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 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1067 struct kvm_mmu *mmu = vcpu->arch.mmu;
1068 unsigned long roots_to_free = 0;
1072 * If neither the current CR3 nor any of the prev_roots use the given
1073 * PCID, then nothing needs to be done here because a resync will
1074 * happen anyway before switching to any other CR3.
1076 if (kvm_get_active_pcid(vcpu) == pcid) {
1077 kvm_mmu_sync_roots(vcpu);
1078 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1081 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1082 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1083 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1085 kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1088 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1090 bool skip_tlb_flush = false;
1091 unsigned long pcid = 0;
1092 #ifdef CONFIG_X86_64
1093 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1096 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1097 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1098 pcid = cr3 & X86_CR3_PCID_MASK;
1102 /* PDPTRs are always reloaded for PAE paging. */
1103 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1104 goto handle_tlb_flush;
1107 * Do not condition the GPA check on long mode, this helper is used to
1108 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1109 * the current vCPU mode is accurate.
1111 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1114 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1117 if (cr3 != kvm_read_cr3(vcpu))
1118 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1120 vcpu->arch.cr3 = cr3;
1121 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1125 * A load of CR3 that flushes the TLB flushes only the current PCID,
1126 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1127 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1128 * and it's impossible to use a non-zero PCID when PCID is disabled,
1129 * i.e. only PCID=0 can be relevant.
1131 if (!skip_tlb_flush)
1132 kvm_invalidate_pcid(vcpu, pcid);
1136 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1138 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1140 if (cr8 & CR8_RESERVED_BITS)
1142 if (lapic_in_kernel(vcpu))
1143 kvm_lapic_set_tpr(vcpu, cr8);
1145 vcpu->arch.cr8 = cr8;
1148 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1150 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1152 if (lapic_in_kernel(vcpu))
1153 return kvm_lapic_get_cr8(vcpu);
1155 return vcpu->arch.cr8;
1157 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1159 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1163 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1164 for (i = 0; i < KVM_NR_DB_REGS; i++)
1165 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1166 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1170 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1174 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1175 dr7 = vcpu->arch.guest_debug_dr7;
1177 dr7 = vcpu->arch.dr7;
1178 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1179 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1180 if (dr7 & DR7_BP_EN_MASK)
1181 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1183 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1185 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1187 u64 fixed = DR6_FIXED_1;
1189 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1192 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1193 fixed |= DR6_BUS_LOCK;
1197 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1199 size_t size = ARRAY_SIZE(vcpu->arch.db);
1203 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1204 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1205 vcpu->arch.eff_db[dr] = val;
1209 if (!kvm_dr6_valid(val))
1211 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1215 if (!kvm_dr7_valid(val))
1217 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1218 kvm_update_dr7(vcpu);
1224 EXPORT_SYMBOL_GPL(kvm_set_dr);
1226 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1228 size_t size = ARRAY_SIZE(vcpu->arch.db);
1232 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1236 *val = vcpu->arch.dr6;
1240 *val = vcpu->arch.dr7;
1244 EXPORT_SYMBOL_GPL(kvm_get_dr);
1246 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1248 u32 ecx = kvm_rcx_read(vcpu);
1251 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1252 kvm_inject_gp(vcpu, 0);
1256 kvm_rax_write(vcpu, (u32)data);
1257 kvm_rdx_write(vcpu, data >> 32);
1258 return kvm_skip_emulated_instruction(vcpu);
1260 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1263 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1264 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1266 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1267 * extract the supported MSRs from the related const lists.
1268 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1269 * capabilities of the host cpu. This capabilities test skips MSRs that are
1270 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1271 * may depend on host virtualization features rather than host cpu features.
1274 static const u32 msrs_to_save_all[] = {
1275 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1277 #ifdef CONFIG_X86_64
1278 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1280 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1281 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1283 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1284 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1285 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1286 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1287 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1288 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1289 MSR_IA32_UMWAIT_CONTROL,
1291 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1292 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1293 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1294 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1295 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1296 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1297 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1298 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1299 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1300 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1301 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1302 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1303 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1304 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1305 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1306 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1307 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1308 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1309 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1310 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1311 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1312 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1315 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1316 static unsigned num_msrs_to_save;
1318 static const u32 emulated_msrs_all[] = {
1319 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1320 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1321 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1322 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1323 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1324 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1325 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1327 HV_X64_MSR_VP_INDEX,
1328 HV_X64_MSR_VP_RUNTIME,
1329 HV_X64_MSR_SCONTROL,
1330 HV_X64_MSR_STIMER0_CONFIG,
1331 HV_X64_MSR_VP_ASSIST_PAGE,
1332 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1333 HV_X64_MSR_TSC_EMULATION_STATUS,
1334 HV_X64_MSR_SYNDBG_OPTIONS,
1335 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1336 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1337 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1339 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1340 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1342 MSR_IA32_TSC_ADJUST,
1343 MSR_IA32_TSC_DEADLINE,
1344 MSR_IA32_ARCH_CAPABILITIES,
1345 MSR_IA32_PERF_CAPABILITIES,
1346 MSR_IA32_MISC_ENABLE,
1347 MSR_IA32_MCG_STATUS,
1349 MSR_IA32_MCG_EXT_CTL,
1353 MSR_MISC_FEATURES_ENABLES,
1354 MSR_AMD64_VIRT_SPEC_CTRL,
1359 * The following list leaves out MSRs whose values are determined
1360 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1361 * We always support the "true" VMX control MSRs, even if the host
1362 * processor does not, so I am putting these registers here rather
1363 * than in msrs_to_save_all.
1366 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1367 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1368 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1369 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1371 MSR_IA32_VMX_CR0_FIXED0,
1372 MSR_IA32_VMX_CR4_FIXED0,
1373 MSR_IA32_VMX_VMCS_ENUM,
1374 MSR_IA32_VMX_PROCBASED_CTLS2,
1375 MSR_IA32_VMX_EPT_VPID_CAP,
1376 MSR_IA32_VMX_VMFUNC,
1379 MSR_KVM_POLL_CONTROL,
1382 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1383 static unsigned num_emulated_msrs;
1386 * List of msr numbers which are used to expose MSR-based features that
1387 * can be used by a hypervisor to validate requested CPU features.
1389 static const u32 msr_based_features_all[] = {
1391 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1392 MSR_IA32_VMX_PINBASED_CTLS,
1393 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1394 MSR_IA32_VMX_PROCBASED_CTLS,
1395 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1396 MSR_IA32_VMX_EXIT_CTLS,
1397 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1398 MSR_IA32_VMX_ENTRY_CTLS,
1400 MSR_IA32_VMX_CR0_FIXED0,
1401 MSR_IA32_VMX_CR0_FIXED1,
1402 MSR_IA32_VMX_CR4_FIXED0,
1403 MSR_IA32_VMX_CR4_FIXED1,
1404 MSR_IA32_VMX_VMCS_ENUM,
1405 MSR_IA32_VMX_PROCBASED_CTLS2,
1406 MSR_IA32_VMX_EPT_VPID_CAP,
1407 MSR_IA32_VMX_VMFUNC,
1411 MSR_IA32_ARCH_CAPABILITIES,
1412 MSR_IA32_PERF_CAPABILITIES,
1415 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1416 static unsigned int num_msr_based_features;
1418 static u64 kvm_get_arch_capabilities(void)
1422 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1423 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1426 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1427 * the nested hypervisor runs with NX huge pages. If it is not,
1428 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1429 * L1 guests, so it need not worry about its own (L2) guests.
1431 data |= ARCH_CAP_PSCHANGE_MC_NO;
1434 * If we're doing cache flushes (either "always" or "cond")
1435 * we will do one whenever the guest does a vmlaunch/vmresume.
1436 * If an outer hypervisor is doing the cache flush for us
1437 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1438 * capability to the guest too, and if EPT is disabled we're not
1439 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1440 * require a nested hypervisor to do a flush of its own.
1442 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1443 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1445 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1446 data |= ARCH_CAP_RDCL_NO;
1447 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1448 data |= ARCH_CAP_SSB_NO;
1449 if (!boot_cpu_has_bug(X86_BUG_MDS))
1450 data |= ARCH_CAP_MDS_NO;
1452 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1454 * If RTM=0 because the kernel has disabled TSX, the host might
1455 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1456 * and therefore knows that there cannot be TAA) but keep
1457 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1458 * and we want to allow migrating those guests to tsx=off hosts.
1460 data &= ~ARCH_CAP_TAA_NO;
1461 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1462 data |= ARCH_CAP_TAA_NO;
1465 * Nothing to do here; we emulate TSX_CTRL if present on the
1466 * host so the guest can choose between disabling TSX or
1467 * using VERW to clear CPU buffers.
1474 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1476 switch (msr->index) {
1477 case MSR_IA32_ARCH_CAPABILITIES:
1478 msr->data = kvm_get_arch_capabilities();
1480 case MSR_IA32_UCODE_REV:
1481 rdmsrl_safe(msr->index, &msr->data);
1484 return static_call(kvm_x86_get_msr_feature)(msr);
1489 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1491 struct kvm_msr_entry msr;
1495 r = kvm_get_msr_feature(&msr);
1497 if (r == KVM_MSR_RET_INVALID) {
1498 /* Unconditionally clear the output for simplicity */
1500 if (kvm_msr_ignored_check(index, 0, false))
1512 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1514 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1517 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1520 if (efer & (EFER_LME | EFER_LMA) &&
1521 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1524 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1530 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1532 if (efer & efer_reserved_bits)
1535 return __kvm_valid_efer(vcpu, efer);
1537 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1539 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1541 u64 old_efer = vcpu->arch.efer;
1542 u64 efer = msr_info->data;
1545 if (efer & efer_reserved_bits)
1548 if (!msr_info->host_initiated) {
1549 if (!__kvm_valid_efer(vcpu, efer))
1552 if (is_paging(vcpu) &&
1553 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1558 efer |= vcpu->arch.efer & EFER_LMA;
1560 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1566 /* Update reserved bits */
1567 if ((efer ^ old_efer) & EFER_NX)
1568 kvm_mmu_reset_context(vcpu);
1573 void kvm_enable_efer_bits(u64 mask)
1575 efer_reserved_bits &= ~mask;
1577 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1579 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1581 struct kvm_x86_msr_filter *msr_filter;
1582 struct msr_bitmap_range *ranges;
1583 struct kvm *kvm = vcpu->kvm;
1588 /* x2APIC MSRs do not support filtering. */
1589 if (index >= 0x800 && index <= 0x8ff)
1592 idx = srcu_read_lock(&kvm->srcu);
1594 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1600 allowed = msr_filter->default_allow;
1601 ranges = msr_filter->ranges;
1603 for (i = 0; i < msr_filter->count; i++) {
1604 u32 start = ranges[i].base;
1605 u32 end = start + ranges[i].nmsrs;
1606 u32 flags = ranges[i].flags;
1607 unsigned long *bitmap = ranges[i].bitmap;
1609 if ((index >= start) && (index < end) && (flags & type)) {
1610 allowed = !!test_bit(index - start, bitmap);
1616 srcu_read_unlock(&kvm->srcu, idx);
1620 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1623 * Write @data into the MSR specified by @index. Select MSR specific fault
1624 * checks are bypassed if @host_initiated is %true.
1625 * Returns 0 on success, non-0 otherwise.
1626 * Assumes vcpu_load() was already called.
1628 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1629 bool host_initiated)
1631 struct msr_data msr;
1633 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1634 return KVM_MSR_RET_FILTERED;
1639 case MSR_KERNEL_GS_BASE:
1642 if (is_noncanonical_address(data, vcpu))
1645 case MSR_IA32_SYSENTER_EIP:
1646 case MSR_IA32_SYSENTER_ESP:
1648 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1649 * non-canonical address is written on Intel but not on
1650 * AMD (which ignores the top 32-bits, because it does
1651 * not implement 64-bit SYSENTER).
1653 * 64-bit code should hence be able to write a non-canonical
1654 * value on AMD. Making the address canonical ensures that
1655 * vmentry does not fail on Intel after writing a non-canonical
1656 * value, and that something deterministic happens if the guest
1657 * invokes 64-bit SYSENTER.
1659 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1662 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1665 if (!host_initiated &&
1666 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1667 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1671 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1672 * incomplete and conflicting architectural behavior. Current
1673 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1674 * reserved and always read as zeros. Enforce Intel's reserved
1675 * bits check if and only if the guest CPU is Intel, and clear
1676 * the bits in all other cases. This ensures cross-vendor
1677 * migration will provide consistent behavior for the guest.
1679 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1688 msr.host_initiated = host_initiated;
1690 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1693 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1694 u32 index, u64 data, bool host_initiated)
1696 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1698 if (ret == KVM_MSR_RET_INVALID)
1699 if (kvm_msr_ignored_check(index, data, true))
1706 * Read the MSR specified by @index into @data. Select MSR specific fault
1707 * checks are bypassed if @host_initiated is %true.
1708 * Returns 0 on success, non-0 otherwise.
1709 * Assumes vcpu_load() was already called.
1711 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1712 bool host_initiated)
1714 struct msr_data msr;
1717 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1718 return KVM_MSR_RET_FILTERED;
1722 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1725 if (!host_initiated &&
1726 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1727 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1733 msr.host_initiated = host_initiated;
1735 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1741 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1742 u32 index, u64 *data, bool host_initiated)
1744 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1746 if (ret == KVM_MSR_RET_INVALID) {
1747 /* Unconditionally clear *data for simplicity */
1749 if (kvm_msr_ignored_check(index, 0, false))
1756 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1758 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1760 EXPORT_SYMBOL_GPL(kvm_get_msr);
1762 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1764 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1766 EXPORT_SYMBOL_GPL(kvm_set_msr);
1768 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1770 int err = vcpu->run->msr.error;
1772 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1773 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1776 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1779 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1781 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1784 static u64 kvm_msr_reason(int r)
1787 case KVM_MSR_RET_INVALID:
1788 return KVM_MSR_EXIT_REASON_UNKNOWN;
1789 case KVM_MSR_RET_FILTERED:
1790 return KVM_MSR_EXIT_REASON_FILTER;
1792 return KVM_MSR_EXIT_REASON_INVAL;
1796 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1797 u32 exit_reason, u64 data,
1798 int (*completion)(struct kvm_vcpu *vcpu),
1801 u64 msr_reason = kvm_msr_reason(r);
1803 /* Check if the user wanted to know about this MSR fault */
1804 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1807 vcpu->run->exit_reason = exit_reason;
1808 vcpu->run->msr.error = 0;
1809 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1810 vcpu->run->msr.reason = msr_reason;
1811 vcpu->run->msr.index = index;
1812 vcpu->run->msr.data = data;
1813 vcpu->arch.complete_userspace_io = completion;
1818 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1820 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1821 complete_emulated_rdmsr, r);
1824 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1826 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1827 complete_emulated_wrmsr, r);
1830 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1832 u32 ecx = kvm_rcx_read(vcpu);
1836 r = kvm_get_msr(vcpu, ecx, &data);
1838 /* MSR read failed? See if we should ask user space */
1839 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1840 /* Bounce to user space */
1845 trace_kvm_msr_read(ecx, data);
1847 kvm_rax_write(vcpu, data & -1u);
1848 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1850 trace_kvm_msr_read_ex(ecx);
1853 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1855 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1857 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1859 u32 ecx = kvm_rcx_read(vcpu);
1860 u64 data = kvm_read_edx_eax(vcpu);
1863 r = kvm_set_msr(vcpu, ecx, data);
1865 /* MSR write failed? See if we should ask user space */
1866 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1867 /* Bounce to user space */
1870 /* Signal all other negative errors to userspace */
1875 trace_kvm_msr_write(ecx, data);
1877 trace_kvm_msr_write_ex(ecx, data);
1879 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1881 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1883 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1885 return kvm_skip_emulated_instruction(vcpu);
1887 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1889 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1891 /* Treat an INVD instruction as a NOP and just skip it. */
1892 return kvm_emulate_as_nop(vcpu);
1894 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1896 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1898 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1899 return kvm_emulate_as_nop(vcpu);
1901 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1903 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1905 kvm_queue_exception(vcpu, UD_VECTOR);
1908 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1910 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1912 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1913 return kvm_emulate_as_nop(vcpu);
1915 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1917 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1919 xfer_to_guest_mode_prepare();
1920 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1921 xfer_to_guest_mode_work_pending();
1925 * The fast path for frequent and performance sensitive wrmsr emulation,
1926 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1927 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1928 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1929 * other cases which must be called after interrupts are enabled on the host.
1931 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1933 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1936 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1937 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1938 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1939 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1942 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1943 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1944 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1945 trace_kvm_apic_write(APIC_ICR, (u32)data);
1952 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1954 if (!kvm_can_use_hv_timer(vcpu))
1957 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1961 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1963 u32 msr = kvm_rcx_read(vcpu);
1965 fastpath_t ret = EXIT_FASTPATH_NONE;
1968 case APIC_BASE_MSR + (APIC_ICR >> 4):
1969 data = kvm_read_edx_eax(vcpu);
1970 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1971 kvm_skip_emulated_instruction(vcpu);
1972 ret = EXIT_FASTPATH_EXIT_HANDLED;
1975 case MSR_IA32_TSC_DEADLINE:
1976 data = kvm_read_edx_eax(vcpu);
1977 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1978 kvm_skip_emulated_instruction(vcpu);
1979 ret = EXIT_FASTPATH_REENTER_GUEST;
1986 if (ret != EXIT_FASTPATH_NONE)
1987 trace_kvm_msr_write(msr, data);
1991 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1994 * Adapt set_msr() to msr_io()'s calling convention
1996 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1998 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2001 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2003 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2006 #ifdef CONFIG_X86_64
2007 struct pvclock_clock {
2017 struct pvclock_gtod_data {
2020 struct pvclock_clock clock; /* extract of a clocksource struct */
2021 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2027 static struct pvclock_gtod_data pvclock_gtod_data;
2029 static void update_pvclock_gtod(struct timekeeper *tk)
2031 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2033 write_seqcount_begin(&vdata->seq);
2035 /* copy pvclock gtod data */
2036 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2037 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2038 vdata->clock.mask = tk->tkr_mono.mask;
2039 vdata->clock.mult = tk->tkr_mono.mult;
2040 vdata->clock.shift = tk->tkr_mono.shift;
2041 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2042 vdata->clock.offset = tk->tkr_mono.base;
2044 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2045 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2046 vdata->raw_clock.mask = tk->tkr_raw.mask;
2047 vdata->raw_clock.mult = tk->tkr_raw.mult;
2048 vdata->raw_clock.shift = tk->tkr_raw.shift;
2049 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2050 vdata->raw_clock.offset = tk->tkr_raw.base;
2052 vdata->wall_time_sec = tk->xtime_sec;
2054 vdata->offs_boot = tk->offs_boot;
2056 write_seqcount_end(&vdata->seq);
2059 static s64 get_kvmclock_base_ns(void)
2061 /* Count up from boot time, but with the frequency of the raw clock. */
2062 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2065 static s64 get_kvmclock_base_ns(void)
2067 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2068 return ktime_get_boottime_ns();
2072 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2076 struct pvclock_wall_clock wc;
2083 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2088 ++version; /* first time write, random junk */
2092 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2096 * The guest calculates current wall clock time by adding
2097 * system time (updated by kvm_guest_time_update below) to the
2098 * wall clock specified here. We do the reverse here.
2100 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2102 wc.nsec = do_div(wall_nsec, 1000000000);
2103 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2104 wc.version = version;
2106 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2109 wc_sec_hi = wall_nsec >> 32;
2110 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2111 &wc_sec_hi, sizeof(wc_sec_hi));
2115 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2118 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2119 bool old_msr, bool host_initiated)
2121 struct kvm_arch *ka = &vcpu->kvm->arch;
2123 if (vcpu->vcpu_id == 0 && !host_initiated) {
2124 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2125 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2127 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2130 vcpu->arch.time = system_time;
2131 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2133 /* we verify if the enable bit is set... */
2134 vcpu->arch.pv_time_enabled = false;
2135 if (!(system_time & 1))
2138 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2139 &vcpu->arch.pv_time, system_time & ~1ULL,
2140 sizeof(struct pvclock_vcpu_time_info)))
2141 vcpu->arch.pv_time_enabled = true;
2146 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2148 do_shl32_div32(dividend, divisor);
2152 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2153 s8 *pshift, u32 *pmultiplier)
2161 scaled64 = scaled_hz;
2162 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2167 tps32 = (uint32_t)tps64;
2168 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2169 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2177 *pmultiplier = div_frac(scaled64, tps32);
2180 #ifdef CONFIG_X86_64
2181 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2184 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2185 static unsigned long max_tsc_khz;
2187 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2189 u64 v = (u64)khz * (1000000 + ppm);
2194 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2196 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2200 /* Guest TSC same frequency as host TSC? */
2202 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2206 /* TSC scaling supported? */
2207 if (!kvm_has_tsc_control) {
2208 if (user_tsc_khz > tsc_khz) {
2209 vcpu->arch.tsc_catchup = 1;
2210 vcpu->arch.tsc_always_catchup = 1;
2213 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2218 /* TSC scaling required - calculate ratio */
2219 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2220 user_tsc_khz, tsc_khz);
2222 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2223 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2228 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2232 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2234 u32 thresh_lo, thresh_hi;
2235 int use_scaling = 0;
2237 /* tsc_khz can be zero if TSC calibration fails */
2238 if (user_tsc_khz == 0) {
2239 /* set tsc_scaling_ratio to a safe value */
2240 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2244 /* Compute a scale to convert nanoseconds in TSC cycles */
2245 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2246 &vcpu->arch.virtual_tsc_shift,
2247 &vcpu->arch.virtual_tsc_mult);
2248 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2251 * Compute the variation in TSC rate which is acceptable
2252 * within the range of tolerance and decide if the
2253 * rate being applied is within that bounds of the hardware
2254 * rate. If so, no scaling or compensation need be done.
2256 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2257 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2258 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2259 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2262 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2265 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2267 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2268 vcpu->arch.virtual_tsc_mult,
2269 vcpu->arch.virtual_tsc_shift);
2270 tsc += vcpu->arch.this_tsc_write;
2274 static inline int gtod_is_based_on_tsc(int mode)
2276 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2279 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2281 #ifdef CONFIG_X86_64
2283 struct kvm_arch *ka = &vcpu->kvm->arch;
2284 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2286 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2287 atomic_read(&vcpu->kvm->online_vcpus));
2290 * Once the masterclock is enabled, always perform request in
2291 * order to update it.
2293 * In order to enable masterclock, the host clocksource must be TSC
2294 * and the vcpus need to have matched TSCs. When that happens,
2295 * perform request to enable masterclock.
2297 if (ka->use_master_clock ||
2298 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2299 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2301 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2302 atomic_read(&vcpu->kvm->online_vcpus),
2303 ka->use_master_clock, gtod->clock.vclock_mode);
2308 * Multiply tsc by a fixed point number represented by ratio.
2310 * The most significant 64-N bits (mult) of ratio represent the
2311 * integral part of the fixed point number; the remaining N bits
2312 * (frac) represent the fractional part, ie. ratio represents a fixed
2313 * point number (mult + frac * 2^(-N)).
2315 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2317 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2319 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2322 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2326 if (ratio != kvm_default_tsc_scaling_ratio)
2327 _tsc = __scale_tsc(ratio, tsc);
2331 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2333 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2337 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2339 return target_tsc - tsc;
2342 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2344 return vcpu->arch.l1_tsc_offset +
2345 kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2347 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2349 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2353 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2354 nested_offset = l1_offset;
2356 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2357 kvm_tsc_scaling_ratio_frac_bits);
2359 nested_offset += l2_offset;
2360 return nested_offset;
2362 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2364 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2366 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2367 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2368 kvm_tsc_scaling_ratio_frac_bits);
2370 return l1_multiplier;
2372 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2374 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2376 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2377 vcpu->arch.l1_tsc_offset,
2380 vcpu->arch.l1_tsc_offset = l1_offset;
2383 * If we are here because L1 chose not to trap WRMSR to TSC then
2384 * according to the spec this should set L1's TSC (as opposed to
2385 * setting L1's offset for L2).
2387 if (is_guest_mode(vcpu))
2388 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2390 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2391 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2393 vcpu->arch.tsc_offset = l1_offset;
2395 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2398 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2400 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2402 /* Userspace is changing the multiplier while L2 is active */
2403 if (is_guest_mode(vcpu))
2404 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2406 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2408 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2410 if (kvm_has_tsc_control)
2411 static_call(kvm_x86_write_tsc_multiplier)(
2412 vcpu, vcpu->arch.tsc_scaling_ratio);
2415 static inline bool kvm_check_tsc_unstable(void)
2417 #ifdef CONFIG_X86_64
2419 * TSC is marked unstable when we're running on Hyper-V,
2420 * 'TSC page' clocksource is good.
2422 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2425 return check_tsc_unstable();
2428 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2430 struct kvm *kvm = vcpu->kvm;
2431 u64 offset, ns, elapsed;
2432 unsigned long flags;
2434 bool already_matched;
2435 bool synchronizing = false;
2437 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2438 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2439 ns = get_kvmclock_base_ns();
2440 elapsed = ns - kvm->arch.last_tsc_nsec;
2442 if (vcpu->arch.virtual_tsc_khz) {
2445 * detection of vcpu initialization -- need to sync
2446 * with other vCPUs. This particularly helps to keep
2447 * kvm_clock stable after CPU hotplug
2449 synchronizing = true;
2451 u64 tsc_exp = kvm->arch.last_tsc_write +
2452 nsec_to_cycles(vcpu, elapsed);
2453 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2455 * Special case: TSC write with a small delta (1 second)
2456 * of virtual cycle time against real time is
2457 * interpreted as an attempt to synchronize the CPU.
2459 synchronizing = data < tsc_exp + tsc_hz &&
2460 data + tsc_hz > tsc_exp;
2465 * For a reliable TSC, we can match TSC offsets, and for an unstable
2466 * TSC, we add elapsed time in this computation. We could let the
2467 * compensation code attempt to catch up if we fall behind, but
2468 * it's better to try to match offsets from the beginning.
2470 if (synchronizing &&
2471 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2472 if (!kvm_check_tsc_unstable()) {
2473 offset = kvm->arch.cur_tsc_offset;
2475 u64 delta = nsec_to_cycles(vcpu, elapsed);
2477 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2480 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2483 * We split periods of matched TSC writes into generations.
2484 * For each generation, we track the original measured
2485 * nanosecond time, offset, and write, so if TSCs are in
2486 * sync, we can match exact offset, and if not, we can match
2487 * exact software computation in compute_guest_tsc()
2489 * These values are tracked in kvm->arch.cur_xxx variables.
2491 kvm->arch.cur_tsc_generation++;
2492 kvm->arch.cur_tsc_nsec = ns;
2493 kvm->arch.cur_tsc_write = data;
2494 kvm->arch.cur_tsc_offset = offset;
2499 * We also track th most recent recorded KHZ, write and time to
2500 * allow the matching interval to be extended at each write.
2502 kvm->arch.last_tsc_nsec = ns;
2503 kvm->arch.last_tsc_write = data;
2504 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2506 vcpu->arch.last_guest_tsc = data;
2508 /* Keep track of which generation this VCPU has synchronized to */
2509 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2510 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2511 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2513 kvm_vcpu_write_tsc_offset(vcpu, offset);
2514 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2516 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2518 kvm->arch.nr_vcpus_matched_tsc = 0;
2519 } else if (!already_matched) {
2520 kvm->arch.nr_vcpus_matched_tsc++;
2523 kvm_track_tsc_matching(vcpu);
2524 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2527 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2530 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2531 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2534 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2536 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2537 WARN_ON(adjustment < 0);
2538 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2539 vcpu->arch.l1_tsc_scaling_ratio);
2540 adjust_tsc_offset_guest(vcpu, adjustment);
2543 #ifdef CONFIG_X86_64
2545 static u64 read_tsc(void)
2547 u64 ret = (u64)rdtsc_ordered();
2548 u64 last = pvclock_gtod_data.clock.cycle_last;
2550 if (likely(ret >= last))
2554 * GCC likes to generate cmov here, but this branch is extremely
2555 * predictable (it's just a function of time and the likely is
2556 * very likely) and there's a data dependence, so force GCC
2557 * to generate a branch instead. I don't barrier() because
2558 * we don't actually need a barrier, and if this function
2559 * ever gets inlined it will generate worse code.
