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>
62 #include <trace/events/kvm.h>
64 #include <asm/debugreg.h>
68 #include <linux/kernel_stat.h>
69 #include <asm/fpu/internal.h> /* Ugh! */
70 #include <asm/pvclock.h>
71 #include <asm/div64.h>
72 #include <asm/irq_remapping.h>
73 #include <asm/mshyperv.h>
74 #include <asm/hypervisor.h>
75 #include <asm/tlbflush.h>
76 #include <asm/intel_pt.h>
77 #include <asm/emulate_prefix.h>
79 #include <clocksource/hyperv_timer.h>
81 #define CREATE_TRACE_POINTS
84 #define MAX_IO_MSRS 256
85 #define KVM_MAX_MCE_BANKS 32
86 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
87 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
89 #define emul_to_vcpu(ctxt) \
90 ((struct kvm_vcpu *)(ctxt)->vcpu)
93 * - enable syscall per default because its emulated by KVM
94 * - enable LME and LMA per default on 64 bit KVM
98 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
100 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
103 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
105 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
106 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
108 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
109 static void process_nmi(struct kvm_vcpu *vcpu);
110 static void process_smi(struct kvm_vcpu *vcpu);
111 static void enter_smm(struct kvm_vcpu *vcpu);
112 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
113 static void store_regs(struct kvm_vcpu *vcpu);
114 static int sync_regs(struct kvm_vcpu *vcpu);
116 struct kvm_x86_ops kvm_x86_ops __read_mostly;
117 EXPORT_SYMBOL_GPL(kvm_x86_ops);
119 #define KVM_X86_OP(func) \
120 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
121 *(((struct kvm_x86_ops *)0)->func));
122 #define KVM_X86_OP_NULL KVM_X86_OP
123 #include <asm/kvm-x86-ops.h>
124 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
125 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
126 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
128 static bool __read_mostly ignore_msrs = 0;
129 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
131 bool __read_mostly report_ignored_msrs = true;
132 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
133 EXPORT_SYMBOL_GPL(report_ignored_msrs);
135 unsigned int min_timer_period_us = 200;
136 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
138 static bool __read_mostly kvmclock_periodic_sync = true;
139 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
141 bool __read_mostly kvm_has_tsc_control;
142 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
143 u32 __read_mostly kvm_max_guest_tsc_khz;
144 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
145 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
146 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
147 u64 __read_mostly kvm_max_tsc_scaling_ratio;
148 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
149 u64 __read_mostly kvm_default_tsc_scaling_ratio;
150 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
151 bool __read_mostly kvm_has_bus_lock_exit;
152 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
154 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
155 static u32 __read_mostly tsc_tolerance_ppm = 250;
156 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
159 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
160 * adaptive tuning starting from default advancement of 1000ns. '0' disables
161 * advancement entirely. Any other value is used as-is and disables adaptive
162 * tuning, i.e. allows privileged userspace to set an exact advancement time.
164 static int __read_mostly lapic_timer_advance_ns = -1;
165 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
167 static bool __read_mostly vector_hashing = true;
168 module_param(vector_hashing, bool, S_IRUGO);
170 bool __read_mostly enable_vmware_backdoor = false;
171 module_param(enable_vmware_backdoor, bool, S_IRUGO);
172 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
174 static bool __read_mostly force_emulation_prefix = false;
175 module_param(force_emulation_prefix, bool, S_IRUGO);
177 int __read_mostly pi_inject_timer = -1;
178 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
181 * Restoring the host value for MSRs that are only consumed when running in
182 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
183 * returns to userspace, i.e. the kernel can run with the guest's value.
185 #define KVM_MAX_NR_USER_RETURN_MSRS 16
187 struct kvm_user_return_msrs {
188 struct user_return_notifier urn;
190 struct kvm_user_return_msr_values {
193 } values[KVM_MAX_NR_USER_RETURN_MSRS];
196 u32 __read_mostly kvm_nr_uret_msrs;
197 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
198 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
199 static struct kvm_user_return_msrs __percpu *user_return_msrs;
201 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
202 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
203 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
204 | XFEATURE_MASK_PKRU)
206 u64 __read_mostly host_efer;
207 EXPORT_SYMBOL_GPL(host_efer);
209 bool __read_mostly allow_smaller_maxphyaddr = 0;
210 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
212 u64 __read_mostly host_xss;
213 EXPORT_SYMBOL_GPL(host_xss);
214 u64 __read_mostly supported_xss;
215 EXPORT_SYMBOL_GPL(supported_xss);
217 struct kvm_stats_debugfs_item debugfs_entries[] = {
218 VCPU_STAT("pf_fixed", pf_fixed),
219 VCPU_STAT("pf_guest", pf_guest),
220 VCPU_STAT("tlb_flush", tlb_flush),
221 VCPU_STAT("invlpg", invlpg),
222 VCPU_STAT("exits", exits),
223 VCPU_STAT("io_exits", io_exits),
224 VCPU_STAT("mmio_exits", mmio_exits),
225 VCPU_STAT("signal_exits", signal_exits),
226 VCPU_STAT("irq_window", irq_window_exits),
227 VCPU_STAT("nmi_window", nmi_window_exits),
228 VCPU_STAT("halt_exits", halt_exits),
229 VCPU_STAT("halt_successful_poll", halt_successful_poll),
230 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
231 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
232 VCPU_STAT("halt_wakeup", halt_wakeup),
233 VCPU_STAT("hypercalls", hypercalls),
234 VCPU_STAT("request_irq", request_irq_exits),
235 VCPU_STAT("irq_exits", irq_exits),
236 VCPU_STAT("host_state_reload", host_state_reload),
237 VCPU_STAT("fpu_reload", fpu_reload),
238 VCPU_STAT("insn_emulation", insn_emulation),
239 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
240 VCPU_STAT("irq_injections", irq_injections),
241 VCPU_STAT("nmi_injections", nmi_injections),
242 VCPU_STAT("req_event", req_event),
243 VCPU_STAT("l1d_flush", l1d_flush),
244 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
245 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
246 VCPU_STAT("nested_run", nested_run),
247 VCPU_STAT("directed_yield_attempted", directed_yield_attempted),
248 VCPU_STAT("directed_yield_successful", directed_yield_successful),
249 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
250 VM_STAT("mmu_pte_write", mmu_pte_write),
251 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
252 VM_STAT("mmu_flooded", mmu_flooded),
253 VM_STAT("mmu_recycled", mmu_recycled),
254 VM_STAT("mmu_cache_miss", mmu_cache_miss),
255 VM_STAT("mmu_unsync", mmu_unsync),
256 VM_STAT("remote_tlb_flush", remote_tlb_flush),
257 VM_STAT("largepages", lpages, .mode = 0444),
258 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
259 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
263 u64 __read_mostly host_xcr0;
264 u64 __read_mostly supported_xcr0;
265 EXPORT_SYMBOL_GPL(supported_xcr0);
267 static struct kmem_cache *x86_fpu_cache;
269 static struct kmem_cache *x86_emulator_cache;
272 * When called, it means the previous get/set msr reached an invalid msr.
273 * Return true if we want to ignore/silent this failed msr access.
275 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
277 const char *op = write ? "wrmsr" : "rdmsr";
280 if (report_ignored_msrs)
281 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
286 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
292 static struct kmem_cache *kvm_alloc_emulator_cache(void)
294 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
295 unsigned int size = sizeof(struct x86_emulate_ctxt);
297 return kmem_cache_create_usercopy("x86_emulator", size,
298 __alignof__(struct x86_emulate_ctxt),
299 SLAB_ACCOUNT, useroffset,
300 size - useroffset, NULL);
303 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
305 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
308 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
309 vcpu->arch.apf.gfns[i] = ~0;
312 static void kvm_on_user_return(struct user_return_notifier *urn)
315 struct kvm_user_return_msrs *msrs
316 = container_of(urn, struct kvm_user_return_msrs, urn);
317 struct kvm_user_return_msr_values *values;
321 * Disabling irqs at this point since the following code could be
322 * interrupted and executed through kvm_arch_hardware_disable()
324 local_irq_save(flags);
325 if (msrs->registered) {
326 msrs->registered = false;
327 user_return_notifier_unregister(urn);
329 local_irq_restore(flags);
330 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
331 values = &msrs->values[slot];
332 if (values->host != values->curr) {
333 wrmsrl(kvm_uret_msrs_list[slot], values->host);
334 values->curr = values->host;
339 static int kvm_probe_user_return_msr(u32 msr)
345 ret = rdmsrl_safe(msr, &val);
348 ret = wrmsrl_safe(msr, val);
354 int kvm_add_user_return_msr(u32 msr)
356 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
358 if (kvm_probe_user_return_msr(msr))
361 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
362 return kvm_nr_uret_msrs++;
364 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
366 int kvm_find_user_return_msr(u32 msr)
370 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
371 if (kvm_uret_msrs_list[i] == msr)
376 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
378 static void kvm_user_return_msr_cpu_online(void)
380 unsigned int cpu = smp_processor_id();
381 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
385 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
386 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
387 msrs->values[i].host = value;
388 msrs->values[i].curr = value;
392 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
394 unsigned int cpu = smp_processor_id();
395 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
398 value = (value & mask) | (msrs->values[slot].host & ~mask);
399 if (value == msrs->values[slot].curr)
401 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
405 msrs->values[slot].curr = value;
406 if (!msrs->registered) {
407 msrs->urn.on_user_return = kvm_on_user_return;
408 user_return_notifier_register(&msrs->urn);
409 msrs->registered = true;
413 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
415 static void drop_user_return_notifiers(void)
417 unsigned int cpu = smp_processor_id();
418 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
420 if (msrs->registered)
421 kvm_on_user_return(&msrs->urn);
424 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
426 return vcpu->arch.apic_base;
428 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
430 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
432 return kvm_apic_mode(kvm_get_apic_base(vcpu));
434 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
436 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
438 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
439 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
440 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
441 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
443 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
445 if (!msr_info->host_initiated) {
446 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
448 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
452 kvm_lapic_set_base(vcpu, msr_info->data);
453 kvm_recalculate_apic_map(vcpu->kvm);
456 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
458 asmlinkage __visible noinstr void kvm_spurious_fault(void)
460 /* Fault while not rebooting. We want the trace. */
461 BUG_ON(!kvm_rebooting);
463 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
465 #define EXCPT_BENIGN 0
466 #define EXCPT_CONTRIBUTORY 1
469 static int exception_class(int vector)
479 return EXCPT_CONTRIBUTORY;
486 #define EXCPT_FAULT 0
488 #define EXCPT_ABORT 2
489 #define EXCPT_INTERRUPT 3
491 static int exception_type(int vector)
495 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
496 return EXCPT_INTERRUPT;
500 /* #DB is trap, as instruction watchpoints are handled elsewhere */
501 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
504 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
507 /* Reserved exceptions will result in fault */
511 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
513 unsigned nr = vcpu->arch.exception.nr;
514 bool has_payload = vcpu->arch.exception.has_payload;
515 unsigned long payload = vcpu->arch.exception.payload;
523 * "Certain debug exceptions may clear bit 0-3. The
524 * remaining contents of the DR6 register are never
525 * cleared by the processor".
527 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
529 * In order to reflect the #DB exception payload in guest
530 * dr6, three components need to be considered: active low
531 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
533 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
534 * In the target guest dr6:
535 * FIXED_1 bits should always be set.
536 * Active low bits should be cleared if 1-setting in payload.
537 * Active high bits should be set if 1-setting in payload.
539 * Note, the payload is compatible with the pending debug
540 * exceptions/exit qualification under VMX, that active_low bits
541 * are active high in payload.
542 * So they need to be flipped for DR6.
544 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
545 vcpu->arch.dr6 |= payload;
546 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
549 * The #DB payload is defined as compatible with the 'pending
550 * debug exceptions' field under VMX, not DR6. While bit 12 is
551 * defined in the 'pending debug exceptions' field (enabled
552 * breakpoint), it is reserved and must be zero in DR6.
554 vcpu->arch.dr6 &= ~BIT(12);
557 vcpu->arch.cr2 = payload;
561 vcpu->arch.exception.has_payload = false;
562 vcpu->arch.exception.payload = 0;
564 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
566 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
567 unsigned nr, bool has_error, u32 error_code,
568 bool has_payload, unsigned long payload, bool reinject)
573 kvm_make_request(KVM_REQ_EVENT, vcpu);
575 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
579 * On vmentry, vcpu->arch.exception.pending is only
580 * true if an event injection was blocked by
581 * nested_run_pending. In that case, however,
582 * vcpu_enter_guest requests an immediate exit,
583 * and the guest shouldn't proceed far enough to
586 WARN_ON_ONCE(vcpu->arch.exception.pending);
587 vcpu->arch.exception.injected = true;
588 if (WARN_ON_ONCE(has_payload)) {
590 * A reinjected event has already
591 * delivered its payload.
597 vcpu->arch.exception.pending = true;
598 vcpu->arch.exception.injected = false;
600 vcpu->arch.exception.has_error_code = has_error;
601 vcpu->arch.exception.nr = nr;
602 vcpu->arch.exception.error_code = error_code;
603 vcpu->arch.exception.has_payload = has_payload;
604 vcpu->arch.exception.payload = payload;
605 if (!is_guest_mode(vcpu))
606 kvm_deliver_exception_payload(vcpu);
610 /* to check exception */
611 prev_nr = vcpu->arch.exception.nr;
612 if (prev_nr == DF_VECTOR) {
613 /* triple fault -> shutdown */
614 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
617 class1 = exception_class(prev_nr);
618 class2 = exception_class(nr);
619 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
620 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
622 * Generate double fault per SDM Table 5-5. Set
623 * exception.pending = true so that the double fault
624 * can trigger a nested vmexit.
626 vcpu->arch.exception.pending = true;
627 vcpu->arch.exception.injected = false;
628 vcpu->arch.exception.has_error_code = true;
629 vcpu->arch.exception.nr = DF_VECTOR;
630 vcpu->arch.exception.error_code = 0;
631 vcpu->arch.exception.has_payload = false;
632 vcpu->arch.exception.payload = 0;
634 /* replace previous exception with a new one in a hope
635 that instruction re-execution will regenerate lost
640 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
642 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
644 EXPORT_SYMBOL_GPL(kvm_queue_exception);
646 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
648 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
650 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
652 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
653 unsigned long payload)
655 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
657 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
659 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
660 u32 error_code, unsigned long payload)
662 kvm_multiple_exception(vcpu, nr, true, error_code,
663 true, payload, false);
666 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
669 kvm_inject_gp(vcpu, 0);
671 return kvm_skip_emulated_instruction(vcpu);
675 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
677 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
679 ++vcpu->stat.pf_guest;
680 vcpu->arch.exception.nested_apf =
681 is_guest_mode(vcpu) && fault->async_page_fault;
682 if (vcpu->arch.exception.nested_apf) {
683 vcpu->arch.apf.nested_apf_token = fault->address;
684 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
686 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
690 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
692 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
693 struct x86_exception *fault)
695 struct kvm_mmu *fault_mmu;
696 WARN_ON_ONCE(fault->vector != PF_VECTOR);
698 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
702 * Invalidate the TLB entry for the faulting address, if it exists,
703 * else the access will fault indefinitely (and to emulate hardware).
705 if ((fault->error_code & PFERR_PRESENT_MASK) &&
706 !(fault->error_code & PFERR_RSVD_MASK))
707 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
708 fault_mmu->root_hpa);
710 fault_mmu->inject_page_fault(vcpu, fault);
711 return fault->nested_page_fault;
713 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
715 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
717 atomic_inc(&vcpu->arch.nmi_queued);
718 kvm_make_request(KVM_REQ_NMI, vcpu);
720 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
722 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
724 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
726 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
728 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
730 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
732 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
735 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
736 * a #GP and return false.
738 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
740 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
742 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
745 EXPORT_SYMBOL_GPL(kvm_require_cpl);
747 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
749 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
752 kvm_queue_exception(vcpu, UD_VECTOR);
755 EXPORT_SYMBOL_GPL(kvm_require_dr);
758 * This function will be used to read from the physical memory of the currently
759 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
760 * can read from guest physical or from the guest's guest physical memory.
762 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
763 gfn_t ngfn, void *data, int offset, int len,
766 struct x86_exception exception;
770 ngpa = gfn_to_gpa(ngfn);
771 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
772 if (real_gfn == UNMAPPED_GVA)
775 real_gfn = gpa_to_gfn(real_gfn);
777 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
779 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
781 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
782 void *data, int offset, int len, u32 access)
784 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
785 data, offset, len, access);
788 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
790 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
794 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
796 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
798 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
799 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
802 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
804 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
805 offset * sizeof(u64), sizeof(pdpte),
806 PFERR_USER_MASK|PFERR_WRITE_MASK);
811 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
812 if ((pdpte[i] & PT_PRESENT_MASK) &&
813 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
820 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
821 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
827 EXPORT_SYMBOL_GPL(load_pdptrs);
829 bool pdptrs_changed(struct kvm_vcpu *vcpu)
831 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
836 if (!is_pae_paging(vcpu))
839 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
842 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
843 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
844 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
845 PFERR_USER_MASK | PFERR_WRITE_MASK);
849 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
851 EXPORT_SYMBOL_GPL(pdptrs_changed);
853 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
855 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
857 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
858 kvm_clear_async_pf_completion_queue(vcpu);
859 kvm_async_pf_hash_reset(vcpu);
862 if ((cr0 ^ old_cr0) & update_bits)
863 kvm_mmu_reset_context(vcpu);
865 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
866 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
867 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
868 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
870 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
872 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
874 unsigned long old_cr0 = kvm_read_cr0(vcpu);
875 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
880 if (cr0 & 0xffffffff00000000UL)
884 cr0 &= ~CR0_RESERVED_BITS;
886 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
889 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
893 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
894 (cr0 & X86_CR0_PG)) {
899 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
904 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
905 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
906 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
909 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
912 static_call(kvm_x86_set_cr0)(vcpu, cr0);
914 kvm_post_set_cr0(vcpu, old_cr0, cr0);
918 EXPORT_SYMBOL_GPL(kvm_set_cr0);
920 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
922 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
924 EXPORT_SYMBOL_GPL(kvm_lmsw);
926 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
928 if (vcpu->arch.guest_state_protected)
931 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
933 if (vcpu->arch.xcr0 != host_xcr0)
934 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
936 if (vcpu->arch.xsaves_enabled &&
937 vcpu->arch.ia32_xss != host_xss)
938 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
941 if (static_cpu_has(X86_FEATURE_PKU) &&
942 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
943 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
944 vcpu->arch.pkru != vcpu->arch.host_pkru)
945 __write_pkru(vcpu->arch.pkru);
947 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
949 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
951 if (vcpu->arch.guest_state_protected)
954 if (static_cpu_has(X86_FEATURE_PKU) &&
955 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
956 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
957 vcpu->arch.pkru = rdpkru();
958 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
959 __write_pkru(vcpu->arch.host_pkru);
962 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
964 if (vcpu->arch.xcr0 != host_xcr0)
965 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
967 if (vcpu->arch.xsaves_enabled &&
968 vcpu->arch.ia32_xss != host_xss)
969 wrmsrl(MSR_IA32_XSS, host_xss);
973 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
975 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
978 u64 old_xcr0 = vcpu->arch.xcr0;
981 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
982 if (index != XCR_XFEATURE_ENABLED_MASK)
984 if (!(xcr0 & XFEATURE_MASK_FP))
986 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
990 * Do not allow the guest to set bits that we do not support
991 * saving. However, xcr0 bit 0 is always set, even if the
992 * emulated CPU does not support XSAVE (see fx_init).
994 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
995 if (xcr0 & ~valid_bits)
998 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
999 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1002 if (xcr0 & XFEATURE_MASK_AVX512) {
1003 if (!(xcr0 & XFEATURE_MASK_YMM))
1005 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1008 vcpu->arch.xcr0 = xcr0;
1010 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1011 kvm_update_cpuid_runtime(vcpu);
1015 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1017 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1018 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1019 kvm_inject_gp(vcpu, 0);
1023 return kvm_skip_emulated_instruction(vcpu);
1025 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1027 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1029 if (cr4 & cr4_reserved_bits)
1032 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1035 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1037 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1039 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1041 unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1042 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
1044 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1045 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1046 kvm_mmu_reset_context(vcpu);
1048 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1050 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1052 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1053 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1056 if (!kvm_is_valid_cr4(vcpu, cr4))
1059 if (is_long_mode(vcpu)) {
1060 if (!(cr4 & X86_CR4_PAE))
1062 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1064 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1065 && ((cr4 ^ old_cr4) & pdptr_bits)
1066 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1067 kvm_read_cr3(vcpu)))
1070 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1071 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1074 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1075 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1079 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1081 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1085 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1087 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1089 bool skip_tlb_flush = false;
1090 #ifdef CONFIG_X86_64
1091 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1094 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1095 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1099 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1100 if (!skip_tlb_flush) {
1101 kvm_mmu_sync_roots(vcpu);
1102 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1108 * Do not condition the GPA check on long mode, this helper is used to
1109 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1110 * the current vCPU mode is accurate.
1112 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1115 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1118 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1119 vcpu->arch.cr3 = cr3;
1120 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1124 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1126 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1128 if (cr8 & CR8_RESERVED_BITS)
1130 if (lapic_in_kernel(vcpu))
1131 kvm_lapic_set_tpr(vcpu, cr8);
1133 vcpu->arch.cr8 = cr8;
1136 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1138 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1140 if (lapic_in_kernel(vcpu))
1141 return kvm_lapic_get_cr8(vcpu);
1143 return vcpu->arch.cr8;
1145 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1147 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1151 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1152 for (i = 0; i < KVM_NR_DB_REGS; i++)
1153 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1154 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1158 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1162 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1163 dr7 = vcpu->arch.guest_debug_dr7;
1165 dr7 = vcpu->arch.dr7;
1166 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1167 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1168 if (dr7 & DR7_BP_EN_MASK)
1169 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1171 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1173 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1175 u64 fixed = DR6_FIXED_1;
1177 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1180 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1181 fixed |= DR6_BUS_LOCK;
1185 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1187 size_t size = ARRAY_SIZE(vcpu->arch.db);
1191 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1192 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1193 vcpu->arch.eff_db[dr] = val;
1197 if (!kvm_dr6_valid(val))
1199 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1203 if (!kvm_dr7_valid(val))
1205 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1206 kvm_update_dr7(vcpu);
1212 EXPORT_SYMBOL_GPL(kvm_set_dr);
1214 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1216 size_t size = ARRAY_SIZE(vcpu->arch.db);
1220 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1224 *val = vcpu->arch.dr6;
1228 *val = vcpu->arch.dr7;
1232 EXPORT_SYMBOL_GPL(kvm_get_dr);
1234 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1236 u32 ecx = kvm_rcx_read(vcpu);
1239 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1240 kvm_inject_gp(vcpu, 0);
1244 kvm_rax_write(vcpu, (u32)data);
1245 kvm_rdx_write(vcpu, data >> 32);
1246 return kvm_skip_emulated_instruction(vcpu);
1248 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1251 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1252 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1254 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1255 * extract the supported MSRs from the related const lists.
1256 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1257 * capabilities of the host cpu. This capabilities test skips MSRs that are
1258 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1259 * may depend on host virtualization features rather than host cpu features.
1262 static const u32 msrs_to_save_all[] = {
1263 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1265 #ifdef CONFIG_X86_64
1266 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1268 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1269 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1271 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1272 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1273 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1274 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1275 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1276 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1277 MSR_IA32_UMWAIT_CONTROL,
1279 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1280 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1281 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1282 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1283 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1284 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1285 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1286 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1287 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1288 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1289 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1290 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1291 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1292 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1293 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1294 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1295 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1296 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1297 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1298 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1299 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1300 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1303 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1304 static unsigned num_msrs_to_save;
1306 static const u32 emulated_msrs_all[] = {
1307 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1308 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1309 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1310 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1311 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1312 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1313 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1315 HV_X64_MSR_VP_INDEX,
1316 HV_X64_MSR_VP_RUNTIME,
1317 HV_X64_MSR_SCONTROL,
1318 HV_X64_MSR_STIMER0_CONFIG,
1319 HV_X64_MSR_VP_ASSIST_PAGE,
1320 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1321 HV_X64_MSR_TSC_EMULATION_STATUS,
1322 HV_X64_MSR_SYNDBG_OPTIONS,
1323 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1324 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1325 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1327 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1328 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1330 MSR_IA32_TSC_ADJUST,
1331 MSR_IA32_TSC_DEADLINE,
1332 MSR_IA32_ARCH_CAPABILITIES,
1333 MSR_IA32_PERF_CAPABILITIES,
1334 MSR_IA32_MISC_ENABLE,
1335 MSR_IA32_MCG_STATUS,
1337 MSR_IA32_MCG_EXT_CTL,
1341 MSR_MISC_FEATURES_ENABLES,
1342 MSR_AMD64_VIRT_SPEC_CTRL,
1347 * The following list leaves out MSRs whose values are determined
1348 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1349 * We always support the "true" VMX control MSRs, even if the host
1350 * processor does not, so I am putting these registers here rather
1351 * than in msrs_to_save_all.
1354 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1355 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1356 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1357 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1359 MSR_IA32_VMX_CR0_FIXED0,
1360 MSR_IA32_VMX_CR4_FIXED0,
1361 MSR_IA32_VMX_VMCS_ENUM,
1362 MSR_IA32_VMX_PROCBASED_CTLS2,
1363 MSR_IA32_VMX_EPT_VPID_CAP,
1364 MSR_IA32_VMX_VMFUNC,
1367 MSR_KVM_POLL_CONTROL,
1370 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1371 static unsigned num_emulated_msrs;
1374 * List of msr numbers which are used to expose MSR-based features that
1375 * can be used by a hypervisor to validate requested CPU features.
1377 static const u32 msr_based_features_all[] = {
1379 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1380 MSR_IA32_VMX_PINBASED_CTLS,
1381 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1382 MSR_IA32_VMX_PROCBASED_CTLS,
1383 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1384 MSR_IA32_VMX_EXIT_CTLS,
1385 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1386 MSR_IA32_VMX_ENTRY_CTLS,
1388 MSR_IA32_VMX_CR0_FIXED0,
1389 MSR_IA32_VMX_CR0_FIXED1,
1390 MSR_IA32_VMX_CR4_FIXED0,
1391 MSR_IA32_VMX_CR4_FIXED1,
1392 MSR_IA32_VMX_VMCS_ENUM,
1393 MSR_IA32_VMX_PROCBASED_CTLS2,
1394 MSR_IA32_VMX_EPT_VPID_CAP,
1395 MSR_IA32_VMX_VMFUNC,
1399 MSR_IA32_ARCH_CAPABILITIES,
1400 MSR_IA32_PERF_CAPABILITIES,
1403 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1404 static unsigned int num_msr_based_features;
1406 static u64 kvm_get_arch_capabilities(void)
1410 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1411 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1414 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1415 * the nested hypervisor runs with NX huge pages. If it is not,
1416 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1417 * L1 guests, so it need not worry about its own (L2) guests.
1419 data |= ARCH_CAP_PSCHANGE_MC_NO;
1422 * If we're doing cache flushes (either "always" or "cond")
1423 * we will do one whenever the guest does a vmlaunch/vmresume.
1424 * If an outer hypervisor is doing the cache flush for us
1425 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1426 * capability to the guest too, and if EPT is disabled we're not
1427 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1428 * require a nested hypervisor to do a flush of its own.
1430 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1431 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1433 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1434 data |= ARCH_CAP_RDCL_NO;
1435 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1436 data |= ARCH_CAP_SSB_NO;
1437 if (!boot_cpu_has_bug(X86_BUG_MDS))
1438 data |= ARCH_CAP_MDS_NO;
1440 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1442 * If RTM=0 because the kernel has disabled TSX, the host might
1443 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1444 * and therefore knows that there cannot be TAA) but keep
1445 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1446 * and we want to allow migrating those guests to tsx=off hosts.
1448 data &= ~ARCH_CAP_TAA_NO;
1449 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1450 data |= ARCH_CAP_TAA_NO;
1453 * Nothing to do here; we emulate TSX_CTRL if present on the
1454 * host so the guest can choose between disabling TSX or
1455 * using VERW to clear CPU buffers.
1462 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1464 switch (msr->index) {
1465 case MSR_IA32_ARCH_CAPABILITIES:
1466 msr->data = kvm_get_arch_capabilities();
1468 case MSR_IA32_UCODE_REV:
1469 rdmsrl_safe(msr->index, &msr->data);
1472 return static_call(kvm_x86_get_msr_feature)(msr);
1477 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1479 struct kvm_msr_entry msr;
1483 r = kvm_get_msr_feature(&msr);
1485 if (r == KVM_MSR_RET_INVALID) {
1486 /* Unconditionally clear the output for simplicity */
1488 if (kvm_msr_ignored_check(index, 0, false))
1500 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1502 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1505 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1508 if (efer & (EFER_LME | EFER_LMA) &&
1509 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1512 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1518 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1520 if (efer & efer_reserved_bits)
1523 return __kvm_valid_efer(vcpu, efer);
1525 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1527 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1529 u64 old_efer = vcpu->arch.efer;
1530 u64 efer = msr_info->data;
1533 if (efer & efer_reserved_bits)
1536 if (!msr_info->host_initiated) {
1537 if (!__kvm_valid_efer(vcpu, efer))
1540 if (is_paging(vcpu) &&
1541 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1546 efer |= vcpu->arch.efer & EFER_LMA;
1548 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1554 /* Update reserved bits */
1555 if ((efer ^ old_efer) & EFER_NX)
1556 kvm_mmu_reset_context(vcpu);
1561 void kvm_enable_efer_bits(u64 mask)
1563 efer_reserved_bits &= ~mask;
1565 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1567 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1569 struct kvm_x86_msr_filter *msr_filter;
1570 struct msr_bitmap_range *ranges;
1571 struct kvm *kvm = vcpu->kvm;
1576 /* x2APIC MSRs do not support filtering. */
1577 if (index >= 0x800 && index <= 0x8ff)
1580 idx = srcu_read_lock(&kvm->srcu);
1582 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1588 allowed = msr_filter->default_allow;
1589 ranges = msr_filter->ranges;
1591 for (i = 0; i < msr_filter->count; i++) {
1592 u32 start = ranges[i].base;
1593 u32 end = start + ranges[i].nmsrs;
1594 u32 flags = ranges[i].flags;
1595 unsigned long *bitmap = ranges[i].bitmap;
1597 if ((index >= start) && (index < end) && (flags & type)) {
1598 allowed = !!test_bit(index - start, bitmap);
1604 srcu_read_unlock(&kvm->srcu, idx);
1608 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1611 * Write @data into the MSR specified by @index. Select MSR specific fault
1612 * checks are bypassed if @host_initiated is %true.
