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>
78 #include <clocksource/hyperv_timer.h>
80 #define CREATE_TRACE_POINTS
83 #define MAX_IO_MSRS 256
84 #define KVM_MAX_MCE_BANKS 32
85 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
86 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
88 #define emul_to_vcpu(ctxt) \
89 ((struct kvm_vcpu *)(ctxt)->vcpu)
92 * - enable syscall per default because its emulated by KVM
93 * - enable LME and LMA per default on 64 bit KVM
97 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
99 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
102 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
104 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
105 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
107 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
108 static void process_nmi(struct kvm_vcpu *vcpu);
109 static void process_smi(struct kvm_vcpu *vcpu);
110 static void enter_smm(struct kvm_vcpu *vcpu);
111 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
112 static void store_regs(struct kvm_vcpu *vcpu);
113 static int sync_regs(struct kvm_vcpu *vcpu);
115 struct kvm_x86_ops kvm_x86_ops __read_mostly;
116 EXPORT_SYMBOL_GPL(kvm_x86_ops);
118 #define KVM_X86_OP(func) \
119 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
120 *(((struct kvm_x86_ops *)0)->func));
121 #define KVM_X86_OP_NULL KVM_X86_OP
122 #include <asm/kvm-x86-ops.h>
123 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
124 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
125 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
127 static bool __read_mostly ignore_msrs = 0;
128 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
130 bool __read_mostly report_ignored_msrs = true;
131 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
132 EXPORT_SYMBOL_GPL(report_ignored_msrs);
134 unsigned int min_timer_period_us = 200;
135 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
137 static bool __read_mostly kvmclock_periodic_sync = true;
138 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
140 bool __read_mostly kvm_has_tsc_control;
141 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
142 u32 __read_mostly kvm_max_guest_tsc_khz;
143 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
144 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
145 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
146 u64 __read_mostly kvm_max_tsc_scaling_ratio;
147 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
148 u64 __read_mostly kvm_default_tsc_scaling_ratio;
149 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
150 bool __read_mostly kvm_has_bus_lock_exit;
151 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
153 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
154 static u32 __read_mostly tsc_tolerance_ppm = 250;
155 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
158 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
159 * adaptive tuning starting from default advancment of 1000ns. '0' disables
160 * advancement entirely. Any other value is used as-is and disables adaptive
161 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
163 static int __read_mostly lapic_timer_advance_ns = -1;
164 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
166 static bool __read_mostly vector_hashing = true;
167 module_param(vector_hashing, bool, S_IRUGO);
169 bool __read_mostly enable_vmware_backdoor = false;
170 module_param(enable_vmware_backdoor, bool, S_IRUGO);
171 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
173 static bool __read_mostly force_emulation_prefix = false;
174 module_param(force_emulation_prefix, bool, S_IRUGO);
176 int __read_mostly pi_inject_timer = -1;
177 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
180 * Restoring the host value for MSRs that are only consumed when running in
181 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
182 * returns to userspace, i.e. the kernel can run with the guest's value.
184 #define KVM_MAX_NR_USER_RETURN_MSRS 16
186 struct kvm_user_return_msrs_global {
188 u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS];
191 struct kvm_user_return_msrs {
192 struct user_return_notifier urn;
194 struct kvm_user_return_msr_values {
197 } values[KVM_MAX_NR_USER_RETURN_MSRS];
200 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global;
201 static struct kvm_user_return_msrs __percpu *user_return_msrs;
203 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
204 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
205 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
206 | XFEATURE_MASK_PKRU)
208 u64 __read_mostly host_efer;
209 EXPORT_SYMBOL_GPL(host_efer);
211 bool __read_mostly allow_smaller_maxphyaddr = 0;
212 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
214 u64 __read_mostly host_xss;
215 EXPORT_SYMBOL_GPL(host_xss);
216 u64 __read_mostly supported_xss;
217 EXPORT_SYMBOL_GPL(supported_xss);
219 struct kvm_stats_debugfs_item debugfs_entries[] = {
220 VCPU_STAT("pf_fixed", pf_fixed),
221 VCPU_STAT("pf_guest", pf_guest),
222 VCPU_STAT("tlb_flush", tlb_flush),
223 VCPU_STAT("invlpg", invlpg),
224 VCPU_STAT("exits", exits),
225 VCPU_STAT("io_exits", io_exits),
226 VCPU_STAT("mmio_exits", mmio_exits),
227 VCPU_STAT("signal_exits", signal_exits),
228 VCPU_STAT("irq_window", irq_window_exits),
229 VCPU_STAT("nmi_window", nmi_window_exits),
230 VCPU_STAT("halt_exits", halt_exits),
231 VCPU_STAT("halt_successful_poll", halt_successful_poll),
232 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
233 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
234 VCPU_STAT("halt_wakeup", halt_wakeup),
235 VCPU_STAT("hypercalls", hypercalls),
236 VCPU_STAT("request_irq", request_irq_exits),
237 VCPU_STAT("irq_exits", irq_exits),
238 VCPU_STAT("host_state_reload", host_state_reload),
239 VCPU_STAT("fpu_reload", fpu_reload),
240 VCPU_STAT("insn_emulation", insn_emulation),
241 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
242 VCPU_STAT("irq_injections", irq_injections),
243 VCPU_STAT("nmi_injections", nmi_injections),
244 VCPU_STAT("req_event", req_event),
245 VCPU_STAT("l1d_flush", l1d_flush),
246 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
247 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
248 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
249 VM_STAT("mmu_pte_write", mmu_pte_write),
250 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
251 VM_STAT("mmu_flooded", mmu_flooded),
252 VM_STAT("mmu_recycled", mmu_recycled),
253 VM_STAT("mmu_cache_miss", mmu_cache_miss),
254 VM_STAT("mmu_unsync", mmu_unsync),
255 VM_STAT("remote_tlb_flush", remote_tlb_flush),
256 VM_STAT("largepages", lpages, .mode = 0444),
257 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
258 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
262 u64 __read_mostly host_xcr0;
263 u64 __read_mostly supported_xcr0;
264 EXPORT_SYMBOL_GPL(supported_xcr0);
266 static struct kmem_cache *x86_fpu_cache;
268 static struct kmem_cache *x86_emulator_cache;
271 * When called, it means the previous get/set msr reached an invalid msr.
272 * Return true if we want to ignore/silent this failed msr access.
274 static bool kvm_msr_ignored_check(struct kvm_vcpu *vcpu, u32 msr,
275 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 < user_return_msrs_global.nr; ++slot) {
331 values = &msrs->values[slot];
332 if (values->host != values->curr) {
333 wrmsrl(user_return_msrs_global.msrs[slot], values->host);
334 values->curr = values->host;
339 void kvm_define_user_return_msr(unsigned slot, u32 msr)
341 BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
342 user_return_msrs_global.msrs[slot] = msr;
343 if (slot >= user_return_msrs_global.nr)
344 user_return_msrs_global.nr = slot + 1;
346 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
348 static void kvm_user_return_msr_cpu_online(void)
350 unsigned int cpu = smp_processor_id();
351 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
355 for (i = 0; i < user_return_msrs_global.nr; ++i) {
356 rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
357 msrs->values[i].host = value;
358 msrs->values[i].curr = value;
362 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
364 unsigned int cpu = smp_processor_id();
365 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
368 value = (value & mask) | (msrs->values[slot].host & ~mask);
369 if (value == msrs->values[slot].curr)
371 err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
375 msrs->values[slot].curr = value;
376 if (!msrs->registered) {
377 msrs->urn.on_user_return = kvm_on_user_return;
378 user_return_notifier_register(&msrs->urn);
379 msrs->registered = true;
383 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
385 static void drop_user_return_notifiers(void)
387 unsigned int cpu = smp_processor_id();
388 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
390 if (msrs->registered)
391 kvm_on_user_return(&msrs->urn);
394 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
396 return vcpu->arch.apic_base;
398 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
400 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
402 return kvm_apic_mode(kvm_get_apic_base(vcpu));
404 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
406 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
408 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
409 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
410 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
411 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
413 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
415 if (!msr_info->host_initiated) {
416 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
418 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
422 kvm_lapic_set_base(vcpu, msr_info->data);
423 kvm_recalculate_apic_map(vcpu->kvm);
426 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
428 asmlinkage __visible noinstr void kvm_spurious_fault(void)
430 /* Fault while not rebooting. We want the trace. */
431 BUG_ON(!kvm_rebooting);
433 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
435 #define EXCPT_BENIGN 0
436 #define EXCPT_CONTRIBUTORY 1
439 static int exception_class(int vector)
449 return EXCPT_CONTRIBUTORY;
456 #define EXCPT_FAULT 0
458 #define EXCPT_ABORT 2
459 #define EXCPT_INTERRUPT 3
461 static int exception_type(int vector)
465 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
466 return EXCPT_INTERRUPT;
470 /* #DB is trap, as instruction watchpoints are handled elsewhere */
471 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
474 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
477 /* Reserved exceptions will result in fault */
481 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
483 unsigned nr = vcpu->arch.exception.nr;
484 bool has_payload = vcpu->arch.exception.has_payload;
485 unsigned long payload = vcpu->arch.exception.payload;
493 * "Certain debug exceptions may clear bit 0-3. The
494 * remaining contents of the DR6 register are never
495 * cleared by the processor".
497 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
499 * In order to reflect the #DB exception payload in guest
500 * dr6, three components need to be considered: active low
501 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
503 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
504 * In the target guest dr6:
505 * FIXED_1 bits should always be set.
506 * Active low bits should be cleared if 1-setting in payload.
507 * Active high bits should be set if 1-setting in payload.
509 * Note, the payload is compatible with the pending debug
510 * exceptions/exit qualification under VMX, that active_low bits
511 * are active high in payload.
512 * So they need to be flipped for DR6.
514 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
515 vcpu->arch.dr6 |= payload;
516 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
519 * The #DB payload is defined as compatible with the 'pending
520 * debug exceptions' field under VMX, not DR6. While bit 12 is
521 * defined in the 'pending debug exceptions' field (enabled
522 * breakpoint), it is reserved and must be zero in DR6.
524 vcpu->arch.dr6 &= ~BIT(12);
527 vcpu->arch.cr2 = payload;
531 vcpu->arch.exception.has_payload = false;
532 vcpu->arch.exception.payload = 0;
534 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
536 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
537 unsigned nr, bool has_error, u32 error_code,
538 bool has_payload, unsigned long payload, bool reinject)
543 kvm_make_request(KVM_REQ_EVENT, vcpu);
545 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
547 if (has_error && !is_protmode(vcpu))
551 * On vmentry, vcpu->arch.exception.pending is only
552 * true if an event injection was blocked by
553 * nested_run_pending. In that case, however,
554 * vcpu_enter_guest requests an immediate exit,
555 * and the guest shouldn't proceed far enough to
558 WARN_ON_ONCE(vcpu->arch.exception.pending);
559 vcpu->arch.exception.injected = true;
560 if (WARN_ON_ONCE(has_payload)) {
562 * A reinjected event has already
563 * delivered its payload.
569 vcpu->arch.exception.pending = true;
570 vcpu->arch.exception.injected = false;
572 vcpu->arch.exception.has_error_code = has_error;
573 vcpu->arch.exception.nr = nr;
574 vcpu->arch.exception.error_code = error_code;
575 vcpu->arch.exception.has_payload = has_payload;
576 vcpu->arch.exception.payload = payload;
577 if (!is_guest_mode(vcpu))
578 kvm_deliver_exception_payload(vcpu);
582 /* to check exception */
583 prev_nr = vcpu->arch.exception.nr;
584 if (prev_nr == DF_VECTOR) {
585 /* triple fault -> shutdown */
586 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
589 class1 = exception_class(prev_nr);
590 class2 = exception_class(nr);
591 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
592 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
594 * Generate double fault per SDM Table 5-5. Set
595 * exception.pending = true so that the double fault
596 * can trigger a nested vmexit.
598 vcpu->arch.exception.pending = true;
599 vcpu->arch.exception.injected = false;
600 vcpu->arch.exception.has_error_code = true;
601 vcpu->arch.exception.nr = DF_VECTOR;
602 vcpu->arch.exception.error_code = 0;
603 vcpu->arch.exception.has_payload = false;
604 vcpu->arch.exception.payload = 0;
606 /* replace previous exception with a new one in a hope
607 that instruction re-execution will regenerate lost
612 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
614 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
616 EXPORT_SYMBOL_GPL(kvm_queue_exception);
618 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
620 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
622 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
624 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
625 unsigned long payload)
627 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
629 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
631 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
632 u32 error_code, unsigned long payload)
634 kvm_multiple_exception(vcpu, nr, true, error_code,
635 true, payload, false);
638 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
641 kvm_inject_gp(vcpu, 0);
643 return kvm_skip_emulated_instruction(vcpu);
647 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
649 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
651 ++vcpu->stat.pf_guest;
652 vcpu->arch.exception.nested_apf =
653 is_guest_mode(vcpu) && fault->async_page_fault;
654 if (vcpu->arch.exception.nested_apf) {
655 vcpu->arch.apf.nested_apf_token = fault->address;
656 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
658 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
662 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
664 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
665 struct x86_exception *fault)
667 struct kvm_mmu *fault_mmu;
668 WARN_ON_ONCE(fault->vector != PF_VECTOR);
670 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
674 * Invalidate the TLB entry for the faulting address, if it exists,
675 * else the access will fault indefinitely (and to emulate hardware).
677 if ((fault->error_code & PFERR_PRESENT_MASK) &&
678 !(fault->error_code & PFERR_RSVD_MASK))
679 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
680 fault_mmu->root_hpa);
682 fault_mmu->inject_page_fault(vcpu, fault);
683 return fault->nested_page_fault;
685 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
687 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
689 atomic_inc(&vcpu->arch.nmi_queued);
690 kvm_make_request(KVM_REQ_NMI, vcpu);
692 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
694 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
696 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
698 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
700 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
702 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
704 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
707 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
708 * a #GP and return false.
710 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
712 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
714 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
717 EXPORT_SYMBOL_GPL(kvm_require_cpl);
719 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
721 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
724 kvm_queue_exception(vcpu, UD_VECTOR);
727 EXPORT_SYMBOL_GPL(kvm_require_dr);
730 * This function will be used to read from the physical memory of the currently
731 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
732 * can read from guest physical or from the guest's guest physical memory.
734 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
735 gfn_t ngfn, void *data, int offset, int len,
738 struct x86_exception exception;
742 ngpa = gfn_to_gpa(ngfn);
743 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
744 if (real_gfn == UNMAPPED_GVA)
747 real_gfn = gpa_to_gfn(real_gfn);
749 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
751 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
753 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
754 void *data, int offset, int len, u32 access)
756 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
757 data, offset, len, access);
760 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
762 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
766 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
768 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
770 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
771 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
774 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
776 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
777 offset * sizeof(u64), sizeof(pdpte),
778 PFERR_USER_MASK|PFERR_WRITE_MASK);
783 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
784 if ((pdpte[i] & PT_PRESENT_MASK) &&
785 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
792 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
793 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
799 EXPORT_SYMBOL_GPL(load_pdptrs);
801 bool pdptrs_changed(struct kvm_vcpu *vcpu)
803 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
808 if (!is_pae_paging(vcpu))
811 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
814 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
815 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
816 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
817 PFERR_USER_MASK | PFERR_WRITE_MASK);
821 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
823 EXPORT_SYMBOL_GPL(pdptrs_changed);
825 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
827 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
829 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
830 kvm_clear_async_pf_completion_queue(vcpu);
831 kvm_async_pf_hash_reset(vcpu);
834 if ((cr0 ^ old_cr0) & update_bits)
835 kvm_mmu_reset_context(vcpu);
837 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
838 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
839 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
840 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
842 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
844 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
846 unsigned long old_cr0 = kvm_read_cr0(vcpu);
847 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
852 if (cr0 & 0xffffffff00000000UL)
856 cr0 &= ~CR0_RESERVED_BITS;
858 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
861 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
865 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
866 (cr0 & X86_CR0_PG)) {
871 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
876 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
877 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
878 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
881 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
884 static_call(kvm_x86_set_cr0)(vcpu, cr0);
886 kvm_post_set_cr0(vcpu, old_cr0, cr0);
890 EXPORT_SYMBOL_GPL(kvm_set_cr0);
892 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
894 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
896 EXPORT_SYMBOL_GPL(kvm_lmsw);
898 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
900 if (vcpu->arch.guest_state_protected)
903 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
905 if (vcpu->arch.xcr0 != host_xcr0)
906 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
908 if (vcpu->arch.xsaves_enabled &&
909 vcpu->arch.ia32_xss != host_xss)
910 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
913 if (static_cpu_has(X86_FEATURE_PKU) &&
914 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
915 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
916 vcpu->arch.pkru != vcpu->arch.host_pkru)
917 __write_pkru(vcpu->arch.pkru);
919 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
921 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
923 if (vcpu->arch.guest_state_protected)
926 if (static_cpu_has(X86_FEATURE_PKU) &&
927 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
928 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
929 vcpu->arch.pkru = rdpkru();
930 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
931 __write_pkru(vcpu->arch.host_pkru);
934 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
936 if (vcpu->arch.xcr0 != host_xcr0)
937 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
939 if (vcpu->arch.xsaves_enabled &&
940 vcpu->arch.ia32_xss != host_xss)
941 wrmsrl(MSR_IA32_XSS, host_xss);
945 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
947 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
950 u64 old_xcr0 = vcpu->arch.xcr0;
953 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
954 if (index != XCR_XFEATURE_ENABLED_MASK)
956 if (!(xcr0 & XFEATURE_MASK_FP))
958 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
962 * Do not allow the guest to set bits that we do not support
963 * saving. However, xcr0 bit 0 is always set, even if the
964 * emulated CPU does not support XSAVE (see fx_init).
966 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
967 if (xcr0 & ~valid_bits)
970 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
971 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
974 if (xcr0 & XFEATURE_MASK_AVX512) {
975 if (!(xcr0 & XFEATURE_MASK_YMM))
977 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
980 vcpu->arch.xcr0 = xcr0;
982 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
983 kvm_update_cpuid_runtime(vcpu);
987 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
989 if (static_call(kvm_x86_get_cpl)(vcpu) == 0)
990 return __kvm_set_xcr(vcpu, index, xcr);
994 EXPORT_SYMBOL_GPL(kvm_set_xcr);
996 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
998 if (cr4 & cr4_reserved_bits)
1001 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1004 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1006 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1008 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1010 unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1011 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
1013 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1014 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1015 kvm_mmu_reset_context(vcpu);
1017 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1019 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1021 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1022 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1025 if (!kvm_is_valid_cr4(vcpu, cr4))
1028 if (is_long_mode(vcpu)) {
1029 if (!(cr4 & X86_CR4_PAE))
1031 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1033 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1034 && ((cr4 ^ old_cr4) & pdptr_bits)
1035 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1036 kvm_read_cr3(vcpu)))
1039 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1040 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1043 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1044 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1048 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1050 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1054 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1056 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1058 bool skip_tlb_flush = false;
1059 #ifdef CONFIG_X86_64
1060 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1063 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1064 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1068 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1069 if (!skip_tlb_flush) {
1070 kvm_mmu_sync_roots(vcpu);
1071 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1076 if (is_long_mode(vcpu) && kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1078 else if (is_pae_paging(vcpu) &&
1079 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1082 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1083 vcpu->arch.cr3 = cr3;
1084 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1088 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1090 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1092 if (cr8 & CR8_RESERVED_BITS)
1094 if (lapic_in_kernel(vcpu))
1095 kvm_lapic_set_tpr(vcpu, cr8);
1097 vcpu->arch.cr8 = cr8;
1100 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1102 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1104 if (lapic_in_kernel(vcpu))
1105 return kvm_lapic_get_cr8(vcpu);
1107 return vcpu->arch.cr8;
1109 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1111 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1115 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1116 for (i = 0; i < KVM_NR_DB_REGS; i++)
1117 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1118 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1122 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1126 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1127 dr7 = vcpu->arch.guest_debug_dr7;
1129 dr7 = vcpu->arch.dr7;
1130 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1131 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1132 if (dr7 & DR7_BP_EN_MASK)
1133 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1135 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1137 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1139 u64 fixed = DR6_FIXED_1;
1141 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1146 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1148 size_t size = ARRAY_SIZE(vcpu->arch.db);
1152 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1153 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1154 vcpu->arch.eff_db[dr] = val;
1158 if (!kvm_dr6_valid(val))
1160 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1164 if (!kvm_dr7_valid(val))
1166 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1167 kvm_update_dr7(vcpu);
1173 EXPORT_SYMBOL_GPL(kvm_set_dr);
1175 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1177 size_t size = ARRAY_SIZE(vcpu->arch.db);
1181 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1185 *val = vcpu->arch.dr6;
1189 *val = vcpu->arch.dr7;
1193 EXPORT_SYMBOL_GPL(kvm_get_dr);
1195 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1197 u32 ecx = kvm_rcx_read(vcpu);
1201 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1204 kvm_rax_write(vcpu, (u32)data);
1205 kvm_rdx_write(vcpu, data >> 32);
1208 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1211 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1212 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1214 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1215 * extract the supported MSRs from the related const lists.
1216 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1217 * capabilities of the host cpu. This capabilities test skips MSRs that are
1218 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1219 * may depend on host virtualization features rather than host cpu features.
1222 static const u32 msrs_to_save_all[] = {
1223 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1225 #ifdef CONFIG_X86_64
1226 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1228 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1229 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1231 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1232 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1233 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1234 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1235 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1236 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1237 MSR_IA32_UMWAIT_CONTROL,
1239 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1240 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1241 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1242 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1243 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1244 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1245 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1246 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1247 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1248 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1249 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1250 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1251 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1252 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1253 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1254 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1255 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1256 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1257 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1258 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1259 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1260 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1263 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1264 static unsigned num_msrs_to_save;
1266 static const u32 emulated_msrs_all[] = {
1267 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1268 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1269 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1270 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1271 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1272 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1273 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1275 HV_X64_MSR_VP_INDEX,
1276 HV_X64_MSR_VP_RUNTIME,
1277 HV_X64_MSR_SCONTROL,
1278 HV_X64_MSR_STIMER0_CONFIG,
1279 HV_X64_MSR_VP_ASSIST_PAGE,
1280 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1281 HV_X64_MSR_TSC_EMULATION_STATUS,
1282 HV_X64_MSR_SYNDBG_OPTIONS,
1283 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1284 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1285 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1287 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1288 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1290 MSR_IA32_TSC_ADJUST,
1291 MSR_IA32_TSCDEADLINE,
1292 MSR_IA32_ARCH_CAPABILITIES,
1293 MSR_IA32_PERF_CAPABILITIES,
1294 MSR_IA32_MISC_ENABLE,
1295 MSR_IA32_MCG_STATUS,
1297 MSR_IA32_MCG_EXT_CTL,
1301 MSR_MISC_FEATURES_ENABLES,
1302 MSR_AMD64_VIRT_SPEC_CTRL,
1307 * The following list leaves out MSRs whose values are determined
1308 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1309 * We always support the "true" VMX control MSRs, even if the host
1310 * processor does not, so I am putting these registers here rather
1311 * than in msrs_to_save_all.
1314 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1315 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1316 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1317 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1319 MSR_IA32_VMX_CR0_FIXED0,
1320 MSR_IA32_VMX_CR4_FIXED0,
1321 MSR_IA32_VMX_VMCS_ENUM,
1322 MSR_IA32_VMX_PROCBASED_CTLS2,
1323 MSR_IA32_VMX_EPT_VPID_CAP,
1324 MSR_IA32_VMX_VMFUNC,
1327 MSR_KVM_POLL_CONTROL,
1330 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1331 static unsigned num_emulated_msrs;
1334 * List of msr numbers which are used to expose MSR-based features that
1335 * can be used by a hypervisor to validate requested CPU features.
1337 static const u32 msr_based_features_all[] = {
1339 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1340 MSR_IA32_VMX_PINBASED_CTLS,
1341 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1342 MSR_IA32_VMX_PROCBASED_CTLS,
1343 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1344 MSR_IA32_VMX_EXIT_CTLS,
1345 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1346 MSR_IA32_VMX_ENTRY_CTLS,
1348 MSR_IA32_VMX_CR0_FIXED0,
1349 MSR_IA32_VMX_CR0_FIXED1,
1350 MSR_IA32_VMX_CR4_FIXED0,
1351 MSR_IA32_VMX_CR4_FIXED1,
1352 MSR_IA32_VMX_VMCS_ENUM,
1353 MSR_IA32_VMX_PROCBASED_CTLS2,
1354 MSR_IA32_VMX_EPT_VPID_CAP,
1355 MSR_IA32_VMX_VMFUNC,
1359 MSR_IA32_ARCH_CAPABILITIES,
1360 MSR_IA32_PERF_CAPABILITIES,
1363 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1364 static unsigned int num_msr_based_features;
1366 static u64 kvm_get_arch_capabilities(void)
1370 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1371 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1374 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1375 * the nested hypervisor runs with NX huge pages. If it is not,
1376 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1377 * L1 guests, so it need not worry about its own (L2) guests.
1379 data |= ARCH_CAP_PSCHANGE_MC_NO;
1382 * If we're doing cache flushes (either "always" or "cond")
1383 * we will do one whenever the guest does a vmlaunch/vmresume.
1384 * If an outer hypervisor is doing the cache flush for us
1385 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1386 * capability to the guest too, and if EPT is disabled we're not
1387 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1388 * require a nested hypervisor to do a flush of its own.
1390 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1391 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1393 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1394 data |= ARCH_CAP_RDCL_NO;
1395 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1396 data |= ARCH_CAP_SSB_NO;
1397 if (!boot_cpu_has_bug(X86_BUG_MDS))
1398 data |= ARCH_CAP_MDS_NO;
1400 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1402 * If RTM=0 because the kernel has disabled TSX, the host might
1403 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1404 * and therefore knows that there cannot be TAA) but keep
1405 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1406 * and we want to allow migrating those guests to tsx=off hosts.
1408 data &= ~ARCH_CAP_TAA_NO;
1409 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1410 data |= ARCH_CAP_TAA_NO;
1413 * Nothing to do here; we emulate TSX_CTRL if present on the
1414 * host so the guest can choose between disabling TSX or
1415 * using VERW to clear CPU buffers.
1422 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1424 switch (msr->index) {
1425 case MSR_IA32_ARCH_CAPABILITIES:
1426 msr->data = kvm_get_arch_capabilities();
1428 case MSR_IA32_UCODE_REV:
1429 rdmsrl_safe(msr->index, &msr->data);
1432 return static_call(kvm_x86_get_msr_feature)(msr);
1437 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1439 struct kvm_msr_entry msr;
1443 r = kvm_get_msr_feature(&msr);
1445 if (r == KVM_MSR_RET_INVALID) {
1446 /* Unconditionally clear the output for simplicity */
1448 if (kvm_msr_ignored_check(vcpu, index, 0, false))
1460 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1462 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1465 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1468 if (efer & (EFER_LME | EFER_LMA) &&
1469 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1472 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1478 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1480 if (efer & efer_reserved_bits)
1483 return __kvm_valid_efer(vcpu, efer);
1485 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1487 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1489 u64 old_efer = vcpu->arch.efer;
1490 u64 efer = msr_info->data;
1493 if (efer & efer_reserved_bits)
1496 if (!msr_info->host_initiated) {
1497 if (!__kvm_valid_efer(vcpu, efer))
1500 if (is_paging(vcpu) &&
1501 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1506 efer |= vcpu->arch.efer & EFER_LMA;
1508 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1514 /* Update reserved bits */
1515 if ((efer ^ old_efer) & EFER_NX)
1516 kvm_mmu_reset_context(vcpu);
1521 void kvm_enable_efer_bits(u64 mask)
1523 efer_reserved_bits &= ~mask;
1525 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1527 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1529 struct kvm *kvm = vcpu->kvm;
1530 struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
1531 u32 count = kvm->arch.msr_filter.count;
1533 bool r = kvm->arch.msr_filter.default_allow;
1536 /* MSR filtering not set up or x2APIC enabled, allow everything */
1537 if (!count || (index >= 0x800 && index <= 0x8ff))
1540 /* Prevent collision with set_msr_filter */
1541 idx = srcu_read_lock(&kvm->srcu);
1543 for (i = 0; i < count; i++) {
1544 u32 start = ranges[i].base;
1545 u32 end = start + ranges[i].nmsrs;
1546 u32 flags = ranges[i].flags;
1547 unsigned long *bitmap = ranges[i].bitmap;
1549 if ((index >= start) && (index < end) && (flags & type)) {
1550 r = !!test_bit(index - start, bitmap);
1555 srcu_read_unlock(&kvm->srcu, idx);
1559 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1562 * Write @data into the MSR specified by @index. Select MSR specific fault
1563 * checks are bypassed if @host_initiated is %true.
1564 * Returns 0 on success, non-0 otherwise.
1565 * Assumes vcpu_load() was already called.
1567 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1568 bool host_initiated)
1570 struct msr_data msr;
1572 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1573 return KVM_MSR_RET_FILTERED;
1578 case MSR_KERNEL_GS_BASE:
1581 if (is_noncanonical_address(data, vcpu))
1584 case MSR_IA32_SYSENTER_EIP:
1585 case MSR_IA32_SYSENTER_ESP:
1587 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1588 * non-canonical address is written on Intel but not on
1589 * AMD (which ignores the top 32-bits, because it does
1590 * not implement 64-bit SYSENTER).