2565 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2571 switch (clock->vclock_mode) {
2572 case VDSO_CLOCKMODE_HVCLOCK:
2573 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2575 if (tsc_pg_val != U64_MAX) {
2576 /* TSC page valid */
2577 *mode = VDSO_CLOCKMODE_HVCLOCK;
2578 v = (tsc_pg_val - clock->cycle_last) &
2581 /* TSC page invalid */
2582 *mode = VDSO_CLOCKMODE_NONE;
2585 case VDSO_CLOCKMODE_TSC:
2586 *mode = VDSO_CLOCKMODE_TSC;
2587 *tsc_timestamp = read_tsc();
2588 v = (*tsc_timestamp - clock->cycle_last) &
2592 *mode = VDSO_CLOCKMODE_NONE;
2595 if (*mode == VDSO_CLOCKMODE_NONE)
2596 *tsc_timestamp = v = 0;
2598 return v * clock->mult;
2601 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2603 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2609 seq = read_seqcount_begin(>od->seq);
2610 ns = gtod->raw_clock.base_cycles;
2611 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2612 ns >>= gtod->raw_clock.shift;
2613 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2614 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2620 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2622 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2628 seq = read_seqcount_begin(>od->seq);
2629 ts->tv_sec = gtod->wall_time_sec;
2630 ns = gtod->clock.base_cycles;
2631 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2632 ns >>= gtod->clock.shift;
2633 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2635 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2641 /* returns true if host is using TSC based clocksource */
2642 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2644 /* checked again under seqlock below */
2645 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2648 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2652 /* returns true if host is using TSC based clocksource */
2653 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2656 /* checked again under seqlock below */
2657 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2660 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2666 * Assuming a stable TSC across physical CPUS, and a stable TSC
2667 * across virtual CPUs, the following condition is possible.
2668 * Each numbered line represents an event visible to both
2669 * CPUs at the next numbered event.
2671 * "timespecX" represents host monotonic time. "tscX" represents
2674 * VCPU0 on CPU0 | VCPU1 on CPU1
2676 * 1. read timespec0,tsc0
2677 * 2. | timespec1 = timespec0 + N
2679 * 3. transition to guest | transition to guest
2680 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2681 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2682 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2684 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2687 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2689 * - 0 < N - M => M < N
2691 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2692 * always the case (the difference between two distinct xtime instances
2693 * might be smaller then the difference between corresponding TSC reads,
2694 * when updating guest vcpus pvclock areas).
2696 * To avoid that problem, do not allow visibility of distinct
2697 * system_timestamp/tsc_timestamp values simultaneously: use a master
2698 * copy of host monotonic time values. Update that master copy
2701 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2705 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2707 #ifdef CONFIG_X86_64
2708 struct kvm_arch *ka = &kvm->arch;
2710 bool host_tsc_clocksource, vcpus_matched;
2712 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2713 atomic_read(&kvm->online_vcpus));
2716 * If the host uses TSC clock, then passthrough TSC as stable
2719 host_tsc_clocksource = kvm_get_time_and_clockread(
2720 &ka->master_kernel_ns,
2721 &ka->master_cycle_now);
2723 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2724 && !ka->backwards_tsc_observed
2725 && !ka->boot_vcpu_runs_old_kvmclock;
2727 if (ka->use_master_clock)
2728 atomic_set(&kvm_guest_has_master_clock, 1);
2730 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2731 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2736 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2738 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2741 static void kvm_gen_update_masterclock(struct kvm *kvm)
2743 #ifdef CONFIG_X86_64
2745 struct kvm_vcpu *vcpu;
2746 struct kvm_arch *ka = &kvm->arch;
2747 unsigned long flags;
2749 kvm_hv_invalidate_tsc_page(kvm);
2751 kvm_make_mclock_inprogress_request(kvm);
2753 /* no guest entries from this point */
2754 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2755 pvclock_update_vm_gtod_copy(kvm);
2756 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2758 kvm_for_each_vcpu(i, vcpu, kvm)
2759 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2761 /* guest entries allowed */
2762 kvm_for_each_vcpu(i, vcpu, kvm)
2763 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2767 u64 get_kvmclock_ns(struct kvm *kvm)
2769 struct kvm_arch *ka = &kvm->arch;
2770 struct pvclock_vcpu_time_info hv_clock;
2771 unsigned long flags;
2774 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2775 if (!ka->use_master_clock) {
2776 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2777 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2780 hv_clock.tsc_timestamp = ka->master_cycle_now;
2781 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2782 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2784 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2787 if (__this_cpu_read(cpu_tsc_khz)) {
2788 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2789 &hv_clock.tsc_shift,
2790 &hv_clock.tsc_to_system_mul);
2791 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2793 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2800 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2801 struct gfn_to_hva_cache *cache,
2802 unsigned int offset)
2804 struct kvm_vcpu_arch *vcpu = &v->arch;
2805 struct pvclock_vcpu_time_info guest_hv_clock;
2807 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2808 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2811 /* This VCPU is paused, but it's legal for a guest to read another
2812 * VCPU's kvmclock, so we really have to follow the specification where
2813 * it says that version is odd if data is being modified, and even after
2816 * Version field updates must be kept separate. This is because
2817 * kvm_write_guest_cached might use a "rep movs" instruction, and
2818 * writes within a string instruction are weakly ordered. So there
2819 * are three writes overall.
2821 * As a small optimization, only write the version field in the first
2822 * and third write. The vcpu->pv_time cache is still valid, because the
2823 * version field is the first in the struct.
2825 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2827 if (guest_hv_clock.version & 1)
2828 ++guest_hv_clock.version; /* first time write, random junk */
2830 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2831 kvm_write_guest_offset_cached(v->kvm, cache,
2832 &vcpu->hv_clock, offset,
2833 sizeof(vcpu->hv_clock.version));
2837 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2838 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2840 if (vcpu->pvclock_set_guest_stopped_request) {
2841 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2842 vcpu->pvclock_set_guest_stopped_request = false;
2845 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2847 kvm_write_guest_offset_cached(v->kvm, cache,
2848 &vcpu->hv_clock, offset,
2849 sizeof(vcpu->hv_clock));
2853 vcpu->hv_clock.version++;
2854 kvm_write_guest_offset_cached(v->kvm, cache,
2855 &vcpu->hv_clock, offset,
2856 sizeof(vcpu->hv_clock.version));
2859 static int kvm_guest_time_update(struct kvm_vcpu *v)
2861 unsigned long flags, tgt_tsc_khz;
2862 struct kvm_vcpu_arch *vcpu = &v->arch;
2863 struct kvm_arch *ka = &v->kvm->arch;
2865 u64 tsc_timestamp, host_tsc;
2867 bool use_master_clock;
2873 * If the host uses TSC clock, then passthrough TSC as stable
2876 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2877 use_master_clock = ka->use_master_clock;
2878 if (use_master_clock) {
2879 host_tsc = ka->master_cycle_now;
2880 kernel_ns = ka->master_kernel_ns;
2882 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2884 /* Keep irq disabled to prevent changes to the clock */
2885 local_irq_save(flags);
2886 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2887 if (unlikely(tgt_tsc_khz == 0)) {
2888 local_irq_restore(flags);
2889 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2892 if (!use_master_clock) {
2894 kernel_ns = get_kvmclock_base_ns();
2897 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2900 * We may have to catch up the TSC to match elapsed wall clock
2901 * time for two reasons, even if kvmclock is used.
2902 * 1) CPU could have been running below the maximum TSC rate
2903 * 2) Broken TSC compensation resets the base at each VCPU
2904 * entry to avoid unknown leaps of TSC even when running
2905 * again on the same CPU. This may cause apparent elapsed
2906 * time to disappear, and the guest to stand still or run
2909 if (vcpu->tsc_catchup) {
2910 u64 tsc = compute_guest_tsc(v, kernel_ns);
2911 if (tsc > tsc_timestamp) {
2912 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2913 tsc_timestamp = tsc;
2917 local_irq_restore(flags);
2919 /* With all the info we got, fill in the values */
2921 if (kvm_has_tsc_control)
2922 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
2923 v->arch.l1_tsc_scaling_ratio);
2925 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2926 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2927 &vcpu->hv_clock.tsc_shift,
2928 &vcpu->hv_clock.tsc_to_system_mul);
2929 vcpu->hw_tsc_khz = tgt_tsc_khz;
2932 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2933 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2934 vcpu->last_guest_tsc = tsc_timestamp;
2936 /* If the host uses TSC clocksource, then it is stable */
2938 if (use_master_clock)
2939 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2941 vcpu->hv_clock.flags = pvclock_flags;
2943 if (vcpu->pv_time_enabled)
2944 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2945 if (vcpu->xen.vcpu_info_set)
2946 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2947 offsetof(struct compat_vcpu_info, time));
2948 if (vcpu->xen.vcpu_time_info_set)
2949 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2950 if (v == kvm_get_vcpu(v->kvm, 0))
2951 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2956 * kvmclock updates which are isolated to a given vcpu, such as
2957 * vcpu->cpu migration, should not allow system_timestamp from
2958 * the rest of the vcpus to remain static. Otherwise ntp frequency
2959 * correction applies to one vcpu's system_timestamp but not
2962 * So in those cases, request a kvmclock update for all vcpus.
2963 * We need to rate-limit these requests though, as they can
2964 * considerably slow guests that have a large number of vcpus.
2965 * The time for a remote vcpu to update its kvmclock is bound
2966 * by the delay we use to rate-limit the updates.
2969 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2971 static void kvmclock_update_fn(struct work_struct *work)
2974 struct delayed_work *dwork = to_delayed_work(work);
2975 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2976 kvmclock_update_work);
2977 struct kvm *kvm = container_of(ka, struct kvm, arch);
2978 struct kvm_vcpu *vcpu;
2980 kvm_for_each_vcpu(i, vcpu, kvm) {
2981 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2982 kvm_vcpu_kick(vcpu);
2986 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2988 struct kvm *kvm = v->kvm;
2990 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2991 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2992 KVMCLOCK_UPDATE_DELAY);
2995 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2997 static void kvmclock_sync_fn(struct work_struct *work)
2999 struct delayed_work *dwork = to_delayed_work(work);
3000 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3001 kvmclock_sync_work);
3002 struct kvm *kvm = container_of(ka, struct kvm, arch);
3004 if (!kvmclock_periodic_sync)
3007 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3008 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3009 KVMCLOCK_SYNC_PERIOD);
3013 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3015 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3017 /* McStatusWrEn enabled? */
3018 if (guest_cpuid_is_amd_or_hygon(vcpu))
3019 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3024 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3026 u64 mcg_cap = vcpu->arch.mcg_cap;
3027 unsigned bank_num = mcg_cap & 0xff;
3028 u32 msr = msr_info->index;
3029 u64 data = msr_info->data;
3032 case MSR_IA32_MCG_STATUS:
3033 vcpu->arch.mcg_status = data;
3035 case MSR_IA32_MCG_CTL:
3036 if (!(mcg_cap & MCG_CTL_P) &&
3037 (data || !msr_info->host_initiated))
3039 if (data != 0 && data != ~(u64)0)
3041 vcpu->arch.mcg_ctl = data;
3044 if (msr >= MSR_IA32_MC0_CTL &&
3045 msr < MSR_IA32_MCx_CTL(bank_num)) {
3046 u32 offset = array_index_nospec(
3047 msr - MSR_IA32_MC0_CTL,
3048 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3050 /* only 0 or all 1s can be written to IA32_MCi_CTL
3051 * some Linux kernels though clear bit 10 in bank 4 to
3052 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3053 * this to avoid an uncatched #GP in the guest
3055 if ((offset & 0x3) == 0 &&
3056 data != 0 && (data | (1 << 10)) != ~(u64)0)
3060 if (!msr_info->host_initiated &&
3061 (offset & 0x3) == 1 && data != 0) {
3062 if (!can_set_mci_status(vcpu))
3066 vcpu->arch.mce_banks[offset] = data;
3074 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3076 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3078 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3081 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3083 gpa_t gpa = data & ~0x3f;
3085 /* Bits 4:5 are reserved, Should be zero */
3089 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3090 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3093 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3094 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3097 if (!lapic_in_kernel(vcpu))
3098 return data ? 1 : 0;
3100 vcpu->arch.apf.msr_en_val = data;
3102 if (!kvm_pv_async_pf_enabled(vcpu)) {
3103 kvm_clear_async_pf_completion_queue(vcpu);
3104 kvm_async_pf_hash_reset(vcpu);
3108 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3112 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3113 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3115 kvm_async_pf_wakeup_all(vcpu);
3120 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3122 /* Bits 8-63 are reserved */
3126 if (!lapic_in_kernel(vcpu))
3129 vcpu->arch.apf.msr_int_val = data;
3131 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3136 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3138 vcpu->arch.pv_time_enabled = false;
3139 vcpu->arch.time = 0;
3142 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3144 ++vcpu->stat.tlb_flush;
3145 static_call(kvm_x86_tlb_flush_all)(vcpu);
3148 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3150 ++vcpu->stat.tlb_flush;
3154 * A TLB flush on behalf of the guest is equivalent to
3155 * INVPCID(all), toggling CR4.PGE, etc., which requires
3156 * a forced sync of the shadow page tables. Unload the
3157 * entire MMU here and the subsequent load will sync the
3158 * shadow page tables, and also flush the TLB.
3160 kvm_mmu_unload(vcpu);
3164 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3167 static void record_steal_time(struct kvm_vcpu *vcpu)
3169 struct kvm_host_map map;
3170 struct kvm_steal_time *st;
3172 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3173 kvm_xen_runstate_set_running(vcpu);
3177 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3180 /* -EAGAIN is returned in atomic context so we can just return. */
3181 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3182 &map, &vcpu->arch.st.cache, false))
3186 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3189 * Doing a TLB flush here, on the guest's behalf, can avoid
3192 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3193 u8 st_preempted = xchg(&st->preempted, 0);
3195 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3196 st_preempted & KVM_VCPU_FLUSH_TLB);
3197 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3198 kvm_vcpu_flush_tlb_guest(vcpu);
3203 vcpu->arch.st.preempted = 0;
3205 if (st->version & 1)
3206 st->version += 1; /* first time write, random junk */
3212 st->steal += current->sched_info.run_delay -
3213 vcpu->arch.st.last_steal;
3214 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3220 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3223 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3226 u32 msr = msr_info->index;
3227 u64 data = msr_info->data;
3229 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3230 return kvm_xen_write_hypercall_page(vcpu, data);
3233 case MSR_AMD64_NB_CFG:
3234 case MSR_IA32_UCODE_WRITE:
3235 case MSR_VM_HSAVE_PA:
3236 case MSR_AMD64_PATCH_LOADER:
3237 case MSR_AMD64_BU_CFG2:
3238 case MSR_AMD64_DC_CFG:
3239 case MSR_F15H_EX_CFG:
3242 case MSR_IA32_UCODE_REV:
3243 if (msr_info->host_initiated)
3244 vcpu->arch.microcode_version = data;
3246 case MSR_IA32_ARCH_CAPABILITIES:
3247 if (!msr_info->host_initiated)
3249 vcpu->arch.arch_capabilities = data;
3251 case MSR_IA32_PERF_CAPABILITIES: {
3252 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3254 if (!msr_info->host_initiated)
3256 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3258 if (data & ~msr_ent.data)
3261 vcpu->arch.perf_capabilities = data;
3266 return set_efer(vcpu, msr_info);
3268 data &= ~(u64)0x40; /* ignore flush filter disable */
3269 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3270 data &= ~(u64)0x8; /* ignore TLB cache disable */
3272 /* Handle McStatusWrEn */
3273 if (data == BIT_ULL(18)) {
3274 vcpu->arch.msr_hwcr = data;
3275 } else if (data != 0) {
3276 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3281 case MSR_FAM10H_MMIO_CONF_BASE:
3283 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3288 case 0x200 ... 0x2ff:
3289 return kvm_mtrr_set_msr(vcpu, msr, data);
3290 case MSR_IA32_APICBASE:
3291 return kvm_set_apic_base(vcpu, msr_info);
3292 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3293 return kvm_x2apic_msr_write(vcpu, msr, data);
3294 case MSR_IA32_TSC_DEADLINE:
3295 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3297 case MSR_IA32_TSC_ADJUST:
3298 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3299 if (!msr_info->host_initiated) {
3300 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3301 adjust_tsc_offset_guest(vcpu, adj);
3303 vcpu->arch.ia32_tsc_adjust_msr = data;
3306 case MSR_IA32_MISC_ENABLE:
3307 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3308 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3309 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3311 vcpu->arch.ia32_misc_enable_msr = data;
3312 kvm_update_cpuid_runtime(vcpu);
3314 vcpu->arch.ia32_misc_enable_msr = data;
3317 case MSR_IA32_SMBASE:
3318 if (!msr_info->host_initiated)
3320 vcpu->arch.smbase = data;
3322 case MSR_IA32_POWER_CTL:
3323 vcpu->arch.msr_ia32_power_ctl = data;
3326 if (msr_info->host_initiated) {
3327 kvm_synchronize_tsc(vcpu, data);
3329 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3330 adjust_tsc_offset_guest(vcpu, adj);
3331 vcpu->arch.ia32_tsc_adjust_msr += adj;
3335 if (!msr_info->host_initiated &&
3336 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3339 * KVM supports exposing PT to the guest, but does not support
3340 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3341 * XSAVES/XRSTORS to save/restore PT MSRs.
3343 if (data & ~supported_xss)
3345 vcpu->arch.ia32_xss = data;
3348 if (!msr_info->host_initiated)
3350 vcpu->arch.smi_count = data;
3352 case MSR_KVM_WALL_CLOCK_NEW:
3353 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3356 vcpu->kvm->arch.wall_clock = data;
3357 kvm_write_wall_clock(vcpu->kvm, data, 0);
3359 case MSR_KVM_WALL_CLOCK:
3360 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3363 vcpu->kvm->arch.wall_clock = data;
3364 kvm_write_wall_clock(vcpu->kvm, data, 0);
3366 case MSR_KVM_SYSTEM_TIME_NEW:
3367 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3370 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3372 case MSR_KVM_SYSTEM_TIME:
3373 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3376 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3378 case MSR_KVM_ASYNC_PF_EN:
3379 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3382 if (kvm_pv_enable_async_pf(vcpu, data))
3385 case MSR_KVM_ASYNC_PF_INT:
3386 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3389 if (kvm_pv_enable_async_pf_int(vcpu, data))
3392 case MSR_KVM_ASYNC_PF_ACK:
3393 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3396 vcpu->arch.apf.pageready_pending = false;
3397 kvm_check_async_pf_completion(vcpu);
3400 case MSR_KVM_STEAL_TIME:
3401 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3404 if (unlikely(!sched_info_on()))
3407 if (data & KVM_STEAL_RESERVED_MASK)
3410 vcpu->arch.st.msr_val = data;
3412 if (!(data & KVM_MSR_ENABLED))
3415 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3418 case MSR_KVM_PV_EOI_EN:
3419 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3422 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3426 case MSR_KVM_POLL_CONTROL:
3427 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3430 /* only enable bit supported */
3431 if (data & (-1ULL << 1))
3434 vcpu->arch.msr_kvm_poll_control = data;
3437 case MSR_IA32_MCG_CTL:
3438 case MSR_IA32_MCG_STATUS:
3439 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3440 return set_msr_mce(vcpu, msr_info);
3442 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3443 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3446 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3447 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3448 if (kvm_pmu_is_valid_msr(vcpu, msr))
3449 return kvm_pmu_set_msr(vcpu, msr_info);
3451 if (pr || data != 0)
3452 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3453 "0x%x data 0x%llx\n", msr, data);
3455 case MSR_K7_CLK_CTL:
3457 * Ignore all writes to this no longer documented MSR.
3458 * Writes are only relevant for old K7 processors,
3459 * all pre-dating SVM, but a recommended workaround from
3460 * AMD for these chips. It is possible to specify the
3461 * affected processor models on the command line, hence
3462 * the need to ignore the workaround.
3465 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3466 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3467 case HV_X64_MSR_SYNDBG_OPTIONS:
3468 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3469 case HV_X64_MSR_CRASH_CTL:
3470 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3471 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3472 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3473 case HV_X64_MSR_TSC_EMULATION_STATUS:
3474 return kvm_hv_set_msr_common(vcpu, msr, data,
3475 msr_info->host_initiated);
3476 case MSR_IA32_BBL_CR_CTL3:
3477 /* Drop writes to this legacy MSR -- see rdmsr
3478 * counterpart for further detail.
3480 if (report_ignored_msrs)
3481 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3484 case MSR_AMD64_OSVW_ID_LENGTH:
3485 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3487 vcpu->arch.osvw.length = data;
3489 case MSR_AMD64_OSVW_STATUS:
3490 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3492 vcpu->arch.osvw.status = data;
3494 case MSR_PLATFORM_INFO:
3495 if (!msr_info->host_initiated ||
3496 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3497 cpuid_fault_enabled(vcpu)))
3499 vcpu->arch.msr_platform_info = data;
3501 case MSR_MISC_FEATURES_ENABLES:
3502 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3503 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3504 !supports_cpuid_fault(vcpu)))
3506 vcpu->arch.msr_misc_features_enables = data;
3509 if (kvm_pmu_is_valid_msr(vcpu, msr))
3510 return kvm_pmu_set_msr(vcpu, msr_info);
3511 return KVM_MSR_RET_INVALID;
3515 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3517 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3520 u64 mcg_cap = vcpu->arch.mcg_cap;
3521 unsigned bank_num = mcg_cap & 0xff;
3524 case MSR_IA32_P5_MC_ADDR:
3525 case MSR_IA32_P5_MC_TYPE:
3528 case MSR_IA32_MCG_CAP:
3529 data = vcpu->arch.mcg_cap;
3531 case MSR_IA32_MCG_CTL:
3532 if (!(mcg_cap & MCG_CTL_P) && !host)
3534 data = vcpu->arch.mcg_ctl;
3536 case MSR_IA32_MCG_STATUS:
3537 data = vcpu->arch.mcg_status;
3540 if (msr >= MSR_IA32_MC0_CTL &&
3541 msr < MSR_IA32_MCx_CTL(bank_num)) {
3542 u32 offset = array_index_nospec(
3543 msr - MSR_IA32_MC0_CTL,
3544 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3546 data = vcpu->arch.mce_banks[offset];
3555 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3557 switch (msr_info->index) {
3558 case MSR_IA32_PLATFORM_ID:
3559 case MSR_IA32_EBL_CR_POWERON:
3560 case MSR_IA32_LASTBRANCHFROMIP:
3561 case MSR_IA32_LASTBRANCHTOIP:
3562 case MSR_IA32_LASTINTFROMIP:
3563 case MSR_IA32_LASTINTTOIP:
3565 case MSR_K8_TSEG_ADDR:
3566 case MSR_K8_TSEG_MASK:
3567 case MSR_VM_HSAVE_PA:
3568 case MSR_K8_INT_PENDING_MSG:
3569 case MSR_AMD64_NB_CFG:
3570 case MSR_FAM10H_MMIO_CONF_BASE:
3571 case MSR_AMD64_BU_CFG2:
3572 case MSR_IA32_PERF_CTL:
3573 case MSR_AMD64_DC_CFG:
3574 case MSR_F15H_EX_CFG:
3576 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3577 * limit) MSRs. Just return 0, as we do not want to expose the host
3578 * data here. Do not conditionalize this on CPUID, as KVM does not do
3579 * so for existing CPU-specific MSRs.
3581 case MSR_RAPL_POWER_UNIT:
3582 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3583 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3584 case MSR_PKG_ENERGY_STATUS: /* Total package */
3585 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3588 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3589 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3590 return kvm_pmu_get_msr(vcpu, msr_info);
3591 if (!msr_info->host_initiated)
3595 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3596 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3597 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3598 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3599 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3600 return kvm_pmu_get_msr(vcpu, msr_info);
3603 case MSR_IA32_UCODE_REV:
3604 msr_info->data = vcpu->arch.microcode_version;
3606 case MSR_IA32_ARCH_CAPABILITIES:
3607 if (!msr_info->host_initiated &&
3608 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3610 msr_info->data = vcpu->arch.arch_capabilities;
3612 case MSR_IA32_PERF_CAPABILITIES:
3613 if (!msr_info->host_initiated &&
3614 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3616 msr_info->data = vcpu->arch.perf_capabilities;
3618 case MSR_IA32_POWER_CTL:
3619 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3621 case MSR_IA32_TSC: {
3623 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3624 * even when not intercepted. AMD manual doesn't explicitly
3625 * state this but appears to behave the same.
3627 * On userspace reads and writes, however, we unconditionally
3628 * return L1's TSC value to ensure backwards-compatible
3629 * behavior for migration.
3633 if (msr_info->host_initiated) {
3634 offset = vcpu->arch.l1_tsc_offset;
3635 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3637 offset = vcpu->arch.tsc_offset;
3638 ratio = vcpu->arch.tsc_scaling_ratio;
3641 msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3645 case 0x200 ... 0x2ff:
3646 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3647 case 0xcd: /* fsb frequency */
3651 * MSR_EBC_FREQUENCY_ID
3652 * Conservative value valid for even the basic CPU models.
3653 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3654 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3655 * and 266MHz for model 3, or 4. Set Core Clock
3656 * Frequency to System Bus Frequency Ratio to 1 (bits
3657 * 31:24) even though these are only valid for CPU
3658 * models > 2, however guests may end up dividing or
3659 * multiplying by zero otherwise.
3661 case MSR_EBC_FREQUENCY_ID:
3662 msr_info->data = 1 << 24;
3664 case MSR_IA32_APICBASE:
3665 msr_info->data = kvm_get_apic_base(vcpu);
3667 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3668 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3669 case MSR_IA32_TSC_DEADLINE:
3670 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3672 case MSR_IA32_TSC_ADJUST:
3673 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3675 case MSR_IA32_MISC_ENABLE:
3676 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3678 case MSR_IA32_SMBASE:
3679 if (!msr_info->host_initiated)
3681 msr_info->data = vcpu->arch.smbase;
3684 msr_info->data = vcpu->arch.smi_count;
3686 case MSR_IA32_PERF_STATUS:
3687 /* TSC increment by tick */
3688 msr_info->data = 1000ULL;
3689 /* CPU multiplier */
3690 msr_info->data |= (((uint64_t)4ULL) << 40);
3693 msr_info->data = vcpu->arch.efer;
3695 case MSR_KVM_WALL_CLOCK:
3696 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3699 msr_info->data = vcpu->kvm->arch.wall_clock;
3701 case MSR_KVM_WALL_CLOCK_NEW:
3702 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3705 msr_info->data = vcpu->kvm->arch.wall_clock;
3707 case MSR_KVM_SYSTEM_TIME:
3708 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3711 msr_info->data = vcpu->arch.time;
3713 case MSR_KVM_SYSTEM_TIME_NEW:
3714 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3717 msr_info->data = vcpu->arch.time;
3719 case MSR_KVM_ASYNC_PF_EN:
3720 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3723 msr_info->data = vcpu->arch.apf.msr_en_val;
3725 case MSR_KVM_ASYNC_PF_INT:
3726 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3729 msr_info->data = vcpu->arch.apf.msr_int_val;
3731 case MSR_KVM_ASYNC_PF_ACK:
3732 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3737 case MSR_KVM_STEAL_TIME:
3738 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3741 msr_info->data = vcpu->arch.st.msr_val;
3743 case MSR_KVM_PV_EOI_EN:
3744 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3747 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3749 case MSR_KVM_POLL_CONTROL:
3750 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3753 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3755 case MSR_IA32_P5_MC_ADDR:
3756 case MSR_IA32_P5_MC_TYPE:
3757 case MSR_IA32_MCG_CAP:
3758 case MSR_IA32_MCG_CTL:
3759 case MSR_IA32_MCG_STATUS:
3760 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3761 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3762 msr_info->host_initiated);
3764 if (!msr_info->host_initiated &&
3765 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3767 msr_info->data = vcpu->arch.ia32_xss;
3769 case MSR_K7_CLK_CTL:
3771 * Provide expected ramp-up count for K7. All other
3772 * are set to zero, indicating minimum divisors for
3775 * This prevents guest kernels on AMD host with CPU
3776 * type 6, model 8 and higher from exploding due to
3777 * the rdmsr failing.
