 for (i = 0; i < 3; i++)
5399 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5400 mutex_unlock(&pit->pit_state.lock);
5404 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5406 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5407 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5408 sizeof(ps->channels));
5409 ps->flags = kvm->arch.vpit->pit_state.flags;
5410 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5411 memset(&ps->reserved, 0, sizeof(ps->reserved));
5415 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5419 u32 prev_legacy, cur_legacy;
5420 struct kvm_pit *pit = kvm->arch.vpit;
5422 mutex_lock(&pit->pit_state.lock);
5423 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5424 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5425 if (!prev_legacy && cur_legacy)
5427 memcpy(&pit->pit_state.channels, &ps->channels,
5428 sizeof(pit->pit_state.channels));
5429 pit->pit_state.flags = ps->flags;
5430 for (i = 0; i < 3; i++)
5431 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5433 mutex_unlock(&pit->pit_state.lock);
5437 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5438 struct kvm_reinject_control *control)
5440 struct kvm_pit *pit = kvm->arch.vpit;
5442 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5443 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5444 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5446 mutex_lock(&pit->pit_state.lock);
5447 kvm_pit_set_reinject(pit, control->pit_reinject);
5448 mutex_unlock(&pit->pit_state.lock);
5453 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5457 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5458 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5459 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5462 struct kvm_vcpu *vcpu;
5465 kvm_for_each_vcpu(i, vcpu, kvm)
5466 kvm_vcpu_kick(vcpu);
5469 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5472 if (!irqchip_in_kernel(kvm))
5475 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5476 irq_event->irq, irq_event->level,
5481 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5482 struct kvm_enable_cap *cap)
5490 case KVM_CAP_DISABLE_QUIRKS:
5491 kvm->arch.disabled_quirks = cap->args[0];
5494 case KVM_CAP_SPLIT_IRQCHIP: {
5495 mutex_lock(&kvm->lock);
5497 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5498 goto split_irqchip_unlock;
5500 if (irqchip_in_kernel(kvm))
5501 goto split_irqchip_unlock;
5502 if (kvm->created_vcpus)
5503 goto split_irqchip_unlock;
5504 r = kvm_setup_empty_irq_routing(kvm);
5506 goto split_irqchip_unlock;
5507 /* Pairs with irqchip_in_kernel. */
5509 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5510 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5512 split_irqchip_unlock:
5513 mutex_unlock(&kvm->lock);
5516 case KVM_CAP_X2APIC_API:
5518 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5521 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5522 kvm->arch.x2apic_format = true;
5523 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5524 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5528 case KVM_CAP_X86_DISABLE_EXITS:
5530 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5533 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5534 kvm_can_mwait_in_guest())
5535 kvm->arch.mwait_in_guest = true;
5536 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5537 kvm->arch.hlt_in_guest = true;
5538 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5539 kvm->arch.pause_in_guest = true;
5540 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5541 kvm->arch.cstate_in_guest = true;
5544 case KVM_CAP_MSR_PLATFORM_INFO:
5545 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5548 case KVM_CAP_EXCEPTION_PAYLOAD:
5549 kvm->arch.exception_payload_enabled = cap->args[0];
5552 case KVM_CAP_X86_USER_SPACE_MSR:
5553 kvm->arch.user_space_msr_mask = cap->args[0];
5556 case KVM_CAP_X86_BUS_LOCK_EXIT:
5558 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5561 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5562 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5565 if (kvm_has_bus_lock_exit &&
5566 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5567 kvm->arch.bus_lock_detection_enabled = true;
5570 #ifdef CONFIG_X86_SGX_KVM
5571 case KVM_CAP_SGX_ATTRIBUTE: {
5572 unsigned long allowed_attributes = 0;
5574 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5578 /* KVM only supports the PROVISIONKEY privileged attribute. */
5579 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5580 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5581 kvm->arch.sgx_provisioning_allowed = true;
5587 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5589 if (kvm_x86_ops.vm_copy_enc_context_from)
5590 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5599 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5601 struct kvm_x86_msr_filter *msr_filter;
5603 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5607 msr_filter->default_allow = default_allow;
5611 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5618 for (i = 0; i < msr_filter->count; i++)
5619 kfree(msr_filter->ranges[i].bitmap);
5624 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5625 struct kvm_msr_filter_range *user_range)
5627 unsigned long *bitmap = NULL;
5630 if (!user_range->nmsrs)
5633 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5636 if (!user_range->flags)
5639 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5640 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5643 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5645 return PTR_ERR(bitmap);
5647 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5648 .flags = user_range->flags,
5649 .base = user_range->base,
5650 .nmsrs = user_range->nmsrs,
5654 msr_filter->count++;
5658 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5660 struct kvm_msr_filter __user *user_msr_filter = argp;
5661 struct kvm_x86_msr_filter *new_filter, *old_filter;
5662 struct kvm_msr_filter filter;
5668 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5671 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5672 empty &= !filter.ranges[i].nmsrs;
5674 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5675 if (empty && !default_allow)
5678 new_filter = kvm_alloc_msr_filter(default_allow);
5682 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5683 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5685 kvm_free_msr_filter(new_filter);
5690 mutex_lock(&kvm->lock);
5692 /* The per-VM filter is protected by kvm->lock... */
5693 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5695 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5696 synchronize_srcu(&kvm->srcu);
5698 kvm_free_msr_filter(old_filter);
5700 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5701 mutex_unlock(&kvm->lock);
5706 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5707 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5709 struct kvm_vcpu *vcpu;
5712 mutex_lock(&kvm->lock);
5713 kvm_for_each_vcpu(i, vcpu, kvm) {
5714 if (!vcpu->arch.pv_time_enabled)
5717 ret = kvm_set_guest_paused(vcpu);
5719 kvm_err("Failed to pause guest VCPU%d: %d\n",
5720 vcpu->vcpu_id, ret);
5724 mutex_unlock(&kvm->lock);
5726 return ret ? NOTIFY_BAD : NOTIFY_DONE;
5729 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5732 case PM_HIBERNATION_PREPARE:
5733 case PM_SUSPEND_PREPARE:
5734 return kvm_arch_suspend_notifier(kvm);
5739 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5741 long kvm_arch_vm_ioctl(struct file *filp,
5742 unsigned int ioctl, unsigned long arg)
5744 struct kvm *kvm = filp->private_data;
5745 void __user *argp = (void __user *)arg;
5748 * This union makes it completely explicit to gcc-3.x
5749 * that these two variables' stack usage should be
5750 * combined, not added together.
5753 struct kvm_pit_state ps;
5754 struct kvm_pit_state2 ps2;
5755 struct kvm_pit_config pit_config;
5759 case KVM_SET_TSS_ADDR:
5760 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5762 case KVM_SET_IDENTITY_MAP_ADDR: {
5765 mutex_lock(&kvm->lock);
5767 if (kvm->created_vcpus)
5768 goto set_identity_unlock;
5770 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5771 goto set_identity_unlock;
5772 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5773 set_identity_unlock:
5774 mutex_unlock(&kvm->lock);
5777 case KVM_SET_NR_MMU_PAGES:
5778 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5780 case KVM_GET_NR_MMU_PAGES:
5781 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5783 case KVM_CREATE_IRQCHIP: {
5784 mutex_lock(&kvm->lock);
5787 if (irqchip_in_kernel(kvm))
5788 goto create_irqchip_unlock;
5791 if (kvm->created_vcpus)
5792 goto create_irqchip_unlock;
5794 r = kvm_pic_init(kvm);
5796 goto create_irqchip_unlock;
5798 r = kvm_ioapic_init(kvm);
5800 kvm_pic_destroy(kvm);
5801 goto create_irqchip_unlock;
5804 r = kvm_setup_default_irq_routing(kvm);
5806 kvm_ioapic_destroy(kvm);
5807 kvm_pic_destroy(kvm);
5808 goto create_irqchip_unlock;
5810 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5812 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5813 create_irqchip_unlock:
5814 mutex_unlock(&kvm->lock);
5817 case KVM_CREATE_PIT:
5818 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5820 case KVM_CREATE_PIT2:
5822 if (copy_from_user(&u.pit_config, argp,
5823 sizeof(struct kvm_pit_config)))
5826 mutex_lock(&kvm->lock);
5829 goto create_pit_unlock;
5831 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5835 mutex_unlock(&kvm->lock);
5837 case KVM_GET_IRQCHIP: {
5838 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5839 struct kvm_irqchip *chip;
5841 chip = memdup_user(argp, sizeof(*chip));
5848 if (!irqchip_kernel(kvm))
5849 goto get_irqchip_out;
5850 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5852 goto get_irqchip_out;
5854 if (copy_to_user(argp, chip, sizeof(*chip)))
5855 goto get_irqchip_out;
5861 case KVM_SET_IRQCHIP: {
5862 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5863 struct kvm_irqchip *chip;
5865 chip = memdup_user(argp, sizeof(*chip));
5872 if (!irqchip_kernel(kvm))
5873 goto set_irqchip_out;
5874 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5881 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5884 if (!kvm->arch.vpit)
5886 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5890 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5897 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5899 mutex_lock(&kvm->lock);
5901 if (!kvm->arch.vpit)
5903 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5905 mutex_unlock(&kvm->lock);
5908 case KVM_GET_PIT2: {
5910 if (!kvm->arch.vpit)
5912 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5916 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5921 case KVM_SET_PIT2: {
5923 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5925 mutex_lock(&kvm->lock);
5927 if (!kvm->arch.vpit)
5929 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5931 mutex_unlock(&kvm->lock);
5934 case KVM_REINJECT_CONTROL: {
5935 struct kvm_reinject_control control;
5937 if (copy_from_user(&control, argp, sizeof(control)))
5940 if (!kvm->arch.vpit)
5942 r = kvm_vm_ioctl_reinject(kvm, &control);
5945 case KVM_SET_BOOT_CPU_ID:
5947 mutex_lock(&kvm->lock);
5948 if (kvm->created_vcpus)
5951 kvm->arch.bsp_vcpu_id = arg;
5952 mutex_unlock(&kvm->lock);
5954 #ifdef CONFIG_KVM_XEN
5955 case KVM_XEN_HVM_CONFIG: {
5956 struct kvm_xen_hvm_config xhc;
5958 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5960 r = kvm_xen_hvm_config(kvm, &xhc);
5963 case KVM_XEN_HVM_GET_ATTR: {
5964 struct kvm_xen_hvm_attr xha;
5967 if (copy_from_user(&xha, argp, sizeof(xha)))
5969 r = kvm_xen_hvm_get_attr(kvm, &xha);
5970 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5974 case KVM_XEN_HVM_SET_ATTR: {
5975 struct kvm_xen_hvm_attr xha;
5978 if (copy_from_user(&xha, argp, sizeof(xha)))
5980 r = kvm_xen_hvm_set_attr(kvm, &xha);
5984 case KVM_SET_CLOCK: {
5985 struct kvm_arch *ka = &kvm->arch;
5986 struct kvm_clock_data user_ns;
5990 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5999 * TODO: userspace has to take care of races with VCPU_RUN, so
6000 * kvm_gen_update_masterclock() can be cut down to locked
6001 * pvclock_update_vm_gtod_copy().
6003 kvm_gen_update_masterclock(kvm);
6006 * This pairs with kvm_guest_time_update(): when masterclock is
6007 * in use, we use master_kernel_ns + kvmclock_offset to set
6008 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6009 * is slightly ahead) here we risk going negative on unsigned
6010 * 'system_time' when 'user_ns.clock' is very small.
6012 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
6013 if (kvm->arch.use_master_clock)
6014 now_ns = ka->master_kernel_ns;
6016 now_ns = get_kvmclock_base_ns();
6017 ka->kvmclock_offset = user_ns.clock - now_ns;
6018 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
6020 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
6023 case KVM_GET_CLOCK: {
6024 struct kvm_clock_data user_ns;
6027 now_ns = get_kvmclock_ns(kvm);
6028 user_ns.clock = now_ns;
6029 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
6030 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
6033 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
6038 case KVM_MEMORY_ENCRYPT_OP: {
6040 if (kvm_x86_ops.mem_enc_op)
6041 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6044 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6045 struct kvm_enc_region region;
6048 if (copy_from_user(®ion, argp, sizeof(region)))
6052 if (kvm_x86_ops.mem_enc_reg_region)
6053 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
6056 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6057 struct kvm_enc_region region;
6060 if (copy_from_user(®ion, argp, sizeof(region)))
6064 if (kvm_x86_ops.mem_enc_unreg_region)
6065 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
6068 case KVM_HYPERV_EVENTFD: {
6069 struct kvm_hyperv_eventfd hvevfd;
6072 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6074 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6077 case KVM_SET_PMU_EVENT_FILTER:
6078 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6080 case KVM_X86_SET_MSR_FILTER:
6081 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6090 static void kvm_init_msr_list(void)
6092 struct x86_pmu_capability x86_pmu;
6096 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6097 "Please update the fixed PMCs in msrs_to_saved_all[]");
6099 perf_get_x86_pmu_capability(&x86_pmu);
6101 num_msrs_to_save = 0;
6102 num_emulated_msrs = 0;
6103 num_msr_based_features = 0;
6105 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6106 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6110 * Even MSRs that are valid in the host may not be exposed
6111 * to the guests in some cases.
6113 switch (msrs_to_save_all[i]) {
6114 case MSR_IA32_BNDCFGS:
6115 if (!kvm_mpx_supported())
6119 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6120 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6123 case MSR_IA32_UMWAIT_CONTROL:
6124 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6127 case MSR_IA32_RTIT_CTL:
6128 case MSR_IA32_RTIT_STATUS:
6129 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6132 case MSR_IA32_RTIT_CR3_MATCH:
6133 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6134 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6137 case MSR_IA32_RTIT_OUTPUT_BASE:
6138 case MSR_IA32_RTIT_OUTPUT_MASK:
6139 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6140 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6141 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6144 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6145 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6146 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6147 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6150 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6151 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6152 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6155 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6156 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6157 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6164 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6167 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6168 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6171 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6174 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6175 struct kvm_msr_entry msr;
6177 msr.index = msr_based_features_all[i];
6178 if (kvm_get_msr_feature(&msr))
6181 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6185 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6193 if (!(lapic_in_kernel(vcpu) &&
6194 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6195 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6206 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6213 if (!(lapic_in_kernel(vcpu) &&
6214 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6216 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6218 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6228 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6229 struct kvm_segment *var, int seg)
6231 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6234 void kvm_get_segment(struct kvm_vcpu *vcpu,
6235 struct kvm_segment *var, int seg)
6237 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6240 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6241 struct x86_exception *exception)
6245 BUG_ON(!mmu_is_nested(vcpu));
6247 /* NPT walks are always user-walks */
6248 access |= PFERR_USER_MASK;
6249 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6254 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6255 struct x86_exception *exception)
6257 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6258 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6260 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6262 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6263 struct x86_exception *exception)
6265 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6266 access |= PFERR_FETCH_MASK;
6267 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6270 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6271 struct x86_exception *exception)
6273 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6274 access |= PFERR_WRITE_MASK;
6275 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6277 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6279 /* uses this to access any guest's mapped memory without checking CPL */
6280 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6281 struct x86_exception *exception)
6283 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6286 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6287 struct kvm_vcpu *vcpu, u32 access,
6288 struct x86_exception *exception)
6291 int r = X86EMUL_CONTINUE;
6294 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6296 unsigned offset = addr & (PAGE_SIZE-1);
6297 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6300 if (gpa == UNMAPPED_GVA)
6301 return X86EMUL_PROPAGATE_FAULT;
6302 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6305 r = X86EMUL_IO_NEEDED;
6317 /* used for instruction fetching */
6318 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6319 gva_t addr, void *val, unsigned int bytes,
6320 struct x86_exception *exception)
6322 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6323 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6327 /* Inline kvm_read_guest_virt_helper for speed. */
6328 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6330 if (unlikely(gpa == UNMAPPED_GVA))
6331 return X86EMUL_PROPAGATE_FAULT;
6333 offset = addr & (PAGE_SIZE-1);
6334 if (WARN_ON(offset + bytes > PAGE_SIZE))
6335 bytes = (unsigned)PAGE_SIZE - offset;
6336 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6338 if (unlikely(ret < 0))
6339 return X86EMUL_IO_NEEDED;
6341 return X86EMUL_CONTINUE;
6344 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6345 gva_t addr, void *val, unsigned int bytes,
6346 struct x86_exception *exception)
6348 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6351 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6352 * is returned, but our callers are not ready for that and they blindly
6353 * call kvm_inject_page_fault. Ensure that they at least do not leak
6354 * uninitialized kernel stack memory into cr2 and error code.