1592 * 64-bit code should hence be able to write a non-canonical
1593 * value on AMD. Making the address canonical ensures that
1594 * vmentry does not fail on Intel after writing a non-canonical
1595 * value, and that something deterministic happens if the guest
1596 * invokes 64-bit SYSENTER.
1598 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1603 msr.host_initiated = host_initiated;
1605 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1608 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1609 u32 index, u64 data, bool host_initiated)
1611 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1613 if (ret == KVM_MSR_RET_INVALID)
1614 if (kvm_msr_ignored_check(vcpu, index, data, true))
1621 * Read the MSR specified by @index into @data. Select MSR specific fault
1622 * checks are bypassed if @host_initiated is %true.
1623 * Returns 0 on success, non-0 otherwise.
1624 * Assumes vcpu_load() was already called.
1626 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1627 bool host_initiated)
1629 struct msr_data msr;
1632 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1633 return KVM_MSR_RET_FILTERED;
1636 msr.host_initiated = host_initiated;
1638 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1644 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1645 u32 index, u64 *data, bool host_initiated)
1647 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1649 if (ret == KVM_MSR_RET_INVALID) {
1650 /* Unconditionally clear *data for simplicity */
1652 if (kvm_msr_ignored_check(vcpu, index, 0, false))
1659 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1661 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1663 EXPORT_SYMBOL_GPL(kvm_get_msr);
1665 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1667 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1669 EXPORT_SYMBOL_GPL(kvm_set_msr);
1671 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1673 int err = vcpu->run->msr.error;
1675 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1676 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1679 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1682 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1684 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1687 static u64 kvm_msr_reason(int r)
1690 case KVM_MSR_RET_INVALID:
1691 return KVM_MSR_EXIT_REASON_UNKNOWN;
1692 case KVM_MSR_RET_FILTERED:
1693 return KVM_MSR_EXIT_REASON_FILTER;
1695 return KVM_MSR_EXIT_REASON_INVAL;
1699 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1700 u32 exit_reason, u64 data,
1701 int (*completion)(struct kvm_vcpu *vcpu),
1704 u64 msr_reason = kvm_msr_reason(r);
1706 /* Check if the user wanted to know about this MSR fault */
1707 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1710 vcpu->run->exit_reason = exit_reason;
1711 vcpu->run->msr.error = 0;
1712 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1713 vcpu->run->msr.reason = msr_reason;
1714 vcpu->run->msr.index = index;
1715 vcpu->run->msr.data = data;
1716 vcpu->arch.complete_userspace_io = completion;
1721 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1723 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1724 complete_emulated_rdmsr, r);
1727 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1729 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1730 complete_emulated_wrmsr, r);
1733 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1735 u32 ecx = kvm_rcx_read(vcpu);
1739 r = kvm_get_msr(vcpu, ecx, &data);
1741 /* MSR read failed? See if we should ask user space */
1742 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1743 /* Bounce to user space */
1748 trace_kvm_msr_read(ecx, data);
1750 kvm_rax_write(vcpu, data & -1u);
1751 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1753 trace_kvm_msr_read_ex(ecx);
1756 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1758 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1760 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1762 u32 ecx = kvm_rcx_read(vcpu);
1763 u64 data = kvm_read_edx_eax(vcpu);
1766 r = kvm_set_msr(vcpu, ecx, data);
1768 /* MSR write failed? See if we should ask user space */
1769 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1770 /* Bounce to user space */
1773 /* Signal all other negative errors to userspace */
1778 trace_kvm_msr_write(ecx, data);
1780 trace_kvm_msr_write_ex(ecx, data);
1782 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1784 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1786 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1788 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1789 xfer_to_guest_mode_work_pending();
1793 * The fast path for frequent and performance sensitive wrmsr emulation,
1794 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1795 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1796 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1797 * other cases which must be called after interrupts are enabled on the host.
1799 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1801 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1804 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1805 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1806 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1807 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1810 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1811 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1812 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1813 trace_kvm_apic_write(APIC_ICR, (u32)data);
1820 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1822 if (!kvm_can_use_hv_timer(vcpu))
1825 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1829 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1831 u32 msr = kvm_rcx_read(vcpu);
1833 fastpath_t ret = EXIT_FASTPATH_NONE;
1836 case APIC_BASE_MSR + (APIC_ICR >> 4):
1837 data = kvm_read_edx_eax(vcpu);
1838 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1839 kvm_skip_emulated_instruction(vcpu);
1840 ret = EXIT_FASTPATH_EXIT_HANDLED;
1843 case MSR_IA32_TSCDEADLINE:
1844 data = kvm_read_edx_eax(vcpu);
1845 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1846 kvm_skip_emulated_instruction(vcpu);
1847 ret = EXIT_FASTPATH_REENTER_GUEST;
1854 if (ret != EXIT_FASTPATH_NONE)
1855 trace_kvm_msr_write(msr, data);
1859 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1862 * Adapt set_msr() to msr_io()'s calling convention
1864 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1866 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1869 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1871 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1874 #ifdef CONFIG_X86_64
1875 struct pvclock_clock {
1885 struct pvclock_gtod_data {
1888 struct pvclock_clock clock; /* extract of a clocksource struct */
1889 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1895 static struct pvclock_gtod_data pvclock_gtod_data;
1897 static void update_pvclock_gtod(struct timekeeper *tk)
1899 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1901 write_seqcount_begin(&vdata->seq);
1903 /* copy pvclock gtod data */
1904 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1905 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1906 vdata->clock.mask = tk->tkr_mono.mask;
1907 vdata->clock.mult = tk->tkr_mono.mult;
1908 vdata->clock.shift = tk->tkr_mono.shift;
1909 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1910 vdata->clock.offset = tk->tkr_mono.base;
1912 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1913 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1914 vdata->raw_clock.mask = tk->tkr_raw.mask;
1915 vdata->raw_clock.mult = tk->tkr_raw.mult;
1916 vdata->raw_clock.shift = tk->tkr_raw.shift;
1917 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1918 vdata->raw_clock.offset = tk->tkr_raw.base;
1920 vdata->wall_time_sec = tk->xtime_sec;
1922 vdata->offs_boot = tk->offs_boot;
1924 write_seqcount_end(&vdata->seq);
1927 static s64 get_kvmclock_base_ns(void)
1929 /* Count up from boot time, but with the frequency of the raw clock. */
1930 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1933 static s64 get_kvmclock_base_ns(void)
1935 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1936 return ktime_get_boottime_ns();
1940 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
1944 struct pvclock_wall_clock wc;
1951 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1956 ++version; /* first time write, random junk */
1960 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1964 * The guest calculates current wall clock time by adding
1965 * system time (updated by kvm_guest_time_update below) to the
1966 * wall clock specified here. We do the reverse here.
1968 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1970 wc.nsec = do_div(wall_nsec, 1000000000);
1971 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1972 wc.version = version;
1974 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1977 wc_sec_hi = wall_nsec >> 32;
1978 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
1979 &wc_sec_hi, sizeof(wc_sec_hi));
1983 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1986 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
1987 bool old_msr, bool host_initiated)
1989 struct kvm_arch *ka = &vcpu->kvm->arch;
1991 if (vcpu->vcpu_id == 0 && !host_initiated) {
1992 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
1993 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1995 ka->boot_vcpu_runs_old_kvmclock = old_msr;
1998 vcpu->arch.time = system_time;
1999 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2001 /* we verify if the enable bit is set... */
2002 vcpu->arch.pv_time_enabled = false;
2003 if (!(system_time & 1))
2006 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2007 &vcpu->arch.pv_time, system_time & ~1ULL,
2008 sizeof(struct pvclock_vcpu_time_info)))
2009 vcpu->arch.pv_time_enabled = true;
2014 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2016 do_shl32_div32(dividend, divisor);
2020 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2021 s8 *pshift, u32 *pmultiplier)
2029 scaled64 = scaled_hz;
2030 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2035 tps32 = (uint32_t)tps64;
2036 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2037 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2045 *pmultiplier = div_frac(scaled64, tps32);
2048 #ifdef CONFIG_X86_64
2049 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2052 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2053 static unsigned long max_tsc_khz;
2055 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2057 u64 v = (u64)khz * (1000000 + ppm);
2062 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2066 /* Guest TSC same frequency as host TSC? */
2068 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2072 /* TSC scaling supported? */
2073 if (!kvm_has_tsc_control) {
2074 if (user_tsc_khz > tsc_khz) {
2075 vcpu->arch.tsc_catchup = 1;
2076 vcpu->arch.tsc_always_catchup = 1;
2079 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2084 /* TSC scaling required - calculate ratio */
2085 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2086 user_tsc_khz, tsc_khz);
2088 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2089 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2094 vcpu->arch.tsc_scaling_ratio = ratio;
2098 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2100 u32 thresh_lo, thresh_hi;
2101 int use_scaling = 0;
2103 /* tsc_khz can be zero if TSC calibration fails */
2104 if (user_tsc_khz == 0) {
2105 /* set tsc_scaling_ratio to a safe value */
2106 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2110 /* Compute a scale to convert nanoseconds in TSC cycles */
2111 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2112 &vcpu->arch.virtual_tsc_shift,
2113 &vcpu->arch.virtual_tsc_mult);
2114 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2117 * Compute the variation in TSC rate which is acceptable
2118 * within the range of tolerance and decide if the
2119 * rate being applied is within that bounds of the hardware
2120 * rate. If so, no scaling or compensation need be done.
2122 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2123 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2124 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2125 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2128 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2131 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2133 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2134 vcpu->arch.virtual_tsc_mult,
2135 vcpu->arch.virtual_tsc_shift);
2136 tsc += vcpu->arch.this_tsc_write;
2140 static inline int gtod_is_based_on_tsc(int mode)
2142 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2145 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2147 #ifdef CONFIG_X86_64
2149 struct kvm_arch *ka = &vcpu->kvm->arch;
2150 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2152 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2153 atomic_read(&vcpu->kvm->online_vcpus));
2156 * Once the masterclock is enabled, always perform request in
2157 * order to update it.
2159 * In order to enable masterclock, the host clocksource must be TSC
2160 * and the vcpus need to have matched TSCs. When that happens,
2161 * perform request to enable masterclock.
2163 if (ka->use_master_clock ||
2164 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2165 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2167 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2168 atomic_read(&vcpu->kvm->online_vcpus),
2169 ka->use_master_clock, gtod->clock.vclock_mode);
2174 * Multiply tsc by a fixed point number represented by ratio.
2176 * The most significant 64-N bits (mult) of ratio represent the
2177 * integral part of the fixed point number; the remaining N bits
2178 * (frac) represent the fractional part, ie. ratio represents a fixed
2179 * point number (mult + frac * 2^(-N)).
2181 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2183 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2185 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2188 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2191 u64 ratio = vcpu->arch.tsc_scaling_ratio;
2193 if (ratio != kvm_default_tsc_scaling_ratio)
2194 _tsc = __scale_tsc(ratio, tsc);
2198 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2200 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2204 tsc = kvm_scale_tsc(vcpu, rdtsc());
2206 return target_tsc - tsc;
2209 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2211 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2213 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2215 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2217 vcpu->arch.l1_tsc_offset = offset;
2218 vcpu->arch.tsc_offset = static_call(kvm_x86_write_l1_tsc_offset)(vcpu, offset);
2221 static inline bool kvm_check_tsc_unstable(void)
2223 #ifdef CONFIG_X86_64
2225 * TSC is marked unstable when we're running on Hyper-V,
2226 * 'TSC page' clocksource is good.
2228 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2231 return check_tsc_unstable();
2234 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2236 struct kvm *kvm = vcpu->kvm;
2237 u64 offset, ns, elapsed;
2238 unsigned long flags;
2240 bool already_matched;
2241 bool synchronizing = false;
2243 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2244 offset = kvm_compute_tsc_offset(vcpu, data);
2245 ns = get_kvmclock_base_ns();
2246 elapsed = ns - kvm->arch.last_tsc_nsec;
2248 if (vcpu->arch.virtual_tsc_khz) {
2251 * detection of vcpu initialization -- need to sync
2252 * with other vCPUs. This particularly helps to keep
2253 * kvm_clock stable after CPU hotplug
2255 synchronizing = true;
2257 u64 tsc_exp = kvm->arch.last_tsc_write +
2258 nsec_to_cycles(vcpu, elapsed);
2259 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2261 * Special case: TSC write with a small delta (1 second)
2262 * of virtual cycle time against real time is
2263 * interpreted as an attempt to synchronize the CPU.
2265 synchronizing = data < tsc_exp + tsc_hz &&
2266 data + tsc_hz > tsc_exp;
2271 * For a reliable TSC, we can match TSC offsets, and for an unstable
2272 * TSC, we add elapsed time in this computation. We could let the
2273 * compensation code attempt to catch up if we fall behind, but
2274 * it's better to try to match offsets from the beginning.
2276 if (synchronizing &&
2277 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2278 if (!kvm_check_tsc_unstable()) {
2279 offset = kvm->arch.cur_tsc_offset;
2281 u64 delta = nsec_to_cycles(vcpu, elapsed);
2283 offset = kvm_compute_tsc_offset(vcpu, data);
2286 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2289 * We split periods of matched TSC writes into generations.
2290 * For each generation, we track the original measured
2291 * nanosecond time, offset, and write, so if TSCs are in
2292 * sync, we can match exact offset, and if not, we can match
2293 * exact software computation in compute_guest_tsc()
2295 * These values are tracked in kvm->arch.cur_xxx variables.
2297 kvm->arch.cur_tsc_generation++;
2298 kvm->arch.cur_tsc_nsec = ns;
2299 kvm->arch.cur_tsc_write = data;
2300 kvm->arch.cur_tsc_offset = offset;
2305 * We also track th most recent recorded KHZ, write and time to
2306 * allow the matching interval to be extended at each write.
2308 kvm->arch.last_tsc_nsec = ns;
2309 kvm->arch.last_tsc_write = data;
2310 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2312 vcpu->arch.last_guest_tsc = data;
2314 /* Keep track of which generation this VCPU has synchronized to */
2315 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2316 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2317 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2319 kvm_vcpu_write_tsc_offset(vcpu, offset);
2320 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2322 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2324 kvm->arch.nr_vcpus_matched_tsc = 0;
2325 } else if (!already_matched) {
2326 kvm->arch.nr_vcpus_matched_tsc++;
2329 kvm_track_tsc_matching(vcpu);
2330 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2333 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2336 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2337 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2340 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2342 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2343 WARN_ON(adjustment < 0);
2344 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2345 adjust_tsc_offset_guest(vcpu, adjustment);
2348 #ifdef CONFIG_X86_64
2350 static u64 read_tsc(void)
2352 u64 ret = (u64)rdtsc_ordered();
2353 u64 last = pvclock_gtod_data.clock.cycle_last;
2355 if (likely(ret >= last))
2359 * GCC likes to generate cmov here, but this branch is extremely
2360 * predictable (it's just a function of time and the likely is
2361 * very likely) and there's a data dependence, so force GCC
2362 * to generate a branch instead. I don't barrier() because
2363 * we don't actually need a barrier, and if this function
2364 * ever gets inlined it will generate worse code.
2370 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2376 switch (clock->vclock_mode) {
2377 case VDSO_CLOCKMODE_HVCLOCK:
2378 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2380 if (tsc_pg_val != U64_MAX) {
2381 /* TSC page valid */
2382 *mode = VDSO_CLOCKMODE_HVCLOCK;
2383 v = (tsc_pg_val - clock->cycle_last) &
2386 /* TSC page invalid */
2387 *mode = VDSO_CLOCKMODE_NONE;
2390 case VDSO_CLOCKMODE_TSC:
2391 *mode = VDSO_CLOCKMODE_TSC;
2392 *tsc_timestamp = read_tsc();
2393 v = (*tsc_timestamp - clock->cycle_last) &
2397 *mode = VDSO_CLOCKMODE_NONE;
2400 if (*mode == VDSO_CLOCKMODE_NONE)
2401 *tsc_timestamp = v = 0;
2403 return v * clock->mult;
2406 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2408 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2414 seq = read_seqcount_begin(>od->seq);
2415 ns = gtod->raw_clock.base_cycles;
2416 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2417 ns >>= gtod->raw_clock.shift;
2418 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2419 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2425 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2427 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2433 seq = read_seqcount_begin(>od->seq);
2434 ts->tv_sec = gtod->wall_time_sec;
2435 ns = gtod->clock.base_cycles;
2436 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2437 ns >>= gtod->clock.shift;
2438 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2440 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2446 /* returns true if host is using TSC based clocksource */
2447 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2449 /* checked again under seqlock below */
2450 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2453 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2457 /* returns true if host is using TSC based clocksource */
2458 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2461 /* checked again under seqlock below */
2462 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2465 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2471 * Assuming a stable TSC across physical CPUS, and a stable TSC
2472 * across virtual CPUs, the following condition is possible.
2473 * Each numbered line represents an event visible to both
2474 * CPUs at the next numbered event.
2476 * "timespecX" represents host monotonic time. "tscX" represents
2479 * VCPU0 on CPU0 | VCPU1 on CPU1
2481 * 1. read timespec0,tsc0
2482 * 2. | timespec1 = timespec0 + N
2484 * 3. transition to guest | transition to guest
2485 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2486 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2487 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2489 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2492 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2494 * - 0 < N - M => M < N
2496 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2497 * always the case (the difference between two distinct xtime instances
2498 * might be smaller then the difference between corresponding TSC reads,
2499 * when updating guest vcpus pvclock areas).
2501 * To avoid that problem, do not allow visibility of distinct
2502 * system_timestamp/tsc_timestamp values simultaneously: use a master
2503 * copy of host monotonic time values. Update that master copy
2506 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2510 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2512 #ifdef CONFIG_X86_64
2513 struct kvm_arch *ka = &kvm->arch;
2515 bool host_tsc_clocksource, vcpus_matched;
2517 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2518 atomic_read(&kvm->online_vcpus));
2521 * If the host uses TSC clock, then passthrough TSC as stable
2524 host_tsc_clocksource = kvm_get_time_and_clockread(
2525 &ka->master_kernel_ns,
2526 &ka->master_cycle_now);
2528 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2529 && !ka->backwards_tsc_observed
2530 && !ka->boot_vcpu_runs_old_kvmclock;
2532 if (ka->use_master_clock)
2533 atomic_set(&kvm_guest_has_master_clock, 1);
2535 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2536 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2541 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2543 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2546 static void kvm_gen_update_masterclock(struct kvm *kvm)
2548 #ifdef CONFIG_X86_64
2550 struct kvm_vcpu *vcpu;
2551 struct kvm_arch *ka = &kvm->arch;
2553 spin_lock(&ka->pvclock_gtod_sync_lock);
2554 kvm_make_mclock_inprogress_request(kvm);
2555 /* no guest entries from this point */
2556 pvclock_update_vm_gtod_copy(kvm);
2558 kvm_for_each_vcpu(i, vcpu, kvm)
2559 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2561 /* guest entries allowed */
2562 kvm_for_each_vcpu(i, vcpu, kvm)
2563 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2565 spin_unlock(&ka->pvclock_gtod_sync_lock);
2569 u64 get_kvmclock_ns(struct kvm *kvm)
2571 struct kvm_arch *ka = &kvm->arch;
2572 struct pvclock_vcpu_time_info hv_clock;
2575 spin_lock(&ka->pvclock_gtod_sync_lock);
2576 if (!ka->use_master_clock) {
2577 spin_unlock(&ka->pvclock_gtod_sync_lock);
2578 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2581 hv_clock.tsc_timestamp = ka->master_cycle_now;
2582 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2583 spin_unlock(&ka->pvclock_gtod_sync_lock);
2585 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2588 if (__this_cpu_read(cpu_tsc_khz)) {
2589 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2590 &hv_clock.tsc_shift,
2591 &hv_clock.tsc_to_system_mul);
2592 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2594 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2601 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2602 struct gfn_to_hva_cache *cache,
2603 unsigned int offset)
2605 struct kvm_vcpu_arch *vcpu = &v->arch;
2606 struct pvclock_vcpu_time_info guest_hv_clock;
2608 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2609 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2612 /* This VCPU is paused, but it's legal for a guest to read another
2613 * VCPU's kvmclock, so we really have to follow the specification where
2614 * it says that version is odd if data is being modified, and even after
2617 * Version field updates must be kept separate. This is because
2618 * kvm_write_guest_cached might use a "rep movs" instruction, and
2619 * writes within a string instruction are weakly ordered. So there
2620 * are three writes overall.
2622 * As a small optimization, only write the version field in the first
2623 * and third write. The vcpu->pv_time cache is still valid, because the
2624 * version field is the first in the struct.
2626 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2628 if (guest_hv_clock.version & 1)
2629 ++guest_hv_clock.version; /* first time write, random junk */
2631 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2632 kvm_write_guest_offset_cached(v->kvm, cache,
2633 &vcpu->hv_clock, offset,
2634 sizeof(vcpu->hv_clock.version));
2638 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2639 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2641 if (vcpu->pvclock_set_guest_stopped_request) {
2642 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2643 vcpu->pvclock_set_guest_stopped_request = false;
2646 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2648 kvm_write_guest_offset_cached(v->kvm, cache,
2649 &vcpu->hv_clock, offset,
2650 sizeof(vcpu->hv_clock));
2654 vcpu->hv_clock.version++;
2655 kvm_write_guest_offset_cached(v->kvm, cache,
2656 &vcpu->hv_clock, offset,
2657 sizeof(vcpu->hv_clock.version));
2660 static int kvm_guest_time_update(struct kvm_vcpu *v)
2662 unsigned long flags, tgt_tsc_khz;
2663 struct kvm_vcpu_arch *vcpu = &v->arch;
2664 struct kvm_arch *ka = &v->kvm->arch;
2666 u64 tsc_timestamp, host_tsc;
2668 bool use_master_clock;
2674 * If the host uses TSC clock, then passthrough TSC as stable
2677 spin_lock(&ka->pvclock_gtod_sync_lock);
2678 use_master_clock = ka->use_master_clock;
2679 if (use_master_clock) {
2680 host_tsc = ka->master_cycle_now;
2681 kernel_ns = ka->master_kernel_ns;
2683 spin_unlock(&ka->pvclock_gtod_sync_lock);
2685 /* Keep irq disabled to prevent changes to the clock */
2686 local_irq_save(flags);
2687 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2688 if (unlikely(tgt_tsc_khz == 0)) {
2689 local_irq_restore(flags);
2690 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2693 if (!use_master_clock) {
2695 kernel_ns = get_kvmclock_base_ns();
2698 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2701 * We may have to catch up the TSC to match elapsed wall clock
2702 * time for two reasons, even if kvmclock is used.
2703 * 1) CPU could have been running below the maximum TSC rate
2704 * 2) Broken TSC compensation resets the base at each VCPU
2705 * entry to avoid unknown leaps of TSC even when running
2706 * again on the same CPU. This may cause apparent elapsed
2707 * time to disappear, and the guest to stand still or run
2710 if (vcpu->tsc_catchup) {
2711 u64 tsc = compute_guest_tsc(v, kernel_ns);
2712 if (tsc > tsc_timestamp) {
2713 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2714 tsc_timestamp = tsc;
2718 local_irq_restore(flags);
2720 /* With all the info we got, fill in the values */
2722 if (kvm_has_tsc_control)
2723 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2725 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2726 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2727 &vcpu->hv_clock.tsc_shift,
2728 &vcpu->hv_clock.tsc_to_system_mul);
2729 vcpu->hw_tsc_khz = tgt_tsc_khz;
2732 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2733 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2734 vcpu->last_guest_tsc = tsc_timestamp;
2736 /* If the host uses TSC clocksource, then it is stable */
2738 if (use_master_clock)
2739 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2741 vcpu->hv_clock.flags = pvclock_flags;
2743 if (vcpu->pv_time_enabled)
2744 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2745 if (vcpu->xen.vcpu_info_set)
2746 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2747 offsetof(struct compat_vcpu_info, time));
2748 if (vcpu->xen.vcpu_time_info_set)
2749 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2750 if (v == kvm_get_vcpu(v->kvm, 0))
2751 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2756 * kvmclock updates which are isolated to a given vcpu, such as
2757 * vcpu->cpu migration, should not allow system_timestamp from
2758 * the rest of the vcpus to remain static. Otherwise ntp frequency
2759 * correction applies to one vcpu's system_timestamp but not
2762 * So in those cases, request a kvmclock update for all vcpus.
2763 * We need to rate-limit these requests though, as they can
2764 * considerably slow guests that have a large number of vcpus.
2765 * The time for a remote vcpu to update its kvmclock is bound
2766 * by the delay we use to rate-limit the updates.
2769 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2771 static void kvmclock_update_fn(struct work_struct *work)
2774 struct delayed_work *dwork = to_delayed_work(work);
2775 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2776 kvmclock_update_work);
2777 struct kvm *kvm = container_of(ka, struct kvm, arch);
2778 struct kvm_vcpu *vcpu;
2780 kvm_for_each_vcpu(i, vcpu, kvm) {
2781 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2782 kvm_vcpu_kick(vcpu);
2786 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2788 struct kvm *kvm = v->kvm;
2790 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2791 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2792 KVMCLOCK_UPDATE_DELAY);
2795 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2797 static void kvmclock_sync_fn(struct work_struct *work)
2799 struct delayed_work *dwork = to_delayed_work(work);
2800 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2801 kvmclock_sync_work);
2802 struct kvm *kvm = container_of(ka, struct kvm, arch);
2804 if (!kvmclock_periodic_sync)
2807 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2808 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2809 KVMCLOCK_SYNC_PERIOD);
2813 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2815 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2817 /* McStatusWrEn enabled? */
2818 if (guest_cpuid_is_amd_or_hygon(vcpu))
2819 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2824 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2826 u64 mcg_cap = vcpu->arch.mcg_cap;
2827 unsigned bank_num = mcg_cap & 0xff;
2828 u32 msr = msr_info->index;
2829 u64 data = msr_info->data;
2832 case MSR_IA32_MCG_STATUS:
2833 vcpu->arch.mcg_status = data;
2835 case MSR_IA32_MCG_CTL:
2836 if (!(mcg_cap & MCG_CTL_P) &&
2837 (data || !msr_info->host_initiated))
2839 if (data != 0 && data != ~(u64)0)
2841 vcpu->arch.mcg_ctl = data;
2844 if (msr >= MSR_IA32_MC0_CTL &&
2845 msr < MSR_IA32_MCx_CTL(bank_num)) {
2846 u32 offset = array_index_nospec(
2847 msr - MSR_IA32_MC0_CTL,
2848 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2850 /* only 0 or all 1s can be written to IA32_MCi_CTL
2851 * some Linux kernels though clear bit 10 in bank 4 to
2852 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2853 * this to avoid an uncatched #GP in the guest
2855 if ((offset & 0x3) == 0 &&
2856 data != 0 && (data | (1 << 10)) != ~(u64)0)
2860 if (!msr_info->host_initiated &&
2861 (offset & 0x3) == 1 && data != 0) {
2862 if (!can_set_mci_status(vcpu))
2866 vcpu->arch.mce_banks[offset] = data;
2874 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2876 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2878 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2881 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2883 gpa_t gpa = data & ~0x3f;
2885 /* Bits 4:5 are reserved, Should be zero */
2889 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2890 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2893 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2894 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2897 if (!lapic_in_kernel(vcpu))
2898 return data ? 1 : 0;
2900 vcpu->arch.apf.msr_en_val = data;
2902 if (!kvm_pv_async_pf_enabled(vcpu)) {
2903 kvm_clear_async_pf_completion_queue(vcpu);
2904 kvm_async_pf_hash_reset(vcpu);
2908 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2912 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2913 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2915 kvm_async_pf_wakeup_all(vcpu);
2920 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2922 /* Bits 8-63 are reserved */
2926 if (!lapic_in_kernel(vcpu))
2929 vcpu->arch.apf.msr_int_val = data;
2931 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2936 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2938 vcpu->arch.pv_time_enabled = false;
2939 vcpu->arch.time = 0;
2942 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2944 ++vcpu->stat.tlb_flush;
2945 static_call(kvm_x86_tlb_flush_all)(vcpu);
2948 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2950 ++vcpu->stat.tlb_flush;
2951 static_call(kvm_x86_tlb_flush_guest)(vcpu);
2954 static void record_steal_time(struct kvm_vcpu *vcpu)
2956 struct kvm_host_map map;
2957 struct kvm_steal_time *st;
2959 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2962 /* -EAGAIN is returned in atomic context so we can just return. */
2963 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2964 &map, &vcpu->arch.st.cache, false))
2968 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2971 * Doing a TLB flush here, on the guest's behalf, can avoid
2974 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
2975 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2976 st->preempted & KVM_VCPU_FLUSH_TLB);
2977 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2978 kvm_vcpu_flush_tlb_guest(vcpu);
2981 vcpu->arch.st.preempted = 0;
2983 if (st->version & 1)
2984 st->version += 1; /* first time write, random junk */
2990 st->steal += current->sched_info.run_delay -
2991 vcpu->arch.st.last_steal;
2992 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2998 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3001 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3004 u32 msr = msr_info->index;
3005 u64 data = msr_info->data;
3007 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3008 return kvm_xen_write_hypercall_page(vcpu, data);
3011 case MSR_AMD64_NB_CFG:
3012 case MSR_IA32_UCODE_WRITE:
3013 case MSR_VM_HSAVE_PA:
3014 case MSR_AMD64_PATCH_LOADER:
3015 case MSR_AMD64_BU_CFG2:
3016 case MSR_AMD64_DC_CFG:
3017 case MSR_F15H_EX_CFG:
3020 case MSR_IA32_UCODE_REV:
3021 if (msr_info->host_initiated)
3022 vcpu->arch.microcode_version = data;
3024 case MSR_IA32_ARCH_CAPABILITIES:
3025 if (!msr_info->host_initiated)
3027 vcpu->arch.arch_capabilities = data;
3029 case MSR_IA32_PERF_CAPABILITIES: {
3030 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3032 if (!msr_info->host_initiated)
3034 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3036 if (data & ~msr_ent.data)
3039 vcpu->arch.perf_capabilities = data;
3044 return set_efer(vcpu, msr_info);
3046 data &= ~(u64)0x40; /* ignore flush filter disable */
3047 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3048 data &= ~(u64)0x8; /* ignore TLB cache disable */
3050 /* Handle McStatusWrEn */
3051 if (data == BIT_ULL(18)) {
3052 vcpu->arch.msr_hwcr = data;
3053 } else if (data != 0) {
3054 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3059 case MSR_FAM10H_MMIO_CONF_BASE:
3061 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3066 case 0x200 ... 0x2ff:
3067 return kvm_mtrr_set_msr(vcpu, msr, data);
3068 case MSR_IA32_APICBASE:
3069 return kvm_set_apic_base(vcpu, msr_info);
3070 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3071 return kvm_x2apic_msr_write(vcpu, msr, data);
3072 case MSR_IA32_TSCDEADLINE:
3073 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3075 case MSR_IA32_TSC_ADJUST:
3076 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3077 if (!msr_info->host_initiated) {
3078 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3079 adjust_tsc_offset_guest(vcpu, adj);
3081 vcpu->arch.ia32_tsc_adjust_msr = data;
3084 case MSR_IA32_MISC_ENABLE:
3085 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3086 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3087 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3089 vcpu->arch.ia32_misc_enable_msr = data;
3090 kvm_update_cpuid_runtime(vcpu);
3092 vcpu->arch.ia32_misc_enable_msr = data;
3095 case MSR_IA32_SMBASE:
3096 if (!msr_info->host_initiated)
3098 vcpu->arch.smbase = data;
3100 case MSR_IA32_POWER_CTL:
3101 vcpu->arch.msr_ia32_power_ctl = data;
3104 if (msr_info->host_initiated) {
3105 kvm_synchronize_tsc(vcpu, data);
3107 u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3108 adjust_tsc_offset_guest(vcpu, adj);
3109 vcpu->arch.ia32_tsc_adjust_msr += adj;
3113 if (!msr_info->host_initiated &&
3114 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3117 * KVM supports exposing PT to the guest, but does not support
3118 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3119 * XSAVES/XRSTORS to save/restore PT MSRs.