3779 msr_info->data = 0x20000000;
3781 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3782 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3783 case HV_X64_MSR_SYNDBG_OPTIONS:
3784 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3785 case HV_X64_MSR_CRASH_CTL:
3786 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3787 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3788 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3789 case HV_X64_MSR_TSC_EMULATION_STATUS:
3790 return kvm_hv_get_msr_common(vcpu,
3791 msr_info->index, &msr_info->data,
3792 msr_info->host_initiated);
3793 case MSR_IA32_BBL_CR_CTL3:
3794 /* This legacy MSR exists but isn't fully documented in current
3795 * silicon. It is however accessed by winxp in very narrow
3796 * scenarios where it sets bit #19, itself documented as
3797 * a "reserved" bit. Best effort attempt to source coherent
3798 * read data here should the balance of the register be
3799 * interpreted by the guest:
3801 * L2 cache control register 3: 64GB range, 256KB size,
3802 * enabled, latency 0x1, configured
3804 msr_info->data = 0xbe702111;
3806 case MSR_AMD64_OSVW_ID_LENGTH:
3807 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3809 msr_info->data = vcpu->arch.osvw.length;
3811 case MSR_AMD64_OSVW_STATUS:
3812 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3814 msr_info->data = vcpu->arch.osvw.status;
3816 case MSR_PLATFORM_INFO:
3817 if (!msr_info->host_initiated &&
3818 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3820 msr_info->data = vcpu->arch.msr_platform_info;
3822 case MSR_MISC_FEATURES_ENABLES:
3823 msr_info->data = vcpu->arch.msr_misc_features_enables;
3826 msr_info->data = vcpu->arch.msr_hwcr;
3829 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3830 return kvm_pmu_get_msr(vcpu, msr_info);
3831 return KVM_MSR_RET_INVALID;
3835 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3838 * Read or write a bunch of msrs. All parameters are kernel addresses.
3840 * @return number of msrs set successfully.
3842 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3843 struct kvm_msr_entry *entries,
3844 int (*do_msr)(struct kvm_vcpu *vcpu,
3845 unsigned index, u64 *data))
3849 for (i = 0; i < msrs->nmsrs; ++i)
3850 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3857 * Read or write a bunch of msrs. Parameters are user addresses.
3859 * @return number of msrs set successfully.
3861 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3862 int (*do_msr)(struct kvm_vcpu *vcpu,
3863 unsigned index, u64 *data),
3866 struct kvm_msrs msrs;
3867 struct kvm_msr_entry *entries;
3872 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3876 if (msrs.nmsrs >= MAX_IO_MSRS)
3879 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3880 entries = memdup_user(user_msrs->entries, size);
3881 if (IS_ERR(entries)) {
3882 r = PTR_ERR(entries);
3886 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3891 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3902 static inline bool kvm_can_mwait_in_guest(void)
3904 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3905 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3906 boot_cpu_has(X86_FEATURE_ARAT);
3909 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3910 struct kvm_cpuid2 __user *cpuid_arg)
3912 struct kvm_cpuid2 cpuid;
3916 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3919 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3924 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3930 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3935 case KVM_CAP_IRQCHIP:
3937 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3938 case KVM_CAP_SET_TSS_ADDR:
3939 case KVM_CAP_EXT_CPUID:
3940 case KVM_CAP_EXT_EMUL_CPUID:
3941 case KVM_CAP_CLOCKSOURCE:
3943 case KVM_CAP_NOP_IO_DELAY:
3944 case KVM_CAP_MP_STATE:
3945 case KVM_CAP_SYNC_MMU:
3946 case KVM_CAP_USER_NMI:
3947 case KVM_CAP_REINJECT_CONTROL:
3948 case KVM_CAP_IRQ_INJECT_STATUS:
3949 case KVM_CAP_IOEVENTFD:
3950 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3952 case KVM_CAP_PIT_STATE2:
3953 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3954 case KVM_CAP_VCPU_EVENTS:
3955 case KVM_CAP_HYPERV:
3956 case KVM_CAP_HYPERV_VAPIC:
3957 case KVM_CAP_HYPERV_SPIN:
3958 case KVM_CAP_HYPERV_SYNIC:
3959 case KVM_CAP_HYPERV_SYNIC2:
3960 case KVM_CAP_HYPERV_VP_INDEX:
3961 case KVM_CAP_HYPERV_EVENTFD:
3962 case KVM_CAP_HYPERV_TLBFLUSH:
3963 case KVM_CAP_HYPERV_SEND_IPI:
3964 case KVM_CAP_HYPERV_CPUID:
3965 case KVM_CAP_HYPERV_ENFORCE_CPUID:
3966 case KVM_CAP_SYS_HYPERV_CPUID:
3967 case KVM_CAP_PCI_SEGMENT:
3968 case KVM_CAP_DEBUGREGS:
3969 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3971 case KVM_CAP_ASYNC_PF:
3972 case KVM_CAP_ASYNC_PF_INT:
3973 case KVM_CAP_GET_TSC_KHZ:
3974 case KVM_CAP_KVMCLOCK_CTRL:
3975 case KVM_CAP_READONLY_MEM:
3976 case KVM_CAP_HYPERV_TIME:
3977 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3978 case KVM_CAP_TSC_DEADLINE_TIMER:
3979 case KVM_CAP_DISABLE_QUIRKS:
3980 case KVM_CAP_SET_BOOT_CPU_ID:
3981 case KVM_CAP_SPLIT_IRQCHIP:
3982 case KVM_CAP_IMMEDIATE_EXIT:
3983 case KVM_CAP_PMU_EVENT_FILTER:
3984 case KVM_CAP_GET_MSR_FEATURES:
3985 case KVM_CAP_MSR_PLATFORM_INFO:
3986 case KVM_CAP_EXCEPTION_PAYLOAD:
3987 case KVM_CAP_SET_GUEST_DEBUG:
3988 case KVM_CAP_LAST_CPU:
3989 case KVM_CAP_X86_USER_SPACE_MSR:
3990 case KVM_CAP_X86_MSR_FILTER:
3991 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3992 #ifdef CONFIG_X86_SGX_KVM
3993 case KVM_CAP_SGX_ATTRIBUTE:
3995 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
3996 case KVM_CAP_SREGS2:
3999 case KVM_CAP_SET_GUEST_DEBUG2:
4000 return KVM_GUESTDBG_VALID_MASK;
4001 #ifdef CONFIG_KVM_XEN
4002 case KVM_CAP_XEN_HVM:
4003 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4004 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4005 KVM_XEN_HVM_CONFIG_SHARED_INFO;
4006 if (sched_info_on())
4007 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4010 case KVM_CAP_SYNC_REGS:
4011 r = KVM_SYNC_X86_VALID_FIELDS;
4013 case KVM_CAP_ADJUST_CLOCK:
4014 r = KVM_CLOCK_TSC_STABLE;
4016 case KVM_CAP_X86_DISABLE_EXITS:
4017 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4018 KVM_X86_DISABLE_EXITS_CSTATE;
4019 if(kvm_can_mwait_in_guest())
4020 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4022 case KVM_CAP_X86_SMM:
4023 /* SMBASE is usually relocated above 1M on modern chipsets,
4024 * and SMM handlers might indeed rely on 4G segment limits,
4025 * so do not report SMM to be available if real mode is
4026 * emulated via vm86 mode. Still, do not go to great lengths
4027 * to avoid userspace's usage of the feature, because it is a
4028 * fringe case that is not enabled except via specific settings
4029 * of the module parameters.
4031 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4034 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4036 case KVM_CAP_NR_VCPUS:
4037 r = KVM_SOFT_MAX_VCPUS;
4039 case KVM_CAP_MAX_VCPUS:
4042 case KVM_CAP_MAX_VCPU_ID:
4043 r = KVM_MAX_VCPU_ID;
4045 case KVM_CAP_PV_MMU: /* obsolete */
4049 r = KVM_MAX_MCE_BANKS;
4052 r = boot_cpu_has(X86_FEATURE_XSAVE);
4054 case KVM_CAP_TSC_CONTROL:
4055 r = kvm_has_tsc_control;
4057 case KVM_CAP_X2APIC_API:
4058 r = KVM_X2APIC_API_VALID_FLAGS;
4060 case KVM_CAP_NESTED_STATE:
4061 r = kvm_x86_ops.nested_ops->get_state ?
4062 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4064 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4065 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4067 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4068 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4070 case KVM_CAP_SMALLER_MAXPHYADDR:
4071 r = (int) allow_smaller_maxphyaddr;
4073 case KVM_CAP_STEAL_TIME:
4074 r = sched_info_on();
4076 case KVM_CAP_X86_BUS_LOCK_EXIT:
4077 if (kvm_has_bus_lock_exit)
4078 r = KVM_BUS_LOCK_DETECTION_OFF |
4079 KVM_BUS_LOCK_DETECTION_EXIT;
4090 long kvm_arch_dev_ioctl(struct file *filp,
4091 unsigned int ioctl, unsigned long arg)
4093 void __user *argp = (void __user *)arg;
4097 case KVM_GET_MSR_INDEX_LIST: {
4098 struct kvm_msr_list __user *user_msr_list = argp;
4099 struct kvm_msr_list msr_list;
4103 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4106 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4107 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4110 if (n < msr_list.nmsrs)
4113 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4114 num_msrs_to_save * sizeof(u32)))
4116 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4118 num_emulated_msrs * sizeof(u32)))
4123 case KVM_GET_SUPPORTED_CPUID:
4124 case KVM_GET_EMULATED_CPUID: {
4125 struct kvm_cpuid2 __user *cpuid_arg = argp;
4126 struct kvm_cpuid2 cpuid;
4129 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4132 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4138 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4143 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4145 if (copy_to_user(argp, &kvm_mce_cap_supported,
4146 sizeof(kvm_mce_cap_supported)))
4150 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4151 struct kvm_msr_list __user *user_msr_list = argp;
4152 struct kvm_msr_list msr_list;
4156 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4159 msr_list.nmsrs = num_msr_based_features;
4160 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4163 if (n < msr_list.nmsrs)
4166 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4167 num_msr_based_features * sizeof(u32)))
4173 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4175 case KVM_GET_SUPPORTED_HV_CPUID:
4176 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4186 static void wbinvd_ipi(void *garbage)
4191 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4193 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4196 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4198 /* Address WBINVD may be executed by guest */
4199 if (need_emulate_wbinvd(vcpu)) {
4200 if (static_call(kvm_x86_has_wbinvd_exit)())
4201 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4202 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4203 smp_call_function_single(vcpu->cpu,
4204 wbinvd_ipi, NULL, 1);
4207 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4209 /* Save host pkru register if supported */
4210 vcpu->arch.host_pkru = read_pkru();
4212 /* Apply any externally detected TSC adjustments (due to suspend) */
4213 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4214 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4215 vcpu->arch.tsc_offset_adjustment = 0;
4216 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4219 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4220 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4221 rdtsc() - vcpu->arch.last_host_tsc;
4223 mark_tsc_unstable("KVM discovered backwards TSC");
4225 if (kvm_check_tsc_unstable()) {
4226 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4227 vcpu->arch.last_guest_tsc);
4228 kvm_vcpu_write_tsc_offset(vcpu, offset);
4229 vcpu->arch.tsc_catchup = 1;
4232 if (kvm_lapic_hv_timer_in_use(vcpu))
4233 kvm_lapic_restart_hv_timer(vcpu);
4236 * On a host with synchronized TSC, there is no need to update
4237 * kvmclock on vcpu->cpu migration
4239 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4240 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4241 if (vcpu->cpu != cpu)
4242 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4246 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4249 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4251 struct kvm_host_map map;
4252 struct kvm_steal_time *st;
4254 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4257 if (vcpu->arch.st.preempted)
4260 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4261 &vcpu->arch.st.cache, true))
4265 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4267 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4269 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4272 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4276 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4277 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4280 * Take the srcu lock as memslots will be accessed to check the gfn
4281 * cache generation against the memslots generation.
4283 idx = srcu_read_lock(&vcpu->kvm->srcu);
4284 if (kvm_xen_msr_enabled(vcpu->kvm))
4285 kvm_xen_runstate_set_preempted(vcpu);
4287 kvm_steal_time_set_preempted(vcpu);
4288 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4290 static_call(kvm_x86_vcpu_put)(vcpu);
4291 vcpu->arch.last_host_tsc = rdtsc();
4293 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4294 * on every vmexit, but if not, we might have a stale dr6 from the
4295 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4300 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4301 struct kvm_lapic_state *s)
4303 if (vcpu->arch.apicv_active)
4304 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4306 return kvm_apic_get_state(vcpu, s);
4309 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4310 struct kvm_lapic_state *s)
4314 r = kvm_apic_set_state(vcpu, s);
4317 update_cr8_intercept(vcpu);
4322 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4325 * We can accept userspace's request for interrupt injection
4326 * as long as we have a place to store the interrupt number.
4327 * The actual injection will happen when the CPU is able to
4328 * deliver the interrupt.
4330 if (kvm_cpu_has_extint(vcpu))
4333 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4334 return (!lapic_in_kernel(vcpu) ||
4335 kvm_apic_accept_pic_intr(vcpu));
4338 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4340 return kvm_arch_interrupt_allowed(vcpu) &&
4341 kvm_cpu_accept_dm_intr(vcpu);
4344 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4345 struct kvm_interrupt *irq)
4347 if (irq->irq >= KVM_NR_INTERRUPTS)
4350 if (!irqchip_in_kernel(vcpu->kvm)) {
4351 kvm_queue_interrupt(vcpu, irq->irq, false);
4352 kvm_make_request(KVM_REQ_EVENT, vcpu);
4357 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4358 * fail for in-kernel 8259.
4360 if (pic_in_kernel(vcpu->kvm))
4363 if (vcpu->arch.pending_external_vector != -1)
4366 vcpu->arch.pending_external_vector = irq->irq;
4367 kvm_make_request(KVM_REQ_EVENT, vcpu);
4371 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4373 kvm_inject_nmi(vcpu);
4378 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4380 kvm_make_request(KVM_REQ_SMI, vcpu);
4385 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4386 struct kvm_tpr_access_ctl *tac)
4390 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4394 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4398 unsigned bank_num = mcg_cap & 0xff, bank;
4401 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4403 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4406 vcpu->arch.mcg_cap = mcg_cap;
4407 /* Init IA32_MCG_CTL to all 1s */
4408 if (mcg_cap & MCG_CTL_P)
4409 vcpu->arch.mcg_ctl = ~(u64)0;
4410 /* Init IA32_MCi_CTL to all 1s */
4411 for (bank = 0; bank < bank_num; bank++)
4412 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4414 static_call(kvm_x86_setup_mce)(vcpu);
4419 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4420 struct kvm_x86_mce *mce)
4422 u64 mcg_cap = vcpu->arch.mcg_cap;
4423 unsigned bank_num = mcg_cap & 0xff;
4424 u64 *banks = vcpu->arch.mce_banks;
4426 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4429 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4430 * reporting is disabled
4432 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4433 vcpu->arch.mcg_ctl != ~(u64)0)
4435 banks += 4 * mce->bank;
4437 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4438 * reporting is disabled for the bank
4440 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4442 if (mce->status & MCI_STATUS_UC) {
4443 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4444 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4445 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4448 if (banks[1] & MCI_STATUS_VAL)
4449 mce->status |= MCI_STATUS_OVER;
4450 banks[2] = mce->addr;
4451 banks[3] = mce->misc;
4452 vcpu->arch.mcg_status = mce->mcg_status;
4453 banks[1] = mce->status;
4454 kvm_queue_exception(vcpu, MC_VECTOR);
4455 } else if (!(banks[1] & MCI_STATUS_VAL)
4456 || !(banks[1] & MCI_STATUS_UC)) {
4457 if (banks[1] & MCI_STATUS_VAL)
4458 mce->status |= MCI_STATUS_OVER;
4459 banks[2] = mce->addr;
4460 banks[3] = mce->misc;
4461 banks[1] = mce->status;
4463 banks[1] |= MCI_STATUS_OVER;
4467 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4468 struct kvm_vcpu_events *events)
4472 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4476 * In guest mode, payload delivery should be deferred,
4477 * so that the L1 hypervisor can intercept #PF before
4478 * CR2 is modified (or intercept #DB before DR6 is
4479 * modified under nVMX). Unless the per-VM capability,
4480 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4481 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4482 * opportunistically defer the exception payload, deliver it if the
4483 * capability hasn't been requested before processing a
4484 * KVM_GET_VCPU_EVENTS.
4486 if (!vcpu->kvm->arch.exception_payload_enabled &&
4487 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4488 kvm_deliver_exception_payload(vcpu);
4491 * The API doesn't provide the instruction length for software
4492 * exceptions, so don't report them. As long as the guest RIP
4493 * isn't advanced, we should expect to encounter the exception
4496 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4497 events->exception.injected = 0;
4498 events->exception.pending = 0;
4500 events->exception.injected = vcpu->arch.exception.injected;
4501 events->exception.pending = vcpu->arch.exception.pending;
4503 * For ABI compatibility, deliberately conflate
4504 * pending and injected exceptions when
4505 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4507 if (!vcpu->kvm->arch.exception_payload_enabled)
4508 events->exception.injected |=
4509 vcpu->arch.exception.pending;
4511 events->exception.nr = vcpu->arch.exception.nr;
4512 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4513 events->exception.error_code = vcpu->arch.exception.error_code;
4514 events->exception_has_payload = vcpu->arch.exception.has_payload;
4515 events->exception_payload = vcpu->arch.exception.payload;
4517 events->interrupt.injected =
4518 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4519 events->interrupt.nr = vcpu->arch.interrupt.nr;
4520 events->interrupt.soft = 0;
4521 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4523 events->nmi.injected = vcpu->arch.nmi_injected;
4524 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4525 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4526 events->nmi.pad = 0;
4528 events->sipi_vector = 0; /* never valid when reporting to user space */
4530 events->smi.smm = is_smm(vcpu);
4531 events->smi.pending = vcpu->arch.smi_pending;
4532 events->smi.smm_inside_nmi =
4533 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4534 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4536 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4537 | KVM_VCPUEVENT_VALID_SHADOW
4538 | KVM_VCPUEVENT_VALID_SMM);
4539 if (vcpu->kvm->arch.exception_payload_enabled)
4540 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4542 memset(&events->reserved, 0, sizeof(events->reserved));
4545 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4547 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4548 struct kvm_vcpu_events *events)
4550 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4551 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4552 | KVM_VCPUEVENT_VALID_SHADOW
4553 | KVM_VCPUEVENT_VALID_SMM
4554 | KVM_VCPUEVENT_VALID_PAYLOAD))
4557 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4558 if (!vcpu->kvm->arch.exception_payload_enabled)
4560 if (events->exception.pending)
4561 events->exception.injected = 0;
4563 events->exception_has_payload = 0;
4565 events->exception.pending = 0;
4566 events->exception_has_payload = 0;
4569 if ((events->exception.injected || events->exception.pending) &&
4570 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4573 /* INITs are latched while in SMM */
4574 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4575 (events->smi.smm || events->smi.pending) &&
4576 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4580 vcpu->arch.exception.injected = events->exception.injected;
4581 vcpu->arch.exception.pending = events->exception.pending;
4582 vcpu->arch.exception.nr = events->exception.nr;
4583 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4584 vcpu->arch.exception.error_code = events->exception.error_code;
4585 vcpu->arch.exception.has_payload = events->exception_has_payload;
4586 vcpu->arch.exception.payload = events->exception_payload;
4588 vcpu->arch.interrupt.injected = events->interrupt.injected;
4589 vcpu->arch.interrupt.nr = events->interrupt.nr;
4590 vcpu->arch.interrupt.soft = events->interrupt.soft;
4591 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4592 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4593 events->interrupt.shadow);
4595 vcpu->arch.nmi_injected = events->nmi.injected;
4596 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4597 vcpu->arch.nmi_pending = events->nmi.pending;
4598 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4600 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4601 lapic_in_kernel(vcpu))
4602 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4604 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4605 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4606 kvm_smm_changed(vcpu, events->smi.smm);
4608 vcpu->arch.smi_pending = events->smi.pending;
4610 if (events->smi.smm) {
4611 if (events->smi.smm_inside_nmi)
4612 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4614 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4617 if (lapic_in_kernel(vcpu)) {
4618 if (events->smi.latched_init)
4619 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4621 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4625 kvm_make_request(KVM_REQ_EVENT, vcpu);
4630 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4631 struct kvm_debugregs *dbgregs)
4635 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4636 kvm_get_dr(vcpu, 6, &val);
4638 dbgregs->dr7 = vcpu->arch.dr7;
4640 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4643 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4644 struct kvm_debugregs *dbgregs)
4649 if (!kvm_dr6_valid(dbgregs->dr6))
4651 if (!kvm_dr7_valid(dbgregs->dr7))
4654 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4655 kvm_update_dr0123(vcpu);
4656 vcpu->arch.dr6 = dbgregs->dr6;
4657 vcpu->arch.dr7 = dbgregs->dr7;
4658 kvm_update_dr7(vcpu);
4663 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4665 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4667 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4668 u64 xstate_bv = xsave->header.xfeatures;
4672 * Copy legacy XSAVE area, to avoid complications with CPUID
4673 * leaves 0 and 1 in the loop below.
4675 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4678 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4679 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4682 * Copy each region from the possibly compacted offset to the
4683 * non-compacted offset.
4685 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4687 u64 xfeature_mask = valid & -valid;
4688 int xfeature_nr = fls64(xfeature_mask) - 1;
4689 void *src = get_xsave_addr(xsave, xfeature_nr);
4692 u32 size, offset, ecx, edx;
4693 cpuid_count(XSTATE_CPUID, xfeature_nr,
4694 &size, &offset, &ecx, &edx);
4695 if (xfeature_nr == XFEATURE_PKRU)
4696 memcpy(dest + offset, &vcpu->arch.pkru,
4697 sizeof(vcpu->arch.pkru));
4699 memcpy(dest + offset, src, size);
4703 valid -= xfeature_mask;
4707 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4709 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4710 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4714 * Copy legacy XSAVE area, to avoid complications with CPUID
4715 * leaves 0 and 1 in the loop below.
4717 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4719 /* Set XSTATE_BV and possibly XCOMP_BV. */
4720 xsave->header.xfeatures = xstate_bv;
4721 if (boot_cpu_has(X86_FEATURE_XSAVES))
4722 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4725 * Copy each region from the non-compacted offset to the
4726 * possibly compacted offset.
4728 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4730 u64 xfeature_mask = valid & -valid;
4731 int xfeature_nr = fls64(xfeature_mask) - 1;
4732 void *dest = get_xsave_addr(xsave, xfeature_nr);
4735 u32 size, offset, ecx, edx;
4736 cpuid_count(XSTATE_CPUID, xfeature_nr,
4737 &size, &offset, &ecx, &edx);
4738 if (xfeature_nr == XFEATURE_PKRU)
4739 memcpy(&vcpu->arch.pkru, src + offset,
4740 sizeof(vcpu->arch.pkru));
4742 memcpy(dest, src + offset, size);
4745 valid -= xfeature_mask;
4749 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4750 struct kvm_xsave *guest_xsave)
4752 if (!vcpu->arch.guest_fpu)
4755 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4756 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4757 fill_xsave((u8 *) guest_xsave->region, vcpu);
4759 memcpy(guest_xsave->region,
4760 &vcpu->arch.guest_fpu->state.fxsave,
4761 sizeof(struct fxregs_state));
4762 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4763 XFEATURE_MASK_FPSSE;
4767 #define XSAVE_MXCSR_OFFSET 24
4769 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4770 struct kvm_xsave *guest_xsave)
4775 if (!vcpu->arch.guest_fpu)
4778 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4779 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4781 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4783 * Here we allow setting states that are not present in
4784 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4785 * with old userspace.
4787 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4789 load_xsave(vcpu, (u8 *)guest_xsave->region);
4791 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4792 mxcsr & ~mxcsr_feature_mask)
4794 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4795 guest_xsave->region, sizeof(struct fxregs_state));
4800 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4801 struct kvm_xcrs *guest_xcrs)
4803 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4804 guest_xcrs->nr_xcrs = 0;
4808 guest_xcrs->nr_xcrs = 1;
4809 guest_xcrs->flags = 0;
4810 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4811 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4814 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4815 struct kvm_xcrs *guest_xcrs)
4819 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4822 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4825 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4826 /* Only support XCR0 currently */
4827 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4828 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4829 guest_xcrs->xcrs[i].value);
4838 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4839 * stopped by the hypervisor. This function will be called from the host only.
4840 * EINVAL is returned when the host attempts to set the flag for a guest that
4841 * does not support pv clocks.