6356 memset(exception, 0, sizeof(*exception));
6357 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6360 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6362 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6363 gva_t addr, void *val, unsigned int bytes,
6364 struct x86_exception *exception, bool system)
6366 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6369 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6370 access |= PFERR_USER_MASK;
6372 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6375 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6376 unsigned long addr, void *val, unsigned int bytes)
6378 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6379 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6381 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6384 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6385 struct kvm_vcpu *vcpu, u32 access,
6386 struct x86_exception *exception)
6389 int r = X86EMUL_CONTINUE;
6392 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6395 unsigned offset = addr & (PAGE_SIZE-1);
6396 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6399 if (gpa == UNMAPPED_GVA)
6400 return X86EMUL_PROPAGATE_FAULT;
6401 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6403 r = X86EMUL_IO_NEEDED;
6415 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6416 unsigned int bytes, struct x86_exception *exception,
6419 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6420 u32 access = PFERR_WRITE_MASK;
6422 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6423 access |= PFERR_USER_MASK;
6425 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6429 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6430 unsigned int bytes, struct x86_exception *exception)
6432 /* kvm_write_guest_virt_system can pull in tons of pages. */
6433 vcpu->arch.l1tf_flush_l1d = true;
6435 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6436 PFERR_WRITE_MASK, exception);
6438 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6440 int handle_ud(struct kvm_vcpu *vcpu)
6442 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6443 int emul_type = EMULTYPE_TRAP_UD;
6444 char sig[5]; /* ud2; .ascii "kvm" */
6445 struct x86_exception e;
6447 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6450 if (force_emulation_prefix &&
6451 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6452 sig, sizeof(sig), &e) == 0 &&
6453 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6454 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6455 emul_type = EMULTYPE_TRAP_UD_FORCED;
6458 return kvm_emulate_instruction(vcpu, emul_type);
6460 EXPORT_SYMBOL_GPL(handle_ud);
6462 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6463 gpa_t gpa, bool write)
6465 /* For APIC access vmexit */
6466 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6469 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6470 trace_vcpu_match_mmio(gva, gpa, write, true);
6477 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6478 gpa_t *gpa, struct x86_exception *exception,
6481 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6482 | (write ? PFERR_WRITE_MASK : 0);
6485 * currently PKRU is only applied to ept enabled guest so
6486 * there is no pkey in EPT page table for L1 guest or EPT
6487 * shadow page table for L2 guest.
6489 if (vcpu_match_mmio_gva(vcpu, gva)
6490 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6491 vcpu->arch.mmio_access, 0, access)) {
6492 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6493 (gva & (PAGE_SIZE - 1));
6494 trace_vcpu_match_mmio(gva, *gpa, write, false);
6498 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6500 if (*gpa == UNMAPPED_GVA)
6503 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6506 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6507 const void *val, int bytes)
6511 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6514 kvm_page_track_write(vcpu, gpa, val, bytes);
6518 struct read_write_emulator_ops {
6519 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6521 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6522 void *val, int bytes);
6523 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6524 int bytes, void *val);
6525 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6526 void *val, int bytes);
6530 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6532 if (vcpu->mmio_read_completed) {
6533 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6534 vcpu->mmio_fragments[0].gpa, val);
6535 vcpu->mmio_read_completed = 0;
6542 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6543 void *val, int bytes)
6545 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6548 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6549 void *val, int bytes)
6551 return emulator_write_phys(vcpu, gpa, val, bytes);
6554 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6556 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6557 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6560 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6561 void *val, int bytes)
6563 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6564 return X86EMUL_IO_NEEDED;
6567 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6568 void *val, int bytes)
6570 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6572 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6573 return X86EMUL_CONTINUE;
6576 static const struct read_write_emulator_ops read_emultor = {
6577 .read_write_prepare = read_prepare,
6578 .read_write_emulate = read_emulate,
6579 .read_write_mmio = vcpu_mmio_read,
6580 .read_write_exit_mmio = read_exit_mmio,
6583 static const struct read_write_emulator_ops write_emultor = {
6584 .read_write_emulate = write_emulate,
6585 .read_write_mmio = write_mmio,
6586 .read_write_exit_mmio = write_exit_mmio,
6590 static int emulator_read_write_onepage(unsigned long addr, void *val,
6592 struct x86_exception *exception,
6593 struct kvm_vcpu *vcpu,
6594 const struct read_write_emulator_ops *ops)
6598 bool write = ops->write;
6599 struct kvm_mmio_fragment *frag;
6600 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6603 * If the exit was due to a NPF we may already have a GPA.
6604 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6605 * Note, this cannot be used on string operations since string
6606 * operation using rep will only have the initial GPA from the NPF
6609 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6610 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6611 gpa = ctxt->gpa_val;
6612 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6614 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6616 return X86EMUL_PROPAGATE_FAULT;
6619 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6620 return X86EMUL_CONTINUE;
6623 * Is this MMIO handled locally?
6625 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6626 if (handled == bytes)
6627 return X86EMUL_CONTINUE;
6633 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6634 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6638 return X86EMUL_CONTINUE;
6641 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6643 void *val, unsigned int bytes,
6644 struct x86_exception *exception,
6645 const struct read_write_emulator_ops *ops)
6647 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6651 if (ops->read_write_prepare &&
6652 ops->read_write_prepare(vcpu, val, bytes))
6653 return X86EMUL_CONTINUE;
6655 vcpu->mmio_nr_fragments = 0;
6657 /* Crossing a page boundary? */
6658 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6661 now = -addr & ~PAGE_MASK;
6662 rc = emulator_read_write_onepage(addr, val, now, exception,
6665 if (rc != X86EMUL_CONTINUE)
6668 if (ctxt->mode != X86EMUL_MODE_PROT64)
6674 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6676 if (rc != X86EMUL_CONTINUE)
6679 if (!vcpu->mmio_nr_fragments)
6682 gpa = vcpu->mmio_fragments[0].gpa;
6684 vcpu->mmio_needed = 1;
6685 vcpu->mmio_cur_fragment = 0;
6687 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6688 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6689 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6690 vcpu->run->mmio.phys_addr = gpa;
6692 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6695 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6699 struct x86_exception *exception)
6701 return emulator_read_write(ctxt, addr, val, bytes,
6702 exception, &read_emultor);
6705 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6709 struct x86_exception *exception)
6711 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6712 exception, &write_emultor);
6715 #define CMPXCHG_TYPE(t, ptr, old, new) \
6716 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6718 #ifdef CONFIG_X86_64
6719 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6721 # define CMPXCHG64(ptr, old, new) \
6722 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6725 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6730 struct x86_exception *exception)
6732 struct kvm_host_map map;
6733 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6739 /* guests cmpxchg8b have to be emulated atomically */
6740 if (bytes > 8 || (bytes & (bytes - 1)))
6743 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6745 if (gpa == UNMAPPED_GVA ||
6746 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6750 * Emulate the atomic as a straight write to avoid #AC if SLD is
6751 * enabled in the host and the access splits a cache line.
6753 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6754 page_line_mask = ~(cache_line_size() - 1);
6756 page_line_mask = PAGE_MASK;
6758 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6761 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6764 kaddr = map.hva + offset_in_page(gpa);
6768 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6771 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6774 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6777 exchanged = CMPXCHG64(kaddr, old, new);
6783 kvm_vcpu_unmap(vcpu, &map, true);
6786 return X86EMUL_CMPXCHG_FAILED;
6788 kvm_page_track_write(vcpu, gpa, new, bytes);
6790 return X86EMUL_CONTINUE;
6793 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6795 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6798 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6802 for (i = 0; i < vcpu->arch.pio.count; i++) {
6803 if (vcpu->arch.pio.in)
6804 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6805 vcpu->arch.pio.size, pd);
6807 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6808 vcpu->arch.pio.port, vcpu->arch.pio.size,
6812 pd += vcpu->arch.pio.size;
6817 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6818 unsigned short port, void *val,
6819 unsigned int count, bool in)
6821 vcpu->arch.pio.port = port;
6822 vcpu->arch.pio.in = in;
6823 vcpu->arch.pio.count = count;
6824 vcpu->arch.pio.size = size;
6826 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6827 vcpu->arch.pio.count = 0;
6831 vcpu->run->exit_reason = KVM_EXIT_IO;
6832 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6833 vcpu->run->io.size = size;
6834 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6835 vcpu->run->io.count = count;
6836 vcpu->run->io.port = port;
6841 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6842 unsigned short port, void *val, unsigned int count)
6846 if (vcpu->arch.pio.count)
6849 memset(vcpu->arch.pio_data, 0, size * count);
6851 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6854 memcpy(val, vcpu->arch.pio_data, size * count);
6855 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6856 vcpu->arch.pio.count = 0;
6863 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6864 int size, unsigned short port, void *val,
6867 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6871 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6872 unsigned short port, const void *val,
6875 memcpy(vcpu->arch.pio_data, val, size * count);
6876 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6877 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6880 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6881 int size, unsigned short port,
6882 const void *val, unsigned int count)
6884 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6887 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6889 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6892 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6894 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6897 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6899 if (!need_emulate_wbinvd(vcpu))
6900 return X86EMUL_CONTINUE;
6902 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6903 int cpu = get_cpu();
6905 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6906 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6907 wbinvd_ipi, NULL, 1);
6909 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6912 return X86EMUL_CONTINUE;
6915 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6917 kvm_emulate_wbinvd_noskip(vcpu);
6918 return kvm_skip_emulated_instruction(vcpu);
6920 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6924 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6926 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6929 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6930 unsigned long *dest)
6932 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6935 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6936 unsigned long value)
6939 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6942 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6944 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6947 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6949 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6950 unsigned long value;
6954 value = kvm_read_cr0(vcpu);
6957 value = vcpu->arch.cr2;
6960 value = kvm_read_cr3(vcpu);
6963 value = kvm_read_cr4(vcpu);
6966 value = kvm_get_cr8(vcpu);
6969 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6976 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6978 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6983 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6986 vcpu->arch.cr2 = val;
6989 res = kvm_set_cr3(vcpu, val);
6992 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6995 res = kvm_set_cr8(vcpu, val);
6998 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7005 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7007 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7010 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7012 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7015 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7017 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7020 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7022 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7025 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7027 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7030 static unsigned long emulator_get_cached_segment_base(
7031 struct x86_emulate_ctxt *ctxt, int seg)
7033 return get_segment_base(emul_to_vcpu(ctxt), seg);
7036 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7037 struct desc_struct *desc, u32 *base3,
7040 struct kvm_segment var;
7042 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7043 *selector = var.selector;
7046 memset(desc, 0, sizeof(*desc));
7054 set_desc_limit(desc, var.limit);
7055 set_desc_base(desc, (unsigned long)var.base);
7056 #ifdef CONFIG_X86_64
7058 *base3 = var.base >> 32;
7060 desc->type = var.type;
7062 desc->dpl = var.dpl;
7063 desc->p = var.present;
7064 desc->avl = var.avl;
7072 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7073 struct desc_struct *desc, u32 base3,
7076 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7077 struct kvm_segment var;
7079 var.selector = selector;
7080 var.base = get_desc_base(desc);
7081 #ifdef CONFIG_X86_64
7082 var.base |= ((u64)base3) << 32;
7084 var.limit = get_desc_limit(desc);
7086 var.limit = (var.limit << 12) | 0xfff;
7087 var.type = desc->type;
7088 var.dpl = desc->dpl;
7093 var.avl = desc->avl;
7094 var.present = desc->p;
7095 var.unusable = !var.present;
7098 kvm_set_segment(vcpu, &var, seg);
7102 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7103 u32 msr_index, u64 *pdata)
7105 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7108 r = kvm_get_msr(vcpu, msr_index, pdata);
7110 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7111 /* Bounce to user space */
7112 return X86EMUL_IO_NEEDED;
7118 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7119 u32 msr_index, u64 data)
7121 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7124 r = kvm_set_msr(vcpu, msr_index, data);
7126 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7127 /* Bounce to user space */
7128 return X86EMUL_IO_NEEDED;
7134 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7136 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7138 return vcpu->arch.smbase;
7141 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7143 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7145 vcpu->arch.smbase = smbase;
7148 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7151 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7154 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7155 u32 pmc, u64 *pdata)
7157 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7160 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7162 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7165 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7166 struct x86_instruction_info *info,
7167 enum x86_intercept_stage stage)
7169 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7173 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7174 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7177 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7180 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7182 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7185 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7187 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7190 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7192 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7195 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7197 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7200 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7202 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7205 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7207 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7210 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7212 return emul_to_vcpu(ctxt)->arch.hflags;
7215 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7217 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7219 kvm_smm_changed(vcpu, false);
7222 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7223 const char *smstate)
7225 return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7228 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7230 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7233 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7235 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7238 static const struct x86_emulate_ops emulate_ops = {
7239 .read_gpr = emulator_read_gpr,
7240 .write_gpr = emulator_write_gpr,
7241 .read_std = emulator_read_std,
7242 .write_std = emulator_write_std,
7243 .read_phys = kvm_read_guest_phys_system,
7244 .fetch = kvm_fetch_guest_virt,
7245 .read_emulated = emulator_read_emulated,
7246 .write_emulated = emulator_write_emulated,
7247 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7248 .invlpg = emulator_invlpg,
7249 .pio_in_emulated = emulator_pio_in_emulated,
7250 .pio_out_emulated = emulator_pio_out_emulated,
7251 .get_segment = emulator_get_segment,
7252 .set_segment = emulator_set_segment,
7253 .get_cached_segment_base = emulator_get_cached_segment_base,
7254 .get_gdt = emulator_get_gdt,
7255 .get_idt = emulator_get_idt,
7256 .set_gdt = emulator_set_gdt,
7257 .set_idt = emulator_set_idt,
7258 .get_cr = emulator_get_cr,
7259 .set_cr = emulator_set_cr,
7260 .cpl = emulator_get_cpl,
7261 .get_dr = emulator_get_dr,
7262 .set_dr = emulator_set_dr,
7263 .get_smbase = emulator_get_smbase,
7264 .set_smbase = emulator_set_smbase,
7265 .set_msr = emulator_set_msr,
7266 .get_msr = emulator_get_msr,
7267 .check_pmc = emulator_check_pmc,
7268 .read_pmc = emulator_read_pmc,
7269 .halt = emulator_halt,
7270 .wbinvd = emulator_wbinvd,
7271 .fix_hypercall = emulator_fix_hypercall,
7272 .intercept = emulator_intercept,
7273 .get_cpuid = emulator_get_cpuid,
7274 .guest_has_long_mode = emulator_guest_has_long_mode,
7275 .guest_has_movbe = emulator_guest_has_movbe,
7276 .guest_has_fxsr = emulator_guest_has_fxsr,
7277 .set_nmi_mask = emulator_set_nmi_mask,
7278 .get_hflags = emulator_get_hflags,
7279 .exiting_smm = emulator_exiting_smm,
7280 .leave_smm = emulator_leave_smm,
7281 .triple_fault = emulator_triple_fault,
7282 .set_xcr = emulator_set_xcr,
7285 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7287 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7289 * an sti; sti; sequence only disable interrupts for the first
7290 * instruction. So, if the last instruction, be it emulated or
7291 * not, left the system with the INT_STI flag enabled, it
7292 * means that the last instruction is an sti. We should not
7293 * leave the flag on in this case. The same goes for mov ss
7295 if (int_shadow & mask)
7297 if (unlikely(int_shadow || mask)) {
7298 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7300 kvm_make_request(KVM_REQ_EVENT, vcpu);
7304 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7306 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7307 if (ctxt->exception.vector == PF_VECTOR)
7308 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7310 if (ctxt->exception.error_code_valid)
7311 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7312 ctxt->exception.error_code);
7314 kvm_queue_exception(vcpu, ctxt->exception.vector);
7318 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7320 struct x86_emulate_ctxt *ctxt;
7322 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7324 pr_err("kvm: failed to allocate vcpu's emulator\n");
7329 ctxt->ops = &emulate_ops;
7330 vcpu->arch.emulate_ctxt = ctxt;
7335 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7337 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7340 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7342 ctxt->gpa_available = false;
7343 ctxt->eflags = kvm_get_rflags(vcpu);
7344 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7346 ctxt->eip = kvm_rip_read(vcpu);
7347 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7348 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7349 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7350 cs_db ? X86EMUL_MODE_PROT32 :
7351 X86EMUL_MODE_PROT16;
7352 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7353 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7354 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7356 ctxt->interruptibility = 0;
7357 ctxt->have_exception = false;
7358 ctxt->exception.vector = -1;
7359 ctxt->perm_ok = false;
7361 init_decode_cache(ctxt);
7362 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7365 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7367 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7370 init_emulate_ctxt(vcpu);
7374 ctxt->_eip = ctxt->eip + inc_eip;
7375 ret = emulate_int_real(ctxt, irq);
7377 if (ret != X86EMUL_CONTINUE) {
7378 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7380 ctxt->eip = ctxt->_eip;
7381 kvm_rip_write(vcpu, ctxt->eip);
7382 kvm_set_rflags(vcpu, ctxt->eflags);
7385 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7387 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7389 ++vcpu->stat.insn_emulation_fail;
7390 trace_kvm_emulate_insn_failed(vcpu);
7392 if (emulation_type & EMULTYPE_VMWARE_GP) {
7393 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7397 if (emulation_type & EMULTYPE_SKIP) {
7398 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7399 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7400 vcpu->run->internal.ndata = 0;
7404 kvm_queue_exception(vcpu, UD_VECTOR);
7406 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7407 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7408 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7409 vcpu->run->internal.ndata = 0;
7416 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7417 bool write_fault_to_shadow_pgtable,
7420 gpa_t gpa = cr2_or_gpa;
7423 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7426 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7427 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7430 if (!vcpu->arch.mmu->direct_map) {
7432 * Write permission should be allowed since only
7433 * write access need to be emulated.