3121 if (data & ~supported_xss)
3123 vcpu->arch.ia32_xss = data;
3126 if (!msr_info->host_initiated)
3128 vcpu->arch.smi_count = data;
3130 case MSR_KVM_WALL_CLOCK_NEW:
3131 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3134 vcpu->kvm->arch.wall_clock = data;
3135 kvm_write_wall_clock(vcpu->kvm, data, 0);
3137 case MSR_KVM_WALL_CLOCK:
3138 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3141 vcpu->kvm->arch.wall_clock = data;
3142 kvm_write_wall_clock(vcpu->kvm, data, 0);
3144 case MSR_KVM_SYSTEM_TIME_NEW:
3145 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3148 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3150 case MSR_KVM_SYSTEM_TIME:
3151 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3154 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3156 case MSR_KVM_ASYNC_PF_EN:
3157 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3160 if (kvm_pv_enable_async_pf(vcpu, data))
3163 case MSR_KVM_ASYNC_PF_INT:
3164 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3167 if (kvm_pv_enable_async_pf_int(vcpu, data))
3170 case MSR_KVM_ASYNC_PF_ACK:
3171 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3174 vcpu->arch.apf.pageready_pending = false;
3175 kvm_check_async_pf_completion(vcpu);
3178 case MSR_KVM_STEAL_TIME:
3179 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3182 if (unlikely(!sched_info_on()))
3185 if (data & KVM_STEAL_RESERVED_MASK)
3188 vcpu->arch.st.msr_val = data;
3190 if (!(data & KVM_MSR_ENABLED))
3193 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3196 case MSR_KVM_PV_EOI_EN:
3197 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3200 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3204 case MSR_KVM_POLL_CONTROL:
3205 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3208 /* only enable bit supported */
3209 if (data & (-1ULL << 1))
3212 vcpu->arch.msr_kvm_poll_control = data;
3215 case MSR_IA32_MCG_CTL:
3216 case MSR_IA32_MCG_STATUS:
3217 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3218 return set_msr_mce(vcpu, msr_info);
3220 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3221 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3224 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3225 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3226 if (kvm_pmu_is_valid_msr(vcpu, msr))
3227 return kvm_pmu_set_msr(vcpu, msr_info);
3229 if (pr || data != 0)
3230 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3231 "0x%x data 0x%llx\n", msr, data);
3233 case MSR_K7_CLK_CTL:
3235 * Ignore all writes to this no longer documented MSR.
3236 * Writes are only relevant for old K7 processors,
3237 * all pre-dating SVM, but a recommended workaround from
3238 * AMD for these chips. It is possible to specify the
3239 * affected processor models on the command line, hence
3240 * the need to ignore the workaround.
3243 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3244 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3245 case HV_X64_MSR_SYNDBG_OPTIONS:
3246 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3247 case HV_X64_MSR_CRASH_CTL:
3248 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3249 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3250 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3251 case HV_X64_MSR_TSC_EMULATION_STATUS:
3252 return kvm_hv_set_msr_common(vcpu, msr, data,
3253 msr_info->host_initiated);
3254 case MSR_IA32_BBL_CR_CTL3:
3255 /* Drop writes to this legacy MSR -- see rdmsr
3256 * counterpart for further detail.
3258 if (report_ignored_msrs)
3259 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3262 case MSR_AMD64_OSVW_ID_LENGTH:
3263 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3265 vcpu->arch.osvw.length = data;
3267 case MSR_AMD64_OSVW_STATUS:
3268 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3270 vcpu->arch.osvw.status = data;
3272 case MSR_PLATFORM_INFO:
3273 if (!msr_info->host_initiated ||
3274 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3275 cpuid_fault_enabled(vcpu)))
3277 vcpu->arch.msr_platform_info = data;
3279 case MSR_MISC_FEATURES_ENABLES:
3280 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3281 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3282 !supports_cpuid_fault(vcpu)))
3284 vcpu->arch.msr_misc_features_enables = data;
3287 if (kvm_pmu_is_valid_msr(vcpu, msr))
3288 return kvm_pmu_set_msr(vcpu, msr_info);
3289 return KVM_MSR_RET_INVALID;
3293 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3295 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3298 u64 mcg_cap = vcpu->arch.mcg_cap;
3299 unsigned bank_num = mcg_cap & 0xff;
3302 case MSR_IA32_P5_MC_ADDR:
3303 case MSR_IA32_P5_MC_TYPE:
3306 case MSR_IA32_MCG_CAP:
3307 data = vcpu->arch.mcg_cap;
3309 case MSR_IA32_MCG_CTL:
3310 if (!(mcg_cap & MCG_CTL_P) && !host)
3312 data = vcpu->arch.mcg_ctl;
3314 case MSR_IA32_MCG_STATUS:
3315 data = vcpu->arch.mcg_status;
3318 if (msr >= MSR_IA32_MC0_CTL &&
3319 msr < MSR_IA32_MCx_CTL(bank_num)) {
3320 u32 offset = array_index_nospec(
3321 msr - MSR_IA32_MC0_CTL,
3322 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3324 data = vcpu->arch.mce_banks[offset];
3333 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3335 switch (msr_info->index) {
3336 case MSR_IA32_PLATFORM_ID:
3337 case MSR_IA32_EBL_CR_POWERON:
3338 case MSR_IA32_LASTBRANCHFROMIP:
3339 case MSR_IA32_LASTBRANCHTOIP:
3340 case MSR_IA32_LASTINTFROMIP:
3341 case MSR_IA32_LASTINTTOIP:
3343 case MSR_K8_TSEG_ADDR:
3344 case MSR_K8_TSEG_MASK:
3345 case MSR_VM_HSAVE_PA:
3346 case MSR_K8_INT_PENDING_MSG:
3347 case MSR_AMD64_NB_CFG:
3348 case MSR_FAM10H_MMIO_CONF_BASE:
3349 case MSR_AMD64_BU_CFG2:
3350 case MSR_IA32_PERF_CTL:
3351 case MSR_AMD64_DC_CFG:
3352 case MSR_F15H_EX_CFG:
3354 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3355 * limit) MSRs. Just return 0, as we do not want to expose the host
3356 * data here. Do not conditionalize this on CPUID, as KVM does not do
3357 * so for existing CPU-specific MSRs.
3359 case MSR_RAPL_POWER_UNIT:
3360 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3361 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3362 case MSR_PKG_ENERGY_STATUS: /* Total package */
3363 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3366 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3367 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3368 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3369 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3370 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3371 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3372 return kvm_pmu_get_msr(vcpu, msr_info);
3375 case MSR_IA32_UCODE_REV:
3376 msr_info->data = vcpu->arch.microcode_version;
3378 case MSR_IA32_ARCH_CAPABILITIES:
3379 if (!msr_info->host_initiated &&
3380 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3382 msr_info->data = vcpu->arch.arch_capabilities;
3384 case MSR_IA32_PERF_CAPABILITIES:
3385 if (!msr_info->host_initiated &&
3386 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3388 msr_info->data = vcpu->arch.perf_capabilities;
3390 case MSR_IA32_POWER_CTL:
3391 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3393 case MSR_IA32_TSC: {
3395 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3396 * even when not intercepted. AMD manual doesn't explicitly
3397 * state this but appears to behave the same.
3399 * On userspace reads and writes, however, we unconditionally
3400 * return L1's TSC value to ensure backwards-compatible
3401 * behavior for migration.
3403 u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3404 vcpu->arch.tsc_offset;
3406 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3410 case 0x200 ... 0x2ff:
3411 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3412 case 0xcd: /* fsb frequency */
3416 * MSR_EBC_FREQUENCY_ID
3417 * Conservative value valid for even the basic CPU models.
3418 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3419 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3420 * and 266MHz for model 3, or 4. Set Core Clock
3421 * Frequency to System Bus Frequency Ratio to 1 (bits
3422 * 31:24) even though these are only valid for CPU
3423 * models > 2, however guests may end up dividing or
3424 * multiplying by zero otherwise.
3426 case MSR_EBC_FREQUENCY_ID:
3427 msr_info->data = 1 << 24;
3429 case MSR_IA32_APICBASE:
3430 msr_info->data = kvm_get_apic_base(vcpu);
3432 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3433 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3434 case MSR_IA32_TSCDEADLINE:
3435 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3437 case MSR_IA32_TSC_ADJUST:
3438 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3440 case MSR_IA32_MISC_ENABLE:
3441 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3443 case MSR_IA32_SMBASE:
3444 if (!msr_info->host_initiated)
3446 msr_info->data = vcpu->arch.smbase;
3449 msr_info->data = vcpu->arch.smi_count;
3451 case MSR_IA32_PERF_STATUS:
3452 /* TSC increment by tick */
3453 msr_info->data = 1000ULL;
3454 /* CPU multiplier */
3455 msr_info->data |= (((uint64_t)4ULL) << 40);
3458 msr_info->data = vcpu->arch.efer;
3460 case MSR_KVM_WALL_CLOCK:
3461 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3464 msr_info->data = vcpu->kvm->arch.wall_clock;
3466 case MSR_KVM_WALL_CLOCK_NEW:
3467 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3470 msr_info->data = vcpu->kvm->arch.wall_clock;
3472 case MSR_KVM_SYSTEM_TIME:
3473 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3476 msr_info->data = vcpu->arch.time;
3478 case MSR_KVM_SYSTEM_TIME_NEW:
3479 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3482 msr_info->data = vcpu->arch.time;
3484 case MSR_KVM_ASYNC_PF_EN:
3485 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3488 msr_info->data = vcpu->arch.apf.msr_en_val;
3490 case MSR_KVM_ASYNC_PF_INT:
3491 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3494 msr_info->data = vcpu->arch.apf.msr_int_val;
3496 case MSR_KVM_ASYNC_PF_ACK:
3497 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3502 case MSR_KVM_STEAL_TIME:
3503 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3506 msr_info->data = vcpu->arch.st.msr_val;
3508 case MSR_KVM_PV_EOI_EN:
3509 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3512 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3514 case MSR_KVM_POLL_CONTROL:
3515 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3518 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3520 case MSR_IA32_P5_MC_ADDR:
3521 case MSR_IA32_P5_MC_TYPE:
3522 case MSR_IA32_MCG_CAP:
3523 case MSR_IA32_MCG_CTL:
3524 case MSR_IA32_MCG_STATUS:
3525 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3526 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3527 msr_info->host_initiated);
3529 if (!msr_info->host_initiated &&
3530 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3532 msr_info->data = vcpu->arch.ia32_xss;
3534 case MSR_K7_CLK_CTL:
3536 * Provide expected ramp-up count for K7. All other
3537 * are set to zero, indicating minimum divisors for
3540 * This prevents guest kernels on AMD host with CPU
3541 * type 6, model 8 and higher from exploding due to
3542 * the rdmsr failing.
3544 msr_info->data = 0x20000000;
3546 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3547 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3548 case HV_X64_MSR_SYNDBG_OPTIONS:
3549 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3550 case HV_X64_MSR_CRASH_CTL:
3551 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3552 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3553 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3554 case HV_X64_MSR_TSC_EMULATION_STATUS:
3555 return kvm_hv_get_msr_common(vcpu,
3556 msr_info->index, &msr_info->data,
3557 msr_info->host_initiated);
3558 case MSR_IA32_BBL_CR_CTL3:
3559 /* This legacy MSR exists but isn't fully documented in current
3560 * silicon. It is however accessed by winxp in very narrow
3561 * scenarios where it sets bit #19, itself documented as
3562 * a "reserved" bit. Best effort attempt to source coherent
3563 * read data here should the balance of the register be
3564 * interpreted by the guest:
3566 * L2 cache control register 3: 64GB range, 256KB size,
3567 * enabled, latency 0x1, configured
3569 msr_info->data = 0xbe702111;
3571 case MSR_AMD64_OSVW_ID_LENGTH:
3572 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3574 msr_info->data = vcpu->arch.osvw.length;
3576 case MSR_AMD64_OSVW_STATUS:
3577 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3579 msr_info->data = vcpu->arch.osvw.status;
3581 case MSR_PLATFORM_INFO:
3582 if (!msr_info->host_initiated &&
3583 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3585 msr_info->data = vcpu->arch.msr_platform_info;
3587 case MSR_MISC_FEATURES_ENABLES:
3588 msr_info->data = vcpu->arch.msr_misc_features_enables;
3591 msr_info->data = vcpu->arch.msr_hwcr;
3594 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3595 return kvm_pmu_get_msr(vcpu, msr_info);
3596 return KVM_MSR_RET_INVALID;
3600 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3603 * Read or write a bunch of msrs. All parameters are kernel addresses.
3605 * @return number of msrs set successfully.
3607 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3608 struct kvm_msr_entry *entries,
3609 int (*do_msr)(struct kvm_vcpu *vcpu,
3610 unsigned index, u64 *data))
3614 for (i = 0; i < msrs->nmsrs; ++i)
3615 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3622 * Read or write a bunch of msrs. Parameters are user addresses.
3624 * @return number of msrs set successfully.
3626 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3627 int (*do_msr)(struct kvm_vcpu *vcpu,
3628 unsigned index, u64 *data),
3631 struct kvm_msrs msrs;
3632 struct kvm_msr_entry *entries;
3637 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3641 if (msrs.nmsrs >= MAX_IO_MSRS)
3644 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3645 entries = memdup_user(user_msrs->entries, size);
3646 if (IS_ERR(entries)) {
3647 r = PTR_ERR(entries);
3651 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3656 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3667 static inline bool kvm_can_mwait_in_guest(void)
3669 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3670 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3671 boot_cpu_has(X86_FEATURE_ARAT);
3674 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3675 struct kvm_cpuid2 __user *cpuid_arg)
3677 struct kvm_cpuid2 cpuid;
3681 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3684 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3689 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3695 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3700 case KVM_CAP_IRQCHIP:
3702 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3703 case KVM_CAP_SET_TSS_ADDR:
3704 case KVM_CAP_EXT_CPUID:
3705 case KVM_CAP_EXT_EMUL_CPUID:
3706 case KVM_CAP_CLOCKSOURCE:
3708 case KVM_CAP_NOP_IO_DELAY:
3709 case KVM_CAP_MP_STATE:
3710 case KVM_CAP_SYNC_MMU:
3711 case KVM_CAP_USER_NMI:
3712 case KVM_CAP_REINJECT_CONTROL:
3713 case KVM_CAP_IRQ_INJECT_STATUS:
3714 case KVM_CAP_IOEVENTFD:
3715 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3717 case KVM_CAP_PIT_STATE2:
3718 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3719 case KVM_CAP_VCPU_EVENTS:
3720 case KVM_CAP_HYPERV:
3721 case KVM_CAP_HYPERV_VAPIC:
3722 case KVM_CAP_HYPERV_SPIN:
3723 case KVM_CAP_HYPERV_SYNIC:
3724 case KVM_CAP_HYPERV_SYNIC2:
3725 case KVM_CAP_HYPERV_VP_INDEX:
3726 case KVM_CAP_HYPERV_EVENTFD:
3727 case KVM_CAP_HYPERV_TLBFLUSH:
3728 case KVM_CAP_HYPERV_SEND_IPI:
3729 case KVM_CAP_HYPERV_CPUID:
3730 case KVM_CAP_SYS_HYPERV_CPUID:
3731 case KVM_CAP_PCI_SEGMENT:
3732 case KVM_CAP_DEBUGREGS:
3733 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3735 case KVM_CAP_ASYNC_PF:
3736 case KVM_CAP_ASYNC_PF_INT:
3737 case KVM_CAP_GET_TSC_KHZ:
3738 case KVM_CAP_KVMCLOCK_CTRL:
3739 case KVM_CAP_READONLY_MEM:
3740 case KVM_CAP_HYPERV_TIME:
3741 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3742 case KVM_CAP_TSC_DEADLINE_TIMER:
3743 case KVM_CAP_DISABLE_QUIRKS:
3744 case KVM_CAP_SET_BOOT_CPU_ID:
3745 case KVM_CAP_SPLIT_IRQCHIP:
3746 case KVM_CAP_IMMEDIATE_EXIT:
3747 case KVM_CAP_PMU_EVENT_FILTER:
3748 case KVM_CAP_GET_MSR_FEATURES:
3749 case KVM_CAP_MSR_PLATFORM_INFO:
3750 case KVM_CAP_EXCEPTION_PAYLOAD:
3751 case KVM_CAP_SET_GUEST_DEBUG:
3752 case KVM_CAP_LAST_CPU:
3753 case KVM_CAP_X86_USER_SPACE_MSR:
3754 case KVM_CAP_X86_MSR_FILTER:
3755 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3758 case KVM_CAP_XEN_HVM:
3759 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
3760 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
3761 KVM_XEN_HVM_CONFIG_SHARED_INFO;
3763 case KVM_CAP_SYNC_REGS:
3764 r = KVM_SYNC_X86_VALID_FIELDS;
3766 case KVM_CAP_ADJUST_CLOCK:
3767 r = KVM_CLOCK_TSC_STABLE;
3769 case KVM_CAP_X86_DISABLE_EXITS:
3770 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3771 KVM_X86_DISABLE_EXITS_CSTATE;
3772 if(kvm_can_mwait_in_guest())
3773 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3775 case KVM_CAP_X86_SMM:
3776 /* SMBASE is usually relocated above 1M on modern chipsets,
3777 * and SMM handlers might indeed rely on 4G segment limits,
3778 * so do not report SMM to be available if real mode is
3779 * emulated via vm86 mode. Still, do not go to great lengths
3780 * to avoid userspace's usage of the feature, because it is a
3781 * fringe case that is not enabled except via specific settings
3782 * of the module parameters.
3784 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
3787 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
3789 case KVM_CAP_NR_VCPUS:
3790 r = KVM_SOFT_MAX_VCPUS;
3792 case KVM_CAP_MAX_VCPUS:
3795 case KVM_CAP_MAX_VCPU_ID:
3796 r = KVM_MAX_VCPU_ID;
3798 case KVM_CAP_PV_MMU: /* obsolete */
3802 r = KVM_MAX_MCE_BANKS;
3805 r = boot_cpu_has(X86_FEATURE_XSAVE);
3807 case KVM_CAP_TSC_CONTROL:
3808 r = kvm_has_tsc_control;
3810 case KVM_CAP_X2APIC_API:
3811 r = KVM_X2APIC_API_VALID_FLAGS;
3813 case KVM_CAP_NESTED_STATE:
3814 r = kvm_x86_ops.nested_ops->get_state ?
3815 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3817 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3818 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3820 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3821 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3823 case KVM_CAP_SMALLER_MAXPHYADDR:
3824 r = (int) allow_smaller_maxphyaddr;
3826 case KVM_CAP_STEAL_TIME:
3827 r = sched_info_on();
3829 case KVM_CAP_X86_BUS_LOCK_EXIT:
3830 if (kvm_has_bus_lock_exit)
3831 r = KVM_BUS_LOCK_DETECTION_OFF |
3832 KVM_BUS_LOCK_DETECTION_EXIT;
3843 long kvm_arch_dev_ioctl(struct file *filp,
3844 unsigned int ioctl, unsigned long arg)
3846 void __user *argp = (void __user *)arg;
3850 case KVM_GET_MSR_INDEX_LIST: {
3851 struct kvm_msr_list __user *user_msr_list = argp;
3852 struct kvm_msr_list msr_list;
3856 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3859 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3860 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3863 if (n < msr_list.nmsrs)
3866 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3867 num_msrs_to_save * sizeof(u32)))
3869 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3871 num_emulated_msrs * sizeof(u32)))
3876 case KVM_GET_SUPPORTED_CPUID:
3877 case KVM_GET_EMULATED_CPUID: {
3878 struct kvm_cpuid2 __user *cpuid_arg = argp;
3879 struct kvm_cpuid2 cpuid;
3882 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3885 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3891 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3896 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3898 if (copy_to_user(argp, &kvm_mce_cap_supported,
3899 sizeof(kvm_mce_cap_supported)))
3903 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3904 struct kvm_msr_list __user *user_msr_list = argp;
3905 struct kvm_msr_list msr_list;
3909 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3912 msr_list.nmsrs = num_msr_based_features;
3913 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3916 if (n < msr_list.nmsrs)
3919 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3920 num_msr_based_features * sizeof(u32)))
3926 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3928 case KVM_GET_SUPPORTED_HV_CPUID:
3929 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
3939 static void wbinvd_ipi(void *garbage)
3944 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3946 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3949 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3951 /* Address WBINVD may be executed by guest */
3952 if (need_emulate_wbinvd(vcpu)) {
3953 if (static_call(kvm_x86_has_wbinvd_exit)())
3954 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3955 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3956 smp_call_function_single(vcpu->cpu,
3957 wbinvd_ipi, NULL, 1);
3960 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
3962 /* Save host pkru register if supported */
3963 vcpu->arch.host_pkru = read_pkru();
3965 /* Apply any externally detected TSC adjustments (due to suspend) */
3966 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3967 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3968 vcpu->arch.tsc_offset_adjustment = 0;
3969 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3972 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3973 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3974 rdtsc() - vcpu->arch.last_host_tsc;
3976 mark_tsc_unstable("KVM discovered backwards TSC");
3978 if (kvm_check_tsc_unstable()) {
3979 u64 offset = kvm_compute_tsc_offset(vcpu,
3980 vcpu->arch.last_guest_tsc);
3981 kvm_vcpu_write_tsc_offset(vcpu, offset);
3982 vcpu->arch.tsc_catchup = 1;
3985 if (kvm_lapic_hv_timer_in_use(vcpu))
3986 kvm_lapic_restart_hv_timer(vcpu);
3989 * On a host with synchronized TSC, there is no need to update
3990 * kvmclock on vcpu->cpu migration
3992 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3993 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3994 if (vcpu->cpu != cpu)
3995 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3999 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4002 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4004 struct kvm_host_map map;
4005 struct kvm_steal_time *st;
4008 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4011 if (vcpu->arch.st.preempted)
4015 * Take the srcu lock as memslots will be accessed to check the gfn
4016 * cache generation against the memslots generation.
4018 idx = srcu_read_lock(&vcpu->kvm->srcu);
4020 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4021 &vcpu->arch.st.cache, true))
4025 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4027 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4029 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4032 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4035 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4037 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4038 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4040 kvm_steal_time_set_preempted(vcpu);
4041 static_call(kvm_x86_vcpu_put)(vcpu);
4042 vcpu->arch.last_host_tsc = rdtsc();
4044 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4045 * on every vmexit, but if not, we might have a stale dr6 from the
4046 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4051 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4052 struct kvm_lapic_state *s)
4054 if (vcpu->arch.apicv_active)
4055 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4057 return kvm_apic_get_state(vcpu, s);
4060 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4061 struct kvm_lapic_state *s)
4065 r = kvm_apic_set_state(vcpu, s);
4068 update_cr8_intercept(vcpu);
4073 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4076 * We can accept userspace's request for interrupt injection
4077 * as long as we have a place to store the interrupt number.
4078 * The actual injection will happen when the CPU is able to
4079 * deliver the interrupt.
4081 if (kvm_cpu_has_extint(vcpu))
4084 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4085 return (!lapic_in_kernel(vcpu) ||
4086 kvm_apic_accept_pic_intr(vcpu));
4089 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4091 return kvm_arch_interrupt_allowed(vcpu) &&
4092 kvm_cpu_accept_dm_intr(vcpu);
4095 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4096 struct kvm_interrupt *irq)
4098 if (irq->irq >= KVM_NR_INTERRUPTS)
4101 if (!irqchip_in_kernel(vcpu->kvm)) {
4102 kvm_queue_interrupt(vcpu, irq->irq, false);
4103 kvm_make_request(KVM_REQ_EVENT, vcpu);
4108 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4109 * fail for in-kernel 8259.
4111 if (pic_in_kernel(vcpu->kvm))
4114 if (vcpu->arch.pending_external_vector != -1)
4117 vcpu->arch.pending_external_vector = irq->irq;
4118 kvm_make_request(KVM_REQ_EVENT, vcpu);
4122 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4124 kvm_inject_nmi(vcpu);
4129 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4131 kvm_make_request(KVM_REQ_SMI, vcpu);
4136 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4137 struct kvm_tpr_access_ctl *tac)
4141 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4145 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4149 unsigned bank_num = mcg_cap & 0xff, bank;
4152 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4154 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4157 vcpu->arch.mcg_cap = mcg_cap;
4158 /* Init IA32_MCG_CTL to all 1s */
4159 if (mcg_cap & MCG_CTL_P)
4160 vcpu->arch.mcg_ctl = ~(u64)0;
4161 /* Init IA32_MCi_CTL to all 1s */
4162 for (bank = 0; bank < bank_num; bank++)
4163 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4165 static_call(kvm_x86_setup_mce)(vcpu);
4170 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4171 struct kvm_x86_mce *mce)
4173 u64 mcg_cap = vcpu->arch.mcg_cap;
4174 unsigned bank_num = mcg_cap & 0xff;
4175 u64 *banks = vcpu->arch.mce_banks;
4177 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4180 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4181 * reporting is disabled
4183 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4184 vcpu->arch.mcg_ctl != ~(u64)0)
4186 banks += 4 * mce->bank;
4188 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4189 * reporting is disabled for the bank
4191 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4193 if (mce->status & MCI_STATUS_UC) {
4194 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4195 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4196 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4199 if (banks[1] & MCI_STATUS_VAL)
4200 mce->status |= MCI_STATUS_OVER;
4201 banks[2] = mce->addr;
4202 banks[3] = mce->misc;
4203 vcpu->arch.mcg_status = mce->mcg_status;
4204 banks[1] = mce->status;
4205 kvm_queue_exception(vcpu, MC_VECTOR);
4206 } else if (!(banks[1] & MCI_STATUS_VAL)
4207 || !(banks[1] & MCI_STATUS_UC)) {
4208 if (banks[1] & MCI_STATUS_VAL)
4209 mce->status |= MCI_STATUS_OVER;
4210 banks[2] = mce->addr;
4211 banks[3] = mce->misc;
4212 banks[1] = mce->status;
4214 banks[1] |= MCI_STATUS_OVER;
4218 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4219 struct kvm_vcpu_events *events)
4223 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4227 * In guest mode, payload delivery should be deferred,
4228 * so that the L1 hypervisor can intercept #PF before
4229 * CR2 is modified (or intercept #DB before DR6 is
4230 * modified under nVMX). Unless the per-VM capability,
4231 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4232 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4233 * opportunistically defer the exception payload, deliver it if the
4234 * capability hasn't been requested before processing a
4235 * KVM_GET_VCPU_EVENTS.