4843 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4845 if (!vcpu->arch.pv_time_enabled)
4847 vcpu->arch.pvclock_set_guest_stopped_request = true;
4848 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4852 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4853 struct kvm_enable_cap *cap)
4856 uint16_t vmcs_version;
4857 void __user *user_ptr;
4863 case KVM_CAP_HYPERV_SYNIC2:
4868 case KVM_CAP_HYPERV_SYNIC:
4869 if (!irqchip_in_kernel(vcpu->kvm))
4871 return kvm_hv_activate_synic(vcpu, cap->cap ==
4872 KVM_CAP_HYPERV_SYNIC2);
4873 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4874 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4876 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4878 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4879 if (copy_to_user(user_ptr, &vmcs_version,
4880 sizeof(vmcs_version)))
4884 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4885 if (!kvm_x86_ops.enable_direct_tlbflush)
4888 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4890 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4891 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
4893 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4894 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4895 if (vcpu->arch.pv_cpuid.enforce)
4896 kvm_update_pv_runtime(vcpu);
4904 long kvm_arch_vcpu_ioctl(struct file *filp,
4905 unsigned int ioctl, unsigned long arg)
4907 struct kvm_vcpu *vcpu = filp->private_data;
4908 void __user *argp = (void __user *)arg;
4911 struct kvm_sregs2 *sregs2;
4912 struct kvm_lapic_state *lapic;
4913 struct kvm_xsave *xsave;
4914 struct kvm_xcrs *xcrs;
4922 case KVM_GET_LAPIC: {
4924 if (!lapic_in_kernel(vcpu))
4926 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4927 GFP_KERNEL_ACCOUNT);
4932 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4936 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4941 case KVM_SET_LAPIC: {
4943 if (!lapic_in_kernel(vcpu))
4945 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4946 if (IS_ERR(u.lapic)) {
4947 r = PTR_ERR(u.lapic);
4951 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4954 case KVM_INTERRUPT: {
4955 struct kvm_interrupt irq;
4958 if (copy_from_user(&irq, argp, sizeof(irq)))
4960 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4964 r = kvm_vcpu_ioctl_nmi(vcpu);
4968 r = kvm_vcpu_ioctl_smi(vcpu);
4971 case KVM_SET_CPUID: {
4972 struct kvm_cpuid __user *cpuid_arg = argp;
4973 struct kvm_cpuid cpuid;
4976 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4978 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4981 case KVM_SET_CPUID2: {
4982 struct kvm_cpuid2 __user *cpuid_arg = argp;
4983 struct kvm_cpuid2 cpuid;
4986 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4988 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4989 cpuid_arg->entries);
4992 case KVM_GET_CPUID2: {
4993 struct kvm_cpuid2 __user *cpuid_arg = argp;
4994 struct kvm_cpuid2 cpuid;
4997 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4999 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5000 cpuid_arg->entries);
5004 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5009 case KVM_GET_MSRS: {
5010 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5011 r = msr_io(vcpu, argp, do_get_msr, 1);
5012 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5015 case KVM_SET_MSRS: {
5016 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5017 r = msr_io(vcpu, argp, do_set_msr, 0);
5018 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5021 case KVM_TPR_ACCESS_REPORTING: {
5022 struct kvm_tpr_access_ctl tac;
5025 if (copy_from_user(&tac, argp, sizeof(tac)))
5027 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5031 if (copy_to_user(argp, &tac, sizeof(tac)))
5036 case KVM_SET_VAPIC_ADDR: {
5037 struct kvm_vapic_addr va;
5041 if (!lapic_in_kernel(vcpu))
5044 if (copy_from_user(&va, argp, sizeof(va)))
5046 idx = srcu_read_lock(&vcpu->kvm->srcu);
5047 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5048 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5051 case KVM_X86_SETUP_MCE: {
5055 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5057 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5060 case KVM_X86_SET_MCE: {
5061 struct kvm_x86_mce mce;
5064 if (copy_from_user(&mce, argp, sizeof(mce)))
5066 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5069 case KVM_GET_VCPU_EVENTS: {
5070 struct kvm_vcpu_events events;
5072 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5075 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5080 case KVM_SET_VCPU_EVENTS: {
5081 struct kvm_vcpu_events events;
5084 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5087 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5090 case KVM_GET_DEBUGREGS: {
5091 struct kvm_debugregs dbgregs;
5093 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5096 if (copy_to_user(argp, &dbgregs,
5097 sizeof(struct kvm_debugregs)))
5102 case KVM_SET_DEBUGREGS: {
5103 struct kvm_debugregs dbgregs;
5106 if (copy_from_user(&dbgregs, argp,
5107 sizeof(struct kvm_debugregs)))
5110 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5113 case KVM_GET_XSAVE: {
5114 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5119 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5122 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5127 case KVM_SET_XSAVE: {
5128 u.xsave = memdup_user(argp, sizeof(*u.xsave));
5129 if (IS_ERR(u.xsave)) {
5130 r = PTR_ERR(u.xsave);
5134 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5137 case KVM_GET_XCRS: {
5138 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5143 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5146 if (copy_to_user(argp, u.xcrs,
5147 sizeof(struct kvm_xcrs)))
5152 case KVM_SET_XCRS: {
5153 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5154 if (IS_ERR(u.xcrs)) {
5155 r = PTR_ERR(u.xcrs);
5159 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5162 case KVM_SET_TSC_KHZ: {
5166 user_tsc_khz = (u32)arg;
5168 if (kvm_has_tsc_control &&
5169 user_tsc_khz >= kvm_max_guest_tsc_khz)
5172 if (user_tsc_khz == 0)
5173 user_tsc_khz = tsc_khz;
5175 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5180 case KVM_GET_TSC_KHZ: {
5181 r = vcpu->arch.virtual_tsc_khz;
5184 case KVM_KVMCLOCK_CTRL: {
5185 r = kvm_set_guest_paused(vcpu);
5188 case KVM_ENABLE_CAP: {
5189 struct kvm_enable_cap cap;
5192 if (copy_from_user(&cap, argp, sizeof(cap)))
5194 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5197 case KVM_GET_NESTED_STATE: {
5198 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5202 if (!kvm_x86_ops.nested_ops->get_state)
5205 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5207 if (get_user(user_data_size, &user_kvm_nested_state->size))
5210 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5215 if (r > user_data_size) {
5216 if (put_user(r, &user_kvm_nested_state->size))
5226 case KVM_SET_NESTED_STATE: {
5227 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5228 struct kvm_nested_state kvm_state;
5232 if (!kvm_x86_ops.nested_ops->set_state)
5236 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5240 if (kvm_state.size < sizeof(kvm_state))
5243 if (kvm_state.flags &
5244 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5245 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5246 | KVM_STATE_NESTED_GIF_SET))
5249 /* nested_run_pending implies guest_mode. */
5250 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5251 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5254 idx = srcu_read_lock(&vcpu->kvm->srcu);
5255 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5256 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5259 case KVM_GET_SUPPORTED_HV_CPUID:
5260 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5262 #ifdef CONFIG_KVM_XEN
5263 case KVM_XEN_VCPU_GET_ATTR: {
5264 struct kvm_xen_vcpu_attr xva;
5267 if (copy_from_user(&xva, argp, sizeof(xva)))
5269 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5270 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5274 case KVM_XEN_VCPU_SET_ATTR: {
5275 struct kvm_xen_vcpu_attr xva;
5278 if (copy_from_user(&xva, argp, sizeof(xva)))
5280 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5284 case KVM_GET_SREGS2: {
5285 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5289 __get_sregs2(vcpu, u.sregs2);
5291 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5296 case KVM_SET_SREGS2: {
5297 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5298 if (IS_ERR(u.sregs2)) {
5299 r = PTR_ERR(u.sregs2);
5303 r = __set_sregs2(vcpu, u.sregs2);
5316 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5318 return VM_FAULT_SIGBUS;
5321 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5325 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5327 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5331 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5334 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5337 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5338 unsigned long kvm_nr_mmu_pages)
5340 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5343 mutex_lock(&kvm->slots_lock);
5345 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5346 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5348 mutex_unlock(&kvm->slots_lock);
5352 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5354 return kvm->arch.n_max_mmu_pages;
5357 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5359 struct kvm_pic *pic = kvm->arch.vpic;
5363 switch (chip->chip_id) {
5364 case KVM_IRQCHIP_PIC_MASTER:
5365 memcpy(&chip->chip.pic, &pic->pics[0],
5366 sizeof(struct kvm_pic_state));
5368 case KVM_IRQCHIP_PIC_SLAVE:
5369 memcpy(&chip->chip.pic, &pic->pics[1],
5370 sizeof(struct kvm_pic_state));
5372 case KVM_IRQCHIP_IOAPIC:
5373 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5382 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5384 struct kvm_pic *pic = kvm->arch.vpic;
5388 switch (chip->chip_id) {
5389 case KVM_IRQCHIP_PIC_MASTER:
5390 spin_lock(&pic->lock);
5391 memcpy(&pic->pics[0], &chip->chip.pic,
5392 sizeof(struct kvm_pic_state));
5393 spin_unlock(&pic->lock);
5395 case KVM_IRQCHIP_PIC_SLAVE:
5396 spin_lock(&pic->lock);
5397 memcpy(&pic->pics[1], &chip->chip.pic,
5398 sizeof(struct kvm_pic_state));
5399 spin_unlock(&pic->lock);
5401 case KVM_IRQCHIP_IOAPIC:
5402 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5408 kvm_pic_update_irq(pic);
5412 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5414 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5416 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5418 mutex_lock(&kps->lock);
5419 memcpy(ps, &kps->channels, sizeof(*ps));
5420 mutex_unlock(&kps->lock);
5424 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5427 struct kvm_pit *pit = kvm->arch.vpit;
5429 mutex_lock(&pit->pit_state.lock);
5430 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5431 for (i = 0; i < 3; i++)
5432 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5433 mutex_unlock(&pit->pit_state.lock);
5437 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5439 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5440 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5441 sizeof(ps->channels));
5442 ps->flags = kvm->arch.vpit->pit_state.flags;
5443 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5444 memset(&ps->reserved, 0, sizeof(ps->reserved));
5448 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5452 u32 prev_legacy, cur_legacy;
5453 struct kvm_pit *pit = kvm->arch.vpit;
5455 mutex_lock(&pit->pit_state.lock);
5456 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5457 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5458 if (!prev_legacy && cur_legacy)
5460 memcpy(&pit->pit_state.channels, &ps->channels,
5461 sizeof(pit->pit_state.channels));
5462 pit->pit_state.flags = ps->flags;
5463 for (i = 0; i < 3; i++)
5464 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5466 mutex_unlock(&pit->pit_state.lock);
5470 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5471 struct kvm_reinject_control *control)
5473 struct kvm_pit *pit = kvm->arch.vpit;
5475 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5476 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5477 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5479 mutex_lock(&pit->pit_state.lock);
5480 kvm_pit_set_reinject(pit, control->pit_reinject);
5481 mutex_unlock(&pit->pit_state.lock);
5486 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5490 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5491 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5492 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5495 struct kvm_vcpu *vcpu;
5498 kvm_for_each_vcpu(i, vcpu, kvm)
5499 kvm_vcpu_kick(vcpu);
5502 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5505 if (!irqchip_in_kernel(kvm))
5508 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5509 irq_event->irq, irq_event->level,
5514 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5515 struct kvm_enable_cap *cap)
5523 case KVM_CAP_DISABLE_QUIRKS:
5524 kvm->arch.disabled_quirks = cap->args[0];
5527 case KVM_CAP_SPLIT_IRQCHIP: {
5528 mutex_lock(&kvm->lock);
5530 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5531 goto split_irqchip_unlock;
5533 if (irqchip_in_kernel(kvm))
5534 goto split_irqchip_unlock;
5535 if (kvm->created_vcpus)
5536 goto split_irqchip_unlock;
5537 r = kvm_setup_empty_irq_routing(kvm);
5539 goto split_irqchip_unlock;
5540 /* Pairs with irqchip_in_kernel. */
5542 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5543 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5545 split_irqchip_unlock:
5546 mutex_unlock(&kvm->lock);
5549 case KVM_CAP_X2APIC_API:
5551 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5554 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5555 kvm->arch.x2apic_format = true;
5556 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5557 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5561 case KVM_CAP_X86_DISABLE_EXITS:
5563 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5566 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5567 kvm_can_mwait_in_guest())
5568 kvm->arch.mwait_in_guest = true;
5569 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5570 kvm->arch.hlt_in_guest = true;
5571 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5572 kvm->arch.pause_in_guest = true;
5573 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5574 kvm->arch.cstate_in_guest = true;
5577 case KVM_CAP_MSR_PLATFORM_INFO:
5578 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5581 case KVM_CAP_EXCEPTION_PAYLOAD:
5582 kvm->arch.exception_payload_enabled = cap->args[0];
5585 case KVM_CAP_X86_USER_SPACE_MSR:
5586 kvm->arch.user_space_msr_mask = cap->args[0];
5589 case KVM_CAP_X86_BUS_LOCK_EXIT:
5591 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5594 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5595 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5598 if (kvm_has_bus_lock_exit &&
5599 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5600 kvm->arch.bus_lock_detection_enabled = true;
5603 #ifdef CONFIG_X86_SGX_KVM
5604 case KVM_CAP_SGX_ATTRIBUTE: {
5605 unsigned long allowed_attributes = 0;
5607 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5611 /* KVM only supports the PROVISIONKEY privileged attribute. */
5612 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5613 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5614 kvm->arch.sgx_provisioning_allowed = true;
5620 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5622 if (kvm_x86_ops.vm_copy_enc_context_from)
5623 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5632 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5634 struct kvm_x86_msr_filter *msr_filter;
5636 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5640 msr_filter->default_allow = default_allow;
5644 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5651 for (i = 0; i < msr_filter->count; i++)
5652 kfree(msr_filter->ranges[i].bitmap);
5657 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5658 struct kvm_msr_filter_range *user_range)
5660 unsigned long *bitmap = NULL;
5663 if (!user_range->nmsrs)
5666 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5669 if (!user_range->flags)
5672 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5673 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5676 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5678 return PTR_ERR(bitmap);
5680 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5681 .flags = user_range->flags,
5682 .base = user_range->base,
5683 .nmsrs = user_range->nmsrs,
5687 msr_filter->count++;
5691 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5693 struct kvm_msr_filter __user *user_msr_filter = argp;
5694 struct kvm_x86_msr_filter *new_filter, *old_filter;
5695 struct kvm_msr_filter filter;
5701 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5704 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5705 empty &= !filter.ranges[i].nmsrs;
5707 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5708 if (empty && !default_allow)
5711 new_filter = kvm_alloc_msr_filter(default_allow);
5715 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5716 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5718 kvm_free_msr_filter(new_filter);
5723 mutex_lock(&kvm->lock);
5725 /* The per-VM filter is protected by kvm->lock... */
5726 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5728 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5729 synchronize_srcu(&kvm->srcu);
5731 kvm_free_msr_filter(old_filter);
5733 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5734 mutex_unlock(&kvm->lock);
5739 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5740 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5742 struct kvm_vcpu *vcpu;
5745 mutex_lock(&kvm->lock);
5746 kvm_for_each_vcpu(i, vcpu, kvm) {
5747 if (!vcpu->arch.pv_time_enabled)
5750 ret = kvm_set_guest_paused(vcpu);
5752 kvm_err("Failed to pause guest VCPU%d: %d\n",
5753 vcpu->vcpu_id, ret);
5757 mutex_unlock(&kvm->lock);
5759 return ret ? NOTIFY_BAD : NOTIFY_DONE;
5762 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5765 case PM_HIBERNATION_PREPARE:
5766 case PM_SUSPEND_PREPARE:
5767 return kvm_arch_suspend_notifier(kvm);
5772 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5774 long kvm_arch_vm_ioctl(struct file *filp,
5775 unsigned int ioctl, unsigned long arg)
5777 struct kvm *kvm = filp->private_data;
5778 void __user *argp = (void __user *)arg;
5781 * This union makes it completely explicit to gcc-3.x
5782 * that these two variables' stack usage should be
5783 * combined, not added together.
5786 struct kvm_pit_state ps;
5787 struct kvm_pit_state2 ps2;
5788 struct kvm_pit_config pit_config;
5792 case KVM_SET_TSS_ADDR:
5793 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5795 case KVM_SET_IDENTITY_MAP_ADDR: {
5798 mutex_lock(&kvm->lock);
5800 if (kvm->created_vcpus)
5801 goto set_identity_unlock;
5803 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5804 goto set_identity_unlock;
5805 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5806 set_identity_unlock:
5807 mutex_unlock(&kvm->lock);
5810 case KVM_SET_NR_MMU_PAGES:
5811 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5813 case KVM_GET_NR_MMU_PAGES:
5814 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5816 case KVM_CREATE_IRQCHIP: {
5817 mutex_lock(&kvm->lock);
5820 if (irqchip_in_kernel(kvm))
5821 goto create_irqchip_unlock;
5824 if (kvm->created_vcpus)
5825 goto create_irqchip_unlock;
5827 r = kvm_pic_init(kvm);
5829 goto create_irqchip_unlock;
5831 r = kvm_ioapic_init(kvm);
5833 kvm_pic_destroy(kvm);
5834 goto create_irqchip_unlock;
5837 r = kvm_setup_default_irq_routing(kvm);
5839 kvm_ioapic_destroy(kvm);
5840 kvm_pic_destroy(kvm);
5841 goto create_irqchip_unlock;
5843 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5845 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5846 create_irqchip_unlock:
5847 mutex_unlock(&kvm->lock);
5850 case KVM_CREATE_PIT:
5851 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5853 case KVM_CREATE_PIT2:
5855 if (copy_from_user(&u.pit_config, argp,
5856 sizeof(struct kvm_pit_config)))
5859 mutex_lock(&kvm->lock);
5862 goto create_pit_unlock;
5864 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5868 mutex_unlock(&kvm->lock);
5870 case KVM_GET_IRQCHIP: {
5871 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5872 struct kvm_irqchip *chip;
5874 chip = memdup_user(argp, sizeof(*chip));
5881 if (!irqchip_kernel(kvm))
5882 goto get_irqchip_out;
5883 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5885 goto get_irqchip_out;
5887 if (copy_to_user(argp, chip, sizeof(*chip)))
5888 goto get_irqchip_out;
5894 case KVM_SET_IRQCHIP: {
5895 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5896 struct kvm_irqchip *chip;
5898 chip = memdup_user(argp, sizeof(*chip));
5905 if (!irqchip_kernel(kvm))
5906 goto set_irqchip_out;
5907 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5914 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5917 if (!kvm->arch.vpit)
5919 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5923 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5930 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5932 mutex_lock(&kvm->lock);
5934 if (!kvm->arch.vpit)
5936 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5938 mutex_unlock(&kvm->lock);
5941 case KVM_GET_PIT2: {
5943 if (!kvm->arch.vpit)
5945 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5949 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5954 case KVM_SET_PIT2: {
5956 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5958 mutex_lock(&kvm->lock);
5960 if (!kvm->arch.vpit)
5962 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5964 mutex_unlock(&kvm->lock);
5967 case KVM_REINJECT_CONTROL: {
5968 struct kvm_reinject_control control;
5970 if (copy_from_user(&control, argp, sizeof(control)))
5973 if (!kvm->arch.vpit)
5975 r = kvm_vm_ioctl_reinject(kvm, &control);
5978 case KVM_SET_BOOT_CPU_ID:
5980 mutex_lock(&kvm->lock);
5981 if (kvm->created_vcpus)
5984 kvm->arch.bsp_vcpu_id = arg;
5985 mutex_unlock(&kvm->lock);
5987 #ifdef CONFIG_KVM_XEN
5988 case KVM_XEN_HVM_CONFIG: {
5989 struct kvm_xen_hvm_config xhc;
5991 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5993 r = kvm_xen_hvm_config(kvm, &xhc);
5996 case KVM_XEN_HVM_GET_ATTR: {
5997 struct kvm_xen_hvm_attr xha;
6000 if (copy_from_user(&xha, argp, sizeof(xha)))
6002 r = kvm_xen_hvm_get_attr(kvm, &xha);
6003 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6007 case KVM_XEN_HVM_SET_ATTR: {
6008 struct kvm_xen_hvm_attr xha;
6011 if (copy_from_user(&xha, argp, sizeof(xha)))
6013 r = kvm_xen_hvm_set_attr(kvm, &xha);
6017 case KVM_SET_CLOCK: {
6018 struct kvm_arch *ka = &kvm->arch;
6019 struct kvm_clock_data user_ns;
6023 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
6032 * TODO: userspace has to take care of races with VCPU_RUN, so
6033 * kvm_gen_update_masterclock() can be cut down to locked
6034 * pvclock_update_vm_gtod_copy().
6036 kvm_gen_update_masterclock(kvm);
6039 * This pairs with kvm_guest_time_update(): when masterclock is
6040 * in use, we use master_kernel_ns + kvmclock_offset to set
6041 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6042 * is slightly ahead) here we risk going negative on unsigned
6043 * 'system_time' when 'user_ns.clock' is very small.
6045 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
6046 if (kvm->arch.use_master_clock)
6047 now_ns = ka->master_kernel_ns;
6049 now_ns = get_kvmclock_base_ns();
6050 ka->kvmclock_offset = user_ns.clock - now_ns;
6051 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
6053 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
6056 case KVM_GET_CLOCK: {
6057 struct kvm_clock_data user_ns;
6060 now_ns = get_kvmclock_ns(kvm);
6061 user_ns.clock = now_ns;
6062 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
6063 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
6066 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
6071 case KVM_MEMORY_ENCRYPT_OP: {
6073 if (kvm_x86_ops.mem_enc_op)
6074 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6077 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6078 struct kvm_enc_region region;
6081 if (copy_from_user(®ion, argp, sizeof(region)))
6085 if (kvm_x86_ops.mem_enc_reg_region)
6086 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
6089 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6090 struct kvm_enc_region region;
6093 if (copy_from_user(®ion, argp, sizeof(region)))
6097 if (kvm_x86_ops.mem_enc_unreg_region)
6098 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
6101 case KVM_HYPERV_EVENTFD: {
6102 struct kvm_hyperv_eventfd hvevfd;
6105 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6107 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6110 case KVM_SET_PMU_EVENT_FILTER:
6111 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6113 case KVM_X86_SET_MSR_FILTER:
6114 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6123 static void kvm_init_msr_list(void)
6125 struct x86_pmu_capability x86_pmu;
6129 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6130 "Please update the fixed PMCs in msrs_to_saved_all[]");
6132 perf_get_x86_pmu_capability(&x86_pmu);
6134 num_msrs_to_save = 0;
6135 num_emulated_msrs = 0;
6136 num_msr_based_features = 0;
6138 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6139 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6143 * Even MSRs that are valid in the host may not be exposed
6144 * to the guests in some cases.
6146 switch (msrs_to_save_all[i]) {
6147 case MSR_IA32_BNDCFGS:
6148 if (!kvm_mpx_supported())
6152 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6153 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6156 case MSR_IA32_UMWAIT_CONTROL:
6157 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6160 case MSR_IA32_RTIT_CTL:
6161 case MSR_IA32_RTIT_STATUS:
6162 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6165 case MSR_IA32_RTIT_CR3_MATCH:
6166 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6167 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6170 case MSR_IA32_RTIT_OUTPUT_BASE:
6171 case MSR_IA32_RTIT_OUTPUT_MASK:
6172 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6173 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6174 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6177 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6178 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6179 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6180 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6183 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6184 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6185 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6188 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6189 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6190 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6197 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6200 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6201 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6204 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6207 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6208 struct kvm_msr_entry msr;
6210 msr.index = msr_based_features_all[i];
6211 if (kvm_get_msr_feature(&msr))
6214 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6218 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6226 if (!(lapic_in_kernel(vcpu) &&
6227 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6228 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6239 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6246 if (!(lapic_in_kernel(vcpu) &&
6247 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6249 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6251 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6261 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6262 struct kvm_segment *var, int seg)
6264 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6267 void kvm_get_segment(struct kvm_vcpu *vcpu,
6268 struct kvm_segment *var, int seg)
6270 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6273 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6274 struct x86_exception *exception)
6278 BUG_ON(!mmu_is_nested(vcpu));
6280 /* NPT walks are always user-walks */
6281 access |= PFERR_USER_MASK;
6282 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6287 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6288 struct x86_exception *exception)
6290 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6291 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6293 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6295 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6296 struct x86_exception *exception)
6298 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6299 access |= PFERR_FETCH_MASK;
6300 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6303 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6304 struct x86_exception *exception)
6306 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6307 access |= PFERR_WRITE_MASK;
6308 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6310 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6312 /* uses this to access any guest's mapped memory without checking CPL */
6313 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6314 struct x86_exception *exception)
6316 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6319 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6320 struct kvm_vcpu *vcpu, u32 access,
6321 struct x86_exception *exception)
6324 int r = X86EMUL_CONTINUE;
6327 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6329 unsigned offset = addr & (PAGE_SIZE-1);
6330 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6333 if (gpa == UNMAPPED_GVA)
6334 return X86EMUL_PROPAGATE_FAULT;
6335 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6338 r = X86EMUL_IO_NEEDED;
6350 /* used for instruction fetching */
6351 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6352 gva_t addr, void *val, unsigned int bytes,
6353 struct x86_exception *exception)
6355 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6356 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6360 /* Inline kvm_read_guest_virt_helper for speed. */
6361 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6363 if (unlikely(gpa == UNMAPPED_GVA))
6364 return X86EMUL_PROPAGATE_FAULT;
6366 offset = addr & (PAGE_SIZE-1);
6367 if (WARN_ON(offset + bytes > PAGE_SIZE))
6368 bytes = (unsigned)PAGE_SIZE - offset;
6369 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6371 if (unlikely(ret < 0))
6372 return X86EMUL_IO_NEEDED;
6374 return X86EMUL_CONTINUE;
6377 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6378 gva_t addr, void *val, unsigned int bytes,
6379 struct x86_exception *exception)
6381 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6384 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6385 * is returned, but our callers are not ready for that and they blindly
6386 * call kvm_inject_page_fault. Ensure that they at least do not leak
6387 * uninitialized kernel stack memory into cr2 and error code.
6389 memset(exception, 0, sizeof(*exception));
6390 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6393 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6395 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6396 gva_t addr, void *val, unsigned int bytes,
6397 struct x86_exception *exception, bool system)
6399 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6402 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6403 access |= PFERR_USER_MASK;
6405 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6408 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6409 unsigned long addr, void *val, unsigned int bytes)
6411 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6412 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6414 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6417 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6418 struct kvm_vcpu *vcpu, u32 access,
6419 struct x86_exception *exception)
6422 int r = X86EMUL_CONTINUE;
6425 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6428 unsigned offset = addr & (PAGE_SIZE-1);
6429 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6432 if (gpa == UNMAPPED_GVA)
6433 return X86EMUL_PROPAGATE_FAULT;
6434 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6436 r = X86EMUL_IO_NEEDED;
6448 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6449 unsigned int bytes, struct x86_exception *exception,
6452 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6453 u32 access = PFERR_WRITE_MASK;
6455 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6456 access |= PFERR_USER_MASK;
6458 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6462 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6463 unsigned int bytes, struct x86_exception *exception)
6465 /* kvm_write_guest_virt_system can pull in tons of pages. */
6466 vcpu->arch.l1tf_flush_l1d = true;
6468 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6469 PFERR_WRITE_MASK, exception);
6471 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6473 int handle_ud(struct kvm_vcpu *vcpu)
6475 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6476 int emul_type = EMULTYPE_TRAP_UD;
6477 char sig[5]; /* ud2; .ascii "kvm" */
6478 struct x86_exception e;
6480 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6483 if (force_emulation_prefix &&
6484 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6485 sig, sizeof(sig), &e) == 0 &&
6486 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6487 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6488 emul_type = EMULTYPE_TRAP_UD_FORCED;
6491 return kvm_emulate_instruction(vcpu, emul_type);
6493 EXPORT_SYMBOL_GPL(handle_ud);
6495 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6496 gpa_t gpa, bool write)
6498 /* For APIC access vmexit */
6499 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6502 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6503 trace_vcpu_match_mmio(gva, gpa, write, true);
6510 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6511 gpa_t *gpa, struct x86_exception *exception,
6514 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6515 | (write ? PFERR_WRITE_MASK : 0);
6518 * currently PKRU is only applied to ept enabled guest so
6519 * there is no pkey in EPT page table for L1 guest or EPT
6520 * shadow page table for L2 guest.
6522 if (vcpu_match_mmio_gva(vcpu, gva)
6523 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6524 vcpu->arch.mmio_access, 0, access)) {
6525 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6526 (gva & (PAGE_SIZE - 1));
6527 trace_vcpu_match_mmio(gva, *gpa, write, false);
6531 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6533 if (*gpa == UNMAPPED_GVA)
6536 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6539 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6540 const void *val, int bytes)
6544 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6547 kvm_page_track_write(vcpu, gpa, val, bytes);
6551 struct read_write_emulator_ops {
6552 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6554 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6555 void *val, int bytes);
6556 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6557 int bytes, void *val);
6558 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6559 void *val, int bytes);
6563 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6565 if (vcpu->mmio_read_completed) {
6566 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6567 vcpu->mmio_fragments[0].gpa, val);
6568 vcpu->mmio_read_completed = 0;
6575 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6576 void *val, int bytes)
6578 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6581 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6582 void *val, int bytes)
6584 return emulator_write_phys(vcpu, gpa, val, bytes);
6587 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6589 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6590 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6593 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6594 void *val, int bytes)
6596 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6597 return X86EMUL_IO_NEEDED;
6600 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6601 void *val, int bytes)
6603 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6605 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6606 return X86EMUL_CONTINUE;
6609 static const struct read_write_emulator_ops read_emultor = {
6610 .read_write_prepare = read_prepare,
6611 .read_write_emulate = read_emulate,
6612 .read_write_mmio = vcpu_mmio_read,
6613 .read_write_exit_mmio = read_exit_mmio,
6616 static const struct read_write_emulator_ops write_emultor = {
6617 .read_write_emulate = write_emulate,
6618 .read_write_mmio = write_mmio,
6619 .read_write_exit_mmio = write_exit_mmio,
6623 static int emulator_read_write_onepage(unsigned long addr, void *val,
6625 struct x86_exception *exception,
6626 struct kvm_vcpu *vcpu,
6627 const struct read_write_emulator_ops *ops)
6631 bool write = ops->write;
6632 struct kvm_mmio_fragment *frag;
6633 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6636 * If the exit was due to a NPF we may already have a GPA.