1613 * Returns 0 on success, non-0 otherwise.
1614 * Assumes vcpu_load() was already called.
1616 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1617 bool host_initiated)
1619 struct msr_data msr;
1621 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1622 return KVM_MSR_RET_FILTERED;
1627 case MSR_KERNEL_GS_BASE:
1630 if (is_noncanonical_address(data, vcpu))
1633 case MSR_IA32_SYSENTER_EIP:
1634 case MSR_IA32_SYSENTER_ESP:
1636 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1637 * non-canonical address is written on Intel but not on
1638 * AMD (which ignores the top 32-bits, because it does
1639 * not implement 64-bit SYSENTER).
1641 * 64-bit code should hence be able to write a non-canonical
1642 * value on AMD. Making the address canonical ensures that
1643 * vmentry does not fail on Intel after writing a non-canonical
1644 * value, and that something deterministic happens if the guest
1645 * invokes 64-bit SYSENTER.
1647 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1650 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1653 if (!host_initiated &&
1654 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1655 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1659 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1660 * incomplete and conflicting architectural behavior. Current
1661 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1662 * reserved and always read as zeros. Enforce Intel's reserved
1663 * bits check if and only if the guest CPU is Intel, and clear
1664 * the bits in all other cases. This ensures cross-vendor
1665 * migration will provide consistent behavior for the guest.
1667 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1676 msr.host_initiated = host_initiated;
1678 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1681 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1682 u32 index, u64 data, bool host_initiated)
1684 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1686 if (ret == KVM_MSR_RET_INVALID)
1687 if (kvm_msr_ignored_check(index, data, true))
1694 * Read the MSR specified by @index into @data. Select MSR specific fault
1695 * checks are bypassed if @host_initiated is %true.
1696 * Returns 0 on success, non-0 otherwise.
1697 * Assumes vcpu_load() was already called.
1699 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1700 bool host_initiated)
1702 struct msr_data msr;
1705 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1706 return KVM_MSR_RET_FILTERED;
1710 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1713 if (!host_initiated &&
1714 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1715 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1721 msr.host_initiated = host_initiated;
1723 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1729 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1730 u32 index, u64 *data, bool host_initiated)
1732 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1734 if (ret == KVM_MSR_RET_INVALID) {
1735 /* Unconditionally clear *data for simplicity */
1737 if (kvm_msr_ignored_check(index, 0, false))
1744 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1746 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1748 EXPORT_SYMBOL_GPL(kvm_get_msr);
1750 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1752 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1754 EXPORT_SYMBOL_GPL(kvm_set_msr);
1756 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1758 int err = vcpu->run->msr.error;
1760 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1761 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1764 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1767 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1769 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1772 static u64 kvm_msr_reason(int r)
1775 case KVM_MSR_RET_INVALID:
1776 return KVM_MSR_EXIT_REASON_UNKNOWN;
1777 case KVM_MSR_RET_FILTERED:
1778 return KVM_MSR_EXIT_REASON_FILTER;
1780 return KVM_MSR_EXIT_REASON_INVAL;
1784 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1785 u32 exit_reason, u64 data,
1786 int (*completion)(struct kvm_vcpu *vcpu),
1789 u64 msr_reason = kvm_msr_reason(r);
1791 /* Check if the user wanted to know about this MSR fault */
1792 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1795 vcpu->run->exit_reason = exit_reason;
1796 vcpu->run->msr.error = 0;
1797 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1798 vcpu->run->msr.reason = msr_reason;
1799 vcpu->run->msr.index = index;
1800 vcpu->run->msr.data = data;
1801 vcpu->arch.complete_userspace_io = completion;
1806 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1808 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1809 complete_emulated_rdmsr, r);
1812 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1814 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1815 complete_emulated_wrmsr, r);
1818 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1820 u32 ecx = kvm_rcx_read(vcpu);
1824 r = kvm_get_msr(vcpu, ecx, &data);
1826 /* MSR read failed? See if we should ask user space */
1827 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1828 /* Bounce to user space */
1833 trace_kvm_msr_read(ecx, data);
1835 kvm_rax_write(vcpu, data & -1u);
1836 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1838 trace_kvm_msr_read_ex(ecx);
1841 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1843 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1845 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1847 u32 ecx = kvm_rcx_read(vcpu);
1848 u64 data = kvm_read_edx_eax(vcpu);
1851 r = kvm_set_msr(vcpu, ecx, data);
1853 /* MSR write failed? See if we should ask user space */
1854 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1855 /* Bounce to user space */
1858 /* Signal all other negative errors to userspace */
1863 trace_kvm_msr_write(ecx, data);
1865 trace_kvm_msr_write_ex(ecx, data);
1867 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1869 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1871 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1873 return kvm_skip_emulated_instruction(vcpu);
1875 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1877 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1879 /* Treat an INVD instruction as a NOP and just skip it. */
1880 return kvm_emulate_as_nop(vcpu);
1882 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1884 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1886 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1887 return kvm_emulate_as_nop(vcpu);
1889 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1891 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1893 kvm_queue_exception(vcpu, UD_VECTOR);
1896 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1898 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1900 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1901 return kvm_emulate_as_nop(vcpu);
1903 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1905 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1907 xfer_to_guest_mode_prepare();
1908 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1909 xfer_to_guest_mode_work_pending();
1913 * The fast path for frequent and performance sensitive wrmsr emulation,
1914 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1915 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1916 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1917 * other cases which must be called after interrupts are enabled on the host.
1919 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1921 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1924 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1925 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1926 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1927 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1930 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1931 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1932 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1933 trace_kvm_apic_write(APIC_ICR, (u32)data);
1940 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1942 if (!kvm_can_use_hv_timer(vcpu))
1945 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1949 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1951 u32 msr = kvm_rcx_read(vcpu);
1953 fastpath_t ret = EXIT_FASTPATH_NONE;
1956 case APIC_BASE_MSR + (APIC_ICR >> 4):
1957 data = kvm_read_edx_eax(vcpu);
1958 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1959 kvm_skip_emulated_instruction(vcpu);
1960 ret = EXIT_FASTPATH_EXIT_HANDLED;
1963 case MSR_IA32_TSC_DEADLINE:
1964 data = kvm_read_edx_eax(vcpu);
1965 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1966 kvm_skip_emulated_instruction(vcpu);
1967 ret = EXIT_FASTPATH_REENTER_GUEST;
1974 if (ret != EXIT_FASTPATH_NONE)
1975 trace_kvm_msr_write(msr, data);
1979 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1982 * Adapt set_msr() to msr_io()'s calling convention
1984 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1986 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1989 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1991 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1994 #ifdef CONFIG_X86_64
1995 struct pvclock_clock {
2005 struct pvclock_gtod_data {
2008 struct pvclock_clock clock; /* extract of a clocksource struct */
2009 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2015 static struct pvclock_gtod_data pvclock_gtod_data;
2017 static void update_pvclock_gtod(struct timekeeper *tk)
2019 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2021 write_seqcount_begin(&vdata->seq);
2023 /* copy pvclock gtod data */
2024 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2025 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2026 vdata->clock.mask = tk->tkr_mono.mask;
2027 vdata->clock.mult = tk->tkr_mono.mult;
2028 vdata->clock.shift = tk->tkr_mono.shift;
2029 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2030 vdata->clock.offset = tk->tkr_mono.base;
2032 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2033 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2034 vdata->raw_clock.mask = tk->tkr_raw.mask;
2035 vdata->raw_clock.mult = tk->tkr_raw.mult;
2036 vdata->raw_clock.shift = tk->tkr_raw.shift;
2037 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2038 vdata->raw_clock.offset = tk->tkr_raw.base;
2040 vdata->wall_time_sec = tk->xtime_sec;
2042 vdata->offs_boot = tk->offs_boot;
2044 write_seqcount_end(&vdata->seq);
2047 static s64 get_kvmclock_base_ns(void)
2049 /* Count up from boot time, but with the frequency of the raw clock. */
2050 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2053 static s64 get_kvmclock_base_ns(void)
2055 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2056 return ktime_get_boottime_ns();
2060 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2064 struct pvclock_wall_clock wc;
2071 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2076 ++version; /* first time write, random junk */
2080 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2084 * The guest calculates current wall clock time by adding
2085 * system time (updated by kvm_guest_time_update below) to the
2086 * wall clock specified here. We do the reverse here.
2088 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2090 wc.nsec = do_div(wall_nsec, 1000000000);
2091 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2092 wc.version = version;
2094 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2097 wc_sec_hi = wall_nsec >> 32;
2098 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2099 &wc_sec_hi, sizeof(wc_sec_hi));
2103 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2106 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2107 bool old_msr, bool host_initiated)
2109 struct kvm_arch *ka = &vcpu->kvm->arch;
2111 if (vcpu->vcpu_id == 0 && !host_initiated) {
2112 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2113 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2115 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2118 vcpu->arch.time = system_time;
2119 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2121 /* we verify if the enable bit is set... */
2122 vcpu->arch.pv_time_enabled = false;
2123 if (!(system_time & 1))
2126 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2127 &vcpu->arch.pv_time, system_time & ~1ULL,
2128 sizeof(struct pvclock_vcpu_time_info)))
2129 vcpu->arch.pv_time_enabled = true;
2134 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2136 do_shl32_div32(dividend, divisor);
2140 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2141 s8 *pshift, u32 *pmultiplier)
2149 scaled64 = scaled_hz;
2150 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2155 tps32 = (uint32_t)tps64;
2156 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2157 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2165 *pmultiplier = div_frac(scaled64, tps32);
2168 #ifdef CONFIG_X86_64
2169 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2172 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2173 static unsigned long max_tsc_khz;
2175 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2177 u64 v = (u64)khz * (1000000 + ppm);
2182 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2186 /* Guest TSC same frequency as host TSC? */
2188 vcpu->arch.l1_tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2189 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2193 /* TSC scaling supported? */
2194 if (!kvm_has_tsc_control) {
2195 if (user_tsc_khz > tsc_khz) {
2196 vcpu->arch.tsc_catchup = 1;
2197 vcpu->arch.tsc_always_catchup = 1;
2200 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2205 /* TSC scaling required - calculate ratio */
2206 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2207 user_tsc_khz, tsc_khz);
2209 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2210 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2215 vcpu->arch.l1_tsc_scaling_ratio = vcpu->arch.tsc_scaling_ratio = ratio;
2219 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2221 u32 thresh_lo, thresh_hi;
2222 int use_scaling = 0;
2224 /* tsc_khz can be zero if TSC calibration fails */
2225 if (user_tsc_khz == 0) {
2226 /* set tsc_scaling_ratio to a safe value */
2227 vcpu->arch.l1_tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2228 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2232 /* Compute a scale to convert nanoseconds in TSC cycles */
2233 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2234 &vcpu->arch.virtual_tsc_shift,
2235 &vcpu->arch.virtual_tsc_mult);
2236 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2239 * Compute the variation in TSC rate which is acceptable
2240 * within the range of tolerance and decide if the
2241 * rate being applied is within that bounds of the hardware
2242 * rate. If so, no scaling or compensation need be done.
2244 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2245 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2246 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2247 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2250 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2253 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2255 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2256 vcpu->arch.virtual_tsc_mult,
2257 vcpu->arch.virtual_tsc_shift);
2258 tsc += vcpu->arch.this_tsc_write;
2262 static inline int gtod_is_based_on_tsc(int mode)
2264 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2267 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2269 #ifdef CONFIG_X86_64
2271 struct kvm_arch *ka = &vcpu->kvm->arch;
2272 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2274 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2275 atomic_read(&vcpu->kvm->online_vcpus));
2278 * Once the masterclock is enabled, always perform request in
2279 * order to update it.
2281 * In order to enable masterclock, the host clocksource must be TSC
2282 * and the vcpus need to have matched TSCs. When that happens,
2283 * perform request to enable masterclock.
2285 if (ka->use_master_clock ||
2286 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2287 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2289 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2290 atomic_read(&vcpu->kvm->online_vcpus),
2291 ka->use_master_clock, gtod->clock.vclock_mode);
2296 * Multiply tsc by a fixed point number represented by ratio.
2298 * The most significant 64-N bits (mult) of ratio represent the
2299 * integral part of the fixed point number; the remaining N bits
2300 * (frac) represent the fractional part, ie. ratio represents a fixed
2301 * point number (mult + frac * 2^(-N)).
2303 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2305 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2307 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2310 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2314 if (ratio != kvm_default_tsc_scaling_ratio)
2315 _tsc = __scale_tsc(ratio, tsc);
2319 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2321 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2325 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2327 return target_tsc - tsc;
2330 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2332 return vcpu->arch.l1_tsc_offset +
2333 kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2335 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2337 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2341 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2342 nested_offset = l1_offset;
2344 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2345 kvm_tsc_scaling_ratio_frac_bits);
2347 nested_offset += l2_offset;
2348 return nested_offset;
2350 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2352 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2354 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2355 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2356 kvm_tsc_scaling_ratio_frac_bits);
2358 return l1_multiplier;
2360 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2362 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2364 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2365 vcpu->arch.l1_tsc_offset,
2368 vcpu->arch.l1_tsc_offset = l1_offset;
2371 * If we are here because L1 chose not to trap WRMSR to TSC then
2372 * according to the spec this should set L1's TSC (as opposed to
2373 * setting L1's offset for L2).
2375 if (is_guest_mode(vcpu))
2376 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2378 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2379 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2381 vcpu->arch.tsc_offset = l1_offset;
2383 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2386 static inline bool kvm_check_tsc_unstable(void)
2388 #ifdef CONFIG_X86_64
2390 * TSC is marked unstable when we're running on Hyper-V,
2391 * 'TSC page' clocksource is good.
2393 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2396 return check_tsc_unstable();
2399 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2401 struct kvm *kvm = vcpu->kvm;
2402 u64 offset, ns, elapsed;
2403 unsigned long flags;
2405 bool already_matched;
2406 bool synchronizing = false;
2408 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2409 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2410 ns = get_kvmclock_base_ns();
2411 elapsed = ns - kvm->arch.last_tsc_nsec;
2413 if (vcpu->arch.virtual_tsc_khz) {
2416 * detection of vcpu initialization -- need to sync
2417 * with other vCPUs. This particularly helps to keep
2418 * kvm_clock stable after CPU hotplug
2420 synchronizing = true;
2422 u64 tsc_exp = kvm->arch.last_tsc_write +
2423 nsec_to_cycles(vcpu, elapsed);
2424 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2426 * Special case: TSC write with a small delta (1 second)
2427 * of virtual cycle time against real time is
2428 * interpreted as an attempt to synchronize the CPU.
2430 synchronizing = data < tsc_exp + tsc_hz &&
2431 data + tsc_hz > tsc_exp;
2436 * For a reliable TSC, we can match TSC offsets, and for an unstable
2437 * TSC, we add elapsed time in this computation. We could let the
2438 * compensation code attempt to catch up if we fall behind, but
2439 * it's better to try to match offsets from the beginning.
2441 if (synchronizing &&
2442 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2443 if (!kvm_check_tsc_unstable()) {
2444 offset = kvm->arch.cur_tsc_offset;
2446 u64 delta = nsec_to_cycles(vcpu, elapsed);
2448 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2451 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2454 * We split periods of matched TSC writes into generations.
2455 * For each generation, we track the original measured
2456 * nanosecond time, offset, and write, so if TSCs are in
2457 * sync, we can match exact offset, and if not, we can match
2458 * exact software computation in compute_guest_tsc()
2460 * These values are tracked in kvm->arch.cur_xxx variables.
2462 kvm->arch.cur_tsc_generation++;
2463 kvm->arch.cur_tsc_nsec = ns;
2464 kvm->arch.cur_tsc_write = data;
2465 kvm->arch.cur_tsc_offset = offset;
2470 * We also track th most recent recorded KHZ, write and time to
2471 * allow the matching interval to be extended at each write.
2473 kvm->arch.last_tsc_nsec = ns;
2474 kvm->arch.last_tsc_write = data;
2475 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2477 vcpu->arch.last_guest_tsc = data;
2479 /* Keep track of which generation this VCPU has synchronized to */
2480 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2481 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2482 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2484 kvm_vcpu_write_tsc_offset(vcpu, offset);
2485 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2487 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2489 kvm->arch.nr_vcpus_matched_tsc = 0;
2490 } else if (!already_matched) {
2491 kvm->arch.nr_vcpus_matched_tsc++;
2494 kvm_track_tsc_matching(vcpu);
2495 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2498 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2501 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2502 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2505 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2507 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2508 WARN_ON(adjustment < 0);
2509 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2510 vcpu->arch.l1_tsc_scaling_ratio);
2511 adjust_tsc_offset_guest(vcpu, adjustment);
2514 #ifdef CONFIG_X86_64
2516 static u64 read_tsc(void)
2518 u64 ret = (u64)rdtsc_ordered();
2519 u64 last = pvclock_gtod_data.clock.cycle_last;
2521 if (likely(ret >= last))
2525 * GCC likes to generate cmov here, but this branch is extremely
2526 * predictable (it's just a function of time and the likely is
2527 * very likely) and there's a data dependence, so force GCC
2528 * to generate a branch instead. I don't barrier() because
2529 * we don't actually need a barrier, and if this function
2530 * ever gets inlined it will generate worse code.
2536 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2542 switch (clock->vclock_mode) {
2543 case VDSO_CLOCKMODE_HVCLOCK:
2544 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2546 if (tsc_pg_val != U64_MAX) {
2547 /* TSC page valid */
2548 *mode = VDSO_CLOCKMODE_HVCLOCK;
2549 v = (tsc_pg_val - clock->cycle_last) &
2552 /* TSC page invalid */
2553 *mode = VDSO_CLOCKMODE_NONE;
2556 case VDSO_CLOCKMODE_TSC:
2557 *mode = VDSO_CLOCKMODE_TSC;
2558 *tsc_timestamp = read_tsc();
2559 v = (*tsc_timestamp - clock->cycle_last) &
2563 *mode = VDSO_CLOCKMODE_NONE;
2566 if (*mode == VDSO_CLOCKMODE_NONE)
2567 *tsc_timestamp = v = 0;
2569 return v * clock->mult;
2572 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2574 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2580 seq = read_seqcount_begin(>od->seq);
2581 ns = gtod->raw_clock.base_cycles;
2582 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2583 ns >>= gtod->raw_clock.shift;
2584 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2585 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2591 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2593 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2599 seq = read_seqcount_begin(>od->seq);
2600 ts->tv_sec = gtod->wall_time_sec;
2601 ns = gtod->clock.base_cycles;
2602 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2603 ns >>= gtod->clock.shift;
2604 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2606 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2612 /* returns true if host is using TSC based clocksource */
2613 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2615 /* checked again under seqlock below */
2616 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2619 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2623 /* returns true if host is using TSC based clocksource */
2624 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2627 /* checked again under seqlock below */
2628 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2631 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2637 * Assuming a stable TSC across physical CPUS, and a stable TSC
2638 * across virtual CPUs, the following condition is possible.
2639 * Each numbered line represents an event visible to both
2640 * CPUs at the next numbered event.
2642 * "timespecX" represents host monotonic time. "tscX" represents
2645 * VCPU0 on CPU0 | VCPU1 on CPU1
2647 * 1. read timespec0,tsc0
2648 * 2. | timespec1 = timespec0 + N
2650 * 3. transition to guest | transition to guest
2651 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2652 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2653 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2655 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2658 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2660 * - 0 < N - M => M < N
2662 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2663 * always the case (the difference between two distinct xtime instances
2664 * might be smaller then the difference between corresponding TSC reads,
2665 * when updating guest vcpus pvclock areas).
2667 * To avoid that problem, do not allow visibility of distinct
2668 * system_timestamp/tsc_timestamp values simultaneously: use a master
2669 * copy of host monotonic time values. Update that master copy
2672 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2676 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2678 #ifdef CONFIG_X86_64
2679 struct kvm_arch *ka = &kvm->arch;
2681 bool host_tsc_clocksource, vcpus_matched;
2683 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2684 atomic_read(&kvm->online_vcpus));
2687 * If the host uses TSC clock, then passthrough TSC as stable
2690 host_tsc_clocksource = kvm_get_time_and_clockread(
2691 &ka->master_kernel_ns,
2692 &ka->master_cycle_now);
2694 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2695 && !ka->backwards_tsc_observed
2696 && !ka->boot_vcpu_runs_old_kvmclock;
2698 if (ka->use_master_clock)
2699 atomic_set(&kvm_guest_has_master_clock, 1);
2701 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2702 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2707 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2709 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2712 static void kvm_gen_update_masterclock(struct kvm *kvm)
2714 #ifdef CONFIG_X86_64
2716 struct kvm_vcpu *vcpu;
2717 struct kvm_arch *ka = &kvm->arch;
2718 unsigned long flags;
2720 kvm_hv_invalidate_tsc_page(kvm);
2722 kvm_make_mclock_inprogress_request(kvm);
2724 /* no guest entries from this point */
2725 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2726 pvclock_update_vm_gtod_copy(kvm);
2727 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2729 kvm_for_each_vcpu(i, vcpu, kvm)
2730 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2732 /* guest entries allowed */
2733 kvm_for_each_vcpu(i, vcpu, kvm)
2734 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2738 u64 get_kvmclock_ns(struct kvm *kvm)
2740 struct kvm_arch *ka = &kvm->arch;
2741 struct pvclock_vcpu_time_info hv_clock;
2742 unsigned long flags;
2745 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2746 if (!ka->use_master_clock) {
2747 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2748 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2751 hv_clock.tsc_timestamp = ka->master_cycle_now;
2752 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2753 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2755 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2758 if (__this_cpu_read(cpu_tsc_khz)) {
2759 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2760 &hv_clock.tsc_shift,
2761 &hv_clock.tsc_to_system_mul);
2762 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2764 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2771 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2772 struct gfn_to_hva_cache *cache,
2773 unsigned int offset)
2775 struct kvm_vcpu_arch *vcpu = &v->arch;
2776 struct pvclock_vcpu_time_info guest_hv_clock;
2778 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2779 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2782 /* This VCPU is paused, but it's legal for a guest to read another
2783 * VCPU's kvmclock, so we really have to follow the specification where
2784 * it says that version is odd if data is being modified, and even after
2787 * Version field updates must be kept separate. This is because
2788 * kvm_write_guest_cached might use a "rep movs" instruction, and
2789 * writes within a string instruction are weakly ordered. So there
2790 * are three writes overall.
2792 * As a small optimization, only write the version field in the first
2793 * and third write. The vcpu->pv_time cache is still valid, because the
2794 * version field is the first in the struct.
2796 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2798 if (guest_hv_clock.version & 1)
2799 ++guest_hv_clock.version; /* first time write, random junk */
2801 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2802 kvm_write_guest_offset_cached(v->kvm, cache,
2803 &vcpu->hv_clock, offset,
2804 sizeof(vcpu->hv_clock.version));
2808 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2809 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2811 if (vcpu->pvclock_set_guest_stopped_request) {
2812 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2813 vcpu->pvclock_set_guest_stopped_request = false;
2816 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2818 kvm_write_guest_offset_cached(v->kvm, cache,
2819 &vcpu->hv_clock, offset,
2820 sizeof(vcpu->hv_clock));
2824 vcpu->hv_clock.version++;
2825 kvm_write_guest_offset_cached(v->kvm, cache,
2826 &vcpu->hv_clock, offset,
2827 sizeof(vcpu->hv_clock.version));
2830 static int kvm_guest_time_update(struct kvm_vcpu *v)
2832 unsigned long flags, tgt_tsc_khz;
2833 struct kvm_vcpu_arch *vcpu = &v->arch;
2834 struct kvm_arch *ka = &v->kvm->arch;
2836 u64 tsc_timestamp, host_tsc;
2838 bool use_master_clock;
2844 * If the host uses TSC clock, then passthrough TSC as stable
2847 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2848 use_master_clock = ka->use_master_clock;
2849 if (use_master_clock) {
2850 host_tsc = ka->master_cycle_now;
2851 kernel_ns = ka->master_kernel_ns;
2853 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2855 /* Keep irq disabled to prevent changes to the clock */
2856 local_irq_save(flags);
2857 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2858 if (unlikely(tgt_tsc_khz == 0)) {
2859 local_irq_restore(flags);
2860 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2863 if (!use_master_clock) {
2865 kernel_ns = get_kvmclock_base_ns();
2868 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2871 * We may have to catch up the TSC to match elapsed wall clock
2872 * time for two reasons, even if kvmclock is used.
2873 * 1) CPU could have been running below the maximum TSC rate
2874 * 2) Broken TSC compensation resets the base at each VCPU
2875 * entry to avoid unknown leaps of TSC even when running
2876 * again on the same CPU. This may cause apparent elapsed
2877 * time to disappear, and the guest to stand still or run
2880 if (vcpu->tsc_catchup) {
2881 u64 tsc = compute_guest_tsc(v, kernel_ns);
2882 if (tsc > tsc_timestamp) {
2883 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2884 tsc_timestamp = tsc;
2888 local_irq_restore(flags);
2890 /* With all the info we got, fill in the values */
2892 if (kvm_has_tsc_control)
2893 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
2894 v->arch.l1_tsc_scaling_ratio);
2896 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2897 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2898 &vcpu->hv_clock.tsc_shift,
2899 &vcpu->hv_clock.tsc_to_system_mul);
2900 vcpu->hw_tsc_khz = tgt_tsc_khz;
2903 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2904 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2905 vcpu->last_guest_tsc = tsc_timestamp;
2907 /* If the host uses TSC clocksource, then it is stable */
2909 if (use_master_clock)
2910 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2912 vcpu->hv_clock.flags = pvclock_flags;
2914 if (vcpu->pv_time_enabled)
2915 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2916 if (vcpu->xen.vcpu_info_set)
2917 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2918 offsetof(struct compat_vcpu_info, time));
2919 if (vcpu->xen.vcpu_time_info_set)
2920 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2921 if (v == kvm_get_vcpu(v->kvm, 0))
2922 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2927 * kvmclock updates which are isolated to a given vcpu, such as
2928 * vcpu->cpu migration, should not allow system_timestamp from
2929 * the rest of the vcpus to remain static. Otherwise ntp frequency
2930 * correction applies to one vcpu's system_timestamp but not
2933 * So in those cases, request a kvmclock update for all vcpus.
2934 * We need to rate-limit these requests though, as they can
2935 * considerably slow guests that have a large number of vcpus.
2936 * The time for a remote vcpu to update its kvmclock is bound
2937 * by the delay we use to rate-limit the updates.
2940 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2942 static void kvmclock_update_fn(struct work_struct *work)
2945 struct delayed_work *dwork = to_delayed_work(work);
2946 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2947 kvmclock_update_work);
2948 struct kvm *kvm = container_of(ka, struct kvm, arch);
2949 struct kvm_vcpu *vcpu;
2951 kvm_for_each_vcpu(i, vcpu, kvm) {
2952 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2953 kvm_vcpu_kick(vcpu);
2957 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2959 struct kvm *kvm = v->kvm;
2961 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2962 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2963 KVMCLOCK_UPDATE_DELAY);
2966 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2968 static void kvmclock_sync_fn(struct work_struct *work)
2970 struct delayed_work *dwork = to_delayed_work(work);
2971 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2972 kvmclock_sync_work);
2973 struct kvm *kvm = container_of(ka, struct kvm, arch);
2975 if (!kvmclock_periodic_sync)
2978 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2979 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2980 KVMCLOCK_SYNC_PERIOD);
2984 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2986 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2988 /* McStatusWrEn enabled? */
2989 if (guest_cpuid_is_amd_or_hygon(vcpu))
2990 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2995 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2997 u64 mcg_cap = vcpu->arch.mcg_cap;
2998 unsigned bank_num = mcg_cap & 0xff;
2999 u32 msr = msr_info->index;
3000 u64 data = msr_info->data;
3003 case MSR_IA32_MCG_STATUS:
3004 vcpu->arch.mcg_status = data;
3006 case MSR_IA32_MCG_CTL:
3007 if (!(mcg_cap & MCG_CTL_P) &&
3008 (data || !msr_info->host_initiated))
3010 if (data != 0 && data != ~(u64)0)
3012 vcpu->arch.mcg_ctl = data;
3015 if (msr >= MSR_IA32_MC0_CTL &&
3016 msr < MSR_IA32_MCx_CTL(bank_num)) {
3017 u32 offset = array_index_nospec(
3018 msr - MSR_IA32_MC0_CTL,
3019 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3021 /* only 0 or all 1s can be written to IA32_MCi_CTL
3022 * some Linux kernels though clear bit 10 in bank 4 to
3023 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3024 * this to avoid an uncatched #GP in the guest
3026 if ((offset & 0x3) == 0 &&
3027 data != 0 && (data | (1 << 10)) != ~(u64)0)
3031 if (!msr_info->host_initiated &&
3032 (offset & 0x3) == 1 && data != 0) {
3033 if (!can_set_mci_status(vcpu))
3037 vcpu->arch.mce_banks[offset] = data;
3045 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3047 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3049 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3052 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3054 gpa_t gpa = data & ~0x3f;
3056 /* Bits 4:5 are reserved, Should be zero */
3060 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3061 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3064 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3065 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3068 if (!lapic_in_kernel(vcpu))
3069 return data ? 1 : 0;
3071 vcpu->arch.apf.msr_en_val = data;
3073 if (!kvm_pv_async_pf_enabled(vcpu)) {
3074 kvm_clear_async_pf_completion_queue(vcpu);
3075 kvm_async_pf_hash_reset(vcpu);
3079 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3083 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3084 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3086 kvm_async_pf_wakeup_all(vcpu);
3091 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3093 /* Bits 8-63 are reserved */
3097 if (!lapic_in_kernel(vcpu))
3100 vcpu->arch.apf.msr_int_val = data;
3102 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3107 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3109 vcpu->arch.pv_time_enabled = false;
3110 vcpu->arch.time = 0;
3113 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3115 ++vcpu->stat.tlb_flush;
3116 static_call(kvm_x86_tlb_flush_all)(vcpu);
3119 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3121 ++vcpu->stat.tlb_flush;
3125 * A TLB flush on behalf of the guest is equivalent to
3126 * INVPCID(all), toggling CR4.PGE, etc., which requires
3127 * a forced sync of the shadow page tables. Unload the
3128 * entire MMU here and the subsequent load will sync the
3129 * shadow page tables, and also flush the TLB.