4237 if (!vcpu->kvm->arch.exception_payload_enabled &&
4238 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4239 kvm_deliver_exception_payload(vcpu);
4242 * The API doesn't provide the instruction length for software
4243 * exceptions, so don't report them. As long as the guest RIP
4244 * isn't advanced, we should expect to encounter the exception
4247 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4248 events->exception.injected = 0;
4249 events->exception.pending = 0;
4251 events->exception.injected = vcpu->arch.exception.injected;
4252 events->exception.pending = vcpu->arch.exception.pending;
4254 * For ABI compatibility, deliberately conflate
4255 * pending and injected exceptions when
4256 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4258 if (!vcpu->kvm->arch.exception_payload_enabled)
4259 events->exception.injected |=
4260 vcpu->arch.exception.pending;
4262 events->exception.nr = vcpu->arch.exception.nr;
4263 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4264 events->exception.error_code = vcpu->arch.exception.error_code;
4265 events->exception_has_payload = vcpu->arch.exception.has_payload;
4266 events->exception_payload = vcpu->arch.exception.payload;
4268 events->interrupt.injected =
4269 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4270 events->interrupt.nr = vcpu->arch.interrupt.nr;
4271 events->interrupt.soft = 0;
4272 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4274 events->nmi.injected = vcpu->arch.nmi_injected;
4275 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4276 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4277 events->nmi.pad = 0;
4279 events->sipi_vector = 0; /* never valid when reporting to user space */
4281 events->smi.smm = is_smm(vcpu);
4282 events->smi.pending = vcpu->arch.smi_pending;
4283 events->smi.smm_inside_nmi =
4284 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4285 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4287 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4288 | KVM_VCPUEVENT_VALID_SHADOW
4289 | KVM_VCPUEVENT_VALID_SMM);
4290 if (vcpu->kvm->arch.exception_payload_enabled)
4291 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4293 memset(&events->reserved, 0, sizeof(events->reserved));
4296 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4298 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4299 struct kvm_vcpu_events *events)
4301 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4302 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4303 | KVM_VCPUEVENT_VALID_SHADOW
4304 | KVM_VCPUEVENT_VALID_SMM
4305 | KVM_VCPUEVENT_VALID_PAYLOAD))
4308 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4309 if (!vcpu->kvm->arch.exception_payload_enabled)
4311 if (events->exception.pending)
4312 events->exception.injected = 0;
4314 events->exception_has_payload = 0;
4316 events->exception.pending = 0;
4317 events->exception_has_payload = 0;
4320 if ((events->exception.injected || events->exception.pending) &&
4321 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4324 /* INITs are latched while in SMM */
4325 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4326 (events->smi.smm || events->smi.pending) &&
4327 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4331 vcpu->arch.exception.injected = events->exception.injected;
4332 vcpu->arch.exception.pending = events->exception.pending;
4333 vcpu->arch.exception.nr = events->exception.nr;
4334 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4335 vcpu->arch.exception.error_code = events->exception.error_code;
4336 vcpu->arch.exception.has_payload = events->exception_has_payload;
4337 vcpu->arch.exception.payload = events->exception_payload;
4339 vcpu->arch.interrupt.injected = events->interrupt.injected;
4340 vcpu->arch.interrupt.nr = events->interrupt.nr;
4341 vcpu->arch.interrupt.soft = events->interrupt.soft;
4342 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4343 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4344 events->interrupt.shadow);
4346 vcpu->arch.nmi_injected = events->nmi.injected;
4347 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4348 vcpu->arch.nmi_pending = events->nmi.pending;
4349 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4351 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4352 lapic_in_kernel(vcpu))
4353 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4355 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4356 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4357 if (events->smi.smm)
4358 vcpu->arch.hflags |= HF_SMM_MASK;
4360 vcpu->arch.hflags &= ~HF_SMM_MASK;
4361 kvm_smm_changed(vcpu);
4364 vcpu->arch.smi_pending = events->smi.pending;
4366 if (events->smi.smm) {
4367 if (events->smi.smm_inside_nmi)
4368 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4370 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4373 if (lapic_in_kernel(vcpu)) {
4374 if (events->smi.latched_init)
4375 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4377 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4381 kvm_make_request(KVM_REQ_EVENT, vcpu);
4386 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4387 struct kvm_debugregs *dbgregs)
4391 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4392 kvm_get_dr(vcpu, 6, &val);
4394 dbgregs->dr7 = vcpu->arch.dr7;
4396 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4399 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4400 struct kvm_debugregs *dbgregs)
4405 if (!kvm_dr6_valid(dbgregs->dr6))
4407 if (!kvm_dr7_valid(dbgregs->dr7))
4410 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4411 kvm_update_dr0123(vcpu);
4412 vcpu->arch.dr6 = dbgregs->dr6;
4413 vcpu->arch.dr7 = dbgregs->dr7;
4414 kvm_update_dr7(vcpu);
4419 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4421 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4423 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4424 u64 xstate_bv = xsave->header.xfeatures;
4428 * Copy legacy XSAVE area, to avoid complications with CPUID
4429 * leaves 0 and 1 in the loop below.
4431 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4434 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4435 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4438 * Copy each region from the possibly compacted offset to the
4439 * non-compacted offset.
4441 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4443 u64 xfeature_mask = valid & -valid;
4444 int xfeature_nr = fls64(xfeature_mask) - 1;
4445 void *src = get_xsave_addr(xsave, xfeature_nr);
4448 u32 size, offset, ecx, edx;
4449 cpuid_count(XSTATE_CPUID, xfeature_nr,
4450 &size, &offset, &ecx, &edx);
4451 if (xfeature_nr == XFEATURE_PKRU)
4452 memcpy(dest + offset, &vcpu->arch.pkru,
4453 sizeof(vcpu->arch.pkru));
4455 memcpy(dest + offset, src, size);
4459 valid -= xfeature_mask;
4463 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4465 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4466 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4470 * Copy legacy XSAVE area, to avoid complications with CPUID
4471 * leaves 0 and 1 in the loop below.
4473 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4475 /* Set XSTATE_BV and possibly XCOMP_BV. */
4476 xsave->header.xfeatures = xstate_bv;
4477 if (boot_cpu_has(X86_FEATURE_XSAVES))
4478 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4481 * Copy each region from the non-compacted offset to the
4482 * possibly compacted offset.
4484 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4486 u64 xfeature_mask = valid & -valid;
4487 int xfeature_nr = fls64(xfeature_mask) - 1;
4488 void *dest = get_xsave_addr(xsave, xfeature_nr);
4491 u32 size, offset, ecx, edx;
4492 cpuid_count(XSTATE_CPUID, xfeature_nr,
4493 &size, &offset, &ecx, &edx);
4494 if (xfeature_nr == XFEATURE_PKRU)
4495 memcpy(&vcpu->arch.pkru, src + offset,
4496 sizeof(vcpu->arch.pkru));
4498 memcpy(dest, src + offset, size);
4501 valid -= xfeature_mask;
4505 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4506 struct kvm_xsave *guest_xsave)
4508 if (!vcpu->arch.guest_fpu)
4511 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4512 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4513 fill_xsave((u8 *) guest_xsave->region, vcpu);
4515 memcpy(guest_xsave->region,
4516 &vcpu->arch.guest_fpu->state.fxsave,
4517 sizeof(struct fxregs_state));
4518 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4519 XFEATURE_MASK_FPSSE;
4523 #define XSAVE_MXCSR_OFFSET 24
4525 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4526 struct kvm_xsave *guest_xsave)
4531 if (!vcpu->arch.guest_fpu)
4534 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4535 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4537 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4539 * Here we allow setting states that are not present in
4540 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4541 * with old userspace.
4543 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4545 load_xsave(vcpu, (u8 *)guest_xsave->region);
4547 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4548 mxcsr & ~mxcsr_feature_mask)
4550 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4551 guest_xsave->region, sizeof(struct fxregs_state));
4556 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4557 struct kvm_xcrs *guest_xcrs)
4559 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4560 guest_xcrs->nr_xcrs = 0;
4564 guest_xcrs->nr_xcrs = 1;
4565 guest_xcrs->flags = 0;
4566 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4567 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4570 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4571 struct kvm_xcrs *guest_xcrs)
4575 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4578 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4581 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4582 /* Only support XCR0 currently */
4583 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4584 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4585 guest_xcrs->xcrs[i].value);
4594 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4595 * stopped by the hypervisor. This function will be called from the host only.
4596 * EINVAL is returned when the host attempts to set the flag for a guest that
4597 * does not support pv clocks.
4599 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4601 if (!vcpu->arch.pv_time_enabled)
4603 vcpu->arch.pvclock_set_guest_stopped_request = true;
4604 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4608 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4609 struct kvm_enable_cap *cap)
4612 uint16_t vmcs_version;
4613 void __user *user_ptr;
4619 case KVM_CAP_HYPERV_SYNIC2:
4624 case KVM_CAP_HYPERV_SYNIC:
4625 if (!irqchip_in_kernel(vcpu->kvm))
4627 return kvm_hv_activate_synic(vcpu, cap->cap ==
4628 KVM_CAP_HYPERV_SYNIC2);
4629 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4630 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4632 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4634 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4635 if (copy_to_user(user_ptr, &vmcs_version,
4636 sizeof(vmcs_version)))
4640 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4641 if (!kvm_x86_ops.enable_direct_tlbflush)
4644 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4646 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4647 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4648 if (vcpu->arch.pv_cpuid.enforce)
4649 kvm_update_pv_runtime(vcpu);
4658 long kvm_arch_vcpu_ioctl(struct file *filp,
4659 unsigned int ioctl, unsigned long arg)
4661 struct kvm_vcpu *vcpu = filp->private_data;
4662 void __user *argp = (void __user *)arg;
4665 struct kvm_lapic_state *lapic;
4666 struct kvm_xsave *xsave;
4667 struct kvm_xcrs *xcrs;
4675 case KVM_GET_LAPIC: {
4677 if (!lapic_in_kernel(vcpu))
4679 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4680 GFP_KERNEL_ACCOUNT);
4685 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4689 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4694 case KVM_SET_LAPIC: {
4696 if (!lapic_in_kernel(vcpu))
4698 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4699 if (IS_ERR(u.lapic)) {
4700 r = PTR_ERR(u.lapic);
4704 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4707 case KVM_INTERRUPT: {
4708 struct kvm_interrupt irq;
4711 if (copy_from_user(&irq, argp, sizeof(irq)))
4713 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4717 r = kvm_vcpu_ioctl_nmi(vcpu);
4721 r = kvm_vcpu_ioctl_smi(vcpu);
4724 case KVM_SET_CPUID: {
4725 struct kvm_cpuid __user *cpuid_arg = argp;
4726 struct kvm_cpuid cpuid;
4729 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4731 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4734 case KVM_SET_CPUID2: {
4735 struct kvm_cpuid2 __user *cpuid_arg = argp;
4736 struct kvm_cpuid2 cpuid;
4739 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4741 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4742 cpuid_arg->entries);
4745 case KVM_GET_CPUID2: {
4746 struct kvm_cpuid2 __user *cpuid_arg = argp;
4747 struct kvm_cpuid2 cpuid;
4750 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4752 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4753 cpuid_arg->entries);
4757 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4762 case KVM_GET_MSRS: {
4763 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4764 r = msr_io(vcpu, argp, do_get_msr, 1);
4765 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4768 case KVM_SET_MSRS: {
4769 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4770 r = msr_io(vcpu, argp, do_set_msr, 0);
4771 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4774 case KVM_TPR_ACCESS_REPORTING: {
4775 struct kvm_tpr_access_ctl tac;
4778 if (copy_from_user(&tac, argp, sizeof(tac)))
4780 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4784 if (copy_to_user(argp, &tac, sizeof(tac)))
4789 case KVM_SET_VAPIC_ADDR: {
4790 struct kvm_vapic_addr va;
4794 if (!lapic_in_kernel(vcpu))
4797 if (copy_from_user(&va, argp, sizeof(va)))
4799 idx = srcu_read_lock(&vcpu->kvm->srcu);
4800 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4801 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4804 case KVM_X86_SETUP_MCE: {
4808 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4810 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4813 case KVM_X86_SET_MCE: {
4814 struct kvm_x86_mce mce;
4817 if (copy_from_user(&mce, argp, sizeof(mce)))
4819 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4822 case KVM_GET_VCPU_EVENTS: {
4823 struct kvm_vcpu_events events;
4825 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4828 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4833 case KVM_SET_VCPU_EVENTS: {
4834 struct kvm_vcpu_events events;
4837 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4840 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4843 case KVM_GET_DEBUGREGS: {
4844 struct kvm_debugregs dbgregs;
4846 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4849 if (copy_to_user(argp, &dbgregs,
4850 sizeof(struct kvm_debugregs)))
4855 case KVM_SET_DEBUGREGS: {
4856 struct kvm_debugregs dbgregs;
4859 if (copy_from_user(&dbgregs, argp,
4860 sizeof(struct kvm_debugregs)))
4863 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4866 case KVM_GET_XSAVE: {
4867 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4872 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4875 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4880 case KVM_SET_XSAVE: {
4881 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4882 if (IS_ERR(u.xsave)) {
4883 r = PTR_ERR(u.xsave);
4887 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4890 case KVM_GET_XCRS: {
4891 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4896 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4899 if (copy_to_user(argp, u.xcrs,
4900 sizeof(struct kvm_xcrs)))
4905 case KVM_SET_XCRS: {
4906 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4907 if (IS_ERR(u.xcrs)) {
4908 r = PTR_ERR(u.xcrs);
4912 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4915 case KVM_SET_TSC_KHZ: {
4919 user_tsc_khz = (u32)arg;
4921 if (kvm_has_tsc_control &&
4922 user_tsc_khz >= kvm_max_guest_tsc_khz)
4925 if (user_tsc_khz == 0)
4926 user_tsc_khz = tsc_khz;
4928 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4933 case KVM_GET_TSC_KHZ: {
4934 r = vcpu->arch.virtual_tsc_khz;
4937 case KVM_KVMCLOCK_CTRL: {
4938 r = kvm_set_guest_paused(vcpu);
4941 case KVM_ENABLE_CAP: {
4942 struct kvm_enable_cap cap;
4945 if (copy_from_user(&cap, argp, sizeof(cap)))
4947 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4950 case KVM_GET_NESTED_STATE: {
4951 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4955 if (!kvm_x86_ops.nested_ops->get_state)
4958 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4960 if (get_user(user_data_size, &user_kvm_nested_state->size))
4963 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4968 if (r > user_data_size) {
4969 if (put_user(r, &user_kvm_nested_state->size))
4979 case KVM_SET_NESTED_STATE: {
4980 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4981 struct kvm_nested_state kvm_state;
4985 if (!kvm_x86_ops.nested_ops->set_state)
4989 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4993 if (kvm_state.size < sizeof(kvm_state))
4996 if (kvm_state.flags &
4997 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4998 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
4999 | KVM_STATE_NESTED_GIF_SET))
5002 /* nested_run_pending implies guest_mode. */
5003 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5004 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5007 idx = srcu_read_lock(&vcpu->kvm->srcu);
5008 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5009 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5012 case KVM_GET_SUPPORTED_HV_CPUID:
5013 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5015 case KVM_XEN_VCPU_GET_ATTR: {
5016 struct kvm_xen_vcpu_attr xva;
5019 if (copy_from_user(&xva, argp, sizeof(xva)))
5021 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5022 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5026 case KVM_XEN_VCPU_SET_ATTR: {
5027 struct kvm_xen_vcpu_attr xva;
5030 if (copy_from_user(&xva, argp, sizeof(xva)))
5032 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5045 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5047 return VM_FAULT_SIGBUS;
5050 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5054 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5056 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5060 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5063 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5066 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5067 unsigned long kvm_nr_mmu_pages)
5069 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5072 mutex_lock(&kvm->slots_lock);
5074 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5075 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5077 mutex_unlock(&kvm->slots_lock);
5081 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5083 return kvm->arch.n_max_mmu_pages;
5086 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5088 struct kvm_pic *pic = kvm->arch.vpic;
5092 switch (chip->chip_id) {
5093 case KVM_IRQCHIP_PIC_MASTER:
5094 memcpy(&chip->chip.pic, &pic->pics[0],
5095 sizeof(struct kvm_pic_state));
5097 case KVM_IRQCHIP_PIC_SLAVE:
5098 memcpy(&chip->chip.pic, &pic->pics[1],
5099 sizeof(struct kvm_pic_state));
5101 case KVM_IRQCHIP_IOAPIC:
5102 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5111 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5113 struct kvm_pic *pic = kvm->arch.vpic;
5117 switch (chip->chip_id) {
5118 case KVM_IRQCHIP_PIC_MASTER:
5119 spin_lock(&pic->lock);
5120 memcpy(&pic->pics[0], &chip->chip.pic,
5121 sizeof(struct kvm_pic_state));
5122 spin_unlock(&pic->lock);
5124 case KVM_IRQCHIP_PIC_SLAVE:
5125 spin_lock(&pic->lock);
5126 memcpy(&pic->pics[1], &chip->chip.pic,
5127 sizeof(struct kvm_pic_state));
5128 spin_unlock(&pic->lock);
5130 case KVM_IRQCHIP_IOAPIC:
5131 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5137 kvm_pic_update_irq(pic);
5141 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5143 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5145 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5147 mutex_lock(&kps->lock);
5148 memcpy(ps, &kps->channels, sizeof(*ps));
5149 mutex_unlock(&kps->lock);
5153 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5156 struct kvm_pit *pit = kvm->arch.vpit;
5158 mutex_lock(&pit->pit_state.lock);
5159 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5160 for (i = 0; i < 3; i++)
5161 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5162 mutex_unlock(&pit->pit_state.lock);
5166 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5168 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5169 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5170 sizeof(ps->channels));
5171 ps->flags = kvm->arch.vpit->pit_state.flags;
5172 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5173 memset(&ps->reserved, 0, sizeof(ps->reserved));
5177 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5181 u32 prev_legacy, cur_legacy;
5182 struct kvm_pit *pit = kvm->arch.vpit;
5184 mutex_lock(&pit->pit_state.lock);
5185 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5186 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5187 if (!prev_legacy && cur_legacy)
5189 memcpy(&pit->pit_state.channels, &ps->channels,
5190 sizeof(pit->pit_state.channels));
5191 pit->pit_state.flags = ps->flags;
5192 for (i = 0; i < 3; i++)
5193 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5195 mutex_unlock(&pit->pit_state.lock);
5199 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5200 struct kvm_reinject_control *control)
5202 struct kvm_pit *pit = kvm->arch.vpit;
5204 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5205 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5206 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5208 mutex_lock(&pit->pit_state.lock);
5209 kvm_pit_set_reinject(pit, control->pit_reinject);
5210 mutex_unlock(&pit->pit_state.lock);
5215 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5218 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
5220 static_call_cond(kvm_x86_flush_log_dirty)(kvm);
5223 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5226 if (!irqchip_in_kernel(kvm))
5229 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5230 irq_event->irq, irq_event->level,
5235 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5236 struct kvm_enable_cap *cap)
5244 case KVM_CAP_DISABLE_QUIRKS:
5245 kvm->arch.disabled_quirks = cap->args[0];
5248 case KVM_CAP_SPLIT_IRQCHIP: {
5249 mutex_lock(&kvm->lock);
5251 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5252 goto split_irqchip_unlock;
5254 if (irqchip_in_kernel(kvm))
5255 goto split_irqchip_unlock;
5256 if (kvm->created_vcpus)
5257 goto split_irqchip_unlock;
5258 r = kvm_setup_empty_irq_routing(kvm);
5260 goto split_irqchip_unlock;
5261 /* Pairs with irqchip_in_kernel. */
5263 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5264 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5266 split_irqchip_unlock:
5267 mutex_unlock(&kvm->lock);
5270 case KVM_CAP_X2APIC_API:
5272 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5275 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5276 kvm->arch.x2apic_format = true;
5277 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5278 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5282 case KVM_CAP_X86_DISABLE_EXITS:
5284 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5287 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5288 kvm_can_mwait_in_guest())
5289 kvm->arch.mwait_in_guest = true;
5290 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5291 kvm->arch.hlt_in_guest = true;
5292 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5293 kvm->arch.pause_in_guest = true;
5294 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5295 kvm->arch.cstate_in_guest = true;
5298 case KVM_CAP_MSR_PLATFORM_INFO:
5299 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5302 case KVM_CAP_EXCEPTION_PAYLOAD:
5303 kvm->arch.exception_payload_enabled = cap->args[0];
5306 case KVM_CAP_X86_USER_SPACE_MSR:
5307 kvm->arch.user_space_msr_mask = cap->args[0];
5310 case KVM_CAP_X86_BUS_LOCK_EXIT:
5312 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5315 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5316 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5319 if (kvm_has_bus_lock_exit &&
5320 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5321 kvm->arch.bus_lock_detection_enabled = true;
5331 static void kvm_clear_msr_filter(struct kvm *kvm)
5334 u32 count = kvm->arch.msr_filter.count;
5335 struct msr_bitmap_range ranges[16];
5337 mutex_lock(&kvm->lock);
5338 kvm->arch.msr_filter.count = 0;
5339 memcpy(ranges, kvm->arch.msr_filter.ranges, count * sizeof(ranges[0]));
5340 mutex_unlock(&kvm->lock);
5341 synchronize_srcu(&kvm->srcu);
5343 for (i = 0; i < count; i++)
5344 kfree(ranges[i].bitmap);
5347 static int kvm_add_msr_filter(struct kvm *kvm, struct kvm_msr_filter_range *user_range)
5349 struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
5350 struct msr_bitmap_range range;
5351 unsigned long *bitmap = NULL;
5355 if (!user_range->nmsrs)
5358 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5359 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5362 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5364 return PTR_ERR(bitmap);
5366 range = (struct msr_bitmap_range) {
5367 .flags = user_range->flags,
5368 .base = user_range->base,
5369 .nmsrs = user_range->nmsrs,
5373 if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) {
5383 /* Everything ok, add this range identifier to our global pool */
5384 ranges[kvm->arch.msr_filter.count] = range;
5385 /* Make sure we filled the array before we tell anyone to walk it */
5387 kvm->arch.msr_filter.count++;
5395 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5397 struct kvm_msr_filter __user *user_msr_filter = argp;
5398 struct kvm_msr_filter filter;
5404 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5407 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5408 empty &= !filter.ranges[i].nmsrs;
5410 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5411 if (empty && !default_allow)
5414 kvm_clear_msr_filter(kvm);
5416 kvm->arch.msr_filter.default_allow = default_allow;
5419 * Protect from concurrent calls to this function that could trigger
5420 * a TOCTOU violation on kvm->arch.msr_filter.count.
5422 mutex_lock(&kvm->lock);
5423 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5424 r = kvm_add_msr_filter(kvm, &filter.ranges[i]);
5429 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5430 mutex_unlock(&kvm->lock);
5435 long kvm_arch_vm_ioctl(struct file *filp,
5436 unsigned int ioctl, unsigned long arg)
5438 struct kvm *kvm = filp->private_data;
5439 void __user *argp = (void __user *)arg;
5442 * This union makes it completely explicit to gcc-3.x
5443 * that these two variables' stack usage should be
5444 * combined, not added together.
5447 struct kvm_pit_state ps;
5448 struct kvm_pit_state2 ps2;
5449 struct kvm_pit_config pit_config;
5453 case KVM_SET_TSS_ADDR:
5454 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5456 case KVM_SET_IDENTITY_MAP_ADDR: {
5459 mutex_lock(&kvm->lock);
5461 if (kvm->created_vcpus)
5462 goto set_identity_unlock;
5464 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5465 goto set_identity_unlock;
5466 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5467 set_identity_unlock:
5468 mutex_unlock(&kvm->lock);
5471 case KVM_SET_NR_MMU_PAGES:
5472 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5474 case KVM_GET_NR_MMU_PAGES:
5475 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5477 case KVM_CREATE_IRQCHIP: {
5478 mutex_lock(&kvm->lock);
5481 if (irqchip_in_kernel(kvm))
5482 goto create_irqchip_unlock;
5485 if (kvm->created_vcpus)
5486 goto create_irqchip_unlock;
5488 r = kvm_pic_init(kvm);
5490 goto create_irqchip_unlock;
5492 r = kvm_ioapic_init(kvm);
5494 kvm_pic_destroy(kvm);
5495 goto create_irqchip_unlock;
5498 r = kvm_setup_default_irq_routing(kvm);
5500 kvm_ioapic_destroy(kvm);
5501 kvm_pic_destroy(kvm);
5502 goto create_irqchip_unlock;
5504 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5506 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5507 create_irqchip_unlock:
5508 mutex_unlock(&kvm->lock);
5511 case KVM_CREATE_PIT:
5512 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5514 case KVM_CREATE_PIT2:
5516 if (copy_from_user(&u.pit_config, argp,
5517 sizeof(struct kvm_pit_config)))
5520 mutex_lock(&kvm->lock);
5523 goto create_pit_unlock;
5525 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5529 mutex_unlock(&kvm->lock);
5531 case KVM_GET_IRQCHIP: {
5532 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5533 struct kvm_irqchip *chip;
5535 chip = memdup_user(argp, sizeof(*chip));
5542 if (!irqchip_kernel(kvm))
5543 goto get_irqchip_out;
5544 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5546 goto get_irqchip_out;
5548 if (copy_to_user(argp, chip, sizeof(*chip)))
5549 goto get_irqchip_out;
5555 case KVM_SET_IRQCHIP: {
5556 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5557 struct kvm_irqchip *chip;
5559 chip = memdup_user(argp, sizeof(*chip));
5566 if (!irqchip_kernel(kvm))
5567 goto set_irqchip_out;
5568 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5575 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5578 if (!kvm->arch.vpit)
5580 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5584 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5591 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5593 mutex_lock(&kvm->lock);
5595 if (!kvm->arch.vpit)
5597 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5599 mutex_unlock(&kvm->lock);
5602 case KVM_GET_PIT2: {
5604 if (!kvm->arch.vpit)
5606 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5610 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5615 case KVM_SET_PIT2: {
5617 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5619 mutex_lock(&kvm->lock);
5621 if (!kvm->arch.vpit)
5623 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5625 mutex_unlock(&kvm->lock);
5628 case KVM_REINJECT_CONTROL: {
5629 struct kvm_reinject_control control;
5631 if (copy_from_user(&control, argp, sizeof(control)))
5634 if (!kvm->arch.vpit)
5636 r = kvm_vm_ioctl_reinject(kvm, &control);
5639 case KVM_SET_BOOT_CPU_ID:
5641 mutex_lock(&kvm->lock);
5642 if (kvm->created_vcpus)
5645 kvm->arch.bsp_vcpu_id = arg;
5646 mutex_unlock(&kvm->lock);
5648 case KVM_XEN_HVM_CONFIG: {
5649 struct kvm_xen_hvm_config xhc;
5651 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5653 r = kvm_xen_hvm_config(kvm, &xhc);
5656 case KVM_XEN_HVM_GET_ATTR: {
5657 struct kvm_xen_hvm_attr xha;
5660 if (copy_from_user(&xha, argp, sizeof(xha)))
5662 r = kvm_xen_hvm_get_attr(kvm, &xha);
5663 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5667 case KVM_XEN_HVM_SET_ATTR: {
5668 struct kvm_xen_hvm_attr xha;
5671 if (copy_from_user(&xha, argp, sizeof(xha)))
5673 r = kvm_xen_hvm_set_attr(kvm, &xha);
5676 case KVM_SET_CLOCK: {
5677 struct kvm_clock_data user_ns;
5681 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5690 * TODO: userspace has to take care of races with VCPU_RUN, so
5691 * kvm_gen_update_masterclock() can be cut down to locked
5692 * pvclock_update_vm_gtod_copy().
5694 kvm_gen_update_masterclock(kvm);
5695 now_ns = get_kvmclock_ns(kvm);
5696 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5697 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5700 case KVM_GET_CLOCK: {
5701 struct kvm_clock_data user_ns;
5704 now_ns = get_kvmclock_ns(kvm);
5705 user_ns.clock = now_ns;
5706 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5707 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5710 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5715 case KVM_MEMORY_ENCRYPT_OP: {
5717 if (kvm_x86_ops.mem_enc_op)
5718 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
5721 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5722 struct kvm_enc_region region;
5725 if (copy_from_user(®ion, argp, sizeof(region)))
5729 if (kvm_x86_ops.mem_enc_reg_region)
5730 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
5733 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5734 struct kvm_enc_region region;
5737 if (copy_from_user(®ion, argp, sizeof(region)))
5741 if (kvm_x86_ops.mem_enc_unreg_region)
5742 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
5745 case KVM_HYPERV_EVENTFD: {
5746 struct kvm_hyperv_eventfd hvevfd;
5749 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5751 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5754 case KVM_SET_PMU_EVENT_FILTER:
5755 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5757 case KVM_X86_SET_MSR_FILTER:
5758 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5767 static void kvm_init_msr_list(void)
5769 struct x86_pmu_capability x86_pmu;
5773 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5774 "Please update the fixed PMCs in msrs_to_saved_all[]");
5776 perf_get_x86_pmu_capability(&x86_pmu);
5778 num_msrs_to_save = 0;
5779 num_emulated_msrs = 0;
5780 num_msr_based_features = 0;
5782 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5783 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5787 * Even MSRs that are valid in the host may not be exposed
5788 * to the guests in some cases.