6637 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6638 * Note, this cannot be used on string operations since string
6639 * operation using rep will only have the initial GPA from the NPF
6642 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6643 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6644 gpa = ctxt->gpa_val;
6645 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6647 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6649 return X86EMUL_PROPAGATE_FAULT;
6652 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6653 return X86EMUL_CONTINUE;
6656 * Is this MMIO handled locally?
6658 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6659 if (handled == bytes)
6660 return X86EMUL_CONTINUE;
6666 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6667 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6671 return X86EMUL_CONTINUE;
6674 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6676 void *val, unsigned int bytes,
6677 struct x86_exception *exception,
6678 const struct read_write_emulator_ops *ops)
6680 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6684 if (ops->read_write_prepare &&
6685 ops->read_write_prepare(vcpu, val, bytes))
6686 return X86EMUL_CONTINUE;
6688 vcpu->mmio_nr_fragments = 0;
6690 /* Crossing a page boundary? */
6691 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6694 now = -addr & ~PAGE_MASK;
6695 rc = emulator_read_write_onepage(addr, val, now, exception,
6698 if (rc != X86EMUL_CONTINUE)
6701 if (ctxt->mode != X86EMUL_MODE_PROT64)
6707 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6709 if (rc != X86EMUL_CONTINUE)
6712 if (!vcpu->mmio_nr_fragments)
6715 gpa = vcpu->mmio_fragments[0].gpa;
6717 vcpu->mmio_needed = 1;
6718 vcpu->mmio_cur_fragment = 0;
6720 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6721 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6722 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6723 vcpu->run->mmio.phys_addr = gpa;
6725 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6728 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6732 struct x86_exception *exception)
6734 return emulator_read_write(ctxt, addr, val, bytes,
6735 exception, &read_emultor);
6738 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6742 struct x86_exception *exception)
6744 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6745 exception, &write_emultor);
6748 #define CMPXCHG_TYPE(t, ptr, old, new) \
6749 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6751 #ifdef CONFIG_X86_64
6752 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6754 # define CMPXCHG64(ptr, old, new) \
6755 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6758 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6763 struct x86_exception *exception)
6765 struct kvm_host_map map;
6766 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6772 /* guests cmpxchg8b have to be emulated atomically */
6773 if (bytes > 8 || (bytes & (bytes - 1)))
6776 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6778 if (gpa == UNMAPPED_GVA ||
6779 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6783 * Emulate the atomic as a straight write to avoid #AC if SLD is
6784 * enabled in the host and the access splits a cache line.
6786 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6787 page_line_mask = ~(cache_line_size() - 1);
6789 page_line_mask = PAGE_MASK;
6791 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6794 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6797 kaddr = map.hva + offset_in_page(gpa);
6801 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6804 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6807 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6810 exchanged = CMPXCHG64(kaddr, old, new);
6816 kvm_vcpu_unmap(vcpu, &map, true);
6819 return X86EMUL_CMPXCHG_FAILED;
6821 kvm_page_track_write(vcpu, gpa, new, bytes);
6823 return X86EMUL_CONTINUE;
6826 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6828 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6831 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6835 for (i = 0; i < vcpu->arch.pio.count; i++) {
6836 if (vcpu->arch.pio.in)
6837 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6838 vcpu->arch.pio.size, pd);
6840 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6841 vcpu->arch.pio.port, vcpu->arch.pio.size,
6845 pd += vcpu->arch.pio.size;
6850 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6851 unsigned short port, void *val,
6852 unsigned int count, bool in)
6854 vcpu->arch.pio.port = port;
6855 vcpu->arch.pio.in = in;
6856 vcpu->arch.pio.count = count;
6857 vcpu->arch.pio.size = size;
6859 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6860 vcpu->arch.pio.count = 0;
6864 vcpu->run->exit_reason = KVM_EXIT_IO;
6865 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6866 vcpu->run->io.size = size;
6867 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6868 vcpu->run->io.count = count;
6869 vcpu->run->io.port = port;
6874 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6875 unsigned short port, void *val, unsigned int count)
6879 if (vcpu->arch.pio.count)
6882 memset(vcpu->arch.pio_data, 0, size * count);
6884 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6887 memcpy(val, vcpu->arch.pio_data, size * count);
6888 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6889 vcpu->arch.pio.count = 0;
6896 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6897 int size, unsigned short port, void *val,
6900 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6904 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6905 unsigned short port, const void *val,
6908 memcpy(vcpu->arch.pio_data, val, size * count);
6909 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6910 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6913 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6914 int size, unsigned short port,
6915 const void *val, unsigned int count)
6917 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6920 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6922 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6925 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6927 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6930 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6932 if (!need_emulate_wbinvd(vcpu))
6933 return X86EMUL_CONTINUE;
6935 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6936 int cpu = get_cpu();
6938 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6939 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6940 wbinvd_ipi, NULL, 1);
6942 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6945 return X86EMUL_CONTINUE;
6948 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6950 kvm_emulate_wbinvd_noskip(vcpu);
6951 return kvm_skip_emulated_instruction(vcpu);
6953 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6957 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6959 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6962 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6963 unsigned long *dest)
6965 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6968 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6969 unsigned long value)
6972 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6975 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6977 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6980 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6982 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6983 unsigned long value;
6987 value = kvm_read_cr0(vcpu);
6990 value = vcpu->arch.cr2;
6993 value = kvm_read_cr3(vcpu);
6996 value = kvm_read_cr4(vcpu);
6999 value = kvm_get_cr8(vcpu);
7002 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7009 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7011 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7016 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7019 vcpu->arch.cr2 = val;
7022 res = kvm_set_cr3(vcpu, val);
7025 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7028 res = kvm_set_cr8(vcpu, val);
7031 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7038 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7040 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7043 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7045 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7048 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7050 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7053 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7055 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7058 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7060 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7063 static unsigned long emulator_get_cached_segment_base(
7064 struct x86_emulate_ctxt *ctxt, int seg)
7066 return get_segment_base(emul_to_vcpu(ctxt), seg);
7069 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7070 struct desc_struct *desc, u32 *base3,
7073 struct kvm_segment var;
7075 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7076 *selector = var.selector;
7079 memset(desc, 0, sizeof(*desc));
7087 set_desc_limit(desc, var.limit);
7088 set_desc_base(desc, (unsigned long)var.base);
7089 #ifdef CONFIG_X86_64
7091 *base3 = var.base >> 32;
7093 desc->type = var.type;
7095 desc->dpl = var.dpl;
7096 desc->p = var.present;
7097 desc->avl = var.avl;
7105 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7106 struct desc_struct *desc, u32 base3,
7109 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7110 struct kvm_segment var;
7112 var.selector = selector;
7113 var.base = get_desc_base(desc);
7114 #ifdef CONFIG_X86_64
7115 var.base |= ((u64)base3) << 32;
7117 var.limit = get_desc_limit(desc);
7119 var.limit = (var.limit << 12) | 0xfff;
7120 var.type = desc->type;
7121 var.dpl = desc->dpl;
7126 var.avl = desc->avl;
7127 var.present = desc->p;
7128 var.unusable = !var.present;
7131 kvm_set_segment(vcpu, &var, seg);
7135 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7136 u32 msr_index, u64 *pdata)
7138 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7141 r = kvm_get_msr(vcpu, msr_index, pdata);
7143 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7144 /* Bounce to user space */
7145 return X86EMUL_IO_NEEDED;
7151 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7152 u32 msr_index, u64 data)
7154 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7157 r = kvm_set_msr(vcpu, msr_index, data);
7159 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7160 /* Bounce to user space */
7161 return X86EMUL_IO_NEEDED;
7167 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7169 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7171 return vcpu->arch.smbase;
7174 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7176 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7178 vcpu->arch.smbase = smbase;
7181 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7184 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7187 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7188 u32 pmc, u64 *pdata)
7190 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7193 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7195 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7198 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7199 struct x86_instruction_info *info,
7200 enum x86_intercept_stage stage)
7202 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7206 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7207 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7210 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7213 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7215 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7218 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7220 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7223 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7225 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7228 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7230 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7233 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7235 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7238 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7240 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7243 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7245 return emul_to_vcpu(ctxt)->arch.hflags;
7248 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7250 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7252 kvm_smm_changed(vcpu, false);
7255 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7256 const char *smstate)
7258 return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7261 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7263 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7266 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7268 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7271 static const struct x86_emulate_ops emulate_ops = {
7272 .read_gpr = emulator_read_gpr,
7273 .write_gpr = emulator_write_gpr,
7274 .read_std = emulator_read_std,
7275 .write_std = emulator_write_std,
7276 .read_phys = kvm_read_guest_phys_system,
7277 .fetch = kvm_fetch_guest_virt,
7278 .read_emulated = emulator_read_emulated,
7279 .write_emulated = emulator_write_emulated,
7280 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7281 .invlpg = emulator_invlpg,
7282 .pio_in_emulated = emulator_pio_in_emulated,
7283 .pio_out_emulated = emulator_pio_out_emulated,
7284 .get_segment = emulator_get_segment,
7285 .set_segment = emulator_set_segment,
7286 .get_cached_segment_base = emulator_get_cached_segment_base,
7287 .get_gdt = emulator_get_gdt,
7288 .get_idt = emulator_get_idt,
7289 .set_gdt = emulator_set_gdt,
7290 .set_idt = emulator_set_idt,
7291 .get_cr = emulator_get_cr,
7292 .set_cr = emulator_set_cr,
7293 .cpl = emulator_get_cpl,
7294 .get_dr = emulator_get_dr,
7295 .set_dr = emulator_set_dr,
7296 .get_smbase = emulator_get_smbase,
7297 .set_smbase = emulator_set_smbase,
7298 .set_msr = emulator_set_msr,
7299 .get_msr = emulator_get_msr,
7300 .check_pmc = emulator_check_pmc,
7301 .read_pmc = emulator_read_pmc,
7302 .halt = emulator_halt,
7303 .wbinvd = emulator_wbinvd,
7304 .fix_hypercall = emulator_fix_hypercall,
7305 .intercept = emulator_intercept,
7306 .get_cpuid = emulator_get_cpuid,
7307 .guest_has_long_mode = emulator_guest_has_long_mode,
7308 .guest_has_movbe = emulator_guest_has_movbe,
7309 .guest_has_fxsr = emulator_guest_has_fxsr,
7310 .set_nmi_mask = emulator_set_nmi_mask,
7311 .get_hflags = emulator_get_hflags,
7312 .exiting_smm = emulator_exiting_smm,
7313 .leave_smm = emulator_leave_smm,
7314 .triple_fault = emulator_triple_fault,
7315 .set_xcr = emulator_set_xcr,
7318 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7320 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7322 * an sti; sti; sequence only disable interrupts for the first
7323 * instruction. So, if the last instruction, be it emulated or
7324 * not, left the system with the INT_STI flag enabled, it
7325 * means that the last instruction is an sti. We should not
7326 * leave the flag on in this case. The same goes for mov ss
7328 if (int_shadow & mask)
7330 if (unlikely(int_shadow || mask)) {
7331 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7333 kvm_make_request(KVM_REQ_EVENT, vcpu);
7337 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7339 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7340 if (ctxt->exception.vector == PF_VECTOR)
7341 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7343 if (ctxt->exception.error_code_valid)
7344 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7345 ctxt->exception.error_code);
7347 kvm_queue_exception(vcpu, ctxt->exception.vector);
7351 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7353 struct x86_emulate_ctxt *ctxt;
7355 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7357 pr_err("kvm: failed to allocate vcpu's emulator\n");
7362 ctxt->ops = &emulate_ops;
7363 vcpu->arch.emulate_ctxt = ctxt;
7368 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7370 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7373 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7375 ctxt->gpa_available = false;
7376 ctxt->eflags = kvm_get_rflags(vcpu);
7377 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7379 ctxt->eip = kvm_rip_read(vcpu);
7380 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7381 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7382 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7383 cs_db ? X86EMUL_MODE_PROT32 :
7384 X86EMUL_MODE_PROT16;
7385 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7386 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7387 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7389 ctxt->interruptibility = 0;
7390 ctxt->have_exception = false;
7391 ctxt->exception.vector = -1;
7392 ctxt->perm_ok = false;
7394 init_decode_cache(ctxt);
7395 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7398 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7400 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7403 init_emulate_ctxt(vcpu);
7407 ctxt->_eip = ctxt->eip + inc_eip;
7408 ret = emulate_int_real(ctxt, irq);
7410 if (ret != X86EMUL_CONTINUE) {
7411 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7413 ctxt->eip = ctxt->_eip;
7414 kvm_rip_write(vcpu, ctxt->eip);
7415 kvm_set_rflags(vcpu, ctxt->eflags);
7418 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7420 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7422 ++vcpu->stat.insn_emulation_fail;
7423 trace_kvm_emulate_insn_failed(vcpu);
7425 if (emulation_type & EMULTYPE_VMWARE_GP) {
7426 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7430 if (emulation_type & EMULTYPE_SKIP) {
7431 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7432 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7433 vcpu->run->internal.ndata = 0;
7437 kvm_queue_exception(vcpu, UD_VECTOR);
7439 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7440 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7441 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7442 vcpu->run->internal.ndata = 0;
7449 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7450 bool write_fault_to_shadow_pgtable,
7453 gpa_t gpa = cr2_or_gpa;
7456 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7459 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7460 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7463 if (!vcpu->arch.mmu->direct_map) {
7465 * Write permission should be allowed since only
7466 * write access need to be emulated.
7468 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7471 * If the mapping is invalid in guest, let cpu retry
7472 * it to generate fault.
7474 if (gpa == UNMAPPED_GVA)
7479 * Do not retry the unhandleable instruction if it faults on the
7480 * readonly host memory, otherwise it will goto a infinite loop:
7481 * retry instruction -> write #PF -> emulation fail -> retry
7482 * instruction -> ...
7484 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7487 * If the instruction failed on the error pfn, it can not be fixed,
7488 * report the error to userspace.
7490 if (is_error_noslot_pfn(pfn))
7493 kvm_release_pfn_clean(pfn);
7495 /* The instructions are well-emulated on direct mmu. */
7496 if (vcpu->arch.mmu->direct_map) {
7497 unsigned int indirect_shadow_pages;
7499 write_lock(&vcpu->kvm->mmu_lock);
7500 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7501 write_unlock(&vcpu->kvm->mmu_lock);
7503 if (indirect_shadow_pages)
7504 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7510 * if emulation was due to access to shadowed page table
7511 * and it failed try to unshadow page and re-enter the
7512 * guest to let CPU execute the instruction.
7514 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7517 * If the access faults on its page table, it can not
7518 * be fixed by unprotecting shadow page and it should
7519 * be reported to userspace.
7521 return !write_fault_to_shadow_pgtable;
7524 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7525 gpa_t cr2_or_gpa, int emulation_type)
7527 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7528 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7530 last_retry_eip = vcpu->arch.last_retry_eip;
7531 last_retry_addr = vcpu->arch.last_retry_addr;
7534 * If the emulation is caused by #PF and it is non-page_table
7535 * writing instruction, it means the VM-EXIT is caused by shadow
7536 * page protected, we can zap the shadow page and retry this
7537 * instruction directly.
7539 * Note: if the guest uses a non-page-table modifying instruction
7540 * on the PDE that points to the instruction, then we will unmap
7541 * the instruction and go to an infinite loop. So, we cache the
7542 * last retried eip and the last fault address, if we meet the eip
7543 * and the address again, we can break out of the potential infinite
7546 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7548 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7551 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7552 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7555 if (x86_page_table_writing_insn(ctxt))
7558 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7561 vcpu->arch.last_retry_eip = ctxt->eip;
7562 vcpu->arch.last_retry_addr = cr2_or_gpa;
7564 if (!vcpu->arch.mmu->direct_map)
7565 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7567 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7572 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7573 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7575 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7577 trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7580 vcpu->arch.hflags |= HF_SMM_MASK;
7582 vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7584 /* Process a latched INIT or SMI, if any. */
7585 kvm_make_request(KVM_REQ_EVENT, vcpu);
7588 kvm_mmu_reset_context(vcpu);
7591 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7600 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7601 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7606 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7608 struct kvm_run *kvm_run = vcpu->run;
7610 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7611 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7612 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7613 kvm_run->debug.arch.exception = DB_VECTOR;
7614 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7617 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7621 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7623 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7626 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7631 * rflags is the old, "raw" value of the flags. The new value has
7632 * not been saved yet.
7634 * This is correct even for TF set by the guest, because "the
7635 * processor will not generate this exception after the instruction
7636 * that sets the TF flag".
7638 if (unlikely(rflags & X86_EFLAGS_TF))
7639 r = kvm_vcpu_do_singlestep(vcpu);
7642 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7644 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7646 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7647 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7648 struct kvm_run *kvm_run = vcpu->run;
7649 unsigned long eip = kvm_get_linear_rip(vcpu);
7650 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7651 vcpu->arch.guest_debug_dr7,
7655 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7656 kvm_run->debug.arch.pc = eip;
7657 kvm_run->debug.arch.exception = DB_VECTOR;
7658 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7664 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7665 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7666 unsigned long eip = kvm_get_linear_rip(vcpu);
7667 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7672 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7681 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7683 switch (ctxt->opcode_len) {
7690 case 0xe6: /* OUT */
7694 case 0x6c: /* INS */
7696 case 0x6e: /* OUTS */
7703 case 0x33: /* RDPMC */
7713 * Decode to be emulated instruction. Return EMULATION_OK if success.
7715 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7716 void *insn, int insn_len)
7718 int r = EMULATION_OK;
7719 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7721 init_emulate_ctxt(vcpu);
7724 * We will reenter on the same instruction since we do not set
7725 * complete_userspace_io. This does not handle watchpoints yet,
7726 * those would be handled in the emulate_ops.
7728 if (!(emulation_type & EMULTYPE_SKIP) &&
7729 kvm_vcpu_check_breakpoint(vcpu, &r))
7732 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7734 trace_kvm_emulate_insn_start(vcpu);
7735 ++vcpu->stat.insn_emulation;
7739 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7741 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7742 int emulation_type, void *insn, int insn_len)
7745 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7746 bool writeback = true;
7747 bool write_fault_to_spt;
7749 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7752 vcpu->arch.l1tf_flush_l1d = true;
7755 * Clear write_fault_to_shadow_pgtable here to ensure it is
7758 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7759 vcpu->arch.write_fault_to_shadow_pgtable = false;
7761 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7762 kvm_clear_exception_queue(vcpu);
7764 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7766 if (r != EMULATION_OK) {
7767 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7768 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7769 kvm_queue_exception(vcpu, UD_VECTOR);
7772 if (reexecute_instruction(vcpu, cr2_or_gpa,
7776 if (ctxt->have_exception) {
7778 * #UD should result in just EMULATION_FAILED, and trap-like
7779 * exception should not be encountered during decode.
7781 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7782 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7783 inject_emulated_exception(vcpu);
7786 return handle_emulation_failure(vcpu, emulation_type);
7790 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7791 !is_vmware_backdoor_opcode(ctxt)) {
7792 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7797 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7798 * for kvm_skip_emulated_instruction(). The caller is responsible for
7799 * updating interruptibility state and injecting single-step #DBs.
7801 if (emulation_type & EMULTYPE_SKIP) {
7802 kvm_rip_write(vcpu, ctxt->_eip);
7803 if (ctxt->eflags & X86_EFLAGS_RF)
7804 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7808 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7811 /* this is needed for vmware backdoor interface to work since it
7812 changes registers values during IO operation */
7813 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7814 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7815 emulator_invalidate_register_cache(ctxt);
7819 if (emulation_type & EMULTYPE_PF) {
7820 /* Save the faulting GPA (cr2) in the address field */
7821 ctxt->exception.address = cr2_or_gpa;
7823 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7824 if (vcpu->arch.mmu->direct_map) {
7825 ctxt->gpa_available = true;
7826 ctxt->gpa_val = cr2_or_gpa;
7829 /* Sanitize the address out of an abundance of paranoia. */
7830 ctxt->exception.address = 0;
7833 r = x86_emulate_insn(ctxt);
7835 if (r == EMULATION_INTERCEPTED)
7838 if (r == EMULATION_FAILED) {
7839 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7843 return handle_emulation_failure(vcpu, emulation_type);
7846 if (ctxt->have_exception) {
7848 if (inject_emulated_exception(vcpu))
7850 } else if (vcpu->arch.pio.count) {
7851 if (!vcpu->arch.pio.in) {
7852 /* FIXME: return into emulator if single-stepping. */
7853 vcpu->arch.pio.count = 0;
7856 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7859 } else if (vcpu->mmio_needed) {
7860 ++vcpu->stat.mmio_exits;
7862 if (!vcpu->mmio_is_write)
7865 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7866 } else if (r == EMULATION_RESTART)
7872 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7873 toggle_interruptibility(vcpu, ctxt->interruptibility);
7874 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7875 if (!ctxt->have_exception ||
7876 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7877 kvm_rip_write(vcpu, ctxt->eip);
7878 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7879 r = kvm_vcpu_do_singlestep(vcpu);
7880 if (kvm_x86_ops.update_emulated_instruction)
7881 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7882 __kvm_set_rflags(vcpu, ctxt->eflags);
7886 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7887 * do nothing, and it will be requested again as soon as
7888 * the shadow expires. But we still need to check here,
7889 * because POPF has no interrupt shadow.
7891 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7892 kvm_make_request(KVM_REQ_EVENT, vcpu);
7894 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7899 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7901 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7903 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7905 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7906 void *insn, int insn_len)
7908 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7910 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7912 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7914 vcpu->arch.pio.count = 0;
7918 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7920 vcpu->arch.pio.count = 0;
7922 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7925 return kvm_skip_emulated_instruction(vcpu);
7928 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7929 unsigned short port)
7931 unsigned long val = kvm_rax_read(vcpu);
7932 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7938 * Workaround userspace that relies on old KVM behavior of %rip being
7939 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7942 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7943 vcpu->arch.complete_userspace_io =
7944 complete_fast_pio_out_port_0x7e;
7945 kvm_skip_emulated_instruction(vcpu);
7947 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7948 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7953 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7957 /* We should only ever be called with arch.pio.count equal to 1 */
7958 BUG_ON(vcpu->arch.pio.count != 1);
7960 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7961 vcpu->arch.pio.count = 0;
7965 /* For size less than 4 we merge, else we zero extend */
7966 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7969 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7970 * the copy and tracing
7972 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7973 kvm_rax_write(vcpu, val);
7975 return kvm_skip_emulated_instruction(vcpu);
7978 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7979 unsigned short port)
7984 /* For size less than 4 we merge, else we zero extend */
7985 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7987 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7989 kvm_rax_write(vcpu, val);
7993 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7994 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7999 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8004 ret = kvm_fast_pio_in(vcpu, size, port);
8006 ret = kvm_fast_pio_out(vcpu, size, port);
8007 return ret && kvm_skip_emulated_instruction(vcpu);
8009 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8011 static int kvmclock_cpu_down_prep(unsigned int cpu)
8013 __this_cpu_write(cpu_tsc_khz, 0);
8017 static void tsc_khz_changed(void *data)
8019 struct cpufreq_freqs *freq = data;
8020 unsigned long khz = 0;
8024 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8025 khz = cpufreq_quick_get(raw_smp_processor_id());
8028 __this_cpu_write(cpu_tsc_khz, khz);
8031 #ifdef CONFIG_X86_64
8032 static void kvm_hyperv_tsc_notifier(void)
8035 struct kvm_vcpu *vcpu;
8037 unsigned long flags;
8039 mutex_lock(&kvm_lock);
8040 list_for_each_entry(kvm, &vm_list, vm_list)
8041 kvm_make_mclock_inprogress_request(kvm);
8043 hyperv_stop_tsc_emulation();
8045 /* TSC frequency always matches when on Hyper-V */
8046 for_each_present_cpu(cpu)
8047 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8048 kvm_max_guest_tsc_khz = tsc_khz;
8050 list_for_each_entry(kvm, &vm_list, vm_list) {
8051 struct kvm_arch *ka = &kvm->arch;
8053 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
8054 pvclock_update_vm_gtod_copy(kvm);
8055 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
8057 kvm_for_each_vcpu(cpu, vcpu, kvm)
8058 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8060 kvm_for_each_vcpu(cpu, vcpu, kvm)
8061 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
8063 mutex_unlock(&kvm_lock);
8067 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8070 struct kvm_vcpu *vcpu;
8071 int i, send_ipi = 0;
8074 * We allow guests to temporarily run on slowing clocks,
8075 * provided we notify them after, or to run on accelerating
8076 * clocks, provided we notify them before. Thus time never
8079 * However, we have a problem. We can't atomically update
8080 * the frequency of a given CPU from this function; it is
8081 * merely a notifier, which can be called from any CPU.
8082 * Changing the TSC frequency at arbitrary points in time
8083 * requires a recomputation of local variables related to
8084 * the TSC for each VCPU. We must flag these local variables
8085 * to be updated and be sure the update takes place with the
8086 * new frequency before any guests proceed.
8088 * Unfortunately, the combination of hotplug CPU and frequency
8089 * change creates an intractable locking scenario; the order
8090 * of when these callouts happen is undefined with respect to
8091 * CPU hotplug, and they can race with each other. As such,
8092 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8093 * undefined; you can actually have a CPU frequency change take
8094 * place in between the computation of X and the setting of the
8095 * variable. To protect against this problem, all updates of
8096 * the per_cpu tsc_khz variable are done in an interrupt
8097 * protected IPI, and all callers wishing to update the value
8098 * must wait for a synchronous IPI to complete (which is trivial
8099 * if the caller is on the CPU already). This establishes the
8100 * necessary total order on variable updates.
8102 * Note that because a guest time update may take place
8103 * anytime after the setting of the VCPU's request bit, the
8104 * correct TSC value must be set before the request. However,
8105 * to ensure the update actually makes it to any guest which
8106 * starts running in hardware virtualization between the set
8107 * and the acquisition of the spinlock, we must also ping the
8108 * CPU after setting the request bit.
8112 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8114 mutex_lock(&kvm_lock);
8115 list_for_each_entry(kvm, &vm_list, vm_list) {
8116 kvm_for_each_vcpu(i, vcpu, kvm) {
8117 if (vcpu->cpu != cpu)
8119 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8120 if (vcpu->cpu != raw_smp_processor_id())
8124 mutex_unlock(&kvm_lock);
8126 if (freq->old < freq->new && send_ipi) {
8128 * We upscale the frequency. Must make the guest
8129 * doesn't see old kvmclock values while running with
8130 * the new frequency, otherwise we risk the guest sees
8131 * time go backwards.
8133 * In case we update the frequency for another cpu
8134 * (which might be in guest context) send an interrupt
8135 * to kick the cpu out of guest context. Next time
8136 * guest context is entered kvmclock will be updated,
8137 * so the guest will not see stale values.
