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_x86_msr_filter *msr_filter;
1530 struct msr_bitmap_range *ranges;
1531 struct kvm *kvm = vcpu->kvm;
1536 /* x2APIC MSRs do not support filtering. */
1537 if (index >= 0x800 && index <= 0x8ff)
1540 idx = srcu_read_lock(&kvm->srcu);
1542 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1548 allowed = msr_filter->default_allow;
1549 ranges = msr_filter->ranges;
1551 for (i = 0; i < msr_filter->count; i++) {
1552 u32 start = ranges[i].base;
1553 u32 end = start + ranges[i].nmsrs;
1554 u32 flags = ranges[i].flags;
1555 unsigned long *bitmap = ranges[i].bitmap;
1557 if ((index >= start) && (index < end) && (flags & type)) {
1558 allowed = !!test_bit(index - start, bitmap);
1564 srcu_read_unlock(&kvm->srcu, idx);
1568 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1571 * Write @data into the MSR specified by @index. Select MSR specific fault
1572 * checks are bypassed if @host_initiated is %true.
1573 * Returns 0 on success, non-0 otherwise.
1574 * Assumes vcpu_load() was already called.
1576 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1577 bool host_initiated)
1579 struct msr_data msr;
1581 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1582 return KVM_MSR_RET_FILTERED;
1587 case MSR_KERNEL_GS_BASE:
1590 if (is_noncanonical_address(data, vcpu))
1593 case MSR_IA32_SYSENTER_EIP:
1594 case MSR_IA32_SYSENTER_ESP:
1596 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1597 * non-canonical address is written on Intel but not on
1598 * AMD (which ignores the top 32-bits, because it does
1599 * not implement 64-bit SYSENTER).
1601 * 64-bit code should hence be able to write a non-canonical
1602 * value on AMD. Making the address canonical ensures that
1603 * vmentry does not fail on Intel after writing a non-canonical
1604 * value, and that something deterministic happens if the guest
1605 * invokes 64-bit SYSENTER.
1607 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1612 msr.host_initiated = host_initiated;
1614 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1617 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1618 u32 index, u64 data, bool host_initiated)
1620 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1622 if (ret == KVM_MSR_RET_INVALID)
1623 if (kvm_msr_ignored_check(vcpu, index, data, true))
1630 * Read the MSR specified by @index into @data. Select MSR specific fault
1631 * checks are bypassed if @host_initiated is %true.
1632 * Returns 0 on success, non-0 otherwise.
1633 * Assumes vcpu_load() was already called.
1635 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1636 bool host_initiated)
1638 struct msr_data msr;
1641 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1642 return KVM_MSR_RET_FILTERED;
1645 msr.host_initiated = host_initiated;
1647 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1653 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1654 u32 index, u64 *data, bool host_initiated)
1656 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1658 if (ret == KVM_MSR_RET_INVALID) {
1659 /* Unconditionally clear *data for simplicity */
1661 if (kvm_msr_ignored_check(vcpu, index, 0, false))
1668 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1670 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1672 EXPORT_SYMBOL_GPL(kvm_get_msr);
1674 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1676 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1678 EXPORT_SYMBOL_GPL(kvm_set_msr);
1680 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1682 int err = vcpu->run->msr.error;
1684 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1685 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1688 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1691 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1693 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1696 static u64 kvm_msr_reason(int r)
1699 case KVM_MSR_RET_INVALID:
1700 return KVM_MSR_EXIT_REASON_UNKNOWN;
1701 case KVM_MSR_RET_FILTERED:
1702 return KVM_MSR_EXIT_REASON_FILTER;
1704 return KVM_MSR_EXIT_REASON_INVAL;
1708 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1709 u32 exit_reason, u64 data,
1710 int (*completion)(struct kvm_vcpu *vcpu),
1713 u64 msr_reason = kvm_msr_reason(r);
1715 /* Check if the user wanted to know about this MSR fault */
1716 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1719 vcpu->run->exit_reason = exit_reason;
1720 vcpu->run->msr.error = 0;
1721 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1722 vcpu->run->msr.reason = msr_reason;
1723 vcpu->run->msr.index = index;
1724 vcpu->run->msr.data = data;
1725 vcpu->arch.complete_userspace_io = completion;
1730 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1732 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1733 complete_emulated_rdmsr, r);
1736 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1738 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1739 complete_emulated_wrmsr, r);
1742 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1744 u32 ecx = kvm_rcx_read(vcpu);
1748 r = kvm_get_msr(vcpu, ecx, &data);
1750 /* MSR read failed? See if we should ask user space */
1751 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1752 /* Bounce to user space */
1757 trace_kvm_msr_read(ecx, data);
1759 kvm_rax_write(vcpu, data & -1u);
1760 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1762 trace_kvm_msr_read_ex(ecx);
1765 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1767 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1769 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1771 u32 ecx = kvm_rcx_read(vcpu);
1772 u64 data = kvm_read_edx_eax(vcpu);
1775 r = kvm_set_msr(vcpu, ecx, data);
1777 /* MSR write failed? See if we should ask user space */
1778 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1779 /* Bounce to user space */
1782 /* Signal all other negative errors to userspace */
1787 trace_kvm_msr_write(ecx, data);
1789 trace_kvm_msr_write_ex(ecx, data);
1791 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1793 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1795 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1797 xfer_to_guest_mode_prepare();
1798 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1799 xfer_to_guest_mode_work_pending();
1803 * The fast path for frequent and performance sensitive wrmsr emulation,
1804 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1805 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1806 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1807 * other cases which must be called after interrupts are enabled on the host.
1809 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1811 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1814 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1815 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1816 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1817 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1820 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1821 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1822 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1823 trace_kvm_apic_write(APIC_ICR, (u32)data);
1830 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1832 if (!kvm_can_use_hv_timer(vcpu))
1835 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1839 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1841 u32 msr = kvm_rcx_read(vcpu);
1843 fastpath_t ret = EXIT_FASTPATH_NONE;
1846 case APIC_BASE_MSR + (APIC_ICR >> 4):
1847 data = kvm_read_edx_eax(vcpu);
1848 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1849 kvm_skip_emulated_instruction(vcpu);
1850 ret = EXIT_FASTPATH_EXIT_HANDLED;
1853 case MSR_IA32_TSCDEADLINE:
1854 data = kvm_read_edx_eax(vcpu);
1855 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1856 kvm_skip_emulated_instruction(vcpu);
1857 ret = EXIT_FASTPATH_REENTER_GUEST;
1864 if (ret != EXIT_FASTPATH_NONE)
1865 trace_kvm_msr_write(msr, data);
1869 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1872 * Adapt set_msr() to msr_io()'s calling convention
1874 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1876 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1879 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1881 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1884 #ifdef CONFIG_X86_64
1885 struct pvclock_clock {
1895 struct pvclock_gtod_data {
1898 struct pvclock_clock clock; /* extract of a clocksource struct */
1899 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1905 static struct pvclock_gtod_data pvclock_gtod_data;
1907 static void update_pvclock_gtod(struct timekeeper *tk)
1909 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1911 write_seqcount_begin(&vdata->seq);
1913 /* copy pvclock gtod data */
1914 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1915 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1916 vdata->clock.mask = tk->tkr_mono.mask;
1917 vdata->clock.mult = tk->tkr_mono.mult;
1918 vdata->clock.shift = tk->tkr_mono.shift;
1919 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1920 vdata->clock.offset = tk->tkr_mono.base;
1922 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1923 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1924 vdata->raw_clock.mask = tk->tkr_raw.mask;
1925 vdata->raw_clock.mult = tk->tkr_raw.mult;
1926 vdata->raw_clock.shift = tk->tkr_raw.shift;
1927 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1928 vdata->raw_clock.offset = tk->tkr_raw.base;
1930 vdata->wall_time_sec = tk->xtime_sec;
1932 vdata->offs_boot = tk->offs_boot;
1934 write_seqcount_end(&vdata->seq);
1937 static s64 get_kvmclock_base_ns(void)
1939 /* Count up from boot time, but with the frequency of the raw clock. */
1940 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1943 static s64 get_kvmclock_base_ns(void)
1945 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1946 return ktime_get_boottime_ns();
1950 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
1954 struct pvclock_wall_clock wc;
1961 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1966 ++version; /* first time write, random junk */
1970 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1974 * The guest calculates current wall clock time by adding
1975 * system time (updated by kvm_guest_time_update below) to the
1976 * wall clock specified here. We do the reverse here.