7435 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7438 * If the mapping is invalid in guest, let cpu retry
7439 * it to generate fault.
7441 if (gpa == UNMAPPED_GVA)
7446 * Do not retry the unhandleable instruction if it faults on the
7447 * readonly host memory, otherwise it will goto a infinite loop:
7448 * retry instruction -> write #PF -> emulation fail -> retry
7449 * instruction -> ...
7451 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7454 * If the instruction failed on the error pfn, it can not be fixed,
7455 * report the error to userspace.
7457 if (is_error_noslot_pfn(pfn))
7460 kvm_release_pfn_clean(pfn);
7462 /* The instructions are well-emulated on direct mmu. */
7463 if (vcpu->arch.mmu->direct_map) {
7464 unsigned int indirect_shadow_pages;
7466 write_lock(&vcpu->kvm->mmu_lock);
7467 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7468 write_unlock(&vcpu->kvm->mmu_lock);
7470 if (indirect_shadow_pages)
7471 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7477 * if emulation was due to access to shadowed page table
7478 * and it failed try to unshadow page and re-enter the
7479 * guest to let CPU execute the instruction.
7481 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7484 * If the access faults on its page table, it can not
7485 * be fixed by unprotecting shadow page and it should
7486 * be reported to userspace.
7488 return !write_fault_to_shadow_pgtable;
7491 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7492 gpa_t cr2_or_gpa, int emulation_type)
7494 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7495 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7497 last_retry_eip = vcpu->arch.last_retry_eip;
7498 last_retry_addr = vcpu->arch.last_retry_addr;
7501 * If the emulation is caused by #PF and it is non-page_table
7502 * writing instruction, it means the VM-EXIT is caused by shadow
7503 * page protected, we can zap the shadow page and retry this
7504 * instruction directly.
7506 * Note: if the guest uses a non-page-table modifying instruction
7507 * on the PDE that points to the instruction, then we will unmap
7508 * the instruction and go to an infinite loop. So, we cache the
7509 * last retried eip and the last fault address, if we meet the eip
7510 * and the address again, we can break out of the potential infinite
7513 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7515 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7518 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7519 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7522 if (x86_page_table_writing_insn(ctxt))
7525 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7528 vcpu->arch.last_retry_eip = ctxt->eip;
7529 vcpu->arch.last_retry_addr = cr2_or_gpa;
7531 if (!vcpu->arch.mmu->direct_map)
7532 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7534 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7539 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7540 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7542 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7544 trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7547 vcpu->arch.hflags |= HF_SMM_MASK;
7549 vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7551 /* Process a latched INIT or SMI, if any. */
7552 kvm_make_request(KVM_REQ_EVENT, vcpu);
7555 kvm_mmu_reset_context(vcpu);
7558 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7567 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7568 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7573 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7575 struct kvm_run *kvm_run = vcpu->run;
7577 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7578 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7579 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7580 kvm_run->debug.arch.exception = DB_VECTOR;
7581 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7584 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7588 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7590 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7593 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7598 * rflags is the old, "raw" value of the flags. The new value has
7599 * not been saved yet.
7601 * This is correct even for TF set by the guest, because "the
7602 * processor will not generate this exception after the instruction
7603 * that sets the TF flag".
7605 if (unlikely(rflags & X86_EFLAGS_TF))
7606 r = kvm_vcpu_do_singlestep(vcpu);
7609 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7611 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7613 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7614 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7615 struct kvm_run *kvm_run = vcpu->run;
7616 unsigned long eip = kvm_get_linear_rip(vcpu);
7617 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7618 vcpu->arch.guest_debug_dr7,
7622 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7623 kvm_run->debug.arch.pc = eip;
7624 kvm_run->debug.arch.exception = DB_VECTOR;
7625 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7631 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7632 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7633 unsigned long eip = kvm_get_linear_rip(vcpu);
7634 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7639 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7648 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7650 switch (ctxt->opcode_len) {
7657 case 0xe6: /* OUT */
7661 case 0x6c: /* INS */
7663 case 0x6e: /* OUTS */
7670 case 0x33: /* RDPMC */
7680 * Decode to be emulated instruction. Return EMULATION_OK if success.
7682 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7683 void *insn, int insn_len)
7685 int r = EMULATION_OK;
7686 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7688 init_emulate_ctxt(vcpu);
7691 * We will reenter on the same instruction since we do not set
7692 * complete_userspace_io. This does not handle watchpoints yet,
7693 * those would be handled in the emulate_ops.
7695 if (!(emulation_type & EMULTYPE_SKIP) &&
7696 kvm_vcpu_check_breakpoint(vcpu, &r))
7699 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7701 trace_kvm_emulate_insn_start(vcpu);
7702 ++vcpu->stat.insn_emulation;
7706 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7708 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7709 int emulation_type, void *insn, int insn_len)
7712 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7713 bool writeback = true;
7714 bool write_fault_to_spt;
7716 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7719 vcpu->arch.l1tf_flush_l1d = true;
7722 * Clear write_fault_to_shadow_pgtable here to ensure it is
7725 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7726 vcpu->arch.write_fault_to_shadow_pgtable = false;
7728 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7729 kvm_clear_exception_queue(vcpu);
7731 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7733 if (r != EMULATION_OK) {
7734 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7735 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7736 kvm_queue_exception(vcpu, UD_VECTOR);
7739 if (reexecute_instruction(vcpu, cr2_or_gpa,
7743 if (ctxt->have_exception) {
7745 * #UD should result in just EMULATION_FAILED, and trap-like
7746 * exception should not be encountered during decode.
7748 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7749 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7750 inject_emulated_exception(vcpu);
7753 return handle_emulation_failure(vcpu, emulation_type);
7757 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7758 !is_vmware_backdoor_opcode(ctxt)) {
7759 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7764 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7765 * for kvm_skip_emulated_instruction(). The caller is responsible for
7766 * updating interruptibility state and injecting single-step #DBs.
7768 if (emulation_type & EMULTYPE_SKIP) {
7769 kvm_rip_write(vcpu, ctxt->_eip);
7770 if (ctxt->eflags & X86_EFLAGS_RF)
7771 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7775 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7778 /* this is needed for vmware backdoor interface to work since it
7779 changes registers values during IO operation */
7780 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7781 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7782 emulator_invalidate_register_cache(ctxt);
7786 if (emulation_type & EMULTYPE_PF) {
7787 /* Save the faulting GPA (cr2) in the address field */
7788 ctxt->exception.address = cr2_or_gpa;
7790 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7791 if (vcpu->arch.mmu->direct_map) {
7792 ctxt->gpa_available = true;
7793 ctxt->gpa_val = cr2_or_gpa;
7796 /* Sanitize the address out of an abundance of paranoia. */
7797 ctxt->exception.address = 0;
7800 r = x86_emulate_insn(ctxt);
7802 if (r == EMULATION_INTERCEPTED)
7805 if (r == EMULATION_FAILED) {
7806 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7810 return handle_emulation_failure(vcpu, emulation_type);
7813 if (ctxt->have_exception) {
7815 if (inject_emulated_exception(vcpu))
7817 } else if (vcpu->arch.pio.count) {
7818 if (!vcpu->arch.pio.in) {
7819 /* FIXME: return into emulator if single-stepping. */
7820 vcpu->arch.pio.count = 0;
7823 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7826 } else if (vcpu->mmio_needed) {
7827 ++vcpu->stat.mmio_exits;
7829 if (!vcpu->mmio_is_write)
7832 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7833 } else if (r == EMULATION_RESTART)
7839 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7840 toggle_interruptibility(vcpu, ctxt->interruptibility);
7841 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7842 if (!ctxt->have_exception ||
7843 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7844 kvm_rip_write(vcpu, ctxt->eip);
7845 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7846 r = kvm_vcpu_do_singlestep(vcpu);
7847 if (kvm_x86_ops.update_emulated_instruction)
7848 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7849 __kvm_set_rflags(vcpu, ctxt->eflags);
7853 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7854 * do nothing, and it will be requested again as soon as
7855 * the shadow expires. But we still need to check here,
7856 * because POPF has no interrupt shadow.
7858 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7859 kvm_make_request(KVM_REQ_EVENT, vcpu);
7861 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7866 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7868 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7870 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7872 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7873 void *insn, int insn_len)
7875 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7877 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7879 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7881 vcpu->arch.pio.count = 0;
7885 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7887 vcpu->arch.pio.count = 0;
7889 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7892 return kvm_skip_emulated_instruction(vcpu);
7895 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7896 unsigned short port)
7898 unsigned long val = kvm_rax_read(vcpu);
7899 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7905 * Workaround userspace that relies on old KVM behavior of %rip being
7906 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7909 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7910 vcpu->arch.complete_userspace_io =
7911 complete_fast_pio_out_port_0x7e;
7912 kvm_skip_emulated_instruction(vcpu);
7914 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7915 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7920 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7924 /* We should only ever be called with arch.pio.count equal to 1 */
7925 BUG_ON(vcpu->arch.pio.count != 1);
7927 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7928 vcpu->arch.pio.count = 0;
7932 /* For size less than 4 we merge, else we zero extend */
7933 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7936 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7937 * the copy and tracing
7939 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7940 kvm_rax_write(vcpu, val);
7942 return kvm_skip_emulated_instruction(vcpu);
7945 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7946 unsigned short port)
7951 /* For size less than 4 we merge, else we zero extend */
7952 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7954 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7956 kvm_rax_write(vcpu, val);
7960 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7961 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7966 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7971 ret = kvm_fast_pio_in(vcpu, size, port);
7973 ret = kvm_fast_pio_out(vcpu, size, port);
7974 return ret && kvm_skip_emulated_instruction(vcpu);
7976 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7978 static int kvmclock_cpu_down_prep(unsigned int cpu)
7980 __this_cpu_write(cpu_tsc_khz, 0);
7984 static void tsc_khz_changed(void *data)
7986 struct cpufreq_freqs *freq = data;
7987 unsigned long khz = 0;
7991 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7992 khz = cpufreq_quick_get(raw_smp_processor_id());
7995 __this_cpu_write(cpu_tsc_khz, khz);
7998 #ifdef CONFIG_X86_64
7999 static void kvm_hyperv_tsc_notifier(void)
8002 struct kvm_vcpu *vcpu;
8004 unsigned long flags;
8006 mutex_lock(&kvm_lock);
8007 list_for_each_entry(kvm, &vm_list, vm_list)
8008 kvm_make_mclock_inprogress_request(kvm);
8010 hyperv_stop_tsc_emulation();
8012 /* TSC frequency always matches when on Hyper-V */
8013 for_each_present_cpu(cpu)
8014 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8015 kvm_max_guest_tsc_khz = tsc_khz;
8017 list_for_each_entry(kvm, &vm_list, vm_list) {
8018 struct kvm_arch *ka = &kvm->arch;
8020 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
8021 pvclock_update_vm_gtod_copy(kvm);
8022 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
8024 kvm_for_each_vcpu(cpu, vcpu, kvm)
8025 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8027 kvm_for_each_vcpu(cpu, vcpu, kvm)
8028 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
8030 mutex_unlock(&kvm_lock);
8034 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8037 struct kvm_vcpu *vcpu;
8038 int i, send_ipi = 0;
8041 * We allow guests to temporarily run on slowing clocks,
8042 * provided we notify them after, or to run on accelerating
8043 * clocks, provided we notify them before. Thus time never
8046 * However, we have a problem. We can't atomically update
8047 * the frequency of a given CPU from this function; it is
8048 * merely a notifier, which can be called from any CPU.
8049 * Changing the TSC frequency at arbitrary points in time
8050 * requires a recomputation of local variables related to
8051 * the TSC for each VCPU. We must flag these local variables
8052 * to be updated and be sure the update takes place with the
8053 * new frequency before any guests proceed.
8055 * Unfortunately, the combination of hotplug CPU and frequency
8056 * change creates an intractable locking scenario; the order
8057 * of when these callouts happen is undefined with respect to
8058 * CPU hotplug, and they can race with each other. As such,
8059 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8060 * undefined; you can actually have a CPU frequency change take
8061 * place in between the computation of X and the setting of the
8062 * variable. To protect against this problem, all updates of
8063 * the per_cpu tsc_khz variable are done in an interrupt
8064 * protected IPI, and all callers wishing to update the value
8065 * must wait for a synchronous IPI to complete (which is trivial
8066 * if the caller is on the CPU already). This establishes the
8067 * necessary total order on variable updates.
8069 * Note that because a guest time update may take place
8070 * anytime after the setting of the VCPU's request bit, the
8071 * correct TSC value must be set before the request. However,
8072 * to ensure the update actually makes it to any guest which
8073 * starts running in hardware virtualization between the set
8074 * and the acquisition of the spinlock, we must also ping the
8075 * CPU after setting the request bit.
8079 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8081 mutex_lock(&kvm_lock);
8082 list_for_each_entry(kvm, &vm_list, vm_list) {
8083 kvm_for_each_vcpu(i, vcpu, kvm) {
8084 if (vcpu->cpu != cpu)
8086 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8087 if (vcpu->cpu != raw_smp_processor_id())
8091 mutex_unlock(&kvm_lock);
8093 if (freq->old < freq->new && send_ipi) {
8095 * We upscale the frequency. Must make the guest
8096 * doesn't see old kvmclock values while running with
8097 * the new frequency, otherwise we risk the guest sees
8098 * time go backwards.
8100 * In case we update the frequency for another cpu
8101 * (which might be in guest context) send an interrupt
8102 * to kick the cpu out of guest context. Next time
8103 * guest context is entered kvmclock will be updated,
8104 * so the guest will not see stale values.
8106 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8110 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8113 struct cpufreq_freqs *freq = data;
8116 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8118 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8121 for_each_cpu(cpu, freq->policy->cpus)
8122 __kvmclock_cpufreq_notifier(freq, cpu);
8127 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8128 .notifier_call = kvmclock_cpufreq_notifier
8131 static int kvmclock_cpu_online(unsigned int cpu)
8133 tsc_khz_changed(NULL);
8137 static void kvm_timer_init(void)
8139 max_tsc_khz = tsc_khz;
8141 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8142 #ifdef CONFIG_CPU_FREQ
8143 struct cpufreq_policy *policy;
8147 policy = cpufreq_cpu_get(cpu);
8149 if (policy->cpuinfo.max_freq)
8150 max_tsc_khz = policy->cpuinfo.max_freq;
8151 cpufreq_cpu_put(policy);
8155 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8156 CPUFREQ_TRANSITION_NOTIFIER);
8159 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8160 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8163 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8164 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8166 int kvm_is_in_guest(void)
8168 return __this_cpu_read(current_vcpu) != NULL;
8171 static int kvm_is_user_mode(void)
8175 if (__this_cpu_read(current_vcpu))
8176 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8178 return user_mode != 0;
8181 static unsigned long kvm_get_guest_ip(void)
8183 unsigned long ip = 0;
8185 if (__this_cpu_read(current_vcpu))
8186 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8191 static void kvm_handle_intel_pt_intr(void)
8193 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8195 kvm_make_request(KVM_REQ_PMI, vcpu);
8196 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8197 (unsigned long *)&vcpu->arch.pmu.global_status);
8200 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8201 .is_in_guest = kvm_is_in_guest,
8202 .is_user_mode = kvm_is_user_mode,
8203 .get_guest_ip = kvm_get_guest_ip,
8204 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8207 #ifdef CONFIG_X86_64
8208 static void pvclock_gtod_update_fn(struct work_struct *work)
8212 struct kvm_vcpu *vcpu;
8215 mutex_lock(&kvm_lock);
8216 list_for_each_entry(kvm, &vm_list, vm_list)
8217 kvm_for_each_vcpu(i, vcpu, kvm)
8218 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8219 atomic_set(&kvm_guest_has_master_clock, 0);
8220 mutex_unlock(&kvm_lock);
8223 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8226 * Indirection to move queue_work() out of the tk_core.seq write held
8227 * region to prevent possible deadlocks against time accessors which
8228 * are invoked with work related locks held.