5790 switch (msrs_to_save_all[i]) {
5791 case MSR_IA32_BNDCFGS:
5792 if (!kvm_mpx_supported())
5796 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5799 case MSR_IA32_UMWAIT_CONTROL:
5800 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5803 case MSR_IA32_RTIT_CTL:
5804 case MSR_IA32_RTIT_STATUS:
5805 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5808 case MSR_IA32_RTIT_CR3_MATCH:
5809 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5810 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5813 case MSR_IA32_RTIT_OUTPUT_BASE:
5814 case MSR_IA32_RTIT_OUTPUT_MASK:
5815 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5816 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5817 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5820 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5821 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5822 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5823 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5826 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5827 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5828 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5831 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5832 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5833 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5840 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5843 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5844 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
5847 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5850 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5851 struct kvm_msr_entry msr;
5853 msr.index = msr_based_features_all[i];
5854 if (kvm_get_msr_feature(&msr))
5857 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5861 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5869 if (!(lapic_in_kernel(vcpu) &&
5870 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5871 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5882 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5889 if (!(lapic_in_kernel(vcpu) &&
5890 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5892 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5894 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5904 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5905 struct kvm_segment *var, int seg)
5907 static_call(kvm_x86_set_segment)(vcpu, var, seg);
5910 void kvm_get_segment(struct kvm_vcpu *vcpu,
5911 struct kvm_segment *var, int seg)
5913 static_call(kvm_x86_get_segment)(vcpu, var, seg);
5916 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5917 struct x86_exception *exception)
5921 BUG_ON(!mmu_is_nested(vcpu));
5923 /* NPT walks are always user-walks */
5924 access |= PFERR_USER_MASK;
5925 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5930 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5931 struct x86_exception *exception)
5933 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
5934 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5937 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5938 struct x86_exception *exception)
5940 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
5941 access |= PFERR_FETCH_MASK;
5942 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5945 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5946 struct x86_exception *exception)
5948 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
5949 access |= PFERR_WRITE_MASK;
5950 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5953 /* uses this to access any guest's mapped memory without checking CPL */
5954 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5955 struct x86_exception *exception)
5957 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5960 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5961 struct kvm_vcpu *vcpu, u32 access,
5962 struct x86_exception *exception)
5965 int r = X86EMUL_CONTINUE;
5968 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5970 unsigned offset = addr & (PAGE_SIZE-1);
5971 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5974 if (gpa == UNMAPPED_GVA)
5975 return X86EMUL_PROPAGATE_FAULT;
5976 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5979 r = X86EMUL_IO_NEEDED;
5991 /* used for instruction fetching */
5992 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5993 gva_t addr, void *val, unsigned int bytes,
5994 struct x86_exception *exception)
5996 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5997 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6001 /* Inline kvm_read_guest_virt_helper for speed. */
6002 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6004 if (unlikely(gpa == UNMAPPED_GVA))
6005 return X86EMUL_PROPAGATE_FAULT;
6007 offset = addr & (PAGE_SIZE-1);
6008 if (WARN_ON(offset + bytes > PAGE_SIZE))
6009 bytes = (unsigned)PAGE_SIZE - offset;
6010 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6012 if (unlikely(ret < 0))
6013 return X86EMUL_IO_NEEDED;
6015 return X86EMUL_CONTINUE;
6018 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6019 gva_t addr, void *val, unsigned int bytes,
6020 struct x86_exception *exception)
6022 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6025 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6026 * is returned, but our callers are not ready for that and they blindly
6027 * call kvm_inject_page_fault. Ensure that they at least do not leak
6028 * uninitialized kernel stack memory into cr2 and error code.
6030 memset(exception, 0, sizeof(*exception));
6031 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6034 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6036 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6037 gva_t addr, void *val, unsigned int bytes,
6038 struct x86_exception *exception, bool system)
6040 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6043 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6044 access |= PFERR_USER_MASK;
6046 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6049 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6050 unsigned long addr, void *val, unsigned int bytes)
6052 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6053 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6055 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6058 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6059 struct kvm_vcpu *vcpu, u32 access,
6060 struct x86_exception *exception)
6063 int r = X86EMUL_CONTINUE;
6066 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6069 unsigned offset = addr & (PAGE_SIZE-1);
6070 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6073 if (gpa == UNMAPPED_GVA)
6074 return X86EMUL_PROPAGATE_FAULT;
6075 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6077 r = X86EMUL_IO_NEEDED;
6089 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6090 unsigned int bytes, struct x86_exception *exception,
6093 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6094 u32 access = PFERR_WRITE_MASK;
6096 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6097 access |= PFERR_USER_MASK;
6099 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6103 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6104 unsigned int bytes, struct x86_exception *exception)
6106 /* kvm_write_guest_virt_system can pull in tons of pages. */
6107 vcpu->arch.l1tf_flush_l1d = true;
6109 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6110 PFERR_WRITE_MASK, exception);
6112 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6114 int handle_ud(struct kvm_vcpu *vcpu)
6116 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6117 int emul_type = EMULTYPE_TRAP_UD;
6118 char sig[5]; /* ud2; .ascii "kvm" */
6119 struct x86_exception e;
6121 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6124 if (force_emulation_prefix &&
6125 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6126 sig, sizeof(sig), &e) == 0 &&
6127 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6128 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6129 emul_type = EMULTYPE_TRAP_UD_FORCED;
6132 return kvm_emulate_instruction(vcpu, emul_type);
6134 EXPORT_SYMBOL_GPL(handle_ud);
6136 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6137 gpa_t gpa, bool write)
6139 /* For APIC access vmexit */
6140 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6143 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6144 trace_vcpu_match_mmio(gva, gpa, write, true);
6151 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6152 gpa_t *gpa, struct x86_exception *exception,
6155 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6156 | (write ? PFERR_WRITE_MASK : 0);
6159 * currently PKRU is only applied to ept enabled guest so
6160 * there is no pkey in EPT page table for L1 guest or EPT
6161 * shadow page table for L2 guest.
6163 if (vcpu_match_mmio_gva(vcpu, gva)
6164 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6165 vcpu->arch.mmio_access, 0, access)) {
6166 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6167 (gva & (PAGE_SIZE - 1));
6168 trace_vcpu_match_mmio(gva, *gpa, write, false);
6172 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6174 if (*gpa == UNMAPPED_GVA)
6177 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6180 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6181 const void *val, int bytes)
6185 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6188 kvm_page_track_write(vcpu, gpa, val, bytes);
6192 struct read_write_emulator_ops {
6193 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6195 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6196 void *val, int bytes);
6197 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6198 int bytes, void *val);
6199 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6200 void *val, int bytes);
6204 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6206 if (vcpu->mmio_read_completed) {
6207 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6208 vcpu->mmio_fragments[0].gpa, val);
6209 vcpu->mmio_read_completed = 0;
6216 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6217 void *val, int bytes)
6219 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6222 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6223 void *val, int bytes)
6225 return emulator_write_phys(vcpu, gpa, val, bytes);
6228 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6230 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6231 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6234 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6235 void *val, int bytes)
6237 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6238 return X86EMUL_IO_NEEDED;
6241 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6242 void *val, int bytes)
6244 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6246 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6247 return X86EMUL_CONTINUE;
6250 static const struct read_write_emulator_ops read_emultor = {
6251 .read_write_prepare = read_prepare,
6252 .read_write_emulate = read_emulate,
6253 .read_write_mmio = vcpu_mmio_read,
6254 .read_write_exit_mmio = read_exit_mmio,
6257 static const struct read_write_emulator_ops write_emultor = {
6258 .read_write_emulate = write_emulate,
6259 .read_write_mmio = write_mmio,
6260 .read_write_exit_mmio = write_exit_mmio,
6264 static int emulator_read_write_onepage(unsigned long addr, void *val,
6266 struct x86_exception *exception,
6267 struct kvm_vcpu *vcpu,
6268 const struct read_write_emulator_ops *ops)
6272 bool write = ops->write;
6273 struct kvm_mmio_fragment *frag;
6274 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6277 * If the exit was due to a NPF we may already have a GPA.
6278 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6279 * Note, this cannot be used on string operations since string
6280 * operation using rep will only have the initial GPA from the NPF
6283 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6284 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6285 gpa = ctxt->gpa_val;
6286 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6288 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6290 return X86EMUL_PROPAGATE_FAULT;
6293 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6294 return X86EMUL_CONTINUE;
6297 * Is this MMIO handled locally?
6299 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6300 if (handled == bytes)
6301 return X86EMUL_CONTINUE;
6307 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6308 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6312 return X86EMUL_CONTINUE;
6315 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6317 void *val, unsigned int bytes,
6318 struct x86_exception *exception,
6319 const struct read_write_emulator_ops *ops)
6321 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6325 if (ops->read_write_prepare &&
6326 ops->read_write_prepare(vcpu, val, bytes))
6327 return X86EMUL_CONTINUE;
6329 vcpu->mmio_nr_fragments = 0;
6331 /* Crossing a page boundary? */
6332 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6335 now = -addr & ~PAGE_MASK;
6336 rc = emulator_read_write_onepage(addr, val, now, exception,
6339 if (rc != X86EMUL_CONTINUE)
6342 if (ctxt->mode != X86EMUL_MODE_PROT64)
6348 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6350 if (rc != X86EMUL_CONTINUE)
6353 if (!vcpu->mmio_nr_fragments)
6356 gpa = vcpu->mmio_fragments[0].gpa;
6358 vcpu->mmio_needed = 1;
6359 vcpu->mmio_cur_fragment = 0;
6361 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6362 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6363 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6364 vcpu->run->mmio.phys_addr = gpa;
6366 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6369 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6373 struct x86_exception *exception)
6375 return emulator_read_write(ctxt, addr, val, bytes,
6376 exception, &read_emultor);
6379 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6383 struct x86_exception *exception)
6385 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6386 exception, &write_emultor);
6389 #define CMPXCHG_TYPE(t, ptr, old, new) \
6390 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6392 #ifdef CONFIG_X86_64
6393 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6395 # define CMPXCHG64(ptr, old, new) \
6396 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6399 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6404 struct x86_exception *exception)
6406 struct kvm_host_map map;
6407 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6413 /* guests cmpxchg8b have to be emulated atomically */
6414 if (bytes > 8 || (bytes & (bytes - 1)))
6417 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6419 if (gpa == UNMAPPED_GVA ||
6420 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6424 * Emulate the atomic as a straight write to avoid #AC if SLD is
6425 * enabled in the host and the access splits a cache line.
6427 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6428 page_line_mask = ~(cache_line_size() - 1);
6430 page_line_mask = PAGE_MASK;
6432 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6435 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6438 kaddr = map.hva + offset_in_page(gpa);
6442 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6445 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6448 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6451 exchanged = CMPXCHG64(kaddr, old, new);
6457 kvm_vcpu_unmap(vcpu, &map, true);
6460 return X86EMUL_CMPXCHG_FAILED;
6462 kvm_page_track_write(vcpu, gpa, new, bytes);
6464 return X86EMUL_CONTINUE;
6467 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6469 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6472 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6476 for (i = 0; i < vcpu->arch.pio.count; i++) {
6477 if (vcpu->arch.pio.in)
6478 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6479 vcpu->arch.pio.size, pd);
6481 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6482 vcpu->arch.pio.port, vcpu->arch.pio.size,
6486 pd += vcpu->arch.pio.size;
6491 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6492 unsigned short port, void *val,
6493 unsigned int count, bool in)
6495 vcpu->arch.pio.port = port;
6496 vcpu->arch.pio.in = in;
6497 vcpu->arch.pio.count = count;
6498 vcpu->arch.pio.size = size;
6500 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6501 vcpu->arch.pio.count = 0;
6505 vcpu->run->exit_reason = KVM_EXIT_IO;
6506 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6507 vcpu->run->io.size = size;
6508 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6509 vcpu->run->io.count = count;
6510 vcpu->run->io.port = port;
6515 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6516 unsigned short port, void *val, unsigned int count)
6520 if (vcpu->arch.pio.count)
6523 memset(vcpu->arch.pio_data, 0, size * count);
6525 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6528 memcpy(val, vcpu->arch.pio_data, size * count);
6529 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6530 vcpu->arch.pio.count = 0;
6537 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6538 int size, unsigned short port, void *val,
6541 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6545 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6546 unsigned short port, const void *val,
6549 memcpy(vcpu->arch.pio_data, val, size * count);
6550 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6551 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6554 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6555 int size, unsigned short port,
6556 const void *val, unsigned int count)
6558 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6561 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6563 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6566 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6568 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6571 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6573 if (!need_emulate_wbinvd(vcpu))
6574 return X86EMUL_CONTINUE;
6576 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6577 int cpu = get_cpu();
6579 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6580 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6581 wbinvd_ipi, NULL, 1);
6583 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6586 return X86EMUL_CONTINUE;
6589 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6591 kvm_emulate_wbinvd_noskip(vcpu);
6592 return kvm_skip_emulated_instruction(vcpu);
6594 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6598 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6600 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6603 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6604 unsigned long *dest)
6606 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6609 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6610 unsigned long value)
6613 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6616 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6618 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6621 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6623 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6624 unsigned long value;
6628 value = kvm_read_cr0(vcpu);
6631 value = vcpu->arch.cr2;
6634 value = kvm_read_cr3(vcpu);
6637 value = kvm_read_cr4(vcpu);
6640 value = kvm_get_cr8(vcpu);
6643 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6650 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6652 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6657 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6660 vcpu->arch.cr2 = val;
6663 res = kvm_set_cr3(vcpu, val);
6666 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6669 res = kvm_set_cr8(vcpu, val);
6672 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6679 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6681 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
6684 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6686 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
6689 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6691 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
6694 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6696 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
6699 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6701 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
6704 static unsigned long emulator_get_cached_segment_base(
6705 struct x86_emulate_ctxt *ctxt, int seg)
6707 return get_segment_base(emul_to_vcpu(ctxt), seg);
6710 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6711 struct desc_struct *desc, u32 *base3,
6714 struct kvm_segment var;
6716 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6717 *selector = var.selector;
6720 memset(desc, 0, sizeof(*desc));
6728 set_desc_limit(desc, var.limit);
6729 set_desc_base(desc, (unsigned long)var.base);
6730 #ifdef CONFIG_X86_64
6732 *base3 = var.base >> 32;
6734 desc->type = var.type;
6736 desc->dpl = var.dpl;
6737 desc->p = var.present;
6738 desc->avl = var.avl;
6746 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6747 struct desc_struct *desc, u32 base3,
6750 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6751 struct kvm_segment var;
6753 var.selector = selector;
6754 var.base = get_desc_base(desc);
6755 #ifdef CONFIG_X86_64
6756 var.base |= ((u64)base3) << 32;
6758 var.limit = get_desc_limit(desc);
6760 var.limit = (var.limit << 12) | 0xfff;
6761 var.type = desc->type;
6762 var.dpl = desc->dpl;
6767 var.avl = desc->avl;
6768 var.present = desc->p;
6769 var.unusable = !var.present;
6772 kvm_set_segment(vcpu, &var, seg);
6776 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6777 u32 msr_index, u64 *pdata)
6779 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6782 r = kvm_get_msr(vcpu, msr_index, pdata);
6784 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6785 /* Bounce to user space */
6786 return X86EMUL_IO_NEEDED;
6792 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6793 u32 msr_index, u64 data)
6795 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6798 r = kvm_set_msr(vcpu, msr_index, data);
6800 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6801 /* Bounce to user space */
6802 return X86EMUL_IO_NEEDED;
6808 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6810 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6812 return vcpu->arch.smbase;
6815 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6817 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6819 vcpu->arch.smbase = smbase;
6822 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6825 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6828 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6829 u32 pmc, u64 *pdata)
6831 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6834 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6836 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6839 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6840 struct x86_instruction_info *info,
6841 enum x86_intercept_stage stage)
6843 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
6847 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6848 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6851 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6854 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6856 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6859 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6861 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6864 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6866 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6869 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6871 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6874 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6876 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6879 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6881 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
6884 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6886 return emul_to_vcpu(ctxt)->arch.hflags;
6889 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6891 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6894 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6895 const char *smstate)
6897 return static_call(kvm_x86_pre_leave_smm)(emul_to_vcpu(ctxt), smstate);
6900 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6902 kvm_smm_changed(emul_to_vcpu(ctxt));
6905 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6907 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6910 static const struct x86_emulate_ops emulate_ops = {
6911 .read_gpr = emulator_read_gpr,
6912 .write_gpr = emulator_write_gpr,
6913 .read_std = emulator_read_std,
6914 .write_std = emulator_write_std,
6915 .read_phys = kvm_read_guest_phys_system,
6916 .fetch = kvm_fetch_guest_virt,
6917 .read_emulated = emulator_read_emulated,
6918 .write_emulated = emulator_write_emulated,
6919 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6920 .invlpg = emulator_invlpg,
6921 .pio_in_emulated = emulator_pio_in_emulated,
6922 .pio_out_emulated = emulator_pio_out_emulated,
6923 .get_segment = emulator_get_segment,
6924 .set_segment = emulator_set_segment,
6925 .get_cached_segment_base = emulator_get_cached_segment_base,
6926 .get_gdt = emulator_get_gdt,
6927 .get_idt = emulator_get_idt,
6928 .set_gdt = emulator_set_gdt,
6929 .set_idt = emulator_set_idt,
6930 .get_cr = emulator_get_cr,
6931 .set_cr = emulator_set_cr,
6932 .cpl = emulator_get_cpl,
6933 .get_dr = emulator_get_dr,
6934 .set_dr = emulator_set_dr,
6935 .get_smbase = emulator_get_smbase,
6936 .set_smbase = emulator_set_smbase,
6937 .set_msr = emulator_set_msr,
6938 .get_msr = emulator_get_msr,
6939 .check_pmc = emulator_check_pmc,
6940 .read_pmc = emulator_read_pmc,
6941 .halt = emulator_halt,
6942 .wbinvd = emulator_wbinvd,
6943 .fix_hypercall = emulator_fix_hypercall,
6944 .intercept = emulator_intercept,
6945 .get_cpuid = emulator_get_cpuid,
6946 .guest_has_long_mode = emulator_guest_has_long_mode,
6947 .guest_has_movbe = emulator_guest_has_movbe,
6948 .guest_has_fxsr = emulator_guest_has_fxsr,
6949 .set_nmi_mask = emulator_set_nmi_mask,
6950 .get_hflags = emulator_get_hflags,
6951 .set_hflags = emulator_set_hflags,
6952 .pre_leave_smm = emulator_pre_leave_smm,
6953 .post_leave_smm = emulator_post_leave_smm,
6954 .set_xcr = emulator_set_xcr,
6957 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6959 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
6961 * an sti; sti; sequence only disable interrupts for the first
6962 * instruction. So, if the last instruction, be it emulated or
6963 * not, left the system with the INT_STI flag enabled, it
6964 * means that the last instruction is an sti. We should not
6965 * leave the flag on in this case. The same goes for mov ss
6967 if (int_shadow & mask)
6969 if (unlikely(int_shadow || mask)) {
6970 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
6972 kvm_make_request(KVM_REQ_EVENT, vcpu);
6976 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6978 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6979 if (ctxt->exception.vector == PF_VECTOR)
6980 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
6982 if (ctxt->exception.error_code_valid)
6983 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6984 ctxt->exception.error_code);
6986 kvm_queue_exception(vcpu, ctxt->exception.vector);
6990 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6992 struct x86_emulate_ctxt *ctxt;
6994 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6996 pr_err("kvm: failed to allocate vcpu's emulator\n");
7001 ctxt->ops = &emulate_ops;
7002 vcpu->arch.emulate_ctxt = ctxt;
7007 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7009 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7012 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7014 ctxt->gpa_available = false;
7015 ctxt->eflags = kvm_get_rflags(vcpu);
7016 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7018 ctxt->eip = kvm_rip_read(vcpu);
7019 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7020 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7021 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7022 cs_db ? X86EMUL_MODE_PROT32 :
7023 X86EMUL_MODE_PROT16;
7024 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7025 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7026 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7028 init_decode_cache(ctxt);
7029 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7032 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7034 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7037 init_emulate_ctxt(vcpu);
7041 ctxt->_eip = ctxt->eip + inc_eip;
7042 ret = emulate_int_real(ctxt, irq);
7044 if (ret != X86EMUL_CONTINUE) {
7045 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7047 ctxt->eip = ctxt->_eip;
7048 kvm_rip_write(vcpu, ctxt->eip);
7049 kvm_set_rflags(vcpu, ctxt->eflags);
7052 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7054 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7056 ++vcpu->stat.insn_emulation_fail;
7057 trace_kvm_emulate_insn_failed(vcpu);
7059 if (emulation_type & EMULTYPE_VMWARE_GP) {
7060 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7064 if (emulation_type & EMULTYPE_SKIP) {
7065 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7066 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7067 vcpu->run->internal.ndata = 0;
7071 kvm_queue_exception(vcpu, UD_VECTOR);
7073 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7074 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7075 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7076 vcpu->run->internal.ndata = 0;
7083 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7084 bool write_fault_to_shadow_pgtable,
7087 gpa_t gpa = cr2_or_gpa;
7090 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7093 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7094 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7097 if (!vcpu->arch.mmu->direct_map) {
7099 * Write permission should be allowed since only
7100 * write access need to be emulated.
7102 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7105 * If the mapping is invalid in guest, let cpu retry
7106 * it to generate fault.
7108 if (gpa == UNMAPPED_GVA)
7113 * Do not retry the unhandleable instruction if it faults on the
7114 * readonly host memory, otherwise it will goto a infinite loop:
7115 * retry instruction -> write #PF -> emulation fail -> retry
7116 * instruction -> ...
7118 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7121 * If the instruction failed on the error pfn, it can not be fixed,
7122 * report the error to userspace.
7124 if (is_error_noslot_pfn(pfn))
7127 kvm_release_pfn_clean(pfn);
7129 /* The instructions are well-emulated on direct mmu. */
7130 if (vcpu->arch.mmu->direct_map) {
7131 unsigned int indirect_shadow_pages;
7133 write_lock(&vcpu->kvm->mmu_lock);
7134 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7135 write_unlock(&vcpu->kvm->mmu_lock);
7137 if (indirect_shadow_pages)
7138 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7144 * if emulation was due to access to shadowed page table
7145 * and it failed try to unshadow page and re-enter the
7146 * guest to let CPU execute the instruction.
7148 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7151 * If the access faults on its page table, it can not
7152 * be fixed by unprotecting shadow page and it should
7153 * be reported to userspace.
7155 return !write_fault_to_shadow_pgtable;
7158 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7159 gpa_t cr2_or_gpa, int emulation_type)
7161 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7162 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7164 last_retry_eip = vcpu->arch.last_retry_eip;
7165 last_retry_addr = vcpu->arch.last_retry_addr;
7168 * If the emulation is caused by #PF and it is non-page_table
7169 * writing instruction, it means the VM-EXIT is caused by shadow
7170 * page protected, we can zap the shadow page and retry this
7171 * instruction directly.
7173 * Note: if the guest uses a non-page-table modifying instruction
7174 * on the PDE that points to the instruction, then we will unmap
7175 * the instruction and go to an infinite loop. So, we cache the
7176 * last retried eip and the last fault address, if we meet the eip
7177 * and the address again, we can break out of the potential infinite
7180 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7182 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7185 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7186 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7189 if (x86_page_table_writing_insn(ctxt))
7192 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7195 vcpu->arch.last_retry_eip = ctxt->eip;
7196 vcpu->arch.last_retry_addr = cr2_or_gpa;
7198 if (!vcpu->arch.mmu->direct_map)
7199 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7201 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7206 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7207 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7209 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7211 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7212 /* This is a good place to trace that we are exiting SMM. */
7213 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7215 /* Process a latched INIT or SMI, if any. */
7216 kvm_make_request(KVM_REQ_EVENT, vcpu);
7219 kvm_mmu_reset_context(vcpu);
7222 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7231 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7232 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7237 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7239 struct kvm_run *kvm_run = vcpu->run;
7241 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7242 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7243 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7244 kvm_run->debug.arch.exception = DB_VECTOR;
7245 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7248 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7252 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7254 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7257 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7262 * rflags is the old, "raw" value of the flags. The new value has
7263 * not been saved yet.
7265 * This is correct even for TF set by the guest, because "the
7266 * processor will not generate this exception after the instruction
7267 * that sets the TF flag".
7269 if (unlikely(rflags & X86_EFLAGS_TF))
7270 r = kvm_vcpu_do_singlestep(vcpu);
7273 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7275 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7277 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7278 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7279 struct kvm_run *kvm_run = vcpu->run;
7280 unsigned long eip = kvm_get_linear_rip(vcpu);
7281 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7282 vcpu->arch.guest_debug_dr7,
7286 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7287 kvm_run->debug.arch.pc = eip;
7288 kvm_run->debug.arch.exception = DB_VECTOR;
7289 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7295 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7296 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7297 unsigned long eip = kvm_get_linear_rip(vcpu);
7298 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7303 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7312 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7314 switch (ctxt->opcode_len) {
7321 case 0xe6: /* OUT */
7325 case 0x6c: /* INS */
7327 case 0x6e: /* OUTS */
7334 case 0x33: /* RDPMC */
7344 * Decode to be emulated instruction. Return EMULATION_OK if success.
7346 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7347 void *insn, int insn_len)
7349 int r = EMULATION_OK;
7350 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7352 init_emulate_ctxt(vcpu);
7355 * We will reenter on the same instruction since we do not set
7356 * complete_userspace_io. This does not handle watchpoints yet,
7357 * those would be handled in the emulate_ops.
7359 if (!(emulation_type & EMULTYPE_SKIP) &&
7360 kvm_vcpu_check_breakpoint(vcpu, &r))
7363 ctxt->interruptibility = 0;
7364 ctxt->have_exception = false;
7365 ctxt->exception.vector = -1;
7366 ctxt->perm_ok = false;
7368 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7370 r = x86_decode_insn(ctxt, insn, insn_len);
7372 trace_kvm_emulate_insn_start(vcpu);
7373 ++vcpu->stat.insn_emulation;
7377 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7379 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7380 int emulation_type, void *insn, int insn_len)
7383 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7384 bool writeback = true;
7385 bool write_fault_to_spt;
7387 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7390 vcpu->arch.l1tf_flush_l1d = true;
7393 * Clear write_fault_to_shadow_pgtable here to ensure it is
7396 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7397 vcpu->arch.write_fault_to_shadow_pgtable = false;
7399 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7400 kvm_clear_exception_queue(vcpu);
7402 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7404 if (r != EMULATION_OK) {
7405 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7406 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7407 kvm_queue_exception(vcpu, UD_VECTOR);
7410 if (reexecute_instruction(vcpu, cr2_or_gpa,
7414 if (ctxt->have_exception) {
7416 * #UD should result in just EMULATION_FAILED, and trap-like
7417 * exception should not be encountered during decode.
7419 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7420 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7421 inject_emulated_exception(vcpu);
7424 return handle_emulation_failure(vcpu, emulation_type);
7428 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7429 !is_vmware_backdoor_opcode(ctxt)) {
7430 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7435 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7436 * for kvm_skip_emulated_instruction(). The caller is responsible for
7437 * updating interruptibility state and injecting single-step #DBs.
7439 if (emulation_type & EMULTYPE_SKIP) {
7440 kvm_rip_write(vcpu, ctxt->_eip);
7441 if (ctxt->eflags & X86_EFLAGS_RF)
7442 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7446 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7449 /* this is needed for vmware backdoor interface to work since it
7450 changes registers values during IO operation */
7451 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7452 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7453 emulator_invalidate_register_cache(ctxt);
7457 if (emulation_type & EMULTYPE_PF) {
7458 /* Save the faulting GPA (cr2) in the address field */
7459 ctxt->exception.address = cr2_or_gpa;
7461 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7462 if (vcpu->arch.mmu->direct_map) {
7463 ctxt->gpa_available = true;
7464 ctxt->gpa_val = cr2_or_gpa;
7467 /* Sanitize the address out of an abundance of paranoia. */
7468 ctxt->exception.address = 0;
7471 r = x86_emulate_insn(ctxt);
7473 if (r == EMULATION_INTERCEPTED)
7476 if (r == EMULATION_FAILED) {
7477 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7481 return handle_emulation_failure(vcpu, emulation_type);
7484 if (ctxt->have_exception) {
7486 if (inject_emulated_exception(vcpu))
7488 } else if (vcpu->arch.pio.count) {
7489 if (!vcpu->arch.pio.in) {
7490 /* FIXME: return into emulator if single-stepping. */
7491 vcpu->arch.pio.count = 0;
7494 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7497 } else if (vcpu->mmio_needed) {
7498 ++vcpu->stat.mmio_exits;
7500 if (!vcpu->mmio_is_write)
7503 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7504 } else if (r == EMULATION_RESTART)
7510 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7511 toggle_interruptibility(vcpu, ctxt->interruptibility);
7512 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7513 if (!ctxt->have_exception ||
7514 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7515 kvm_rip_write(vcpu, ctxt->eip);
7516 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7517 r = kvm_vcpu_do_singlestep(vcpu);
7518 if (kvm_x86_ops.update_emulated_instruction)
7519 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7520 __kvm_set_rflags(vcpu, ctxt->eflags);
7524 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7525 * do nothing, and it will be requested again as soon as
7526 * the shadow expires. But we still need to check here,
7527 * because POPF has no interrupt shadow.