1978 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1980 wc.nsec = do_div(wall_nsec, 1000000000);
1981 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1982 wc.version = version;
1984 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1987 wc_sec_hi = wall_nsec >> 32;
1988 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
1989 &wc_sec_hi, sizeof(wc_sec_hi));
1993 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1996 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
1997 bool old_msr, bool host_initiated)
1999 struct kvm_arch *ka = &vcpu->kvm->arch;
2001 if (vcpu->vcpu_id == 0 && !host_initiated) {
2002 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2003 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2005 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2008 vcpu->arch.time = system_time;
2009 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2011 /* we verify if the enable bit is set... */
2012 vcpu->arch.pv_time_enabled = false;
2013 if (!(system_time & 1))
2016 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2017 &vcpu->arch.pv_time, system_time & ~1ULL,
2018 sizeof(struct pvclock_vcpu_time_info)))
2019 vcpu->arch.pv_time_enabled = true;
2024 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2026 do_shl32_div32(dividend, divisor);
2030 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2031 s8 *pshift, u32 *pmultiplier)
2039 scaled64 = scaled_hz;
2040 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2045 tps32 = (uint32_t)tps64;
2046 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2047 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2055 *pmultiplier = div_frac(scaled64, tps32);
2058 #ifdef CONFIG_X86_64
2059 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2062 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2063 static unsigned long max_tsc_khz;
2065 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2067 u64 v = (u64)khz * (1000000 + ppm);
2072 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2076 /* Guest TSC same frequency as host TSC? */
2078 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2082 /* TSC scaling supported? */
2083 if (!kvm_has_tsc_control) {
2084 if (user_tsc_khz > tsc_khz) {
2085 vcpu->arch.tsc_catchup = 1;
2086 vcpu->arch.tsc_always_catchup = 1;
2089 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2094 /* TSC scaling required - calculate ratio */
2095 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2096 user_tsc_khz, tsc_khz);
2098 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2099 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2104 vcpu->arch.tsc_scaling_ratio = ratio;
2108 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2110 u32 thresh_lo, thresh_hi;
2111 int use_scaling = 0;
2113 /* tsc_khz can be zero if TSC calibration fails */
2114 if (user_tsc_khz == 0) {
2115 /* set tsc_scaling_ratio to a safe value */
2116 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2120 /* Compute a scale to convert nanoseconds in TSC cycles */
2121 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2122 &vcpu->arch.virtual_tsc_shift,
2123 &vcpu->arch.virtual_tsc_mult);
2124 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2127 * Compute the variation in TSC rate which is acceptable
2128 * within the range of tolerance and decide if the
2129 * rate being applied is within that bounds of the hardware
2130 * rate. If so, no scaling or compensation need be done.
2132 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2133 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2134 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2135 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2138 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2141 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2143 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2144 vcpu->arch.virtual_tsc_mult,
2145 vcpu->arch.virtual_tsc_shift);
2146 tsc += vcpu->arch.this_tsc_write;
2150 static inline int gtod_is_based_on_tsc(int mode)
2152 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2155 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2157 #ifdef CONFIG_X86_64
2159 struct kvm_arch *ka = &vcpu->kvm->arch;
2160 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2162 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2163 atomic_read(&vcpu->kvm->online_vcpus));
2166 * Once the masterclock is enabled, always perform request in
2167 * order to update it.
2169 * In order to enable masterclock, the host clocksource must be TSC
2170 * and the vcpus need to have matched TSCs. When that happens,
2171 * perform request to enable masterclock.
2173 if (ka->use_master_clock ||
2174 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2175 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2177 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2178 atomic_read(&vcpu->kvm->online_vcpus),
2179 ka->use_master_clock, gtod->clock.vclock_mode);
2184 * Multiply tsc by a fixed point number represented by ratio.
2186 * The most significant 64-N bits (mult) of ratio represent the
2187 * integral part of the fixed point number; the remaining N bits
2188 * (frac) represent the fractional part, ie. ratio represents a fixed
2189 * point number (mult + frac * 2^(-N)).
2191 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2193 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2195 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2198 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2201 u64 ratio = vcpu->arch.tsc_scaling_ratio;
2203 if (ratio != kvm_default_tsc_scaling_ratio)
2204 _tsc = __scale_tsc(ratio, tsc);
2208 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2210 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2214 tsc = kvm_scale_tsc(vcpu, rdtsc());
2216 return target_tsc - tsc;
2219 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2221 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2223 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2225 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2227 vcpu->arch.l1_tsc_offset = offset;
2228 vcpu->arch.tsc_offset = static_call(kvm_x86_write_l1_tsc_offset)(vcpu, offset);
2231 static inline bool kvm_check_tsc_unstable(void)
2233 #ifdef CONFIG_X86_64
2235 * TSC is marked unstable when we're running on Hyper-V,
2236 * 'TSC page' clocksource is good.
2238 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2241 return check_tsc_unstable();
2244 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2246 struct kvm *kvm = vcpu->kvm;
2247 u64 offset, ns, elapsed;
2248 unsigned long flags;
2250 bool already_matched;
2251 bool synchronizing = false;
2253 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2254 offset = kvm_compute_tsc_offset(vcpu, data);
2255 ns = get_kvmclock_base_ns();
2256 elapsed = ns - kvm->arch.last_tsc_nsec;
2258 if (vcpu->arch.virtual_tsc_khz) {
2261 * detection of vcpu initialization -- need to sync
2262 * with other vCPUs. This particularly helps to keep
2263 * kvm_clock stable after CPU hotplug
2265 synchronizing = true;
2267 u64 tsc_exp = kvm->arch.last_tsc_write +
2268 nsec_to_cycles(vcpu, elapsed);
2269 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2271 * Special case: TSC write with a small delta (1 second)
2272 * of virtual cycle time against real time is
2273 * interpreted as an attempt to synchronize the CPU.
2275 synchronizing = data < tsc_exp + tsc_hz &&
2276 data + tsc_hz > tsc_exp;
2281 * For a reliable TSC, we can match TSC offsets, and for an unstable
2282 * TSC, we add elapsed time in this computation. We could let the
2283 * compensation code attempt to catch up if we fall behind, but
2284 * it's better to try to match offsets from the beginning.
2286 if (synchronizing &&
2287 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2288 if (!kvm_check_tsc_unstable()) {
2289 offset = kvm->arch.cur_tsc_offset;
2291 u64 delta = nsec_to_cycles(vcpu, elapsed);
2293 offset = kvm_compute_tsc_offset(vcpu, data);
2296 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2299 * We split periods of matched TSC writes into generations.
2300 * For each generation, we track the original measured
2301 * nanosecond time, offset, and write, so if TSCs are in
2302 * sync, we can match exact offset, and if not, we can match
2303 * exact software computation in compute_guest_tsc()
2305 * These values are tracked in kvm->arch.cur_xxx variables.
2307 kvm->arch.cur_tsc_generation++;
2308 kvm->arch.cur_tsc_nsec = ns;
2309 kvm->arch.cur_tsc_write = data;
2310 kvm->arch.cur_tsc_offset = offset;
2315 * We also track th most recent recorded KHZ, write and time to
2316 * allow the matching interval to be extended at each write.
2318 kvm->arch.last_tsc_nsec = ns;
2319 kvm->arch.last_tsc_write = data;
2320 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2322 vcpu->arch.last_guest_tsc = data;
2324 /* Keep track of which generation this VCPU has synchronized to */
2325 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2326 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2327 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2329 kvm_vcpu_write_tsc_offset(vcpu, offset);
2330 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2332 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2334 kvm->arch.nr_vcpus_matched_tsc = 0;
2335 } else if (!already_matched) {
2336 kvm->arch.nr_vcpus_matched_tsc++;
2339 kvm_track_tsc_matching(vcpu);
2340 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2343 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2346 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2347 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2350 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2352 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2353 WARN_ON(adjustment < 0);
2354 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2355 adjust_tsc_offset_guest(vcpu, adjustment);
2358 #ifdef CONFIG_X86_64
2360 static u64 read_tsc(void)
2362 u64 ret = (u64)rdtsc_ordered();
2363 u64 last = pvclock_gtod_data.clock.cycle_last;
2365 if (likely(ret >= last))
2369 * GCC likes to generate cmov here, but this branch is extremely
2370 * predictable (it's just a function of time and the likely is
2371 * very likely) and there's a data dependence, so force GCC
2372 * to generate a branch instead. I don't barrier() because
2373 * we don't actually need a barrier, and if this function
2374 * ever gets inlined it will generate worse code.
2380 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2386 switch (clock->vclock_mode) {
2387 case VDSO_CLOCKMODE_HVCLOCK:
2388 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2390 if (tsc_pg_val != U64_MAX) {
2391 /* TSC page valid */
2392 *mode = VDSO_CLOCKMODE_HVCLOCK;
2393 v = (tsc_pg_val - clock->cycle_last) &
2396 /* TSC page invalid */
2397 *mode = VDSO_CLOCKMODE_NONE;
2400 case VDSO_CLOCKMODE_TSC:
2401 *mode = VDSO_CLOCKMODE_TSC;
2402 *tsc_timestamp = read_tsc();
2403 v = (*tsc_timestamp - clock->cycle_last) &
2407 *mode = VDSO_CLOCKMODE_NONE;
2410 if (*mode == VDSO_CLOCKMODE_NONE)
2411 *tsc_timestamp = v = 0;
2413 return v * clock->mult;
2416 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2418 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2424 seq = read_seqcount_begin(>od->seq);
2425 ns = gtod->raw_clock.base_cycles;
2426 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2427 ns >>= gtod->raw_clock.shift;
2428 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2429 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2435 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2437 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2443 seq = read_seqcount_begin(>od->seq);
2444 ts->tv_sec = gtod->wall_time_sec;
2445 ns = gtod->clock.base_cycles;
2446 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2447 ns >>= gtod->clock.shift;
2448 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2450 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2456 /* returns true if host is using TSC based clocksource */
2457 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2459 /* checked again under seqlock below */
2460 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2463 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2467 /* returns true if host is using TSC based clocksource */
2468 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2471 /* checked again under seqlock below */
2472 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2475 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2481 * Assuming a stable TSC across physical CPUS, and a stable TSC
2482 * across virtual CPUs, the following condition is possible.
2483 * Each numbered line represents an event visible to both
2484 * CPUs at the next numbered event.
2486 * "timespecX" represents host monotonic time. "tscX" represents
2489 * VCPU0 on CPU0 | VCPU1 on CPU1
2491 * 1. read timespec0,tsc0
2492 * 2. | timespec1 = timespec0 + N
2494 * 3. transition to guest | transition to guest
2495 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2496 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2497 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2499 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2502 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2504 * - 0 < N - M => M < N
2506 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2507 * always the case (the difference between two distinct xtime instances
2508 * might be smaller then the difference between corresponding TSC reads,
2509 * when updating guest vcpus pvclock areas).