8230 static void pvclock_irq_work_fn(struct irq_work *w)
8232 queue_work(system_long_wq, &pvclock_gtod_work);
8235 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8238 * Notification about pvclock gtod data update.
8240 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8243 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8244 struct timekeeper *tk = priv;
8246 update_pvclock_gtod(tk);
8249 * Disable master clock if host does not trust, or does not use,
8250 * TSC based clocksource. Delegate queue_work() to irq_work as
8251 * this is invoked with tk_core.seq write held.
8253 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8254 atomic_read(&kvm_guest_has_master_clock) != 0)
8255 irq_work_queue(&pvclock_irq_work);
8259 static struct notifier_block pvclock_gtod_notifier = {
8260 .notifier_call = pvclock_gtod_notify,
8264 int kvm_arch_init(void *opaque)
8266 struct kvm_x86_init_ops *ops = opaque;
8269 if (kvm_x86_ops.hardware_enable) {
8270 printk(KERN_ERR "kvm: already loaded the other module\n");
8275 if (!ops->cpu_has_kvm_support()) {
8276 pr_err_ratelimited("kvm: no hardware support\n");
8280 if (ops->disabled_by_bios()) {
8281 pr_err_ratelimited("kvm: disabled by bios\n");
8287 * KVM explicitly assumes that the guest has an FPU and
8288 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8289 * vCPU's FPU state as a fxregs_state struct.
8291 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8292 printk(KERN_ERR "kvm: inadequate fpu\n");
8298 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8299 __alignof__(struct fpu), SLAB_ACCOUNT,
8301 if (!x86_fpu_cache) {
8302 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8306 x86_emulator_cache = kvm_alloc_emulator_cache();
8307 if (!x86_emulator_cache) {
8308 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8309 goto out_free_x86_fpu_cache;
8312 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8313 if (!user_return_msrs) {
8314 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8315 goto out_free_x86_emulator_cache;
8317 kvm_nr_uret_msrs = 0;
8319 r = kvm_mmu_module_init();
8321 goto out_free_percpu;
8325 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8327 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8328 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8329 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8332 if (pi_inject_timer == -1)
8333 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8334 #ifdef CONFIG_X86_64
8335 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8337 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8338 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8344 free_percpu(user_return_msrs);
8345 out_free_x86_emulator_cache:
8346 kmem_cache_destroy(x86_emulator_cache);
8347 out_free_x86_fpu_cache:
8348 kmem_cache_destroy(x86_fpu_cache);
8353 void kvm_arch_exit(void)
8355 #ifdef CONFIG_X86_64
8356 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8357 clear_hv_tscchange_cb();
8360 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8362 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8363 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8364 CPUFREQ_TRANSITION_NOTIFIER);
8365 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8366 #ifdef CONFIG_X86_64
8367 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8368 irq_work_sync(&pvclock_irq_work);
8369 cancel_work_sync(&pvclock_gtod_work);
8371 kvm_x86_ops.hardware_enable = NULL;
8372 kvm_mmu_module_exit();
8373 free_percpu(user_return_msrs);
8374 kmem_cache_destroy(x86_emulator_cache);
8375 kmem_cache_destroy(x86_fpu_cache);
8376 #ifdef CONFIG_KVM_XEN
8377 static_key_deferred_flush(&kvm_xen_enabled);
8378 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8382 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8384 ++vcpu->stat.halt_exits;
8385 if (lapic_in_kernel(vcpu)) {
8386 vcpu->arch.mp_state = state;
8389 vcpu->run->exit_reason = reason;
8394 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8396 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8398 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8400 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8402 int ret = kvm_skip_emulated_instruction(vcpu);
8404 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8405 * KVM_EXIT_DEBUG here.
8407 return kvm_vcpu_halt(vcpu) && ret;
8409 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8411 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8413 int ret = kvm_skip_emulated_instruction(vcpu);
8415 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8417 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8419 #ifdef CONFIG_X86_64
8420 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8421 unsigned long clock_type)
8423 struct kvm_clock_pairing clock_pairing;
8424 struct timespec64 ts;
8428 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8429 return -KVM_EOPNOTSUPP;
8431 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8432 return -KVM_EOPNOTSUPP;
8434 clock_pairing.sec = ts.tv_sec;
8435 clock_pairing.nsec = ts.tv_nsec;
8436 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8437 clock_pairing.flags = 0;
8438 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8441 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8442 sizeof(struct kvm_clock_pairing)))
8450 * kvm_pv_kick_cpu_op: Kick a vcpu.
8452 * @apicid - apicid of vcpu to be kicked.
8454 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8456 struct kvm_lapic_irq lapic_irq;
8458 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8459 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8460 lapic_irq.level = 0;
8461 lapic_irq.dest_id = apicid;
8462 lapic_irq.msi_redir_hint = false;
8464 lapic_irq.delivery_mode = APIC_DM_REMRD;
8465 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8468 bool kvm_apicv_activated(struct kvm *kvm)
8470 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8472 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8474 static void kvm_apicv_init(struct kvm *kvm)
8477 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8478 &kvm->arch.apicv_inhibit_reasons);
8480 set_bit(APICV_INHIBIT_REASON_DISABLE,
8481 &kvm->arch.apicv_inhibit_reasons);
8484 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8486 struct kvm_vcpu *target = NULL;
8487 struct kvm_apic_map *map;
8489 vcpu->stat.directed_yield_attempted++;
8491 if (single_task_running())
8495 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8497 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8498 target = map->phys_map[dest_id]->vcpu;
8502 if (!target || !READ_ONCE(target->ready))
8505 /* Ignore requests to yield to self */
8509 if (kvm_vcpu_yield_to(target) <= 0)
8512 vcpu->stat.directed_yield_successful++;
8518 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8520 unsigned long nr, a0, a1, a2, a3, ret;
8523 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8524 return kvm_xen_hypercall(vcpu);
8526 if (kvm_hv_hypercall_enabled(vcpu))
8527 return kvm_hv_hypercall(vcpu);
8529 nr = kvm_rax_read(vcpu);
8530 a0 = kvm_rbx_read(vcpu);
8531 a1 = kvm_rcx_read(vcpu);
8532 a2 = kvm_rdx_read(vcpu);
8533 a3 = kvm_rsi_read(vcpu);
8535 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8537 op_64_bit = is_64_bit_mode(vcpu);
8546 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8554 case KVM_HC_VAPIC_POLL_IRQ:
8557 case KVM_HC_KICK_CPU:
8558 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8561 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8562 kvm_sched_yield(vcpu, a1);
8565 #ifdef CONFIG_X86_64
8566 case KVM_HC_CLOCK_PAIRING:
8567 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8570 case KVM_HC_SEND_IPI:
8571 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8574 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8576 case KVM_HC_SCHED_YIELD:
8577 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8580 kvm_sched_yield(vcpu, a0);
8590 kvm_rax_write(vcpu, ret);
8592 ++vcpu->stat.hypercalls;
8593 return kvm_skip_emulated_instruction(vcpu);
8595 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8597 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8599 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8600 char instruction[3];
8601 unsigned long rip = kvm_rip_read(vcpu);
8603 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8605 return emulator_write_emulated(ctxt, rip, instruction, 3,
8609 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8611 return vcpu->run->request_interrupt_window &&
8612 likely(!pic_in_kernel(vcpu->kvm));
8615 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8617 struct kvm_run *kvm_run = vcpu->run;
8620 * if_flag is obsolete and useless, so do not bother
8621 * setting it for SEV-ES guests. Userspace can just
8622 * use kvm_run->ready_for_interrupt_injection.
8624 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8625 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8627 kvm_run->cr8 = kvm_get_cr8(vcpu);
8628 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8629 kvm_run->ready_for_interrupt_injection =
8630 pic_in_kernel(vcpu->kvm) ||
8631 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8634 kvm_run->flags |= KVM_RUN_X86_SMM;
8637 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8641 if (!kvm_x86_ops.update_cr8_intercept)
8644 if (!lapic_in_kernel(vcpu))
8647 if (vcpu->arch.apicv_active)
8650 if (!vcpu->arch.apic->vapic_addr)
8651 max_irr = kvm_lapic_find_highest_irr(vcpu);
8658 tpr = kvm_lapic_get_cr8(vcpu);
8660 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8664 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8666 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8667 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8671 return kvm_x86_ops.nested_ops->check_events(vcpu);
8674 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8676 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8677 vcpu->arch.exception.error_code = false;
8678 static_call(kvm_x86_queue_exception)(vcpu);
8681 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8684 bool can_inject = true;
8686 /* try to reinject previous events if any */
8688 if (vcpu->arch.exception.injected) {
8689 kvm_inject_exception(vcpu);
8693 * Do not inject an NMI or interrupt if there is a pending
8694 * exception. Exceptions and interrupts are recognized at
8695 * instruction boundaries, i.e. the start of an instruction.
8696 * Trap-like exceptions, e.g. #DB, have higher priority than
8697 * NMIs and interrupts, i.e. traps are recognized before an
8698 * NMI/interrupt that's pending on the same instruction.
8699 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8700 * priority, but are only generated (pended) during instruction
8701 * execution, i.e. a pending fault-like exception means the
8702 * fault occurred on the *previous* instruction and must be
8703 * serviced prior to recognizing any new events in order to
8704 * fully complete the previous instruction.
8706 else if (!vcpu->arch.exception.pending) {
8707 if (vcpu->arch.nmi_injected) {
8708 static_call(kvm_x86_set_nmi)(vcpu);
8710 } else if (vcpu->arch.interrupt.injected) {
8711 static_call(kvm_x86_set_irq)(vcpu);
8716 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8717 vcpu->arch.exception.pending);
8720 * Call check_nested_events() even if we reinjected a previous event
8721 * in order for caller to determine if it should require immediate-exit
8722 * from L2 to L1 due to pending L1 events which require exit
8725 if (is_guest_mode(vcpu)) {
8726 r = kvm_check_nested_events(vcpu);
8731 /* try to inject new event if pending */
8732 if (vcpu->arch.exception.pending) {
8733 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8734 vcpu->arch.exception.has_error_code,
8735 vcpu->arch.exception.error_code);
8737 vcpu->arch.exception.pending = false;
8738 vcpu->arch.exception.injected = true;
8740 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8741 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8744 if (vcpu->arch.exception.nr == DB_VECTOR) {
8745 kvm_deliver_exception_payload(vcpu);
8746 if (vcpu->arch.dr7 & DR7_GD) {
8747 vcpu->arch.dr7 &= ~DR7_GD;
8748 kvm_update_dr7(vcpu);
8752 kvm_inject_exception(vcpu);
8757 * Finally, inject interrupt events. If an event cannot be injected
8758 * due to architectural conditions (e.g. IF=0) a window-open exit
8759 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8760 * and can architecturally be injected, but we cannot do it right now:
8761 * an interrupt could have arrived just now and we have to inject it
8762 * as a vmexit, or there could already an event in the queue, which is
8763 * indicated by can_inject. In that case we request an immediate exit
8764 * in order to make progress and get back here for another iteration.
8765 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8767 if (vcpu->arch.smi_pending) {
8768 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8772 vcpu->arch.smi_pending = false;
8773 ++vcpu->arch.smi_count;
8777 static_call(kvm_x86_enable_smi_window)(vcpu);
8780 if (vcpu->arch.nmi_pending) {
8781 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8785 --vcpu->arch.nmi_pending;
8786 vcpu->arch.nmi_injected = true;
8787 static_call(kvm_x86_set_nmi)(vcpu);
8789 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8791 if (vcpu->arch.nmi_pending)
8792 static_call(kvm_x86_enable_nmi_window)(vcpu);
8795 if (kvm_cpu_has_injectable_intr(vcpu)) {
8796 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8800 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8801 static_call(kvm_x86_set_irq)(vcpu);
8802 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8804 if (kvm_cpu_has_injectable_intr(vcpu))
8805 static_call(kvm_x86_enable_irq_window)(vcpu);
8808 if (is_guest_mode(vcpu) &&
8809 kvm_x86_ops.nested_ops->hv_timer_pending &&
8810 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8811 *req_immediate_exit = true;
8813 WARN_ON(vcpu->arch.exception.pending);
8818 *req_immediate_exit = true;
8824 static void process_nmi(struct kvm_vcpu *vcpu)
8829 * x86 is limited to one NMI running, and one NMI pending after it.
8830 * If an NMI is already in progress, limit further NMIs to just one.
8831 * Otherwise, allow two (and we'll inject the first one immediately).
8833 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8836 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8837 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8838 kvm_make_request(KVM_REQ_EVENT, vcpu);
8841 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8844 flags |= seg->g << 23;
8845 flags |= seg->db << 22;
8846 flags |= seg->l << 21;
8847 flags |= seg->avl << 20;
8848 flags |= seg->present << 15;
8849 flags |= seg->dpl << 13;
8850 flags |= seg->s << 12;
8851 flags |= seg->type << 8;
8855 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8857 struct kvm_segment seg;
8860 kvm_get_segment(vcpu, &seg, n);
8861 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8864 offset = 0x7f84 + n * 12;
8866 offset = 0x7f2c + (n - 3) * 12;
8868 put_smstate(u32, buf, offset + 8, seg.base);
8869 put_smstate(u32, buf, offset + 4, seg.limit);
8870 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8873 #ifdef CONFIG_X86_64
8874 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8876 struct kvm_segment seg;
8880 kvm_get_segment(vcpu, &seg, n);
8881 offset = 0x7e00 + n * 16;
8883 flags = enter_smm_get_segment_flags(&seg) >> 8;
8884 put_smstate(u16, buf, offset, seg.selector);
8885 put_smstate(u16, buf, offset + 2, flags);
8886 put_smstate(u32, buf, offset + 4, seg.limit);
8887 put_smstate(u64, buf, offset + 8, seg.base);
8891 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8894 struct kvm_segment seg;
8898 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8899 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8900 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8901 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8903 for (i = 0; i < 8; i++)
8904 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
8906 kvm_get_dr(vcpu, 6, &val);
8907 put_smstate(u32, buf, 0x7fcc, (u32)val);
8908 kvm_get_dr(vcpu, 7, &val);
8909 put_smstate(u32, buf, 0x7fc8, (u32)val);
8911 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8912 put_smstate(u32, buf, 0x7fc4, seg.selector);
8913 put_smstate(u32, buf, 0x7f64, seg.base);
8914 put_smstate(u32, buf, 0x7f60, seg.limit);
8915 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8917 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8918 put_smstate(u32, buf, 0x7fc0, seg.selector);
8919 put_smstate(u32, buf, 0x7f80, seg.base);
8920 put_smstate(u32, buf, 0x7f7c, seg.limit);
8921 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8923 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8924 put_smstate(u32, buf, 0x7f74, dt.address);
8925 put_smstate(u32, buf, 0x7f70, dt.size);
8927 static_call(kvm_x86_get_idt)(vcpu, &dt);
8928 put_smstate(u32, buf, 0x7f58, dt.address);
8929 put_smstate(u32, buf, 0x7f54, dt.size);
8931 for (i = 0; i < 6; i++)
8932 enter_smm_save_seg_32(vcpu, buf, i);
8934 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8937 put_smstate(u32, buf, 0x7efc, 0x00020000);
8938 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8941 #ifdef CONFIG_X86_64
8942 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8945 struct kvm_segment seg;
8949 for (i = 0; i < 16; i++)
8950 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
8952 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8953 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8955 kvm_get_dr(vcpu, 6, &val);
8956 put_smstate(u64, buf, 0x7f68, val);
8957 kvm_get_dr(vcpu, 7, &val);
8958 put_smstate(u64, buf, 0x7f60, val);
8960 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8961 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8962 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8964 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8967 put_smstate(u32, buf, 0x7efc, 0x00020064);
8969 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8971 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8972 put_smstate(u16, buf, 0x7e90, seg.selector);
8973 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8974 put_smstate(u32, buf, 0x7e94, seg.limit);
8975 put_smstate(u64, buf, 0x7e98, seg.base);
8977 static_call(kvm_x86_get_idt)(vcpu, &dt);
8978 put_smstate(u32, buf, 0x7e84, dt.size);
8979 put_smstate(u64, buf, 0x7e88, dt.address);
8981 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8982 put_smstate(u16, buf, 0x7e70, seg.selector);
8983 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8984 put_smstate(u32, buf, 0x7e74, seg.limit);
8985 put_smstate(u64, buf, 0x7e78, seg.base);
8987 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8988 put_smstate(u32, buf, 0x7e64, dt.size);
8989 put_smstate(u64, buf, 0x7e68, dt.address);
8991 for (i = 0; i < 6; i++)
8992 enter_smm_save_seg_64(vcpu, buf, i);
8996 static void enter_smm(struct kvm_vcpu *vcpu)
8998 struct kvm_segment cs, ds;
9003 memset(buf, 0, 512);
9004 #ifdef CONFIG_X86_64
9005 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9006 enter_smm_save_state_64(vcpu, buf);
9009 enter_smm_save_state_32(vcpu, buf);
9012 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9013 * state (e.g. leave guest mode) after we've saved the state into the
9014 * SMM state-save area.