3131 kvm_mmu_unload(vcpu);
3135 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3138 static void record_steal_time(struct kvm_vcpu *vcpu)
3140 struct kvm_host_map map;
3141 struct kvm_steal_time *st;
3143 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3144 kvm_xen_runstate_set_running(vcpu);
3148 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3151 /* -EAGAIN is returned in atomic context so we can just return. */
3152 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3153 &map, &vcpu->arch.st.cache, false))
3157 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3160 * Doing a TLB flush here, on the guest's behalf, can avoid
3163 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3164 u8 st_preempted = xchg(&st->preempted, 0);
3166 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3167 st_preempted & KVM_VCPU_FLUSH_TLB);
3168 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3169 kvm_vcpu_flush_tlb_guest(vcpu);
3174 vcpu->arch.st.preempted = 0;
3176 if (st->version & 1)
3177 st->version += 1; /* first time write, random junk */
3183 st->steal += current->sched_info.run_delay -
3184 vcpu->arch.st.last_steal;
3185 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3191 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3194 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3197 u32 msr = msr_info->index;
3198 u64 data = msr_info->data;
3200 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3201 return kvm_xen_write_hypercall_page(vcpu, data);
3204 case MSR_AMD64_NB_CFG:
3205 case MSR_IA32_UCODE_WRITE:
3206 case MSR_VM_HSAVE_PA:
3207 case MSR_AMD64_PATCH_LOADER:
3208 case MSR_AMD64_BU_CFG2:
3209 case MSR_AMD64_DC_CFG:
3210 case MSR_F15H_EX_CFG:
3213 case MSR_IA32_UCODE_REV:
3214 if (msr_info->host_initiated)
3215 vcpu->arch.microcode_version = data;
3217 case MSR_IA32_ARCH_CAPABILITIES:
3218 if (!msr_info->host_initiated)
3220 vcpu->arch.arch_capabilities = data;
3222 case MSR_IA32_PERF_CAPABILITIES: {
3223 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3225 if (!msr_info->host_initiated)
3227 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3229 if (data & ~msr_ent.data)
3232 vcpu->arch.perf_capabilities = data;
3237 return set_efer(vcpu, msr_info);
3239 data &= ~(u64)0x40; /* ignore flush filter disable */
3240 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3241 data &= ~(u64)0x8; /* ignore TLB cache disable */
3243 /* Handle McStatusWrEn */
3244 if (data == BIT_ULL(18)) {
3245 vcpu->arch.msr_hwcr = data;
3246 } else if (data != 0) {
3247 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3252 case MSR_FAM10H_MMIO_CONF_BASE:
3254 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3259 case 0x200 ... 0x2ff:
3260 return kvm_mtrr_set_msr(vcpu, msr, data);
3261 case MSR_IA32_APICBASE:
3262 return kvm_set_apic_base(vcpu, msr_info);
3263 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3264 return kvm_x2apic_msr_write(vcpu, msr, data);
3265 case MSR_IA32_TSC_DEADLINE:
3266 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3268 case MSR_IA32_TSC_ADJUST:
3269 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3270 if (!msr_info->host_initiated) {
3271 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3272 adjust_tsc_offset_guest(vcpu, adj);
3274 vcpu->arch.ia32_tsc_adjust_msr = data;
3277 case MSR_IA32_MISC_ENABLE:
3278 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3279 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3280 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3282 vcpu->arch.ia32_misc_enable_msr = data;
3283 kvm_update_cpuid_runtime(vcpu);
3285 vcpu->arch.ia32_misc_enable_msr = data;
3288 case MSR_IA32_SMBASE:
3289 if (!msr_info->host_initiated)
3291 vcpu->arch.smbase = data;
3293 case MSR_IA32_POWER_CTL:
3294 vcpu->arch.msr_ia32_power_ctl = data;
3297 if (msr_info->host_initiated) {
3298 kvm_synchronize_tsc(vcpu, data);
3300 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3301 adjust_tsc_offset_guest(vcpu, adj);
3302 vcpu->arch.ia32_tsc_adjust_msr += adj;
3306 if (!msr_info->host_initiated &&
3307 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3310 * KVM supports exposing PT to the guest, but does not support
3311 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3312 * XSAVES/XRSTORS to save/restore PT MSRs.
3314 if (data & ~supported_xss)
3316 vcpu->arch.ia32_xss = data;
3319 if (!msr_info->host_initiated)
3321 vcpu->arch.smi_count = data;
3323 case MSR_KVM_WALL_CLOCK_NEW:
3324 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3327 vcpu->kvm->arch.wall_clock = data;
3328 kvm_write_wall_clock(vcpu->kvm, data, 0);
3330 case MSR_KVM_WALL_CLOCK:
3331 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3334 vcpu->kvm->arch.wall_clock = data;
3335 kvm_write_wall_clock(vcpu->kvm, data, 0);
3337 case MSR_KVM_SYSTEM_TIME_NEW:
3338 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3341 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3343 case MSR_KVM_SYSTEM_TIME:
3344 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3347 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3349 case MSR_KVM_ASYNC_PF_EN:
3350 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3353 if (kvm_pv_enable_async_pf(vcpu, data))
3356 case MSR_KVM_ASYNC_PF_INT:
3357 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3360 if (kvm_pv_enable_async_pf_int(vcpu, data))
3363 case MSR_KVM_ASYNC_PF_ACK:
3364 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3367 vcpu->arch.apf.pageready_pending = false;
3368 kvm_check_async_pf_completion(vcpu);
3371 case MSR_KVM_STEAL_TIME:
3372 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3375 if (unlikely(!sched_info_on()))
3378 if (data & KVM_STEAL_RESERVED_MASK)
3381 vcpu->arch.st.msr_val = data;
3383 if (!(data & KVM_MSR_ENABLED))
3386 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3389 case MSR_KVM_PV_EOI_EN:
3390 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3393 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3397 case MSR_KVM_POLL_CONTROL:
3398 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3401 /* only enable bit supported */
3402 if (data & (-1ULL << 1))
3405 vcpu->arch.msr_kvm_poll_control = data;
3408 case MSR_IA32_MCG_CTL:
3409 case MSR_IA32_MCG_STATUS:
3410 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3411 return set_msr_mce(vcpu, msr_info);
3413 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3414 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3417 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3418 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3419 if (kvm_pmu_is_valid_msr(vcpu, msr))
3420 return kvm_pmu_set_msr(vcpu, msr_info);
3422 if (pr || data != 0)
3423 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3424 "0x%x data 0x%llx\n", msr, data);
3426 case MSR_K7_CLK_CTL:
3428 * Ignore all writes to this no longer documented MSR.
3429 * Writes are only relevant for old K7 processors,
3430 * all pre-dating SVM, but a recommended workaround from
3431 * AMD for these chips. It is possible to specify the
3432 * affected processor models on the command line, hence
3433 * the need to ignore the workaround.
3436 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3437 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3438 case HV_X64_MSR_SYNDBG_OPTIONS:
3439 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3440 case HV_X64_MSR_CRASH_CTL:
3441 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3442 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3443 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3444 case HV_X64_MSR_TSC_EMULATION_STATUS:
3445 return kvm_hv_set_msr_common(vcpu, msr, data,
3446 msr_info->host_initiated);
3447 case MSR_IA32_BBL_CR_CTL3:
3448 /* Drop writes to this legacy MSR -- see rdmsr
3449 * counterpart for further detail.
3451 if (report_ignored_msrs)
3452 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3455 case MSR_AMD64_OSVW_ID_LENGTH:
3456 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3458 vcpu->arch.osvw.length = data;
3460 case MSR_AMD64_OSVW_STATUS:
3461 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3463 vcpu->arch.osvw.status = data;
3465 case MSR_PLATFORM_INFO:
3466 if (!msr_info->host_initiated ||
3467 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3468 cpuid_fault_enabled(vcpu)))
3470 vcpu->arch.msr_platform_info = data;
3472 case MSR_MISC_FEATURES_ENABLES:
3473 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3474 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3475 !supports_cpuid_fault(vcpu)))
3477 vcpu->arch.msr_misc_features_enables = data;
3480 if (kvm_pmu_is_valid_msr(vcpu, msr))
3481 return kvm_pmu_set_msr(vcpu, msr_info);
3482 return KVM_MSR_RET_INVALID;
3486 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3488 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3491 u64 mcg_cap = vcpu->arch.mcg_cap;
3492 unsigned bank_num = mcg_cap & 0xff;
3495 case MSR_IA32_P5_MC_ADDR:
3496 case MSR_IA32_P5_MC_TYPE:
3499 case MSR_IA32_MCG_CAP:
3500 data = vcpu->arch.mcg_cap;
3502 case MSR_IA32_MCG_CTL:
3503 if (!(mcg_cap & MCG_CTL_P) && !host)
3505 data = vcpu->arch.mcg_ctl;
3507 case MSR_IA32_MCG_STATUS:
3508 data = vcpu->arch.mcg_status;
3511 if (msr >= MSR_IA32_MC0_CTL &&
3512 msr < MSR_IA32_MCx_CTL(bank_num)) {
3513 u32 offset = array_index_nospec(
3514 msr - MSR_IA32_MC0_CTL,
3515 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3517 data = vcpu->arch.mce_banks[offset];
3526 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3528 switch (msr_info->index) {
3529 case MSR_IA32_PLATFORM_ID:
3530 case MSR_IA32_EBL_CR_POWERON:
3531 case MSR_IA32_LASTBRANCHFROMIP:
3532 case MSR_IA32_LASTBRANCHTOIP:
3533 case MSR_IA32_LASTINTFROMIP:
3534 case MSR_IA32_LASTINTTOIP:
3536 case MSR_K8_TSEG_ADDR:
3537 case MSR_K8_TSEG_MASK:
3538 case MSR_VM_HSAVE_PA:
3539 case MSR_K8_INT_PENDING_MSG:
3540 case MSR_AMD64_NB_CFG:
3541 case MSR_FAM10H_MMIO_CONF_BASE:
3542 case MSR_AMD64_BU_CFG2:
3543 case MSR_IA32_PERF_CTL:
3544 case MSR_AMD64_DC_CFG:
3545 case MSR_F15H_EX_CFG:
3547 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3548 * limit) MSRs. Just return 0, as we do not want to expose the host
3549 * data here. Do not conditionalize this on CPUID, as KVM does not do
3550 * so for existing CPU-specific MSRs.
3552 case MSR_RAPL_POWER_UNIT:
3553 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3554 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3555 case MSR_PKG_ENERGY_STATUS: /* Total package */
3556 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3559 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3560 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3561 return kvm_pmu_get_msr(vcpu, msr_info);
3562 if (!msr_info->host_initiated)
3566 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3567 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3568 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3569 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3570 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3571 return kvm_pmu_get_msr(vcpu, msr_info);
3574 case MSR_IA32_UCODE_REV:
3575 msr_info->data = vcpu->arch.microcode_version;
3577 case MSR_IA32_ARCH_CAPABILITIES:
3578 if (!msr_info->host_initiated &&
3579 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3581 msr_info->data = vcpu->arch.arch_capabilities;
3583 case MSR_IA32_PERF_CAPABILITIES:
3584 if (!msr_info->host_initiated &&
3585 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3587 msr_info->data = vcpu->arch.perf_capabilities;
3589 case MSR_IA32_POWER_CTL:
3590 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3592 case MSR_IA32_TSC: {
3594 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3595 * even when not intercepted. AMD manual doesn't explicitly
3596 * state this but appears to behave the same.
3598 * On userspace reads and writes, however, we unconditionally
3599 * return L1's TSC value to ensure backwards-compatible
3600 * behavior for migration.
3604 if (msr_info->host_initiated) {
3605 offset = vcpu->arch.l1_tsc_offset;
3606 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3608 offset = vcpu->arch.tsc_offset;
3609 ratio = vcpu->arch.tsc_scaling_ratio;
3612 msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3616 case 0x200 ... 0x2ff:
3617 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3618 case 0xcd: /* fsb frequency */
3622 * MSR_EBC_FREQUENCY_ID
3623 * Conservative value valid for even the basic CPU models.
3624 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3625 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3626 * and 266MHz for model 3, or 4. Set Core Clock
3627 * Frequency to System Bus Frequency Ratio to 1 (bits
3628 * 31:24) even though these are only valid for CPU
3629 * models > 2, however guests may end up dividing or
3630 * multiplying by zero otherwise.
3632 case MSR_EBC_FREQUENCY_ID:
3633 msr_info->data = 1 << 24;
3635 case MSR_IA32_APICBASE:
3636 msr_info->data = kvm_get_apic_base(vcpu);
3638 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3639 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3640 case MSR_IA32_TSC_DEADLINE:
3641 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3643 case MSR_IA32_TSC_ADJUST:
3644 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3646 case MSR_IA32_MISC_ENABLE:
3647 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3649 case MSR_IA32_SMBASE:
3650 if (!msr_info->host_initiated)
3652 msr_info->data = vcpu->arch.smbase;
3655 msr_info->data = vcpu->arch.smi_count;
3657 case MSR_IA32_PERF_STATUS:
3658 /* TSC increment by tick */
3659 msr_info->data = 1000ULL;
3660 /* CPU multiplier */
3661 msr_info->data |= (((uint64_t)4ULL) << 40);
3664 msr_info->data = vcpu->arch.efer;
3666 case MSR_KVM_WALL_CLOCK:
3667 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3670 msr_info->data = vcpu->kvm->arch.wall_clock;
3672 case MSR_KVM_WALL_CLOCK_NEW:
3673 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3676 msr_info->data = vcpu->kvm->arch.wall_clock;
3678 case MSR_KVM_SYSTEM_TIME:
3679 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3682 msr_info->data = vcpu->arch.time;
3684 case MSR_KVM_SYSTEM_TIME_NEW:
3685 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3688 msr_info->data = vcpu->arch.time;
3690 case MSR_KVM_ASYNC_PF_EN:
3691 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3694 msr_info->data = vcpu->arch.apf.msr_en_val;
3696 case MSR_KVM_ASYNC_PF_INT:
3697 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3700 msr_info->data = vcpu->arch.apf.msr_int_val;
3702 case MSR_KVM_ASYNC_PF_ACK:
3703 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3708 case MSR_KVM_STEAL_TIME:
3709 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3712 msr_info->data = vcpu->arch.st.msr_val;
3714 case MSR_KVM_PV_EOI_EN:
3715 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3718 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3720 case MSR_KVM_POLL_CONTROL:
3721 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3724 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3726 case MSR_IA32_P5_MC_ADDR:
3727 case MSR_IA32_P5_MC_TYPE:
3728 case MSR_IA32_MCG_CAP:
3729 case MSR_IA32_MCG_CTL:
3730 case MSR_IA32_MCG_STATUS:
3731 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3732 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3733 msr_info->host_initiated);
3735 if (!msr_info->host_initiated &&
3736 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3738 msr_info->data = vcpu->arch.ia32_xss;
3740 case MSR_K7_CLK_CTL:
3742 * Provide expected ramp-up count for K7. All other
3743 * are set to zero, indicating minimum divisors for
3746 * This prevents guest kernels on AMD host with CPU
3747 * type 6, model 8 and higher from exploding due to
3748 * the rdmsr failing.
3750 msr_info->data = 0x20000000;
3752 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3753 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3754 case HV_X64_MSR_SYNDBG_OPTIONS:
3755 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3756 case HV_X64_MSR_CRASH_CTL:
3757 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3758 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3759 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3760 case HV_X64_MSR_TSC_EMULATION_STATUS:
3761 return kvm_hv_get_msr_common(vcpu,
3762 msr_info->index, &msr_info->data,
3763 msr_info->host_initiated);
3764 case MSR_IA32_BBL_CR_CTL3:
3765 /* This legacy MSR exists but isn't fully documented in current
3766 * silicon. It is however accessed by winxp in very narrow
3767 * scenarios where it sets bit #19, itself documented as
3768 * a "reserved" bit. Best effort attempt to source coherent
3769 * read data here should the balance of the register be
3770 * interpreted by the guest:
3772 * L2 cache control register 3: 64GB range, 256KB size,
3773 * enabled, latency 0x1, configured
3775 msr_info->data = 0xbe702111;
3777 case MSR_AMD64_OSVW_ID_LENGTH:
3778 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3780 msr_info->data = vcpu->arch.osvw.length;
3782 case MSR_AMD64_OSVW_STATUS:
3783 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3785 msr_info->data = vcpu->arch.osvw.status;
3787 case MSR_PLATFORM_INFO:
3788 if (!msr_info->host_initiated &&
3789 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3791 msr_info->data = vcpu->arch.msr_platform_info;
3793 case MSR_MISC_FEATURES_ENABLES:
3794 msr_info->data = vcpu->arch.msr_misc_features_enables;
3797 msr_info->data = vcpu->arch.msr_hwcr;
3800 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3801 return kvm_pmu_get_msr(vcpu, msr_info);
3802 return KVM_MSR_RET_INVALID;
3806 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3809 * Read or write a bunch of msrs. All parameters are kernel addresses.
3811 * @return number of msrs set successfully.
3813 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3814 struct kvm_msr_entry *entries,
3815 int (*do_msr)(struct kvm_vcpu *vcpu,
3816 unsigned index, u64 *data))
3820 for (i = 0; i < msrs->nmsrs; ++i)
3821 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3828 * Read or write a bunch of msrs. Parameters are user addresses.
3830 * @return number of msrs set successfully.
3832 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3833 int (*do_msr)(struct kvm_vcpu *vcpu,
3834 unsigned index, u64 *data),
3837 struct kvm_msrs msrs;
3838 struct kvm_msr_entry *entries;
3843 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3847 if (msrs.nmsrs >= MAX_IO_MSRS)
3850 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3851 entries = memdup_user(user_msrs->entries, size);
3852 if (IS_ERR(entries)) {
3853 r = PTR_ERR(entries);
3857 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3862 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3873 static inline bool kvm_can_mwait_in_guest(void)
3875 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3876 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3877 boot_cpu_has(X86_FEATURE_ARAT);
3880 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3881 struct kvm_cpuid2 __user *cpuid_arg)
3883 struct kvm_cpuid2 cpuid;
3887 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3890 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3895 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3901 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3906 case KVM_CAP_IRQCHIP:
3908 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3909 case KVM_CAP_SET_TSS_ADDR:
3910 case KVM_CAP_EXT_CPUID:
3911 case KVM_CAP_EXT_EMUL_CPUID:
3912 case KVM_CAP_CLOCKSOURCE:
3914 case KVM_CAP_NOP_IO_DELAY:
3915 case KVM_CAP_MP_STATE:
3916 case KVM_CAP_SYNC_MMU:
3917 case KVM_CAP_USER_NMI:
3918 case KVM_CAP_REINJECT_CONTROL:
3919 case KVM_CAP_IRQ_INJECT_STATUS:
3920 case KVM_CAP_IOEVENTFD:
3921 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3923 case KVM_CAP_PIT_STATE2:
3924 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3925 case KVM_CAP_VCPU_EVENTS:
3926 case KVM_CAP_HYPERV:
3927 case KVM_CAP_HYPERV_VAPIC:
3928 case KVM_CAP_HYPERV_SPIN:
3929 case KVM_CAP_HYPERV_SYNIC:
3930 case KVM_CAP_HYPERV_SYNIC2:
3931 case KVM_CAP_HYPERV_VP_INDEX:
3932 case KVM_CAP_HYPERV_EVENTFD:
3933 case KVM_CAP_HYPERV_TLBFLUSH:
3934 case KVM_CAP_HYPERV_SEND_IPI:
3935 case KVM_CAP_HYPERV_CPUID:
3936 case KVM_CAP_SYS_HYPERV_CPUID:
3937 case KVM_CAP_PCI_SEGMENT:
3938 case KVM_CAP_DEBUGREGS:
3939 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3941 case KVM_CAP_ASYNC_PF:
3942 case KVM_CAP_ASYNC_PF_INT:
3943 case KVM_CAP_GET_TSC_KHZ:
3944 case KVM_CAP_KVMCLOCK_CTRL:
3945 case KVM_CAP_READONLY_MEM:
3946 case KVM_CAP_HYPERV_TIME:
3947 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3948 case KVM_CAP_TSC_DEADLINE_TIMER:
3949 case KVM_CAP_DISABLE_QUIRKS:
3950 case KVM_CAP_SET_BOOT_CPU_ID:
3951 case KVM_CAP_SPLIT_IRQCHIP:
3952 case KVM_CAP_IMMEDIATE_EXIT:
3953 case KVM_CAP_PMU_EVENT_FILTER:
3954 case KVM_CAP_GET_MSR_FEATURES:
3955 case KVM_CAP_MSR_PLATFORM_INFO:
3956 case KVM_CAP_EXCEPTION_PAYLOAD:
3957 case KVM_CAP_SET_GUEST_DEBUG:
3958 case KVM_CAP_LAST_CPU:
3959 case KVM_CAP_X86_USER_SPACE_MSR:
3960 case KVM_CAP_X86_MSR_FILTER:
3961 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3962 #ifdef CONFIG_X86_SGX_KVM
3963 case KVM_CAP_SGX_ATTRIBUTE:
3965 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
3968 case KVM_CAP_SET_GUEST_DEBUG2:
3969 return KVM_GUESTDBG_VALID_MASK;
3970 #ifdef CONFIG_KVM_XEN
3971 case KVM_CAP_XEN_HVM:
3972 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
3973 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
3974 KVM_XEN_HVM_CONFIG_SHARED_INFO;
3975 if (sched_info_on())
3976 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
3979 case KVM_CAP_SYNC_REGS:
3980 r = KVM_SYNC_X86_VALID_FIELDS;
3982 case KVM_CAP_ADJUST_CLOCK:
3983 r = KVM_CLOCK_TSC_STABLE;
3985 case KVM_CAP_X86_DISABLE_EXITS:
3986 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3987 KVM_X86_DISABLE_EXITS_CSTATE;
3988 if(kvm_can_mwait_in_guest())
3989 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3991 case KVM_CAP_X86_SMM:
3992 /* SMBASE is usually relocated above 1M on modern chipsets,
3993 * and SMM handlers might indeed rely on 4G segment limits,
3994 * so do not report SMM to be available if real mode is
3995 * emulated via vm86 mode. Still, do not go to great lengths
3996 * to avoid userspace's usage of the feature, because it is a
3997 * fringe case that is not enabled except via specific settings
3998 * of the module parameters.
4000 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4003 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4005 case KVM_CAP_NR_VCPUS:
4006 r = KVM_SOFT_MAX_VCPUS;
4008 case KVM_CAP_MAX_VCPUS:
4011 case KVM_CAP_MAX_VCPU_ID:
4012 r = KVM_MAX_VCPU_ID;
4014 case KVM_CAP_PV_MMU: /* obsolete */
4018 r = KVM_MAX_MCE_BANKS;
4021 r = boot_cpu_has(X86_FEATURE_XSAVE);
4023 case KVM_CAP_TSC_CONTROL:
4024 r = kvm_has_tsc_control;
4026 case KVM_CAP_X2APIC_API:
4027 r = KVM_X2APIC_API_VALID_FLAGS;
4029 case KVM_CAP_NESTED_STATE:
4030 r = kvm_x86_ops.nested_ops->get_state ?
4031 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4033 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4034 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4036 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4037 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4039 case KVM_CAP_SMALLER_MAXPHYADDR:
4040 r = (int) allow_smaller_maxphyaddr;
4042 case KVM_CAP_STEAL_TIME:
4043 r = sched_info_on();
4045 case KVM_CAP_X86_BUS_LOCK_EXIT:
4046 if (kvm_has_bus_lock_exit)
4047 r = KVM_BUS_LOCK_DETECTION_OFF |
4048 KVM_BUS_LOCK_DETECTION_EXIT;
4059 long kvm_arch_dev_ioctl(struct file *filp,
4060 unsigned int ioctl, unsigned long arg)
4062 void __user *argp = (void __user *)arg;
4066 case KVM_GET_MSR_INDEX_LIST: {
4067 struct kvm_msr_list __user *user_msr_list = argp;
4068 struct kvm_msr_list msr_list;
4072 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4075 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4076 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4079 if (n < msr_list.nmsrs)
4082 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4083 num_msrs_to_save * sizeof(u32)))
4085 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4087 num_emulated_msrs * sizeof(u32)))
4092 case KVM_GET_SUPPORTED_CPUID:
4093 case KVM_GET_EMULATED_CPUID: {
4094 struct kvm_cpuid2 __user *cpuid_arg = argp;
4095 struct kvm_cpuid2 cpuid;
4098 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4101 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4107 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4112 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4114 if (copy_to_user(argp, &kvm_mce_cap_supported,
4115 sizeof(kvm_mce_cap_supported)))
4119 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4120 struct kvm_msr_list __user *user_msr_list = argp;
4121 struct kvm_msr_list msr_list;
4125 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4128 msr_list.nmsrs = num_msr_based_features;
4129 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4132 if (n < msr_list.nmsrs)
4135 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4136 num_msr_based_features * sizeof(u32)))
4142 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4144 case KVM_GET_SUPPORTED_HV_CPUID:
4145 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4155 static void wbinvd_ipi(void *garbage)
4160 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4162 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4165 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4167 /* Address WBINVD may be executed by guest */
4168 if (need_emulate_wbinvd(vcpu)) {
4169 if (static_call(kvm_x86_has_wbinvd_exit)())
4170 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4171 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4172 smp_call_function_single(vcpu->cpu,
4173 wbinvd_ipi, NULL, 1);
4176 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4178 /* Save host pkru register if supported */
4179 vcpu->arch.host_pkru = read_pkru();
4181 /* Apply any externally detected TSC adjustments (due to suspend) */
4182 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4183 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4184 vcpu->arch.tsc_offset_adjustment = 0;
4185 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4188 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4189 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4190 rdtsc() - vcpu->arch.last_host_tsc;
4192 mark_tsc_unstable("KVM discovered backwards TSC");
4194 if (kvm_check_tsc_unstable()) {
4195 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4196 vcpu->arch.last_guest_tsc);
4197 kvm_vcpu_write_tsc_offset(vcpu, offset);
4198 vcpu->arch.tsc_catchup = 1;
4201 if (kvm_lapic_hv_timer_in_use(vcpu))
4202 kvm_lapic_restart_hv_timer(vcpu);
4205 * On a host with synchronized TSC, there is no need to update
4206 * kvmclock on vcpu->cpu migration
4208 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4209 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4210 if (vcpu->cpu != cpu)
4211 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4215 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4218 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4220 struct kvm_host_map map;
4221 struct kvm_steal_time *st;
4223 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4226 if (vcpu->arch.st.preempted)
4229 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4230 &vcpu->arch.st.cache, true))
4234 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4236 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4238 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4241 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4245 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4246 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4249 * Take the srcu lock as memslots will be accessed to check the gfn
4250 * cache generation against the memslots generation.
4252 idx = srcu_read_lock(&vcpu->kvm->srcu);
4253 if (kvm_xen_msr_enabled(vcpu->kvm))
4254 kvm_xen_runstate_set_preempted(vcpu);
4256 kvm_steal_time_set_preempted(vcpu);
4257 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4259 static_call(kvm_x86_vcpu_put)(vcpu);
4260 vcpu->arch.last_host_tsc = rdtsc();
4262 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4263 * on every vmexit, but if not, we might have a stale dr6 from the
4264 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4269 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4270 struct kvm_lapic_state *s)
4272 if (vcpu->arch.apicv_active)
4273 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4275 return kvm_apic_get_state(vcpu, s);
4278 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4279 struct kvm_lapic_state *s)
4283 r = kvm_apic_set_state(vcpu, s);
4286 update_cr8_intercept(vcpu);
4291 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4294 * We can accept userspace's request for interrupt injection
4295 * as long as we have a place to store the interrupt number.
4296 * The actual injection will happen when the CPU is able to
4297 * deliver the interrupt.
4299 if (kvm_cpu_has_extint(vcpu))
4302 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4303 return (!lapic_in_kernel(vcpu) ||
4304 kvm_apic_accept_pic_intr(vcpu));
4307 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4309 return kvm_arch_interrupt_allowed(vcpu) &&
4310 kvm_cpu_accept_dm_intr(vcpu);
4313 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4314 struct kvm_interrupt *irq)
4316 if (irq->irq >= KVM_NR_INTERRUPTS)
4319 if (!irqchip_in_kernel(vcpu->kvm)) {
4320 kvm_queue_interrupt(vcpu, irq->irq, false);
4321 kvm_make_request(KVM_REQ_EVENT, vcpu);
4326 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4327 * fail for in-kernel 8259.