8139 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8143 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8146 struct cpufreq_freqs *freq = data;
8149 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8151 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8154 for_each_cpu(cpu, freq->policy->cpus)
8155 __kvmclock_cpufreq_notifier(freq, cpu);
8160 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8161 .notifier_call = kvmclock_cpufreq_notifier
8164 static int kvmclock_cpu_online(unsigned int cpu)
8166 tsc_khz_changed(NULL);
8170 static void kvm_timer_init(void)
8172 max_tsc_khz = tsc_khz;
8174 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8175 #ifdef CONFIG_CPU_FREQ
8176 struct cpufreq_policy *policy;
8180 policy = cpufreq_cpu_get(cpu);
8182 if (policy->cpuinfo.max_freq)
8183 max_tsc_khz = policy->cpuinfo.max_freq;
8184 cpufreq_cpu_put(policy);
8188 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8189 CPUFREQ_TRANSITION_NOTIFIER);
8192 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8193 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8196 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8197 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8199 int kvm_is_in_guest(void)
8201 return __this_cpu_read(current_vcpu) != NULL;
8204 static int kvm_is_user_mode(void)
8208 if (__this_cpu_read(current_vcpu))
8209 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8211 return user_mode != 0;
8214 static unsigned long kvm_get_guest_ip(void)
8216 unsigned long ip = 0;
8218 if (__this_cpu_read(current_vcpu))
8219 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8224 static void kvm_handle_intel_pt_intr(void)
8226 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8228 kvm_make_request(KVM_REQ_PMI, vcpu);
8229 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8230 (unsigned long *)&vcpu->arch.pmu.global_status);
8233 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8234 .is_in_guest = kvm_is_in_guest,
8235 .is_user_mode = kvm_is_user_mode,
8236 .get_guest_ip = kvm_get_guest_ip,
8237 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8240 #ifdef CONFIG_X86_64
8241 static void pvclock_gtod_update_fn(struct work_struct *work)
8245 struct kvm_vcpu *vcpu;
8248 mutex_lock(&kvm_lock);
8249 list_for_each_entry(kvm, &vm_list, vm_list)
8250 kvm_for_each_vcpu(i, vcpu, kvm)
8251 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8252 atomic_set(&kvm_guest_has_master_clock, 0);
8253 mutex_unlock(&kvm_lock);
8256 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8259 * Indirection to move queue_work() out of the tk_core.seq write held
8260 * region to prevent possible deadlocks against time accessors which
8261 * are invoked with work related locks held.
8263 static void pvclock_irq_work_fn(struct irq_work *w)
8265 queue_work(system_long_wq, &pvclock_gtod_work);
8268 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8271 * Notification about pvclock gtod data update.
8273 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8276 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8277 struct timekeeper *tk = priv;
8279 update_pvclock_gtod(tk);
8282 * Disable master clock if host does not trust, or does not use,
8283 * TSC based clocksource. Delegate queue_work() to irq_work as
8284 * this is invoked with tk_core.seq write held.
8286 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8287 atomic_read(&kvm_guest_has_master_clock) != 0)
8288 irq_work_queue(&pvclock_irq_work);
8292 static struct notifier_block pvclock_gtod_notifier = {
8293 .notifier_call = pvclock_gtod_notify,
8297 int kvm_arch_init(void *opaque)
8299 struct kvm_x86_init_ops *ops = opaque;
8302 if (kvm_x86_ops.hardware_enable) {
8303 printk(KERN_ERR "kvm: already loaded the other module\n");
8308 if (!ops->cpu_has_kvm_support()) {
8309 pr_err_ratelimited("kvm: no hardware support\n");
8313 if (ops->disabled_by_bios()) {
8314 pr_err_ratelimited("kvm: disabled by bios\n");
8320 * KVM explicitly assumes that the guest has an FPU and
8321 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8322 * vCPU's FPU state as a fxregs_state struct.
8324 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8325 printk(KERN_ERR "kvm: inadequate fpu\n");
8331 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8332 __alignof__(struct fpu), SLAB_ACCOUNT,
8334 if (!x86_fpu_cache) {
8335 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8339 x86_emulator_cache = kvm_alloc_emulator_cache();
8340 if (!x86_emulator_cache) {
8341 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8342 goto out_free_x86_fpu_cache;
8345 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8346 if (!user_return_msrs) {
8347 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8348 goto out_free_x86_emulator_cache;
8350 kvm_nr_uret_msrs = 0;
8352 r = kvm_mmu_module_init();
8354 goto out_free_percpu;
8358 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8360 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8361 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8362 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8365 if (pi_inject_timer == -1)
8366 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8367 #ifdef CONFIG_X86_64
8368 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8370 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8371 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8377 free_percpu(user_return_msrs);
8378 out_free_x86_emulator_cache:
8379 kmem_cache_destroy(x86_emulator_cache);
8380 out_free_x86_fpu_cache:
8381 kmem_cache_destroy(x86_fpu_cache);
8386 void kvm_arch_exit(void)
8388 #ifdef CONFIG_X86_64
8389 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8390 clear_hv_tscchange_cb();
8393 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8395 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8396 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8397 CPUFREQ_TRANSITION_NOTIFIER);
8398 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8399 #ifdef CONFIG_X86_64
8400 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8401 irq_work_sync(&pvclock_irq_work);
8402 cancel_work_sync(&pvclock_gtod_work);
8404 kvm_x86_ops.hardware_enable = NULL;
8405 kvm_mmu_module_exit();
8406 free_percpu(user_return_msrs);
8407 kmem_cache_destroy(x86_emulator_cache);
8408 kmem_cache_destroy(x86_fpu_cache);
8409 #ifdef CONFIG_KVM_XEN
8410 static_key_deferred_flush(&kvm_xen_enabled);
8411 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8415 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8417 ++vcpu->stat.halt_exits;
8418 if (lapic_in_kernel(vcpu)) {
8419 vcpu->arch.mp_state = state;
8422 vcpu->run->exit_reason = reason;
8427 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8429 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8431 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8433 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8435 int ret = kvm_skip_emulated_instruction(vcpu);
8437 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8438 * KVM_EXIT_DEBUG here.
8440 return kvm_vcpu_halt(vcpu) && ret;
8442 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8444 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8446 int ret = kvm_skip_emulated_instruction(vcpu);
8448 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8450 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8452 #ifdef CONFIG_X86_64
8453 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8454 unsigned long clock_type)
8456 struct kvm_clock_pairing clock_pairing;
8457 struct timespec64 ts;
8461 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8462 return -KVM_EOPNOTSUPP;
8464 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8465 return -KVM_EOPNOTSUPP;
8467 clock_pairing.sec = ts.tv_sec;
8468 clock_pairing.nsec = ts.tv_nsec;
8469 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8470 clock_pairing.flags = 0;
8471 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8474 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8475 sizeof(struct kvm_clock_pairing)))
8483 * kvm_pv_kick_cpu_op: Kick a vcpu.
8485 * @apicid - apicid of vcpu to be kicked.
8487 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8489 struct kvm_lapic_irq lapic_irq;
8491 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8492 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8493 lapic_irq.level = 0;
8494 lapic_irq.dest_id = apicid;
8495 lapic_irq.msi_redir_hint = false;
8497 lapic_irq.delivery_mode = APIC_DM_REMRD;
8498 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8501 bool kvm_apicv_activated(struct kvm *kvm)
8503 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8505 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8507 static void kvm_apicv_init(struct kvm *kvm)
8510 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8511 &kvm->arch.apicv_inhibit_reasons);
8513 set_bit(APICV_INHIBIT_REASON_DISABLE,
8514 &kvm->arch.apicv_inhibit_reasons);
8517 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8519 struct kvm_vcpu *target = NULL;
8520 struct kvm_apic_map *map;
8522 vcpu->stat.directed_yield_attempted++;
8524 if (single_task_running())
8528 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8530 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8531 target = map->phys_map[dest_id]->vcpu;
8535 if (!target || !READ_ONCE(target->ready))
8538 /* Ignore requests to yield to self */
8542 if (kvm_vcpu_yield_to(target) <= 0)
8545 vcpu->stat.directed_yield_successful++;
8551 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8553 unsigned long nr, a0, a1, a2, a3, ret;
8556 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8557 return kvm_xen_hypercall(vcpu);
8559 if (kvm_hv_hypercall_enabled(vcpu))
8560 return kvm_hv_hypercall(vcpu);
8562 nr = kvm_rax_read(vcpu);
8563 a0 = kvm_rbx_read(vcpu);
8564 a1 = kvm_rcx_read(vcpu);
8565 a2 = kvm_rdx_read(vcpu);
8566 a3 = kvm_rsi_read(vcpu);
8568 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8570 op_64_bit = is_64_bit_mode(vcpu);
8579 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8587 case KVM_HC_VAPIC_POLL_IRQ:
8590 case KVM_HC_KICK_CPU:
8591 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8594 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8595 kvm_sched_yield(vcpu, a1);
8598 #ifdef CONFIG_X86_64
8599 case KVM_HC_CLOCK_PAIRING:
8600 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8603 case KVM_HC_SEND_IPI:
8604 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8607 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8609 case KVM_HC_SCHED_YIELD:
8610 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8613 kvm_sched_yield(vcpu, a0);
8623 kvm_rax_write(vcpu, ret);
8625 ++vcpu->stat.hypercalls;
8626 return kvm_skip_emulated_instruction(vcpu);
8628 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8630 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8632 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8633 char instruction[3];
8634 unsigned long rip = kvm_rip_read(vcpu);
8636 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8638 return emulator_write_emulated(ctxt, rip, instruction, 3,
8642 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8644 return vcpu->run->request_interrupt_window &&
8645 likely(!pic_in_kernel(vcpu->kvm));
8648 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8650 struct kvm_run *kvm_run = vcpu->run;
8653 * if_flag is obsolete and useless, so do not bother
8654 * setting it for SEV-ES guests. Userspace can just
8655 * use kvm_run->ready_for_interrupt_injection.
8657 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8658 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8660 kvm_run->cr8 = kvm_get_cr8(vcpu);
8661 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8662 kvm_run->ready_for_interrupt_injection =
8663 pic_in_kernel(vcpu->kvm) ||
8664 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8667 kvm_run->flags |= KVM_RUN_X86_SMM;
8670 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8674 if (!kvm_x86_ops.update_cr8_intercept)
8677 if (!lapic_in_kernel(vcpu))
8680 if (vcpu->arch.apicv_active)
8683 if (!vcpu->arch.apic->vapic_addr)
8684 max_irr = kvm_lapic_find_highest_irr(vcpu);
8691 tpr = kvm_lapic_get_cr8(vcpu);
8693 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8697 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8699 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8700 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8704 return kvm_x86_ops.nested_ops->check_events(vcpu);
8707 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8709 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8710 vcpu->arch.exception.error_code = false;
8711 static_call(kvm_x86_queue_exception)(vcpu);
8714 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8717 bool can_inject = true;
8719 /* try to reinject previous events if any */
8721 if (vcpu->arch.exception.injected) {
8722 kvm_inject_exception(vcpu);
8726 * Do not inject an NMI or interrupt if there is a pending
8727 * exception. Exceptions and interrupts are recognized at
8728 * instruction boundaries, i.e. the start of an instruction.
8729 * Trap-like exceptions, e.g. #DB, have higher priority than
8730 * NMIs and interrupts, i.e. traps are recognized before an
8731 * NMI/interrupt that's pending on the same instruction.
8732 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8733 * priority, but are only generated (pended) during instruction
8734 * execution, i.e. a pending fault-like exception means the
8735 * fault occurred on the *previous* instruction and must be
8736 * serviced prior to recognizing any new events in order to
8737 * fully complete the previous instruction.
8739 else if (!vcpu->arch.exception.pending) {
8740 if (vcpu->arch.nmi_injected) {
8741 static_call(kvm_x86_set_nmi)(vcpu);
8743 } else if (vcpu->arch.interrupt.injected) {
8744 static_call(kvm_x86_set_irq)(vcpu);
8749 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8750 vcpu->arch.exception.pending);
8753 * Call check_nested_events() even if we reinjected a previous event
8754 * in order for caller to determine if it should require immediate-exit
8755 * from L2 to L1 due to pending L1 events which require exit
8758 if (is_guest_mode(vcpu)) {
8759 r = kvm_check_nested_events(vcpu);
8764 /* try to inject new event if pending */
8765 if (vcpu->arch.exception.pending) {
8766 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8767 vcpu->arch.exception.has_error_code,
8768 vcpu->arch.exception.error_code);
8770 vcpu->arch.exception.pending = false;
8771 vcpu->arch.exception.injected = true;
8773 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8774 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8777 if (vcpu->arch.exception.nr == DB_VECTOR) {
8778 kvm_deliver_exception_payload(vcpu);
8779 if (vcpu->arch.dr7 & DR7_GD) {
8780 vcpu->arch.dr7 &= ~DR7_GD;
8781 kvm_update_dr7(vcpu);
8785 kvm_inject_exception(vcpu);
8790 * Finally, inject interrupt events. If an event cannot be injected
8791 * due to architectural conditions (e.g. IF=0) a window-open exit
8792 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8793 * and can architecturally be injected, but we cannot do it right now:
8794 * an interrupt could have arrived just now and we have to inject it
8795 * as a vmexit, or there could already an event in the queue, which is
8796 * indicated by can_inject. In that case we request an immediate exit
8797 * in order to make progress and get back here for another iteration.
8798 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8800 if (vcpu->arch.smi_pending) {
8801 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8805 vcpu->arch.smi_pending = false;
8806 ++vcpu->arch.smi_count;
8810 static_call(kvm_x86_enable_smi_window)(vcpu);
8813 if (vcpu->arch.nmi_pending) {
8814 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8818 --vcpu->arch.nmi_pending;
8819 vcpu->arch.nmi_injected = true;
8820 static_call(kvm_x86_set_nmi)(vcpu);
8822 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8824 if (vcpu->arch.nmi_pending)
8825 static_call(kvm_x86_enable_nmi_window)(vcpu);
8828 if (kvm_cpu_has_injectable_intr(vcpu)) {
8829 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8833 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8834 static_call(kvm_x86_set_irq)(vcpu);
8835 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8837 if (kvm_cpu_has_injectable_intr(vcpu))
8838 static_call(kvm_x86_enable_irq_window)(vcpu);
8841 if (is_guest_mode(vcpu) &&
8842 kvm_x86_ops.nested_ops->hv_timer_pending &&
8843 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8844 *req_immediate_exit = true;
8846 WARN_ON(vcpu->arch.exception.pending);
8851 *req_immediate_exit = true;
8857 static void process_nmi(struct kvm_vcpu *vcpu)
8862 * x86 is limited to one NMI running, and one NMI pending after it.
8863 * If an NMI is already in progress, limit further NMIs to just one.
8864 * Otherwise, allow two (and we'll inject the first one immediately).
8866 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8869 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8870 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8871 kvm_make_request(KVM_REQ_EVENT, vcpu);
8874 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8877 flags |= seg->g << 23;
8878 flags |= seg->db << 22;
8879 flags |= seg->l << 21;
8880 flags |= seg->avl << 20;
8881 flags |= seg->present << 15;
8882 flags |= seg->dpl << 13;
8883 flags |= seg->s << 12;
8884 flags |= seg->type << 8;
8888 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8890 struct kvm_segment seg;
8893 kvm_get_segment(vcpu, &seg, n);
8894 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8897 offset = 0x7f84 + n * 12;
8899 offset = 0x7f2c + (n - 3) * 12;
8901 put_smstate(u32, buf, offset + 8, seg.base);
8902 put_smstate(u32, buf, offset + 4, seg.limit);
8903 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8906 #ifdef CONFIG_X86_64
8907 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8909 struct kvm_segment seg;
8913 kvm_get_segment(vcpu, &seg, n);
8914 offset = 0x7e00 + n * 16;
8916 flags = enter_smm_get_segment_flags(&seg) >> 8;
8917 put_smstate(u16, buf, offset, seg.selector);
8918 put_smstate(u16, buf, offset + 2, flags);
8919 put_smstate(u32, buf, offset + 4, seg.limit);
8920 put_smstate(u64, buf, offset + 8, seg.base);
8924 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8927 struct kvm_segment seg;
8931 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8932 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8933 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8934 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8936 for (i = 0; i < 8; i++)
8937 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
8939 kvm_get_dr(vcpu, 6, &val);
8940 put_smstate(u32, buf, 0x7fcc, (u32)val);
8941 kvm_get_dr(vcpu, 7, &val);
8942 put_smstate(u32, buf, 0x7fc8, (u32)val);
8944 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8945 put_smstate(u32, buf, 0x7fc4, seg.selector);
8946 put_smstate(u32, buf, 0x7f64, seg.base);
8947 put_smstate(u32, buf, 0x7f60, seg.limit);
8948 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8950 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8951 put_smstate(u32, buf, 0x7fc0, seg.selector);
8952 put_smstate(u32, buf, 0x7f80, seg.base);
8953 put_smstate(u32, buf, 0x7f7c, seg.limit);
8954 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8956 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8957 put_smstate(u32, buf, 0x7f74, dt.address);
8958 put_smstate(u32, buf, 0x7f70, dt.size);
8960 static_call(kvm_x86_get_idt)(vcpu, &dt);
8961 put_smstate(u32, buf, 0x7f58, dt.address);
8962 put_smstate(u32, buf, 0x7f54, dt.size);
8964 for (i = 0; i < 6; i++)
8965 enter_smm_save_seg_32(vcpu, buf, i);
8967 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8970 put_smstate(u32, buf, 0x7efc, 0x00020000);
8971 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8974 #ifdef CONFIG_X86_64
8975 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8978 struct kvm_segment seg;
8982 for (i = 0; i < 16; i++)
8983 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
8985 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8986 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8988 kvm_get_dr(vcpu, 6, &val);
8989 put_smstate(u64, buf, 0x7f68, val);
8990 kvm_get_dr(vcpu, 7, &val);
8991 put_smstate(u64, buf, 0x7f60, val);
8993 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8994 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8995 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8997 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9000 put_smstate(u32, buf, 0x7efc, 0x00020064);
9002 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9004 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9005 put_smstate(u16, buf, 0x7e90, seg.selector);
9006 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9007 put_smstate(u32, buf, 0x7e94, seg.limit);
9008 put_smstate(u64, buf, 0x7e98, seg.base);
9010 static_call(kvm_x86_get_idt)(vcpu, &dt);
9011 put_smstate(u32, buf, 0x7e84, dt.size);
9012 put_smstate(u64, buf, 0x7e88, dt.address);
9014 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9015 put_smstate(u16, buf, 0x7e70, seg.selector);
9016 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9017 put_smstate(u32, buf, 0x7e74, seg.limit);
9018 put_smstate(u64, buf, 0x7e78, seg.base);
9020 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9021 put_smstate(u32, buf, 0x7e64, dt.size);
9022 put_smstate(u64, buf, 0x7e68, dt.address);
9024 for (i = 0; i < 6; i++)
9025 enter_smm_save_seg_64(vcpu, buf, i);
9029 static void enter_smm(struct kvm_vcpu *vcpu)
9031 struct kvm_segment cs, ds;
9036 memset(buf, 0, 512);
9037 #ifdef CONFIG_X86_64
9038 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9039 enter_smm_save_state_64(vcpu, buf);
9042 enter_smm_save_state_32(vcpu, buf);
9045 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9046 * state (e.g. leave guest mode) after we've saved the state into the
9047 * SMM state-save area.
9049 static_call(kvm_x86_enter_smm)(vcpu, buf);
9051 kvm_smm_changed(vcpu, true);
9052 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9054 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9055 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9057 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9059 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9060 kvm_rip_write(vcpu, 0x8000);
9062 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9063 static_call(kvm_x86_set_cr0)(vcpu, cr0);
9064 vcpu->arch.cr0 = cr0;
9066 static_call(kvm_x86_set_cr4)(vcpu, 0);
9068 /* Undocumented: IDT limit is set to zero on entry to SMM. */
9069 dt.address = dt.size = 0;
9070 static_call(kvm_x86_set_idt)(vcpu, &dt);
9072 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9074 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9075 cs.base = vcpu->arch.smbase;
9080 cs.limit = ds.limit = 0xffffffff;
9081 cs.type = ds.type = 0x3;
9082 cs.dpl = ds.dpl = 0;
9087 cs.avl = ds.avl = 0;
9088 cs.present = ds.present = 1;
9089 cs.unusable = ds.unusable = 0;
9090 cs.padding = ds.padding = 0;
9092 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9093 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9094 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9095 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9096 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9097 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9099 #ifdef CONFIG_X86_64
9100 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9101 static_call(kvm_x86_set_efer)(vcpu, 0);
9104 kvm_update_cpuid_runtime(vcpu);
9105 kvm_mmu_reset_context(vcpu);
9108 static void process_smi(struct kvm_vcpu *vcpu)
9110 vcpu->arch.smi_pending = true;
9111 kvm_make_request(KVM_REQ_EVENT, vcpu);
9114 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9115 unsigned long *vcpu_bitmap)
9119 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
9121 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
9122 NULL, vcpu_bitmap, cpus);
9124 free_cpumask_var(cpus);
9127 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9129 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9132 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9134 if (!lapic_in_kernel(vcpu))
9137 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
9138 kvm_apic_update_apicv(vcpu);
9139 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9141 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9144 * NOTE: Do not hold any lock prior to calling this.
9146 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
9147 * locked, because it calls __x86_set_memory_region() which does
9148 * synchronize_srcu(&kvm->srcu).
9150 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9152 struct kvm_vcpu *except;
9153 unsigned long old, new, expected;
9155 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9156 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9159 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
9161 expected = new = old;
9163 __clear_bit(bit, &new);
9165 __set_bit(bit, &new);
9168 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
9169 } while (old != expected);
9174 trace_kvm_apicv_update_request(activate, bit);
9175 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
9176 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
9179 * Sending request to update APICV for all other vcpus,
9180 * while update the calling vcpu immediately instead of
9181 * waiting for another #VMEXIT to handle the request.
9183 except = kvm_get_running_vcpu();
9184 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
9187 kvm_vcpu_update_apicv(except);
9189 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9191 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9193 if (!kvm_apic_present(vcpu))
9196 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9198 if (irqchip_split(vcpu->kvm))
9199 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9201 if (vcpu->arch.apicv_active)
9202 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9203 if (ioapic_in_kernel(vcpu->kvm))
9204 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9207 if (is_guest_mode(vcpu))
9208 vcpu->arch.load_eoi_exitmap_pending = true;
9210 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9213 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9215 u64 eoi_exit_bitmap[4];
9217 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9220 if (to_hv_vcpu(vcpu))
9221 bitmap_or((ulong *)eoi_exit_bitmap,
9222 vcpu->arch.ioapic_handled_vectors,
9223 to_hv_synic(vcpu)->vec_bitmap, 256);
9225 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9228 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9229 unsigned long start, unsigned long end)
9231 unsigned long apic_address;
9234 * The physical address of apic access page is stored in the VMCS.
9235 * Update it when it becomes invalid.
9237 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9238 if (start <= apic_address && apic_address < end)
9239 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9242 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9244 if (!lapic_in_kernel(vcpu))
9247 if (!kvm_x86_ops.set_apic_access_page_addr)
9250 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9253 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9255 smp_send_reschedule(vcpu->cpu);
9257 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9260 * Returns 1 to let vcpu_run() continue the guest execution loop without
9261 * exiting to the userspace. Otherwise, the value will be returned to the
9264 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9268 dm_request_for_irq_injection(vcpu) &&
9269 kvm_cpu_accept_dm_intr(vcpu);
9270 fastpath_t exit_fastpath;
9272 bool req_immediate_exit = false;
9274 /* Forbid vmenter if vcpu dirty ring is soft-full */
9275 if (unlikely(vcpu->kvm->dirty_ring_size &&
9276 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9277 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9278 trace_kvm_dirty_ring_exit(vcpu);
9283 if (kvm_request_pending(vcpu)) {
9284 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9285 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9290 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9291 kvm_mmu_unload(vcpu);
9292 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9293 __kvm_migrate_timers(vcpu);
9294 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9295 kvm_gen_update_masterclock(vcpu->kvm);
9296 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9297 kvm_gen_kvmclock_update(vcpu);
9298 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9299 r = kvm_guest_time_update(vcpu);
9303 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9304 kvm_mmu_sync_roots(vcpu);
9305 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9306 kvm_mmu_load_pgd(vcpu);
9307 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9308 kvm_vcpu_flush_tlb_all(vcpu);
9310 /* Flushing all ASIDs flushes the current ASID... */
9311 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9313 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9314 kvm_vcpu_flush_tlb_current(vcpu);
9315 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9316 kvm_vcpu_flush_tlb_guest(vcpu);
9318 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9319 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9323 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9324 if (is_guest_mode(vcpu)) {
9325 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9327 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9328 vcpu->mmio_needed = 0;
9333 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9334 /* Page is swapped out. Do synthetic halt */
9335 vcpu->arch.apf.halted = true;
9339 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9340 record_steal_time(vcpu);
9341 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9343 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9345 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9346 kvm_pmu_handle_event(vcpu);
9347 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9348 kvm_pmu_deliver_pmi(vcpu);
9349 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9350 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9351 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9352 vcpu->arch.ioapic_handled_vectors)) {
9353 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9354 vcpu->run->eoi.vector =
9355 vcpu->arch.pending_ioapic_eoi;
9360 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9361 vcpu_scan_ioapic(vcpu);
9362 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9363 vcpu_load_eoi_exitmap(vcpu);
9364 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9365 kvm_vcpu_reload_apic_access_page(vcpu);
9366 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9367 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9368 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9372 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9373 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9374 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9378 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9379 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9381 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9382 vcpu->run->hyperv = hv_vcpu->exit;
9388 * KVM_REQ_HV_STIMER has to be processed after
9389 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9390 * depend on the guest clock being up-to-date
9392 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9393 kvm_hv_process_stimers(vcpu);
9394 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9395 kvm_vcpu_update_apicv(vcpu);
9396 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9397 kvm_check_async_pf_completion(vcpu);
9398 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9399 static_call(kvm_x86_msr_filter_changed)(vcpu);
9401 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9402 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9405 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9406 kvm_xen_has_interrupt(vcpu)) {
9407 ++vcpu->stat.req_event;
9408 r = kvm_apic_accept_events(vcpu);
9413 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9418 r = inject_pending_event(vcpu, &req_immediate_exit);
9424 static_call(kvm_x86_enable_irq_window)(vcpu);
9426 if (kvm_lapic_enabled(vcpu)) {
9427 update_cr8_intercept(vcpu);
9428 kvm_lapic_sync_to_vapic(vcpu);
9432 r = kvm_mmu_reload(vcpu);
9434 goto cancel_injection;
9439 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9442 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9443 * IPI are then delayed after guest entry, which ensures that they
9444 * result in virtual interrupt delivery.
9446 local_irq_disable();
9447 vcpu->mode = IN_GUEST_MODE;
9449 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9452 * 1) We should set ->mode before checking ->requests. Please see
9453 * the comment in kvm_vcpu_exiting_guest_mode().
9455 * 2) For APICv, we should set ->mode before checking PID.ON. This
9456 * pairs with the memory barrier implicit in pi_test_and_set_on
9457 * (see vmx_deliver_posted_interrupt).
9459 * 3) This also orders the write to mode from any reads to the page
9460 * tables done while the VCPU is running. Please see the comment
9461 * in kvm_flush_remote_tlbs.
9463 smp_mb__after_srcu_read_unlock();
9466 * This handles the case where a posted interrupt was
9467 * notified with kvm_vcpu_kick.
9469 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9470 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9472 if (kvm_vcpu_exit_request(vcpu)) {
9473 vcpu->mode = OUTSIDE_GUEST_MODE;
9477 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9479 goto cancel_injection;
9482 if (req_immediate_exit) {
9483 kvm_make_request(KVM_REQ_EVENT, vcpu);
9484 static_call(kvm_x86_request_immediate_exit)(vcpu);
9487 fpregs_assert_state_consistent();
9488 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9489 switch_fpu_return();
9491 if (unlikely(vcpu->arch.switch_db_regs)) {
9493 set_debugreg(vcpu->arch.eff_db[0], 0);
9494 set_debugreg(vcpu->arch.eff_db[1], 1);
9495 set_debugreg(vcpu->arch.eff_db[2], 2);
9496 set_debugreg(vcpu->arch.eff_db[3], 3);
9497 set_debugreg(vcpu->arch.dr6, 6);
9498 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9502 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9503 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9506 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9507 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9511 if (vcpu->arch.apicv_active)
9512 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9516 * Do this here before restoring debug registers on the host. And
9517 * since we do this before handling the vmexit, a DR access vmexit
9518 * can (a) read the correct value of the debug registers, (b) set
9519 * KVM_DEBUGREG_WONT_EXIT again.
9521 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9522 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9523 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9524 kvm_update_dr0123(vcpu);
9525 kvm_update_dr7(vcpu);
9526 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9530 * If the guest has used debug registers, at least dr7
9531 * will be disabled while returning to the host.