9016 static_call(kvm_x86_enter_smm)(vcpu, buf);
9018 kvm_smm_changed(vcpu, true);
9019 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9021 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9022 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9024 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9026 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9027 kvm_rip_write(vcpu, 0x8000);
9029 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9030 static_call(kvm_x86_set_cr0)(vcpu, cr0);
9031 vcpu->arch.cr0 = cr0;
9033 static_call(kvm_x86_set_cr4)(vcpu, 0);
9035 /* Undocumented: IDT limit is set to zero on entry to SMM. */
9036 dt.address = dt.size = 0;
9037 static_call(kvm_x86_set_idt)(vcpu, &dt);
9039 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9041 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9042 cs.base = vcpu->arch.smbase;
9047 cs.limit = ds.limit = 0xffffffff;
9048 cs.type = ds.type = 0x3;
9049 cs.dpl = ds.dpl = 0;
9054 cs.avl = ds.avl = 0;
9055 cs.present = ds.present = 1;
9056 cs.unusable = ds.unusable = 0;
9057 cs.padding = ds.padding = 0;
9059 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9060 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9061 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9062 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9063 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9064 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9066 #ifdef CONFIG_X86_64
9067 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9068 static_call(kvm_x86_set_efer)(vcpu, 0);
9071 kvm_update_cpuid_runtime(vcpu);
9072 kvm_mmu_reset_context(vcpu);
9075 static void process_smi(struct kvm_vcpu *vcpu)
9077 vcpu->arch.smi_pending = true;
9078 kvm_make_request(KVM_REQ_EVENT, vcpu);
9081 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9082 unsigned long *vcpu_bitmap)
9086 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
9088 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
9089 NULL, vcpu_bitmap, cpus);
9091 free_cpumask_var(cpus);
9094 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9096 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9099 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9101 if (!lapic_in_kernel(vcpu))
9104 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
9105 kvm_apic_update_apicv(vcpu);
9106 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9108 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9111 * NOTE: Do not hold any lock prior to calling this.
9113 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
9114 * locked, because it calls __x86_set_memory_region() which does
9115 * synchronize_srcu(&kvm->srcu).
9117 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9119 struct kvm_vcpu *except;
9120 unsigned long old, new, expected;
9122 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9123 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9126 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
9128 expected = new = old;
9130 __clear_bit(bit, &new);
9132 __set_bit(bit, &new);
9135 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
9136 } while (old != expected);
9141 trace_kvm_apicv_update_request(activate, bit);
9142 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
9143 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
9146 * Sending request to update APICV for all other vcpus,
9147 * while update the calling vcpu immediately instead of
9148 * waiting for another #VMEXIT to handle the request.
9150 except = kvm_get_running_vcpu();
9151 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
9154 kvm_vcpu_update_apicv(except);
9156 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9158 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9160 if (!kvm_apic_present(vcpu))
9163 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9165 if (irqchip_split(vcpu->kvm))
9166 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9168 if (vcpu->arch.apicv_active)
9169 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9170 if (ioapic_in_kernel(vcpu->kvm))
9171 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9174 if (is_guest_mode(vcpu))
9175 vcpu->arch.load_eoi_exitmap_pending = true;
9177 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9180 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9182 u64 eoi_exit_bitmap[4];
9184 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9187 if (to_hv_vcpu(vcpu))
9188 bitmap_or((ulong *)eoi_exit_bitmap,
9189 vcpu->arch.ioapic_handled_vectors,
9190 to_hv_synic(vcpu)->vec_bitmap, 256);
9192 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9195 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9196 unsigned long start, unsigned long end)
9198 unsigned long apic_address;
9201 * The physical address of apic access page is stored in the VMCS.
9202 * Update it when it becomes invalid.
9204 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9205 if (start <= apic_address && apic_address < end)
9206 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9209 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9211 if (!lapic_in_kernel(vcpu))
9214 if (!kvm_x86_ops.set_apic_access_page_addr)
9217 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9220 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9222 smp_send_reschedule(vcpu->cpu);
9224 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9227 * Returns 1 to let vcpu_run() continue the guest execution loop without
9228 * exiting to the userspace. Otherwise, the value will be returned to the
9231 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9235 dm_request_for_irq_injection(vcpu) &&
9236 kvm_cpu_accept_dm_intr(vcpu);
9237 fastpath_t exit_fastpath;
9239 bool req_immediate_exit = false;
9241 /* Forbid vmenter if vcpu dirty ring is soft-full */
9242 if (unlikely(vcpu->kvm->dirty_ring_size &&
9243 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9244 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9245 trace_kvm_dirty_ring_exit(vcpu);
9250 if (kvm_request_pending(vcpu)) {
9251 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9252 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9257 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9258 kvm_mmu_unload(vcpu);
9259 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9260 __kvm_migrate_timers(vcpu);
9261 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9262 kvm_gen_update_masterclock(vcpu->kvm);
9263 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9264 kvm_gen_kvmclock_update(vcpu);
9265 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9266 r = kvm_guest_time_update(vcpu);
9270 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9271 kvm_mmu_sync_roots(vcpu);
9272 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9273 kvm_mmu_load_pgd(vcpu);
9274 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9275 kvm_vcpu_flush_tlb_all(vcpu);
9277 /* Flushing all ASIDs flushes the current ASID... */
9278 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9280 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9281 kvm_vcpu_flush_tlb_current(vcpu);
9282 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
9283 kvm_vcpu_flush_tlb_guest(vcpu);
9285 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9286 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9290 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9291 if (is_guest_mode(vcpu)) {
9292 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9294 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9295 vcpu->mmio_needed = 0;
9300 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9301 /* Page is swapped out. Do synthetic halt */
9302 vcpu->arch.apf.halted = true;
9306 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9307 record_steal_time(vcpu);
9308 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9310 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9312 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9313 kvm_pmu_handle_event(vcpu);
9314 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9315 kvm_pmu_deliver_pmi(vcpu);
9316 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9317 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9318 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9319 vcpu->arch.ioapic_handled_vectors)) {
9320 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9321 vcpu->run->eoi.vector =
9322 vcpu->arch.pending_ioapic_eoi;
9327 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9328 vcpu_scan_ioapic(vcpu);
9329 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9330 vcpu_load_eoi_exitmap(vcpu);
9331 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9332 kvm_vcpu_reload_apic_access_page(vcpu);
9333 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9334 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9335 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9339 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9340 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9341 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9345 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9346 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9348 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9349 vcpu->run->hyperv = hv_vcpu->exit;
9355 * KVM_REQ_HV_STIMER has to be processed after
9356 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9357 * depend on the guest clock being up-to-date
9359 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9360 kvm_hv_process_stimers(vcpu);
9361 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9362 kvm_vcpu_update_apicv(vcpu);
9363 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9364 kvm_check_async_pf_completion(vcpu);
9365 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9366 static_call(kvm_x86_msr_filter_changed)(vcpu);
9368 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9369 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9372 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9373 kvm_xen_has_interrupt(vcpu)) {
9374 ++vcpu->stat.req_event;
9375 r = kvm_apic_accept_events(vcpu);
9380 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9385 r = inject_pending_event(vcpu, &req_immediate_exit);
9391 static_call(kvm_x86_enable_irq_window)(vcpu);
9393 if (kvm_lapic_enabled(vcpu)) {
9394 update_cr8_intercept(vcpu);
9395 kvm_lapic_sync_to_vapic(vcpu);
9399 r = kvm_mmu_reload(vcpu);
9401 goto cancel_injection;
9406 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9409 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9410 * IPI are then delayed after guest entry, which ensures that they
9411 * result in virtual interrupt delivery.
9413 local_irq_disable();
9414 vcpu->mode = IN_GUEST_MODE;
9416 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9419 * 1) We should set ->mode before checking ->requests. Please see
9420 * the comment in kvm_vcpu_exiting_guest_mode().
9422 * 2) For APICv, we should set ->mode before checking PID.ON. This
9423 * pairs with the memory barrier implicit in pi_test_and_set_on
9424 * (see vmx_deliver_posted_interrupt).
9426 * 3) This also orders the write to mode from any reads to the page
9427 * tables done while the VCPU is running. Please see the comment
9428 * in kvm_flush_remote_tlbs.
9430 smp_mb__after_srcu_read_unlock();
9433 * This handles the case where a posted interrupt was
9434 * notified with kvm_vcpu_kick.
9436 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9437 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9439 if (kvm_vcpu_exit_request(vcpu)) {
9440 vcpu->mode = OUTSIDE_GUEST_MODE;
9444 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9446 goto cancel_injection;
9449 if (req_immediate_exit) {
9450 kvm_make_request(KVM_REQ_EVENT, vcpu);
9451 static_call(kvm_x86_request_immediate_exit)(vcpu);
9454 fpregs_assert_state_consistent();
9455 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9456 switch_fpu_return();
9458 if (unlikely(vcpu->arch.switch_db_regs)) {
9460 set_debugreg(vcpu->arch.eff_db[0], 0);
9461 set_debugreg(vcpu->arch.eff_db[1], 1);
9462 set_debugreg(vcpu->arch.eff_db[2], 2);
9463 set_debugreg(vcpu->arch.eff_db[3], 3);
9464 set_debugreg(vcpu->arch.dr6, 6);
9465 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9469 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9470 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9473 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9474 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9478 if (vcpu->arch.apicv_active)
9479 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9483 * Do this here before restoring debug registers on the host. And
9484 * since we do this before handling the vmexit, a DR access vmexit
9485 * can (a) read the correct value of the debug registers, (b) set
9486 * KVM_DEBUGREG_WONT_EXIT again.
9488 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9489 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9490 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9491 kvm_update_dr0123(vcpu);
9492 kvm_update_dr7(vcpu);
9493 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9497 * If the guest has used debug registers, at least dr7
9498 * will be disabled while returning to the host.
9499 * If we don't have active breakpoints in the host, we don't
9500 * care about the messed up debug address registers. But if
9501 * we have some of them active, restore the old state.
9503 if (hw_breakpoint_active())
9504 hw_breakpoint_restore();
9506 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9507 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9509 vcpu->mode = OUTSIDE_GUEST_MODE;
9512 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9515 * Consume any pending interrupts, including the possible source of
9516 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9517 * An instruction is required after local_irq_enable() to fully unblock
9518 * interrupts on processors that implement an interrupt shadow, the
9519 * stat.exits increment will do nicely.
9521 kvm_before_interrupt(vcpu);
9524 local_irq_disable();
9525 kvm_after_interrupt(vcpu);
9528 * Wait until after servicing IRQs to account guest time so that any
9529 * ticks that occurred while running the guest are properly accounted
9530 * to the guest. Waiting until IRQs are enabled degrades the accuracy
9531 * of accounting via context tracking, but the loss of accuracy is
9532 * acceptable for all known use cases.
9534 vtime_account_guest_exit();
9536 if (lapic_in_kernel(vcpu)) {
9537 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9538 if (delta != S64_MIN) {
9539 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9540 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9547 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9550 * Profile KVM exit RIPs:
9552 if (unlikely(prof_on == KVM_PROFILING)) {
9553 unsigned long rip = kvm_rip_read(vcpu);
9554 profile_hit(KVM_PROFILING, (void *)rip);
9557 if (unlikely(vcpu->arch.tsc_always_catchup))
9558 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9560 if (vcpu->arch.apic_attention)
9561 kvm_lapic_sync_from_vapic(vcpu);
9563 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9567 if (req_immediate_exit)
9568 kvm_make_request(KVM_REQ_EVENT, vcpu);
9569 static_call(kvm_x86_cancel_injection)(vcpu);
9570 if (unlikely(vcpu->arch.apic_attention))
9571 kvm_lapic_sync_from_vapic(vcpu);
9576 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9578 if (!kvm_arch_vcpu_runnable(vcpu) &&
9579 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9580 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9581 kvm_vcpu_block(vcpu);
9582 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9584 if (kvm_x86_ops.post_block)
9585 static_call(kvm_x86_post_block)(vcpu);
9587 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9591 if (kvm_apic_accept_events(vcpu) < 0)
9593 switch(vcpu->arch.mp_state) {
9594 case KVM_MP_STATE_HALTED:
9595 case KVM_MP_STATE_AP_RESET_HOLD:
9596 vcpu->arch.pv.pv_unhalted = false;
9597 vcpu->arch.mp_state =
9598 KVM_MP_STATE_RUNNABLE;
9600 case KVM_MP_STATE_RUNNABLE:
9601 vcpu->arch.apf.halted = false;
9603 case KVM_MP_STATE_INIT_RECEIVED:
9611 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9613 if (is_guest_mode(vcpu))
9614 kvm_check_nested_events(vcpu);
9616 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9617 !vcpu->arch.apf.halted);
9620 static int vcpu_run(struct kvm_vcpu *vcpu)
9623 struct kvm *kvm = vcpu->kvm;
9625 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9626 vcpu->arch.l1tf_flush_l1d = true;
9629 if (kvm_vcpu_running(vcpu)) {
9630 r = vcpu_enter_guest(vcpu);
9632 r = vcpu_block(kvm, vcpu);
9638 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9639 if (kvm_cpu_has_pending_timer(vcpu))
9640 kvm_inject_pending_timer_irqs(vcpu);
9642 if (dm_request_for_irq_injection(vcpu) &&
9643 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9645 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9646 ++vcpu->stat.request_irq_exits;
9650 if (__xfer_to_guest_mode_work_pending()) {
9651 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9652 r = xfer_to_guest_mode_handle_work(vcpu);
9655 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9659 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9664 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9668 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9669 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9670 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9674 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9676 BUG_ON(!vcpu->arch.pio.count);
9678 return complete_emulated_io(vcpu);
9682 * Implements the following, as a state machine:
9686 * for each mmio piece in the fragment
9694 * for each mmio piece in the fragment
9699 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9701 struct kvm_run *run = vcpu->run;
9702 struct kvm_mmio_fragment *frag;
9705 BUG_ON(!vcpu->mmio_needed);
9707 /* Complete previous fragment */
9708 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9709 len = min(8u, frag->len);
9710 if (!vcpu->mmio_is_write)
9711 memcpy(frag->data, run->mmio.data, len);
9713 if (frag->len <= 8) {
9714 /* Switch to the next fragment. */
9716 vcpu->mmio_cur_fragment++;
9718 /* Go forward to the next mmio piece. */
9724 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9725 vcpu->mmio_needed = 0;
9727 /* FIXME: return into emulator if single-stepping. */
9728 if (vcpu->mmio_is_write)
9730 vcpu->mmio_read_completed = 1;
9731 return complete_emulated_io(vcpu);
9734 run->exit_reason = KVM_EXIT_MMIO;
9735 run->mmio.phys_addr = frag->gpa;
9736 if (vcpu->mmio_is_write)
9737 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9738 run->mmio.len = min(8u, frag->len);
9739 run->mmio.is_write = vcpu->mmio_is_write;
9740 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9744 static void kvm_save_current_fpu(struct fpu *fpu)
9747 * If the target FPU state is not resident in the CPU registers, just
9748 * memcpy() from current, else save CPU state directly to the target.
9750 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9751 memcpy(&fpu->state, ¤t->thread.fpu.state,
9752 fpu_kernel_xstate_size);
9754 copy_fpregs_to_fpstate(fpu);
9757 /* Swap (qemu) user FPU context for the guest FPU context. */
9758 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9762 kvm_save_current_fpu(vcpu->arch.user_fpu);
9765 * Guests with protected state can't have it set by the hypervisor,
9766 * so skip trying to set it.
9768 if (vcpu->arch.guest_fpu)
9769 /* PKRU is separately restored in kvm_x86_ops.run. */
9770 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9771 ~XFEATURE_MASK_PKRU);
9773 fpregs_mark_activate();
9779 /* When vcpu_run ends, restore user space FPU context. */
9780 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9785 * Guests with protected state can't have it read by the hypervisor,
9786 * so skip trying to save it.