7529 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7530 kvm_make_request(KVM_REQ_EVENT, vcpu);
7532 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7537 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7539 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7541 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7543 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7544 void *insn, int insn_len)
7546 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7548 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7550 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7552 vcpu->arch.pio.count = 0;
7556 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7558 vcpu->arch.pio.count = 0;
7560 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7563 return kvm_skip_emulated_instruction(vcpu);
7566 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7567 unsigned short port)
7569 unsigned long val = kvm_rax_read(vcpu);
7570 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7576 * Workaround userspace that relies on old KVM behavior of %rip being
7577 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7580 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7581 vcpu->arch.complete_userspace_io =
7582 complete_fast_pio_out_port_0x7e;
7583 kvm_skip_emulated_instruction(vcpu);
7585 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7586 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7591 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7595 /* We should only ever be called with arch.pio.count equal to 1 */
7596 BUG_ON(vcpu->arch.pio.count != 1);
7598 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7599 vcpu->arch.pio.count = 0;
7603 /* For size less than 4 we merge, else we zero extend */
7604 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7607 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7608 * the copy and tracing
7610 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7611 kvm_rax_write(vcpu, val);
7613 return kvm_skip_emulated_instruction(vcpu);
7616 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7617 unsigned short port)
7622 /* For size less than 4 we merge, else we zero extend */
7623 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7625 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7627 kvm_rax_write(vcpu, val);
7631 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7632 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7637 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7642 ret = kvm_fast_pio_in(vcpu, size, port);
7644 ret = kvm_fast_pio_out(vcpu, size, port);
7645 return ret && kvm_skip_emulated_instruction(vcpu);
7647 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7649 static int kvmclock_cpu_down_prep(unsigned int cpu)
7651 __this_cpu_write(cpu_tsc_khz, 0);
7655 static void tsc_khz_changed(void *data)
7657 struct cpufreq_freqs *freq = data;
7658 unsigned long khz = 0;
7662 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7663 khz = cpufreq_quick_get(raw_smp_processor_id());
7666 __this_cpu_write(cpu_tsc_khz, khz);
7669 #ifdef CONFIG_X86_64
7670 static void kvm_hyperv_tsc_notifier(void)
7673 struct kvm_vcpu *vcpu;
7676 mutex_lock(&kvm_lock);
7677 list_for_each_entry(kvm, &vm_list, vm_list)
7678 kvm_make_mclock_inprogress_request(kvm);
7680 hyperv_stop_tsc_emulation();
7682 /* TSC frequency always matches when on Hyper-V */
7683 for_each_present_cpu(cpu)
7684 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7685 kvm_max_guest_tsc_khz = tsc_khz;
7687 list_for_each_entry(kvm, &vm_list, vm_list) {
7688 struct kvm_arch *ka = &kvm->arch;
7690 spin_lock(&ka->pvclock_gtod_sync_lock);
7692 pvclock_update_vm_gtod_copy(kvm);
7694 kvm_for_each_vcpu(cpu, vcpu, kvm)
7695 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7697 kvm_for_each_vcpu(cpu, vcpu, kvm)
7698 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7700 spin_unlock(&ka->pvclock_gtod_sync_lock);
7702 mutex_unlock(&kvm_lock);
7706 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7709 struct kvm_vcpu *vcpu;
7710 int i, send_ipi = 0;
7713 * We allow guests to temporarily run on slowing clocks,
7714 * provided we notify them after, or to run on accelerating
7715 * clocks, provided we notify them before. Thus time never
7718 * However, we have a problem. We can't atomically update
7719 * the frequency of a given CPU from this function; it is
7720 * merely a notifier, which can be called from any CPU.
7721 * Changing the TSC frequency at arbitrary points in time
7722 * requires a recomputation of local variables related to
7723 * the TSC for each VCPU. We must flag these local variables
7724 * to be updated and be sure the update takes place with the
7725 * new frequency before any guests proceed.
7727 * Unfortunately, the combination of hotplug CPU and frequency
7728 * change creates an intractable locking scenario; the order
7729 * of when these callouts happen is undefined with respect to
7730 * CPU hotplug, and they can race with each other. As such,
7731 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7732 * undefined; you can actually have a CPU frequency change take
7733 * place in between the computation of X and the setting of the
7734 * variable. To protect against this problem, all updates of
7735 * the per_cpu tsc_khz variable are done in an interrupt
7736 * protected IPI, and all callers wishing to update the value
7737 * must wait for a synchronous IPI to complete (which is trivial
7738 * if the caller is on the CPU already). This establishes the
7739 * necessary total order on variable updates.
7741 * Note that because a guest time update may take place
7742 * anytime after the setting of the VCPU's request bit, the
7743 * correct TSC value must be set before the request. However,
7744 * to ensure the update actually makes it to any guest which
7745 * starts running in hardware virtualization between the set
7746 * and the acquisition of the spinlock, we must also ping the
7747 * CPU after setting the request bit.
7751 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7753 mutex_lock(&kvm_lock);
7754 list_for_each_entry(kvm, &vm_list, vm_list) {
7755 kvm_for_each_vcpu(i, vcpu, kvm) {
7756 if (vcpu->cpu != cpu)
7758 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7759 if (vcpu->cpu != raw_smp_processor_id())
7763 mutex_unlock(&kvm_lock);
7765 if (freq->old < freq->new && send_ipi) {
7767 * We upscale the frequency. Must make the guest
7768 * doesn't see old kvmclock values while running with
7769 * the new frequency, otherwise we risk the guest sees
7770 * time go backwards.
7772 * In case we update the frequency for another cpu
7773 * (which might be in guest context) send an interrupt
7774 * to kick the cpu out of guest context. Next time
7775 * guest context is entered kvmclock will be updated,
7776 * so the guest will not see stale values.
7778 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7782 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7785 struct cpufreq_freqs *freq = data;
7788 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7790 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7793 for_each_cpu(cpu, freq->policy->cpus)
7794 __kvmclock_cpufreq_notifier(freq, cpu);
7799 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7800 .notifier_call = kvmclock_cpufreq_notifier
7803 static int kvmclock_cpu_online(unsigned int cpu)
7805 tsc_khz_changed(NULL);
7809 static void kvm_timer_init(void)
7811 max_tsc_khz = tsc_khz;
7813 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7814 #ifdef CONFIG_CPU_FREQ
7815 struct cpufreq_policy *policy;
7819 policy = cpufreq_cpu_get(cpu);
7821 if (policy->cpuinfo.max_freq)
7822 max_tsc_khz = policy->cpuinfo.max_freq;
7823 cpufreq_cpu_put(policy);
7827 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7828 CPUFREQ_TRANSITION_NOTIFIER);
7831 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7832 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7835 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7836 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7838 int kvm_is_in_guest(void)
7840 return __this_cpu_read(current_vcpu) != NULL;
7843 static int kvm_is_user_mode(void)
7847 if (__this_cpu_read(current_vcpu))
7848 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
7850 return user_mode != 0;
7853 static unsigned long kvm_get_guest_ip(void)
7855 unsigned long ip = 0;
7857 if (__this_cpu_read(current_vcpu))
7858 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7863 static void kvm_handle_intel_pt_intr(void)
7865 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7867 kvm_make_request(KVM_REQ_PMI, vcpu);
7868 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7869 (unsigned long *)&vcpu->arch.pmu.global_status);
7872 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7873 .is_in_guest = kvm_is_in_guest,
7874 .is_user_mode = kvm_is_user_mode,
7875 .get_guest_ip = kvm_get_guest_ip,
7876 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7879 #ifdef CONFIG_X86_64
7880 static void pvclock_gtod_update_fn(struct work_struct *work)
7884 struct kvm_vcpu *vcpu;
7887 mutex_lock(&kvm_lock);
7888 list_for_each_entry(kvm, &vm_list, vm_list)
7889 kvm_for_each_vcpu(i, vcpu, kvm)
7890 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7891 atomic_set(&kvm_guest_has_master_clock, 0);
7892 mutex_unlock(&kvm_lock);
7895 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7898 * Notification about pvclock gtod data update.
7900 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7903 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7904 struct timekeeper *tk = priv;
7906 update_pvclock_gtod(tk);
7908 /* disable master clock if host does not trust, or does not
7909 * use, TSC based clocksource.
7911 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7912 atomic_read(&kvm_guest_has_master_clock) != 0)
7913 queue_work(system_long_wq, &pvclock_gtod_work);
7918 static struct notifier_block pvclock_gtod_notifier = {
7919 .notifier_call = pvclock_gtod_notify,
7923 int kvm_arch_init(void *opaque)
7925 struct kvm_x86_init_ops *ops = opaque;
7928 if (kvm_x86_ops.hardware_enable) {
7929 printk(KERN_ERR "kvm: already loaded the other module\n");
7934 if (!ops->cpu_has_kvm_support()) {
7935 pr_err_ratelimited("kvm: no hardware support\n");
7939 if (ops->disabled_by_bios()) {
7940 pr_err_ratelimited("kvm: disabled by bios\n");
7946 * KVM explicitly assumes that the guest has an FPU and
7947 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7948 * vCPU's FPU state as a fxregs_state struct.
7950 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7951 printk(KERN_ERR "kvm: inadequate fpu\n");
7957 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7958 __alignof__(struct fpu), SLAB_ACCOUNT,
7960 if (!x86_fpu_cache) {
7961 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7965 x86_emulator_cache = kvm_alloc_emulator_cache();
7966 if (!x86_emulator_cache) {
7967 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7968 goto out_free_x86_fpu_cache;
7971 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
7972 if (!user_return_msrs) {
7973 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
7974 goto out_free_x86_emulator_cache;
7977 r = kvm_mmu_module_init();
7979 goto out_free_percpu;
7981 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7982 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7983 PT_PRESENT_MASK, 0, sme_me_mask);
7986 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7988 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7989 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7990 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7993 if (pi_inject_timer == -1)
7994 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7995 #ifdef CONFIG_X86_64
7996 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7998 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7999 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8005 free_percpu(user_return_msrs);
8006 out_free_x86_emulator_cache:
8007 kmem_cache_destroy(x86_emulator_cache);
8008 out_free_x86_fpu_cache:
8009 kmem_cache_destroy(x86_fpu_cache);
8014 void kvm_arch_exit(void)
8016 #ifdef CONFIG_X86_64
8017 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8018 clear_hv_tscchange_cb();
8021 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8023 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8024 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8025 CPUFREQ_TRANSITION_NOTIFIER);
8026 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8027 #ifdef CONFIG_X86_64
8028 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8030 kvm_x86_ops.hardware_enable = NULL;
8031 kvm_mmu_module_exit();
8032 free_percpu(user_return_msrs);
8033 kmem_cache_destroy(x86_fpu_cache);
8034 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8037 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8039 ++vcpu->stat.halt_exits;
8040 if (lapic_in_kernel(vcpu)) {
8041 vcpu->arch.mp_state = state;
8044 vcpu->run->exit_reason = reason;
8049 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8051 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8053 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8055 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8057 int ret = kvm_skip_emulated_instruction(vcpu);
8059 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8060 * KVM_EXIT_DEBUG here.
8062 return kvm_vcpu_halt(vcpu) && ret;
8064 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8066 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8068 int ret = kvm_skip_emulated_instruction(vcpu);
8070 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8072 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8074 #ifdef CONFIG_X86_64
8075 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8076 unsigned long clock_type)
8078 struct kvm_clock_pairing clock_pairing;
8079 struct timespec64 ts;
8083 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8084 return -KVM_EOPNOTSUPP;
8086 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8087 return -KVM_EOPNOTSUPP;
8089 clock_pairing.sec = ts.tv_sec;
8090 clock_pairing.nsec = ts.tv_nsec;
8091 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8092 clock_pairing.flags = 0;
8093 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8096 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8097 sizeof(struct kvm_clock_pairing)))
8105 * kvm_pv_kick_cpu_op: Kick a vcpu.
8107 * @apicid - apicid of vcpu to be kicked.
8109 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8111 struct kvm_lapic_irq lapic_irq;
8113 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8114 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8115 lapic_irq.level = 0;
8116 lapic_irq.dest_id = apicid;
8117 lapic_irq.msi_redir_hint = false;
8119 lapic_irq.delivery_mode = APIC_DM_REMRD;
8120 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8123 bool kvm_apicv_activated(struct kvm *kvm)
8125 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8127 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8129 void kvm_apicv_init(struct kvm *kvm, bool enable)
8132 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8133 &kvm->arch.apicv_inhibit_reasons);
8135 set_bit(APICV_INHIBIT_REASON_DISABLE,
8136 &kvm->arch.apicv_inhibit_reasons);
8138 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8140 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
8142 struct kvm_vcpu *target = NULL;
8143 struct kvm_apic_map *map;
8146 map = rcu_dereference(kvm->arch.apic_map);
8148 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8149 target = map->phys_map[dest_id]->vcpu;
8153 if (target && READ_ONCE(target->ready))
8154 kvm_vcpu_yield_to(target);
8157 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8159 unsigned long nr, a0, a1, a2, a3, ret;
8162 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8163 return kvm_xen_hypercall(vcpu);
8165 if (kvm_hv_hypercall_enabled(vcpu->kvm))
8166 return kvm_hv_hypercall(vcpu);
8168 nr = kvm_rax_read(vcpu);
8169 a0 = kvm_rbx_read(vcpu);
8170 a1 = kvm_rcx_read(vcpu);
8171 a2 = kvm_rdx_read(vcpu);
8172 a3 = kvm_rsi_read(vcpu);
8174 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8176 op_64_bit = is_64_bit_mode(vcpu);
8185 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8193 case KVM_HC_VAPIC_POLL_IRQ:
8196 case KVM_HC_KICK_CPU:
8197 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8200 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8201 kvm_sched_yield(vcpu->kvm, a1);
8204 #ifdef CONFIG_X86_64
8205 case KVM_HC_CLOCK_PAIRING:
8206 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8209 case KVM_HC_SEND_IPI:
8210 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8213 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8215 case KVM_HC_SCHED_YIELD:
8216 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8219 kvm_sched_yield(vcpu->kvm, a0);
8229 kvm_rax_write(vcpu, ret);
8231 ++vcpu->stat.hypercalls;
8232 return kvm_skip_emulated_instruction(vcpu);
8234 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8236 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8238 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8239 char instruction[3];
8240 unsigned long rip = kvm_rip_read(vcpu);
8242 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8244 return emulator_write_emulated(ctxt, rip, instruction, 3,
8248 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8250 return vcpu->run->request_interrupt_window &&
8251 likely(!pic_in_kernel(vcpu->kvm));
8254 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8256 struct kvm_run *kvm_run = vcpu->run;
8259 * if_flag is obsolete and useless, so do not bother
8260 * setting it for SEV-ES guests. Userspace can just
8261 * use kvm_run->ready_for_interrupt_injection.
8263 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8264 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8266 kvm_run->cr8 = kvm_get_cr8(vcpu);
8267 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8268 kvm_run->ready_for_interrupt_injection =
8269 pic_in_kernel(vcpu->kvm) ||
8270 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8273 kvm_run->flags |= KVM_RUN_X86_SMM;
8276 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8280 if (!kvm_x86_ops.update_cr8_intercept)
8283 if (!lapic_in_kernel(vcpu))
8286 if (vcpu->arch.apicv_active)
8289 if (!vcpu->arch.apic->vapic_addr)
8290 max_irr = kvm_lapic_find_highest_irr(vcpu);
8297 tpr = kvm_lapic_get_cr8(vcpu);
8299 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8302 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8305 bool can_inject = true;
8307 /* try to reinject previous events if any */
8309 if (vcpu->arch.exception.injected) {
8310 static_call(kvm_x86_queue_exception)(vcpu);
8314 * Do not inject an NMI or interrupt if there is a pending
8315 * exception. Exceptions and interrupts are recognized at
8316 * instruction boundaries, i.e. the start of an instruction.
8317 * Trap-like exceptions, e.g. #DB, have higher priority than
8318 * NMIs and interrupts, i.e. traps are recognized before an
8319 * NMI/interrupt that's pending on the same instruction.
8320 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8321 * priority, but are only generated (pended) during instruction
8322 * execution, i.e. a pending fault-like exception means the
8323 * fault occurred on the *previous* instruction and must be
8324 * serviced prior to recognizing any new events in order to
8325 * fully complete the previous instruction.
8327 else if (!vcpu->arch.exception.pending) {
8328 if (vcpu->arch.nmi_injected) {
8329 static_call(kvm_x86_set_nmi)(vcpu);
8331 } else if (vcpu->arch.interrupt.injected) {
8332 static_call(kvm_x86_set_irq)(vcpu);
8337 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8338 vcpu->arch.exception.pending);
8341 * Call check_nested_events() even if we reinjected a previous event
8342 * in order for caller to determine if it should require immediate-exit
8343 * from L2 to L1 due to pending L1 events which require exit
8346 if (is_guest_mode(vcpu)) {
8347 r = kvm_x86_ops.nested_ops->check_events(vcpu);
8352 /* try to inject new event if pending */
8353 if (vcpu->arch.exception.pending) {
8354 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8355 vcpu->arch.exception.has_error_code,
8356 vcpu->arch.exception.error_code);
8358 vcpu->arch.exception.pending = false;
8359 vcpu->arch.exception.injected = true;
8361 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8362 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8365 if (vcpu->arch.exception.nr == DB_VECTOR) {
8366 kvm_deliver_exception_payload(vcpu);
8367 if (vcpu->arch.dr7 & DR7_GD) {
8368 vcpu->arch.dr7 &= ~DR7_GD;
8369 kvm_update_dr7(vcpu);
8373 static_call(kvm_x86_queue_exception)(vcpu);
8378 * Finally, inject interrupt events. If an event cannot be injected
8379 * due to architectural conditions (e.g. IF=0) a window-open exit
8380 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8381 * and can architecturally be injected, but we cannot do it right now:
8382 * an interrupt could have arrived just now and we have to inject it
8383 * as a vmexit, or there could already an event in the queue, which is
8384 * indicated by can_inject. In that case we request an immediate exit
8385 * in order to make progress and get back here for another iteration.
8386 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8388 if (vcpu->arch.smi_pending) {
8389 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8393 vcpu->arch.smi_pending = false;
8394 ++vcpu->arch.smi_count;
8398 static_call(kvm_x86_enable_smi_window)(vcpu);
8401 if (vcpu->arch.nmi_pending) {
8402 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8406 --vcpu->arch.nmi_pending;
8407 vcpu->arch.nmi_injected = true;
8408 static_call(kvm_x86_set_nmi)(vcpu);
8410 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8412 if (vcpu->arch.nmi_pending)
8413 static_call(kvm_x86_enable_nmi_window)(vcpu);
8416 if (kvm_cpu_has_injectable_intr(vcpu)) {
8417 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8421 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8422 static_call(kvm_x86_set_irq)(vcpu);
8423 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8425 if (kvm_cpu_has_injectable_intr(vcpu))
8426 static_call(kvm_x86_enable_irq_window)(vcpu);
8429 if (is_guest_mode(vcpu) &&
8430 kvm_x86_ops.nested_ops->hv_timer_pending &&
8431 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8432 *req_immediate_exit = true;
8434 WARN_ON(vcpu->arch.exception.pending);
8438 *req_immediate_exit = true;
8442 static void process_nmi(struct kvm_vcpu *vcpu)
8447 * x86 is limited to one NMI running, and one NMI pending after it.
8448 * If an NMI is already in progress, limit further NMIs to just one.
8449 * Otherwise, allow two (and we'll inject the first one immediately).
8451 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8454 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8455 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8456 kvm_make_request(KVM_REQ_EVENT, vcpu);
8459 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8462 flags |= seg->g << 23;
8463 flags |= seg->db << 22;
8464 flags |= seg->l << 21;
8465 flags |= seg->avl << 20;
8466 flags |= seg->present << 15;
8467 flags |= seg->dpl << 13;
8468 flags |= seg->s << 12;
8469 flags |= seg->type << 8;
8473 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8475 struct kvm_segment seg;
8478 kvm_get_segment(vcpu, &seg, n);
8479 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8482 offset = 0x7f84 + n * 12;
8484 offset = 0x7f2c + (n - 3) * 12;
8486 put_smstate(u32, buf, offset + 8, seg.base);
8487 put_smstate(u32, buf, offset + 4, seg.limit);
8488 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8491 #ifdef CONFIG_X86_64
8492 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8494 struct kvm_segment seg;
8498 kvm_get_segment(vcpu, &seg, n);
8499 offset = 0x7e00 + n * 16;
8501 flags = enter_smm_get_segment_flags(&seg) >> 8;
8502 put_smstate(u16, buf, offset, seg.selector);
8503 put_smstate(u16, buf, offset + 2, flags);
8504 put_smstate(u32, buf, offset + 4, seg.limit);
8505 put_smstate(u64, buf, offset + 8, seg.base);
8509 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8512 struct kvm_segment seg;
8516 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8517 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8518 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8519 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8521 for (i = 0; i < 8; i++)
8522 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
8524 kvm_get_dr(vcpu, 6, &val);
8525 put_smstate(u32, buf, 0x7fcc, (u32)val);
8526 kvm_get_dr(vcpu, 7, &val);
8527 put_smstate(u32, buf, 0x7fc8, (u32)val);
8529 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8530 put_smstate(u32, buf, 0x7fc4, seg.selector);
8531 put_smstate(u32, buf, 0x7f64, seg.base);
8532 put_smstate(u32, buf, 0x7f60, seg.limit);
8533 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8535 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8536 put_smstate(u32, buf, 0x7fc0, seg.selector);
8537 put_smstate(u32, buf, 0x7f80, seg.base);
8538 put_smstate(u32, buf, 0x7f7c, seg.limit);
8539 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8541 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8542 put_smstate(u32, buf, 0x7f74, dt.address);
8543 put_smstate(u32, buf, 0x7f70, dt.size);
8545 static_call(kvm_x86_get_idt)(vcpu, &dt);
8546 put_smstate(u32, buf, 0x7f58, dt.address);
8547 put_smstate(u32, buf, 0x7f54, dt.size);
8549 for (i = 0; i < 6; i++)
8550 enter_smm_save_seg_32(vcpu, buf, i);
8552 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8555 put_smstate(u32, buf, 0x7efc, 0x00020000);
8556 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8559 #ifdef CONFIG_X86_64
8560 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8563 struct kvm_segment seg;
8567 for (i = 0; i < 16; i++)
8568 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8570 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8571 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8573 kvm_get_dr(vcpu, 6, &val);
8574 put_smstate(u64, buf, 0x7f68, val);
8575 kvm_get_dr(vcpu, 7, &val);
8576 put_smstate(u64, buf, 0x7f60, val);
8578 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8579 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8580 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8582 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8585 put_smstate(u32, buf, 0x7efc, 0x00020064);
8587 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8589 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8590 put_smstate(u16, buf, 0x7e90, seg.selector);
8591 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8592 put_smstate(u32, buf, 0x7e94, seg.limit);
8593 put_smstate(u64, buf, 0x7e98, seg.base);
8595 static_call(kvm_x86_get_idt)(vcpu, &dt);
8596 put_smstate(u32, buf, 0x7e84, dt.size);
8597 put_smstate(u64, buf, 0x7e88, dt.address);
8599 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8600 put_smstate(u16, buf, 0x7e70, seg.selector);
8601 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8602 put_smstate(u32, buf, 0x7e74, seg.limit);
8603 put_smstate(u64, buf, 0x7e78, seg.base);
8605 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8606 put_smstate(u32, buf, 0x7e64, dt.size);
8607 put_smstate(u64, buf, 0x7e68, dt.address);
8609 for (i = 0; i < 6; i++)
8610 enter_smm_save_seg_64(vcpu, buf, i);
8614 static void enter_smm(struct kvm_vcpu *vcpu)
8616 struct kvm_segment cs, ds;
8621 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8622 memset(buf, 0, 512);
8623 #ifdef CONFIG_X86_64
8624 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8625 enter_smm_save_state_64(vcpu, buf);
8628 enter_smm_save_state_32(vcpu, buf);
8631 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8632 * vCPU state (e.g. leave guest mode) after we've saved the state into
8633 * the SMM state-save area.
8635 static_call(kvm_x86_pre_enter_smm)(vcpu, buf);
8637 vcpu->arch.hflags |= HF_SMM_MASK;
8638 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8640 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
8641 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8643 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
8645 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8646 kvm_rip_write(vcpu, 0x8000);
8648 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8649 static_call(kvm_x86_set_cr0)(vcpu, cr0);
8650 vcpu->arch.cr0 = cr0;
8652 static_call(kvm_x86_set_cr4)(vcpu, 0);
8654 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8655 dt.address = dt.size = 0;
8656 static_call(kvm_x86_set_idt)(vcpu, &dt);
8658 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8660 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8661 cs.base = vcpu->arch.smbase;
8666 cs.limit = ds.limit = 0xffffffff;
8667 cs.type = ds.type = 0x3;
8668 cs.dpl = ds.dpl = 0;
8673 cs.avl = ds.avl = 0;
8674 cs.present = ds.present = 1;
8675 cs.unusable = ds.unusable = 0;
8676 cs.padding = ds.padding = 0;
8678 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8679 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8680 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8681 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8682 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8683 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8685 #ifdef CONFIG_X86_64
8686 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8687 static_call(kvm_x86_set_efer)(vcpu, 0);
8690 kvm_update_cpuid_runtime(vcpu);
8691 kvm_mmu_reset_context(vcpu);
8694 static void process_smi(struct kvm_vcpu *vcpu)
8696 vcpu->arch.smi_pending = true;
8697 kvm_make_request(KVM_REQ_EVENT, vcpu);
8700 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8701 unsigned long *vcpu_bitmap)
8705 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8707 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8708 NULL, vcpu_bitmap, cpus);
8710 free_cpumask_var(cpus);
8713 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8715 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8718 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8720 if (!lapic_in_kernel(vcpu))
8723 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8724 kvm_apic_update_apicv(vcpu);
8725 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
8727 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8730 * NOTE: Do not hold any lock prior to calling this.
8732 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8733 * locked, because it calls __x86_set_memory_region() which does
8734 * synchronize_srcu(&kvm->srcu).
8736 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8738 struct kvm_vcpu *except;
8739 unsigned long old, new, expected;
8741 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8742 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
8745 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8747 expected = new = old;
8749 __clear_bit(bit, &new);
8751 __set_bit(bit, &new);
8754 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8755 } while (old != expected);
8760 trace_kvm_apicv_update_request(activate, bit);
8761 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8762 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
8765 * Sending request to update APICV for all other vcpus,
8766 * while update the calling vcpu immediately instead of
8767 * waiting for another #VMEXIT to handle the request.
8769 except = kvm_get_running_vcpu();
8770 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8773 kvm_vcpu_update_apicv(except);
8775 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8777 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8779 if (!kvm_apic_present(vcpu))
8782 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8784 if (irqchip_split(vcpu->kvm))
8785 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8787 if (vcpu->arch.apicv_active)
8788 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
8789 if (ioapic_in_kernel(vcpu->kvm))
8790 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8793 if (is_guest_mode(vcpu))
8794 vcpu->arch.load_eoi_exitmap_pending = true;
8796 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8799 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8801 u64 eoi_exit_bitmap[4];
8803 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8806 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8807 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8808 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
8811 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8812 unsigned long start, unsigned long end)
8814 unsigned long apic_address;
8817 * The physical address of apic access page is stored in the VMCS.
8818 * Update it when it becomes invalid.