2511 * To avoid that problem, do not allow visibility of distinct
2512 * system_timestamp/tsc_timestamp values simultaneously: use a master
2513 * copy of host monotonic time values. Update that master copy
2516 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2520 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2522 #ifdef CONFIG_X86_64
2523 struct kvm_arch *ka = &kvm->arch;
2525 bool host_tsc_clocksource, vcpus_matched;
2527 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2528 atomic_read(&kvm->online_vcpus));
2531 * If the host uses TSC clock, then passthrough TSC as stable
2534 host_tsc_clocksource = kvm_get_time_and_clockread(
2535 &ka->master_kernel_ns,
2536 &ka->master_cycle_now);
2538 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2539 && !ka->backwards_tsc_observed
2540 && !ka->boot_vcpu_runs_old_kvmclock;
2542 if (ka->use_master_clock)
2543 atomic_set(&kvm_guest_has_master_clock, 1);
2545 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2546 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2551 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2553 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2556 static void kvm_gen_update_masterclock(struct kvm *kvm)
2558 #ifdef CONFIG_X86_64
2560 struct kvm_vcpu *vcpu;
2561 struct kvm_arch *ka = &kvm->arch;
2563 kvm_hv_invalidate_tsc_page(kvm);
2565 spin_lock(&ka->pvclock_gtod_sync_lock);
2566 kvm_make_mclock_inprogress_request(kvm);
2567 /* no guest entries from this point */
2568 pvclock_update_vm_gtod_copy(kvm);
2570 kvm_for_each_vcpu(i, vcpu, kvm)
2571 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2573 /* guest entries allowed */
2574 kvm_for_each_vcpu(i, vcpu, kvm)
2575 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2577 spin_unlock(&ka->pvclock_gtod_sync_lock);
2581 u64 get_kvmclock_ns(struct kvm *kvm)
2583 struct kvm_arch *ka = &kvm->arch;
2584 struct pvclock_vcpu_time_info hv_clock;
2587 spin_lock(&ka->pvclock_gtod_sync_lock);
2588 if (!ka->use_master_clock) {
2589 spin_unlock(&ka->pvclock_gtod_sync_lock);
2590 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2593 hv_clock.tsc_timestamp = ka->master_cycle_now;
2594 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2595 spin_unlock(&ka->pvclock_gtod_sync_lock);
2597 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2600 if (__this_cpu_read(cpu_tsc_khz)) {
2601 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2602 &hv_clock.tsc_shift,
2603 &hv_clock.tsc_to_system_mul);
2604 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2606 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2613 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2614 struct gfn_to_hva_cache *cache,
2615 unsigned int offset)
2617 struct kvm_vcpu_arch *vcpu = &v->arch;
2618 struct pvclock_vcpu_time_info guest_hv_clock;
2620 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2621 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2624 /* This VCPU is paused, but it's legal for a guest to read another
2625 * VCPU's kvmclock, so we really have to follow the specification where
2626 * it says that version is odd if data is being modified, and even after
2629 * Version field updates must be kept separate. This is because
2630 * kvm_write_guest_cached might use a "rep movs" instruction, and
2631 * writes within a string instruction are weakly ordered. So there
2632 * are three writes overall.
2634 * As a small optimization, only write the version field in the first
2635 * and third write. The vcpu->pv_time cache is still valid, because the
2636 * version field is the first in the struct.
2638 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2640 if (guest_hv_clock.version & 1)
2641 ++guest_hv_clock.version; /* first time write, random junk */
2643 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2644 kvm_write_guest_offset_cached(v->kvm, cache,
2645 &vcpu->hv_clock, offset,
2646 sizeof(vcpu->hv_clock.version));
2650 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2651 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2653 if (vcpu->pvclock_set_guest_stopped_request) {
2654 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2655 vcpu->pvclock_set_guest_stopped_request = false;
2658 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2660 kvm_write_guest_offset_cached(v->kvm, cache,
2661 &vcpu->hv_clock, offset,
2662 sizeof(vcpu->hv_clock));
2666 vcpu->hv_clock.version++;
2667 kvm_write_guest_offset_cached(v->kvm, cache,
2668 &vcpu->hv_clock, offset,
2669 sizeof(vcpu->hv_clock.version));
2672 static int kvm_guest_time_update(struct kvm_vcpu *v)
2674 unsigned long flags, tgt_tsc_khz;
2675 struct kvm_vcpu_arch *vcpu = &v->arch;
2676 struct kvm_arch *ka = &v->kvm->arch;
2678 u64 tsc_timestamp, host_tsc;
2680 bool use_master_clock;
2686 * If the host uses TSC clock, then passthrough TSC as stable
2689 spin_lock(&ka->pvclock_gtod_sync_lock);
2690 use_master_clock = ka->use_master_clock;
2691 if (use_master_clock) {
2692 host_tsc = ka->master_cycle_now;
2693 kernel_ns = ka->master_kernel_ns;
2695 spin_unlock(&ka->pvclock_gtod_sync_lock);
2697 /* Keep irq disabled to prevent changes to the clock */
2698 local_irq_save(flags);
2699 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2700 if (unlikely(tgt_tsc_khz == 0)) {
2701 local_irq_restore(flags);
2702 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2705 if (!use_master_clock) {
2707 kernel_ns = get_kvmclock_base_ns();
2710 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2713 * We may have to catch up the TSC to match elapsed wall clock
2714 * time for two reasons, even if kvmclock is used.
2715 * 1) CPU could have been running below the maximum TSC rate
2716 * 2) Broken TSC compensation resets the base at each VCPU
2717 * entry to avoid unknown leaps of TSC even when running
2718 * again on the same CPU. This may cause apparent elapsed
2719 * time to disappear, and the guest to stand still or run
2722 if (vcpu->tsc_catchup) {
2723 u64 tsc = compute_guest_tsc(v, kernel_ns);
2724 if (tsc > tsc_timestamp) {
2725 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2726 tsc_timestamp = tsc;
2730 local_irq_restore(flags);
2732 /* With all the info we got, fill in the values */
2734 if (kvm_has_tsc_control)
2735 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2737 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2738 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2739 &vcpu->hv_clock.tsc_shift,
2740 &vcpu->hv_clock.tsc_to_system_mul);
2741 vcpu->hw_tsc_khz = tgt_tsc_khz;
2744 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2745 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2746 vcpu->last_guest_tsc = tsc_timestamp;
2748 /* If the host uses TSC clocksource, then it is stable */
2750 if (use_master_clock)
2751 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2753 vcpu->hv_clock.flags = pvclock_flags;
2755 if (vcpu->pv_time_enabled)
2756 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2757 if (vcpu->xen.vcpu_info_set)
2758 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2759 offsetof(struct compat_vcpu_info, time));
2760 if (vcpu->xen.vcpu_time_info_set)
2761 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2762 if (v == kvm_get_vcpu(v->kvm, 0))
2763 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2768 * kvmclock updates which are isolated to a given vcpu, such as
2769 * vcpu->cpu migration, should not allow system_timestamp from
2770 * the rest of the vcpus to remain static. Otherwise ntp frequency
2771 * correction applies to one vcpu's system_timestamp but not
2774 * So in those cases, request a kvmclock update for all vcpus.
2775 * We need to rate-limit these requests though, as they can
2776 * considerably slow guests that have a large number of vcpus.
2777 * The time for a remote vcpu to update its kvmclock is bound
2778 * by the delay we use to rate-limit the updates.