9788 if (vcpu->arch.guest_fpu)
9789 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9791 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9793 fpregs_mark_activate();
9796 ++vcpu->stat.fpu_reload;
9800 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9802 struct kvm_run *kvm_run = vcpu->run;
9806 kvm_sigset_activate(vcpu);
9808 kvm_load_guest_fpu(vcpu);
9810 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9811 if (kvm_run->immediate_exit) {
9815 kvm_vcpu_block(vcpu);
9816 if (kvm_apic_accept_events(vcpu) < 0) {
9820 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9822 if (signal_pending(current)) {
9824 kvm_run->exit_reason = KVM_EXIT_INTR;
9825 ++vcpu->stat.signal_exits;
9830 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9835 if (kvm_run->kvm_dirty_regs) {
9836 r = sync_regs(vcpu);
9841 /* re-sync apic's tpr */
9842 if (!lapic_in_kernel(vcpu)) {
9843 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9849 if (unlikely(vcpu->arch.complete_userspace_io)) {
9850 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9851 vcpu->arch.complete_userspace_io = NULL;
9856 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9858 if (kvm_run->immediate_exit)
9864 kvm_put_guest_fpu(vcpu);
9865 if (kvm_run->kvm_valid_regs)
9867 post_kvm_run_save(vcpu);
9868 kvm_sigset_deactivate(vcpu);
9874 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9876 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9878 * We are here if userspace calls get_regs() in the middle of
9879 * instruction emulation. Registers state needs to be copied
9880 * back from emulation context to vcpu. Userspace shouldn't do
9881 * that usually, but some bad designed PV devices (vmware
9882 * backdoor interface) need this to work
9884 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9885 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9887 regs->rax = kvm_rax_read(vcpu);
9888 regs->rbx = kvm_rbx_read(vcpu);
9889 regs->rcx = kvm_rcx_read(vcpu);
9890 regs->rdx = kvm_rdx_read(vcpu);
9891 regs->rsi = kvm_rsi_read(vcpu);
9892 regs->rdi = kvm_rdi_read(vcpu);
9893 regs->rsp = kvm_rsp_read(vcpu);
9894 regs->rbp = kvm_rbp_read(vcpu);
9895 #ifdef CONFIG_X86_64
9896 regs->r8 = kvm_r8_read(vcpu);
9897 regs->r9 = kvm_r9_read(vcpu);
9898 regs->r10 = kvm_r10_read(vcpu);
9899 regs->r11 = kvm_r11_read(vcpu);
9900 regs->r12 = kvm_r12_read(vcpu);
9901 regs->r13 = kvm_r13_read(vcpu);
9902 regs->r14 = kvm_r14_read(vcpu);
9903 regs->r15 = kvm_r15_read(vcpu);
9906 regs->rip = kvm_rip_read(vcpu);
9907 regs->rflags = kvm_get_rflags(vcpu);
9910 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9913 __get_regs(vcpu, regs);
9918 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9920 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9921 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9923 kvm_rax_write(vcpu, regs->rax);
9924 kvm_rbx_write(vcpu, regs->rbx);
9925 kvm_rcx_write(vcpu, regs->rcx);
9926 kvm_rdx_write(vcpu, regs->rdx);
9927 kvm_rsi_write(vcpu, regs->rsi);
9928 kvm_rdi_write(vcpu, regs->rdi);
9929 kvm_rsp_write(vcpu, regs->rsp);
9930 kvm_rbp_write(vcpu, regs->rbp);
9931 #ifdef CONFIG_X86_64
9932 kvm_r8_write(vcpu, regs->r8);
9933 kvm_r9_write(vcpu, regs->r9);
9934 kvm_r10_write(vcpu, regs->r10);
9935 kvm_r11_write(vcpu, regs->r11);
9936 kvm_r12_write(vcpu, regs->r12);
9937 kvm_r13_write(vcpu, regs->r13);
9938 kvm_r14_write(vcpu, regs->r14);
9939 kvm_r15_write(vcpu, regs->r15);
9942 kvm_rip_write(vcpu, regs->rip);
9943 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9945 vcpu->arch.exception.pending = false;
9947 kvm_make_request(KVM_REQ_EVENT, vcpu);
9950 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9953 __set_regs(vcpu, regs);
9958 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9960 struct kvm_segment cs;
9962 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9966 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9968 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9972 if (vcpu->arch.guest_state_protected)
9973 goto skip_protected_regs;
9975 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9976 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9977 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9978 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9979 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9980 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9982 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9983 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9985 static_call(kvm_x86_get_idt)(vcpu, &dt);
9986 sregs->idt.limit = dt.size;
9987 sregs->idt.base = dt.address;
9988 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9989 sregs->gdt.limit = dt.size;
9990 sregs->gdt.base = dt.address;
9992 sregs->cr2 = vcpu->arch.cr2;
9993 sregs->cr3 = kvm_read_cr3(vcpu);
9995 skip_protected_regs:
9996 sregs->cr0 = kvm_read_cr0(vcpu);
9997 sregs->cr4 = kvm_read_cr4(vcpu);
9998 sregs->cr8 = kvm_get_cr8(vcpu);
9999 sregs->efer = vcpu->arch.efer;
10000 sregs->apic_base = kvm_get_apic_base(vcpu);
10003 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10005 __get_sregs_common(vcpu, sregs);
10007 if (vcpu->arch.guest_state_protected)
10010 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10011 set_bit(vcpu->arch.interrupt.nr,
10012 (unsigned long *)sregs->interrupt_bitmap);
10015 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10019 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10021 if (vcpu->arch.guest_state_protected)
10024 if (is_pae_paging(vcpu)) {
10025 for (i = 0 ; i < 4 ; i++)
10026 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10027 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10031 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10032 struct kvm_sregs *sregs)
10035 __get_sregs(vcpu, sregs);
10040 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10041 struct kvm_mp_state *mp_state)
10046 if (kvm_mpx_supported())
10047 kvm_load_guest_fpu(vcpu);
10049 r = kvm_apic_accept_events(vcpu);
10054 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10055 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10056 vcpu->arch.pv.pv_unhalted)
10057 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10059 mp_state->mp_state = vcpu->arch.mp_state;
10062 if (kvm_mpx_supported())
10063 kvm_put_guest_fpu(vcpu);
10068 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10069 struct kvm_mp_state *mp_state)
10075 if (!lapic_in_kernel(vcpu) &&
10076 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10080 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10081 * INIT state; latched init should be reported using
10082 * KVM_SET_VCPU_EVENTS, so reject it here.
10084 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10085 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10086 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10089 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10090 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10091 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10093 vcpu->arch.mp_state = mp_state->mp_state;
10094 kvm_make_request(KVM_REQ_EVENT, vcpu);
10102 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10103 int reason, bool has_error_code, u32 error_code)
10105 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10108 init_emulate_ctxt(vcpu);
10110 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10111 has_error_code, error_code);
10113 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10114 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10115 vcpu->run->internal.ndata = 0;
10119 kvm_rip_write(vcpu, ctxt->eip);
10120 kvm_set_rflags(vcpu, ctxt->eflags);
10123 EXPORT_SYMBOL_GPL(kvm_task_switch);
10125 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10127 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10129 * When EFER.LME and CR0.PG are set, the processor is in
10130 * 64-bit mode (though maybe in a 32-bit code segment).
10131 * CR4.PAE and EFER.LMA must be set.
10133 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10135 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10139 * Not in 64-bit mode: EFER.LMA is clear and the code
10140 * segment cannot be 64-bit.
10142 if (sregs->efer & EFER_LMA || sregs->cs.l)
10146 return kvm_is_valid_cr4(vcpu, sregs->cr4);
10149 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10150 int *mmu_reset_needed, bool update_pdptrs)
10152 struct msr_data apic_base_msr;
10154 struct desc_ptr dt;
10156 if (!kvm_is_valid_sregs(vcpu, sregs))
10159 apic_base_msr.data = sregs->apic_base;
10160 apic_base_msr.host_initiated = true;
10161 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10164 if (vcpu->arch.guest_state_protected)
10167 dt.size = sregs->idt.limit;
10168 dt.address = sregs->idt.base;
10169 static_call(kvm_x86_set_idt)(vcpu, &dt);
10170 dt.size = sregs->gdt.limit;
10171 dt.address = sregs->gdt.base;
10172 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10174 vcpu->arch.cr2 = sregs->cr2;
10175 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10176 vcpu->arch.cr3 = sregs->cr3;
10177 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10179 kvm_set_cr8(vcpu, sregs->cr8);
10181 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10182 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10184 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10185 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10186 vcpu->arch.cr0 = sregs->cr0;
10188 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10189 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10191 if (update_pdptrs) {
10192 idx = srcu_read_lock(&vcpu->kvm->srcu);
10193 if (is_pae_paging(vcpu)) {
10194 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10195 *mmu_reset_needed = 1;
10197 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10200 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10201 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10202 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10203 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10204 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10205 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10207 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10208 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10210 update_cr8_intercept(vcpu);
10212 /* Older userspace won't unhalt the vcpu on reset. */
10213 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10214 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10215 !is_protmode(vcpu))
10216 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10221 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10223 int pending_vec, max_bits;
10224 int mmu_reset_needed = 0;
10225 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10230 if (mmu_reset_needed)
10231 kvm_mmu_reset_context(vcpu);
10233 max_bits = KVM_NR_INTERRUPTS;
10234 pending_vec = find_first_bit(
10235 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10237 if (pending_vec < max_bits) {
10238 kvm_queue_interrupt(vcpu, pending_vec, false);
10239 pr_debug("Set back pending irq %d\n", pending_vec);
10240 kvm_make_request(KVM_REQ_EVENT, vcpu);
10245 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10247 int mmu_reset_needed = 0;
10248 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10249 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10250 !(sregs2->efer & EFER_LMA);
10253 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10256 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10259 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10260 &mmu_reset_needed, !valid_pdptrs);
10264 if (valid_pdptrs) {
10265 for (i = 0; i < 4 ; i++)
10266 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10268 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10269 mmu_reset_needed = 1;
10270 vcpu->arch.pdptrs_from_userspace = true;
10272 if (mmu_reset_needed)
10273 kvm_mmu_reset_context(vcpu);
10277 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10278 struct kvm_sregs *sregs)
10283 ret = __set_sregs(vcpu, sregs);
10288 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10289 struct kvm_guest_debug *dbg)
10291 unsigned long rflags;
10294 if (vcpu->arch.guest_state_protected)
10299 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10301 if (vcpu->arch.exception.pending)
10303 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10304 kvm_queue_exception(vcpu, DB_VECTOR);
10306 kvm_queue_exception(vcpu, BP_VECTOR);
10310 * Read rflags as long as potentially injected trace flags are still
10313 rflags = kvm_get_rflags(vcpu);
10315 vcpu->guest_debug = dbg->control;
10316 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10317 vcpu->guest_debug = 0;
10319 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10320 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10321 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10322 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10324 for (i = 0; i < KVM_NR_DB_REGS; i++)
10325 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10327 kvm_update_dr7(vcpu);
10329 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10330 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10333 * Trigger an rflags update that will inject or remove the trace
10336 kvm_set_rflags(vcpu, rflags);
10338 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10348 * Translate a guest virtual address to a guest physical address.
10350 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10351 struct kvm_translation *tr)
10353 unsigned long vaddr = tr->linear_address;
10359 idx = srcu_read_lock(&vcpu->kvm->srcu);
10360 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10361 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10362 tr->physical_address = gpa;
10363 tr->valid = gpa != UNMAPPED_GVA;
10371 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10373 struct fxregs_state *fxsave;
10375 if (!vcpu->arch.guest_fpu)
10380 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10381 memcpy(fpu->fpr, fxsave->st_space, 128);
10382 fpu->fcw = fxsave->cwd;
10383 fpu->fsw = fxsave->swd;
10384 fpu->ftwx = fxsave->twd;
10385 fpu->last_opcode = fxsave->fop;
10386 fpu->last_ip = fxsave->rip;
10387 fpu->last_dp = fxsave->rdp;
10388 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10394 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10396 struct fxregs_state *fxsave;
10398 if (!vcpu->arch.guest_fpu)
10403 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10405 memcpy(fxsave->st_space, fpu->fpr, 128);
10406 fxsave->cwd = fpu->fcw;
10407 fxsave->swd = fpu->fsw;
10408 fxsave->twd = fpu->ftwx;
10409 fxsave->fop = fpu->last_opcode;
10410 fxsave->rip = fpu->last_ip;
10411 fxsave->rdp = fpu->last_dp;
10412 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10418 static void store_regs(struct kvm_vcpu *vcpu)
10420 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10422 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10423 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10425 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10426 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10428 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10429 kvm_vcpu_ioctl_x86_get_vcpu_events(
10430 vcpu, &vcpu->run->s.regs.events);
10433 static int sync_regs(struct kvm_vcpu *vcpu)
10435 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10438 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10439 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10440 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10442 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10443 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10445 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10447 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10448 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10449 vcpu, &vcpu->run->s.regs.events))
10451 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10457 static void fx_init(struct kvm_vcpu *vcpu)
10459 if (!vcpu->arch.guest_fpu)
10462 fpstate_init(&vcpu->arch.guest_fpu->state);
10463 if (boot_cpu_has(X86_FEATURE_XSAVES))
10464 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10465 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10468 * Ensure guest xcr0 is valid for loading
10470 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10472 vcpu->arch.cr0 |= X86_CR0_ET;
10475 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10477 if (vcpu->arch.guest_fpu) {
10478 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10479 vcpu->arch.guest_fpu = NULL;
10482 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10484 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10486 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10487 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10488 "guest TSC will not be reliable\n");
10493 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10498 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10499 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10501 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10503 r = kvm_mmu_create(vcpu);
10507 if (irqchip_in_kernel(vcpu->kvm)) {
10508 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10510 goto fail_mmu_destroy;
10511 if (kvm_apicv_activated(vcpu->kvm))
10512 vcpu->arch.apicv_active = true;
10514 static_branch_inc(&kvm_has_noapic_vcpu);
10518 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10520 goto fail_free_lapic;
10521 vcpu->arch.pio_data = page_address(page);
10523 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10524 GFP_KERNEL_ACCOUNT);
10525 if (!vcpu->arch.mce_banks)
10526 goto fail_free_pio_data;
10527 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10529 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10530 GFP_KERNEL_ACCOUNT))
10531 goto fail_free_mce_banks;
10533 if (!alloc_emulate_ctxt(vcpu))
10534 goto free_wbinvd_dirty_mask;
10536 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10537 GFP_KERNEL_ACCOUNT);
10538 if (!vcpu->arch.user_fpu) {
10539 pr_err("kvm: failed to allocate userspace's fpu\n");
10540 goto free_emulate_ctxt;
10543 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10544 GFP_KERNEL_ACCOUNT);
10545 if (!vcpu->arch.guest_fpu) {
10546 pr_err("kvm: failed to allocate vcpu's fpu\n");
10547 goto free_user_fpu;
10551 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10552 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10554 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10556 kvm_async_pf_hash_reset(vcpu);
10557 kvm_pmu_init(vcpu);
10559 vcpu->arch.pending_external_vector = -1;
10560 vcpu->arch.preempted_in_kernel = false;
10562 #if IS_ENABLED(CONFIG_HYPERV)
10563 vcpu->arch.hv_root_tdp = INVALID_PAGE;
10566 r = static_call(kvm_x86_vcpu_create)(vcpu);
10568 goto free_guest_fpu;
10570 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10571 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10572 kvm_vcpu_mtrr_init(vcpu);
10574 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10575 kvm_vcpu_reset(vcpu, false);
10576 kvm_init_mmu(vcpu, false);
10581 kvm_free_guest_fpu(vcpu);
10583 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10585 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10586 free_wbinvd_dirty_mask:
10587 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10588 fail_free_mce_banks:
10589 kfree(vcpu->arch.mce_banks);
10590 fail_free_pio_data:
10591 free_page((unsigned long)vcpu->arch.pio_data);
10593 kvm_free_lapic(vcpu);
10595 kvm_mmu_destroy(vcpu);
10599 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10601 struct kvm *kvm = vcpu->kvm;
10603 if (mutex_lock_killable(&vcpu->mutex))
10606 kvm_synchronize_tsc(vcpu, 0);
10609 /* poll control enabled by default */
10610 vcpu->arch.msr_kvm_poll_control = 1;
10612 mutex_unlock(&vcpu->mutex);
10614 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10615 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10616 KVMCLOCK_SYNC_PERIOD);
10619 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10621 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10624 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10626 kvmclock_reset(vcpu);
10628 static_call(kvm_x86_vcpu_free)(vcpu);
10630 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10631 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10632 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10633 kvm_free_guest_fpu(vcpu);
10635 kvm_hv_vcpu_uninit(vcpu);
10636 kvm_pmu_destroy(vcpu);
10637 kfree(vcpu->arch.mce_banks);
10638 kvm_free_lapic(vcpu);
10639 idx = srcu_read_lock(&vcpu->kvm->srcu);
10640 kvm_mmu_destroy(vcpu);
10641 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10642 free_page((unsigned long)vcpu->arch.pio_data);
10643 kvfree(vcpu->arch.cpuid_entries);
10644 if (!lapic_in_kernel(vcpu))
10645 static_branch_dec(&kvm_has_noapic_vcpu);
10648 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10650 kvm_lapic_reset(vcpu, init_event);
10652 vcpu->arch.hflags = 0;
10654 vcpu->arch.smi_pending = 0;
10655 vcpu->arch.smi_count = 0;
10656 atomic_set(&vcpu->arch.nmi_queued, 0);
10657 vcpu->arch.nmi_pending = 0;
10658 vcpu->arch.nmi_injected = false;
10659 kvm_clear_interrupt_queue(vcpu);
10660 kvm_clear_exception_queue(vcpu);
10662 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10663 kvm_update_dr0123(vcpu);
10664 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10665 vcpu->arch.dr7 = DR7_FIXED_1;
10666 kvm_update_dr7(vcpu);
10668 vcpu->arch.cr2 = 0;
10670 kvm_make_request(KVM_REQ_EVENT, vcpu);
10671 vcpu->arch.apf.msr_en_val = 0;
10672 vcpu->arch.apf.msr_int_val = 0;
10673 vcpu->arch.st.msr_val = 0;
10675 kvmclock_reset(vcpu);
10677 kvm_clear_async_pf_completion_queue(vcpu);
10678 kvm_async_pf_hash_reset(vcpu);
10679 vcpu->arch.apf.halted = false;
10681 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10682 void *mpx_state_buffer;
10685 * To avoid have the INIT path from kvm_apic_has_events() that be
10686 * called with loaded FPU and does not let userspace fix the state.