4329 if (pic_in_kernel(vcpu->kvm))
4332 if (vcpu->arch.pending_external_vector != -1)
4335 vcpu->arch.pending_external_vector = irq->irq;
4336 kvm_make_request(KVM_REQ_EVENT, vcpu);
4340 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4342 kvm_inject_nmi(vcpu);
4347 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4349 kvm_make_request(KVM_REQ_SMI, vcpu);
4354 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4355 struct kvm_tpr_access_ctl *tac)
4359 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4363 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4367 unsigned bank_num = mcg_cap & 0xff, bank;
4370 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4372 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4375 vcpu->arch.mcg_cap = mcg_cap;
4376 /* Init IA32_MCG_CTL to all 1s */
4377 if (mcg_cap & MCG_CTL_P)
4378 vcpu->arch.mcg_ctl = ~(u64)0;
4379 /* Init IA32_MCi_CTL to all 1s */
4380 for (bank = 0; bank < bank_num; bank++)
4381 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4383 static_call(kvm_x86_setup_mce)(vcpu);
4388 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4389 struct kvm_x86_mce *mce)
4391 u64 mcg_cap = vcpu->arch.mcg_cap;
4392 unsigned bank_num = mcg_cap & 0xff;
4393 u64 *banks = vcpu->arch.mce_banks;
4395 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4398 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4399 * reporting is disabled
4401 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4402 vcpu->arch.mcg_ctl != ~(u64)0)
4404 banks += 4 * mce->bank;
4406 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4407 * reporting is disabled for the bank
4409 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4411 if (mce->status & MCI_STATUS_UC) {
4412 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4413 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4414 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4417 if (banks[1] & MCI_STATUS_VAL)
4418 mce->status |= MCI_STATUS_OVER;
4419 banks[2] = mce->addr;
4420 banks[3] = mce->misc;
4421 vcpu->arch.mcg_status = mce->mcg_status;
4422 banks[1] = mce->status;
4423 kvm_queue_exception(vcpu, MC_VECTOR);
4424 } else if (!(banks[1] & MCI_STATUS_VAL)
4425 || !(banks[1] & MCI_STATUS_UC)) {
4426 if (banks[1] & MCI_STATUS_VAL)
4427 mce->status |= MCI_STATUS_OVER;
4428 banks[2] = mce->addr;
4429 banks[3] = mce->misc;
4430 banks[1] = mce->status;
4432 banks[1] |= MCI_STATUS_OVER;
4436 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4437 struct kvm_vcpu_events *events)
4441 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4445 * In guest mode, payload delivery should be deferred,
4446 * so that the L1 hypervisor can intercept #PF before
4447 * CR2 is modified (or intercept #DB before DR6 is
4448 * modified under nVMX). Unless the per-VM capability,
4449 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4450 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4451 * opportunistically defer the exception payload, deliver it if the
4452 * capability hasn't been requested before processing a
4453 * KVM_GET_VCPU_EVENTS.
4455 if (!vcpu->kvm->arch.exception_payload_enabled &&
4456 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4457 kvm_deliver_exception_payload(vcpu);
4460 * The API doesn't provide the instruction length for software
4461 * exceptions, so don't report them. As long as the guest RIP
4462 * isn't advanced, we should expect to encounter the exception
4465 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4466 events->exception.injected = 0;
4467 events->exception.pending = 0;
4469 events->exception.injected = vcpu->arch.exception.injected;
4470 events->exception.pending = vcpu->arch.exception.pending;
4472 * For ABI compatibility, deliberately conflate
4473 * pending and injected exceptions when
4474 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4476 if (!vcpu->kvm->arch.exception_payload_enabled)
4477 events->exception.injected |=
4478 vcpu->arch.exception.pending;
4480 events->exception.nr = vcpu->arch.exception.nr;
4481 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4482 events->exception.error_code = vcpu->arch.exception.error_code;
4483 events->exception_has_payload = vcpu->arch.exception.has_payload;
4484 events->exception_payload = vcpu->arch.exception.payload;
4486 events->interrupt.injected =
4487 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4488 events->interrupt.nr = vcpu->arch.interrupt.nr;
4489 events->interrupt.soft = 0;
4490 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4492 events->nmi.injected = vcpu->arch.nmi_injected;
4493 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4494 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4495 events->nmi.pad = 0;
4497 events->sipi_vector = 0; /* never valid when reporting to user space */
4499 events->smi.smm = is_smm(vcpu);
4500 events->smi.pending = vcpu->arch.smi_pending;
4501 events->smi.smm_inside_nmi =
4502 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4503 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4505 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4506 | KVM_VCPUEVENT_VALID_SHADOW
4507 | KVM_VCPUEVENT_VALID_SMM);
4508 if (vcpu->kvm->arch.exception_payload_enabled)
4509 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4511 memset(&events->reserved, 0, sizeof(events->reserved));
4514 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4516 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4517 struct kvm_vcpu_events *events)
4519 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4520 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4521 | KVM_VCPUEVENT_VALID_SHADOW
4522 | KVM_VCPUEVENT_VALID_SMM
4523 | KVM_VCPUEVENT_VALID_PAYLOAD))
4526 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4527 if (!vcpu->kvm->arch.exception_payload_enabled)
4529 if (events->exception.pending)
4530 events->exception.injected = 0;
4532 events->exception_has_payload = 0;
4534 events->exception.pending = 0;
4535 events->exception_has_payload = 0;
4538 if ((events->exception.injected || events->exception.pending) &&
4539 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4542 /* INITs are latched while in SMM */
4543 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4544 (events->smi.smm || events->smi.pending) &&
4545 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4549 vcpu->arch.exception.injected = events->exception.injected;
4550 vcpu->arch.exception.pending = events->exception.pending;
4551 vcpu->arch.exception.nr = events->exception.nr;
4552 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4553 vcpu->arch.exception.error_code = events->exception.error_code;
4554 vcpu->arch.exception.has_payload = events->exception_has_payload;
4555 vcpu->arch.exception.payload = events->exception_payload;
4557 vcpu->arch.interrupt.injected = events->interrupt.injected;
4558 vcpu->arch.interrupt.nr = events->interrupt.nr;
4559 vcpu->arch.interrupt.soft = events->interrupt.soft;
4560 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4561 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4562 events->interrupt.shadow);
4564 vcpu->arch.nmi_injected = events->nmi.injected;
4565 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4566 vcpu->arch.nmi_pending = events->nmi.pending;
4567 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4569 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4570 lapic_in_kernel(vcpu))
4571 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4573 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4574 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4575 if (events->smi.smm)
4576 vcpu->arch.hflags |= HF_SMM_MASK;
4578 vcpu->arch.hflags &= ~HF_SMM_MASK;
4579 kvm_smm_changed(vcpu);
4582 vcpu->arch.smi_pending = events->smi.pending;
4584 if (events->smi.smm) {
4585 if (events->smi.smm_inside_nmi)
4586 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4588 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4591 if (lapic_in_kernel(vcpu)) {
4592 if (events->smi.latched_init)
4593 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4595 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4599 kvm_make_request(KVM_REQ_EVENT, vcpu);
4604 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4605 struct kvm_debugregs *dbgregs)
4609 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4610 kvm_get_dr(vcpu, 6, &val);
4612 dbgregs->dr7 = vcpu->arch.dr7;
4614 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4617 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4618 struct kvm_debugregs *dbgregs)
4623 if (!kvm_dr6_valid(dbgregs->dr6))
4625 if (!kvm_dr7_valid(dbgregs->dr7))
4628 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4629 kvm_update_dr0123(vcpu);
4630 vcpu->arch.dr6 = dbgregs->dr6;
4631 vcpu->arch.dr7 = dbgregs->dr7;
4632 kvm_update_dr7(vcpu);
4637 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4639 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4641 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4642 u64 xstate_bv = xsave->header.xfeatures;
4646 * Copy legacy XSAVE area, to avoid complications with CPUID
4647 * leaves 0 and 1 in the loop below.
4649 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4652 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4653 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4656 * Copy each region from the possibly compacted offset to the
4657 * non-compacted offset.
4659 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4661 u64 xfeature_mask = valid & -valid;
4662 int xfeature_nr = fls64(xfeature_mask) - 1;
4663 void *src = get_xsave_addr(xsave, xfeature_nr);
4666 u32 size, offset, ecx, edx;
4667 cpuid_count(XSTATE_CPUID, xfeature_nr,
4668 &size, &offset, &ecx, &edx);
4669 if (xfeature_nr == XFEATURE_PKRU)
4670 memcpy(dest + offset, &vcpu->arch.pkru,
4671 sizeof(vcpu->arch.pkru));
4673 memcpy(dest + offset, src, size);
4677 valid -= xfeature_mask;
4681 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4683 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4684 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4688 * Copy legacy XSAVE area, to avoid complications with CPUID
4689 * leaves 0 and 1 in the loop below.
4691 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4693 /* Set XSTATE_BV and possibly XCOMP_BV. */
4694 xsave->header.xfeatures = xstate_bv;
4695 if (boot_cpu_has(X86_FEATURE_XSAVES))
4696 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4699 * Copy each region from the non-compacted offset to the
4700 * possibly compacted offset.
4702 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4704 u64 xfeature_mask = valid & -valid;
4705 int xfeature_nr = fls64(xfeature_mask) - 1;
4706 void *dest = get_xsave_addr(xsave, xfeature_nr);
4709 u32 size, offset, ecx, edx;
4710 cpuid_count(XSTATE_CPUID, xfeature_nr,
4711 &size, &offset, &ecx, &edx);
4712 if (xfeature_nr == XFEATURE_PKRU)
4713 memcpy(&vcpu->arch.pkru, src + offset,
4714 sizeof(vcpu->arch.pkru));
4716 memcpy(dest, src + offset, size);
4719 valid -= xfeature_mask;
4723 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4724 struct kvm_xsave *guest_xsave)
4726 if (!vcpu->arch.guest_fpu)
4729 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4730 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4731 fill_xsave((u8 *) guest_xsave->region, vcpu);
4733 memcpy(guest_xsave->region,
4734 &vcpu->arch.guest_fpu->state.fxsave,
4735 sizeof(struct fxregs_state));
4736 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4737 XFEATURE_MASK_FPSSE;
4741 #define XSAVE_MXCSR_OFFSET 24
4743 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4744 struct kvm_xsave *guest_xsave)
4749 if (!vcpu->arch.guest_fpu)
4752 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4753 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4755 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4757 * Here we allow setting states that are not present in
4758 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4759 * with old userspace.
4761 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4763 load_xsave(vcpu, (u8 *)guest_xsave->region);
4765 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4766 mxcsr & ~mxcsr_feature_mask)
4768 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4769 guest_xsave->region, sizeof(struct fxregs_state));
4774 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4775 struct kvm_xcrs *guest_xcrs)
4777 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4778 guest_xcrs->nr_xcrs = 0;
4782 guest_xcrs->nr_xcrs = 1;
4783 guest_xcrs->flags = 0;
4784 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4785 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4788 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4789 struct kvm_xcrs *guest_xcrs)
4793 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4796 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4799 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4800 /* Only support XCR0 currently */
4801 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4802 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4803 guest_xcrs->xcrs[i].value);
4812 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4813 * stopped by the hypervisor. This function will be called from the host only.
4814 * EINVAL is returned when the host attempts to set the flag for a guest that
4815 * does not support pv clocks.
4817 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4819 if (!vcpu->arch.pv_time_enabled)
4821 vcpu->arch.pvclock_set_guest_stopped_request = true;
4822 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4826 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4827 struct kvm_enable_cap *cap)
4830 uint16_t vmcs_version;
4831 void __user *user_ptr;
4837 case KVM_CAP_HYPERV_SYNIC2:
4842 case KVM_CAP_HYPERV_SYNIC:
4843 if (!irqchip_in_kernel(vcpu->kvm))
4845 return kvm_hv_activate_synic(vcpu, cap->cap ==
4846 KVM_CAP_HYPERV_SYNIC2);
4847 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4848 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4850 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4852 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4853 if (copy_to_user(user_ptr, &vmcs_version,
4854 sizeof(vmcs_version)))
4858 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4859 if (!kvm_x86_ops.enable_direct_tlbflush)
4862 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4864 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4865 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4866 if (vcpu->arch.pv_cpuid.enforce)
4867 kvm_update_pv_runtime(vcpu);
4875 long kvm_arch_vcpu_ioctl(struct file *filp,
4876 unsigned int ioctl, unsigned long arg)
4878 struct kvm_vcpu *vcpu = filp->private_data;
4879 void __user *argp = (void __user *)arg;
4882 struct kvm_lapic_state *lapic;
4883 struct kvm_xsave *xsave;
4884 struct kvm_xcrs *xcrs;
4892 case KVM_GET_LAPIC: {
4894 if (!lapic_in_kernel(vcpu))
4896 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4897 GFP_KERNEL_ACCOUNT);
4902 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4906 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4911 case KVM_SET_LAPIC: {
4913 if (!lapic_in_kernel(vcpu))
4915 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4916 if (IS_ERR(u.lapic)) {
4917 r = PTR_ERR(u.lapic);
4921 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4924 case KVM_INTERRUPT: {
4925 struct kvm_interrupt irq;
4928 if (copy_from_user(&irq, argp, sizeof(irq)))
4930 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4934 r = kvm_vcpu_ioctl_nmi(vcpu);
4938 r = kvm_vcpu_ioctl_smi(vcpu);
4941 case KVM_SET_CPUID: {
4942 struct kvm_cpuid __user *cpuid_arg = argp;
4943 struct kvm_cpuid cpuid;
4946 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4948 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4951 case KVM_SET_CPUID2: {
4952 struct kvm_cpuid2 __user *cpuid_arg = argp;
4953 struct kvm_cpuid2 cpuid;
4956 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4958 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4959 cpuid_arg->entries);
4962 case KVM_GET_CPUID2: {
4963 struct kvm_cpuid2 __user *cpuid_arg = argp;
4964 struct kvm_cpuid2 cpuid;
4967 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4969 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4970 cpuid_arg->entries);
4974 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4979 case KVM_GET_MSRS: {
4980 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4981 r = msr_io(vcpu, argp, do_get_msr, 1);
4982 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4985 case KVM_SET_MSRS: {
4986 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4987 r = msr_io(vcpu, argp, do_set_msr, 0);
4988 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4991 case KVM_TPR_ACCESS_REPORTING: {
4992 struct kvm_tpr_access_ctl tac;
4995 if (copy_from_user(&tac, argp, sizeof(tac)))
4997 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5001 if (copy_to_user(argp, &tac, sizeof(tac)))
5006 case KVM_SET_VAPIC_ADDR: {
5007 struct kvm_vapic_addr va;
5011 if (!lapic_in_kernel(vcpu))
5014 if (copy_from_user(&va, argp, sizeof(va)))
5016 idx = srcu_read_lock(&vcpu->kvm->srcu);
5017 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5018 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5021 case KVM_X86_SETUP_MCE: {
5025 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5027 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5030 case KVM_X86_SET_MCE: {
5031 struct kvm_x86_mce mce;
5034 if (copy_from_user(&mce, argp, sizeof(mce)))
5036 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5039 case KVM_GET_VCPU_EVENTS: {
5040 struct kvm_vcpu_events events;
5042 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5045 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5050 case KVM_SET_VCPU_EVENTS: {
5051 struct kvm_vcpu_events events;
5054 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5057 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5060 case KVM_GET_DEBUGREGS: {
5061 struct kvm_debugregs dbgregs;
5063 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5066 if (copy_to_user(argp, &dbgregs,
5067 sizeof(struct kvm_debugregs)))
5072 case KVM_SET_DEBUGREGS: {
5073 struct kvm_debugregs dbgregs;
5076 if (copy_from_user(&dbgregs, argp,
5077 sizeof(struct kvm_debugregs)))
5080 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5083 case KVM_GET_XSAVE: {
5084 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5089 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5092 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5097 case KVM_SET_XSAVE: {
5098 u.xsave = memdup_user(argp, sizeof(*u.xsave));
5099 if (IS_ERR(u.xsave)) {
5100 r = PTR_ERR(u.xsave);
5104 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5107 case KVM_GET_XCRS: {
5108 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5113 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5116 if (copy_to_user(argp, u.xcrs,
5117 sizeof(struct kvm_xcrs)))
5122 case KVM_SET_XCRS: {
5123 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5124 if (IS_ERR(u.xcrs)) {
5125 r = PTR_ERR(u.xcrs);
5129 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5132 case KVM_SET_TSC_KHZ: {
5136 user_tsc_khz = (u32)arg;
5138 if (kvm_has_tsc_control &&
5139 user_tsc_khz >= kvm_max_guest_tsc_khz)
5142 if (user_tsc_khz == 0)
5143 user_tsc_khz = tsc_khz;
5145 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5150 case KVM_GET_TSC_KHZ: {
5151 r = vcpu->arch.virtual_tsc_khz;
5154 case KVM_KVMCLOCK_CTRL: {
5155 r = kvm_set_guest_paused(vcpu);
5158 case KVM_ENABLE_CAP: {
5159 struct kvm_enable_cap cap;
5162 if (copy_from_user(&cap, argp, sizeof(cap)))
5164 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5167 case KVM_GET_NESTED_STATE: {
5168 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5172 if (!kvm_x86_ops.nested_ops->get_state)
5175 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5177 if (get_user(user_data_size, &user_kvm_nested_state->size))
5180 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5185 if (r > user_data_size) {
5186 if (put_user(r, &user_kvm_nested_state->size))
5196 case KVM_SET_NESTED_STATE: {
5197 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5198 struct kvm_nested_state kvm_state;
5202 if (!kvm_x86_ops.nested_ops->set_state)
5206 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5210 if (kvm_state.size < sizeof(kvm_state))
5213 if (kvm_state.flags &
5214 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5215 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5216 | KVM_STATE_NESTED_GIF_SET))
5219 /* nested_run_pending implies guest_mode. */
5220 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5221 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5224 idx = srcu_read_lock(&vcpu->kvm->srcu);
5225 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5226 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5229 case KVM_GET_SUPPORTED_HV_CPUID:
5230 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5232 #ifdef CONFIG_KVM_XEN
5233 case KVM_XEN_VCPU_GET_ATTR: {
5234 struct kvm_xen_vcpu_attr xva;
5237 if (copy_from_user(&xva, argp, sizeof(xva)))
5239 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5240 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5244 case KVM_XEN_VCPU_SET_ATTR: {
5245 struct kvm_xen_vcpu_attr xva;
5248 if (copy_from_user(&xva, argp, sizeof(xva)))
5250 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5264 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5266 return VM_FAULT_SIGBUS;
5269 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5273 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5275 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5279 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5282 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5285 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5286 unsigned long kvm_nr_mmu_pages)
5288 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5291 mutex_lock(&kvm->slots_lock);
5293 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5294 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5296 mutex_unlock(&kvm->slots_lock);
5300 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5302 return kvm->arch.n_max_mmu_pages;
5305 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5307 struct kvm_pic *pic = kvm->arch.vpic;
5311 switch (chip->chip_id) {
5312 case KVM_IRQCHIP_PIC_MASTER:
5313 memcpy(&chip->chip.pic, &pic->pics[0],
5314 sizeof(struct kvm_pic_state));
5316 case KVM_IRQCHIP_PIC_SLAVE:
5317 memcpy(&chip->chip.pic, &pic->pics[1],
5318 sizeof(struct kvm_pic_state));
5320 case KVM_IRQCHIP_IOAPIC:
5321 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5330 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5332 struct kvm_pic *pic = kvm->arch.vpic;
5336 switch (chip->chip_id) {
5337 case KVM_IRQCHIP_PIC_MASTER:
5338 spin_lock(&pic->lock);
5339 memcpy(&pic->pics[0], &chip->chip.pic,
5340 sizeof(struct kvm_pic_state));
5341 spin_unlock(&pic->lock);
5343 case KVM_IRQCHIP_PIC_SLAVE:
5344 spin_lock(&pic->lock);
5345 memcpy(&pic->pics[1], &chip->chip.pic,
5346 sizeof(struct kvm_pic_state));
5347 spin_unlock(&pic->lock);
5349 case KVM_IRQCHIP_IOAPIC:
5350 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5356 kvm_pic_update_irq(pic);
5360 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5362 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5364 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5366 mutex_lock(&kps->lock);
5367 memcpy(ps, &kps->channels, sizeof(*ps));
5368 mutex_unlock(&kps->lock);
5372 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5375 struct kvm_pit *pit = kvm->arch.vpit;
5377 mutex_lock(&pit->pit_state.lock);
5378 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5379 for (i = 0; i < 3; i++)
5380 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5381 mutex_unlock(&pit->pit_state.lock);
5385 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5387 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5388 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5389 sizeof(ps->channels));
5390 ps->flags = kvm->arch.vpit->pit_state.flags;
5391 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5392 memset(&ps->reserved, 0, sizeof(ps->reserved));
5396 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5400 u32 prev_legacy, cur_legacy;
5401 struct kvm_pit *pit = kvm->arch.vpit;
5403 mutex_lock(&pit->pit_state.lock);
5404 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5405 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5406 if (!prev_legacy && cur_legacy)
5408 memcpy(&pit->pit_state.channels, &ps->channels,
5409 sizeof(pit->pit_state.channels));
5410 pit->pit_state.flags = ps->flags;
5411 for (i = 0; i < 3; i++)
5412 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5414 mutex_unlock(&pit->pit_state.lock);
5418 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5419 struct kvm_reinject_control *control)
5421 struct kvm_pit *pit = kvm->arch.vpit;
5423 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5424 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5425 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5427 mutex_lock(&pit->pit_state.lock);
5428 kvm_pit_set_reinject(pit, control->pit_reinject);
5429 mutex_unlock(&pit->pit_state.lock);
5434 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5438 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5439 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5440 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5443 struct kvm_vcpu *vcpu;
5446 kvm_for_each_vcpu(i, vcpu, kvm)
5447 kvm_vcpu_kick(vcpu);
5450 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5453 if (!irqchip_in_kernel(kvm))
5456 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5457 irq_event->irq, irq_event->level,
5462 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5463 struct kvm_enable_cap *cap)
5471 case KVM_CAP_DISABLE_QUIRKS:
5472 kvm->arch.disabled_quirks = cap->args[0];
5475 case KVM_CAP_SPLIT_IRQCHIP: {
5476 mutex_lock(&kvm->lock);
5478 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5479 goto split_irqchip_unlock;
5481 if (irqchip_in_kernel(kvm))
5482 goto split_irqchip_unlock;
5483 if (kvm->created_vcpus)
5484 goto split_irqchip_unlock;
5485 r = kvm_setup_empty_irq_routing(kvm);
5487 goto split_irqchip_unlock;
5488 /* Pairs with irqchip_in_kernel. */
5490 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5491 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5493 split_irqchip_unlock:
5494 mutex_unlock(&kvm->lock);
5497 case KVM_CAP_X2APIC_API:
5499 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5502 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5503 kvm->arch.x2apic_format = true;
5504 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5505 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5509 case KVM_CAP_X86_DISABLE_EXITS:
5511 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5514 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5515 kvm_can_mwait_in_guest())
5516 kvm->arch.mwait_in_guest = true;
5517 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5518 kvm->arch.hlt_in_guest = true;
5519 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5520 kvm->arch.pause_in_guest = true;
5521 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5522 kvm->arch.cstate_in_guest = true;
5525 case KVM_CAP_MSR_PLATFORM_INFO:
5526 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5529 case KVM_CAP_EXCEPTION_PAYLOAD:
5530 kvm->arch.exception_payload_enabled = cap->args[0];
5533 case KVM_CAP_X86_USER_SPACE_MSR:
5534 kvm->arch.user_space_msr_mask = cap->args[0];
5537 case KVM_CAP_X86_BUS_LOCK_EXIT:
5539 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5542 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5543 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5546 if (kvm_has_bus_lock_exit &&
5547 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5548 kvm->arch.bus_lock_detection_enabled = true;
5551 #ifdef CONFIG_X86_SGX_KVM
5552 case KVM_CAP_SGX_ATTRIBUTE: {
5553 unsigned long allowed_attributes = 0;
5555 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5559 /* KVM only supports the PROVISIONKEY privileged attribute. */
5560 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5561 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5562 kvm->arch.sgx_provisioning_allowed = true;
5568 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5570 if (kvm_x86_ops.vm_copy_enc_context_from)
5571 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5580 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5582 struct kvm_x86_msr_filter *msr_filter;
5584 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5588 msr_filter->default_allow = default_allow;
5592 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5599 for (i = 0; i < msr_filter->count; i++)
5600 kfree(msr_filter->ranges[i].bitmap);
5605 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5606 struct kvm_msr_filter_range *user_range)
5608 unsigned long *bitmap = NULL;
5611 if (!user_range->nmsrs)
5614 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5617 if (!user_range->flags)
5620 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5621 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5624 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5626 return PTR_ERR(bitmap);
5628 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5629 .flags = user_range->flags,
5630 .base = user_range->base,
5631 .nmsrs = user_range->nmsrs,
5635 msr_filter->count++;
5639 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5641 struct kvm_msr_filter __user *user_msr_filter = argp;
5642 struct kvm_x86_msr_filter *new_filter, *old_filter;
5643 struct kvm_msr_filter filter;
5649 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5652 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5653 empty &= !filter.ranges[i].nmsrs;
5655 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5656 if (empty && !default_allow)
5659 new_filter = kvm_alloc_msr_filter(default_allow);
5663 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5664 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5666 kvm_free_msr_filter(new_filter);
5671 mutex_lock(&kvm->lock);
5673 /* The per-VM filter is protected by kvm->lock... */
5674 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5676 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5677 synchronize_srcu(&kvm->srcu);
5679 kvm_free_msr_filter(old_filter);
5681 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5682 mutex_unlock(&kvm->lock);
5687 long kvm_arch_vm_ioctl(struct file *filp,
5688 unsigned int ioctl, unsigned long arg)
5690 struct kvm *kvm = filp->private_data;
5691 void __user *argp = (void __user *)arg;
5694 * This union makes it completely explicit to gcc-3.x
5695 * that these two variables' stack usage should be
5696 * combined, not added together.
5699 struct kvm_pit_state ps;
5700 struct kvm_pit_state2 ps2;
5701 struct kvm_pit_config pit_config;
5705 case KVM_SET_TSS_ADDR:
5706 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5708 case KVM_SET_IDENTITY_MAP_ADDR: {
5711 mutex_lock(&kvm->lock);
5713 if (kvm->created_vcpus)
5714 goto set_identity_unlock;
5716 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5717 goto set_identity_unlock;
5718 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5719 set_identity_unlock:
5720 mutex_unlock(&kvm->lock);
5723 case KVM_SET_NR_MMU_PAGES:
5724 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5726 case KVM_GET_NR_MMU_PAGES:
5727 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5729 case KVM_CREATE_IRQCHIP: {
5730 mutex_lock(&kvm->lock);
5733 if (irqchip_in_kernel(kvm))
5734 goto create_irqchip_unlock;
5737 if (kvm->created_vcpus)
5738 goto create_irqchip_unlock;
5740 r = kvm_pic_init(kvm);
5742 goto create_irqchip_unlock;
5744 r = kvm_ioapic_init(kvm);
5746 kvm_pic_destroy(kvm);
5747 goto create_irqchip_unlock;
5750 r = kvm_setup_default_irq_routing(kvm);
5752 kvm_ioapic_destroy(kvm);
5753 kvm_pic_destroy(kvm);
5754 goto create_irqchip_unlock;
5756 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5758 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5759 create_irqchip_unlock:
5760 mutex_unlock(&kvm->lock);
5763 case KVM_CREATE_PIT:
5764 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5766 case KVM_CREATE_PIT2:
5768 if (copy_from_user(&u.pit_config, argp,
5769 sizeof(struct kvm_pit_config)))
5772 mutex_lock(&kvm->lock);
5775 goto create_pit_unlock;
5777 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5781 mutex_unlock(&kvm->lock);
5783 case KVM_GET_IRQCHIP: {
5784 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5785 struct kvm_irqchip *chip;
5787 chip = memdup_user(argp, sizeof(*chip));
5794 if (!irqchip_kernel(kvm))
5795 goto get_irqchip_out;
5796 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5798 goto get_irqchip_out;
5800 if (copy_to_user(argp, chip, sizeof(*chip)))
5801 goto get_irqchip_out;
5807 case KVM_SET_IRQCHIP: {
5808 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5809 struct kvm_irqchip *chip;
5811 chip = memdup_user(argp, sizeof(*chip));
5818 if (!irqchip_kernel(kvm))
5819 goto set_irqchip_out;
5820 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5827 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5830 if (!kvm->arch.vpit)
5832 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5836 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5843 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5845 mutex_lock(&kvm->lock);
5847 if (!kvm->arch.vpit)
5849 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5851 mutex_unlock(&kvm->lock);
5854 case KVM_GET_PIT2: {
5856 if (!kvm->arch.vpit)
5858 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5862 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5867 case KVM_SET_PIT2: {
5869 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5871 mutex_lock(&kvm->lock);
5873 if (!kvm->arch.vpit)
5875 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5877 mutex_unlock(&kvm->lock);
5880 case KVM_REINJECT_CONTROL: {
5881 struct kvm_reinject_control control;
5883 if (copy_from_user(&control, argp, sizeof(control)))
5886 if (!kvm->arch.vpit)
5888 r = kvm_vm_ioctl_reinject(kvm, &control);
5891 case KVM_SET_BOOT_CPU_ID:
5893 mutex_lock(&kvm->lock);
5894 if (kvm->created_vcpus)
5897 kvm->arch.bsp_vcpu_id = arg;
5898 mutex_unlock(&kvm->lock);
5900 #ifdef CONFIG_KVM_XEN
5901 case KVM_XEN_HVM_CONFIG: {
5902 struct kvm_xen_hvm_config xhc;
5904 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5906 r = kvm_xen_hvm_config(kvm, &xhc);
5909 case KVM_XEN_HVM_GET_ATTR: {
5910 struct kvm_xen_hvm_attr xha;
5913 if (copy_from_user(&xha, argp, sizeof(xha)))
5915 r = kvm_xen_hvm_get_attr(kvm, &xha);
5916 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5920 case KVM_XEN_HVM_SET_ATTR: {
5921 struct kvm_xen_hvm_attr xha;
5924 if (copy_from_user(&xha, argp, sizeof(xha)))
5926 r = kvm_xen_hvm_set_attr(kvm, &xha);
5930 case KVM_SET_CLOCK: {
5931 struct kvm_arch *ka = &kvm->arch;
5932 struct kvm_clock_data user_ns;
5936 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5945 * TODO: userspace has to take care of races with VCPU_RUN, so
5946 * kvm_gen_update_masterclock() can be cut down to locked
5947 * pvclock_update_vm_gtod_copy().