2781 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2783 static void kvmclock_update_fn(struct work_struct *work)
2786 struct delayed_work *dwork = to_delayed_work(work);
2787 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2788 kvmclock_update_work);
2789 struct kvm *kvm = container_of(ka, struct kvm, arch);
2790 struct kvm_vcpu *vcpu;
2792 kvm_for_each_vcpu(i, vcpu, kvm) {
2793 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2794 kvm_vcpu_kick(vcpu);
2798 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2800 struct kvm *kvm = v->kvm;
2802 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2803 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2804 KVMCLOCK_UPDATE_DELAY);
2807 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2809 static void kvmclock_sync_fn(struct work_struct *work)
2811 struct delayed_work *dwork = to_delayed_work(work);
2812 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2813 kvmclock_sync_work);
2814 struct kvm *kvm = container_of(ka, struct kvm, arch);
2816 if (!kvmclock_periodic_sync)
2819 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2820 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2821 KVMCLOCK_SYNC_PERIOD);
2825 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2827 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2829 /* McStatusWrEn enabled? */
2830 if (guest_cpuid_is_amd_or_hygon(vcpu))
2831 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2836 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2838 u64 mcg_cap = vcpu->arch.mcg_cap;
2839 unsigned bank_num = mcg_cap & 0xff;
2840 u32 msr = msr_info->index;
2841 u64 data = msr_info->data;
2844 case MSR_IA32_MCG_STATUS:
2845 vcpu->arch.mcg_status = data;
2847 case MSR_IA32_MCG_CTL:
2848 if (!(mcg_cap & MCG_CTL_P) &&
2849 (data || !msr_info->host_initiated))
2851 if (data != 0 && data != ~(u64)0)
2853 vcpu->arch.mcg_ctl = data;
2856 if (msr >= MSR_IA32_MC0_CTL &&
2857 msr < MSR_IA32_MCx_CTL(bank_num)) {
2858 u32 offset = array_index_nospec(
2859 msr - MSR_IA32_MC0_CTL,
2860 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2862 /* only 0 or all 1s can be written to IA32_MCi_CTL
2863 * some Linux kernels though clear bit 10 in bank 4 to
2864 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2865 * this to avoid an uncatched #GP in the guest
2867 if ((offset & 0x3) == 0 &&
2868 data != 0 && (data | (1 << 10)) != ~(u64)0)
2872 if (!msr_info->host_initiated &&
2873 (offset & 0x3) == 1 && data != 0) {
2874 if (!can_set_mci_status(vcpu))
2878 vcpu->arch.mce_banks[offset] = data;
2886 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2888 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2890 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2893 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2895 gpa_t gpa = data & ~0x3f;
2897 /* Bits 4:5 are reserved, Should be zero */
2901 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2902 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2905 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2906 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2909 if (!lapic_in_kernel(vcpu))
2910 return data ? 1 : 0;
2912 vcpu->arch.apf.msr_en_val = data;
2914 if (!kvm_pv_async_pf_enabled(vcpu)) {
2915 kvm_clear_async_pf_completion_queue(vcpu);
2916 kvm_async_pf_hash_reset(vcpu);
2920 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2924 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2925 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2927 kvm_async_pf_wakeup_all(vcpu);
2932 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2934 /* Bits 8-63 are reserved */
2938 if (!lapic_in_kernel(vcpu))
2941 vcpu->arch.apf.msr_int_val = data;
2943 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2948 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2950 vcpu->arch.pv_time_enabled = false;
2951 vcpu->arch.time = 0;
2954 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2956 ++vcpu->stat.tlb_flush;
2957 static_call(kvm_x86_tlb_flush_all)(vcpu);
2960 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2962 ++vcpu->stat.tlb_flush;
2963 static_call(kvm_x86_tlb_flush_guest)(vcpu);
2966 static void record_steal_time(struct kvm_vcpu *vcpu)
2968 struct kvm_host_map map;
2969 struct kvm_steal_time *st;
2971 if (kvm_xen_msr_enabled(vcpu->kvm)) {
2972 kvm_xen_runstate_set_running(vcpu);
2976 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2979 /* -EAGAIN is returned in atomic context so we can just return. */
2980 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2981 &map, &vcpu->arch.st.cache, false))
2985 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2988 * Doing a TLB flush here, on the guest's behalf, can avoid
2991 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
2992 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2993 st->preempted & KVM_VCPU_FLUSH_TLB);
2994 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2995 kvm_vcpu_flush_tlb_guest(vcpu);
2998 vcpu->arch.st.preempted = 0;
3000 if (st->version & 1)
3001 st->version += 1; /* first time write, random junk */
3007 st->steal += current->sched_info.run_delay -
3008 vcpu->arch.st.last_steal;
3009 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3015 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3018 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3021 u32 msr = msr_info->index;
3022 u64 data = msr_info->data;
3024 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3025 return kvm_xen_write_hypercall_page(vcpu, data);
3028 case MSR_AMD64_NB_CFG:
3029 case MSR_IA32_UCODE_WRITE:
3030 case MSR_VM_HSAVE_PA:
3031 case MSR_AMD64_PATCH_LOADER:
3032 case MSR_AMD64_BU_CFG2:
3033 case MSR_AMD64_DC_CFG:
3034 case MSR_F15H_EX_CFG:
3037 case MSR_IA32_UCODE_REV:
3038 if (msr_info->host_initiated)
3039 vcpu->arch.microcode_version = data;
3041 case MSR_IA32_ARCH_CAPABILITIES:
3042 if (!msr_info->host_initiated)
3044 vcpu->arch.arch_capabilities = data;
3046 case MSR_IA32_PERF_CAPABILITIES: {
3047 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3049 if (!msr_info->host_initiated)
3051 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3053 if (data & ~msr_ent.data)
3056 vcpu->arch.perf_capabilities = data;
3061 return set_efer(vcpu, msr_info);
3063 data &= ~(u64)0x40; /* ignore flush filter disable */
3064 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3065 data &= ~(u64)0x8; /* ignore TLB cache disable */
3067 /* Handle McStatusWrEn */
3068 if (data == BIT_ULL(18)) {
3069 vcpu->arch.msr_hwcr = data;
3070 } else if (data != 0) {
3071 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3076 case MSR_FAM10H_MMIO_CONF_BASE:
3078 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3083 case 0x200 ... 0x2ff:
3084 return kvm_mtrr_set_msr(vcpu, msr, data);
3085 case MSR_IA32_APICBASE:
3086 return kvm_set_apic_base(vcpu, msr_info);
3087 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3088 return kvm_x2apic_msr_write(vcpu, msr, data);
3089 case MSR_IA32_TSCDEADLINE:
3090 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3092 case MSR_IA32_TSC_ADJUST:
3093 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3094 if (!msr_info->host_initiated) {
3095 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3096 adjust_tsc_offset_guest(vcpu, adj);
3098 vcpu->arch.ia32_tsc_adjust_msr = data;
3101 case MSR_IA32_MISC_ENABLE:
3102 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3103 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3104 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3106 vcpu->arch.ia32_misc_enable_msr = data;
3107 kvm_update_cpuid_runtime(vcpu);
3109 vcpu->arch.ia32_misc_enable_msr = data;
3112 case MSR_IA32_SMBASE:
3113 if (!msr_info->host_initiated)
3115 vcpu->arch.smbase = data;
3117 case MSR_IA32_POWER_CTL:
3118 vcpu->arch.msr_ia32_power_ctl = data;
3121 if (msr_info->host_initiated) {
3122 kvm_synchronize_tsc(vcpu, data);
3124 u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3125 adjust_tsc_offset_guest(vcpu, adj);
3126 vcpu->arch.ia32_tsc_adjust_msr += adj;
3130 if (!msr_info->host_initiated &&
3131 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3134 * KVM supports exposing PT to the guest, but does not support
3135 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3136 * XSAVES/XRSTORS to save/restore PT MSRs.
3138 if (data & ~supported_xss)
3140 vcpu->arch.ia32_xss = data;
3143 if (!msr_info->host_initiated)
3145 vcpu->arch.smi_count = data;
3147 case MSR_KVM_WALL_CLOCK_NEW:
3148 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3151 vcpu->kvm->arch.wall_clock = data;
3152 kvm_write_wall_clock(vcpu->kvm, data, 0);
3154 case MSR_KVM_WALL_CLOCK:
3155 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3158 vcpu->kvm->arch.wall_clock = data;
3159 kvm_write_wall_clock(vcpu->kvm, data, 0);
3161 case MSR_KVM_SYSTEM_TIME_NEW:
3162 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3165 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3167 case MSR_KVM_SYSTEM_TIME:
3168 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3171 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3173 case MSR_KVM_ASYNC_PF_EN:
3174 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3177 if (kvm_pv_enable_async_pf(vcpu, data))
3180 case MSR_KVM_ASYNC_PF_INT:
3181 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3184 if (kvm_pv_enable_async_pf_int(vcpu, data))
3187 case MSR_KVM_ASYNC_PF_ACK:
3188 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3191 vcpu->arch.apf.pageready_pending = false;
3192 kvm_check_async_pf_completion(vcpu);
3195 case MSR_KVM_STEAL_TIME:
3196 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3199 if (unlikely(!sched_info_on()))
3202 if (data & KVM_STEAL_RESERVED_MASK)
3205 vcpu->arch.st.msr_val = data;
3207 if (!(data & KVM_MSR_ENABLED))
3210 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3213 case MSR_KVM_PV_EOI_EN:
3214 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3217 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3221 case MSR_KVM_POLL_CONTROL:
3222 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3225 /* only enable bit supported */
3226 if (data & (-1ULL << 1))
3229 vcpu->arch.msr_kvm_poll_control = data;
3232 case MSR_IA32_MCG_CTL:
3233 case MSR_IA32_MCG_STATUS:
3234 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3235 return set_msr_mce(vcpu, msr_info);
3237 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3238 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3241 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3242 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3243 if (kvm_pmu_is_valid_msr(vcpu, msr))
3244 return kvm_pmu_set_msr(vcpu, msr_info);
3246 if (pr || data != 0)
3247 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3248 "0x%x data 0x%llx\n", msr, data);
3250 case MSR_K7_CLK_CTL:
3252 * Ignore all writes to this no longer documented MSR.
3253 * Writes are only relevant for old K7 processors,
3254 * all pre-dating SVM, but a recommended workaround from
3255 * AMD for these chips. It is possible to specify the
3256 * affected processor models on the command line, hence
3257 * the need to ignore the workaround.
3260 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3261 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3262 case HV_X64_MSR_SYNDBG_OPTIONS:
3263 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3264 case HV_X64_MSR_CRASH_CTL:
3265 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3266 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3267 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3268 case HV_X64_MSR_TSC_EMULATION_STATUS:
3269 return kvm_hv_set_msr_common(vcpu, msr, data,
3270 msr_info->host_initiated);
3271 case MSR_IA32_BBL_CR_CTL3:
3272 /* Drop writes to this legacy MSR -- see rdmsr
3273 * counterpart for further detail.