8820 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8821 if (start <= apic_address && apic_address < end)
8822 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8825 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8827 if (!lapic_in_kernel(vcpu))
8830 if (!kvm_x86_ops.set_apic_access_page_addr)
8833 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
8836 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8838 smp_send_reschedule(vcpu->cpu);
8840 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8843 * Returns 1 to let vcpu_run() continue the guest execution loop without
8844 * exiting to the userspace. Otherwise, the value will be returned to the
8847 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8851 dm_request_for_irq_injection(vcpu) &&
8852 kvm_cpu_accept_dm_intr(vcpu);
8853 fastpath_t exit_fastpath;
8855 bool req_immediate_exit = false;
8857 /* Forbid vmenter if vcpu dirty ring is soft-full */
8858 if (unlikely(vcpu->kvm->dirty_ring_size &&
8859 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
8860 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
8861 trace_kvm_dirty_ring_exit(vcpu);
8866 if (kvm_request_pending(vcpu)) {
8867 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
8868 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
8873 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8874 kvm_mmu_unload(vcpu);
8875 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8876 __kvm_migrate_timers(vcpu);
8877 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8878 kvm_gen_update_masterclock(vcpu->kvm);
8879 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8880 kvm_gen_kvmclock_update(vcpu);
8881 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8882 r = kvm_guest_time_update(vcpu);
8886 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8887 kvm_mmu_sync_roots(vcpu);
8888 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8889 kvm_mmu_load_pgd(vcpu);
8890 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8891 kvm_vcpu_flush_tlb_all(vcpu);
8893 /* Flushing all ASIDs flushes the current ASID... */
8894 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8896 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8897 kvm_vcpu_flush_tlb_current(vcpu);
8898 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8899 kvm_vcpu_flush_tlb_guest(vcpu);
8901 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8902 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8906 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8907 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8908 vcpu->mmio_needed = 0;
8912 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8913 /* Page is swapped out. Do synthetic halt */
8914 vcpu->arch.apf.halted = true;
8918 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8919 record_steal_time(vcpu);
8920 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8922 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8924 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8925 kvm_pmu_handle_event(vcpu);
8926 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8927 kvm_pmu_deliver_pmi(vcpu);
8928 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8929 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8930 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8931 vcpu->arch.ioapic_handled_vectors)) {
8932 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8933 vcpu->run->eoi.vector =
8934 vcpu->arch.pending_ioapic_eoi;
8939 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8940 vcpu_scan_ioapic(vcpu);
8941 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8942 vcpu_load_eoi_exitmap(vcpu);
8943 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8944 kvm_vcpu_reload_apic_access_page(vcpu);
8945 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8946 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8947 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8951 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8952 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8953 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8957 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8958 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8959 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8965 * KVM_REQ_HV_STIMER has to be processed after
8966 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8967 * depend on the guest clock being up-to-date
8969 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8970 kvm_hv_process_stimers(vcpu);
8971 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8972 kvm_vcpu_update_apicv(vcpu);
8973 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
8974 kvm_check_async_pf_completion(vcpu);
8975 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
8976 static_call(kvm_x86_msr_filter_changed)(vcpu);
8979 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
8980 kvm_xen_has_interrupt(vcpu)) {
8981 ++vcpu->stat.req_event;
8982 kvm_apic_accept_events(vcpu);
8983 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8988 inject_pending_event(vcpu, &req_immediate_exit);
8990 static_call(kvm_x86_enable_irq_window)(vcpu);
8992 if (kvm_lapic_enabled(vcpu)) {
8993 update_cr8_intercept(vcpu);
8994 kvm_lapic_sync_to_vapic(vcpu);
8998 r = kvm_mmu_reload(vcpu);
9000 goto cancel_injection;
9005 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9008 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9009 * IPI are then delayed after guest entry, which ensures that they
9010 * result in virtual interrupt delivery.
9012 local_irq_disable();
9013 vcpu->mode = IN_GUEST_MODE;
9015 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9018 * 1) We should set ->mode before checking ->requests. Please see
9019 * the comment in kvm_vcpu_exiting_guest_mode().
9021 * 2) For APICv, we should set ->mode before checking PID.ON. This
9022 * pairs with the memory barrier implicit in pi_test_and_set_on
9023 * (see vmx_deliver_posted_interrupt).
9025 * 3) This also orders the write to mode from any reads to the page
9026 * tables done while the VCPU is running. Please see the comment
9027 * in kvm_flush_remote_tlbs.
9029 smp_mb__after_srcu_read_unlock();
9032 * This handles the case where a posted interrupt was
9033 * notified with kvm_vcpu_kick.
9035 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9036 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9038 if (kvm_vcpu_exit_request(vcpu)) {
9039 vcpu->mode = OUTSIDE_GUEST_MODE;
9043 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9045 goto cancel_injection;
9048 if (req_immediate_exit) {
9049 kvm_make_request(KVM_REQ_EVENT, vcpu);
9050 static_call(kvm_x86_request_immediate_exit)(vcpu);
9053 fpregs_assert_state_consistent();
9054 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9055 switch_fpu_return();
9057 if (unlikely(vcpu->arch.switch_db_regs)) {
9059 set_debugreg(vcpu->arch.eff_db[0], 0);
9060 set_debugreg(vcpu->arch.eff_db[1], 1);
9061 set_debugreg(vcpu->arch.eff_db[2], 2);
9062 set_debugreg(vcpu->arch.eff_db[3], 3);
9063 set_debugreg(vcpu->arch.dr6, 6);
9064 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9068 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9069 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9072 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9073 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9077 if (vcpu->arch.apicv_active)
9078 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9082 * Do this here before restoring debug registers on the host. And
9083 * since we do this before handling the vmexit, a DR access vmexit
9084 * can (a) read the correct value of the debug registers, (b) set
9085 * KVM_DEBUGREG_WONT_EXIT again.
9087 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9088 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9089 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9090 kvm_update_dr0123(vcpu);
9091 kvm_update_dr7(vcpu);
9092 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9096 * If the guest has used debug registers, at least dr7
9097 * will be disabled while returning to the host.
9098 * If we don't have active breakpoints in the host, we don't
9099 * care about the messed up debug address registers. But if
9100 * we have some of them active, restore the old state.
9102 if (hw_breakpoint_active())
9103 hw_breakpoint_restore();
9105 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9106 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9108 vcpu->mode = OUTSIDE_GUEST_MODE;
9111 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9114 * Consume any pending interrupts, including the possible source of
9115 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9116 * An instruction is required after local_irq_enable() to fully unblock
9117 * interrupts on processors that implement an interrupt shadow, the
9118 * stat.exits increment will do nicely.
9120 kvm_before_interrupt(vcpu);
9123 local_irq_disable();
9124 kvm_after_interrupt(vcpu);
9126 if (lapic_in_kernel(vcpu)) {
9127 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9128 if (delta != S64_MIN) {
9129 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9130 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9137 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9140 * Profile KVM exit RIPs:
9142 if (unlikely(prof_on == KVM_PROFILING)) {
9143 unsigned long rip = kvm_rip_read(vcpu);
9144 profile_hit(KVM_PROFILING, (void *)rip);
9147 if (unlikely(vcpu->arch.tsc_always_catchup))
9148 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9150 if (vcpu->arch.apic_attention)
9151 kvm_lapic_sync_from_vapic(vcpu);
9153 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9157 if (req_immediate_exit)
9158 kvm_make_request(KVM_REQ_EVENT, vcpu);
9159 static_call(kvm_x86_cancel_injection)(vcpu);
9160 if (unlikely(vcpu->arch.apic_attention))
9161 kvm_lapic_sync_from_vapic(vcpu);
9166 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9168 if (!kvm_arch_vcpu_runnable(vcpu) &&
9169 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9170 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9171 kvm_vcpu_block(vcpu);
9172 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9174 if (kvm_x86_ops.post_block)
9175 static_call(kvm_x86_post_block)(vcpu);
9177 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9181 kvm_apic_accept_events(vcpu);
9182 switch(vcpu->arch.mp_state) {
9183 case KVM_MP_STATE_HALTED:
9184 case KVM_MP_STATE_AP_RESET_HOLD:
9185 vcpu->arch.pv.pv_unhalted = false;
9186 vcpu->arch.mp_state =
9187 KVM_MP_STATE_RUNNABLE;
9189 case KVM_MP_STATE_RUNNABLE:
9190 vcpu->arch.apf.halted = false;
9192 case KVM_MP_STATE_INIT_RECEIVED:
9200 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9202 if (is_guest_mode(vcpu))
9203 kvm_x86_ops.nested_ops->check_events(vcpu);
9205 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9206 !vcpu->arch.apf.halted);
9209 static int vcpu_run(struct kvm_vcpu *vcpu)
9212 struct kvm *kvm = vcpu->kvm;
9214 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9215 vcpu->arch.l1tf_flush_l1d = true;
9218 if (kvm_vcpu_running(vcpu)) {
9219 r = vcpu_enter_guest(vcpu);
9221 r = vcpu_block(kvm, vcpu);
9227 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9228 if (kvm_cpu_has_pending_timer(vcpu))
9229 kvm_inject_pending_timer_irqs(vcpu);
9231 if (dm_request_for_irq_injection(vcpu) &&
9232 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9234 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9235 ++vcpu->stat.request_irq_exits;
9239 if (__xfer_to_guest_mode_work_pending()) {
9240 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9241 r = xfer_to_guest_mode_handle_work(vcpu);
9244 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9248 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9253 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9257 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9258 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9259 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9263 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9265 BUG_ON(!vcpu->arch.pio.count);
9267 return complete_emulated_io(vcpu);
9271 * Implements the following, as a state machine:
9275 * for each mmio piece in the fragment
9283 * for each mmio piece in the fragment
9288 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9290 struct kvm_run *run = vcpu->run;
9291 struct kvm_mmio_fragment *frag;
9294 BUG_ON(!vcpu->mmio_needed);
9296 /* Complete previous fragment */
9297 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9298 len = min(8u, frag->len);
9299 if (!vcpu->mmio_is_write)
9300 memcpy(frag->data, run->mmio.data, len);
9302 if (frag->len <= 8) {
9303 /* Switch to the next fragment. */
9305 vcpu->mmio_cur_fragment++;
9307 /* Go forward to the next mmio piece. */
9313 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9314 vcpu->mmio_needed = 0;
9316 /* FIXME: return into emulator if single-stepping. */
9317 if (vcpu->mmio_is_write)
9319 vcpu->mmio_read_completed = 1;
9320 return complete_emulated_io(vcpu);
9323 run->exit_reason = KVM_EXIT_MMIO;
9324 run->mmio.phys_addr = frag->gpa;
9325 if (vcpu->mmio_is_write)
9326 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9327 run->mmio.len = min(8u, frag->len);
9328 run->mmio.is_write = vcpu->mmio_is_write;
9329 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9333 static void kvm_save_current_fpu(struct fpu *fpu)
9336 * If the target FPU state is not resident in the CPU registers, just
9337 * memcpy() from current, else save CPU state directly to the target.
9339 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9340 memcpy(&fpu->state, ¤t->thread.fpu.state,
9341 fpu_kernel_xstate_size);
9343 copy_fpregs_to_fpstate(fpu);
9346 /* Swap (qemu) user FPU context for the guest FPU context. */
9347 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9351 kvm_save_current_fpu(vcpu->arch.user_fpu);
9354 * Guests with protected state can't have it set by the hypervisor,
9355 * so skip trying to set it.
9357 if (vcpu->arch.guest_fpu)
9358 /* PKRU is separately restored in kvm_x86_ops.run. */
9359 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9360 ~XFEATURE_MASK_PKRU);
9362 fpregs_mark_activate();
9368 /* When vcpu_run ends, restore user space FPU context. */
9369 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9374 * Guests with protected state can't have it read by the hypervisor,
9375 * so skip trying to save it.
9377 if (vcpu->arch.guest_fpu)
9378 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9380 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9382 fpregs_mark_activate();
9385 ++vcpu->stat.fpu_reload;
9389 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9391 struct kvm_run *kvm_run = vcpu->run;
9395 kvm_sigset_activate(vcpu);
9397 kvm_load_guest_fpu(vcpu);
9399 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9400 if (kvm_run->immediate_exit) {
9404 kvm_vcpu_block(vcpu);
9405 kvm_apic_accept_events(vcpu);
9406 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9408 if (signal_pending(current)) {
9410 kvm_run->exit_reason = KVM_EXIT_INTR;
9411 ++vcpu->stat.signal_exits;
9416 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9421 if (kvm_run->kvm_dirty_regs) {
9422 r = sync_regs(vcpu);
9427 /* re-sync apic's tpr */
9428 if (!lapic_in_kernel(vcpu)) {
9429 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9435 if (unlikely(vcpu->arch.complete_userspace_io)) {
9436 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9437 vcpu->arch.complete_userspace_io = NULL;
9442 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9444 if (kvm_run->immediate_exit)
9450 kvm_put_guest_fpu(vcpu);
9451 if (kvm_run->kvm_valid_regs)
9453 post_kvm_run_save(vcpu);
9454 kvm_sigset_deactivate(vcpu);
9460 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9462 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9464 * We are here if userspace calls get_regs() in the middle of
9465 * instruction emulation. Registers state needs to be copied
9466 * back from emulation context to vcpu. Userspace shouldn't do
9467 * that usually, but some bad designed PV devices (vmware
9468 * backdoor interface) need this to work
9470 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9471 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9473 regs->rax = kvm_rax_read(vcpu);
9474 regs->rbx = kvm_rbx_read(vcpu);
9475 regs->rcx = kvm_rcx_read(vcpu);
9476 regs->rdx = kvm_rdx_read(vcpu);
9477 regs->rsi = kvm_rsi_read(vcpu);
9478 regs->rdi = kvm_rdi_read(vcpu);
9479 regs->rsp = kvm_rsp_read(vcpu);
9480 regs->rbp = kvm_rbp_read(vcpu);
9481 #ifdef CONFIG_X86_64
9482 regs->r8 = kvm_r8_read(vcpu);
9483 regs->r9 = kvm_r9_read(vcpu);
9484 regs->r10 = kvm_r10_read(vcpu);
9485 regs->r11 = kvm_r11_read(vcpu);
9486 regs->r12 = kvm_r12_read(vcpu);
9487 regs->r13 = kvm_r13_read(vcpu);
9488 regs->r14 = kvm_r14_read(vcpu);
9489 regs->r15 = kvm_r15_read(vcpu);
9492 regs->rip = kvm_rip_read(vcpu);
9493 regs->rflags = kvm_get_rflags(vcpu);
9496 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9499 __get_regs(vcpu, regs);
9504 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9506 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9507 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9509 kvm_rax_write(vcpu, regs->rax);
9510 kvm_rbx_write(vcpu, regs->rbx);
9511 kvm_rcx_write(vcpu, regs->rcx);
9512 kvm_rdx_write(vcpu, regs->rdx);
9513 kvm_rsi_write(vcpu, regs->rsi);
9514 kvm_rdi_write(vcpu, regs->rdi);
9515 kvm_rsp_write(vcpu, regs->rsp);
9516 kvm_rbp_write(vcpu, regs->rbp);
9517 #ifdef CONFIG_X86_64
9518 kvm_r8_write(vcpu, regs->r8);
9519 kvm_r9_write(vcpu, regs->r9);
9520 kvm_r10_write(vcpu, regs->r10);
9521 kvm_r11_write(vcpu, regs->r11);
9522 kvm_r12_write(vcpu, regs->r12);
9523 kvm_r13_write(vcpu, regs->r13);
9524 kvm_r14_write(vcpu, regs->r14);
9525 kvm_r15_write(vcpu, regs->r15);
9528 kvm_rip_write(vcpu, regs->rip);
9529 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9531 vcpu->arch.exception.pending = false;
9533 kvm_make_request(KVM_REQ_EVENT, vcpu);
9536 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9539 __set_regs(vcpu, regs);
9544 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9546 struct kvm_segment cs;
9548 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9552 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9554 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9558 if (vcpu->arch.guest_state_protected)
9559 goto skip_protected_regs;
9561 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9562 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9563 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9564 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9565 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9566 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9568 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9569 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9571 static_call(kvm_x86_get_idt)(vcpu, &dt);
9572 sregs->idt.limit = dt.size;
9573 sregs->idt.base = dt.address;
9574 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9575 sregs->gdt.limit = dt.size;
9576 sregs->gdt.base = dt.address;
9578 sregs->cr2 = vcpu->arch.cr2;
9579 sregs->cr3 = kvm_read_cr3(vcpu);
9581 skip_protected_regs:
9582 sregs->cr0 = kvm_read_cr0(vcpu);
9583 sregs->cr4 = kvm_read_cr4(vcpu);
9584 sregs->cr8 = kvm_get_cr8(vcpu);
9585 sregs->efer = vcpu->arch.efer;
9586 sregs->apic_base = kvm_get_apic_base(vcpu);
9588 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9590 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9591 set_bit(vcpu->arch.interrupt.nr,
9592 (unsigned long *)sregs->interrupt_bitmap);
9595 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9596 struct kvm_sregs *sregs)
9599 __get_sregs(vcpu, sregs);
9604 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9605 struct kvm_mp_state *mp_state)
9608 if (kvm_mpx_supported())
9609 kvm_load_guest_fpu(vcpu);
9611 kvm_apic_accept_events(vcpu);
9612 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9613 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9614 vcpu->arch.pv.pv_unhalted)
9615 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9617 mp_state->mp_state = vcpu->arch.mp_state;
9619 if (kvm_mpx_supported())
9620 kvm_put_guest_fpu(vcpu);
9625 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9626 struct kvm_mp_state *mp_state)
9632 if (!lapic_in_kernel(vcpu) &&
9633 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9637 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9638 * INIT state; latched init should be reported using
9639 * KVM_SET_VCPU_EVENTS, so reject it here.
9641 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9642 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9643 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9646 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9647 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9648 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9650 vcpu->arch.mp_state = mp_state->mp_state;
9651 kvm_make_request(KVM_REQ_EVENT, vcpu);
9659 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9660 int reason, bool has_error_code, u32 error_code)
9662 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9665 init_emulate_ctxt(vcpu);
9667 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9668 has_error_code, error_code);
9670 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9671 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9672 vcpu->run->internal.ndata = 0;
9676 kvm_rip_write(vcpu, ctxt->eip);
9677 kvm_set_rflags(vcpu, ctxt->eflags);
9680 EXPORT_SYMBOL_GPL(kvm_task_switch);
9682 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9684 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9686 * When EFER.LME and CR0.PG are set, the processor is in
9687 * 64-bit mode (though maybe in a 32-bit code segment).
9688 * CR4.PAE and EFER.LMA must be set.
9690 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9692 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
9696 * Not in 64-bit mode: EFER.LMA is clear and the code
9697 * segment cannot be 64-bit.
9699 if (sregs->efer & EFER_LMA || sregs->cs.l)
9703 return kvm_is_valid_cr4(vcpu, sregs->cr4);
9706 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9708 struct msr_data apic_base_msr;
9709 int mmu_reset_needed = 0;
9710 int pending_vec, max_bits, idx;
9714 if (!kvm_is_valid_sregs(vcpu, sregs))
9717 apic_base_msr.data = sregs->apic_base;
9718 apic_base_msr.host_initiated = true;
9719 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9722 if (vcpu->arch.guest_state_protected)
9723 goto skip_protected_regs;
9725 dt.size = sregs->idt.limit;
9726 dt.address = sregs->idt.base;
9727 static_call(kvm_x86_set_idt)(vcpu, &dt);
9728 dt.size = sregs->gdt.limit;
9729 dt.address = sregs->gdt.base;
9730 static_call(kvm_x86_set_gdt)(vcpu, &dt);
9732 vcpu->arch.cr2 = sregs->cr2;
9733 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9734 vcpu->arch.cr3 = sregs->cr3;
9735 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9737 kvm_set_cr8(vcpu, sregs->cr8);
9739 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9740 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
9742 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9743 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
9744 vcpu->arch.cr0 = sregs->cr0;
9746 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9747 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
9749 idx = srcu_read_lock(&vcpu->kvm->srcu);
9750 if (is_pae_paging(vcpu)) {
9751 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9752 mmu_reset_needed = 1;
9754 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9756 if (mmu_reset_needed)
9757 kvm_mmu_reset_context(vcpu);
9759 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9760 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9761 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9762 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9763 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9764 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9766 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9767 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9769 update_cr8_intercept(vcpu);
9771 /* Older userspace won't unhalt the vcpu on reset. */
9772 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9773 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9775 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9777 skip_protected_regs:
9778 max_bits = KVM_NR_INTERRUPTS;
9779 pending_vec = find_first_bit(
9780 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9781 if (pending_vec < max_bits) {
9782 kvm_queue_interrupt(vcpu, pending_vec, false);
9783 pr_debug("Set back pending irq %d\n", pending_vec);
9786 kvm_make_request(KVM_REQ_EVENT, vcpu);
9793 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9794 struct kvm_sregs *sregs)
9799 ret = __set_sregs(vcpu, sregs);
9804 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9805 struct kvm_guest_debug *dbg)
9807 unsigned long rflags;
9810 if (vcpu->arch.guest_state_protected)
9815 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9817 if (vcpu->arch.exception.pending)
9819 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9820 kvm_queue_exception(vcpu, DB_VECTOR);
9822 kvm_queue_exception(vcpu, BP_VECTOR);
9826 * Read rflags as long as potentially injected trace flags are still
9829 rflags = kvm_get_rflags(vcpu);
9831 vcpu->guest_debug = dbg->control;
9832 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9833 vcpu->guest_debug = 0;
9835 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9836 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9837 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9838 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9840 for (i = 0; i < KVM_NR_DB_REGS; i++)
9841 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9843 kvm_update_dr7(vcpu);
9845 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9846 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9847 get_segment_base(vcpu, VCPU_SREG_CS);
9850 * Trigger an rflags update that will inject or remove the trace
9853 kvm_set_rflags(vcpu, rflags);
9855 static_call(kvm_x86_update_exception_bitmap)(vcpu);
9865 * Translate a guest virtual address to a guest physical address.
9867 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9868 struct kvm_translation *tr)
9870 unsigned long vaddr = tr->linear_address;
9876 idx = srcu_read_lock(&vcpu->kvm->srcu);
9877 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9878 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9879 tr->physical_address = gpa;
9880 tr->valid = gpa != UNMAPPED_GVA;
9888 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9890 struct fxregs_state *fxsave;
9892 if (!vcpu->arch.guest_fpu)
9897 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9898 memcpy(fpu->fpr, fxsave->st_space, 128);
9899 fpu->fcw = fxsave->cwd;
9900 fpu->fsw = fxsave->swd;
9901 fpu->ftwx = fxsave->twd;
9902 fpu->last_opcode = fxsave->fop;
9903 fpu->last_ip = fxsave->rip;
9904 fpu->last_dp = fxsave->rdp;
9905 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9911 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9913 struct fxregs_state *fxsave;
9915 if (!vcpu->arch.guest_fpu)
9920 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9922 memcpy(fxsave->st_space, fpu->fpr, 128);
9923 fxsave->cwd = fpu->fcw;
9924 fxsave->swd = fpu->fsw;
9925 fxsave->twd = fpu->ftwx;
9926 fxsave->fop = fpu->last_opcode;
9927 fxsave->rip = fpu->last_ip;
9928 fxsave->rdp = fpu->last_dp;
9929 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9935 static void store_regs(struct kvm_vcpu *vcpu)
9937 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9939 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9940 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9942 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9943 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9945 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9946 kvm_vcpu_ioctl_x86_get_vcpu_events(
9947 vcpu, &vcpu->run->s.regs.events);
9950 static int sync_regs(struct kvm_vcpu *vcpu)
9952 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9955 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9956 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9957 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9959 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9960 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9962 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9964 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9965 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9966 vcpu, &vcpu->run->s.regs.events))
9968 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9974 static void fx_init(struct kvm_vcpu *vcpu)
9976 if (!vcpu->arch.guest_fpu)
9979 fpstate_init(&vcpu->arch.guest_fpu->state);
9980 if (boot_cpu_has(X86_FEATURE_XSAVES))
9981 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9982 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9985 * Ensure guest xcr0 is valid for loading
9987 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9989 vcpu->arch.cr0 |= X86_CR0_ET;
9992 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
9994 if (vcpu->arch.guest_fpu) {
9995 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9996 vcpu->arch.guest_fpu = NULL;
9999 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10001 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10003 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10004 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10005 "guest TSC will not be reliable\n");
10010 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10015 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10016 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10018 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10020 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10022 r = kvm_mmu_create(vcpu);
10026 if (irqchip_in_kernel(vcpu->kvm)) {
10027 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10029 goto fail_mmu_destroy;
10030 if (kvm_apicv_activated(vcpu->kvm))
10031 vcpu->arch.apicv_active = true;
10033 static_branch_inc(&kvm_has_noapic_vcpu);
10037 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10039 goto fail_free_lapic;
10040 vcpu->arch.pio_data = page_address(page);
10042 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10043 GFP_KERNEL_ACCOUNT);
10044 if (!vcpu->arch.mce_banks)
10045 goto fail_free_pio_data;
10046 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10048 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10049 GFP_KERNEL_ACCOUNT))
10050 goto fail_free_mce_banks;
10052 if (!alloc_emulate_ctxt(vcpu))
10053 goto free_wbinvd_dirty_mask;
10055 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10056 GFP_KERNEL_ACCOUNT);
10057 if (!vcpu->arch.user_fpu) {
10058 pr_err("kvm: failed to allocate userspace's fpu\n");
10059 goto free_emulate_ctxt;
10062 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10063 GFP_KERNEL_ACCOUNT);
10064 if (!vcpu->arch.guest_fpu) {
10065 pr_err("kvm: failed to allocate vcpu's fpu\n");
10066 goto free_user_fpu;
10070 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10071 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10073 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10075 kvm_async_pf_hash_reset(vcpu);
10076 kvm_pmu_init(vcpu);
10078 vcpu->arch.pending_external_vector = -1;
10079 vcpu->arch.preempted_in_kernel = false;
10081 kvm_hv_vcpu_init(vcpu);
10083 r = static_call(kvm_x86_vcpu_create)(vcpu);
10085 goto free_guest_fpu;
10087 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10088 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10089 kvm_vcpu_mtrr_init(vcpu);
10091 kvm_vcpu_reset(vcpu, false);
10092 kvm_init_mmu(vcpu, false);
10097 kvm_free_guest_fpu(vcpu);
10099 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10101 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10102 free_wbinvd_dirty_mask:
10103 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10104 fail_free_mce_banks:
10105 kfree(vcpu->arch.mce_banks);
10106 fail_free_pio_data:
10107 free_page((unsigned long)vcpu->arch.pio_data);
10109 kvm_free_lapic(vcpu);
10111 kvm_mmu_destroy(vcpu);
10115 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10117 struct kvm *kvm = vcpu->kvm;
10119 kvm_hv_vcpu_postcreate(vcpu);
10121 if (mutex_lock_killable(&vcpu->mutex))
10124 kvm_synchronize_tsc(vcpu, 0);
10127 /* poll control enabled by default */
10128 vcpu->arch.msr_kvm_poll_control = 1;
10130 mutex_unlock(&vcpu->mutex);
10132 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10133 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10134 KVMCLOCK_SYNC_PERIOD);
10137 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10139 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10142 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10144 kvmclock_reset(vcpu);
10146 static_call(kvm_x86_vcpu_free)(vcpu);
10148 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10149 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10150 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10151 kvm_free_guest_fpu(vcpu);
10153 kvm_hv_vcpu_uninit(vcpu);
10154 kvm_pmu_destroy(vcpu);
10155 kfree(vcpu->arch.mce_banks);
10156 kvm_free_lapic(vcpu);
10157 idx = srcu_read_lock(&vcpu->kvm->srcu);
10158 kvm_mmu_destroy(vcpu);
10159 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10160 free_page((unsigned long)vcpu->arch.pio_data);
10161 kvfree(vcpu->arch.cpuid_entries);
10162 if (!lapic_in_kernel(vcpu))
10163 static_branch_dec(&kvm_has_noapic_vcpu);
10166 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10168 kvm_lapic_reset(vcpu, init_event);
10170 vcpu->arch.hflags = 0;
10172 vcpu->arch.smi_pending = 0;
10173 vcpu->arch.smi_count = 0;
10174 atomic_set(&vcpu->arch.nmi_queued, 0);
10175 vcpu->arch.nmi_pending = 0;
10176 vcpu->arch.nmi_injected = false;
10177 kvm_clear_interrupt_queue(vcpu);
10178 kvm_clear_exception_queue(vcpu);
10180 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10181 kvm_update_dr0123(vcpu);
10182 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10183 vcpu->arch.dr7 = DR7_FIXED_1;
10184 kvm_update_dr7(vcpu);
10186 vcpu->arch.cr2 = 0;
10188 kvm_make_request(KVM_REQ_EVENT, vcpu);
10189 vcpu->arch.apf.msr_en_val = 0;
10190 vcpu->arch.apf.msr_int_val = 0;
10191 vcpu->arch.st.msr_val = 0;
10193 kvmclock_reset(vcpu);
10195 kvm_clear_async_pf_completion_queue(vcpu);
10196 kvm_async_pf_hash_reset(vcpu);
10197 vcpu->arch.apf.halted = false;
10199 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10200 void *mpx_state_buffer;
10203 * To avoid have the INIT path from kvm_apic_has_events() that be
10204 * called with loaded FPU and does not let userspace fix the state.
10207 kvm_put_guest_fpu(vcpu);
10208 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10210 if (mpx_state_buffer)
10211 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10212 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10214 if (mpx_state_buffer)
10215 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10217 kvm_load_guest_fpu(vcpu);
10221 kvm_pmu_reset(vcpu);
10222 vcpu->arch.smbase = 0x30000;
10224 vcpu->arch.msr_misc_features_enables = 0;
10226 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10229 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10230 vcpu->arch.regs_avail = ~0;
10231 vcpu->arch.regs_dirty = ~0;
10233 vcpu->arch.ia32_xss = 0;
10235 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10238 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10240 struct kvm_segment cs;
10242 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10243 cs.selector = vector << 8;
10244 cs.base = vector << 12;
10245 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10246 kvm_rip_write(vcpu, 0);
10248 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10250 int kvm_arch_hardware_enable(void)
10253 struct kvm_vcpu *vcpu;
10258 bool stable, backwards_tsc = false;
10260 kvm_user_return_msr_cpu_online();
10261 ret = static_call(kvm_x86_hardware_enable)();
10265 local_tsc = rdtsc();
10266 stable = !kvm_check_tsc_unstable();
10267 list_for_each_entry(kvm, &vm_list, vm_list) {
10268 kvm_for_each_vcpu(i, vcpu, kvm) {
10269 if (!stable && vcpu->cpu == smp_processor_id())
10270 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10271 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10272 backwards_tsc = true;
10273 if (vcpu->arch.last_host_tsc > max_tsc)
10274 max_tsc = vcpu->arch.last_host_tsc;
10280 * Sometimes, even reliable TSCs go backwards. This happens on
10281 * platforms that reset TSC during suspend or hibernate actions, but
10282 * maintain synchronization. We must compensate. Fortunately, we can
10283 * detect that condition here, which happens early in CPU bringup,
10284 * before any KVM threads can be running. Unfortunately, we can't
10285 * bring the TSCs fully up to date with real time, as we aren't yet far
10286 * enough into CPU bringup that we know how much real time has actually
10287 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10288 * variables that haven't been updated yet.