5949 kvm_gen_update_masterclock(kvm);
5952 * This pairs with kvm_guest_time_update(): when masterclock is
5953 * in use, we use master_kernel_ns + kvmclock_offset to set
5954 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
5955 * is slightly ahead) here we risk going negative on unsigned
5956 * 'system_time' when 'user_ns.clock' is very small.
5958 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
5959 if (kvm->arch.use_master_clock)
5960 now_ns = ka->master_kernel_ns;
5962 now_ns = get_kvmclock_base_ns();
5963 ka->kvmclock_offset = user_ns.clock - now_ns;
5964 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
5966 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5969 case KVM_GET_CLOCK: {
5970 struct kvm_clock_data user_ns;
5973 now_ns = get_kvmclock_ns(kvm);
5974 user_ns.clock = now_ns;
5975 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5976 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5979 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5984 case KVM_MEMORY_ENCRYPT_OP: {
5986 if (kvm_x86_ops.mem_enc_op)
5987 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
5990 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5991 struct kvm_enc_region region;
5994 if (copy_from_user(®ion, argp, sizeof(region)))
5998 if (kvm_x86_ops.mem_enc_reg_region)
5999 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
6002 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6003 struct kvm_enc_region region;
6006 if (copy_from_user(®ion, argp, sizeof(region)))
6010 if (kvm_x86_ops.mem_enc_unreg_region)
6011 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
6014 case KVM_HYPERV_EVENTFD: {
6015 struct kvm_hyperv_eventfd hvevfd;
6018 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6020 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6023 case KVM_SET_PMU_EVENT_FILTER:
6024 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6026 case KVM_X86_SET_MSR_FILTER:
6027 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6036 static void kvm_init_msr_list(void)
6038 struct x86_pmu_capability x86_pmu;
6042 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6043 "Please update the fixed PMCs in msrs_to_saved_all[]");
6045 perf_get_x86_pmu_capability(&x86_pmu);
6047 num_msrs_to_save = 0;
6048 num_emulated_msrs = 0;
6049 num_msr_based_features = 0;
6051 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6052 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6056 * Even MSRs that are valid in the host may not be exposed
6057 * to the guests in some cases.
6059 switch (msrs_to_save_all[i]) {
6060 case MSR_IA32_BNDCFGS:
6061 if (!kvm_mpx_supported())
6065 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6066 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6069 case MSR_IA32_UMWAIT_CONTROL:
6070 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6073 case MSR_IA32_RTIT_CTL:
6074 case MSR_IA32_RTIT_STATUS:
6075 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6078 case MSR_IA32_RTIT_CR3_MATCH:
6079 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6080 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6083 case MSR_IA32_RTIT_OUTPUT_BASE:
6084 case MSR_IA32_RTIT_OUTPUT_MASK:
6085 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6086 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6087 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6090 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6091 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6092 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6093 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6096 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6097 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6098 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6101 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6102 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6103 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6110 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6113 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6114 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6117 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6120 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6121 struct kvm_msr_entry msr;
6123 msr.index = msr_based_features_all[i];
6124 if (kvm_get_msr_feature(&msr))
6127 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6131 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6139 if (!(lapic_in_kernel(vcpu) &&
6140 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6141 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6152 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6159 if (!(lapic_in_kernel(vcpu) &&
6160 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6162 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6164 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6174 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6175 struct kvm_segment *var, int seg)
6177 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6180 void kvm_get_segment(struct kvm_vcpu *vcpu,
6181 struct kvm_segment *var, int seg)
6183 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6186 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6187 struct x86_exception *exception)
6191 BUG_ON(!mmu_is_nested(vcpu));
6193 /* NPT walks are always user-walks */
6194 access |= PFERR_USER_MASK;
6195 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6200 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6201 struct x86_exception *exception)
6203 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6204 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6206 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6208 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6209 struct x86_exception *exception)
6211 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6212 access |= PFERR_FETCH_MASK;
6213 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6216 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6217 struct x86_exception *exception)
6219 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6220 access |= PFERR_WRITE_MASK;
6221 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6223 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6225 /* uses this to access any guest's mapped memory without checking CPL */
6226 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6227 struct x86_exception *exception)
6229 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6232 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6233 struct kvm_vcpu *vcpu, u32 access,
6234 struct x86_exception *exception)
6237 int r = X86EMUL_CONTINUE;
6240 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6242 unsigned offset = addr & (PAGE_SIZE-1);
6243 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6246 if (gpa == UNMAPPED_GVA)
6247 return X86EMUL_PROPAGATE_FAULT;
6248 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6251 r = X86EMUL_IO_NEEDED;
6263 /* used for instruction fetching */
6264 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6265 gva_t addr, void *val, unsigned int bytes,
6266 struct x86_exception *exception)
6268 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6269 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6273 /* Inline kvm_read_guest_virt_helper for speed. */
6274 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6276 if (unlikely(gpa == UNMAPPED_GVA))
6277 return X86EMUL_PROPAGATE_FAULT;
6279 offset = addr & (PAGE_SIZE-1);
6280 if (WARN_ON(offset + bytes > PAGE_SIZE))
6281 bytes = (unsigned)PAGE_SIZE - offset;
6282 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6284 if (unlikely(ret < 0))
6285 return X86EMUL_IO_NEEDED;
6287 return X86EMUL_CONTINUE;
6290 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6291 gva_t addr, void *val, unsigned int bytes,
6292 struct x86_exception *exception)
6294 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6297 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6298 * is returned, but our callers are not ready for that and they blindly
6299 * call kvm_inject_page_fault. Ensure that they at least do not leak
6300 * uninitialized kernel stack memory into cr2 and error code.
6302 memset(exception, 0, sizeof(*exception));
6303 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6306 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6308 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6309 gva_t addr, void *val, unsigned int bytes,
6310 struct x86_exception *exception, bool system)
6312 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6315 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6316 access |= PFERR_USER_MASK;
6318 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6321 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6322 unsigned long addr, void *val, unsigned int bytes)
6324 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6325 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6327 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6330 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6331 struct kvm_vcpu *vcpu, u32 access,
6332 struct x86_exception *exception)
6335 int r = X86EMUL_CONTINUE;
6338 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6341 unsigned offset = addr & (PAGE_SIZE-1);
6342 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6345 if (gpa == UNMAPPED_GVA)
6346 return X86EMUL_PROPAGATE_FAULT;
6347 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6349 r = X86EMUL_IO_NEEDED;
6361 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6362 unsigned int bytes, struct x86_exception *exception,
6365 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6366 u32 access = PFERR_WRITE_MASK;
6368 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6369 access |= PFERR_USER_MASK;
6371 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6375 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6376 unsigned int bytes, struct x86_exception *exception)
6378 /* kvm_write_guest_virt_system can pull in tons of pages. */
6379 vcpu->arch.l1tf_flush_l1d = true;
6381 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6382 PFERR_WRITE_MASK, exception);
6384 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6386 int handle_ud(struct kvm_vcpu *vcpu)
6388 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6389 int emul_type = EMULTYPE_TRAP_UD;
6390 char sig[5]; /* ud2; .ascii "kvm" */
6391 struct x86_exception e;
6393 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6396 if (force_emulation_prefix &&
6397 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6398 sig, sizeof(sig), &e) == 0 &&
6399 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6400 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6401 emul_type = EMULTYPE_TRAP_UD_FORCED;
6404 return kvm_emulate_instruction(vcpu, emul_type);
6406 EXPORT_SYMBOL_GPL(handle_ud);
6408 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6409 gpa_t gpa, bool write)
6411 /* For APIC access vmexit */
6412 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6415 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6416 trace_vcpu_match_mmio(gva, gpa, write, true);
6423 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6424 gpa_t *gpa, struct x86_exception *exception,
6427 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6428 | (write ? PFERR_WRITE_MASK : 0);
6431 * currently PKRU is only applied to ept enabled guest so
6432 * there is no pkey in EPT page table for L1 guest or EPT
6433 * shadow page table for L2 guest.
6435 if (vcpu_match_mmio_gva(vcpu, gva)
6436 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6437 vcpu->arch.mmio_access, 0, access)) {
6438 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6439 (gva & (PAGE_SIZE - 1));
6440 trace_vcpu_match_mmio(gva, *gpa, write, false);
6444 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6446 if (*gpa == UNMAPPED_GVA)
6449 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6452 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6453 const void *val, int bytes)
6457 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6460 kvm_page_track_write(vcpu, gpa, val, bytes);
6464 struct read_write_emulator_ops {
6465 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6467 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6468 void *val, int bytes);
6469 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6470 int bytes, void *val);
6471 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6472 void *val, int bytes);
6476 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6478 if (vcpu->mmio_read_completed) {
6479 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6480 vcpu->mmio_fragments[0].gpa, val);
6481 vcpu->mmio_read_completed = 0;
6488 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6489 void *val, int bytes)
6491 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6494 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6495 void *val, int bytes)
6497 return emulator_write_phys(vcpu, gpa, val, bytes);
6500 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6502 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6503 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6506 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6507 void *val, int bytes)
6509 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6510 return X86EMUL_IO_NEEDED;
6513 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6514 void *val, int bytes)
6516 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6518 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6519 return X86EMUL_CONTINUE;
6522 static const struct read_write_emulator_ops read_emultor = {
6523 .read_write_prepare = read_prepare,
6524 .read_write_emulate = read_emulate,
6525 .read_write_mmio = vcpu_mmio_read,
6526 .read_write_exit_mmio = read_exit_mmio,
6529 static const struct read_write_emulator_ops write_emultor = {
6530 .read_write_emulate = write_emulate,
6531 .read_write_mmio = write_mmio,
6532 .read_write_exit_mmio = write_exit_mmio,
6536 static int emulator_read_write_onepage(unsigned long addr, void *val,
6538 struct x86_exception *exception,
6539 struct kvm_vcpu *vcpu,
6540 const struct read_write_emulator_ops *ops)
6544 bool write = ops->write;
6545 struct kvm_mmio_fragment *frag;
6546 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6549 * If the exit was due to a NPF we may already have a GPA.
6550 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6551 * Note, this cannot be used on string operations since string
6552 * operation using rep will only have the initial GPA from the NPF
6555 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6556 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6557 gpa = ctxt->gpa_val;
6558 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6560 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6562 return X86EMUL_PROPAGATE_FAULT;
6565 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6566 return X86EMUL_CONTINUE;
6569 * Is this MMIO handled locally?
6571 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6572 if (handled == bytes)
6573 return X86EMUL_CONTINUE;
6579 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6580 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6584 return X86EMUL_CONTINUE;
6587 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6589 void *val, unsigned int bytes,
6590 struct x86_exception *exception,
6591 const struct read_write_emulator_ops *ops)
6593 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6597 if (ops->read_write_prepare &&
6598 ops->read_write_prepare(vcpu, val, bytes))
6599 return X86EMUL_CONTINUE;
6601 vcpu->mmio_nr_fragments = 0;
6603 /* Crossing a page boundary? */
6604 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6607 now = -addr & ~PAGE_MASK;
6608 rc = emulator_read_write_onepage(addr, val, now, exception,
6611 if (rc != X86EMUL_CONTINUE)
6614 if (ctxt->mode != X86EMUL_MODE_PROT64)
6620 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6622 if (rc != X86EMUL_CONTINUE)
6625 if (!vcpu->mmio_nr_fragments)
6628 gpa = vcpu->mmio_fragments[0].gpa;
6630 vcpu->mmio_needed = 1;
6631 vcpu->mmio_cur_fragment = 0;
6633 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6634 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6635 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6636 vcpu->run->mmio.phys_addr = gpa;
6638 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6641 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6645 struct x86_exception *exception)
6647 return emulator_read_write(ctxt, addr, val, bytes,
6648 exception, &read_emultor);
6651 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6655 struct x86_exception *exception)
6657 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6658 exception, &write_emultor);
6661 #define CMPXCHG_TYPE(t, ptr, old, new) \
6662 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6664 #ifdef CONFIG_X86_64
6665 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6667 # define CMPXCHG64(ptr, old, new) \
6668 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6671 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6676 struct x86_exception *exception)
6678 struct kvm_host_map map;
6679 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6685 /* guests cmpxchg8b have to be emulated atomically */
6686 if (bytes > 8 || (bytes & (bytes - 1)))
6689 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6691 if (gpa == UNMAPPED_GVA ||
6692 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6696 * Emulate the atomic as a straight write to avoid #AC if SLD is
6697 * enabled in the host and the access splits a cache line.
6699 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6700 page_line_mask = ~(cache_line_size() - 1);
6702 page_line_mask = PAGE_MASK;
6704 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6707 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6710 kaddr = map.hva + offset_in_page(gpa);
6714 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6717 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6720 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6723 exchanged = CMPXCHG64(kaddr, old, new);
6729 kvm_vcpu_unmap(vcpu, &map, true);
6732 return X86EMUL_CMPXCHG_FAILED;
6734 kvm_page_track_write(vcpu, gpa, new, bytes);
6736 return X86EMUL_CONTINUE;
6739 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6741 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6744 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6748 for (i = 0; i < vcpu->arch.pio.count; i++) {
6749 if (vcpu->arch.pio.in)
6750 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6751 vcpu->arch.pio.size, pd);
6753 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6754 vcpu->arch.pio.port, vcpu->arch.pio.size,
6758 pd += vcpu->arch.pio.size;
6763 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6764 unsigned short port, void *val,
6765 unsigned int count, bool in)
6767 vcpu->arch.pio.port = port;
6768 vcpu->arch.pio.in = in;
6769 vcpu->arch.pio.count = count;
6770 vcpu->arch.pio.size = size;
6772 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6773 vcpu->arch.pio.count = 0;
6777 vcpu->run->exit_reason = KVM_EXIT_IO;
6778 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6779 vcpu->run->io.size = size;
6780 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6781 vcpu->run->io.count = count;
6782 vcpu->run->io.port = port;
6787 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6788 unsigned short port, void *val, unsigned int count)
6792 if (vcpu->arch.pio.count)
6795 memset(vcpu->arch.pio_data, 0, size * count);
6797 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6800 memcpy(val, vcpu->arch.pio_data, size * count);
6801 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6802 vcpu->arch.pio.count = 0;
6809 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6810 int size, unsigned short port, void *val,
6813 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6817 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6818 unsigned short port, const void *val,
6821 memcpy(vcpu->arch.pio_data, val, size * count);
6822 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6823 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6826 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6827 int size, unsigned short port,
6828 const void *val, unsigned int count)
6830 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6833 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6835 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6838 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6840 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6843 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6845 if (!need_emulate_wbinvd(vcpu))
6846 return X86EMUL_CONTINUE;
6848 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6849 int cpu = get_cpu();
6851 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6852 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6853 wbinvd_ipi, NULL, 1);
6855 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6858 return X86EMUL_CONTINUE;
6861 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6863 kvm_emulate_wbinvd_noskip(vcpu);
6864 return kvm_skip_emulated_instruction(vcpu);
6866 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6870 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6872 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6875 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6876 unsigned long *dest)
6878 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6881 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6882 unsigned long value)
6885 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6888 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6890 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6893 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6895 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6896 unsigned long value;
6900 value = kvm_read_cr0(vcpu);
6903 value = vcpu->arch.cr2;
6906 value = kvm_read_cr3(vcpu);
6909 value = kvm_read_cr4(vcpu);
6912 value = kvm_get_cr8(vcpu);
6915 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6922 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6924 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6929 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6932 vcpu->arch.cr2 = val;
6935 res = kvm_set_cr3(vcpu, val);
6938 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6941 res = kvm_set_cr8(vcpu, val);
6944 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6951 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6953 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
6956 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6958 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
6961 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6963 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
6966 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6968 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
6971 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6973 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
6976 static unsigned long emulator_get_cached_segment_base(
6977 struct x86_emulate_ctxt *ctxt, int seg)
6979 return get_segment_base(emul_to_vcpu(ctxt), seg);
6982 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6983 struct desc_struct *desc, u32 *base3,
6986 struct kvm_segment var;
6988 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6989 *selector = var.selector;
6992 memset(desc, 0, sizeof(*desc));
7000 set_desc_limit(desc, var.limit);
7001 set_desc_base(desc, (unsigned long)var.base);
7002 #ifdef CONFIG_X86_64
7004 *base3 = var.base >> 32;
7006 desc->type = var.type;
7008 desc->dpl = var.dpl;
7009 desc->p = var.present;
7010 desc->avl = var.avl;
7018 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7019 struct desc_struct *desc, u32 base3,
7022 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7023 struct kvm_segment var;
7025 var.selector = selector;
7026 var.base = get_desc_base(desc);
7027 #ifdef CONFIG_X86_64
7028 var.base |= ((u64)base3) << 32;
7030 var.limit = get_desc_limit(desc);
7032 var.limit = (var.limit << 12) | 0xfff;
7033 var.type = desc->type;
7034 var.dpl = desc->dpl;
7039 var.avl = desc->avl;
7040 var.present = desc->p;
7041 var.unusable = !var.present;
7044 kvm_set_segment(vcpu, &var, seg);
7048 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7049 u32 msr_index, u64 *pdata)
7051 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7054 r = kvm_get_msr(vcpu, msr_index, pdata);
7056 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7057 /* Bounce to user space */
7058 return X86EMUL_IO_NEEDED;
7064 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7065 u32 msr_index, u64 data)
7067 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7070 r = kvm_set_msr(vcpu, msr_index, data);
7072 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7073 /* Bounce to user space */
7074 return X86EMUL_IO_NEEDED;
7080 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7082 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7084 return vcpu->arch.smbase;
7087 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7089 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7091 vcpu->arch.smbase = smbase;
7094 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7097 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7100 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7101 u32 pmc, u64 *pdata)
7103 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7106 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7108 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7111 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7112 struct x86_instruction_info *info,
7113 enum x86_intercept_stage stage)
7115 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7119 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7120 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7123 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7126 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7128 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7131 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7133 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7136 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7138 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7141 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7143 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7146 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7148 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7151 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7153 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7156 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7158 return emul_to_vcpu(ctxt)->arch.hflags;
7161 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
7163 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7165 vcpu->arch.hflags = emul_flags;
7166 kvm_mmu_reset_context(vcpu);
7169 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
7170 const char *smstate)
7172 return static_call(kvm_x86_pre_leave_smm)(emul_to_vcpu(ctxt), smstate);
7175 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
7177 kvm_smm_changed(emul_to_vcpu(ctxt));
7180 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7182 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7185 static const struct x86_emulate_ops emulate_ops = {
7186 .read_gpr = emulator_read_gpr,
7187 .write_gpr = emulator_write_gpr,
7188 .read_std = emulator_read_std,
7189 .write_std = emulator_write_std,
7190 .read_phys = kvm_read_guest_phys_system,
7191 .fetch = kvm_fetch_guest_virt,
7192 .read_emulated = emulator_read_emulated,
7193 .write_emulated = emulator_write_emulated,
7194 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7195 .invlpg = emulator_invlpg,
7196 .pio_in_emulated = emulator_pio_in_emulated,
7197 .pio_out_emulated = emulator_pio_out_emulated,
7198 .get_segment = emulator_get_segment,
7199 .set_segment = emulator_set_segment,
7200 .get_cached_segment_base = emulator_get_cached_segment_base,
7201 .get_gdt = emulator_get_gdt,
7202 .get_idt = emulator_get_idt,
7203 .set_gdt = emulator_set_gdt,
7204 .set_idt = emulator_set_idt,
7205 .get_cr = emulator_get_cr,
7206 .set_cr = emulator_set_cr,
7207 .cpl = emulator_get_cpl,
7208 .get_dr = emulator_get_dr,
7209 .set_dr = emulator_set_dr,
7210 .get_smbase = emulator_get_smbase,
7211 .set_smbase = emulator_set_smbase,
7212 .set_msr = emulator_set_msr,
7213 .get_msr = emulator_get_msr,
7214 .check_pmc = emulator_check_pmc,
7215 .read_pmc = emulator_read_pmc,
7216 .halt = emulator_halt,
7217 .wbinvd = emulator_wbinvd,
7218 .fix_hypercall = emulator_fix_hypercall,
7219 .intercept = emulator_intercept,
7220 .get_cpuid = emulator_get_cpuid,
7221 .guest_has_long_mode = emulator_guest_has_long_mode,
7222 .guest_has_movbe = emulator_guest_has_movbe,
7223 .guest_has_fxsr = emulator_guest_has_fxsr,
7224 .set_nmi_mask = emulator_set_nmi_mask,
7225 .get_hflags = emulator_get_hflags,
7226 .set_hflags = emulator_set_hflags,
7227 .pre_leave_smm = emulator_pre_leave_smm,
7228 .post_leave_smm = emulator_post_leave_smm,
7229 .set_xcr = emulator_set_xcr,
7232 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7234 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7236 * an sti; sti; sequence only disable interrupts for the first
7237 * instruction. So, if the last instruction, be it emulated or
7238 * not, left the system with the INT_STI flag enabled, it
7239 * means that the last instruction is an sti. We should not
7240 * leave the flag on in this case. The same goes for mov ss
7242 if (int_shadow & mask)
7244 if (unlikely(int_shadow || mask)) {
7245 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7247 kvm_make_request(KVM_REQ_EVENT, vcpu);
7251 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7253 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7254 if (ctxt->exception.vector == PF_VECTOR)
7255 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7257 if (ctxt->exception.error_code_valid)
7258 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7259 ctxt->exception.error_code);
7261 kvm_queue_exception(vcpu, ctxt->exception.vector);
7265 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7267 struct x86_emulate_ctxt *ctxt;
7269 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7271 pr_err("kvm: failed to allocate vcpu's emulator\n");
7276 ctxt->ops = &emulate_ops;
7277 vcpu->arch.emulate_ctxt = ctxt;
7282 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7284 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7287 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7289 ctxt->gpa_available = false;
7290 ctxt->eflags = kvm_get_rflags(vcpu);
7291 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7293 ctxt->eip = kvm_rip_read(vcpu);
7294 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7295 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7296 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7297 cs_db ? X86EMUL_MODE_PROT32 :
7298 X86EMUL_MODE_PROT16;
7299 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7300 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7301 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7303 ctxt->interruptibility = 0;
7304 ctxt->have_exception = false;
7305 ctxt->exception.vector = -1;
7306 ctxt->perm_ok = false;
7308 init_decode_cache(ctxt);
7309 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7312 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7314 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7317 init_emulate_ctxt(vcpu);
7321 ctxt->_eip = ctxt->eip + inc_eip;
7322 ret = emulate_int_real(ctxt, irq);
7324 if (ret != X86EMUL_CONTINUE) {
7325 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7327 ctxt->eip = ctxt->_eip;
7328 kvm_rip_write(vcpu, ctxt->eip);
7329 kvm_set_rflags(vcpu, ctxt->eflags);
7332 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7334 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7336 ++vcpu->stat.insn_emulation_fail;
7337 trace_kvm_emulate_insn_failed(vcpu);
7339 if (emulation_type & EMULTYPE_VMWARE_GP) {
7340 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7344 if (emulation_type & EMULTYPE_SKIP) {
7345 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7346 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7347 vcpu->run->internal.ndata = 0;
7351 kvm_queue_exception(vcpu, UD_VECTOR);
7353 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7354 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7355 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7356 vcpu->run->internal.ndata = 0;
7363 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7364 bool write_fault_to_shadow_pgtable,
7367 gpa_t gpa = cr2_or_gpa;
7370 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7373 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7374 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7377 if (!vcpu->arch.mmu->direct_map) {
7379 * Write permission should be allowed since only
7380 * write access need to be emulated.
7382 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7385 * If the mapping is invalid in guest, let cpu retry
7386 * it to generate fault.
7388 if (gpa == UNMAPPED_GVA)
7393 * Do not retry the unhandleable instruction if it faults on the
7394 * readonly host memory, otherwise it will goto a infinite loop:
7395 * retry instruction -> write #PF -> emulation fail -> retry
7396 * instruction -> ...
7398 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7401 * If the instruction failed on the error pfn, it can not be fixed,
7402 * report the error to userspace.
7404 if (is_error_noslot_pfn(pfn))
7407 kvm_release_pfn_clean(pfn);
7409 /* The instructions are well-emulated on direct mmu. */
7410 if (vcpu->arch.mmu->direct_map) {
7411 unsigned int indirect_shadow_pages;
7413 write_lock(&vcpu->kvm->mmu_lock);
7414 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7415 write_unlock(&vcpu->kvm->mmu_lock);
7417 if (indirect_shadow_pages)
7418 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7424 * if emulation was due to access to shadowed page table
7425 * and it failed try to unshadow page and re-enter the
7426 * guest to let CPU execute the instruction.
7428 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7431 * If the access faults on its page table, it can not
7432 * be fixed by unprotecting shadow page and it should
7433 * be reported to userspace.
7435 return !write_fault_to_shadow_pgtable;
7438 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7439 gpa_t cr2_or_gpa, int emulation_type)
7441 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7442 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7444 last_retry_eip = vcpu->arch.last_retry_eip;
7445 last_retry_addr = vcpu->arch.last_retry_addr;
7448 * If the emulation is caused by #PF and it is non-page_table
7449 * writing instruction, it means the VM-EXIT is caused by shadow
7450 * page protected, we can zap the shadow page and retry this
7451 * instruction directly.
7453 * Note: if the guest uses a non-page-table modifying instruction
7454 * on the PDE that points to the instruction, then we will unmap
7455 * the instruction and go to an infinite loop. So, we cache the
7456 * last retried eip and the last fault address, if we meet the eip
7457 * and the address again, we can break out of the potential infinite
7460 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7462 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7465 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7466 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7469 if (x86_page_table_writing_insn(ctxt))
7472 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7475 vcpu->arch.last_retry_eip = ctxt->eip;
7476 vcpu->arch.last_retry_addr = cr2_or_gpa;
7478 if (!vcpu->arch.mmu->direct_map)
7479 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7481 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7486 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7487 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7489 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7491 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7492 /* This is a good place to trace that we are exiting SMM. */
7493 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7495 /* Process a latched INIT or SMI, if any. */
7496 kvm_make_request(KVM_REQ_EVENT, vcpu);
7499 kvm_mmu_reset_context(vcpu);
7502 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7511 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7512 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7517 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7519 struct kvm_run *kvm_run = vcpu->run;
7521 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7522 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7523 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7524 kvm_run->debug.arch.exception = DB_VECTOR;
7525 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7528 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7532 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7534 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7537 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7542 * rflags is the old, "raw" value of the flags. The new value has
7543 * not been saved yet.
7545 * This is correct even for TF set by the guest, because "the
7546 * processor will not generate this exception after the instruction
7547 * that sets the TF flag".
7549 if (unlikely(rflags & X86_EFLAGS_TF))
7550 r = kvm_vcpu_do_singlestep(vcpu);
7553 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7555 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7557 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7558 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7559 struct kvm_run *kvm_run = vcpu->run;
7560 unsigned long eip = kvm_get_linear_rip(vcpu);
7561 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7562 vcpu->arch.guest_debug_dr7,
7566 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7567 kvm_run->debug.arch.pc = eip;
7568 kvm_run->debug.arch.exception = DB_VECTOR;
7569 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7575 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7576 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7577 unsigned long eip = kvm_get_linear_rip(vcpu);
7578 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7583 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7592 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7594 switch (ctxt->opcode_len) {
7601 case 0xe6: /* OUT */
7605 case 0x6c: /* INS */
7607 case 0x6e: /* OUTS */
7614 case 0x33: /* RDPMC */
7624 * Decode to be emulated instruction. Return EMULATION_OK if success.
7626 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7627 void *insn, int insn_len)
7629 int r = EMULATION_OK;
7630 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7632 init_emulate_ctxt(vcpu);
7635 * We will reenter on the same instruction since we do not set
7636 * complete_userspace_io. This does not handle watchpoints yet,
7637 * those would be handled in the emulate_ops.
7639 if (!(emulation_type & EMULTYPE_SKIP) &&
7640 kvm_vcpu_check_breakpoint(vcpu, &r))
7643 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7645 trace_kvm_emulate_insn_start(vcpu);
7646 ++vcpu->stat.insn_emulation;
7650 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7652 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7653 int emulation_type, void *insn, int insn_len)
7656 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7657 bool writeback = true;
7658 bool write_fault_to_spt;
7660 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7663 vcpu->arch.l1tf_flush_l1d = true;
7666 * Clear write_fault_to_shadow_pgtable here to ensure it is
7669 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7670 vcpu->arch.write_fault_to_shadow_pgtable = false;
7672 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7673 kvm_clear_exception_queue(vcpu);
7675 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7677 if (r != EMULATION_OK) {
7678 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7679 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7680 kvm_queue_exception(vcpu, UD_VECTOR);
7683 if (reexecute_instruction(vcpu, cr2_or_gpa,
7687 if (ctxt->have_exception) {
7689 * #UD should result in just EMULATION_FAILED, and trap-like
7690 * exception should not be encountered during decode.
7692 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7693 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7694 inject_emulated_exception(vcpu);
7697 return handle_emulation_failure(vcpu, emulation_type);
7701 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7702 !is_vmware_backdoor_opcode(ctxt)) {
7703 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7708 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7709 * for kvm_skip_emulated_instruction(). The caller is responsible for
7710 * updating interruptibility state and injecting single-step #DBs.