3275 if (report_ignored_msrs)
3276 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3279 case MSR_AMD64_OSVW_ID_LENGTH:
3280 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3282 vcpu->arch.osvw.length = data;
3284 case MSR_AMD64_OSVW_STATUS:
3285 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3287 vcpu->arch.osvw.status = data;
3289 case MSR_PLATFORM_INFO:
3290 if (!msr_info->host_initiated ||
3291 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3292 cpuid_fault_enabled(vcpu)))
3294 vcpu->arch.msr_platform_info = data;
3296 case MSR_MISC_FEATURES_ENABLES:
3297 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3298 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3299 !supports_cpuid_fault(vcpu)))
3301 vcpu->arch.msr_misc_features_enables = data;
3304 if (kvm_pmu_is_valid_msr(vcpu, msr))
3305 return kvm_pmu_set_msr(vcpu, msr_info);
3306 return KVM_MSR_RET_INVALID;
3310 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3312 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3315 u64 mcg_cap = vcpu->arch.mcg_cap;
3316 unsigned bank_num = mcg_cap & 0xff;
3319 case MSR_IA32_P5_MC_ADDR:
3320 case MSR_IA32_P5_MC_TYPE:
3323 case MSR_IA32_MCG_CAP:
3324 data = vcpu->arch.mcg_cap;
3326 case MSR_IA32_MCG_CTL:
3327 if (!(mcg_cap & MCG_CTL_P) && !host)
3329 data = vcpu->arch.mcg_ctl;
3331 case MSR_IA32_MCG_STATUS:
3332 data = vcpu->arch.mcg_status;
3335 if (msr >= MSR_IA32_MC0_CTL &&
3336 msr < MSR_IA32_MCx_CTL(bank_num)) {
3337 u32 offset = array_index_nospec(
3338 msr - MSR_IA32_MC0_CTL,
3339 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3341 data = vcpu->arch.mce_banks[offset];
3350 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3352 switch (msr_info->index) {
3353 case MSR_IA32_PLATFORM_ID:
3354 case MSR_IA32_EBL_CR_POWERON:
3355 case MSR_IA32_LASTBRANCHFROMIP:
3356 case MSR_IA32_LASTBRANCHTOIP:
3357 case MSR_IA32_LASTINTFROMIP:
3358 case MSR_IA32_LASTINTTOIP:
3360 case MSR_K8_TSEG_ADDR:
3361 case MSR_K8_TSEG_MASK:
3362 case MSR_VM_HSAVE_PA:
3363 case MSR_K8_INT_PENDING_MSG:
3364 case MSR_AMD64_NB_CFG:
3365 case MSR_FAM10H_MMIO_CONF_BASE:
3366 case MSR_AMD64_BU_CFG2:
3367 case MSR_IA32_PERF_CTL:
3368 case MSR_AMD64_DC_CFG:
3369 case MSR_F15H_EX_CFG:
3371 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3372 * limit) MSRs. Just return 0, as we do not want to expose the host
3373 * data here. Do not conditionalize this on CPUID, as KVM does not do
3374 * so for existing CPU-specific MSRs.
3376 case MSR_RAPL_POWER_UNIT:
3377 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3378 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3379 case MSR_PKG_ENERGY_STATUS: /* Total package */
3380 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3383 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3384 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3385 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3386 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3387 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3388 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3389 return kvm_pmu_get_msr(vcpu, msr_info);
3392 case MSR_IA32_UCODE_REV:
3393 msr_info->data = vcpu->arch.microcode_version;
3395 case MSR_IA32_ARCH_CAPABILITIES:
3396 if (!msr_info->host_initiated &&
3397 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3399 msr_info->data = vcpu->arch.arch_capabilities;
3401 case MSR_IA32_PERF_CAPABILITIES:
3402 if (!msr_info->host_initiated &&
3403 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3405 msr_info->data = vcpu->arch.perf_capabilities;
3407 case MSR_IA32_POWER_CTL:
3408 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3410 case MSR_IA32_TSC: {
3412 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3413 * even when not intercepted. AMD manual doesn't explicitly
3414 * state this but appears to behave the same.
3416 * On userspace reads and writes, however, we unconditionally
3417 * return L1's TSC value to ensure backwards-compatible
3418 * behavior for migration.
3420 u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3421 vcpu->arch.tsc_offset;
3423 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3427 case 0x200 ... 0x2ff:
3428 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3429 case 0xcd: /* fsb frequency */
3433 * MSR_EBC_FREQUENCY_ID
3434 * Conservative value valid for even the basic CPU models.
3435 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3436 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3437 * and 266MHz for model 3, or 4. Set Core Clock
3438 * Frequency to System Bus Frequency Ratio to 1 (bits
3439 * 31:24) even though these are only valid for CPU
3440 * models > 2, however guests may end up dividing or
3441 * multiplying by zero otherwise.
3443 case MSR_EBC_FREQUENCY_ID:
3444 msr_info->data = 1 << 24;
3446 case MSR_IA32_APICBASE:
3447 msr_info->data = kvm_get_apic_base(vcpu);
3449 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3450 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3451 case MSR_IA32_TSCDEADLINE:
3452 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3454 case MSR_IA32_TSC_ADJUST:
3455 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3457 case MSR_IA32_MISC_ENABLE:
3458 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3460 case MSR_IA32_SMBASE:
3461 if (!msr_info->host_initiated)
3463 msr_info->data = vcpu->arch.smbase;
3466 msr_info->data = vcpu->arch.smi_count;
3468 case MSR_IA32_PERF_STATUS:
3469 /* TSC increment by tick */
3470 msr_info->data = 1000ULL;
3471 /* CPU multiplier */
3472 msr_info->data |= (((uint64_t)4ULL) << 40);
3475 msr_info->data = vcpu->arch.efer;
3477 case MSR_KVM_WALL_CLOCK:
3478 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3481 msr_info->data = vcpu->kvm->arch.wall_clock;
3483 case MSR_KVM_WALL_CLOCK_NEW:
3484 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3487 msr_info->data = vcpu->kvm->arch.wall_clock;
3489 case MSR_KVM_SYSTEM_TIME:
3490 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3493 msr_info->data = vcpu->arch.time;
3495 case MSR_KVM_SYSTEM_TIME_NEW:
3496 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3499 msr_info->data = vcpu->arch.time;
3501 case MSR_KVM_ASYNC_PF_EN:
3502 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3505 msr_info->data = vcpu->arch.apf.msr_en_val;
3507 case MSR_KVM_ASYNC_PF_INT:
3508 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3511 msr_info->data = vcpu->arch.apf.msr_int_val;
3513 case MSR_KVM_ASYNC_PF_ACK:
3514 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3519 case MSR_KVM_STEAL_TIME:
3520 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3523 msr_info->data = vcpu->arch.st.msr_val;
3525 case MSR_KVM_PV_EOI_EN:
3526 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3529 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3531 case MSR_KVM_POLL_CONTROL:
3532 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3535 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3537 case MSR_IA32_P5_MC_ADDR:
3538 case MSR_IA32_P5_MC_TYPE:
3539 case MSR_IA32_MCG_CAP:
3540 case MSR_IA32_MCG_CTL:
3541 case MSR_IA32_MCG_STATUS:
3542 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3543 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3544 msr_info->host_initiated);
3546 if (!msr_info->host_initiated &&
3547 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3549 msr_info->data = vcpu->arch.ia32_xss;
3551 case MSR_K7_CLK_CTL:
3553 * Provide expected ramp-up count for K7. All other
3554 * are set to zero, indicating minimum divisors for
3557 * This prevents guest kernels on AMD host with CPU
3558 * type 6, model 8 and higher from exploding due to
3559 * the rdmsr failing.
3561 msr_info->data = 0x20000000;
3563 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3564 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3565 case HV_X64_MSR_SYNDBG_OPTIONS:
3566 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3567 case HV_X64_MSR_CRASH_CTL:
3568 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3569 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3570 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3571 case HV_X64_MSR_TSC_EMULATION_STATUS:
3572 return kvm_hv_get_msr_common(vcpu,
3573 msr_info->index, &msr_info->data,
3574 msr_info->host_initiated);
3575 case MSR_IA32_BBL_CR_CTL3:
3576 /* This legacy MSR exists but isn't fully documented in current
3577 * silicon. It is however accessed by winxp in very narrow
3578 * scenarios where it sets bit #19, itself documented as
3579 * a "reserved" bit. Best effort attempt to source coherent
3580 * read data here should the balance of the register be
3581 * interpreted by the guest:
3583 * L2 cache control register 3: 64GB range, 256KB size,
3584 * enabled, latency 0x1, configured
3586 msr_info->data = 0xbe702111;
3588 case MSR_AMD64_OSVW_ID_LENGTH:
3589 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3591 msr_info->data = vcpu->arch.osvw.length;
3593 case MSR_AMD64_OSVW_STATUS:
3594 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3596 msr_info->data = vcpu->arch.osvw.status;
3598 case MSR_PLATFORM_INFO:
3599 if (!msr_info->host_initiated &&
3600 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3602 msr_info->data = vcpu->arch.msr_platform_info;
3604 case MSR_MISC_FEATURES_ENABLES:
3605 msr_info->data = vcpu->arch.msr_misc_features_enables;
3608 msr_info->data = vcpu->arch.msr_hwcr;
3611 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3612 return kvm_pmu_get_msr(vcpu, msr_info);
3613 return KVM_MSR_RET_INVALID;
3617 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3620 * Read or write a bunch of msrs. All parameters are kernel addresses.
3622 * @return number of msrs set successfully.
3624 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3625 struct kvm_msr_entry *entries,
3626 int (*do_msr)(struct kvm_vcpu *vcpu,
3627 unsigned index, u64 *data))
3631 for (i = 0; i < msrs->nmsrs; ++i)
3632 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3639 * Read or write a bunch of msrs. Parameters are user addresses.
3641 * @return number of msrs set successfully.