9532 * If we don't have active breakpoints in the host, we don't
9533 * care about the messed up debug address registers. But if
9534 * we have some of them active, restore the old state.
9536 if (hw_breakpoint_active())
9537 hw_breakpoint_restore();
9539 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9540 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9542 vcpu->mode = OUTSIDE_GUEST_MODE;
9545 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9548 * Consume any pending interrupts, including the possible source of
9549 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9550 * An instruction is required after local_irq_enable() to fully unblock
9551 * interrupts on processors that implement an interrupt shadow, the
9552 * stat.exits increment will do nicely.
9554 kvm_before_interrupt(vcpu);
9557 local_irq_disable();
9558 kvm_after_interrupt(vcpu);
9561 * Wait until after servicing IRQs to account guest time so that any
9562 * ticks that occurred while running the guest are properly accounted
9563 * to the guest. Waiting until IRQs are enabled degrades the accuracy
9564 * of accounting via context tracking, but the loss of accuracy is
9565 * acceptable for all known use cases.
9567 vtime_account_guest_exit();
9569 if (lapic_in_kernel(vcpu)) {
9570 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9571 if (delta != S64_MIN) {
9572 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9573 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9580 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9583 * Profile KVM exit RIPs:
9585 if (unlikely(prof_on == KVM_PROFILING)) {
9586 unsigned long rip = kvm_rip_read(vcpu);
9587 profile_hit(KVM_PROFILING, (void *)rip);
9590 if (unlikely(vcpu->arch.tsc_always_catchup))
9591 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9593 if (vcpu->arch.apic_attention)
9594 kvm_lapic_sync_from_vapic(vcpu);
9596 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9600 if (req_immediate_exit)
9601 kvm_make_request(KVM_REQ_EVENT, vcpu);
9602 static_call(kvm_x86_cancel_injection)(vcpu);
9603 if (unlikely(vcpu->arch.apic_attention))
9604 kvm_lapic_sync_from_vapic(vcpu);
9609 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9611 if (!kvm_arch_vcpu_runnable(vcpu) &&
9612 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9613 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9614 kvm_vcpu_block(vcpu);
9615 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9617 if (kvm_x86_ops.post_block)
9618 static_call(kvm_x86_post_block)(vcpu);
9620 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9624 if (kvm_apic_accept_events(vcpu) < 0)
9626 switch(vcpu->arch.mp_state) {
9627 case KVM_MP_STATE_HALTED:
9628 case KVM_MP_STATE_AP_RESET_HOLD:
9629 vcpu->arch.pv.pv_unhalted = false;
9630 vcpu->arch.mp_state =
9631 KVM_MP_STATE_RUNNABLE;
9633 case KVM_MP_STATE_RUNNABLE:
9634 vcpu->arch.apf.halted = false;
9636 case KVM_MP_STATE_INIT_RECEIVED:
9644 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9646 if (is_guest_mode(vcpu))
9647 kvm_check_nested_events(vcpu);
9649 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9650 !vcpu->arch.apf.halted);
9653 static int vcpu_run(struct kvm_vcpu *vcpu)
9656 struct kvm *kvm = vcpu->kvm;
9658 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9659 vcpu->arch.l1tf_flush_l1d = true;
9662 if (kvm_vcpu_running(vcpu)) {
9663 r = vcpu_enter_guest(vcpu);
9665 r = vcpu_block(kvm, vcpu);
9671 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9672 if (kvm_cpu_has_pending_timer(vcpu))
9673 kvm_inject_pending_timer_irqs(vcpu);
9675 if (dm_request_for_irq_injection(vcpu) &&
9676 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9678 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9679 ++vcpu->stat.request_irq_exits;
9683 if (__xfer_to_guest_mode_work_pending()) {
9684 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9685 r = xfer_to_guest_mode_handle_work(vcpu);
9688 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9692 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9697 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9701 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9702 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9703 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9707 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9709 BUG_ON(!vcpu->arch.pio.count);
9711 return complete_emulated_io(vcpu);
9715 * Implements the following, as a state machine:
9719 * for each mmio piece in the fragment
9727 * for each mmio piece in the fragment
9732 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9734 struct kvm_run *run = vcpu->run;
9735 struct kvm_mmio_fragment *frag;
9738 BUG_ON(!vcpu->mmio_needed);
9740 /* Complete previous fragment */
9741 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9742 len = min(8u, frag->len);
9743 if (!vcpu->mmio_is_write)
9744 memcpy(frag->data, run->mmio.data, len);
9746 if (frag->len <= 8) {
9747 /* Switch to the next fragment. */
9749 vcpu->mmio_cur_fragment++;
9751 /* Go forward to the next mmio piece. */
9757 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9758 vcpu->mmio_needed = 0;
9760 /* FIXME: return into emulator if single-stepping. */
9761 if (vcpu->mmio_is_write)
9763 vcpu->mmio_read_completed = 1;
9764 return complete_emulated_io(vcpu);
9767 run->exit_reason = KVM_EXIT_MMIO;
9768 run->mmio.phys_addr = frag->gpa;
9769 if (vcpu->mmio_is_write)
9770 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9771 run->mmio.len = min(8u, frag->len);
9772 run->mmio.is_write = vcpu->mmio_is_write;
9773 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9777 static void kvm_save_current_fpu(struct fpu *fpu)
9780 * If the target FPU state is not resident in the CPU registers, just
9781 * memcpy() from current, else save CPU state directly to the target.
9783 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9784 memcpy(&fpu->state, ¤t->thread.fpu.state,
9785 fpu_kernel_xstate_size);
9787 copy_fpregs_to_fpstate(fpu);
9790 /* Swap (qemu) user FPU context for the guest FPU context. */
9791 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9795 kvm_save_current_fpu(vcpu->arch.user_fpu);
9798 * Guests with protected state can't have it set by the hypervisor,
9799 * so skip trying to set it.
9801 if (vcpu->arch.guest_fpu)
9802 /* PKRU is separately restored in kvm_x86_ops.run. */
9803 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9804 ~XFEATURE_MASK_PKRU);
9806 fpregs_mark_activate();
9812 /* When vcpu_run ends, restore user space FPU context. */
9813 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9818 * Guests with protected state can't have it read by the hypervisor,
9819 * so skip trying to save it.
9821 if (vcpu->arch.guest_fpu)
9822 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9824 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9826 fpregs_mark_activate();
9829 ++vcpu->stat.fpu_reload;
9833 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9835 struct kvm_run *kvm_run = vcpu->run;
9839 kvm_sigset_activate(vcpu);
9841 kvm_load_guest_fpu(vcpu);
9843 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9844 if (kvm_run->immediate_exit) {
9848 kvm_vcpu_block(vcpu);
9849 if (kvm_apic_accept_events(vcpu) < 0) {
9853 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9855 if (signal_pending(current)) {
9857 kvm_run->exit_reason = KVM_EXIT_INTR;
9858 ++vcpu->stat.signal_exits;
9863 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9868 if (kvm_run->kvm_dirty_regs) {
9869 r = sync_regs(vcpu);
9874 /* re-sync apic's tpr */
9875 if (!lapic_in_kernel(vcpu)) {
9876 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9882 if (unlikely(vcpu->arch.complete_userspace_io)) {
9883 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9884 vcpu->arch.complete_userspace_io = NULL;
9889 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9891 if (kvm_run->immediate_exit)
9897 kvm_put_guest_fpu(vcpu);
9898 if (kvm_run->kvm_valid_regs)
9900 post_kvm_run_save(vcpu);
9901 kvm_sigset_deactivate(vcpu);
9907 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9909 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9911 * We are here if userspace calls get_regs() in the middle of
9912 * instruction emulation. Registers state needs to be copied
9913 * back from emulation context to vcpu. Userspace shouldn't do
9914 * that usually, but some bad designed PV devices (vmware
9915 * backdoor interface) need this to work
9917 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9918 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9920 regs->rax = kvm_rax_read(vcpu);
9921 regs->rbx = kvm_rbx_read(vcpu);
9922 regs->rcx = kvm_rcx_read(vcpu);
9923 regs->rdx = kvm_rdx_read(vcpu);
9924 regs->rsi = kvm_rsi_read(vcpu);
9925 regs->rdi = kvm_rdi_read(vcpu);
9926 regs->rsp = kvm_rsp_read(vcpu);
9927 regs->rbp = kvm_rbp_read(vcpu);
9928 #ifdef CONFIG_X86_64
9929 regs->r8 = kvm_r8_read(vcpu);
9930 regs->r9 = kvm_r9_read(vcpu);
9931 regs->r10 = kvm_r10_read(vcpu);
9932 regs->r11 = kvm_r11_read(vcpu);
9933 regs->r12 = kvm_r12_read(vcpu);
9934 regs->r13 = kvm_r13_read(vcpu);
9935 regs->r14 = kvm_r14_read(vcpu);
9936 regs->r15 = kvm_r15_read(vcpu);
9939 regs->rip = kvm_rip_read(vcpu);
9940 regs->rflags = kvm_get_rflags(vcpu);
9943 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9946 __get_regs(vcpu, regs);
9951 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9953 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9954 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9956 kvm_rax_write(vcpu, regs->rax);
9957 kvm_rbx_write(vcpu, regs->rbx);
9958 kvm_rcx_write(vcpu, regs->rcx);
9959 kvm_rdx_write(vcpu, regs->rdx);
9960 kvm_rsi_write(vcpu, regs->rsi);
9961 kvm_rdi_write(vcpu, regs->rdi);
9962 kvm_rsp_write(vcpu, regs->rsp);
9963 kvm_rbp_write(vcpu, regs->rbp);
9964 #ifdef CONFIG_X86_64
9965 kvm_r8_write(vcpu, regs->r8);
9966 kvm_r9_write(vcpu, regs->r9);
9967 kvm_r10_write(vcpu, regs->r10);
9968 kvm_r11_write(vcpu, regs->r11);
9969 kvm_r12_write(vcpu, regs->r12);
9970 kvm_r13_write(vcpu, regs->r13);
9971 kvm_r14_write(vcpu, regs->r14);
9972 kvm_r15_write(vcpu, regs->r15);
9975 kvm_rip_write(vcpu, regs->rip);
9976 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9978 vcpu->arch.exception.pending = false;
9980 kvm_make_request(KVM_REQ_EVENT, vcpu);
9983 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9986 __set_regs(vcpu, regs);
9991 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9993 struct kvm_segment cs;
9995 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9999 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10001 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10003 struct desc_ptr dt;
10005 if (vcpu->arch.guest_state_protected)
10006 goto skip_protected_regs;
10008 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10009 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10010 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10011 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10012 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10013 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10015 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10016 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10018 static_call(kvm_x86_get_idt)(vcpu, &dt);
10019 sregs->idt.limit = dt.size;
10020 sregs->idt.base = dt.address;
10021 static_call(kvm_x86_get_gdt)(vcpu, &dt);
10022 sregs->gdt.limit = dt.size;
10023 sregs->gdt.base = dt.address;
10025 sregs->cr2 = vcpu->arch.cr2;
10026 sregs->cr3 = kvm_read_cr3(vcpu);
10028 skip_protected_regs:
10029 sregs->cr0 = kvm_read_cr0(vcpu);
10030 sregs->cr4 = kvm_read_cr4(vcpu);
10031 sregs->cr8 = kvm_get_cr8(vcpu);
10032 sregs->efer = vcpu->arch.efer;
10033 sregs->apic_base = kvm_get_apic_base(vcpu);
10036 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10038 __get_sregs_common(vcpu, sregs);
10040 if (vcpu->arch.guest_state_protected)
10043 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10044 set_bit(vcpu->arch.interrupt.nr,
10045 (unsigned long *)sregs->interrupt_bitmap);
10048 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10052 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10054 if (vcpu->arch.guest_state_protected)
10057 if (is_pae_paging(vcpu)) {
10058 for (i = 0 ; i < 4 ; i++)
10059 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10060 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10064 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10065 struct kvm_sregs *sregs)
10068 __get_sregs(vcpu, sregs);
10073 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10074 struct kvm_mp_state *mp_state)
10079 if (kvm_mpx_supported())
10080 kvm_load_guest_fpu(vcpu);
10082 r = kvm_apic_accept_events(vcpu);
10087 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10088 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10089 vcpu->arch.pv.pv_unhalted)
10090 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10092 mp_state->mp_state = vcpu->arch.mp_state;
10095 if (kvm_mpx_supported())
10096 kvm_put_guest_fpu(vcpu);
10101 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10102 struct kvm_mp_state *mp_state)
10108 if (!lapic_in_kernel(vcpu) &&
10109 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10113 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10114 * INIT state; latched init should be reported using
10115 * KVM_SET_VCPU_EVENTS, so reject it here.
10117 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10118 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10119 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10122 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10123 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10124 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10126 vcpu->arch.mp_state = mp_state->mp_state;
10127 kvm_make_request(KVM_REQ_EVENT, vcpu);
10135 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10136 int reason, bool has_error_code, u32 error_code)
10138 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10141 init_emulate_ctxt(vcpu);
10143 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10144 has_error_code, error_code);
10146 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10147 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10148 vcpu->run->internal.ndata = 0;
10152 kvm_rip_write(vcpu, ctxt->eip);
10153 kvm_set_rflags(vcpu, ctxt->eflags);
10156 EXPORT_SYMBOL_GPL(kvm_task_switch);
10158 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10160 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10162 * When EFER.LME and CR0.PG are set, the processor is in
10163 * 64-bit mode (though maybe in a 32-bit code segment).
10164 * CR4.PAE and EFER.LMA must be set.
10166 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10168 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10172 * Not in 64-bit mode: EFER.LMA is clear and the code
10173 * segment cannot be 64-bit.
10175 if (sregs->efer & EFER_LMA || sregs->cs.l)
10179 return kvm_is_valid_cr4(vcpu, sregs->cr4);
10182 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10183 int *mmu_reset_needed, bool update_pdptrs)
10185 struct msr_data apic_base_msr;
10187 struct desc_ptr dt;
10189 if (!kvm_is_valid_sregs(vcpu, sregs))
10192 apic_base_msr.data = sregs->apic_base;
10193 apic_base_msr.host_initiated = true;
10194 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10197 if (vcpu->arch.guest_state_protected)
10200 dt.size = sregs->idt.limit;
10201 dt.address = sregs->idt.base;
10202 static_call(kvm_x86_set_idt)(vcpu, &dt);
10203 dt.size = sregs->gdt.limit;
10204 dt.address = sregs->gdt.base;
10205 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10207 vcpu->arch.cr2 = sregs->cr2;
10208 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10209 vcpu->arch.cr3 = sregs->cr3;
10210 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10212 kvm_set_cr8(vcpu, sregs->cr8);
10214 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10215 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10217 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10218 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10219 vcpu->arch.cr0 = sregs->cr0;
10221 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10222 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10224 if (update_pdptrs) {
10225 idx = srcu_read_lock(&vcpu->kvm->srcu);
10226 if (is_pae_paging(vcpu)) {
10227 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10228 *mmu_reset_needed = 1;
10230 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10233 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10234 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10235 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10236 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10237 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10238 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10240 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10241 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10243 update_cr8_intercept(vcpu);
10245 /* Older userspace won't unhalt the vcpu on reset. */
10246 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10247 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10248 !is_protmode(vcpu))
10249 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10254 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10256 int pending_vec, max_bits;
10257 int mmu_reset_needed = 0;
10258 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10263 if (mmu_reset_needed)
10264 kvm_mmu_reset_context(vcpu);
10266 max_bits = KVM_NR_INTERRUPTS;
10267 pending_vec = find_first_bit(
10268 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10270 if (pending_vec < max_bits) {
10271 kvm_queue_interrupt(vcpu, pending_vec, false);
10272 pr_debug("Set back pending irq %d\n", pending_vec);
10273 kvm_make_request(KVM_REQ_EVENT, vcpu);
10278 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10280 int mmu_reset_needed = 0;
10281 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10282 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10283 !(sregs2->efer & EFER_LMA);
10286 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10289 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10292 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10293 &mmu_reset_needed, !valid_pdptrs);
10297 if (valid_pdptrs) {
10298 for (i = 0; i < 4 ; i++)
10299 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10301 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10302 mmu_reset_needed = 1;
10303 vcpu->arch.pdptrs_from_userspace = true;
10305 if (mmu_reset_needed)
10306 kvm_mmu_reset_context(vcpu);
10310 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10311 struct kvm_sregs *sregs)
10316 ret = __set_sregs(vcpu, sregs);
10321 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10322 struct kvm_guest_debug *dbg)
10324 unsigned long rflags;
10327 if (vcpu->arch.guest_state_protected)
10332 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10334 if (vcpu->arch.exception.pending)
10336 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10337 kvm_queue_exception(vcpu, DB_VECTOR);
10339 kvm_queue_exception(vcpu, BP_VECTOR);
10343 * Read rflags as long as potentially injected trace flags are still
10346 rflags = kvm_get_rflags(vcpu);
10348 vcpu->guest_debug = dbg->control;
10349 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10350 vcpu->guest_debug = 0;
10352 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10353 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10354 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10355 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10357 for (i = 0; i < KVM_NR_DB_REGS; i++)
10358 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10360 kvm_update_dr7(vcpu);
10362 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10363 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10366 * Trigger an rflags update that will inject or remove the trace
10369 kvm_set_rflags(vcpu, rflags);
10371 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10381 * Translate a guest virtual address to a guest physical address.
10383 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10384 struct kvm_translation *tr)
10386 unsigned long vaddr = tr->linear_address;
10392 idx = srcu_read_lock(&vcpu->kvm->srcu);
10393 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10394 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10395 tr->physical_address = gpa;
10396 tr->valid = gpa != UNMAPPED_GVA;
10404 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10406 struct fxregs_state *fxsave;
10408 if (!vcpu->arch.guest_fpu)
10413 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10414 memcpy(fpu->fpr, fxsave->st_space, 128);
10415 fpu->fcw = fxsave->cwd;
10416 fpu->fsw = fxsave->swd;
10417 fpu->ftwx = fxsave->twd;
10418 fpu->last_opcode = fxsave->fop;
10419 fpu->last_ip = fxsave->rip;
10420 fpu->last_dp = fxsave->rdp;
10421 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10427 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10429 struct fxregs_state *fxsave;
10431 if (!vcpu->arch.guest_fpu)
10436 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10438 memcpy(fxsave->st_space, fpu->fpr, 128);
10439 fxsave->cwd = fpu->fcw;
10440 fxsave->swd = fpu->fsw;
10441 fxsave->twd = fpu->ftwx;
10442 fxsave->fop = fpu->last_opcode;
10443 fxsave->rip = fpu->last_ip;
10444 fxsave->rdp = fpu->last_dp;
10445 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10451 static void store_regs(struct kvm_vcpu *vcpu)
10453 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10455 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10456 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10458 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10459 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10461 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10462 kvm_vcpu_ioctl_x86_get_vcpu_events(
10463 vcpu, &vcpu->run->s.regs.events);
10466 static int sync_regs(struct kvm_vcpu *vcpu)
10468 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10471 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10472 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10473 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10475 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10476 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10478 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10480 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10481 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10482 vcpu, &vcpu->run->s.regs.events))
10484 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10490 static void fx_init(struct kvm_vcpu *vcpu)
10492 if (!vcpu->arch.guest_fpu)
10495 fpstate_init(&vcpu->arch.guest_fpu->state);
10496 if (boot_cpu_has(X86_FEATURE_XSAVES))
10497 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10498 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10501 * Ensure guest xcr0 is valid for loading
10503 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10505 vcpu->arch.cr0 |= X86_CR0_ET;
10508 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10510 if (vcpu->arch.guest_fpu) {
10511 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10512 vcpu->arch.guest_fpu = NULL;
10515 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10517 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10519 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10520 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10521 "guest TSC will not be reliable\n");
10526 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10531 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10532 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10534 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10536 r = kvm_mmu_create(vcpu);
10540 if (irqchip_in_kernel(vcpu->kvm)) {
10541 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10543 goto fail_mmu_destroy;
10544 if (kvm_apicv_activated(vcpu->kvm))
10545 vcpu->arch.apicv_active = true;
10547 static_branch_inc(&kvm_has_noapic_vcpu);
10551 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10553 goto fail_free_lapic;
10554 vcpu->arch.pio_data = page_address(page);
10556 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10557 GFP_KERNEL_ACCOUNT);
10558 if (!vcpu->arch.mce_banks)
10559 goto fail_free_pio_data;
10560 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10562 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10563 GFP_KERNEL_ACCOUNT))
10564 goto fail_free_mce_banks;
10566 if (!alloc_emulate_ctxt(vcpu))
10567 goto free_wbinvd_dirty_mask;
10569 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10570 GFP_KERNEL_ACCOUNT);
10571 if (!vcpu->arch.user_fpu) {
10572 pr_err("kvm: failed to allocate userspace's fpu\n");
10573 goto free_emulate_ctxt;
10576 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10577 GFP_KERNEL_ACCOUNT);
10578 if (!vcpu->arch.guest_fpu) {
10579 pr_err("kvm: failed to allocate vcpu's fpu\n");
10580 goto free_user_fpu;
10584 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10585 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10587 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10589 kvm_async_pf_hash_reset(vcpu);
10590 kvm_pmu_init(vcpu);
10592 vcpu->arch.pending_external_vector = -1;
10593 vcpu->arch.preempted_in_kernel = false;
10595 #if IS_ENABLED(CONFIG_HYPERV)
10596 vcpu->arch.hv_root_tdp = INVALID_PAGE;
10599 r = static_call(kvm_x86_vcpu_create)(vcpu);
10601 goto free_guest_fpu;
10603 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10604 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10605 kvm_vcpu_mtrr_init(vcpu);
10607 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10608 kvm_vcpu_reset(vcpu, false);
10609 kvm_init_mmu(vcpu, false);
10614 kvm_free_guest_fpu(vcpu);
10616 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10618 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10619 free_wbinvd_dirty_mask:
10620 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10621 fail_free_mce_banks:
10622 kfree(vcpu->arch.mce_banks);
10623 fail_free_pio_data:
10624 free_page((unsigned long)vcpu->arch.pio_data);
10626 kvm_free_lapic(vcpu);
10628 kvm_mmu_destroy(vcpu);
10632 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10634 struct kvm *kvm = vcpu->kvm;
10636 if (mutex_lock_killable(&vcpu->mutex))
10639 kvm_synchronize_tsc(vcpu, 0);
10642 /* poll control enabled by default */
10643 vcpu->arch.msr_kvm_poll_control = 1;
10645 mutex_unlock(&vcpu->mutex);
10647 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10648 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10649 KVMCLOCK_SYNC_PERIOD);
10652 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10654 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10657 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10659 kvmclock_reset(vcpu);
10661 static_call(kvm_x86_vcpu_free)(vcpu);
10663 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10664 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10665 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10666 kvm_free_guest_fpu(vcpu);
10668 kvm_hv_vcpu_uninit(vcpu);
10669 kvm_pmu_destroy(vcpu);
10670 kfree(vcpu->arch.mce_banks);
10671 kvm_free_lapic(vcpu);
10672 idx = srcu_read_lock(&vcpu->kvm->srcu);
10673 kvm_mmu_destroy(vcpu);
10674 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10675 free_page((unsigned long)vcpu->arch.pio_data);
10676 kvfree(vcpu->arch.cpuid_entries);
10677 if (!lapic_in_kernel(vcpu))
10678 static_branch_dec(&kvm_has_noapic_vcpu);
10681 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10683 kvm_lapic_reset(vcpu, init_event);
10685 vcpu->arch.hflags = 0;
10687 vcpu->arch.smi_pending = 0;
10688 vcpu->arch.smi_count = 0;
10689 atomic_set(&vcpu->arch.nmi_queued, 0);
10690 vcpu->arch.nmi_pending = 0;
10691 vcpu->arch.nmi_injected = false;
10692 kvm_clear_interrupt_queue(vcpu);
10693 kvm_clear_exception_queue(vcpu);
10695 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10696 kvm_update_dr0123(vcpu);
10697 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10698 vcpu->arch.dr7 = DR7_FIXED_1;
10699 kvm_update_dr7(vcpu);
10701 vcpu->arch.cr2 = 0;
10703 kvm_make_request(KVM_REQ_EVENT, vcpu);
10704 vcpu->arch.apf.msr_en_val = 0;
10705 vcpu->arch.apf.msr_int_val = 0;
10706 vcpu->arch.st.msr_val = 0;
10708 kvmclock_reset(vcpu);
10710 kvm_clear_async_pf_completion_queue(vcpu);
10711 kvm_async_pf_hash_reset(vcpu);
10712 vcpu->arch.apf.halted = false;
10714 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10715 void *mpx_state_buffer;
10718 * To avoid have the INIT path from kvm_apic_has_events() that be
10719 * called with loaded FPU and does not let userspace fix the state.
10722 kvm_put_guest_fpu(vcpu);
10723 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10725 if (mpx_state_buffer)
10726 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10727 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10729 if (mpx_state_buffer)
10730 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10732 kvm_load_guest_fpu(vcpu);
10736 kvm_pmu_reset(vcpu);
10737 vcpu->arch.smbase = 0x30000;
10739 vcpu->arch.msr_misc_features_enables = 0;
10741 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10744 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10745 vcpu->arch.regs_avail = ~0;
10746 vcpu->arch.regs_dirty = ~0;
10748 vcpu->arch.ia32_xss = 0;
10750 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10753 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10755 struct kvm_segment cs;
10757 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10758 cs.selector = vector << 8;
10759 cs.base = vector << 12;
10760 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10761 kvm_rip_write(vcpu, 0);
10763 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10765 int kvm_arch_hardware_enable(void)
10768 struct kvm_vcpu *vcpu;
10773 bool stable, backwards_tsc = false;
10775 kvm_user_return_msr_cpu_online();
10776 ret = static_call(kvm_x86_hardware_enable)();
10780 local_tsc = rdtsc();
10781 stable = !kvm_check_tsc_unstable();
10782 list_for_each_entry(kvm, &vm_list, vm_list) {
10783 kvm_for_each_vcpu(i, vcpu, kvm) {
10784 if (!stable && vcpu->cpu == smp_processor_id())
10785 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10786 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10787 backwards_tsc = true;
10788 if (vcpu->arch.last_host_tsc > max_tsc)
10789 max_tsc = vcpu->arch.last_host_tsc;
10795 * Sometimes, even reliable TSCs go backwards. This happens on
10796 * platforms that reset TSC during suspend or hibernate actions, but
10797 * maintain synchronization. We must compensate. Fortunately, we can
10798 * detect that condition here, which happens early in CPU bringup,
10799 * before any KVM threads can be running. Unfortunately, we can't
10800 * bring the TSCs fully up to date with real time, as we aren't yet far
10801 * enough into CPU bringup that we know how much real time has actually
10802 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10803 * variables that haven't been updated yet.
10805 * So we simply find the maximum observed TSC above, then record the
10806 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10807 * the adjustment will be applied. Note that we accumulate
10808 * adjustments, in case multiple suspend cycles happen before some VCPU
10809 * gets a chance to run again. In the event that no KVM threads get a
10810 * chance to run, we will miss the entire elapsed period, as we'll have
10811 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10812 * loose cycle time. This isn't too big a deal, since the loss will be
10813 * uniform across all VCPUs (not to mention the scenario is extremely
10814 * unlikely). It is possible that a second hibernate recovery happens
10815 * much faster than a first, causing the observed TSC here to be
10816 * smaller; this would require additional padding adjustment, which is
10817 * why we set last_host_tsc to the local tsc observed here.
10819 * N.B. - this code below runs only on platforms with reliable TSC,
10820 * as that is the only way backwards_tsc is set above. Also note
10821 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10822 * have the same delta_cyc adjustment applied if backwards_tsc
10823 * is detected. Note further, this adjustment is only done once,
10824 * as we reset last_host_tsc on all VCPUs to stop this from being
10825 * called multiple times (one for each physical CPU bringup).