7712 if (emulation_type & EMULTYPE_SKIP) {
7713 kvm_rip_write(vcpu, ctxt->_eip);
7714 if (ctxt->eflags & X86_EFLAGS_RF)
7715 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7719 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7722 /* this is needed for vmware backdoor interface to work since it
7723 changes registers values during IO operation */
7724 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7725 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7726 emulator_invalidate_register_cache(ctxt);
7730 if (emulation_type & EMULTYPE_PF) {
7731 /* Save the faulting GPA (cr2) in the address field */
7732 ctxt->exception.address = cr2_or_gpa;
7734 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7735 if (vcpu->arch.mmu->direct_map) {
7736 ctxt->gpa_available = true;
7737 ctxt->gpa_val = cr2_or_gpa;
7740 /* Sanitize the address out of an abundance of paranoia. */
7741 ctxt->exception.address = 0;
7744 r = x86_emulate_insn(ctxt);
7746 if (r == EMULATION_INTERCEPTED)
7749 if (r == EMULATION_FAILED) {
7750 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7754 return handle_emulation_failure(vcpu, emulation_type);
7757 if (ctxt->have_exception) {
7759 if (inject_emulated_exception(vcpu))
7761 } else if (vcpu->arch.pio.count) {
7762 if (!vcpu->arch.pio.in) {
7763 /* FIXME: return into emulator if single-stepping. */
7764 vcpu->arch.pio.count = 0;
7767 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7770 } else if (vcpu->mmio_needed) {
7771 ++vcpu->stat.mmio_exits;
7773 if (!vcpu->mmio_is_write)
7776 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7777 } else if (r == EMULATION_RESTART)
7783 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7784 toggle_interruptibility(vcpu, ctxt->interruptibility);
7785 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7786 if (!ctxt->have_exception ||
7787 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7788 kvm_rip_write(vcpu, ctxt->eip);
7789 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7790 r = kvm_vcpu_do_singlestep(vcpu);
7791 if (kvm_x86_ops.update_emulated_instruction)
7792 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7793 __kvm_set_rflags(vcpu, ctxt->eflags);
7797 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7798 * do nothing, and it will be requested again as soon as
7799 * the shadow expires. But we still need to check here,
7800 * because POPF has no interrupt shadow.
7802 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7803 kvm_make_request(KVM_REQ_EVENT, vcpu);
7805 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7810 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7812 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7814 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7816 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7817 void *insn, int insn_len)
7819 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7821 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7823 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7825 vcpu->arch.pio.count = 0;
7829 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7831 vcpu->arch.pio.count = 0;
7833 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7836 return kvm_skip_emulated_instruction(vcpu);
7839 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7840 unsigned short port)
7842 unsigned long val = kvm_rax_read(vcpu);
7843 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7849 * Workaround userspace that relies on old KVM behavior of %rip being
7850 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7853 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7854 vcpu->arch.complete_userspace_io =
7855 complete_fast_pio_out_port_0x7e;
7856 kvm_skip_emulated_instruction(vcpu);
7858 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7859 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7864 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7868 /* We should only ever be called with arch.pio.count equal to 1 */
7869 BUG_ON(vcpu->arch.pio.count != 1);
7871 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7872 vcpu->arch.pio.count = 0;
7876 /* For size less than 4 we merge, else we zero extend */
7877 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7880 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7881 * the copy and tracing
7883 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7884 kvm_rax_write(vcpu, val);
7886 return kvm_skip_emulated_instruction(vcpu);
7889 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7890 unsigned short port)
7895 /* For size less than 4 we merge, else we zero extend */
7896 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7898 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7900 kvm_rax_write(vcpu, val);
7904 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7905 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7910 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7915 ret = kvm_fast_pio_in(vcpu, size, port);
7917 ret = kvm_fast_pio_out(vcpu, size, port);
7918 return ret && kvm_skip_emulated_instruction(vcpu);
7920 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7922 static int kvmclock_cpu_down_prep(unsigned int cpu)
7924 __this_cpu_write(cpu_tsc_khz, 0);
7928 static void tsc_khz_changed(void *data)
7930 struct cpufreq_freqs *freq = data;
7931 unsigned long khz = 0;
7935 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7936 khz = cpufreq_quick_get(raw_smp_processor_id());
7939 __this_cpu_write(cpu_tsc_khz, khz);
7942 #ifdef CONFIG_X86_64
7943 static void kvm_hyperv_tsc_notifier(void)
7946 struct kvm_vcpu *vcpu;
7948 unsigned long flags;
7950 mutex_lock(&kvm_lock);
7951 list_for_each_entry(kvm, &vm_list, vm_list)
7952 kvm_make_mclock_inprogress_request(kvm);
7954 hyperv_stop_tsc_emulation();
7956 /* TSC frequency always matches when on Hyper-V */
7957 for_each_present_cpu(cpu)
7958 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7959 kvm_max_guest_tsc_khz = tsc_khz;
7961 list_for_each_entry(kvm, &vm_list, vm_list) {
7962 struct kvm_arch *ka = &kvm->arch;
7964 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
7965 pvclock_update_vm_gtod_copy(kvm);
7966 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
7968 kvm_for_each_vcpu(cpu, vcpu, kvm)
7969 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7971 kvm_for_each_vcpu(cpu, vcpu, kvm)
7972 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7974 mutex_unlock(&kvm_lock);
7978 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7981 struct kvm_vcpu *vcpu;
7982 int i, send_ipi = 0;
7985 * We allow guests to temporarily run on slowing clocks,
7986 * provided we notify them after, or to run on accelerating
7987 * clocks, provided we notify them before. Thus time never
7990 * However, we have a problem. We can't atomically update
7991 * the frequency of a given CPU from this function; it is
7992 * merely a notifier, which can be called from any CPU.
7993 * Changing the TSC frequency at arbitrary points in time
7994 * requires a recomputation of local variables related to
7995 * the TSC for each VCPU. We must flag these local variables
7996 * to be updated and be sure the update takes place with the
7997 * new frequency before any guests proceed.
7999 * Unfortunately, the combination of hotplug CPU and frequency
8000 * change creates an intractable locking scenario; the order
8001 * of when these callouts happen is undefined with respect to
8002 * CPU hotplug, and they can race with each other. As such,
8003 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8004 * undefined; you can actually have a CPU frequency change take
8005 * place in between the computation of X and the setting of the
8006 * variable. To protect against this problem, all updates of
8007 * the per_cpu tsc_khz variable are done in an interrupt
8008 * protected IPI, and all callers wishing to update the value
8009 * must wait for a synchronous IPI to complete (which is trivial
8010 * if the caller is on the CPU already). This establishes the
8011 * necessary total order on variable updates.
8013 * Note that because a guest time update may take place
8014 * anytime after the setting of the VCPU's request bit, the
8015 * correct TSC value must be set before the request. However,
8016 * to ensure the update actually makes it to any guest which
8017 * starts running in hardware virtualization between the set
8018 * and the acquisition of the spinlock, we must also ping the
8019 * CPU after setting the request bit.
8023 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8025 mutex_lock(&kvm_lock);
8026 list_for_each_entry(kvm, &vm_list, vm_list) {
8027 kvm_for_each_vcpu(i, vcpu, kvm) {
8028 if (vcpu->cpu != cpu)
8030 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8031 if (vcpu->cpu != raw_smp_processor_id())
8035 mutex_unlock(&kvm_lock);
8037 if (freq->old < freq->new && send_ipi) {
8039 * We upscale the frequency. Must make the guest
8040 * doesn't see old kvmclock values while running with
8041 * the new frequency, otherwise we risk the guest sees
8042 * time go backwards.
8044 * In case we update the frequency for another cpu
8045 * (which might be in guest context) send an interrupt
8046 * to kick the cpu out of guest context. Next time
8047 * guest context is entered kvmclock will be updated,
8048 * so the guest will not see stale values.
8050 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8054 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8057 struct cpufreq_freqs *freq = data;
8060 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8062 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8065 for_each_cpu(cpu, freq->policy->cpus)
8066 __kvmclock_cpufreq_notifier(freq, cpu);
8071 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8072 .notifier_call = kvmclock_cpufreq_notifier
8075 static int kvmclock_cpu_online(unsigned int cpu)
8077 tsc_khz_changed(NULL);
8081 static void kvm_timer_init(void)
8083 max_tsc_khz = tsc_khz;
8085 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8086 #ifdef CONFIG_CPU_FREQ
8087 struct cpufreq_policy *policy;
8091 policy = cpufreq_cpu_get(cpu);
8093 if (policy->cpuinfo.max_freq)
8094 max_tsc_khz = policy->cpuinfo.max_freq;
8095 cpufreq_cpu_put(policy);
8099 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8100 CPUFREQ_TRANSITION_NOTIFIER);
8103 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8104 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8107 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8108 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8110 int kvm_is_in_guest(void)
8112 return __this_cpu_read(current_vcpu) != NULL;
8115 static int kvm_is_user_mode(void)
8119 if (__this_cpu_read(current_vcpu))
8120 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8122 return user_mode != 0;
8125 static unsigned long kvm_get_guest_ip(void)
8127 unsigned long ip = 0;
8129 if (__this_cpu_read(current_vcpu))
8130 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8135 static void kvm_handle_intel_pt_intr(void)
8137 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8139 kvm_make_request(KVM_REQ_PMI, vcpu);
8140 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8141 (unsigned long *)&vcpu->arch.pmu.global_status);
8144 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8145 .is_in_guest = kvm_is_in_guest,
8146 .is_user_mode = kvm_is_user_mode,
8147 .get_guest_ip = kvm_get_guest_ip,
8148 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8151 #ifdef CONFIG_X86_64
8152 static void pvclock_gtod_update_fn(struct work_struct *work)
8156 struct kvm_vcpu *vcpu;
8159 mutex_lock(&kvm_lock);
8160 list_for_each_entry(kvm, &vm_list, vm_list)
8161 kvm_for_each_vcpu(i, vcpu, kvm)
8162 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8163 atomic_set(&kvm_guest_has_master_clock, 0);
8164 mutex_unlock(&kvm_lock);
8167 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8170 * Indirection to move queue_work() out of the tk_core.seq write held
8171 * region to prevent possible deadlocks against time accessors which
8172 * are invoked with work related locks held.
8174 static void pvclock_irq_work_fn(struct irq_work *w)
8176 queue_work(system_long_wq, &pvclock_gtod_work);
8179 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8182 * Notification about pvclock gtod data update.
8184 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8187 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8188 struct timekeeper *tk = priv;
8190 update_pvclock_gtod(tk);
8193 * Disable master clock if host does not trust, or does not use,
8194 * TSC based clocksource. Delegate queue_work() to irq_work as
8195 * this is invoked with tk_core.seq write held.
8197 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8198 atomic_read(&kvm_guest_has_master_clock) != 0)
8199 irq_work_queue(&pvclock_irq_work);
8203 static struct notifier_block pvclock_gtod_notifier = {
8204 .notifier_call = pvclock_gtod_notify,
8208 int kvm_arch_init(void *opaque)
8210 struct kvm_x86_init_ops *ops = opaque;
8213 if (kvm_x86_ops.hardware_enable) {
8214 printk(KERN_ERR "kvm: already loaded the other module\n");
8219 if (!ops->cpu_has_kvm_support()) {
8220 pr_err_ratelimited("kvm: no hardware support\n");
8224 if (ops->disabled_by_bios()) {
8225 pr_err_ratelimited("kvm: disabled by bios\n");
8231 * KVM explicitly assumes that the guest has an FPU and
8232 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8233 * vCPU's FPU state as a fxregs_state struct.
8235 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8236 printk(KERN_ERR "kvm: inadequate fpu\n");
8242 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8243 __alignof__(struct fpu), SLAB_ACCOUNT,
8245 if (!x86_fpu_cache) {
8246 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8250 x86_emulator_cache = kvm_alloc_emulator_cache();
8251 if (!x86_emulator_cache) {
8252 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8253 goto out_free_x86_fpu_cache;
8256 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8257 if (!user_return_msrs) {
8258 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8259 goto out_free_x86_emulator_cache;
8261 kvm_nr_uret_msrs = 0;
8263 r = kvm_mmu_module_init();
8265 goto out_free_percpu;
8269 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8271 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8272 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8273 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8276 if (pi_inject_timer == -1)
8277 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8278 #ifdef CONFIG_X86_64
8279 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8281 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8282 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8288 free_percpu(user_return_msrs);
8289 out_free_x86_emulator_cache:
8290 kmem_cache_destroy(x86_emulator_cache);
8291 out_free_x86_fpu_cache:
8292 kmem_cache_destroy(x86_fpu_cache);
8297 void kvm_arch_exit(void)
8299 #ifdef CONFIG_X86_64
8300 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8301 clear_hv_tscchange_cb();
8304 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8306 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8307 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8308 CPUFREQ_TRANSITION_NOTIFIER);
8309 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8310 #ifdef CONFIG_X86_64
8311 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8312 irq_work_sync(&pvclock_irq_work);
8313 cancel_work_sync(&pvclock_gtod_work);
8315 kvm_x86_ops.hardware_enable = NULL;
8316 kvm_mmu_module_exit();
8317 free_percpu(user_return_msrs);
8318 kmem_cache_destroy(x86_emulator_cache);
8319 kmem_cache_destroy(x86_fpu_cache);
8320 #ifdef CONFIG_KVM_XEN
8321 static_key_deferred_flush(&kvm_xen_enabled);
8322 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8326 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8328 ++vcpu->stat.halt_exits;
8329 if (lapic_in_kernel(vcpu)) {
8330 vcpu->arch.mp_state = state;
8333 vcpu->run->exit_reason = reason;
8338 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8340 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8342 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8344 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8346 int ret = kvm_skip_emulated_instruction(vcpu);
8348 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8349 * KVM_EXIT_DEBUG here.
8351 return kvm_vcpu_halt(vcpu) && ret;
8353 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8355 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8357 int ret = kvm_skip_emulated_instruction(vcpu);
8359 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8361 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8363 #ifdef CONFIG_X86_64
8364 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8365 unsigned long clock_type)
8367 struct kvm_clock_pairing clock_pairing;
8368 struct timespec64 ts;
8372 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8373 return -KVM_EOPNOTSUPP;
8375 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8376 return -KVM_EOPNOTSUPP;
8378 clock_pairing.sec = ts.tv_sec;
8379 clock_pairing.nsec = ts.tv_nsec;
8380 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8381 clock_pairing.flags = 0;
8382 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8385 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8386 sizeof(struct kvm_clock_pairing)))
8394 * kvm_pv_kick_cpu_op: Kick a vcpu.
8396 * @apicid - apicid of vcpu to be kicked.
8398 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8400 struct kvm_lapic_irq lapic_irq;
8402 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8403 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8404 lapic_irq.level = 0;
8405 lapic_irq.dest_id = apicid;
8406 lapic_irq.msi_redir_hint = false;
8408 lapic_irq.delivery_mode = APIC_DM_REMRD;
8409 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8412 bool kvm_apicv_activated(struct kvm *kvm)
8414 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8416 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8418 void kvm_apicv_init(struct kvm *kvm, bool enable)
8421 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8422 &kvm->arch.apicv_inhibit_reasons);
8424 set_bit(APICV_INHIBIT_REASON_DISABLE,
8425 &kvm->arch.apicv_inhibit_reasons);
8427 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8429 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8431 struct kvm_vcpu *target = NULL;
8432 struct kvm_apic_map *map;
8434 vcpu->stat.directed_yield_attempted++;
8436 if (single_task_running())
8440 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8442 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8443 target = map->phys_map[dest_id]->vcpu;
8447 if (!target || !READ_ONCE(target->ready))
8450 /* Ignore requests to yield to self */
8454 if (kvm_vcpu_yield_to(target) <= 0)
8457 vcpu->stat.directed_yield_successful++;
8463 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8465 unsigned long nr, a0, a1, a2, a3, ret;
8468 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8469 return kvm_xen_hypercall(vcpu);
8471 if (kvm_hv_hypercall_enabled(vcpu))
8472 return kvm_hv_hypercall(vcpu);
8474 nr = kvm_rax_read(vcpu);
8475 a0 = kvm_rbx_read(vcpu);
8476 a1 = kvm_rcx_read(vcpu);
8477 a2 = kvm_rdx_read(vcpu);
8478 a3 = kvm_rsi_read(vcpu);
8480 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8482 op_64_bit = is_64_bit_mode(vcpu);
8491 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8499 case KVM_HC_VAPIC_POLL_IRQ:
8502 case KVM_HC_KICK_CPU:
8503 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8506 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8507 kvm_sched_yield(vcpu, a1);
8510 #ifdef CONFIG_X86_64
8511 case KVM_HC_CLOCK_PAIRING:
8512 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8515 case KVM_HC_SEND_IPI:
8516 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8519 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8521 case KVM_HC_SCHED_YIELD:
8522 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8525 kvm_sched_yield(vcpu, a0);
8535 kvm_rax_write(vcpu, ret);
8537 ++vcpu->stat.hypercalls;
8538 return kvm_skip_emulated_instruction(vcpu);
8540 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8542 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8544 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8545 char instruction[3];
8546 unsigned long rip = kvm_rip_read(vcpu);
8548 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8550 return emulator_write_emulated(ctxt, rip, instruction, 3,
8554 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8556 return vcpu->run->request_interrupt_window &&
8557 likely(!pic_in_kernel(vcpu->kvm));
8560 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8562 struct kvm_run *kvm_run = vcpu->run;
8565 * if_flag is obsolete and useless, so do not bother
8566 * setting it for SEV-ES guests. Userspace can just
8567 * use kvm_run->ready_for_interrupt_injection.
8569 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8570 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8572 kvm_run->cr8 = kvm_get_cr8(vcpu);
8573 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8574 kvm_run->ready_for_interrupt_injection =
8575 pic_in_kernel(vcpu->kvm) ||
8576 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8579 kvm_run->flags |= KVM_RUN_X86_SMM;
8582 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8586 if (!kvm_x86_ops.update_cr8_intercept)
8589 if (!lapic_in_kernel(vcpu))
8592 if (vcpu->arch.apicv_active)
8595 if (!vcpu->arch.apic->vapic_addr)
8596 max_irr = kvm_lapic_find_highest_irr(vcpu);
8603 tpr = kvm_lapic_get_cr8(vcpu);
8605 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8609 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8611 if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
8614 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8615 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8619 return kvm_x86_ops.nested_ops->check_events(vcpu);
8622 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8624 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8625 vcpu->arch.exception.error_code = false;
8626 static_call(kvm_x86_queue_exception)(vcpu);
8629 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8632 bool can_inject = true;
8634 /* try to reinject previous events if any */
8636 if (vcpu->arch.exception.injected) {
8637 kvm_inject_exception(vcpu);
8641 * Do not inject an NMI or interrupt if there is a pending
8642 * exception. Exceptions and interrupts are recognized at
8643 * instruction boundaries, i.e. the start of an instruction.
8644 * Trap-like exceptions, e.g. #DB, have higher priority than
8645 * NMIs and interrupts, i.e. traps are recognized before an
8646 * NMI/interrupt that's pending on the same instruction.
8647 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8648 * priority, but are only generated (pended) during instruction
8649 * execution, i.e. a pending fault-like exception means the
8650 * fault occurred on the *previous* instruction and must be
8651 * serviced prior to recognizing any new events in order to
8652 * fully complete the previous instruction.
8654 else if (!vcpu->arch.exception.pending) {
8655 if (vcpu->arch.nmi_injected) {
8656 static_call(kvm_x86_set_nmi)(vcpu);
8658 } else if (vcpu->arch.interrupt.injected) {
8659 static_call(kvm_x86_set_irq)(vcpu);
8664 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8665 vcpu->arch.exception.pending);
8668 * Call check_nested_events() even if we reinjected a previous event
8669 * in order for caller to determine if it should require immediate-exit
8670 * from L2 to L1 due to pending L1 events which require exit
8673 if (is_guest_mode(vcpu)) {
8674 r = kvm_check_nested_events(vcpu);
8679 /* try to inject new event if pending */
8680 if (vcpu->arch.exception.pending) {
8681 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8682 vcpu->arch.exception.has_error_code,
8683 vcpu->arch.exception.error_code);
8685 vcpu->arch.exception.pending = false;
8686 vcpu->arch.exception.injected = true;
8688 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8689 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8692 if (vcpu->arch.exception.nr == DB_VECTOR) {
8693 kvm_deliver_exception_payload(vcpu);
8694 if (vcpu->arch.dr7 & DR7_GD) {
8695 vcpu->arch.dr7 &= ~DR7_GD;
8696 kvm_update_dr7(vcpu);
8700 kvm_inject_exception(vcpu);
8705 * Finally, inject interrupt events. If an event cannot be injected
8706 * due to architectural conditions (e.g. IF=0) a window-open exit
8707 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8708 * and can architecturally be injected, but we cannot do it right now:
8709 * an interrupt could have arrived just now and we have to inject it
8710 * as a vmexit, or there could already an event in the queue, which is
8711 * indicated by can_inject. In that case we request an immediate exit
8712 * in order to make progress and get back here for another iteration.
8713 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8715 if (vcpu->arch.smi_pending) {
8716 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8720 vcpu->arch.smi_pending = false;
8721 ++vcpu->arch.smi_count;
8725 static_call(kvm_x86_enable_smi_window)(vcpu);
8728 if (vcpu->arch.nmi_pending) {
8729 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8733 --vcpu->arch.nmi_pending;
8734 vcpu->arch.nmi_injected = true;
8735 static_call(kvm_x86_set_nmi)(vcpu);
8737 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8739 if (vcpu->arch.nmi_pending)
8740 static_call(kvm_x86_enable_nmi_window)(vcpu);
8743 if (kvm_cpu_has_injectable_intr(vcpu)) {
8744 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8748 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8749 static_call(kvm_x86_set_irq)(vcpu);
8750 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8752 if (kvm_cpu_has_injectable_intr(vcpu))
8753 static_call(kvm_x86_enable_irq_window)(vcpu);
8756 if (is_guest_mode(vcpu) &&
8757 kvm_x86_ops.nested_ops->hv_timer_pending &&
8758 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8759 *req_immediate_exit = true;
8761 WARN_ON(vcpu->arch.exception.pending);
8765 *req_immediate_exit = true;
8769 static void process_nmi(struct kvm_vcpu *vcpu)
8774 * x86 is limited to one NMI running, and one NMI pending after it.
8775 * If an NMI is already in progress, limit further NMIs to just one.
8776 * Otherwise, allow two (and we'll inject the first one immediately).
8778 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8781 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8782 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8783 kvm_make_request(KVM_REQ_EVENT, vcpu);
8786 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8789 flags |= seg->g << 23;
8790 flags |= seg->db << 22;
8791 flags |= seg->l << 21;
8792 flags |= seg->avl << 20;
8793 flags |= seg->present << 15;
8794 flags |= seg->dpl << 13;
8795 flags |= seg->s << 12;
8796 flags |= seg->type << 8;
8800 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8802 struct kvm_segment seg;
8805 kvm_get_segment(vcpu, &seg, n);
8806 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8809 offset = 0x7f84 + n * 12;
8811 offset = 0x7f2c + (n - 3) * 12;
8813 put_smstate(u32, buf, offset + 8, seg.base);
8814 put_smstate(u32, buf, offset + 4, seg.limit);
8815 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8818 #ifdef CONFIG_X86_64
8819 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8821 struct kvm_segment seg;
8825 kvm_get_segment(vcpu, &seg, n);
8826 offset = 0x7e00 + n * 16;
8828 flags = enter_smm_get_segment_flags(&seg) >> 8;
8829 put_smstate(u16, buf, offset, seg.selector);
8830 put_smstate(u16, buf, offset + 2, flags);
8831 put_smstate(u32, buf, offset + 4, seg.limit);
8832 put_smstate(u64, buf, offset + 8, seg.base);
8836 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8839 struct kvm_segment seg;
8843 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8844 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8845 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8846 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8848 for (i = 0; i < 8; i++)
8849 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
8851 kvm_get_dr(vcpu, 6, &val);
8852 put_smstate(u32, buf, 0x7fcc, (u32)val);
8853 kvm_get_dr(vcpu, 7, &val);
8854 put_smstate(u32, buf, 0x7fc8, (u32)val);
8856 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8857 put_smstate(u32, buf, 0x7fc4, seg.selector);
8858 put_smstate(u32, buf, 0x7f64, seg.base);
8859 put_smstate(u32, buf, 0x7f60, seg.limit);
8860 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8862 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8863 put_smstate(u32, buf, 0x7fc0, seg.selector);
8864 put_smstate(u32, buf, 0x7f80, seg.base);
8865 put_smstate(u32, buf, 0x7f7c, seg.limit);
8866 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8868 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8869 put_smstate(u32, buf, 0x7f74, dt.address);
8870 put_smstate(u32, buf, 0x7f70, dt.size);
8872 static_call(kvm_x86_get_idt)(vcpu, &dt);
8873 put_smstate(u32, buf, 0x7f58, dt.address);
8874 put_smstate(u32, buf, 0x7f54, dt.size);
8876 for (i = 0; i < 6; i++)
8877 enter_smm_save_seg_32(vcpu, buf, i);
8879 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8882 put_smstate(u32, buf, 0x7efc, 0x00020000);
8883 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8886 #ifdef CONFIG_X86_64
8887 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8890 struct kvm_segment seg;
8894 for (i = 0; i < 16; i++)
8895 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
8897 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8898 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8900 kvm_get_dr(vcpu, 6, &val);
8901 put_smstate(u64, buf, 0x7f68, val);
8902 kvm_get_dr(vcpu, 7, &val);
8903 put_smstate(u64, buf, 0x7f60, val);
8905 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8906 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8907 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8909 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8912 put_smstate(u32, buf, 0x7efc, 0x00020064);
8914 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8916 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8917 put_smstate(u16, buf, 0x7e90, seg.selector);
8918 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8919 put_smstate(u32, buf, 0x7e94, seg.limit);
8920 put_smstate(u64, buf, 0x7e98, seg.base);
8922 static_call(kvm_x86_get_idt)(vcpu, &dt);
8923 put_smstate(u32, buf, 0x7e84, dt.size);
8924 put_smstate(u64, buf, 0x7e88, dt.address);
8926 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8927 put_smstate(u16, buf, 0x7e70, seg.selector);
8928 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8929 put_smstate(u32, buf, 0x7e74, seg.limit);
8930 put_smstate(u64, buf, 0x7e78, seg.base);
8932 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8933 put_smstate(u32, buf, 0x7e64, dt.size);
8934 put_smstate(u64, buf, 0x7e68, dt.address);
8936 for (i = 0; i < 6; i++)
8937 enter_smm_save_seg_64(vcpu, buf, i);
8941 static void enter_smm(struct kvm_vcpu *vcpu)
8943 struct kvm_segment cs, ds;
8948 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8949 memset(buf, 0, 512);
8950 #ifdef CONFIG_X86_64
8951 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8952 enter_smm_save_state_64(vcpu, buf);
8955 enter_smm_save_state_32(vcpu, buf);
8958 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8959 * vCPU state (e.g. leave guest mode) after we've saved the state into
8960 * the SMM state-save area.
8962 static_call(kvm_x86_pre_enter_smm)(vcpu, buf);
8964 vcpu->arch.hflags |= HF_SMM_MASK;
8965 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8967 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
8968 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8970 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
8972 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8973 kvm_rip_write(vcpu, 0x8000);
8975 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8976 static_call(kvm_x86_set_cr0)(vcpu, cr0);
8977 vcpu->arch.cr0 = cr0;
8979 static_call(kvm_x86_set_cr4)(vcpu, 0);
8981 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8982 dt.address = dt.size = 0;
8983 static_call(kvm_x86_set_idt)(vcpu, &dt);
8985 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8987 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8988 cs.base = vcpu->arch.smbase;
8993 cs.limit = ds.limit = 0xffffffff;
8994 cs.type = ds.type = 0x3;
8995 cs.dpl = ds.dpl = 0;
9000 cs.avl = ds.avl = 0;
9001 cs.present = ds.present = 1;
9002 cs.unusable = ds.unusable = 0;
9003 cs.padding = ds.padding = 0;
9005 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9006 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9007 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9008 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9009 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9010 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9012 #ifdef CONFIG_X86_64
9013 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9014 static_call(kvm_x86_set_efer)(vcpu, 0);
9017 kvm_update_cpuid_runtime(vcpu);
9018 kvm_mmu_reset_context(vcpu);
9021 static void process_smi(struct kvm_vcpu *vcpu)
9023 vcpu->arch.smi_pending = true;
9024 kvm_make_request(KVM_REQ_EVENT, vcpu);
9027 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9028 unsigned long *vcpu_bitmap)
9032 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
9034 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
9035 NULL, vcpu_bitmap, cpus);
9037 free_cpumask_var(cpus);
9040 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9042 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9045 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9047 if (!lapic_in_kernel(vcpu))
9050 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
9051 kvm_apic_update_apicv(vcpu);
9052 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9054 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9057 * NOTE: Do not hold any lock prior to calling this.
9059 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
9060 * locked, because it calls __x86_set_memory_region() which does
9061 * synchronize_srcu(&kvm->srcu).
9063 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9065 struct kvm_vcpu *except;
9066 unsigned long old, new, expected;
9068 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9069 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9072 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
9074 expected = new = old;
9076 __clear_bit(bit, &new);
9078 __set_bit(bit, &new);
9081 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
9082 } while (old != expected);
9087 trace_kvm_apicv_update_request(activate, bit);
9088 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
9089 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
9092 * Sending request to update APICV for all other vcpus,
9093 * while update the calling vcpu immediately instead of
9094 * waiting for another #VMEXIT to handle the request.
9096 except = kvm_get_running_vcpu();
9097 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
9100 kvm_vcpu_update_apicv(except);
9102 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9104 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9106 if (!kvm_apic_present(vcpu))
9109 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9111 if (irqchip_split(vcpu->kvm))
9112 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9114 if (vcpu->arch.apicv_active)
9115 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9116 if (ioapic_in_kernel(vcpu->kvm))
9117 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9120 if (is_guest_mode(vcpu))
9121 vcpu->arch.load_eoi_exitmap_pending = true;
9123 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9126 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9128 u64 eoi_exit_bitmap[4];
9130 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9133 if (to_hv_vcpu(vcpu))
9134 bitmap_or((ulong *)eoi_exit_bitmap,
9135 vcpu->arch.ioapic_handled_vectors,
9136 to_hv_synic(vcpu)->vec_bitmap, 256);
9138 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9141 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9142 unsigned long start, unsigned long end)
9144 unsigned long apic_address;
9147 * The physical address of apic access page is stored in the VMCS.
9148 * Update it when it becomes invalid.