10689 kvm_put_guest_fpu(vcpu);
10690 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10692 if (mpx_state_buffer)
10693 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10694 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10696 if (mpx_state_buffer)
10697 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10699 kvm_load_guest_fpu(vcpu);
10703 kvm_pmu_reset(vcpu);
10704 vcpu->arch.smbase = 0x30000;
10706 vcpu->arch.msr_misc_features_enables = 0;
10708 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10711 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10712 vcpu->arch.regs_avail = ~0;
10713 vcpu->arch.regs_dirty = ~0;
10715 vcpu->arch.ia32_xss = 0;
10717 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10720 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10722 struct kvm_segment cs;
10724 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10725 cs.selector = vector << 8;
10726 cs.base = vector << 12;
10727 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10728 kvm_rip_write(vcpu, 0);
10730 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10732 int kvm_arch_hardware_enable(void)
10735 struct kvm_vcpu *vcpu;
10740 bool stable, backwards_tsc = false;
10742 kvm_user_return_msr_cpu_online();
10743 ret = static_call(kvm_x86_hardware_enable)();
10747 local_tsc = rdtsc();
10748 stable = !kvm_check_tsc_unstable();
10749 list_for_each_entry(kvm, &vm_list, vm_list) {
10750 kvm_for_each_vcpu(i, vcpu, kvm) {
10751 if (!stable && vcpu->cpu == smp_processor_id())
10752 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10753 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10754 backwards_tsc = true;
10755 if (vcpu->arch.last_host_tsc > max_tsc)
10756 max_tsc = vcpu->arch.last_host_tsc;
10762 * Sometimes, even reliable TSCs go backwards. This happens on
10763 * platforms that reset TSC during suspend or hibernate actions, but
10764 * maintain synchronization. We must compensate. Fortunately, we can
10765 * detect that condition here, which happens early in CPU bringup,
10766 * before any KVM threads can be running. Unfortunately, we can't
10767 * bring the TSCs fully up to date with real time, as we aren't yet far
10768 * enough into CPU bringup that we know how much real time has actually
10769 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10770 * variables that haven't been updated yet.
10772 * So we simply find the maximum observed TSC above, then record the
10773 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10774 * the adjustment will be applied. Note that we accumulate
10775 * adjustments, in case multiple suspend cycles happen before some VCPU
10776 * gets a chance to run again. In the event that no KVM threads get a
10777 * chance to run, we will miss the entire elapsed period, as we'll have
10778 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10779 * loose cycle time. This isn't too big a deal, since the loss will be
10780 * uniform across all VCPUs (not to mention the scenario is extremely
10781 * unlikely). It is possible that a second hibernate recovery happens
10782 * much faster than a first, causing the observed TSC here to be
10783 * smaller; this would require additional padding adjustment, which is
10784 * why we set last_host_tsc to the local tsc observed here.
10786 * N.B. - this code below runs only on platforms with reliable TSC,
10787 * as that is the only way backwards_tsc is set above. Also note
10788 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10789 * have the same delta_cyc adjustment applied if backwards_tsc
10790 * is detected. Note further, this adjustment is only done once,
10791 * as we reset last_host_tsc on all VCPUs to stop this from being
10792 * called multiple times (one for each physical CPU bringup).
10794 * Platforms with unreliable TSCs don't have to deal with this, they
10795 * will be compensated by the logic in vcpu_load, which sets the TSC to
10796 * catchup mode. This will catchup all VCPUs to real time, but cannot
10797 * guarantee that they stay in perfect synchronization.
10799 if (backwards_tsc) {
10800 u64 delta_cyc = max_tsc - local_tsc;
10801 list_for_each_entry(kvm, &vm_list, vm_list) {
10802 kvm->arch.backwards_tsc_observed = true;
10803 kvm_for_each_vcpu(i, vcpu, kvm) {
10804 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10805 vcpu->arch.last_host_tsc = local_tsc;
10806 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10810 * We have to disable TSC offset matching.. if you were
10811 * booting a VM while issuing an S4 host suspend....
10812 * you may have some problem. Solving this issue is
10813 * left as an exercise to the reader.
10815 kvm->arch.last_tsc_nsec = 0;
10816 kvm->arch.last_tsc_write = 0;
10823 void kvm_arch_hardware_disable(void)
10825 static_call(kvm_x86_hardware_disable)();
10826 drop_user_return_notifiers();
10829 int kvm_arch_hardware_setup(void *opaque)
10831 struct kvm_x86_init_ops *ops = opaque;
10834 rdmsrl_safe(MSR_EFER, &host_efer);
10836 if (boot_cpu_has(X86_FEATURE_XSAVES))
10837 rdmsrl(MSR_IA32_XSS, host_xss);
10839 r = ops->hardware_setup();
10843 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10844 kvm_ops_static_call_update();
10846 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10849 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10850 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10851 #undef __kvm_cpu_cap_has
10853 if (kvm_has_tsc_control) {
10855 * Make sure the user can only configure tsc_khz values that
10856 * fit into a signed integer.
10857 * A min value is not calculated because it will always
10858 * be 1 on all machines.
10860 u64 max = min(0x7fffffffULL,
10861 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10862 kvm_max_guest_tsc_khz = max;
10864 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10867 kvm_init_msr_list();
10871 void kvm_arch_hardware_unsetup(void)
10873 static_call(kvm_x86_hardware_unsetup)();
10876 int kvm_arch_check_processor_compat(void *opaque)
10878 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10879 struct kvm_x86_init_ops *ops = opaque;
10881 WARN_ON(!irqs_disabled());
10883 if (__cr4_reserved_bits(cpu_has, c) !=
10884 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10887 return ops->check_processor_compatibility();
10890 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10892 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10894 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10896 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10898 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10901 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10902 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10904 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10906 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10908 vcpu->arch.l1tf_flush_l1d = true;
10909 if (pmu->version && unlikely(pmu->event_count)) {
10910 pmu->need_cleanup = true;
10911 kvm_make_request(KVM_REQ_PMU, vcpu);
10913 static_call(kvm_x86_sched_in)(vcpu, cpu);
10916 void kvm_arch_free_vm(struct kvm *kvm)
10918 kfree(to_kvm_hv(kvm)->hv_pa_pg);
10923 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10928 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10929 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10930 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10931 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10932 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10933 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10935 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10936 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10937 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10938 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10939 &kvm->arch.irq_sources_bitmap);
10941 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10942 mutex_init(&kvm->arch.apic_map_lock);
10943 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10945 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10946 pvclock_update_vm_gtod_copy(kvm);
10948 kvm->arch.guest_can_read_msr_platform_info = true;
10950 #if IS_ENABLED(CONFIG_HYPERV)
10951 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
10952 kvm->arch.hv_root_tdp = INVALID_PAGE;
10955 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10956 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10958 kvm_apicv_init(kvm);
10959 kvm_hv_init_vm(kvm);
10960 kvm_page_track_init(kvm);
10961 kvm_mmu_init_vm(kvm);
10963 return static_call(kvm_x86_vm_init)(kvm);
10966 int kvm_arch_post_init_vm(struct kvm *kvm)
10968 return kvm_mmu_post_init_vm(kvm);
10971 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10974 kvm_mmu_unload(vcpu);
10978 static void kvm_free_vcpus(struct kvm *kvm)
10981 struct kvm_vcpu *vcpu;
10984 * Unpin any mmu pages first.
10986 kvm_for_each_vcpu(i, vcpu, kvm) {
10987 kvm_clear_async_pf_completion_queue(vcpu);
10988 kvm_unload_vcpu_mmu(vcpu);
10990 kvm_for_each_vcpu(i, vcpu, kvm)
10991 kvm_vcpu_destroy(vcpu);
10993 mutex_lock(&kvm->lock);
10994 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10995 kvm->vcpus[i] = NULL;
10997 atomic_set(&kvm->online_vcpus, 0);
10998 mutex_unlock(&kvm->lock);
11001 void kvm_arch_sync_events(struct kvm *kvm)
11003 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11004 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11008 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
11011 * __x86_set_memory_region: Setup KVM internal memory slot
11013 * @kvm: the kvm pointer to the VM.
11014 * @id: the slot ID to setup.
11015 * @gpa: the GPA to install the slot (unused when @size == 0).
11016 * @size: the size of the slot. Set to zero to uninstall a slot.
11018 * This function helps to setup a KVM internal memory slot. Specify
11019 * @size > 0 to install a new slot, while @size == 0 to uninstall a
11020 * slot. The return code can be one of the following:
11022 * HVA: on success (uninstall will return a bogus HVA)
11025 * The caller should always use IS_ERR() to check the return value
11026 * before use. Note, the KVM internal memory slots are guaranteed to
11027 * remain valid and unchanged until the VM is destroyed, i.e., the
11028 * GPA->HVA translation will not change. However, the HVA is a user
11029 * address, i.e. its accessibility is not guaranteed, and must be
11030 * accessed via __copy_{to,from}_user().
11032 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11036 unsigned long hva, old_npages;
11037 struct kvm_memslots *slots = kvm_memslots(kvm);
11038 struct kvm_memory_slot *slot;
11040 /* Called with kvm->slots_lock held. */
11041 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11042 return ERR_PTR_USR(-EINVAL);
11044 slot = id_to_memslot(slots, id);
11046 if (slot && slot->npages)
11047 return ERR_PTR_USR(-EEXIST);
11050 * MAP_SHARED to prevent internal slot pages from being moved
11053 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11054 MAP_SHARED | MAP_ANONYMOUS, 0);
11055 if (IS_ERR((void *)hva))
11056 return (void __user *)hva;
11058 if (!slot || !slot->npages)
11061 old_npages = slot->npages;
11062 hva = slot->userspace_addr;
11065 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11066 struct kvm_userspace_memory_region m;
11068 m.slot = id | (i << 16);
11070 m.guest_phys_addr = gpa;
11071 m.userspace_addr = hva;
11072 m.memory_size = size;
11073 r = __kvm_set_memory_region(kvm, &m);
11075 return ERR_PTR_USR(r);
11079 vm_munmap(hva, old_npages * PAGE_SIZE);
11081 return (void __user *)hva;
11083 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11085 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11087 kvm_mmu_pre_destroy_vm(kvm);
11090 void kvm_arch_destroy_vm(struct kvm *kvm)
11092 if (current->mm == kvm->mm) {
11094 * Free memory regions allocated on behalf of userspace,
11095 * unless the the memory map has changed due to process exit
11098 mutex_lock(&kvm->slots_lock);
11099 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11101 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11103 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11104 mutex_unlock(&kvm->slots_lock);
11106 static_call_cond(kvm_x86_vm_destroy)(kvm);
11107 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11108 kvm_pic_destroy(kvm);
11109 kvm_ioapic_destroy(kvm);
11110 kvm_free_vcpus(kvm);
11111 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11112 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11113 kvm_mmu_uninit_vm(kvm);
11114 kvm_page_track_cleanup(kvm);
11115 kvm_xen_destroy_vm(kvm);
11116 kvm_hv_destroy_vm(kvm);
11119 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11123 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11124 kvfree(slot->arch.rmap[i]);
11125 slot->arch.rmap[i] = NULL;
11129 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11133 memslot_rmap_free(slot);
11135 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11136 kvfree(slot->arch.lpage_info[i - 1]);
11137 slot->arch.lpage_info[i - 1] = NULL;
11140 kvm_page_track_free_memslot(slot);
11143 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
11144 unsigned long npages)
11146 const int sz = sizeof(*slot->arch.rmap[0]);
11149 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11151 int lpages = gfn_to_index(slot->base_gfn + npages - 1,
11152 slot->base_gfn, level) + 1;
11154 WARN_ON(slot->arch.rmap[i]);
11156 slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11157 if (!slot->arch.rmap[i]) {
11158 memslot_rmap_free(slot);
11166 int alloc_all_memslots_rmaps(struct kvm *kvm)
11168 struct kvm_memslots *slots;
11169 struct kvm_memory_slot *slot;
11173 * Check if memslots alreday have rmaps early before acquiring
11174 * the slots_arch_lock below.
11176 if (kvm_memslots_have_rmaps(kvm))
11179 mutex_lock(&kvm->slots_arch_lock);
11182 * Read memslots_have_rmaps again, under the slots arch lock,
11183 * before allocating the rmaps
11185 if (kvm_memslots_have_rmaps(kvm)) {
11186 mutex_unlock(&kvm->slots_arch_lock);
11190 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11191 slots = __kvm_memslots(kvm, i);
11192 kvm_for_each_memslot(slot, slots) {
11193 r = memslot_rmap_alloc(slot, slot->npages);
11195 mutex_unlock(&kvm->slots_arch_lock);
11202 * Ensure that memslots_have_rmaps becomes true strictly after
11203 * all the rmap pointers are set.
11205 smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11206 mutex_unlock(&kvm->slots_arch_lock);
11210 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11211 struct kvm_memory_slot *slot,
11212 unsigned long npages)
11217 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
11218 * old arrays will be freed by __kvm_set_memory_region() if installing
11219 * the new memslot is successful.
11221 memset(&slot->arch, 0, sizeof(slot->arch));
11223 if (kvm_memslots_have_rmaps(kvm)) {
11224 r = memslot_rmap_alloc(slot, npages);
11229 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11230 struct kvm_lpage_info *linfo;
11231 unsigned long ugfn;
11235 lpages = gfn_to_index(slot->base_gfn + npages - 1,
11236 slot->base_gfn, level) + 1;
11238 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11242 slot->arch.lpage_info[i - 1] = linfo;
11244 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11245 linfo[0].disallow_lpage = 1;
11246 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11247 linfo[lpages - 1].disallow_lpage = 1;
11248 ugfn = slot->userspace_addr >> PAGE_SHIFT;
11250 * If the gfn and userspace address are not aligned wrt each
11251 * other, disable large page support for this slot.
11253 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11256 for (j = 0; j < lpages; ++j)
11257 linfo[j].disallow_lpage = 1;
11261 if (kvm_page_track_create_memslot(slot, npages))
11267 memslot_rmap_free(slot);
11269 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11270 kvfree(slot->arch.lpage_info[i - 1]);
11271 slot->arch.lpage_info[i - 1] = NULL;
11276 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11278 struct kvm_vcpu *vcpu;
11282 * memslots->generation has been incremented.
11283 * mmio generation may have reached its maximum value.
11285 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11287 /* Force re-initialization of steal_time cache */
11288 kvm_for_each_vcpu(i, vcpu, kvm)
11289 kvm_vcpu_kick(vcpu);
11292 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11293 struct kvm_memory_slot *memslot,
11294 const struct kvm_userspace_memory_region *mem,
11295 enum kvm_mr_change change)
11297 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11298 return kvm_alloc_memslot_metadata(kvm, memslot,
11299 mem->memory_size >> PAGE_SHIFT);
11304 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11306 struct kvm_arch *ka = &kvm->arch;
11308 if (!kvm_x86_ops.cpu_dirty_log_size)
11311 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11312 (!enable && --ka->cpu_dirty_logging_count == 0))
11313 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11315 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11318 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11319 struct kvm_memory_slot *old,
11320 struct kvm_memory_slot *new,
11321 enum kvm_mr_change change)
11323 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11326 * Update CPU dirty logging if dirty logging is being toggled. This
11327 * applies to all operations.