10290 * So we simply find the maximum observed TSC above, then record the
10291 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10292 * the adjustment will be applied. Note that we accumulate
10293 * adjustments, in case multiple suspend cycles happen before some VCPU
10294 * gets a chance to run again. In the event that no KVM threads get a
10295 * chance to run, we will miss the entire elapsed period, as we'll have
10296 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10297 * loose cycle time. This isn't too big a deal, since the loss will be
10298 * uniform across all VCPUs (not to mention the scenario is extremely
10299 * unlikely). It is possible that a second hibernate recovery happens
10300 * much faster than a first, causing the observed TSC here to be
10301 * smaller; this would require additional padding adjustment, which is
10302 * why we set last_host_tsc to the local tsc observed here.
10304 * N.B. - this code below runs only on platforms with reliable TSC,
10305 * as that is the only way backwards_tsc is set above. Also note
10306 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10307 * have the same delta_cyc adjustment applied if backwards_tsc
10308 * is detected. Note further, this adjustment is only done once,
10309 * as we reset last_host_tsc on all VCPUs to stop this from being
10310 * called multiple times (one for each physical CPU bringup).
10312 * Platforms with unreliable TSCs don't have to deal with this, they
10313 * will be compensated by the logic in vcpu_load, which sets the TSC to
10314 * catchup mode. This will catchup all VCPUs to real time, but cannot
10315 * guarantee that they stay in perfect synchronization.
10317 if (backwards_tsc) {
10318 u64 delta_cyc = max_tsc - local_tsc;
10319 list_for_each_entry(kvm, &vm_list, vm_list) {
10320 kvm->arch.backwards_tsc_observed = true;
10321 kvm_for_each_vcpu(i, vcpu, kvm) {
10322 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10323 vcpu->arch.last_host_tsc = local_tsc;
10324 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10328 * We have to disable TSC offset matching.. if you were
10329 * booting a VM while issuing an S4 host suspend....
10330 * you may have some problem. Solving this issue is
10331 * left as an exercise to the reader.
10333 kvm->arch.last_tsc_nsec = 0;
10334 kvm->arch.last_tsc_write = 0;
10341 void kvm_arch_hardware_disable(void)
10343 static_call(kvm_x86_hardware_disable)();
10344 drop_user_return_notifiers();
10347 int kvm_arch_hardware_setup(void *opaque)
10349 struct kvm_x86_init_ops *ops = opaque;
10352 rdmsrl_safe(MSR_EFER, &host_efer);
10354 if (boot_cpu_has(X86_FEATURE_XSAVES))
10355 rdmsrl(MSR_IA32_XSS, host_xss);
10357 r = ops->hardware_setup();
10361 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10362 kvm_ops_static_call_update();
10364 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10367 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10368 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10369 #undef __kvm_cpu_cap_has
10371 if (kvm_has_tsc_control) {
10373 * Make sure the user can only configure tsc_khz values that
10374 * fit into a signed integer.
10375 * A min value is not calculated because it will always
10376 * be 1 on all machines.
10378 u64 max = min(0x7fffffffULL,
10379 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10380 kvm_max_guest_tsc_khz = max;
10382 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10385 kvm_init_msr_list();
10389 void kvm_arch_hardware_unsetup(void)
10391 static_call(kvm_x86_hardware_unsetup)();
10394 int kvm_arch_check_processor_compat(void *opaque)
10396 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10397 struct kvm_x86_init_ops *ops = opaque;
10399 WARN_ON(!irqs_disabled());
10401 if (__cr4_reserved_bits(cpu_has, c) !=
10402 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10405 return ops->check_processor_compatibility();
10408 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10410 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10412 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10414 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10416 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10419 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10420 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10422 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10424 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10426 vcpu->arch.l1tf_flush_l1d = true;
10427 if (pmu->version && unlikely(pmu->event_count)) {
10428 pmu->need_cleanup = true;
10429 kvm_make_request(KVM_REQ_PMU, vcpu);
10431 static_call(kvm_x86_sched_in)(vcpu, cpu);
10434 void kvm_arch_free_vm(struct kvm *kvm)
10436 kfree(kvm->arch.hyperv.hv_pa_pg);
10441 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10446 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10447 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10448 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10449 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10450 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10451 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10453 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10454 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10455 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10456 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10457 &kvm->arch.irq_sources_bitmap);
10459 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10460 mutex_init(&kvm->arch.apic_map_lock);
10461 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10463 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10464 pvclock_update_vm_gtod_copy(kvm);
10466 kvm->arch.guest_can_read_msr_platform_info = true;
10468 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10469 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10471 kvm_hv_init_vm(kvm);
10472 kvm_page_track_init(kvm);
10473 kvm_mmu_init_vm(kvm);
10475 return static_call(kvm_x86_vm_init)(kvm);
10478 int kvm_arch_post_init_vm(struct kvm *kvm)
10480 return kvm_mmu_post_init_vm(kvm);
10483 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10486 kvm_mmu_unload(vcpu);
10490 static void kvm_free_vcpus(struct kvm *kvm)
10493 struct kvm_vcpu *vcpu;
10496 * Unpin any mmu pages first.
10498 kvm_for_each_vcpu(i, vcpu, kvm) {
10499 kvm_clear_async_pf_completion_queue(vcpu);
10500 kvm_unload_vcpu_mmu(vcpu);
10502 kvm_for_each_vcpu(i, vcpu, kvm)
10503 kvm_vcpu_destroy(vcpu);
10505 mutex_lock(&kvm->lock);
10506 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10507 kvm->vcpus[i] = NULL;
10509 atomic_set(&kvm->online_vcpus, 0);
10510 mutex_unlock(&kvm->lock);
10513 void kvm_arch_sync_events(struct kvm *kvm)
10515 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10516 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10520 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
10523 * __x86_set_memory_region: Setup KVM internal memory slot
10525 * @kvm: the kvm pointer to the VM.
10526 * @id: the slot ID to setup.
10527 * @gpa: the GPA to install the slot (unused when @size == 0).
10528 * @size: the size of the slot. Set to zero to uninstall a slot.
10530 * This function helps to setup a KVM internal memory slot. Specify
10531 * @size > 0 to install a new slot, while @size == 0 to uninstall a
10532 * slot. The return code can be one of the following:
10534 * HVA: on success (uninstall will return a bogus HVA)
10537 * The caller should always use IS_ERR() to check the return value
10538 * before use. Note, the KVM internal memory slots are guaranteed to
10539 * remain valid and unchanged until the VM is destroyed, i.e., the
10540 * GPA->HVA translation will not change. However, the HVA is a user
10541 * address, i.e. its accessibility is not guaranteed, and must be
10542 * accessed via __copy_{to,from}_user().
10544 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10548 unsigned long hva, old_npages;
10549 struct kvm_memslots *slots = kvm_memslots(kvm);
10550 struct kvm_memory_slot *slot;
10552 /* Called with kvm->slots_lock held. */
10553 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10554 return ERR_PTR_USR(-EINVAL);
10556 slot = id_to_memslot(slots, id);
10558 if (slot && slot->npages)
10559 return ERR_PTR_USR(-EEXIST);
10562 * MAP_SHARED to prevent internal slot pages from being moved
10565 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10566 MAP_SHARED | MAP_ANONYMOUS, 0);
10567 if (IS_ERR((void *)hva))
10568 return (void __user *)hva;
10570 if (!slot || !slot->npages)
10573 old_npages = slot->npages;
10574 hva = slot->userspace_addr;
10577 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10578 struct kvm_userspace_memory_region m;
10580 m.slot = id | (i << 16);
10582 m.guest_phys_addr = gpa;
10583 m.userspace_addr = hva;
10584 m.memory_size = size;
10585 r = __kvm_set_memory_region(kvm, &m);
10587 return ERR_PTR_USR(r);
10591 vm_munmap(hva, old_npages * PAGE_SIZE);
10593 return (void __user *)hva;
10595 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10597 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10599 kvm_mmu_pre_destroy_vm(kvm);
10602 void kvm_arch_destroy_vm(struct kvm *kvm)
10606 if (current->mm == kvm->mm) {
10608 * Free memory regions allocated on behalf of userspace,
10609 * unless the the memory map has changed due to process exit
10612 mutex_lock(&kvm->slots_lock);
10613 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10615 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10617 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10618 mutex_unlock(&kvm->slots_lock);
10620 static_call_cond(kvm_x86_vm_destroy)(kvm);
10621 for (i = 0; i < kvm->arch.msr_filter.count; i++)
10622 kfree(kvm->arch.msr_filter.ranges[i].bitmap);
10623 kvm_pic_destroy(kvm);
10624 kvm_ioapic_destroy(kvm);
10625 kvm_free_vcpus(kvm);
10626 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10627 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10628 kvm_mmu_uninit_vm(kvm);
10629 kvm_page_track_cleanup(kvm);
10630 kvm_xen_destroy_vm(kvm);
10631 kvm_hv_destroy_vm(kvm);
10634 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10638 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10639 kvfree(slot->arch.rmap[i]);
10640 slot->arch.rmap[i] = NULL;
10645 kvfree(slot->arch.lpage_info[i - 1]);
10646 slot->arch.lpage_info[i - 1] = NULL;
10649 kvm_page_track_free_memslot(slot);
10652 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10653 unsigned long npages)
10658 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10659 * old arrays will be freed by __kvm_set_memory_region() if installing
10660 * the new memslot is successful.
10662 memset(&slot->arch, 0, sizeof(slot->arch));
10664 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10665 struct kvm_lpage_info *linfo;
10666 unsigned long ugfn;
10670 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10671 slot->base_gfn, level) + 1;
10673 slot->arch.rmap[i] =
10674 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10675 GFP_KERNEL_ACCOUNT);
10676 if (!slot->arch.rmap[i])
10681 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10685 slot->arch.lpage_info[i - 1] = linfo;
10687 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10688 linfo[0].disallow_lpage = 1;
10689 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10690 linfo[lpages - 1].disallow_lpage = 1;
10691 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10693 * If the gfn and userspace address are not aligned wrt each
10694 * other, disable large page support for this slot.
10696 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10699 for (j = 0; j < lpages; ++j)
10700 linfo[j].disallow_lpage = 1;
10704 if (kvm_page_track_create_memslot(slot, npages))
10710 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10711 kvfree(slot->arch.rmap[i]);
10712 slot->arch.rmap[i] = NULL;
10716 kvfree(slot->arch.lpage_info[i - 1]);
10717 slot->arch.lpage_info[i - 1] = NULL;
10722 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10724 struct kvm_vcpu *vcpu;
10728 * memslots->generation has been incremented.
10729 * mmio generation may have reached its maximum value.
10731 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10733 /* Force re-initialization of steal_time cache */
10734 kvm_for_each_vcpu(i, vcpu, kvm)
10735 kvm_vcpu_kick(vcpu);
10738 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10739 struct kvm_memory_slot *memslot,
10740 const struct kvm_userspace_memory_region *mem,
10741 enum kvm_mr_change change)
10743 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10744 return kvm_alloc_memslot_metadata(memslot,
10745 mem->memory_size >> PAGE_SHIFT);
10749 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10750 struct kvm_memory_slot *old,
10751 struct kvm_memory_slot *new,
10752 enum kvm_mr_change change)
10755 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
10756 * See comments below.
10758 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10762 * Dirty logging tracks sptes in 4k granularity, meaning that large
10763 * sptes have to be split. If live migration is successful, the guest
10764 * in the source machine will be destroyed and large sptes will be
10765 * created in the destination. However, if the guest continues to run
10766 * in the source machine (for example if live migration fails), small
10767 * sptes will remain around and cause bad performance.
10769 * Scan sptes if dirty logging has been stopped, dropping those
10770 * which can be collapsed into a single large-page spte. Later
10771 * page faults will create the large-page sptes.
10773 * There is no need to do this in any of the following cases:
10774 * CREATE: No dirty mappings will already exist.
10775 * MOVE/DELETE: The old mappings will already have been cleaned up by
10776 * kvm_arch_flush_shadow_memslot()
10778 if ((old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10779 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10780 kvm_mmu_zap_collapsible_sptes(kvm, new);
10783 * Enable or disable dirty logging for the slot.
10785 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old
10786 * slot have been zapped so no dirty logging updates are needed for
10788 * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible
10789 * any mappings that might be created in it will consume the
10790 * properties of the new slot and do not need to be updated here.
10792 * When PML is enabled, the kvm_x86_ops dirty logging hooks are
10793 * called to enable/disable dirty logging.
10795 * When disabling dirty logging with PML enabled, the D-bit is set
10796 * for sptes in the slot in order to prevent unnecessary GPA
10797 * logging in the PML buffer (and potential PML buffer full VMEXIT).
10798 * This guarantees leaving PML enabled for the guest's lifetime
10799 * won't have any additional overhead from PML when the guest is
10800 * running with dirty logging disabled.
10802 * When enabling dirty logging, large sptes are write-protected
10803 * so they can be split on first write. New large sptes cannot
10804 * be created for this slot until the end of the logging.
10805 * See the comments in fast_page_fault().
10806 * For small sptes, nothing is done if the dirty log is in the
10807 * initial-all-set state. Otherwise, depending on whether pml
10808 * is enabled the D-bit or the W-bit will be cleared.
10810 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10811 if (kvm_x86_ops.slot_enable_log_dirty) {
10812 static_call(kvm_x86_slot_enable_log_dirty)(kvm, new);
10815 kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10816 PG_LEVEL_2M : PG_LEVEL_4K;
10819 * If we're with initial-all-set, we don't need
10820 * to write protect any small page because
10821 * they're reported as dirty already. However
10822 * we still need to write-protect huge pages
10823 * so that the page split can happen lazily on
10824 * the first write to the huge page.
10826 kvm_mmu_slot_remove_write_access(kvm, new, level);
10829 static_call_cond(kvm_x86_slot_disable_log_dirty)(kvm, new);
10833 void kvm_arch_commit_memory_region(struct kvm *kvm,
10834 const struct kvm_userspace_memory_region *mem,
10835 struct kvm_memory_slot *old,
10836 const struct kvm_memory_slot *new,
10837 enum kvm_mr_change change)
10839 if (!kvm->arch.n_requested_mmu_pages)
10840 kvm_mmu_change_mmu_pages(kvm,
10841 kvm_mmu_calculate_default_mmu_pages(kvm));
10844 * FIXME: const-ify all uses of struct kvm_memory_slot.
10846 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
10848 /* Free the arrays associated with the old memslot. */
10849 if (change == KVM_MR_MOVE)
10850 kvm_arch_free_memslot(kvm, old);
10853 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10855 kvm_mmu_zap_all(kvm);
10858 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10859 struct kvm_memory_slot *slot)
10861 kvm_page_track_flush_slot(kvm, slot);
10864 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10866 return (is_guest_mode(vcpu) &&
10867 kvm_x86_ops.guest_apic_has_interrupt &&
10868 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
10871 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10873 if (!list_empty_careful(&vcpu->async_pf.done))
10876 if (kvm_apic_has_events(vcpu))
10879 if (vcpu->arch.pv.pv_unhalted)
10882 if (vcpu->arch.exception.pending)
10885 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10886 (vcpu->arch.nmi_pending &&
10887 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
10890 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10891 (vcpu->arch.smi_pending &&
10892 static_call(kvm_x86_smi_allowed)(vcpu, false)))
10895 if (kvm_arch_interrupt_allowed(vcpu) &&
10896 (kvm_cpu_has_interrupt(vcpu) ||
10897 kvm_guest_apic_has_interrupt(vcpu)))
10900 if (kvm_hv_has_stimer_pending(vcpu))
10903 if (is_guest_mode(vcpu) &&
10904 kvm_x86_ops.nested_ops->hv_timer_pending &&
10905 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10911 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10913 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10916 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10918 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10921 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10922 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10923 kvm_test_request(KVM_REQ_EVENT, vcpu))
10926 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
10932 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10934 return vcpu->arch.preempted_in_kernel;
10937 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10939 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10942 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10944 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
10947 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10949 /* Can't read the RIP when guest state is protected, just return 0 */
10950 if (vcpu->arch.guest_state_protected)
10953 if (is_64_bit_mode(vcpu))
10954 return kvm_rip_read(vcpu);
10955 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10956 kvm_rip_read(vcpu));
10958 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10960 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10962 return kvm_get_linear_rip(vcpu) == linear_rip;
10964 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10966 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10968 unsigned long rflags;
10970 rflags = static_call(kvm_x86_get_rflags)(vcpu);
10971 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10972 rflags &= ~X86_EFLAGS_TF;
10975 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10977 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10979 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10980 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10981 rflags |= X86_EFLAGS_TF;
10982 static_call(kvm_x86_set_rflags)(vcpu, rflags);
10985 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10987 __kvm_set_rflags(vcpu, rflags);
10988 kvm_make_request(KVM_REQ_EVENT, vcpu);
10990 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10992 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10996 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11000 r = kvm_mmu_reload(vcpu);
11004 if (!vcpu->arch.mmu->direct_map &&
11005 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11008 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11011 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11013 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11015 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11018 static inline u32 kvm_async_pf_next_probe(u32 key)
11020 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11023 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11025 u32 key = kvm_async_pf_hash_fn(gfn);
11027 while (vcpu->arch.apf.gfns[key] != ~0)
11028 key = kvm_async_pf_next_probe(key);
11030 vcpu->arch.apf.gfns[key] = gfn;
11033 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11036 u32 key = kvm_async_pf_hash_fn(gfn);
11038 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11039 (vcpu->arch.apf.gfns[key] != gfn &&
11040 vcpu->arch.apf.gfns[key] != ~0); i++)
11041 key = kvm_async_pf_next_probe(key);
11046 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11048 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11051 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11055 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11057 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11061 vcpu->arch.apf.gfns[i] = ~0;
11063 j = kvm_async_pf_next_probe(j);
11064 if (vcpu->arch.apf.gfns[j] == ~0)
11066 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11068 * k lies cyclically in ]i,j]
11070 * |....j i.k.| or |.k..j i...|
11072 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11073 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11078 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11080 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11082 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11086 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11088 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11090 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11091 &token, offset, sizeof(token));
11094 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11096 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11099 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11100 &val, offset, sizeof(val)))
11106 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11108 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11111 if (!kvm_pv_async_pf_enabled(vcpu) ||
11112 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11118 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11120 if (unlikely(!lapic_in_kernel(vcpu) ||
11121 kvm_event_needs_reinjection(vcpu) ||
11122 vcpu->arch.exception.pending))
11125 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11129 * If interrupts are off we cannot even use an artificial
11132 return kvm_arch_interrupt_allowed(vcpu);
11135 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11136 struct kvm_async_pf *work)
11138 struct x86_exception fault;
11140 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11141 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11143 if (kvm_can_deliver_async_pf(vcpu) &&
11144 !apf_put_user_notpresent(vcpu)) {
11145 fault.vector = PF_VECTOR;
11146 fault.error_code_valid = true;
11147 fault.error_code = 0;
11148 fault.nested_page_fault = false;
11149 fault.address = work->arch.token;
11150 fault.async_page_fault = true;
11151 kvm_inject_page_fault(vcpu, &fault);
11155 * It is not possible to deliver a paravirtualized asynchronous
11156 * page fault, but putting the guest in an artificial halt state
11157 * can be beneficial nevertheless: if an interrupt arrives, we
11158 * can deliver it timely and perhaps the guest will schedule
11159 * another process. When the instruction that triggered a page
11160 * fault is retried, hopefully the page will be ready in the host.
11162 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11167 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11168 struct kvm_async_pf *work)
11170 struct kvm_lapic_irq irq = {
11171 .delivery_mode = APIC_DM_FIXED,
11172 .vector = vcpu->arch.apf.vec
11175 if (work->wakeup_all)
11176 work->arch.token = ~0; /* broadcast wakeup */
11178 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11179 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11181 if ((work->wakeup_all || work->notpresent_injected) &&
11182 kvm_pv_async_pf_enabled(vcpu) &&
11183 !apf_put_user_ready(vcpu, work->arch.token)) {
11184 vcpu->arch.apf.pageready_pending = true;
11185 kvm_apic_set_irq(vcpu, &irq, NULL);
11188 vcpu->arch.apf.halted = false;
11189 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11192 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11194 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11195 if (!vcpu->arch.apf.pageready_pending)
11196 kvm_vcpu_kick(vcpu);
11199 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11201 if (!kvm_pv_async_pf_enabled(vcpu))
11204 return apf_pageready_slot_free(vcpu);
11207 void kvm_arch_start_assignment(struct kvm *kvm)
11209 atomic_inc(&kvm->arch.assigned_device_count);
11211 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11213 void kvm_arch_end_assignment(struct kvm *kvm)
11215 atomic_dec(&kvm->arch.assigned_device_count);
11217 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11219 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11221 return atomic_read(&kvm->arch.assigned_device_count);
11223 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11225 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11227 atomic_inc(&kvm->arch.noncoherent_dma_count);
11229 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11231 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11233 atomic_dec(&kvm->arch.noncoherent_dma_count);
11235 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11237 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11239 return atomic_read(&kvm->arch.noncoherent_dma_count);
11241 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11243 bool kvm_arch_has_irq_bypass(void)
11248 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11249 struct irq_bypass_producer *prod)
11251 struct kvm_kernel_irqfd *irqfd =
11252 container_of(cons, struct kvm_kernel_irqfd, consumer);
11255 irqfd->producer = prod;
11256 kvm_arch_start_assignment(irqfd->kvm);
11257 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11258 prod->irq, irqfd->gsi, 1);
11261 kvm_arch_end_assignment(irqfd->kvm);
11266 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11267 struct irq_bypass_producer *prod)
11270 struct kvm_kernel_irqfd *irqfd =
11271 container_of(cons, struct kvm_kernel_irqfd, consumer);
11273 WARN_ON(irqfd->producer != prod);
11274 irqfd->producer = NULL;
11277 * When producer of consumer is unregistered, we change back to
11278 * remapped mode, so we can re-use the current implementation
11279 * when the irq is masked/disabled or the consumer side (KVM
11280 * int this case doesn't want to receive the interrupts.
11282 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11284 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11285 " fails: %d\n", irqfd->consumer.token, ret);
11287 kvm_arch_end_assignment(irqfd->kvm);
11290 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11291 uint32_t guest_irq, bool set)
11293 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11296 bool kvm_vector_hashing_enabled(void)
11298 return vector_hashing;
11301 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11303 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11305 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11308 int kvm_spec_ctrl_test_value(u64 value)
11311 * test that setting IA32_SPEC_CTRL to given value
11312 * is allowed by the host processor
11316 unsigned long flags;
11319 local_irq_save(flags);
11321 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11323 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11326 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11328 local_irq_restore(flags);
11332 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11334 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11336 struct x86_exception fault;
11337 u32 access = error_code &
11338 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11340 if (!(error_code & PFERR_PRESENT_MASK) ||
11341 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11343 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11344 * tables probably do not match the TLB. Just proceed
11345 * with the error code that the processor gave.
11347 fault.vector = PF_VECTOR;
11348 fault.error_code_valid = true;
11349 fault.error_code = error_code;
11350 fault.nested_page_fault = false;
11351 fault.address = gva;
11353 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11355 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11358 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11359 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11360 * indicates whether exit to userspace is needed.
11362 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11363 struct x86_exception *e)
11365 if (r == X86EMUL_PROPAGATE_FAULT) {
11366 kvm_inject_emulated_page_fault(vcpu, e);
11371 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11372 * while handling a VMX instruction KVM could've handled the request
11373 * correctly by exiting to userspace and performing I/O but there
11374 * doesn't seem to be a real use-case behind such requests, just return
11375 * KVM_EXIT_INTERNAL_ERROR for now.
11377 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11378 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11379 vcpu->run->internal.ndata = 0;
11383 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11385 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11388 struct x86_exception e;
11390 unsigned long roots_to_free = 0;
11397 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11398 if (r != X86EMUL_CONTINUE)
11399 return kvm_handle_memory_failure(vcpu, r, &e);
11401 if (operand.pcid >> 12 != 0) {
11402 kvm_inject_gp(vcpu, 0);
11406 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11409 case INVPCID_TYPE_INDIV_ADDR:
11410 if ((!pcid_enabled && (operand.pcid != 0)) ||
11411 is_noncanonical_address(operand.gla, vcpu)) {
11412 kvm_inject_gp(vcpu, 0);
11415 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11416 return kvm_skip_emulated_instruction(vcpu);
11418 case INVPCID_TYPE_SINGLE_CTXT:
11419 if (!pcid_enabled && (operand.pcid != 0)) {
11420 kvm_inject_gp(vcpu, 0);
11424 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11425 kvm_mmu_sync_roots(vcpu);
11426 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11429 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11430 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11432 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11434 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11436 * If neither the current cr3 nor any of the prev_roots use the
11437 * given PCID, then nothing needs to be done here because a
11438 * resync will happen anyway before switching to any other CR3.
11441 return kvm_skip_emulated_instruction(vcpu);
11443 case INVPCID_TYPE_ALL_NON_GLOBAL:
11445 * Currently, KVM doesn't mark global entries in the shadow
11446 * page tables, so a non-global flush just degenerates to a
11447 * global flush. If needed, we could optimize this later by
11448 * keeping track of global entries in shadow page tables.
11452 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11453 kvm_mmu_unload(vcpu);
11454 return kvm_skip_emulated_instruction(vcpu);
11457 BUG(); /* We have already checked above that type <= 3 */
11460 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11462 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11464 struct kvm_run *run = vcpu->run;
11465 struct kvm_mmio_fragment *frag;
11468 BUG_ON(!vcpu->mmio_needed);
11470 /* Complete previous fragment */
11471 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11472 len = min(8u, frag->len);
11473 if (!vcpu->mmio_is_write)
11474 memcpy(frag->data, run->mmio.data, len);
11476 if (frag->len <= 8) {
11477 /* Switch to the next fragment. */
11479 vcpu->mmio_cur_fragment++;
11481 /* Go forward to the next mmio piece. */
11487 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11488 vcpu->mmio_needed = 0;
11490 // VMG change, at this point, we're always done
11491 // RIP has already been advanced
11495 // More MMIO is needed
11496 run->mmio.phys_addr = frag->gpa;
11497 run->mmio.len = min(8u, frag->len);
11498 run->mmio.is_write = vcpu->mmio_is_write;
11499 if (run->mmio.is_write)
11500 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11501 run->exit_reason = KVM_EXIT_MMIO;
11503 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11508 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11512 struct kvm_mmio_fragment *frag;
11517 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11518 if (handled == bytes)
11525 /*TODO: Check if need to increment number of frags */
11526 frag = vcpu->mmio_fragments;
11527 vcpu->mmio_nr_fragments = 1;
11532 vcpu->mmio_needed = 1;
11533 vcpu->mmio_cur_fragment = 0;
11535 vcpu->run->mmio.phys_addr = gpa;
11536 vcpu->run->mmio.len = min(8u, frag->len);
11537 vcpu->run->mmio.is_write = 1;
11538 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11539 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11541 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11545 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11547 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11551 struct kvm_mmio_fragment *frag;
11556 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11557 if (handled == bytes)
11564 /*TODO: Check if need to increment number of frags */
11565 frag = vcpu->mmio_fragments;
11566 vcpu->mmio_nr_fragments = 1;
11571 vcpu->mmio_needed = 1;
11572 vcpu->mmio_cur_fragment = 0;
11574 vcpu->run->mmio.phys_addr = gpa;
11575 vcpu->run->mmio.len = min(8u, frag->len);
11576 vcpu->run->mmio.is_write = 0;
11577 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11579 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11583 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11585 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11587 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11588 vcpu->arch.pio.count * vcpu->arch.pio.size);
11589 vcpu->arch.pio.count = 0;
11594 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11595 unsigned int port, void *data, unsigned int count)
11599 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11604 vcpu->arch.pio.count = 0;
11609 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11610 unsigned int port, void *data, unsigned int count)
11614 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11617 vcpu->arch.pio.count = 0;
11619 vcpu->arch.guest_ins_data = data;
11620 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11626 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11627 unsigned int port, void *data, unsigned int count,
11630 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11631 : kvm_sev_es_outs(vcpu, size, port, data, count);
11633 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11635 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11636 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11637 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11638 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11639 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11640 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11641 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11642 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11643 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11644 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11645 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11646 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11647 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11648 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11649 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11650 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11651 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11652 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11653 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11654 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11655 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11656 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11657 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11658 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11659 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11660 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11661 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);