9150 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9151 if (start <= apic_address && apic_address < end)
9152 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9155 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9157 if (!lapic_in_kernel(vcpu))
9160 if (!kvm_x86_ops.set_apic_access_page_addr)
9163 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9166 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9168 smp_send_reschedule(vcpu->cpu);
9170 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9173 * Returns 1 to let vcpu_run() continue the guest execution loop without
9174 * exiting to the userspace. Otherwise, the value will be returned to the
9177 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9181 dm_request_for_irq_injection(vcpu) &&
9182 kvm_cpu_accept_dm_intr(vcpu);
9183 fastpath_t exit_fastpath;
9185 bool req_immediate_exit = false;
9187 /* Forbid vmenter if vcpu dirty ring is soft-full */
9188 if (unlikely(vcpu->kvm->dirty_ring_size &&
9189 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9190 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9191 trace_kvm_dirty_ring_exit(vcpu);
9196 if (kvm_request_pending(vcpu)) {
9197 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9198 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9203 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9204 kvm_mmu_unload(vcpu);
9205 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9206 __kvm_migrate_timers(vcpu);
9207 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9208 kvm_gen_update_masterclock(vcpu->kvm);
9209 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9210 kvm_gen_kvmclock_update(vcpu);
9211 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9212 r = kvm_guest_time_update(vcpu);
9216 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9217 kvm_mmu_sync_roots(vcpu);
9218 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9219 kvm_mmu_load_pgd(vcpu);
9220 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9221 kvm_vcpu_flush_tlb_all(vcpu);
9223 /* Flushing all ASIDs flushes the current ASID... */
9224 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9226 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9227 kvm_vcpu_flush_tlb_current(vcpu);
9228 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
9229 kvm_vcpu_flush_tlb_guest(vcpu);
9231 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9232 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9236 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9237 if (is_guest_mode(vcpu)) {
9238 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9240 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9241 vcpu->mmio_needed = 0;
9246 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9247 /* Page is swapped out. Do synthetic halt */
9248 vcpu->arch.apf.halted = true;
9252 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9253 record_steal_time(vcpu);
9254 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9256 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9258 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9259 kvm_pmu_handle_event(vcpu);
9260 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9261 kvm_pmu_deliver_pmi(vcpu);
9262 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9263 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9264 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9265 vcpu->arch.ioapic_handled_vectors)) {
9266 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9267 vcpu->run->eoi.vector =
9268 vcpu->arch.pending_ioapic_eoi;
9273 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9274 vcpu_scan_ioapic(vcpu);
9275 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9276 vcpu_load_eoi_exitmap(vcpu);
9277 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9278 kvm_vcpu_reload_apic_access_page(vcpu);
9279 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9280 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9281 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9285 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9286 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9287 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9291 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9292 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9294 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9295 vcpu->run->hyperv = hv_vcpu->exit;
9301 * KVM_REQ_HV_STIMER has to be processed after
9302 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9303 * depend on the guest clock being up-to-date
9305 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9306 kvm_hv_process_stimers(vcpu);
9307 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9308 kvm_vcpu_update_apicv(vcpu);
9309 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9310 kvm_check_async_pf_completion(vcpu);
9311 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9312 static_call(kvm_x86_msr_filter_changed)(vcpu);
9314 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9315 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9318 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9319 kvm_xen_has_interrupt(vcpu)) {
9320 ++vcpu->stat.req_event;
9321 kvm_apic_accept_events(vcpu);
9322 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9327 inject_pending_event(vcpu, &req_immediate_exit);
9329 static_call(kvm_x86_enable_irq_window)(vcpu);
9331 if (kvm_lapic_enabled(vcpu)) {
9332 update_cr8_intercept(vcpu);
9333 kvm_lapic_sync_to_vapic(vcpu);
9337 r = kvm_mmu_reload(vcpu);
9339 goto cancel_injection;
9344 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9347 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9348 * IPI are then delayed after guest entry, which ensures that they
9349 * result in virtual interrupt delivery.
9351 local_irq_disable();
9352 vcpu->mode = IN_GUEST_MODE;
9354 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9357 * 1) We should set ->mode before checking ->requests. Please see
9358 * the comment in kvm_vcpu_exiting_guest_mode().
9360 * 2) For APICv, we should set ->mode before checking PID.ON. This
9361 * pairs with the memory barrier implicit in pi_test_and_set_on
9362 * (see vmx_deliver_posted_interrupt).
9364 * 3) This also orders the write to mode from any reads to the page
9365 * tables done while the VCPU is running. Please see the comment
9366 * in kvm_flush_remote_tlbs.
9368 smp_mb__after_srcu_read_unlock();
9371 * This handles the case where a posted interrupt was
9372 * notified with kvm_vcpu_kick.
9374 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9375 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9377 if (kvm_vcpu_exit_request(vcpu)) {
9378 vcpu->mode = OUTSIDE_GUEST_MODE;
9382 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9384 goto cancel_injection;
9387 if (req_immediate_exit) {
9388 kvm_make_request(KVM_REQ_EVENT, vcpu);
9389 static_call(kvm_x86_request_immediate_exit)(vcpu);
9392 fpregs_assert_state_consistent();
9393 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9394 switch_fpu_return();
9396 if (unlikely(vcpu->arch.switch_db_regs)) {
9398 set_debugreg(vcpu->arch.eff_db[0], 0);
9399 set_debugreg(vcpu->arch.eff_db[1], 1);
9400 set_debugreg(vcpu->arch.eff_db[2], 2);
9401 set_debugreg(vcpu->arch.eff_db[3], 3);
9402 set_debugreg(vcpu->arch.dr6, 6);
9403 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9407 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9408 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9411 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9412 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9416 if (vcpu->arch.apicv_active)
9417 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9421 * Do this here before restoring debug registers on the host. And
9422 * since we do this before handling the vmexit, a DR access vmexit
9423 * can (a) read the correct value of the debug registers, (b) set
9424 * KVM_DEBUGREG_WONT_EXIT again.
9426 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9427 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9428 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9429 kvm_update_dr0123(vcpu);
9430 kvm_update_dr7(vcpu);
9431 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9435 * If the guest has used debug registers, at least dr7
9436 * will be disabled while returning to the host.
9437 * If we don't have active breakpoints in the host, we don't
9438 * care about the messed up debug address registers. But if
9439 * we have some of them active, restore the old state.
9441 if (hw_breakpoint_active())
9442 hw_breakpoint_restore();
9444 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9445 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9447 vcpu->mode = OUTSIDE_GUEST_MODE;
9450 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9453 * Consume any pending interrupts, including the possible source of
9454 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9455 * An instruction is required after local_irq_enable() to fully unblock
9456 * interrupts on processors that implement an interrupt shadow, the
9457 * stat.exits increment will do nicely.
9459 kvm_before_interrupt(vcpu);
9462 local_irq_disable();
9463 kvm_after_interrupt(vcpu);
9466 * Wait until after servicing IRQs to account guest time so that any
9467 * ticks that occurred while running the guest are properly accounted
9468 * to the guest. Waiting until IRQs are enabled degrades the accuracy
9469 * of accounting via context tracking, but the loss of accuracy is
9470 * acceptable for all known use cases.
9472 vtime_account_guest_exit();
9474 if (lapic_in_kernel(vcpu)) {
9475 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9476 if (delta != S64_MIN) {
9477 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9478 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9485 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9488 * Profile KVM exit RIPs:
9490 if (unlikely(prof_on == KVM_PROFILING)) {
9491 unsigned long rip = kvm_rip_read(vcpu);
9492 profile_hit(KVM_PROFILING, (void *)rip);
9495 if (unlikely(vcpu->arch.tsc_always_catchup))
9496 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9498 if (vcpu->arch.apic_attention)
9499 kvm_lapic_sync_from_vapic(vcpu);
9501 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9505 if (req_immediate_exit)
9506 kvm_make_request(KVM_REQ_EVENT, vcpu);
9507 static_call(kvm_x86_cancel_injection)(vcpu);
9508 if (unlikely(vcpu->arch.apic_attention))
9509 kvm_lapic_sync_from_vapic(vcpu);
9514 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9516 if (!kvm_arch_vcpu_runnable(vcpu) &&
9517 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9518 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9519 kvm_vcpu_block(vcpu);
9520 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9522 if (kvm_x86_ops.post_block)
9523 static_call(kvm_x86_post_block)(vcpu);
9525 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9529 kvm_apic_accept_events(vcpu);
9530 switch(vcpu->arch.mp_state) {
9531 case KVM_MP_STATE_HALTED:
9532 case KVM_MP_STATE_AP_RESET_HOLD:
9533 vcpu->arch.pv.pv_unhalted = false;
9534 vcpu->arch.mp_state =
9535 KVM_MP_STATE_RUNNABLE;
9537 case KVM_MP_STATE_RUNNABLE:
9538 vcpu->arch.apf.halted = false;
9540 case KVM_MP_STATE_INIT_RECEIVED:
9548 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9550 if (is_guest_mode(vcpu))
9551 kvm_check_nested_events(vcpu);
9553 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9554 !vcpu->arch.apf.halted);
9557 static int vcpu_run(struct kvm_vcpu *vcpu)
9560 struct kvm *kvm = vcpu->kvm;
9562 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9563 vcpu->arch.l1tf_flush_l1d = true;
9566 if (kvm_vcpu_running(vcpu)) {
9567 r = vcpu_enter_guest(vcpu);
9569 r = vcpu_block(kvm, vcpu);
9575 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9576 if (kvm_cpu_has_pending_timer(vcpu))
9577 kvm_inject_pending_timer_irqs(vcpu);
9579 if (dm_request_for_irq_injection(vcpu) &&
9580 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9582 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9583 ++vcpu->stat.request_irq_exits;
9587 if (__xfer_to_guest_mode_work_pending()) {
9588 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9589 r = xfer_to_guest_mode_handle_work(vcpu);
9592 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9596 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9601 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9605 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9606 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9607 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9611 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9613 BUG_ON(!vcpu->arch.pio.count);
9615 return complete_emulated_io(vcpu);
9619 * Implements the following, as a state machine:
9623 * for each mmio piece in the fragment
9631 * for each mmio piece in the fragment
9636 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9638 struct kvm_run *run = vcpu->run;
9639 struct kvm_mmio_fragment *frag;
9642 BUG_ON(!vcpu->mmio_needed);
9644 /* Complete previous fragment */
9645 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9646 len = min(8u, frag->len);
9647 if (!vcpu->mmio_is_write)
9648 memcpy(frag->data, run->mmio.data, len);
9650 if (frag->len <= 8) {
9651 /* Switch to the next fragment. */
9653 vcpu->mmio_cur_fragment++;
9655 /* Go forward to the next mmio piece. */
9661 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9662 vcpu->mmio_needed = 0;
9664 /* FIXME: return into emulator if single-stepping. */
9665 if (vcpu->mmio_is_write)
9667 vcpu->mmio_read_completed = 1;
9668 return complete_emulated_io(vcpu);
9671 run->exit_reason = KVM_EXIT_MMIO;
9672 run->mmio.phys_addr = frag->gpa;
9673 if (vcpu->mmio_is_write)
9674 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9675 run->mmio.len = min(8u, frag->len);
9676 run->mmio.is_write = vcpu->mmio_is_write;
9677 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9681 static void kvm_save_current_fpu(struct fpu *fpu)
9684 * If the target FPU state is not resident in the CPU registers, just
9685 * memcpy() from current, else save CPU state directly to the target.
9687 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9688 memcpy(&fpu->state, ¤t->thread.fpu.state,
9689 fpu_kernel_xstate_size);
9691 copy_fpregs_to_fpstate(fpu);
9694 /* Swap (qemu) user FPU context for the guest FPU context. */
9695 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9699 kvm_save_current_fpu(vcpu->arch.user_fpu);
9702 * Guests with protected state can't have it set by the hypervisor,
9703 * so skip trying to set it.
9705 if (vcpu->arch.guest_fpu)
9706 /* PKRU is separately restored in kvm_x86_ops.run. */
9707 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9708 ~XFEATURE_MASK_PKRU);
9710 fpregs_mark_activate();
9716 /* When vcpu_run ends, restore user space FPU context. */
9717 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9722 * Guests with protected state can't have it read by the hypervisor,
9723 * so skip trying to save it.
9725 if (vcpu->arch.guest_fpu)
9726 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9728 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9730 fpregs_mark_activate();
9733 ++vcpu->stat.fpu_reload;
9737 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9739 struct kvm_run *kvm_run = vcpu->run;
9743 kvm_sigset_activate(vcpu);
9745 kvm_load_guest_fpu(vcpu);
9747 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9748 if (kvm_run->immediate_exit) {
9752 kvm_vcpu_block(vcpu);
9753 kvm_apic_accept_events(vcpu);
9754 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9756 if (signal_pending(current)) {
9758 kvm_run->exit_reason = KVM_EXIT_INTR;
9759 ++vcpu->stat.signal_exits;
9764 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9769 if (kvm_run->kvm_dirty_regs) {
9770 r = sync_regs(vcpu);
9775 /* re-sync apic's tpr */
9776 if (!lapic_in_kernel(vcpu)) {
9777 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9783 if (unlikely(vcpu->arch.complete_userspace_io)) {
9784 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9785 vcpu->arch.complete_userspace_io = NULL;
9790 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9792 if (kvm_run->immediate_exit)
9798 kvm_put_guest_fpu(vcpu);
9799 if (kvm_run->kvm_valid_regs)
9801 post_kvm_run_save(vcpu);
9802 kvm_sigset_deactivate(vcpu);
9808 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9810 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9812 * We are here if userspace calls get_regs() in the middle of
9813 * instruction emulation. Registers state needs to be copied
9814 * back from emulation context to vcpu. Userspace shouldn't do
9815 * that usually, but some bad designed PV devices (vmware
9816 * backdoor interface) need this to work
9818 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9819 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9821 regs->rax = kvm_rax_read(vcpu);
9822 regs->rbx = kvm_rbx_read(vcpu);
9823 regs->rcx = kvm_rcx_read(vcpu);
9824 regs->rdx = kvm_rdx_read(vcpu);
9825 regs->rsi = kvm_rsi_read(vcpu);
9826 regs->rdi = kvm_rdi_read(vcpu);
9827 regs->rsp = kvm_rsp_read(vcpu);
9828 regs->rbp = kvm_rbp_read(vcpu);
9829 #ifdef CONFIG_X86_64
9830 regs->r8 = kvm_r8_read(vcpu);
9831 regs->r9 = kvm_r9_read(vcpu);
9832 regs->r10 = kvm_r10_read(vcpu);
9833 regs->r11 = kvm_r11_read(vcpu);
9834 regs->r12 = kvm_r12_read(vcpu);
9835 regs->r13 = kvm_r13_read(vcpu);
9836 regs->r14 = kvm_r14_read(vcpu);
9837 regs->r15 = kvm_r15_read(vcpu);
9840 regs->rip = kvm_rip_read(vcpu);
9841 regs->rflags = kvm_get_rflags(vcpu);
9844 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9847 __get_regs(vcpu, regs);
9852 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9854 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9855 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9857 kvm_rax_write(vcpu, regs->rax);
9858 kvm_rbx_write(vcpu, regs->rbx);
9859 kvm_rcx_write(vcpu, regs->rcx);
9860 kvm_rdx_write(vcpu, regs->rdx);
9861 kvm_rsi_write(vcpu, regs->rsi);
9862 kvm_rdi_write(vcpu, regs->rdi);
9863 kvm_rsp_write(vcpu, regs->rsp);
9864 kvm_rbp_write(vcpu, regs->rbp);
9865 #ifdef CONFIG_X86_64
9866 kvm_r8_write(vcpu, regs->r8);
9867 kvm_r9_write(vcpu, regs->r9);
9868 kvm_r10_write(vcpu, regs->r10);
9869 kvm_r11_write(vcpu, regs->r11);
9870 kvm_r12_write(vcpu, regs->r12);
9871 kvm_r13_write(vcpu, regs->r13);
9872 kvm_r14_write(vcpu, regs->r14);
9873 kvm_r15_write(vcpu, regs->r15);
9876 kvm_rip_write(vcpu, regs->rip);
9877 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9879 vcpu->arch.exception.pending = false;
9881 kvm_make_request(KVM_REQ_EVENT, vcpu);
9884 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9887 __set_regs(vcpu, regs);
9892 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9894 struct kvm_segment cs;
9896 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9900 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9902 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9906 if (vcpu->arch.guest_state_protected)
9907 goto skip_protected_regs;
9909 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9910 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9911 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9912 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9913 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9914 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9916 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9917 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9919 static_call(kvm_x86_get_idt)(vcpu, &dt);
9920 sregs->idt.limit = dt.size;
9921 sregs->idt.base = dt.address;
9922 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9923 sregs->gdt.limit = dt.size;
9924 sregs->gdt.base = dt.address;
9926 sregs->cr2 = vcpu->arch.cr2;
9927 sregs->cr3 = kvm_read_cr3(vcpu);
9929 skip_protected_regs:
9930 sregs->cr0 = kvm_read_cr0(vcpu);
9931 sregs->cr4 = kvm_read_cr4(vcpu);
9932 sregs->cr8 = kvm_get_cr8(vcpu);
9933 sregs->efer = vcpu->arch.efer;
9934 sregs->apic_base = kvm_get_apic_base(vcpu);
9936 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9938 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9939 set_bit(vcpu->arch.interrupt.nr,
9940 (unsigned long *)sregs->interrupt_bitmap);
9943 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9944 struct kvm_sregs *sregs)
9947 __get_sregs(vcpu, sregs);
9952 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9953 struct kvm_mp_state *mp_state)
9956 if (kvm_mpx_supported())
9957 kvm_load_guest_fpu(vcpu);
9959 kvm_apic_accept_events(vcpu);
9960 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9961 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9962 vcpu->arch.pv.pv_unhalted)
9963 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9965 mp_state->mp_state = vcpu->arch.mp_state;
9967 if (kvm_mpx_supported())
9968 kvm_put_guest_fpu(vcpu);
9973 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9974 struct kvm_mp_state *mp_state)
9980 if (!lapic_in_kernel(vcpu) &&
9981 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9985 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9986 * INIT state; latched init should be reported using
9987 * KVM_SET_VCPU_EVENTS, so reject it here.
9989 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9990 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9991 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9994 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9995 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9996 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9998 vcpu->arch.mp_state = mp_state->mp_state;
9999 kvm_make_request(KVM_REQ_EVENT, vcpu);
10007 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10008 int reason, bool has_error_code, u32 error_code)
10010 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10013 init_emulate_ctxt(vcpu);
10015 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10016 has_error_code, error_code);
10018 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10019 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10020 vcpu->run->internal.ndata = 0;
10024 kvm_rip_write(vcpu, ctxt->eip);
10025 kvm_set_rflags(vcpu, ctxt->eflags);
10028 EXPORT_SYMBOL_GPL(kvm_task_switch);
10030 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10032 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10034 * When EFER.LME and CR0.PG are set, the processor is in
10035 * 64-bit mode (though maybe in a 32-bit code segment).
10036 * CR4.PAE and EFER.LMA must be set.
10038 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10040 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10044 * Not in 64-bit mode: EFER.LMA is clear and the code
10045 * segment cannot be 64-bit.
10047 if (sregs->efer & EFER_LMA || sregs->cs.l)
10051 return kvm_is_valid_cr4(vcpu, sregs->cr4);
10054 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10056 struct msr_data apic_base_msr;
10057 int mmu_reset_needed = 0;
10058 int pending_vec, max_bits, idx;
10059 struct desc_ptr dt;
10062 if (!kvm_is_valid_sregs(vcpu, sregs))
10065 apic_base_msr.data = sregs->apic_base;
10066 apic_base_msr.host_initiated = true;
10067 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10070 if (vcpu->arch.guest_state_protected)
10071 goto skip_protected_regs;
10073 dt.size = sregs->idt.limit;
10074 dt.address = sregs->idt.base;
10075 static_call(kvm_x86_set_idt)(vcpu, &dt);
10076 dt.size = sregs->gdt.limit;
10077 dt.address = sregs->gdt.base;
10078 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10080 vcpu->arch.cr2 = sregs->cr2;
10081 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10082 vcpu->arch.cr3 = sregs->cr3;
10083 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10085 kvm_set_cr8(vcpu, sregs->cr8);
10087 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10088 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10090 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10091 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10092 vcpu->arch.cr0 = sregs->cr0;
10094 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10095 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10097 idx = srcu_read_lock(&vcpu->kvm->srcu);
10098 if (is_pae_paging(vcpu)) {
10099 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10100 mmu_reset_needed = 1;
10102 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10104 if (mmu_reset_needed)
10105 kvm_mmu_reset_context(vcpu);
10107 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10108 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10109 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10110 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10111 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10112 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10114 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10115 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10117 update_cr8_intercept(vcpu);
10119 /* Older userspace won't unhalt the vcpu on reset. */
10120 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10121 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10122 !is_protmode(vcpu))
10123 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10125 skip_protected_regs:
10126 max_bits = KVM_NR_INTERRUPTS;
10127 pending_vec = find_first_bit(
10128 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10129 if (pending_vec < max_bits) {
10130 kvm_queue_interrupt(vcpu, pending_vec, false);
10131 pr_debug("Set back pending irq %d\n", pending_vec);
10134 kvm_make_request(KVM_REQ_EVENT, vcpu);
10141 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10142 struct kvm_sregs *sregs)
10147 ret = __set_sregs(vcpu, sregs);
10152 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10153 struct kvm_guest_debug *dbg)
10155 unsigned long rflags;
10158 if (vcpu->arch.guest_state_protected)
10163 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10165 if (vcpu->arch.exception.pending)
10167 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10168 kvm_queue_exception(vcpu, DB_VECTOR);
10170 kvm_queue_exception(vcpu, BP_VECTOR);
10174 * Read rflags as long as potentially injected trace flags are still
10177 rflags = kvm_get_rflags(vcpu);
10179 vcpu->guest_debug = dbg->control;
10180 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10181 vcpu->guest_debug = 0;
10183 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10184 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10185 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10186 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10188 for (i = 0; i < KVM_NR_DB_REGS; i++)
10189 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10191 kvm_update_dr7(vcpu);
10193 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10194 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10197 * Trigger an rflags update that will inject or remove the trace
10200 kvm_set_rflags(vcpu, rflags);
10202 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10212 * Translate a guest virtual address to a guest physical address.
10214 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10215 struct kvm_translation *tr)
10217 unsigned long vaddr = tr->linear_address;
10223 idx = srcu_read_lock(&vcpu->kvm->srcu);
10224 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10225 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10226 tr->physical_address = gpa;
10227 tr->valid = gpa != UNMAPPED_GVA;
10235 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10237 struct fxregs_state *fxsave;
10239 if (!vcpu->arch.guest_fpu)
10244 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10245 memcpy(fpu->fpr, fxsave->st_space, 128);
10246 fpu->fcw = fxsave->cwd;
10247 fpu->fsw = fxsave->swd;
10248 fpu->ftwx = fxsave->twd;
10249 fpu->last_opcode = fxsave->fop;
10250 fpu->last_ip = fxsave->rip;
10251 fpu->last_dp = fxsave->rdp;
10252 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10258 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10260 struct fxregs_state *fxsave;
10262 if (!vcpu->arch.guest_fpu)
10267 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10269 memcpy(fxsave->st_space, fpu->fpr, 128);
10270 fxsave->cwd = fpu->fcw;
10271 fxsave->swd = fpu->fsw;
10272 fxsave->twd = fpu->ftwx;
10273 fxsave->fop = fpu->last_opcode;
10274 fxsave->rip = fpu->last_ip;
10275 fxsave->rdp = fpu->last_dp;
10276 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10282 static void store_regs(struct kvm_vcpu *vcpu)
10284 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10286 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10287 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10289 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10290 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10292 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10293 kvm_vcpu_ioctl_x86_get_vcpu_events(
10294 vcpu, &vcpu->run->s.regs.events);
10297 static int sync_regs(struct kvm_vcpu *vcpu)
10299 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10302 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10303 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10304 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10306 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10307 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10309 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10311 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10312 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10313 vcpu, &vcpu->run->s.regs.events))
10315 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10321 static void fx_init(struct kvm_vcpu *vcpu)
10323 if (!vcpu->arch.guest_fpu)
10326 fpstate_init(&vcpu->arch.guest_fpu->state);
10327 if (boot_cpu_has(X86_FEATURE_XSAVES))
10328 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10329 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10332 * Ensure guest xcr0 is valid for loading
10334 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10336 vcpu->arch.cr0 |= X86_CR0_ET;
10339 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10341 if (vcpu->arch.guest_fpu) {
10342 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10343 vcpu->arch.guest_fpu = NULL;
10346 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10348 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10350 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10351 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10352 "guest TSC will not be reliable\n");
10357 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10362 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10363 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10365 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10367 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10369 r = kvm_mmu_create(vcpu);
10373 if (irqchip_in_kernel(vcpu->kvm)) {
10374 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10376 goto fail_mmu_destroy;
10377 if (kvm_apicv_activated(vcpu->kvm))
10378 vcpu->arch.apicv_active = true;
10380 static_branch_inc(&kvm_has_noapic_vcpu);
10384 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10386 goto fail_free_lapic;
10387 vcpu->arch.pio_data = page_address(page);
10389 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10390 GFP_KERNEL_ACCOUNT);
10391 if (!vcpu->arch.mce_banks)
10392 goto fail_free_pio_data;
10393 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10395 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10396 GFP_KERNEL_ACCOUNT))
10397 goto fail_free_mce_banks;
10399 if (!alloc_emulate_ctxt(vcpu))
10400 goto free_wbinvd_dirty_mask;
10402 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10403 GFP_KERNEL_ACCOUNT);
10404 if (!vcpu->arch.user_fpu) {
10405 pr_err("kvm: failed to allocate userspace's fpu\n");
10406 goto free_emulate_ctxt;
10409 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10410 GFP_KERNEL_ACCOUNT);
10411 if (!vcpu->arch.guest_fpu) {
10412 pr_err("kvm: failed to allocate vcpu's fpu\n");
10413 goto free_user_fpu;
10417 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10418 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10420 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10422 kvm_async_pf_hash_reset(vcpu);
10423 kvm_pmu_init(vcpu);
10425 vcpu->arch.pending_external_vector = -1;
10426 vcpu->arch.preempted_in_kernel = false;
10428 r = static_call(kvm_x86_vcpu_create)(vcpu);
10430 goto free_guest_fpu;
10432 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10433 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10434 kvm_vcpu_mtrr_init(vcpu);
10436 kvm_vcpu_reset(vcpu, false);
10437 kvm_init_mmu(vcpu, false);
10442 kvm_free_guest_fpu(vcpu);
10444 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10446 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10447 free_wbinvd_dirty_mask:
10448 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10449 fail_free_mce_banks:
10450 kfree(vcpu->arch.mce_banks);
10451 fail_free_pio_data:
10452 free_page((unsigned long)vcpu->arch.pio_data);
10454 kvm_free_lapic(vcpu);
10456 kvm_mmu_destroy(vcpu);
10460 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10462 struct kvm *kvm = vcpu->kvm;
10464 if (mutex_lock_killable(&vcpu->mutex))
10467 kvm_synchronize_tsc(vcpu, 0);
10470 /* poll control enabled by default */
10471 vcpu->arch.msr_kvm_poll_control = 1;
10473 mutex_unlock(&vcpu->mutex);
10475 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10476 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10477 KVMCLOCK_SYNC_PERIOD);
10480 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10482 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10485 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10487 kvmclock_reset(vcpu);
10489 static_call(kvm_x86_vcpu_free)(vcpu);
10491 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10492 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10493 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10494 kvm_free_guest_fpu(vcpu);
10496 kvm_hv_vcpu_uninit(vcpu);
10497 kvm_pmu_destroy(vcpu);
10498 kfree(vcpu->arch.mce_banks);
10499 kvm_free_lapic(vcpu);
10500 idx = srcu_read_lock(&vcpu->kvm->srcu);
10501 kvm_mmu_destroy(vcpu);
10502 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10503 free_page((unsigned long)vcpu->arch.pio_data);
10504 kvfree(vcpu->arch.cpuid_entries);
10505 if (!lapic_in_kernel(vcpu))
10506 static_branch_dec(&kvm_has_noapic_vcpu);
10509 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10511 kvm_lapic_reset(vcpu, init_event);
10513 vcpu->arch.hflags = 0;
10515 vcpu->arch.smi_pending = 0;
10516 vcpu->arch.smi_count = 0;
10517 atomic_set(&vcpu->arch.nmi_queued, 0);
10518 vcpu->arch.nmi_pending = 0;
10519 vcpu->arch.nmi_injected = false;
10520 kvm_clear_interrupt_queue(vcpu);
10521 kvm_clear_exception_queue(vcpu);
10523 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10524 kvm_update_dr0123(vcpu);
10525 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10526 vcpu->arch.dr7 = DR7_FIXED_1;
10527 kvm_update_dr7(vcpu);
10529 vcpu->arch.cr2 = 0;
10531 kvm_make_request(KVM_REQ_EVENT, vcpu);
10532 vcpu->arch.apf.msr_en_val = 0;
10533 vcpu->arch.apf.msr_int_val = 0;
10534 vcpu->arch.st.msr_val = 0;
10536 kvmclock_reset(vcpu);
10538 kvm_clear_async_pf_completion_queue(vcpu);
10539 kvm_async_pf_hash_reset(vcpu);
10540 vcpu->arch.apf.halted = false;
10542 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10543 void *mpx_state_buffer;
10546 * To avoid have the INIT path from kvm_apic_has_events() that be
10547 * called with loaded FPU and does not let userspace fix the state.
10550 kvm_put_guest_fpu(vcpu);
10551 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10553 if (mpx_state_buffer)
10554 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10555 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10557 if (mpx_state_buffer)
10558 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10560 kvm_load_guest_fpu(vcpu);
10564 kvm_pmu_reset(vcpu);
10565 vcpu->arch.smbase = 0x30000;
10567 vcpu->arch.msr_misc_features_enables = 0;
10569 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10572 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10573 vcpu->arch.regs_avail = ~0;
10574 vcpu->arch.regs_dirty = ~0;
10576 vcpu->arch.ia32_xss = 0;
10578 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10581 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10583 struct kvm_segment cs;
10585 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10586 cs.selector = vector << 8;
10587 cs.base = vector << 12;
10588 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10589 kvm_rip_write(vcpu, 0);
10591 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10593 int kvm_arch_hardware_enable(void)
10596 struct kvm_vcpu *vcpu;
10601 bool stable, backwards_tsc = false;
10603 kvm_user_return_msr_cpu_online();
10604 ret = static_call(kvm_x86_hardware_enable)();
10608 local_tsc = rdtsc();
10609 stable = !kvm_check_tsc_unstable();
10610 list_for_each_entry(kvm, &vm_list, vm_list) {
10611 kvm_for_each_vcpu(i, vcpu, kvm) {
10612 if (!stable && vcpu->cpu == smp_processor_id())
10613 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10614 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10615 backwards_tsc = true;
10616 if (vcpu->arch.last_host_tsc > max_tsc)
10617 max_tsc = vcpu->arch.last_host_tsc;
10623 * Sometimes, even reliable TSCs go backwards. This happens on
10624 * platforms that reset TSC during suspend or hibernate actions, but
10625 * maintain synchronization. We must compensate. Fortunately, we can
10626 * detect that condition here, which happens early in CPU bringup,
10627 * before any KVM threads can be running. Unfortunately, we can't
10628 * bring the TSCs fully up to date with real time, as we aren't yet far
10629 * enough into CPU bringup that we know how much real time has actually
10630 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10631 * variables that haven't been updated yet.