3643 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3644 int (*do_msr)(struct kvm_vcpu *vcpu,
3645 unsigned index, u64 *data),
3648 struct kvm_msrs msrs;
3649 struct kvm_msr_entry *entries;
3654 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3658 if (msrs.nmsrs >= MAX_IO_MSRS)
3661 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3662 entries = memdup_user(user_msrs->entries, size);
3663 if (IS_ERR(entries)) {
3664 r = PTR_ERR(entries);
3668 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3673 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3684 static inline bool kvm_can_mwait_in_guest(void)
3686 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3687 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3688 boot_cpu_has(X86_FEATURE_ARAT);
3691 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3692 struct kvm_cpuid2 __user *cpuid_arg)
3694 struct kvm_cpuid2 cpuid;
3698 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3701 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3706 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3712 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3717 case KVM_CAP_IRQCHIP:
3719 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3720 case KVM_CAP_SET_TSS_ADDR:
3721 case KVM_CAP_EXT_CPUID:
3722 case KVM_CAP_EXT_EMUL_CPUID:
3723 case KVM_CAP_CLOCKSOURCE:
3725 case KVM_CAP_NOP_IO_DELAY:
3726 case KVM_CAP_MP_STATE:
3727 case KVM_CAP_SYNC_MMU:
3728 case KVM_CAP_USER_NMI:
3729 case KVM_CAP_REINJECT_CONTROL:
3730 case KVM_CAP_IRQ_INJECT_STATUS:
3731 case KVM_CAP_IOEVENTFD:
3732 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3734 case KVM_CAP_PIT_STATE2:
3735 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3736 case KVM_CAP_VCPU_EVENTS:
3737 case KVM_CAP_HYPERV:
3738 case KVM_CAP_HYPERV_VAPIC:
3739 case KVM_CAP_HYPERV_SPIN:
3740 case KVM_CAP_HYPERV_SYNIC:
3741 case KVM_CAP_HYPERV_SYNIC2:
3742 case KVM_CAP_HYPERV_VP_INDEX:
3743 case KVM_CAP_HYPERV_EVENTFD:
3744 case KVM_CAP_HYPERV_TLBFLUSH:
3745 case KVM_CAP_HYPERV_SEND_IPI:
3746 case KVM_CAP_HYPERV_CPUID:
3747 case KVM_CAP_SYS_HYPERV_CPUID:
3748 case KVM_CAP_PCI_SEGMENT:
3749 case KVM_CAP_DEBUGREGS:
3750 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3752 case KVM_CAP_ASYNC_PF:
3753 case KVM_CAP_ASYNC_PF_INT:
3754 case KVM_CAP_GET_TSC_KHZ:
3755 case KVM_CAP_KVMCLOCK_CTRL:
3756 case KVM_CAP_READONLY_MEM:
3757 case KVM_CAP_HYPERV_TIME:
3758 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3759 case KVM_CAP_TSC_DEADLINE_TIMER:
3760 case KVM_CAP_DISABLE_QUIRKS:
3761 case KVM_CAP_SET_BOOT_CPU_ID:
3762 case KVM_CAP_SPLIT_IRQCHIP:
3763 case KVM_CAP_IMMEDIATE_EXIT:
3764 case KVM_CAP_PMU_EVENT_FILTER:
3765 case KVM_CAP_GET_MSR_FEATURES:
3766 case KVM_CAP_MSR_PLATFORM_INFO:
3767 case KVM_CAP_EXCEPTION_PAYLOAD:
3768 case KVM_CAP_SET_GUEST_DEBUG:
3769 case KVM_CAP_LAST_CPU:
3770 case KVM_CAP_X86_USER_SPACE_MSR:
3771 case KVM_CAP_X86_MSR_FILTER:
3772 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3775 #ifdef CONFIG_KVM_XEN
3776 case KVM_CAP_XEN_HVM:
3777 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
3778 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
3779 KVM_XEN_HVM_CONFIG_SHARED_INFO;
3780 if (sched_info_on())
3781 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
3784 case KVM_CAP_SYNC_REGS:
3785 r = KVM_SYNC_X86_VALID_FIELDS;
3787 case KVM_CAP_ADJUST_CLOCK:
3788 r = KVM_CLOCK_TSC_STABLE;
3790 case KVM_CAP_X86_DISABLE_EXITS:
3791 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3792 KVM_X86_DISABLE_EXITS_CSTATE;
3793 if(kvm_can_mwait_in_guest())
3794 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3796 case KVM_CAP_X86_SMM:
3797 /* SMBASE is usually relocated above 1M on modern chipsets,
3798 * and SMM handlers might indeed rely on 4G segment limits,
3799 * so do not report SMM to be available if real mode is
3800 * emulated via vm86 mode. Still, do not go to great lengths
3801 * to avoid userspace's usage of the feature, because it is a
3802 * fringe case that is not enabled except via specific settings
3803 * of the module parameters.
3805 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
3808 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
3810 case KVM_CAP_NR_VCPUS:
3811 r = KVM_SOFT_MAX_VCPUS;
3813 case KVM_CAP_MAX_VCPUS:
3816 case KVM_CAP_MAX_VCPU_ID:
3817 r = KVM_MAX_VCPU_ID;
3819 case KVM_CAP_PV_MMU: /* obsolete */
3823 r = KVM_MAX_MCE_BANKS;
3826 r = boot_cpu_has(X86_FEATURE_XSAVE);
3828 case KVM_CAP_TSC_CONTROL:
3829 r = kvm_has_tsc_control;
3831 case KVM_CAP_X2APIC_API:
3832 r = KVM_X2APIC_API_VALID_FLAGS;
3834 case KVM_CAP_NESTED_STATE:
3835 r = kvm_x86_ops.nested_ops->get_state ?
3836 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3838 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3839 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3841 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3842 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3844 case KVM_CAP_SMALLER_MAXPHYADDR:
3845 r = (int) allow_smaller_maxphyaddr;
3847 case KVM_CAP_STEAL_TIME:
3848 r = sched_info_on();
3850 case KVM_CAP_X86_BUS_LOCK_EXIT:
3851 if (kvm_has_bus_lock_exit)
3852 r = KVM_BUS_LOCK_DETECTION_OFF |
3853 KVM_BUS_LOCK_DETECTION_EXIT;
3864 long kvm_arch_dev_ioctl(struct file *filp,
3865 unsigned int ioctl, unsigned long arg)
3867 void __user *argp = (void __user *)arg;
3871 case KVM_GET_MSR_INDEX_LIST: {
3872 struct kvm_msr_list __user *user_msr_list = argp;
3873 struct kvm_msr_list msr_list;
3877 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3880 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3881 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3884 if (n < msr_list.nmsrs)
3887 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3888 num_msrs_to_save * sizeof(u32)))
3890 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3892 num_emulated_msrs * sizeof(u32)))
3897 case KVM_GET_SUPPORTED_CPUID:
3898 case KVM_GET_EMULATED_CPUID: {
3899 struct kvm_cpuid2 __user *cpuid_arg = argp;
3900 struct kvm_cpuid2 cpuid;
3903 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3906 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3912 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3917 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3919 if (copy_to_user(argp, &kvm_mce_cap_supported,
3920 sizeof(kvm_mce_cap_supported)))
3924 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3925 struct kvm_msr_list __user *user_msr_list = argp;
3926 struct kvm_msr_list msr_list;
3930 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3933 msr_list.nmsrs = num_msr_based_features;
3934 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3937 if (n < msr_list.nmsrs)
3940 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3941 num_msr_based_features * sizeof(u32)))
3947 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3949 case KVM_GET_SUPPORTED_HV_CPUID:
3950 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
3960 static void wbinvd_ipi(void *garbage)
3965 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3967 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3970 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3972 /* Address WBINVD may be executed by guest */
3973 if (need_emulate_wbinvd(vcpu)) {
3974 if (static_call(kvm_x86_has_wbinvd_exit)())
3975 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3976 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3977 smp_call_function_single(vcpu->cpu,
3978 wbinvd_ipi, NULL, 1);
3981 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
3983 /* Save host pkru register if supported */
3984 vcpu->arch.host_pkru = read_pkru();
3986 /* Apply any externally detected TSC adjustments (due to suspend) */
3987 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3988 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3989 vcpu->arch.tsc_offset_adjustment = 0;
3990 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3993 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3994 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3995 rdtsc() - vcpu->arch.last_host_tsc;
3997 mark_tsc_unstable("KVM discovered backwards TSC");
3999 if (kvm_check_tsc_unstable()) {
4000 u64 offset = kvm_compute_tsc_offset(vcpu,
4001 vcpu->arch.last_guest_tsc);
4002 kvm_vcpu_write_tsc_offset(vcpu, offset);
4003 vcpu->arch.tsc_catchup = 1;
4006 if (kvm_lapic_hv_timer_in_use(vcpu))
4007 kvm_lapic_restart_hv_timer(vcpu);
4010 * On a host with synchronized TSC, there is no need to update
4011 * kvmclock on vcpu->cpu migration
4013 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4014 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4015 if (vcpu->cpu != cpu)
4016 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4020 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4023 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4025 struct kvm_host_map map;
4026 struct kvm_steal_time *st;
4029 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4032 if (vcpu->arch.st.preempted)
4036 * Take the srcu lock as memslots will be accessed to check the gfn
4037 * cache generation against the memslots generation.
4039 idx = srcu_read_lock(&vcpu->kvm->srcu);
4041 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4042 &vcpu->arch.st.cache, true))
4046 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4048 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4050 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4053 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4056 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4058 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4059 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4061 if (kvm_xen_msr_enabled(vcpu->kvm))
4062 kvm_xen_runstate_set_preempted(vcpu);
4064 kvm_steal_time_set_preempted(vcpu);
4066 static_call(kvm_x86_vcpu_put)(vcpu);
4067 vcpu->arch.last_host_tsc = rdtsc();
4069 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4070 * on every vmexit, but if not, we might have a stale dr6 from the
4071 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4076 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4077 struct kvm_lapic_state *s)
4079 if (vcpu->arch.apicv_active)
4080 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4082 return kvm_apic_get_state(vcpu, s);
4085 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4086 struct kvm_lapic_state *s)
4090 r = kvm_apic_set_state(vcpu, s);
4093 update_cr8_intercept(vcpu);
4098 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4101 * We can accept userspace's request for interrupt injection
4102 * as long as we have a place to store the interrupt number.