11329 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11330 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11333 * Nothing more to do for RO slots (which can't be dirtied and can't be
11334 * made writable) or CREATE/MOVE/DELETE of a slot.
11336 * For a memslot with dirty logging disabled:
11337 * CREATE: No dirty mappings will already exist.
11338 * MOVE/DELETE: The old mappings will already have been cleaned up by
11339 * kvm_arch_flush_shadow_memslot()
11341 * For a memslot with dirty logging enabled:
11342 * CREATE: No shadow pages exist, thus nothing to write-protect
11343 * and no dirty bits to clear.
11344 * MOVE/DELETE: The old mappings will already have been cleaned up by
11345 * kvm_arch_flush_shadow_memslot().
11347 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11351 * READONLY and non-flags changes were filtered out above, and the only
11352 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11353 * logging isn't being toggled on or off.
11355 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11358 if (!log_dirty_pages) {
11360 * Dirty logging tracks sptes in 4k granularity, meaning that
11361 * large sptes have to be split. If live migration succeeds,
11362 * the guest in the source machine will be destroyed and large
11363 * sptes will be created in the destination. However, if the
11364 * guest continues to run in the source machine (for example if
11365 * live migration fails), small sptes will remain around and
11366 * cause bad performance.
11368 * Scan sptes if dirty logging has been stopped, dropping those
11369 * which can be collapsed into a single large-page spte. Later
11370 * page faults will create the large-page sptes.
11372 kvm_mmu_zap_collapsible_sptes(kvm, new);
11375 * Initially-all-set does not require write protecting any page,
11376 * because they're all assumed to be dirty.
11378 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11381 if (kvm_x86_ops.cpu_dirty_log_size) {
11382 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11383 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11385 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11390 void kvm_arch_commit_memory_region(struct kvm *kvm,
11391 const struct kvm_userspace_memory_region *mem,
11392 struct kvm_memory_slot *old,
11393 const struct kvm_memory_slot *new,
11394 enum kvm_mr_change change)
11396 if (!kvm->arch.n_requested_mmu_pages)
11397 kvm_mmu_change_mmu_pages(kvm,
11398 kvm_mmu_calculate_default_mmu_pages(kvm));
11401 * FIXME: const-ify all uses of struct kvm_memory_slot.
11403 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11405 /* Free the arrays associated with the old memslot. */
11406 if (change == KVM_MR_MOVE)
11407 kvm_arch_free_memslot(kvm, old);
11410 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11412 kvm_mmu_zap_all(kvm);
11415 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11416 struct kvm_memory_slot *slot)
11418 kvm_page_track_flush_slot(kvm, slot);
11421 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11423 return (is_guest_mode(vcpu) &&
11424 kvm_x86_ops.guest_apic_has_interrupt &&
11425 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11428 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11430 if (!list_empty_careful(&vcpu->async_pf.done))
11433 if (kvm_apic_has_events(vcpu))
11436 if (vcpu->arch.pv.pv_unhalted)
11439 if (vcpu->arch.exception.pending)
11442 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11443 (vcpu->arch.nmi_pending &&
11444 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11447 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11448 (vcpu->arch.smi_pending &&
11449 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11452 if (kvm_arch_interrupt_allowed(vcpu) &&
11453 (kvm_cpu_has_interrupt(vcpu) ||
11454 kvm_guest_apic_has_interrupt(vcpu)))
11457 if (kvm_hv_has_stimer_pending(vcpu))
11460 if (is_guest_mode(vcpu) &&
11461 kvm_x86_ops.nested_ops->hv_timer_pending &&
11462 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11468 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11470 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11473 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11475 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11481 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11483 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11486 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11487 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11488 kvm_test_request(KVM_REQ_EVENT, vcpu))
11491 return kvm_arch_dy_has_pending_interrupt(vcpu);
11494 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11496 if (vcpu->arch.guest_state_protected)
11499 return vcpu->arch.preempted_in_kernel;
11502 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11504 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11507 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11509 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11512 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11514 /* Can't read the RIP when guest state is protected, just return 0 */
11515 if (vcpu->arch.guest_state_protected)
11518 if (is_64_bit_mode(vcpu))
11519 return kvm_rip_read(vcpu);
11520 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11521 kvm_rip_read(vcpu));
11523 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11525 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11527 return kvm_get_linear_rip(vcpu) == linear_rip;
11529 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11531 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11533 unsigned long rflags;
11535 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11536 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11537 rflags &= ~X86_EFLAGS_TF;
11540 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11542 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11544 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11545 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11546 rflags |= X86_EFLAGS_TF;
11547 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11550 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11552 __kvm_set_rflags(vcpu, rflags);
11553 kvm_make_request(KVM_REQ_EVENT, vcpu);
11555 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11557 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11561 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11565 r = kvm_mmu_reload(vcpu);
11569 if (!vcpu->arch.mmu->direct_map &&
11570 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11573 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11576 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11578 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11580 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11583 static inline u32 kvm_async_pf_next_probe(u32 key)
11585 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11588 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11590 u32 key = kvm_async_pf_hash_fn(gfn);
11592 while (vcpu->arch.apf.gfns[key] != ~0)
11593 key = kvm_async_pf_next_probe(key);
11595 vcpu->arch.apf.gfns[key] = gfn;
11598 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11601 u32 key = kvm_async_pf_hash_fn(gfn);
11603 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11604 (vcpu->arch.apf.gfns[key] != gfn &&
11605 vcpu->arch.apf.gfns[key] != ~0); i++)
11606 key = kvm_async_pf_next_probe(key);
11611 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11613 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11616 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11620 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11622 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11626 vcpu->arch.apf.gfns[i] = ~0;
11628 j = kvm_async_pf_next_probe(j);
11629 if (vcpu->arch.apf.gfns[j] == ~0)
11631 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11633 * k lies cyclically in ]i,j]
11635 * |....j i.k.| or |.k..j i...|
11637 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11638 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11643 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11645 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11647 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11651 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11653 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11655 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11656 &token, offset, sizeof(token));
11659 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11661 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11664 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11665 &val, offset, sizeof(val)))
11671 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11673 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11676 if (!kvm_pv_async_pf_enabled(vcpu) ||
11677 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11683 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11685 if (unlikely(!lapic_in_kernel(vcpu) ||
11686 kvm_event_needs_reinjection(vcpu) ||
11687 vcpu->arch.exception.pending))
11690 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11694 * If interrupts are off we cannot even use an artificial
11697 return kvm_arch_interrupt_allowed(vcpu);
11700 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11701 struct kvm_async_pf *work)
11703 struct x86_exception fault;
11705 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11706 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11708 if (kvm_can_deliver_async_pf(vcpu) &&
11709 !apf_put_user_notpresent(vcpu)) {
11710 fault.vector = PF_VECTOR;
11711 fault.error_code_valid = true;
11712 fault.error_code = 0;
11713 fault.nested_page_fault = false;
11714 fault.address = work->arch.token;
11715 fault.async_page_fault = true;
11716 kvm_inject_page_fault(vcpu, &fault);
11720 * It is not possible to deliver a paravirtualized asynchronous
11721 * page fault, but putting the guest in an artificial halt state
11722 * can be beneficial nevertheless: if an interrupt arrives, we
11723 * can deliver it timely and perhaps the guest will schedule
11724 * another process. When the instruction that triggered a page
11725 * fault is retried, hopefully the page will be ready in the host.
11727 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11732 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11733 struct kvm_async_pf *work)
11735 struct kvm_lapic_irq irq = {
11736 .delivery_mode = APIC_DM_FIXED,
11737 .vector = vcpu->arch.apf.vec
11740 if (work->wakeup_all)
11741 work->arch.token = ~0; /* broadcast wakeup */
11743 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11744 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11746 if ((work->wakeup_all || work->notpresent_injected) &&
11747 kvm_pv_async_pf_enabled(vcpu) &&
11748 !apf_put_user_ready(vcpu, work->arch.token)) {
11749 vcpu->arch.apf.pageready_pending = true;
11750 kvm_apic_set_irq(vcpu, &irq, NULL);
11753 vcpu->arch.apf.halted = false;
11754 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11757 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11759 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11760 if (!vcpu->arch.apf.pageready_pending)
11761 kvm_vcpu_kick(vcpu);
11764 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11766 if (!kvm_pv_async_pf_enabled(vcpu))
11769 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11772 void kvm_arch_start_assignment(struct kvm *kvm)
11774 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
11775 static_call_cond(kvm_x86_start_assignment)(kvm);
11777 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11779 void kvm_arch_end_assignment(struct kvm *kvm)
11781 atomic_dec(&kvm->arch.assigned_device_count);
11783 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11785 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11787 return atomic_read(&kvm->arch.assigned_device_count);
11789 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11791 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11793 atomic_inc(&kvm->arch.noncoherent_dma_count);
11795 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11797 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11799 atomic_dec(&kvm->arch.noncoherent_dma_count);
11801 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11803 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11805 return atomic_read(&kvm->arch.noncoherent_dma_count);
11807 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11809 bool kvm_arch_has_irq_bypass(void)
11814 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11815 struct irq_bypass_producer *prod)
11817 struct kvm_kernel_irqfd *irqfd =
11818 container_of(cons, struct kvm_kernel_irqfd, consumer);
11821 irqfd->producer = prod;
11822 kvm_arch_start_assignment(irqfd->kvm);
11823 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11824 prod->irq, irqfd->gsi, 1);
11827 kvm_arch_end_assignment(irqfd->kvm);
11832 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11833 struct irq_bypass_producer *prod)
11836 struct kvm_kernel_irqfd *irqfd =
11837 container_of(cons, struct kvm_kernel_irqfd, consumer);
11839 WARN_ON(irqfd->producer != prod);
11840 irqfd->producer = NULL;
11843 * When producer of consumer is unregistered, we change back to
11844 * remapped mode, so we can re-use the current implementation
11845 * when the irq is masked/disabled or the consumer side (KVM
11846 * int this case doesn't want to receive the interrupts.
11848 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11850 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11851 " fails: %d\n", irqfd->consumer.token, ret);
11853 kvm_arch_end_assignment(irqfd->kvm);
11856 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11857 uint32_t guest_irq, bool set)
11859 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11862 bool kvm_vector_hashing_enabled(void)
11864 return vector_hashing;
11867 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11869 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11871 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11874 int kvm_spec_ctrl_test_value(u64 value)
11877 * test that setting IA32_SPEC_CTRL to given value
11878 * is allowed by the host processor
11882 unsigned long flags;
11885 local_irq_save(flags);
11887 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11889 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11892 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11894 local_irq_restore(flags);
11898 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11900 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11902 struct x86_exception fault;
11903 u32 access = error_code &
11904 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11906 if (!(error_code & PFERR_PRESENT_MASK) ||
11907 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11909 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11910 * tables probably do not match the TLB. Just proceed
11911 * with the error code that the processor gave.
11913 fault.vector = PF_VECTOR;
11914 fault.error_code_valid = true;
11915 fault.error_code = error_code;
11916 fault.nested_page_fault = false;
11917 fault.address = gva;
11919 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11921 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11924 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11925 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11926 * indicates whether exit to userspace is needed.
11928 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11929 struct x86_exception *e)
11931 if (r == X86EMUL_PROPAGATE_FAULT) {
11932 kvm_inject_emulated_page_fault(vcpu, e);
11937 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11938 * while handling a VMX instruction KVM could've handled the request
11939 * correctly by exiting to userspace and performing I/O but there
11940 * doesn't seem to be a real use-case behind such requests, just return
11941 * KVM_EXIT_INTERNAL_ERROR for now.
11943 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11944 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11945 vcpu->run->internal.ndata = 0;
11949 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11951 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11954 struct x86_exception e;
11956 unsigned long roots_to_free = 0;
11963 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11964 if (r != X86EMUL_CONTINUE)
11965 return kvm_handle_memory_failure(vcpu, r, &e);
11967 if (operand.pcid >> 12 != 0) {
11968 kvm_inject_gp(vcpu, 0);
11972 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11975 case INVPCID_TYPE_INDIV_ADDR:
11976 if ((!pcid_enabled && (operand.pcid != 0)) ||
11977 is_noncanonical_address(operand.gla, vcpu)) {
11978 kvm_inject_gp(vcpu, 0);
11981 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11982 return kvm_skip_emulated_instruction(vcpu);
11984 case INVPCID_TYPE_SINGLE_CTXT:
11985 if (!pcid_enabled && (operand.pcid != 0)) {
11986 kvm_inject_gp(vcpu, 0);
11990 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11991 kvm_mmu_sync_roots(vcpu);
11992 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11995 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11996 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11998 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
12000 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
12002 * If neither the current cr3 nor any of the prev_roots use the
12003 * given PCID, then nothing needs to be done here because a
12004 * resync will happen anyway before switching to any other CR3.
12007 return kvm_skip_emulated_instruction(vcpu);
12009 case INVPCID_TYPE_ALL_NON_GLOBAL:
12011 * Currently, KVM doesn't mark global entries in the shadow
12012 * page tables, so a non-global flush just degenerates to a
12013 * global flush. If needed, we could optimize this later by
12014 * keeping track of global entries in shadow page tables.
12018 case INVPCID_TYPE_ALL_INCL_GLOBAL:
12019 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
12020 return kvm_skip_emulated_instruction(vcpu);
12023 BUG(); /* We have already checked above that type <= 3 */
12026 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12028 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12030 struct kvm_run *run = vcpu->run;
12031 struct kvm_mmio_fragment *frag;
12034 BUG_ON(!vcpu->mmio_needed);
12036 /* Complete previous fragment */
12037 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12038 len = min(8u, frag->len);
12039 if (!vcpu->mmio_is_write)
12040 memcpy(frag->data, run->mmio.data, len);
12042 if (frag->len <= 8) {
12043 /* Switch to the next fragment. */
12045 vcpu->mmio_cur_fragment++;
12047 /* Go forward to the next mmio piece. */
12053 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12054 vcpu->mmio_needed = 0;
12056 // VMG change, at this point, we're always done
12057 // RIP has already been advanced
12061 // More MMIO is needed
12062 run->mmio.phys_addr = frag->gpa;
12063 run->mmio.len = min(8u, frag->len);
12064 run->mmio.is_write = vcpu->mmio_is_write;
12065 if (run->mmio.is_write)
12066 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12067 run->exit_reason = KVM_EXIT_MMIO;
12069 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12074 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12078 struct kvm_mmio_fragment *frag;
12083 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12084 if (handled == bytes)
12091 /*TODO: Check if need to increment number of frags */
12092 frag = vcpu->mmio_fragments;
12093 vcpu->mmio_nr_fragments = 1;
12098 vcpu->mmio_needed = 1;
12099 vcpu->mmio_cur_fragment = 0;
12101 vcpu->run->mmio.phys_addr = gpa;
12102 vcpu->run->mmio.len = min(8u, frag->len);
12103 vcpu->run->mmio.is_write = 1;
12104 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12105 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12107 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12111 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12113 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12117 struct kvm_mmio_fragment *frag;
12122 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12123 if (handled == bytes)
12130 /*TODO: Check if need to increment number of frags */
12131 frag = vcpu->mmio_fragments;
12132 vcpu->mmio_nr_fragments = 1;
12137 vcpu->mmio_needed = 1;
12138 vcpu->mmio_cur_fragment = 0;
12140 vcpu->run->mmio.phys_addr = gpa;
12141 vcpu->run->mmio.len = min(8u, frag->len);
12142 vcpu->run->mmio.is_write = 0;
12143 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12145 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12149 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12151 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12153 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
12154 vcpu->arch.pio.count * vcpu->arch.pio.size);
12155 vcpu->arch.pio.count = 0;
12160 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12161 unsigned int port, void *data, unsigned int count)
12165 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
12170 vcpu->arch.pio.count = 0;
12175 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12176 unsigned int port, void *data, unsigned int count)
12180 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
12183 vcpu->arch.pio.count = 0;
12185 vcpu->arch.guest_ins_data = data;
12186 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12192 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12193 unsigned int port, void *data, unsigned int count,
12196 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
12197 : kvm_sev_es_outs(vcpu, size, port, data, count);
12199 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12201 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12202 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12203 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12204 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12205 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12206 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12207 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12208 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12209 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12210 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12211 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12212 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12213 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12214 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12215 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12216 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12217 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12218 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12219 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12220 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12221 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12222 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12223 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12224 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12225 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12226 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12227 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);