10827 * Platforms with unreliable TSCs don't have to deal with this, they
10828 * will be compensated by the logic in vcpu_load, which sets the TSC to
10829 * catchup mode. This will catchup all VCPUs to real time, but cannot
10830 * guarantee that they stay in perfect synchronization.
10832 if (backwards_tsc) {
10833 u64 delta_cyc = max_tsc - local_tsc;
10834 list_for_each_entry(kvm, &vm_list, vm_list) {
10835 kvm->arch.backwards_tsc_observed = true;
10836 kvm_for_each_vcpu(i, vcpu, kvm) {
10837 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10838 vcpu->arch.last_host_tsc = local_tsc;
10839 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10843 * We have to disable TSC offset matching.. if you were
10844 * booting a VM while issuing an S4 host suspend....
10845 * you may have some problem. Solving this issue is
10846 * left as an exercise to the reader.
10848 kvm->arch.last_tsc_nsec = 0;
10849 kvm->arch.last_tsc_write = 0;
10856 void kvm_arch_hardware_disable(void)
10858 static_call(kvm_x86_hardware_disable)();
10859 drop_user_return_notifiers();
10862 int kvm_arch_hardware_setup(void *opaque)
10864 struct kvm_x86_init_ops *ops = opaque;
10867 rdmsrl_safe(MSR_EFER, &host_efer);
10869 if (boot_cpu_has(X86_FEATURE_XSAVES))
10870 rdmsrl(MSR_IA32_XSS, host_xss);
10872 r = ops->hardware_setup();
10876 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10877 kvm_ops_static_call_update();
10879 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10882 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10883 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10884 #undef __kvm_cpu_cap_has
10886 if (kvm_has_tsc_control) {
10888 * Make sure the user can only configure tsc_khz values that
10889 * fit into a signed integer.
10890 * A min value is not calculated because it will always
10891 * be 1 on all machines.
10893 u64 max = min(0x7fffffffULL,
10894 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10895 kvm_max_guest_tsc_khz = max;
10897 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10900 kvm_init_msr_list();
10904 void kvm_arch_hardware_unsetup(void)
10906 static_call(kvm_x86_hardware_unsetup)();
10909 int kvm_arch_check_processor_compat(void *opaque)
10911 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10912 struct kvm_x86_init_ops *ops = opaque;
10914 WARN_ON(!irqs_disabled());
10916 if (__cr4_reserved_bits(cpu_has, c) !=
10917 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10920 return ops->check_processor_compatibility();
10923 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10925 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10927 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10929 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10931 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10934 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10935 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10937 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10939 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10941 vcpu->arch.l1tf_flush_l1d = true;
10942 if (pmu->version && unlikely(pmu->event_count)) {
10943 pmu->need_cleanup = true;
10944 kvm_make_request(KVM_REQ_PMU, vcpu);
10946 static_call(kvm_x86_sched_in)(vcpu, cpu);
10949 void kvm_arch_free_vm(struct kvm *kvm)
10951 kfree(to_kvm_hv(kvm)->hv_pa_pg);
10956 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10961 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10962 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10963 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10964 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10965 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10966 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10968 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10969 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10970 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10971 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10972 &kvm->arch.irq_sources_bitmap);
10974 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10975 mutex_init(&kvm->arch.apic_map_lock);
10976 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10978 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10979 pvclock_update_vm_gtod_copy(kvm);
10981 kvm->arch.guest_can_read_msr_platform_info = true;
10983 #if IS_ENABLED(CONFIG_HYPERV)
10984 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
10985 kvm->arch.hv_root_tdp = INVALID_PAGE;
10988 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10989 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10991 kvm_apicv_init(kvm);
10992 kvm_hv_init_vm(kvm);
10993 kvm_page_track_init(kvm);
10994 kvm_mmu_init_vm(kvm);
10996 return static_call(kvm_x86_vm_init)(kvm);
10999 int kvm_arch_post_init_vm(struct kvm *kvm)
11001 return kvm_mmu_post_init_vm(kvm);
11004 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11007 kvm_mmu_unload(vcpu);
11011 static void kvm_free_vcpus(struct kvm *kvm)
11014 struct kvm_vcpu *vcpu;
11017 * Unpin any mmu pages first.
11019 kvm_for_each_vcpu(i, vcpu, kvm) {
11020 kvm_clear_async_pf_completion_queue(vcpu);
11021 kvm_unload_vcpu_mmu(vcpu);
11023 kvm_for_each_vcpu(i, vcpu, kvm)
11024 kvm_vcpu_destroy(vcpu);
11026 mutex_lock(&kvm->lock);
11027 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11028 kvm->vcpus[i] = NULL;
11030 atomic_set(&kvm->online_vcpus, 0);
11031 mutex_unlock(&kvm->lock);
11034 void kvm_arch_sync_events(struct kvm *kvm)
11036 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11037 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11041 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
11044 * __x86_set_memory_region: Setup KVM internal memory slot
11046 * @kvm: the kvm pointer to the VM.
11047 * @id: the slot ID to setup.
11048 * @gpa: the GPA to install the slot (unused when @size == 0).
11049 * @size: the size of the slot. Set to zero to uninstall a slot.
11051 * This function helps to setup a KVM internal memory slot. Specify
11052 * @size > 0 to install a new slot, while @size == 0 to uninstall a
11053 * slot. The return code can be one of the following:
11055 * HVA: on success (uninstall will return a bogus HVA)
11058 * The caller should always use IS_ERR() to check the return value
11059 * before use. Note, the KVM internal memory slots are guaranteed to
11060 * remain valid and unchanged until the VM is destroyed, i.e., the
11061 * GPA->HVA translation will not change. However, the HVA is a user
11062 * address, i.e. its accessibility is not guaranteed, and must be
11063 * accessed via __copy_{to,from}_user().
11065 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11069 unsigned long hva, old_npages;
11070 struct kvm_memslots *slots = kvm_memslots(kvm);
11071 struct kvm_memory_slot *slot;
11073 /* Called with kvm->slots_lock held. */
11074 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11075 return ERR_PTR_USR(-EINVAL);
11077 slot = id_to_memslot(slots, id);
11079 if (slot && slot->npages)
11080 return ERR_PTR_USR(-EEXIST);
11083 * MAP_SHARED to prevent internal slot pages from being moved
11086 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11087 MAP_SHARED | MAP_ANONYMOUS, 0);
11088 if (IS_ERR((void *)hva))
11089 return (void __user *)hva;
11091 if (!slot || !slot->npages)
11094 old_npages = slot->npages;
11095 hva = slot->userspace_addr;
11098 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11099 struct kvm_userspace_memory_region m;
11101 m.slot = id | (i << 16);
11103 m.guest_phys_addr = gpa;
11104 m.userspace_addr = hva;
11105 m.memory_size = size;
11106 r = __kvm_set_memory_region(kvm, &m);
11108 return ERR_PTR_USR(r);
11112 vm_munmap(hva, old_npages * PAGE_SIZE);
11114 return (void __user *)hva;
11116 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11118 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11120 kvm_mmu_pre_destroy_vm(kvm);
11123 void kvm_arch_destroy_vm(struct kvm *kvm)
11125 if (current->mm == kvm->mm) {
11127 * Free memory regions allocated on behalf of userspace,
11128 * unless the the memory map has changed due to process exit
11131 mutex_lock(&kvm->slots_lock);
11132 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11134 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11136 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11137 mutex_unlock(&kvm->slots_lock);
11139 static_call_cond(kvm_x86_vm_destroy)(kvm);
11140 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11141 kvm_pic_destroy(kvm);
11142 kvm_ioapic_destroy(kvm);
11143 kvm_free_vcpus(kvm);
11144 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11145 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11146 kvm_mmu_uninit_vm(kvm);
11147 kvm_page_track_cleanup(kvm);
11148 kvm_xen_destroy_vm(kvm);
11149 kvm_hv_destroy_vm(kvm);
11152 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11156 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11157 kvfree(slot->arch.rmap[i]);
11158 slot->arch.rmap[i] = NULL;
11162 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11166 memslot_rmap_free(slot);
11168 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11169 kvfree(slot->arch.lpage_info[i - 1]);
11170 slot->arch.lpage_info[i - 1] = NULL;
11173 kvm_page_track_free_memslot(slot);
11176 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
11177 unsigned long npages)
11179 const int sz = sizeof(*slot->arch.rmap[0]);
11182 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11184 int lpages = gfn_to_index(slot->base_gfn + npages - 1,
11185 slot->base_gfn, level) + 1;
11187 WARN_ON(slot->arch.rmap[i]);
11189 slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11190 if (!slot->arch.rmap[i]) {
11191 memslot_rmap_free(slot);
11199 int alloc_all_memslots_rmaps(struct kvm *kvm)
11201 struct kvm_memslots *slots;
11202 struct kvm_memory_slot *slot;
11206 * Check if memslots alreday have rmaps early before acquiring
11207 * the slots_arch_lock below.
11209 if (kvm_memslots_have_rmaps(kvm))
11212 mutex_lock(&kvm->slots_arch_lock);
11215 * Read memslots_have_rmaps again, under the slots arch lock,
11216 * before allocating the rmaps
11218 if (kvm_memslots_have_rmaps(kvm)) {
11219 mutex_unlock(&kvm->slots_arch_lock);
11223 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11224 slots = __kvm_memslots(kvm, i);
11225 kvm_for_each_memslot(slot, slots) {
11226 r = memslot_rmap_alloc(slot, slot->npages);
11228 mutex_unlock(&kvm->slots_arch_lock);
11235 * Ensure that memslots_have_rmaps becomes true strictly after
11236 * all the rmap pointers are set.
11238 smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11239 mutex_unlock(&kvm->slots_arch_lock);
11243 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11244 struct kvm_memory_slot *slot,
11245 unsigned long npages)
11250 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
11251 * old arrays will be freed by __kvm_set_memory_region() if installing
11252 * the new memslot is successful.
11254 memset(&slot->arch, 0, sizeof(slot->arch));
11256 if (kvm_memslots_have_rmaps(kvm)) {
11257 r = memslot_rmap_alloc(slot, npages);
11262 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11263 struct kvm_lpage_info *linfo;
11264 unsigned long ugfn;
11268 lpages = gfn_to_index(slot->base_gfn + npages - 1,
11269 slot->base_gfn, level) + 1;
11271 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11275 slot->arch.lpage_info[i - 1] = linfo;
11277 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11278 linfo[0].disallow_lpage = 1;
11279 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11280 linfo[lpages - 1].disallow_lpage = 1;
11281 ugfn = slot->userspace_addr >> PAGE_SHIFT;
11283 * If the gfn and userspace address are not aligned wrt each
11284 * other, disable large page support for this slot.
11286 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11289 for (j = 0; j < lpages; ++j)
11290 linfo[j].disallow_lpage = 1;
11294 if (kvm_page_track_create_memslot(slot, npages))
11300 memslot_rmap_free(slot);
11302 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11303 kvfree(slot->arch.lpage_info[i - 1]);
11304 slot->arch.lpage_info[i - 1] = NULL;
11309 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11311 struct kvm_vcpu *vcpu;
11315 * memslots->generation has been incremented.
11316 * mmio generation may have reached its maximum value.
11318 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11320 /* Force re-initialization of steal_time cache */
11321 kvm_for_each_vcpu(i, vcpu, kvm)
11322 kvm_vcpu_kick(vcpu);
11325 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11326 struct kvm_memory_slot *memslot,
11327 const struct kvm_userspace_memory_region *mem,
11328 enum kvm_mr_change change)
11330 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11331 return kvm_alloc_memslot_metadata(kvm, memslot,
11332 mem->memory_size >> PAGE_SHIFT);
11337 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11339 struct kvm_arch *ka = &kvm->arch;
11341 if (!kvm_x86_ops.cpu_dirty_log_size)
11344 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11345 (!enable && --ka->cpu_dirty_logging_count == 0))
11346 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11348 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11351 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11352 struct kvm_memory_slot *old,
11353 struct kvm_memory_slot *new,
11354 enum kvm_mr_change change)
11356 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11359 * Update CPU dirty logging if dirty logging is being toggled. This
11360 * applies to all operations.
11362 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11363 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11366 * Nothing more to do for RO slots (which can't be dirtied and can't be
11367 * made writable) or CREATE/MOVE/DELETE of a slot.
11369 * For a memslot with dirty logging disabled:
11370 * CREATE: No dirty mappings will already exist.
11371 * MOVE/DELETE: The old mappings will already have been cleaned up by
11372 * kvm_arch_flush_shadow_memslot()
11374 * For a memslot with dirty logging enabled:
11375 * CREATE: No shadow pages exist, thus nothing to write-protect
11376 * and no dirty bits to clear.
11377 * MOVE/DELETE: The old mappings will already have been cleaned up by
11378 * kvm_arch_flush_shadow_memslot().
11380 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11384 * READONLY and non-flags changes were filtered out above, and the only
11385 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11386 * logging isn't being toggled on or off.
11388 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11391 if (!log_dirty_pages) {
11393 * Dirty logging tracks sptes in 4k granularity, meaning that
11394 * large sptes have to be split. If live migration succeeds,
11395 * the guest in the source machine will be destroyed and large
11396 * sptes will be created in the destination. However, if the
11397 * guest continues to run in the source machine (for example if
11398 * live migration fails), small sptes will remain around and
11399 * cause bad performance.
11401 * Scan sptes if dirty logging has been stopped, dropping those
11402 * which can be collapsed into a single large-page spte. Later
11403 * page faults will create the large-page sptes.
11405 kvm_mmu_zap_collapsible_sptes(kvm, new);
11408 * Initially-all-set does not require write protecting any page,
11409 * because they're all assumed to be dirty.
11411 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11414 if (kvm_x86_ops.cpu_dirty_log_size) {
11415 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11416 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11418 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11423 void kvm_arch_commit_memory_region(struct kvm *kvm,
11424 const struct kvm_userspace_memory_region *mem,
11425 struct kvm_memory_slot *old,
11426 const struct kvm_memory_slot *new,
11427 enum kvm_mr_change change)
11429 if (!kvm->arch.n_requested_mmu_pages)
11430 kvm_mmu_change_mmu_pages(kvm,
11431 kvm_mmu_calculate_default_mmu_pages(kvm));
11434 * FIXME: const-ify all uses of struct kvm_memory_slot.
11436 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11438 /* Free the arrays associated with the old memslot. */
11439 if (change == KVM_MR_MOVE)
11440 kvm_arch_free_memslot(kvm, old);
11443 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11445 kvm_mmu_zap_all(kvm);
11448 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11449 struct kvm_memory_slot *slot)
11451 kvm_page_track_flush_slot(kvm, slot);
11454 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11456 return (is_guest_mode(vcpu) &&
11457 kvm_x86_ops.guest_apic_has_interrupt &&
11458 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11461 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11463 if (!list_empty_careful(&vcpu->async_pf.done))
11466 if (kvm_apic_has_events(vcpu))
11469 if (vcpu->arch.pv.pv_unhalted)
11472 if (vcpu->arch.exception.pending)
11475 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11476 (vcpu->arch.nmi_pending &&
11477 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11480 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11481 (vcpu->arch.smi_pending &&
11482 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11485 if (kvm_arch_interrupt_allowed(vcpu) &&
11486 (kvm_cpu_has_interrupt(vcpu) ||
11487 kvm_guest_apic_has_interrupt(vcpu)))
11490 if (kvm_hv_has_stimer_pending(vcpu))
11493 if (is_guest_mode(vcpu) &&
11494 kvm_x86_ops.nested_ops->hv_timer_pending &&
11495 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11501 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11503 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11506 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11508 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11514 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11516 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11519 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11520 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11521 kvm_test_request(KVM_REQ_EVENT, vcpu))
11524 return kvm_arch_dy_has_pending_interrupt(vcpu);
11527 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11529 if (vcpu->arch.guest_state_protected)
11532 return vcpu->arch.preempted_in_kernel;
11535 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11537 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11540 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11542 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11545 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11547 /* Can't read the RIP when guest state is protected, just return 0 */
11548 if (vcpu->arch.guest_state_protected)
11551 if (is_64_bit_mode(vcpu))
11552 return kvm_rip_read(vcpu);
11553 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11554 kvm_rip_read(vcpu));
11556 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11558 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11560 return kvm_get_linear_rip(vcpu) == linear_rip;
11562 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11564 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11566 unsigned long rflags;
11568 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11569 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11570 rflags &= ~X86_EFLAGS_TF;
11573 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11575 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11577 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11578 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11579 rflags |= X86_EFLAGS_TF;
11580 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11583 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11585 __kvm_set_rflags(vcpu, rflags);
11586 kvm_make_request(KVM_REQ_EVENT, vcpu);
11588 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11590 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11594 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11598 r = kvm_mmu_reload(vcpu);
11602 if (!vcpu->arch.mmu->direct_map &&
11603 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11606 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11609 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11611 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11613 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11616 static inline u32 kvm_async_pf_next_probe(u32 key)
11618 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11621 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11623 u32 key = kvm_async_pf_hash_fn(gfn);
11625 while (vcpu->arch.apf.gfns[key] != ~0)
11626 key = kvm_async_pf_next_probe(key);
11628 vcpu->arch.apf.gfns[key] = gfn;
11631 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11634 u32 key = kvm_async_pf_hash_fn(gfn);
11636 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11637 (vcpu->arch.apf.gfns[key] != gfn &&
11638 vcpu->arch.apf.gfns[key] != ~0); i++)
11639 key = kvm_async_pf_next_probe(key);
11644 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11646 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11649 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11653 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11655 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11659 vcpu->arch.apf.gfns[i] = ~0;
11661 j = kvm_async_pf_next_probe(j);
11662 if (vcpu->arch.apf.gfns[j] == ~0)
11664 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11666 * k lies cyclically in ]i,j]
11668 * |....j i.k.| or |.k..j i...|
11670 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11671 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11676 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11678 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11680 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11684 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11686 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11688 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11689 &token, offset, sizeof(token));
11692 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11694 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11697 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11698 &val, offset, sizeof(val)))
11704 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11706 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11709 if (!kvm_pv_async_pf_enabled(vcpu) ||
11710 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11716 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11718 if (unlikely(!lapic_in_kernel(vcpu) ||
11719 kvm_event_needs_reinjection(vcpu) ||
11720 vcpu->arch.exception.pending))
11723 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11727 * If interrupts are off we cannot even use an artificial
11730 return kvm_arch_interrupt_allowed(vcpu);
11733 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11734 struct kvm_async_pf *work)
11736 struct x86_exception fault;
11738 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11739 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11741 if (kvm_can_deliver_async_pf(vcpu) &&
11742 !apf_put_user_notpresent(vcpu)) {
11743 fault.vector = PF_VECTOR;
11744 fault.error_code_valid = true;
11745 fault.error_code = 0;
11746 fault.nested_page_fault = false;
11747 fault.address = work->arch.token;
11748 fault.async_page_fault = true;
11749 kvm_inject_page_fault(vcpu, &fault);
11753 * It is not possible to deliver a paravirtualized asynchronous
11754 * page fault, but putting the guest in an artificial halt state
11755 * can be beneficial nevertheless: if an interrupt arrives, we
11756 * can deliver it timely and perhaps the guest will schedule
11757 * another process. When the instruction that triggered a page
11758 * fault is retried, hopefully the page will be ready in the host.
11760 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11765 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11766 struct kvm_async_pf *work)
11768 struct kvm_lapic_irq irq = {
11769 .delivery_mode = APIC_DM_FIXED,
11770 .vector = vcpu->arch.apf.vec
11773 if (work->wakeup_all)
11774 work->arch.token = ~0; /* broadcast wakeup */
11776 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11777 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11779 if ((work->wakeup_all || work->notpresent_injected) &&
11780 kvm_pv_async_pf_enabled(vcpu) &&
11781 !apf_put_user_ready(vcpu, work->arch.token)) {
11782 vcpu->arch.apf.pageready_pending = true;
11783 kvm_apic_set_irq(vcpu, &irq, NULL);
11786 vcpu->arch.apf.halted = false;
11787 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11790 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11792 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11793 if (!vcpu->arch.apf.pageready_pending)
11794 kvm_vcpu_kick(vcpu);
11797 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11799 if (!kvm_pv_async_pf_enabled(vcpu))
11802 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11805 void kvm_arch_start_assignment(struct kvm *kvm)
11807 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
11808 static_call_cond(kvm_x86_start_assignment)(kvm);
11810 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11812 void kvm_arch_end_assignment(struct kvm *kvm)
11814 atomic_dec(&kvm->arch.assigned_device_count);
11816 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11818 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11820 return atomic_read(&kvm->arch.assigned_device_count);
11822 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11824 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11826 atomic_inc(&kvm->arch.noncoherent_dma_count);
11828 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11830 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11832 atomic_dec(&kvm->arch.noncoherent_dma_count);
11834 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11836 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11838 return atomic_read(&kvm->arch.noncoherent_dma_count);
11840 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11842 bool kvm_arch_has_irq_bypass(void)
11847 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11848 struct irq_bypass_producer *prod)
11850 struct kvm_kernel_irqfd *irqfd =
11851 container_of(cons, struct kvm_kernel_irqfd, consumer);
11854 irqfd->producer = prod;
11855 kvm_arch_start_assignment(irqfd->kvm);
11856 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11857 prod->irq, irqfd->gsi, 1);
11860 kvm_arch_end_assignment(irqfd->kvm);
11865 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11866 struct irq_bypass_producer *prod)
11869 struct kvm_kernel_irqfd *irqfd =
11870 container_of(cons, struct kvm_kernel_irqfd, consumer);
11872 WARN_ON(irqfd->producer != prod);
11873 irqfd->producer = NULL;
11876 * When producer of consumer is unregistered, we change back to
11877 * remapped mode, so we can re-use the current implementation
11878 * when the irq is masked/disabled or the consumer side (KVM
11879 * int this case doesn't want to receive the interrupts.
11881 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11883 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11884 " fails: %d\n", irqfd->consumer.token, ret);
11886 kvm_arch_end_assignment(irqfd->kvm);
11889 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11890 uint32_t guest_irq, bool set)
11892 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11895 bool kvm_vector_hashing_enabled(void)
11897 return vector_hashing;
11900 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11902 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11904 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11907 int kvm_spec_ctrl_test_value(u64 value)
11910 * test that setting IA32_SPEC_CTRL to given value
11911 * is allowed by the host processor
11915 unsigned long flags;
11918 local_irq_save(flags);
11920 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11922 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11925 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11927 local_irq_restore(flags);
11931 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11933 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11935 struct x86_exception fault;
11936 u32 access = error_code &
11937 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11939 if (!(error_code & PFERR_PRESENT_MASK) ||
11940 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11942 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11943 * tables probably do not match the TLB. Just proceed
11944 * with the error code that the processor gave.
11946 fault.vector = PF_VECTOR;
11947 fault.error_code_valid = true;
11948 fault.error_code = error_code;
11949 fault.nested_page_fault = false;
11950 fault.address = gva;
11952 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11954 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11957 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11958 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11959 * indicates whether exit to userspace is needed.
11961 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11962 struct x86_exception *e)
11964 if (r == X86EMUL_PROPAGATE_FAULT) {
11965 kvm_inject_emulated_page_fault(vcpu, e);
11970 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11971 * while handling a VMX instruction KVM could've handled the request
11972 * correctly by exiting to userspace and performing I/O but there
11973 * doesn't seem to be a real use-case behind such requests, just return
11974 * KVM_EXIT_INTERNAL_ERROR for now.
11976 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11977 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11978 vcpu->run->internal.ndata = 0;
11982 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11984 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11987 struct x86_exception e;
11994 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11995 if (r != X86EMUL_CONTINUE)
11996 return kvm_handle_memory_failure(vcpu, r, &e);
11998 if (operand.pcid >> 12 != 0) {
11999 kvm_inject_gp(vcpu, 0);
12003 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12006 case INVPCID_TYPE_INDIV_ADDR:
12007 if ((!pcid_enabled && (operand.pcid != 0)) ||
12008 is_noncanonical_address(operand.gla, vcpu)) {
12009 kvm_inject_gp(vcpu, 0);
12012 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12013 return kvm_skip_emulated_instruction(vcpu);
12015 case INVPCID_TYPE_SINGLE_CTXT:
12016 if (!pcid_enabled && (operand.pcid != 0)) {
12017 kvm_inject_gp(vcpu, 0);
12021 kvm_invalidate_pcid(vcpu, operand.pcid);
12022 return kvm_skip_emulated_instruction(vcpu);
12024 case INVPCID_TYPE_ALL_NON_GLOBAL:
12026 * Currently, KVM doesn't mark global entries in the shadow
12027 * page tables, so a non-global flush just degenerates to a
12028 * global flush. If needed, we could optimize this later by
12029 * keeping track of global entries in shadow page tables.
12033 case INVPCID_TYPE_ALL_INCL_GLOBAL:
12034 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
12035 return kvm_skip_emulated_instruction(vcpu);
12038 BUG(); /* We have already checked above that type <= 3 */
12041 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12043 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12045 struct kvm_run *run = vcpu->run;
12046 struct kvm_mmio_fragment *frag;
12049 BUG_ON(!vcpu->mmio_needed);
12051 /* Complete previous fragment */
12052 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12053 len = min(8u, frag->len);
12054 if (!vcpu->mmio_is_write)
12055 memcpy(frag->data, run->mmio.data, len);
12057 if (frag->len <= 8) {
12058 /* Switch to the next fragment. */
12060 vcpu->mmio_cur_fragment++;
12062 /* Go forward to the next mmio piece. */
12068 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12069 vcpu->mmio_needed = 0;
12071 // VMG change, at this point, we're always done
12072 // RIP has already been advanced
12076 // More MMIO is needed
12077 run->mmio.phys_addr = frag->gpa;
12078 run->mmio.len = min(8u, frag->len);
12079 run->mmio.is_write = vcpu->mmio_is_write;
12080 if (run->mmio.is_write)
12081 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12082 run->exit_reason = KVM_EXIT_MMIO;
12084 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12089 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12093 struct kvm_mmio_fragment *frag;
12098 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12099 if (handled == bytes)
12106 /*TODO: Check if need to increment number of frags */
12107 frag = vcpu->mmio_fragments;
12108 vcpu->mmio_nr_fragments = 1;
12113 vcpu->mmio_needed = 1;
12114 vcpu->mmio_cur_fragment = 0;
12116 vcpu->run->mmio.phys_addr = gpa;
12117 vcpu->run->mmio.len = min(8u, frag->len);
12118 vcpu->run->mmio.is_write = 1;
12119 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12120 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12122 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12126 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12128 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12132 struct kvm_mmio_fragment *frag;
12137 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12138 if (handled == bytes)
12145 /*TODO: Check if need to increment number of frags */
12146 frag = vcpu->mmio_fragments;
12147 vcpu->mmio_nr_fragments = 1;
12152 vcpu->mmio_needed = 1;
12153 vcpu->mmio_cur_fragment = 0;
12155 vcpu->run->mmio.phys_addr = gpa;
12156 vcpu->run->mmio.len = min(8u, frag->len);
12157 vcpu->run->mmio.is_write = 0;
12158 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12160 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12164 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12166 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12168 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
12169 vcpu->arch.pio.count * vcpu->arch.pio.size);
12170 vcpu->arch.pio.count = 0;
12175 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12176 unsigned int port, void *data, unsigned int count)
12180 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
12185 vcpu->arch.pio.count = 0;
12190 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12191 unsigned int port, void *data, unsigned int count)
12195 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
12198 vcpu->arch.pio.count = 0;
12200 vcpu->arch.guest_ins_data = data;
12201 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12207 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12208 unsigned int port, void *data, unsigned int count,
12211 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
12212 : kvm_sev_es_outs(vcpu, size, port, data, count);
12214 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12216 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12217 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12218 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12219 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12220 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12221 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12222 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12223 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12224 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12225 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12226 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12227 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12228 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12229 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12230 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12231 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12232 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12233 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12234 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12235 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12236 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12237 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12238 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12239 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12240 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12241 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12242 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);