4103 * The actual injection will happen when the CPU is able to
4104 * deliver the interrupt.
4106 if (kvm_cpu_has_extint(vcpu))
4109 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4110 return (!lapic_in_kernel(vcpu) ||
4111 kvm_apic_accept_pic_intr(vcpu));
4114 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4116 return kvm_arch_interrupt_allowed(vcpu) &&
4117 kvm_cpu_accept_dm_intr(vcpu);
4120 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4121 struct kvm_interrupt *irq)
4123 if (irq->irq >= KVM_NR_INTERRUPTS)
4126 if (!irqchip_in_kernel(vcpu->kvm)) {
4127 kvm_queue_interrupt(vcpu, irq->irq, false);
4128 kvm_make_request(KVM_REQ_EVENT, vcpu);
4133 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4134 * fail for in-kernel 8259.
4136 if (pic_in_kernel(vcpu->kvm))
4139 if (vcpu->arch.pending_external_vector != -1)
4142 vcpu->arch.pending_external_vector = irq->irq;
4143 kvm_make_request(KVM_REQ_EVENT, vcpu);
4147 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4149 kvm_inject_nmi(vcpu);
4154 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4156 kvm_make_request(KVM_REQ_SMI, vcpu);
4161 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4162 struct kvm_tpr_access_ctl *tac)
4166 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4170 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4174 unsigned bank_num = mcg_cap & 0xff, bank;
4177 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4179 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4182 vcpu->arch.mcg_cap = mcg_cap;
4183 /* Init IA32_MCG_CTL to all 1s */
4184 if (mcg_cap & MCG_CTL_P)
4185 vcpu->arch.mcg_ctl = ~(u64)0;
4186 /* Init IA32_MCi_CTL to all 1s */
4187 for (bank = 0; bank < bank_num; bank++)
4188 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4190 static_call(kvm_x86_setup_mce)(vcpu);
4195 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4196 struct kvm_x86_mce *mce)
4198 u64 mcg_cap = vcpu->arch.mcg_cap;
4199 unsigned bank_num = mcg_cap & 0xff;
4200 u64 *banks = vcpu->arch.mce_banks;
4202 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4205 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4206 * reporting is disabled
4208 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4209 vcpu->arch.mcg_ctl != ~(u64)0)
4211 banks += 4 * mce->bank;
4213 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4214 * reporting is disabled for the bank
4216 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4218 if (mce->status & MCI_STATUS_UC) {
4219 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4220 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4221 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4224 if (banks[1] & MCI_STATUS_VAL)
4225 mce->status |= MCI_STATUS_OVER;
4226 banks[2] = mce->addr;
4227 banks[3] = mce->misc;
4228 vcpu->arch.mcg_status = mce->mcg_status;
4229 banks[1] = mce->status;
4230 kvm_queue_exception(vcpu, MC_VECTOR);
4231 } else if (!(banks[1] & MCI_STATUS_VAL)
4232 || !(banks[1] & MCI_STATUS_UC)) {
4233 if (banks[1] & MCI_STATUS_VAL)
4234 mce->status |= MCI_STATUS_OVER;
4235 banks[2] = mce->addr;
4236 banks[3] = mce->misc;
4237 banks[1] = mce->status;
4239 banks[1] |= MCI_STATUS_OVER;
4243 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4244 struct kvm_vcpu_events *events)
4248 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4252 * In guest mode, payload delivery should be deferred,
4253 * so that the L1 hypervisor can intercept #PF before
4254 * CR2 is modified (or intercept #DB before DR6 is
4255 * modified under nVMX). Unless the per-VM capability,
4256 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4257 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4258 * opportunistically defer the exception payload, deliver it if the
4259 * capability hasn't been requested before processing a
4260 * KVM_GET_VCPU_EVENTS.
4262 if (!vcpu->kvm->arch.exception_payload_enabled &&
4263 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4264 kvm_deliver_exception_payload(vcpu);
4267 * The API doesn't provide the instruction length for software
4268 * exceptions, so don't report them. As long as the guest RIP
4269 * isn't advanced, we should expect to encounter the exception
4272 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4273 events->exception.injected = 0;
4274 events->exception.pending = 0;
4276 events->exception.injected = vcpu->arch.exception.injected;
4277 events->exception.pending = vcpu->arch.exception.pending;
4279 * For ABI compatibility, deliberately conflate
4280 * pending and injected exceptions when
4281 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4283 if (!vcpu->kvm->arch.exception_payload_enabled)
4284 events->exception.injected |=
4285 vcpu->arch.exception.pending;
4287 events->exception.nr = vcpu->arch.exception.nr;
4288 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4289 events->exception.error_code = vcpu->arch.exception.error_code;
4290 events->exception_has_payload = vcpu->arch.exception.has_payload;
4291 events->exception_payload = vcpu->arch.exception.payload;
4293 events->interrupt.injected =
4294 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4295 events->interrupt.nr = vcpu->arch.interrupt.nr;
4296 events->interrupt.soft = 0;
4297 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4299 events->nmi.injected = vcpu->arch.nmi_injected;
4300 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4301 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4302 events->nmi.pad = 0;
4304 events->sipi_vector = 0; /* never valid when reporting to user space */
4306 events->smi.smm = is_smm(vcpu);
4307 events->smi.pending = vcpu->arch.smi_pending;
4308 events->smi.smm_inside_nmi =
4309 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4310 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4312 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4313 | KVM_VCPUEVENT_VALID_SHADOW
4314 | KVM_VCPUEVENT_VALID_SMM);
4315 if (vcpu->kvm->arch.exception_payload_enabled)
4316 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4318 memset(&events->reserved, 0, sizeof(events->reserved));
4321 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4323 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4324 struct kvm_vcpu_events *events)
4326 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4327 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4328 | KVM_VCPUEVENT_VALID_SHADOW
4329 | KVM_VCPUEVENT_VALID_SMM
4330 | KVM_VCPUEVENT_VALID_PAYLOAD))
4333 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4334 if (!vcpu->kvm->arch.exception_payload_enabled)
4336 if (events->exception.pending)
4337 events->exception.injected = 0;
4339 events->exception_has_payload = 0;
4341 events->exception.pending = 0;
4342 events->exception_has_payload = 0;
4345 if ((events->exception.injected || events->exception.pending) &&
4346 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4349 /* INITs are latched while in SMM */
4350 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4351 (events->smi.smm || events->smi.pending) &&
4352 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4356 vcpu->arch.exception.injected = events->exception.injected;
4357 vcpu->arch.exception.pending = events->exception.pending;
4358 vcpu->arch.exception.nr = events->exception.nr;
4359 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4360 vcpu->arch.exception.error_code = events->exception.error_code;
4361 vcpu->arch.exception.has_payload = events->exception_has_payload;
4362 vcpu->arch.exception.payload = events->exception_payload;
4364 vcpu->arch.interrupt.injected = events->interrupt.injected;
4365 vcpu->arch.interrupt.nr = events->interrupt.nr;
4366 vcpu->arch.interrupt.soft = events->interrupt.soft;
4367 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4368 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4369 events->interrupt.shadow);
4371 vcpu->arch.nmi_injected = events->nmi.injected;
4372 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4373 vcpu->arch.nmi_pending = events->nmi.pending;
4374 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4376 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4377 lapic_in_kernel(vcpu))
4378 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4380 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4381 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4382 if (events->smi.smm)
4383 vcpu->arch.hflags |= HF_SMM_MASK;
4385 vcpu->arch.hflags &= ~HF_SMM_MASK;
4386 kvm_smm_changed(vcpu);
4389 vcpu->arch.smi_pending = events->smi.pending;
4391 if (events->smi.smm) {
4392 if (events->smi.smm_inside_nmi)
4393 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4395 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4398 if (lapic_in_kernel(vcpu)) {
4399 if (events->smi.latched_init)
4400 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4402 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4406 kvm_make_request(KVM_REQ_EVENT, vcpu);
4411 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4412 struct kvm_debugregs *dbgregs)
4416 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4417 kvm_get_dr(vcpu, 6, &val);
4419 dbgregs->dr7 = vcpu->arch.dr7;
4421 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4424 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4425 struct kvm_debugregs *dbgregs)
4430 if (!kvm_dr6_valid(dbgregs->dr6))
4432 if (!kvm_dr7_valid(dbgregs->dr7))
4435 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4436 kvm_update_dr0123(vcpu);
4437 vcpu->arch.dr6 = dbgregs->dr6;
4438 vcpu->arch.dr7 = dbgregs->dr7;
4439 kvm_update_dr7(vcpu);
4444 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4446 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4448 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4449 u64 xstate_bv = xsave->header.xfeatures;
4453 * Copy legacy XSAVE area, to avoid complications with CPUID
4454 * leaves 0 and 1 in the loop below.
4456 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4459 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4460 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4463 * Copy each region from the possibly compacted offset to the
4464 * non-compacted offset.
4466 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4468 u64 xfeature_mask = valid & -valid;
4469 int xfeature_nr = fls64(xfeature_mask) - 1;
4470 void *src = get_xsave_addr(xsave, xfeature_nr);
4473 u32 size, offset, ecx, edx;
4474 cpuid_count(XSTATE_CPUID, xfeature_nr,
4475 &size, &offset, &ecx, &edx);
4476 if (xfeature_nr == XFEATURE_PKRU)
4477 memcpy(dest + offset, &vcpu->arch.pkru,
4478 sizeof(vcpu->arch.pkru));
4480 memcpy(dest + offset, src, size);
4484 valid -= xfeature_mask;
4488 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4490 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4491 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4495 * Copy legacy XSAVE area, to avoid complications with CPUID
4496 * leaves 0 and 1 in the loop below.