10633 * So we simply find the maximum observed TSC above, then record the
10634 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10635 * the adjustment will be applied. Note that we accumulate
10636 * adjustments, in case multiple suspend cycles happen before some VCPU
10637 * gets a chance to run again. In the event that no KVM threads get a
10638 * chance to run, we will miss the entire elapsed period, as we'll have
10639 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10640 * loose cycle time. This isn't too big a deal, since the loss will be
10641 * uniform across all VCPUs (not to mention the scenario is extremely
10642 * unlikely). It is possible that a second hibernate recovery happens
10643 * much faster than a first, causing the observed TSC here to be
10644 * smaller; this would require additional padding adjustment, which is
10645 * why we set last_host_tsc to the local tsc observed here.
10647 * N.B. - this code below runs only on platforms with reliable TSC,
10648 * as that is the only way backwards_tsc is set above. Also note
10649 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10650 * have the same delta_cyc adjustment applied if backwards_tsc
10651 * is detected. Note further, this adjustment is only done once,
10652 * as we reset last_host_tsc on all VCPUs to stop this from being
10653 * called multiple times (one for each physical CPU bringup).
10655 * Platforms with unreliable TSCs don't have to deal with this, they
10656 * will be compensated by the logic in vcpu_load, which sets the TSC to
10657 * catchup mode. This will catchup all VCPUs to real time, but cannot
10658 * guarantee that they stay in perfect synchronization.
10660 if (backwards_tsc) {
10661 u64 delta_cyc = max_tsc - local_tsc;
10662 list_for_each_entry(kvm, &vm_list, vm_list) {
10663 kvm->arch.backwards_tsc_observed = true;
10664 kvm_for_each_vcpu(i, vcpu, kvm) {
10665 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10666 vcpu->arch.last_host_tsc = local_tsc;
10667 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10671 * We have to disable TSC offset matching.. if you were
10672 * booting a VM while issuing an S4 host suspend....
10673 * you may have some problem. Solving this issue is
10674 * left as an exercise to the reader.
10676 kvm->arch.last_tsc_nsec = 0;
10677 kvm->arch.last_tsc_write = 0;
10684 void kvm_arch_hardware_disable(void)
10686 static_call(kvm_x86_hardware_disable)();
10687 drop_user_return_notifiers();
10690 int kvm_arch_hardware_setup(void *opaque)
10692 struct kvm_x86_init_ops *ops = opaque;
10695 rdmsrl_safe(MSR_EFER, &host_efer);
10697 if (boot_cpu_has(X86_FEATURE_XSAVES))
10698 rdmsrl(MSR_IA32_XSS, host_xss);
10700 r = ops->hardware_setup();
10704 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10705 kvm_ops_static_call_update();
10707 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10710 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10711 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10712 #undef __kvm_cpu_cap_has
10714 if (kvm_has_tsc_control) {
10716 * Make sure the user can only configure tsc_khz values that
10717 * fit into a signed integer.
10718 * A min value is not calculated because it will always
10719 * be 1 on all machines.
10721 u64 max = min(0x7fffffffULL,
10722 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10723 kvm_max_guest_tsc_khz = max;
10725 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10728 kvm_init_msr_list();
10732 void kvm_arch_hardware_unsetup(void)
10734 static_call(kvm_x86_hardware_unsetup)();
10737 int kvm_arch_check_processor_compat(void *opaque)
10739 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10740 struct kvm_x86_init_ops *ops = opaque;
10742 WARN_ON(!irqs_disabled());
10744 if (__cr4_reserved_bits(cpu_has, c) !=
10745 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10748 return ops->check_processor_compatibility();
10751 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10753 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10755 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10757 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10759 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10762 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10763 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10765 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10767 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10769 vcpu->arch.l1tf_flush_l1d = true;
10770 if (pmu->version && unlikely(pmu->event_count)) {
10771 pmu->need_cleanup = true;
10772 kvm_make_request(KVM_REQ_PMU, vcpu);
10774 static_call(kvm_x86_sched_in)(vcpu, cpu);
10777 void kvm_arch_free_vm(struct kvm *kvm)
10779 kfree(to_kvm_hv(kvm)->hv_pa_pg);
10784 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10789 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10790 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10791 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10792 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10793 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10794 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10796 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10797 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10798 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10799 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10800 &kvm->arch.irq_sources_bitmap);
10802 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10803 mutex_init(&kvm->arch.apic_map_lock);
10804 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10806 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10807 pvclock_update_vm_gtod_copy(kvm);
10809 kvm->arch.guest_can_read_msr_platform_info = true;
10811 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10812 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10814 kvm_hv_init_vm(kvm);
10815 kvm_page_track_init(kvm);
10816 kvm_mmu_init_vm(kvm);
10818 return static_call(kvm_x86_vm_init)(kvm);
10821 int kvm_arch_post_init_vm(struct kvm *kvm)
10823 return kvm_mmu_post_init_vm(kvm);
10826 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10829 kvm_mmu_unload(vcpu);
10833 static void kvm_free_vcpus(struct kvm *kvm)
10836 struct kvm_vcpu *vcpu;
10839 * Unpin any mmu pages first.
10841 kvm_for_each_vcpu(i, vcpu, kvm) {
10842 kvm_clear_async_pf_completion_queue(vcpu);
10843 kvm_unload_vcpu_mmu(vcpu);
10845 kvm_for_each_vcpu(i, vcpu, kvm)
10846 kvm_vcpu_destroy(vcpu);
10848 mutex_lock(&kvm->lock);
10849 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10850 kvm->vcpus[i] = NULL;
10852 atomic_set(&kvm->online_vcpus, 0);
10853 mutex_unlock(&kvm->lock);
10856 void kvm_arch_sync_events(struct kvm *kvm)
10858 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10859 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10863 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
10866 * __x86_set_memory_region: Setup KVM internal memory slot
10868 * @kvm: the kvm pointer to the VM.
10869 * @id: the slot ID to setup.
10870 * @gpa: the GPA to install the slot (unused when @size == 0).
10871 * @size: the size of the slot. Set to zero to uninstall a slot.
10873 * This function helps to setup a KVM internal memory slot. Specify
10874 * @size > 0 to install a new slot, while @size == 0 to uninstall a
10875 * slot. The return code can be one of the following:
10877 * HVA: on success (uninstall will return a bogus HVA)
10880 * The caller should always use IS_ERR() to check the return value
10881 * before use. Note, the KVM internal memory slots are guaranteed to
10882 * remain valid and unchanged until the VM is destroyed, i.e., the
10883 * GPA->HVA translation will not change. However, the HVA is a user
10884 * address, i.e. its accessibility is not guaranteed, and must be
10885 * accessed via __copy_{to,from}_user().
10887 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10891 unsigned long hva, old_npages;
10892 struct kvm_memslots *slots = kvm_memslots(kvm);
10893 struct kvm_memory_slot *slot;
10895 /* Called with kvm->slots_lock held. */
10896 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10897 return ERR_PTR_USR(-EINVAL);
10899 slot = id_to_memslot(slots, id);
10901 if (slot && slot->npages)
10902 return ERR_PTR_USR(-EEXIST);
10905 * MAP_SHARED to prevent internal slot pages from being moved
10908 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10909 MAP_SHARED | MAP_ANONYMOUS, 0);
10910 if (IS_ERR((void *)hva))
10911 return (void __user *)hva;
10913 if (!slot || !slot->npages)
10916 old_npages = slot->npages;
10917 hva = slot->userspace_addr;
10920 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10921 struct kvm_userspace_memory_region m;
10923 m.slot = id | (i << 16);
10925 m.guest_phys_addr = gpa;
10926 m.userspace_addr = hva;
10927 m.memory_size = size;
10928 r = __kvm_set_memory_region(kvm, &m);
10930 return ERR_PTR_USR(r);
10934 vm_munmap(hva, old_npages * PAGE_SIZE);
10936 return (void __user *)hva;
10938 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10940 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10942 kvm_mmu_pre_destroy_vm(kvm);
10945 void kvm_arch_destroy_vm(struct kvm *kvm)
10947 if (current->mm == kvm->mm) {
10949 * Free memory regions allocated on behalf of userspace,
10950 * unless the the memory map has changed due to process exit
10953 mutex_lock(&kvm->slots_lock);
10954 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10956 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10958 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10959 mutex_unlock(&kvm->slots_lock);
10961 static_call_cond(kvm_x86_vm_destroy)(kvm);
10962 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
10963 kvm_pic_destroy(kvm);
10964 kvm_ioapic_destroy(kvm);
10965 kvm_free_vcpus(kvm);
10966 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10967 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10968 kvm_mmu_uninit_vm(kvm);
10969 kvm_page_track_cleanup(kvm);
10970 kvm_xen_destroy_vm(kvm);
10971 kvm_hv_destroy_vm(kvm);
10974 static void memslot_rmap_free(struct kvm_memory_slot *slot)
10978 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10979 kvfree(slot->arch.rmap[i]);
10980 slot->arch.rmap[i] = NULL;
10984 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10988 memslot_rmap_free(slot);
10990 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
10991 kvfree(slot->arch.lpage_info[i - 1]);
10992 slot->arch.lpage_info[i - 1] = NULL;
10995 kvm_page_track_free_memslot(slot);
10998 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
10999 unsigned long npages)
11001 const int sz = sizeof(*slot->arch.rmap[0]);
11004 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11006 int lpages = gfn_to_index(slot->base_gfn + npages - 1,
11007 slot->base_gfn, level) + 1;
11009 WARN_ON(slot->arch.rmap[i]);
11011 slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11012 if (!slot->arch.rmap[i]) {
11013 memslot_rmap_free(slot);
11021 int alloc_all_memslots_rmaps(struct kvm *kvm)
11023 struct kvm_memslots *slots;
11024 struct kvm_memory_slot *slot;
11028 * Check if memslots alreday have rmaps early before acquiring
11029 * the slots_arch_lock below.
11031 if (kvm_memslots_have_rmaps(kvm))
11034 mutex_lock(&kvm->slots_arch_lock);
11037 * Read memslots_have_rmaps again, under the slots arch lock,
11038 * before allocating the rmaps
11040 if (kvm_memslots_have_rmaps(kvm)) {
11041 mutex_unlock(&kvm->slots_arch_lock);
11045 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11046 slots = __kvm_memslots(kvm, i);
11047 kvm_for_each_memslot(slot, slots) {
11048 r = memslot_rmap_alloc(slot, slot->npages);
11050 mutex_unlock(&kvm->slots_arch_lock);
11057 * Ensure that memslots_have_rmaps becomes true strictly after
11058 * all the rmap pointers are set.
11060 smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11061 mutex_unlock(&kvm->slots_arch_lock);
11065 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11066 struct kvm_memory_slot *slot,
11067 unsigned long npages)
11072 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
11073 * old arrays will be freed by __kvm_set_memory_region() if installing
11074 * the new memslot is successful.
11076 memset(&slot->arch, 0, sizeof(slot->arch));
11078 if (kvm_memslots_have_rmaps(kvm)) {
11079 r = memslot_rmap_alloc(slot, npages);
11084 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11085 struct kvm_lpage_info *linfo;
11086 unsigned long ugfn;
11090 lpages = gfn_to_index(slot->base_gfn + npages - 1,
11091 slot->base_gfn, level) + 1;
11093 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11097 slot->arch.lpage_info[i - 1] = linfo;
11099 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11100 linfo[0].disallow_lpage = 1;
11101 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11102 linfo[lpages - 1].disallow_lpage = 1;
11103 ugfn = slot->userspace_addr >> PAGE_SHIFT;
11105 * If the gfn and userspace address are not aligned wrt each
11106 * other, disable large page support for this slot.
11108 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11111 for (j = 0; j < lpages; ++j)
11112 linfo[j].disallow_lpage = 1;
11116 if (kvm_page_track_create_memslot(slot, npages))
11122 memslot_rmap_free(slot);
11124 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11125 kvfree(slot->arch.lpage_info[i - 1]);
11126 slot->arch.lpage_info[i - 1] = NULL;
11131 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11133 struct kvm_vcpu *vcpu;
11137 * memslots->generation has been incremented.
11138 * mmio generation may have reached its maximum value.
11140 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11142 /* Force re-initialization of steal_time cache */
11143 kvm_for_each_vcpu(i, vcpu, kvm)
11144 kvm_vcpu_kick(vcpu);
11147 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11148 struct kvm_memory_slot *memslot,
11149 const struct kvm_userspace_memory_region *mem,
11150 enum kvm_mr_change change)
11152 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11153 return kvm_alloc_memslot_metadata(kvm, memslot,
11154 mem->memory_size >> PAGE_SHIFT);
11159 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11161 struct kvm_arch *ka = &kvm->arch;
11163 if (!kvm_x86_ops.cpu_dirty_log_size)
11166 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11167 (!enable && --ka->cpu_dirty_logging_count == 0))
11168 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11170 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11173 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11174 struct kvm_memory_slot *old,
11175 struct kvm_memory_slot *new,
11176 enum kvm_mr_change change)
11178 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11181 * Update CPU dirty logging if dirty logging is being toggled. This
11182 * applies to all operations.
11184 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11185 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11188 * Nothing more to do for RO slots (which can't be dirtied and can't be
11189 * made writable) or CREATE/MOVE/DELETE of a slot.
11191 * For a memslot with dirty logging disabled:
11192 * CREATE: No dirty mappings will already exist.
11193 * MOVE/DELETE: The old mappings will already have been cleaned up by
11194 * kvm_arch_flush_shadow_memslot()
11196 * For a memslot with dirty logging enabled:
11197 * CREATE: No shadow pages exist, thus nothing to write-protect
11198 * and no dirty bits to clear.
11199 * MOVE/DELETE: The old mappings will already have been cleaned up by
11200 * kvm_arch_flush_shadow_memslot().
11202 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11206 * READONLY and non-flags changes were filtered out above, and the only
11207 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11208 * logging isn't being toggled on or off.
11210 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11213 if (!log_dirty_pages) {
11215 * Dirty logging tracks sptes in 4k granularity, meaning that
11216 * large sptes have to be split. If live migration succeeds,
11217 * the guest in the source machine will be destroyed and large
11218 * sptes will be created in the destination. However, if the
11219 * guest continues to run in the source machine (for example if
11220 * live migration fails), small sptes will remain around and
11221 * cause bad performance.
11223 * Scan sptes if dirty logging has been stopped, dropping those
11224 * which can be collapsed into a single large-page spte. Later
11225 * page faults will create the large-page sptes.
11227 kvm_mmu_zap_collapsible_sptes(kvm, new);
11230 * Initially-all-set does not require write protecting any page,
11231 * because they're all assumed to be dirty.
11233 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11236 if (kvm_x86_ops.cpu_dirty_log_size) {
11237 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11238 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11240 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11245 void kvm_arch_commit_memory_region(struct kvm *kvm,
11246 const struct kvm_userspace_memory_region *mem,
11247 struct kvm_memory_slot *old,
11248 const struct kvm_memory_slot *new,
11249 enum kvm_mr_change change)
11251 if (!kvm->arch.n_requested_mmu_pages)
11252 kvm_mmu_change_mmu_pages(kvm,
11253 kvm_mmu_calculate_default_mmu_pages(kvm));
11256 * FIXME: const-ify all uses of struct kvm_memory_slot.
11258 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11260 /* Free the arrays associated with the old memslot. */
11261 if (change == KVM_MR_MOVE)
11262 kvm_arch_free_memslot(kvm, old);
11265 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11267 kvm_mmu_zap_all(kvm);
11270 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11271 struct kvm_memory_slot *slot)
11273 kvm_page_track_flush_slot(kvm, slot);
11276 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11278 return (is_guest_mode(vcpu) &&
11279 kvm_x86_ops.guest_apic_has_interrupt &&
11280 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11283 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11285 if (!list_empty_careful(&vcpu->async_pf.done))
11288 if (kvm_apic_has_events(vcpu))
11291 if (vcpu->arch.pv.pv_unhalted)
11294 if (vcpu->arch.exception.pending)
11297 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11298 (vcpu->arch.nmi_pending &&
11299 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11302 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11303 (vcpu->arch.smi_pending &&
11304 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11307 if (kvm_arch_interrupt_allowed(vcpu) &&
11308 (kvm_cpu_has_interrupt(vcpu) ||
11309 kvm_guest_apic_has_interrupt(vcpu)))
11312 if (kvm_hv_has_stimer_pending(vcpu))
11315 if (is_guest_mode(vcpu) &&
11316 kvm_x86_ops.nested_ops->hv_timer_pending &&
11317 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11323 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11325 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11328 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11330 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11336 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11338 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11341 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11342 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11343 kvm_test_request(KVM_REQ_EVENT, vcpu))
11346 return kvm_arch_dy_has_pending_interrupt(vcpu);
11349 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11351 if (vcpu->arch.guest_state_protected)
11354 return vcpu->arch.preempted_in_kernel;
11357 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11359 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11362 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11364 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11367 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11369 /* Can't read the RIP when guest state is protected, just return 0 */
11370 if (vcpu->arch.guest_state_protected)
11373 if (is_64_bit_mode(vcpu))
11374 return kvm_rip_read(vcpu);
11375 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11376 kvm_rip_read(vcpu));
11378 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11380 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11382 return kvm_get_linear_rip(vcpu) == linear_rip;
11384 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11386 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11388 unsigned long rflags;
11390 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11391 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11392 rflags &= ~X86_EFLAGS_TF;
11395 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11397 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11399 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11400 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11401 rflags |= X86_EFLAGS_TF;
11402 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11405 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11407 __kvm_set_rflags(vcpu, rflags);
11408 kvm_make_request(KVM_REQ_EVENT, vcpu);
11410 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11412 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11416 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11420 r = kvm_mmu_reload(vcpu);
11424 if (!vcpu->arch.mmu->direct_map &&
11425 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11428 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11431 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11433 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11435 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11438 static inline u32 kvm_async_pf_next_probe(u32 key)
11440 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11443 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11445 u32 key = kvm_async_pf_hash_fn(gfn);
11447 while (vcpu->arch.apf.gfns[key] != ~0)
11448 key = kvm_async_pf_next_probe(key);
11450 vcpu->arch.apf.gfns[key] = gfn;
11453 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11456 u32 key = kvm_async_pf_hash_fn(gfn);
11458 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11459 (vcpu->arch.apf.gfns[key] != gfn &&
11460 vcpu->arch.apf.gfns[key] != ~0); i++)
11461 key = kvm_async_pf_next_probe(key);
11466 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11468 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11471 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11475 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11477 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11481 vcpu->arch.apf.gfns[i] = ~0;
11483 j = kvm_async_pf_next_probe(j);
11484 if (vcpu->arch.apf.gfns[j] == ~0)
11486 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11488 * k lies cyclically in ]i,j]
11490 * |....j i.k.| or |.k..j i...|
11492 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11493 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11498 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11500 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11502 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11506 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11508 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11510 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11511 &token, offset, sizeof(token));
11514 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11516 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11519 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11520 &val, offset, sizeof(val)))
11526 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11528 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11531 if (!kvm_pv_async_pf_enabled(vcpu) ||
11532 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11538 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11540 if (unlikely(!lapic_in_kernel(vcpu) ||
11541 kvm_event_needs_reinjection(vcpu) ||
11542 vcpu->arch.exception.pending))
11545 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11549 * If interrupts are off we cannot even use an artificial
11552 return kvm_arch_interrupt_allowed(vcpu);
11555 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11556 struct kvm_async_pf *work)
11558 struct x86_exception fault;
11560 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11561 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11563 if (kvm_can_deliver_async_pf(vcpu) &&
11564 !apf_put_user_notpresent(vcpu)) {
11565 fault.vector = PF_VECTOR;
11566 fault.error_code_valid = true;
11567 fault.error_code = 0;
11568 fault.nested_page_fault = false;
11569 fault.address = work->arch.token;
11570 fault.async_page_fault = true;
11571 kvm_inject_page_fault(vcpu, &fault);
11575 * It is not possible to deliver a paravirtualized asynchronous
11576 * page fault, but putting the guest in an artificial halt state
11577 * can be beneficial nevertheless: if an interrupt arrives, we
11578 * can deliver it timely and perhaps the guest will schedule
11579 * another process. When the instruction that triggered a page
11580 * fault is retried, hopefully the page will be ready in the host.
11582 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11587 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11588 struct kvm_async_pf *work)
11590 struct kvm_lapic_irq irq = {
11591 .delivery_mode = APIC_DM_FIXED,
11592 .vector = vcpu->arch.apf.vec
11595 if (work->wakeup_all)
11596 work->arch.token = ~0; /* broadcast wakeup */
11598 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11599 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11601 if ((work->wakeup_all || work->notpresent_injected) &&
11602 kvm_pv_async_pf_enabled(vcpu) &&
11603 !apf_put_user_ready(vcpu, work->arch.token)) {
11604 vcpu->arch.apf.pageready_pending = true;
11605 kvm_apic_set_irq(vcpu, &irq, NULL);
11608 vcpu->arch.apf.halted = false;
11609 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11612 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11614 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11615 if (!vcpu->arch.apf.pageready_pending)
11616 kvm_vcpu_kick(vcpu);
11619 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11621 if (!kvm_pv_async_pf_enabled(vcpu))
11624 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11627 void kvm_arch_start_assignment(struct kvm *kvm)
11629 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
11630 static_call_cond(kvm_x86_start_assignment)(kvm);
11632 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11634 void kvm_arch_end_assignment(struct kvm *kvm)
11636 atomic_dec(&kvm->arch.assigned_device_count);
11638 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11640 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11642 return atomic_read(&kvm->arch.assigned_device_count);
11644 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11646 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11648 atomic_inc(&kvm->arch.noncoherent_dma_count);
11650 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11652 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11654 atomic_dec(&kvm->arch.noncoherent_dma_count);
11656 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11658 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11660 return atomic_read(&kvm->arch.noncoherent_dma_count);
11662 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11664 bool kvm_arch_has_irq_bypass(void)
11669 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11670 struct irq_bypass_producer *prod)
11672 struct kvm_kernel_irqfd *irqfd =
11673 container_of(cons, struct kvm_kernel_irqfd, consumer);
11676 irqfd->producer = prod;
11677 kvm_arch_start_assignment(irqfd->kvm);
11678 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11679 prod->irq, irqfd->gsi, 1);
11682 kvm_arch_end_assignment(irqfd->kvm);
11687 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11688 struct irq_bypass_producer *prod)
11691 struct kvm_kernel_irqfd *irqfd =
11692 container_of(cons, struct kvm_kernel_irqfd, consumer);
11694 WARN_ON(irqfd->producer != prod);
11695 irqfd->producer = NULL;
11698 * When producer of consumer is unregistered, we change back to
11699 * remapped mode, so we can re-use the current implementation
11700 * when the irq is masked/disabled or the consumer side (KVM
11701 * int this case doesn't want to receive the interrupts.
11703 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11705 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11706 " fails: %d\n", irqfd->consumer.token, ret);
11708 kvm_arch_end_assignment(irqfd->kvm);
11711 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11712 uint32_t guest_irq, bool set)
11714 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11717 bool kvm_vector_hashing_enabled(void)
11719 return vector_hashing;
11722 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11724 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11726 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11729 int kvm_spec_ctrl_test_value(u64 value)
11732 * test that setting IA32_SPEC_CTRL to given value
11733 * is allowed by the host processor
11737 unsigned long flags;
11740 local_irq_save(flags);
11742 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11744 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11747 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11749 local_irq_restore(flags);
11753 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11755 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11757 struct x86_exception fault;
11758 u32 access = error_code &
11759 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11761 if (!(error_code & PFERR_PRESENT_MASK) ||
11762 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11764 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11765 * tables probably do not match the TLB. Just proceed
11766 * with the error code that the processor gave.
11768 fault.vector = PF_VECTOR;
11769 fault.error_code_valid = true;
11770 fault.error_code = error_code;
11771 fault.nested_page_fault = false;
11772 fault.address = gva;
11774 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11776 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11779 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11780 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11781 * indicates whether exit to userspace is needed.
11783 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11784 struct x86_exception *e)
11786 if (r == X86EMUL_PROPAGATE_FAULT) {
11787 kvm_inject_emulated_page_fault(vcpu, e);
11792 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11793 * while handling a VMX instruction KVM could've handled the request
11794 * correctly by exiting to userspace and performing I/O but there
11795 * doesn't seem to be a real use-case behind such requests, just return
11796 * KVM_EXIT_INTERNAL_ERROR for now.
11798 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11799 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11800 vcpu->run->internal.ndata = 0;
11804 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11806 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11809 struct x86_exception e;
11811 unsigned long roots_to_free = 0;
11818 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11819 if (r != X86EMUL_CONTINUE)
11820 return kvm_handle_memory_failure(vcpu, r, &e);
11822 if (operand.pcid >> 12 != 0) {
11823 kvm_inject_gp(vcpu, 0);
11827 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11830 case INVPCID_TYPE_INDIV_ADDR:
11831 if ((!pcid_enabled && (operand.pcid != 0)) ||
11832 is_noncanonical_address(operand.gla, vcpu)) {
11833 kvm_inject_gp(vcpu, 0);
11836 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11837 return kvm_skip_emulated_instruction(vcpu);
11839 case INVPCID_TYPE_SINGLE_CTXT:
11840 if (!pcid_enabled && (operand.pcid != 0)) {
11841 kvm_inject_gp(vcpu, 0);
11845 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11846 kvm_mmu_sync_roots(vcpu);
11847 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11850 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11851 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11853 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11855 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11857 * If neither the current cr3 nor any of the prev_roots use the
11858 * given PCID, then nothing needs to be done here because a
11859 * resync will happen anyway before switching to any other CR3.
11862 return kvm_skip_emulated_instruction(vcpu);
11864 case INVPCID_TYPE_ALL_NON_GLOBAL:
11866 * Currently, KVM doesn't mark global entries in the shadow
11867 * page tables, so a non-global flush just degenerates to a
11868 * global flush. If needed, we could optimize this later by
11869 * keeping track of global entries in shadow page tables.
11873 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11874 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
11875 return kvm_skip_emulated_instruction(vcpu);
11878 BUG(); /* We have already checked above that type <= 3 */
11881 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11883 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11885 struct kvm_run *run = vcpu->run;
11886 struct kvm_mmio_fragment *frag;
11889 BUG_ON(!vcpu->mmio_needed);
11891 /* Complete previous fragment */
11892 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11893 len = min(8u, frag->len);
11894 if (!vcpu->mmio_is_write)
11895 memcpy(frag->data, run->mmio.data, len);
11897 if (frag->len <= 8) {
11898 /* Switch to the next fragment. */
11900 vcpu->mmio_cur_fragment++;
11902 /* Go forward to the next mmio piece. */
11908 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11909 vcpu->mmio_needed = 0;
11911 // VMG change, at this point, we're always done
11912 // RIP has already been advanced
11916 // More MMIO is needed
11917 run->mmio.phys_addr = frag->gpa;
11918 run->mmio.len = min(8u, frag->len);
11919 run->mmio.is_write = vcpu->mmio_is_write;
11920 if (run->mmio.is_write)
11921 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11922 run->exit_reason = KVM_EXIT_MMIO;
11924 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11929 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11933 struct kvm_mmio_fragment *frag;
11938 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11939 if (handled == bytes)
11946 /*TODO: Check if need to increment number of frags */
11947 frag = vcpu->mmio_fragments;
11948 vcpu->mmio_nr_fragments = 1;
11953 vcpu->mmio_needed = 1;
11954 vcpu->mmio_cur_fragment = 0;
11956 vcpu->run->mmio.phys_addr = gpa;
11957 vcpu->run->mmio.len = min(8u, frag->len);
11958 vcpu->run->mmio.is_write = 1;
11959 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11960 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11962 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11966 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11968 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11972 struct kvm_mmio_fragment *frag;
11977 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11978 if (handled == bytes)
11985 /*TODO: Check if need to increment number of frags */
11986 frag = vcpu->mmio_fragments;
11987 vcpu->mmio_nr_fragments = 1;
11992 vcpu->mmio_needed = 1;
11993 vcpu->mmio_cur_fragment = 0;
11995 vcpu->run->mmio.phys_addr = gpa;
11996 vcpu->run->mmio.len = min(8u, frag->len);
11997 vcpu->run->mmio.is_write = 0;
11998 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12000 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12004 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12006 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12008 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
12009 vcpu->arch.pio.count * vcpu->arch.pio.size);
12010 vcpu->arch.pio.count = 0;
12015 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12016 unsigned int port, void *data, unsigned int count)
12020 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
12025 vcpu->arch.pio.count = 0;
12030 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12031 unsigned int port, void *data, unsigned int count)
12035 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
12038 vcpu->arch.pio.count = 0;
12040 vcpu->arch.guest_ins_data = data;
12041 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12047 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12048 unsigned int port, void *data, unsigned int count,
12051 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
12052 : kvm_sev_es_outs(vcpu, size, port, data, count);
12054 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12056 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12057 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12058 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12059 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12060 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12061 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12062 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12063 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12064 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12065 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12066 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12067 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12068 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12069 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12070 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12071 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12072 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12073 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12074 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12075 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12076 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12077 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12078 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12079 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12080 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12081 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12082 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);