4498 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4500 /* Set XSTATE_BV and possibly XCOMP_BV. */
4501 xsave->header.xfeatures = xstate_bv;
4502 if (boot_cpu_has(X86_FEATURE_XSAVES))
4503 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4506 * Copy each region from the non-compacted offset to the
4507 * possibly compacted offset.
4509 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4511 u64 xfeature_mask = valid & -valid;
4512 int xfeature_nr = fls64(xfeature_mask) - 1;
4513 void *dest = get_xsave_addr(xsave, xfeature_nr);
4516 u32 size, offset, ecx, edx;
4517 cpuid_count(XSTATE_CPUID, xfeature_nr,
4518 &size, &offset, &ecx, &edx);
4519 if (xfeature_nr == XFEATURE_PKRU)
4520 memcpy(&vcpu->arch.pkru, src + offset,
4521 sizeof(vcpu->arch.pkru));
4523 memcpy(dest, src + offset, size);
4526 valid -= xfeature_mask;
4530 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4531 struct kvm_xsave *guest_xsave)
4533 if (!vcpu->arch.guest_fpu)
4536 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4537 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4538 fill_xsave((u8 *) guest_xsave->region, vcpu);
4540 memcpy(guest_xsave->region,
4541 &vcpu->arch.guest_fpu->state.fxsave,
4542 sizeof(struct fxregs_state));
4543 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4544 XFEATURE_MASK_FPSSE;
4548 #define XSAVE_MXCSR_OFFSET 24
4550 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4551 struct kvm_xsave *guest_xsave)
4556 if (!vcpu->arch.guest_fpu)
4559 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4560 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4562 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4564 * Here we allow setting states that are not present in
4565 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4566 * with old userspace.
4568 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4570 load_xsave(vcpu, (u8 *)guest_xsave->region);
4572 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4573 mxcsr & ~mxcsr_feature_mask)
4575 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4576 guest_xsave->region, sizeof(struct fxregs_state));
4581 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4582 struct kvm_xcrs *guest_xcrs)
4584 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4585 guest_xcrs->nr_xcrs = 0;
4589 guest_xcrs->nr_xcrs = 1;
4590 guest_xcrs->flags = 0;
4591 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4592 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4595 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4596 struct kvm_xcrs *guest_xcrs)
4600 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4603 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4606 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4607 /* Only support XCR0 currently */
4608 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4609 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4610 guest_xcrs->xcrs[i].value);
4619 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4620 * stopped by the hypervisor. This function will be called from the host only.
4621 * EINVAL is returned when the host attempts to set the flag for a guest that
4622 * does not support pv clocks.
4624 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4626 if (!vcpu->arch.pv_time_enabled)
4628 vcpu->arch.pvclock_set_guest_stopped_request = true;
4629 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4633 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4634 struct kvm_enable_cap *cap)
4637 uint16_t vmcs_version;
4638 void __user *user_ptr;
4644 case KVM_CAP_HYPERV_SYNIC2:
4649 case KVM_CAP_HYPERV_SYNIC:
4650 if (!irqchip_in_kernel(vcpu->kvm))
4652 return kvm_hv_activate_synic(vcpu, cap->cap ==
4653 KVM_CAP_HYPERV_SYNIC2);
4654 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4655 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4657 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4659 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4660 if (copy_to_user(user_ptr, &vmcs_version,
4661 sizeof(vmcs_version)))
4665 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4666 if (!kvm_x86_ops.enable_direct_tlbflush)
4669 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4671 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4672 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4673 if (vcpu->arch.pv_cpuid.enforce)
4674 kvm_update_pv_runtime(vcpu);
4683 long kvm_arch_vcpu_ioctl(struct file *filp,
4684 unsigned int ioctl, unsigned long arg)
4686 struct kvm_vcpu *vcpu = filp->private_data;
4687 void __user *argp = (void __user *)arg;
4690 struct kvm_lapic_state *lapic;
4691 struct kvm_xsave *xsave;
4692 struct kvm_xcrs *xcrs;
4700 case KVM_GET_LAPIC: {
4702 if (!lapic_in_kernel(vcpu))
4704 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4705 GFP_KERNEL_ACCOUNT);
4710 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4714 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4719 case KVM_SET_LAPIC: {
4721 if (!lapic_in_kernel(vcpu))
4723 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4724 if (IS_ERR(u.lapic)) {
4725 r = PTR_ERR(u.lapic);
4729 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4732 case KVM_INTERRUPT: {
4733 struct kvm_interrupt irq;
4736 if (copy_from_user(&irq, argp, sizeof(irq)))
4738 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4742 r = kvm_vcpu_ioctl_nmi(vcpu);
4746 r = kvm_vcpu_ioctl_smi(vcpu);
4749 case KVM_SET_CPUID: {
4750 struct kvm_cpuid __user *cpuid_arg = argp;
4751 struct kvm_cpuid cpuid;
4754 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4756 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4759 case KVM_SET_CPUID2: {
4760 struct kvm_cpuid2 __user *cpuid_arg = argp;
4761 struct kvm_cpuid2 cpuid;
4764 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4766 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4767 cpuid_arg->entries);
4770 case KVM_GET_CPUID2: {
4771 struct kvm_cpuid2 __user *cpuid_arg = argp;
4772 struct kvm_cpuid2 cpuid;
4775 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4777 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4778 cpuid_arg->entries);
4782 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4787 case KVM_GET_MSRS: {
4788 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4789 r = msr_io(vcpu, argp, do_get_msr, 1);
4790 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4793 case KVM_SET_MSRS: {
4794 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4795 r = msr_io(vcpu, argp, do_set_msr, 0);
4796 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4799 case KVM_TPR_ACCESS_REPORTING: {
4800 struct kvm_tpr_access_ctl tac;
4803 if (copy_from_user(&tac, argp, sizeof(tac)))
4805 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4809 if (copy_to_user(argp, &tac, sizeof(tac)))
4814 case KVM_SET_VAPIC_ADDR: {
4815 struct kvm_vapic_addr va;
4819 if (!lapic_in_kernel(vcpu))
4822 if (copy_from_user(&va, argp, sizeof(va)))
4824 idx = srcu_read_lock(&vcpu->kvm->srcu);
4825 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4826 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4829 case KVM_X86_SETUP_MCE: {
4833 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4835 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4838 case KVM_X86_SET_MCE: {
4839 struct kvm_x86_mce mce;
4842 if (copy_from_user(&mce, argp, sizeof(mce)))
4844 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4847 case KVM_GET_VCPU_EVENTS: {
4848 struct kvm_vcpu_events events;
4850 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4853 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4858 case KVM_SET_VCPU_EVENTS: {
4859 struct kvm_vcpu_events events;
4862 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4865 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4868 case KVM_GET_DEBUGREGS: {
4869 struct kvm_debugregs dbgregs;
4871 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4874 if (copy_to_user(argp, &dbgregs,
4875 sizeof(struct kvm_debugregs)))
4880 case KVM_SET_DEBUGREGS: {
4881 struct kvm_debugregs dbgregs;
4884 if (copy_from_user(&dbgregs, argp,
4885 sizeof(struct kvm_debugregs)))
4888 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4891 case KVM_GET_XSAVE: {
4892 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4897 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4900 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4905 case KVM_SET_XSAVE: {
4906 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4907 if (IS_ERR(u.xsave)) {
4908 r = PTR_ERR(u.xsave);
4912 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4915 case KVM_GET_XCRS: {
4916 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4921 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4924 if (copy_to_user(argp, u.xcrs,
4925 sizeof(struct kvm_xcrs)))
4930 case KVM_SET_XCRS: {
4931 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4932 if (IS_ERR(u.xcrs)) {
4933 r = PTR_ERR(u.xcrs);
4937 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4940 case KVM_SET_TSC_KHZ: {
4944 user_tsc_khz = (u32)arg;
4946 if (kvm_has_tsc_control &&
4947 user_tsc_khz >= kvm_max_guest_tsc_khz)
4950 if (user_tsc_khz == 0)
4951 user_tsc_khz = tsc_khz;
4953 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4958 case KVM_GET_TSC_KHZ: {
4959 r = vcpu->arch.virtual_tsc_khz;
4962 case KVM_KVMCLOCK_CTRL: {
4963 r = kvm_set_guest_paused(vcpu);
4966 case KVM_ENABLE_CAP: {
4967 struct kvm_enable_cap cap;
4970 if (copy_from_user(&cap, argp, sizeof(cap)))
4972 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4975 case KVM_GET_NESTED_STATE: {
4976 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4980 if (!kvm_x86_ops.nested_ops->get_state)
4983 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4985 if (get_user(user_data_size, &user_kvm_nested_state->size))
4988 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4993 if (r > user_data_size) {
4994 if (put_user(r, &user_kvm_nested_state->size))
5004 case KVM_SET_NESTED_STATE: {
5005 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5006 struct kvm_nested_state kvm_state;
5010 if (!kvm_x86_ops.nested_ops->set_state)
5014 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5018 if (kvm_state.size < sizeof(kvm_state))
5021 if (kvm_state.flags &
5022 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5023 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5024 | KVM_STATE_NESTED_GIF_SET))
5027 /* nested_run_pending implies guest_mode. */
5028 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5029 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5032 idx = srcu_read_lock(&vcpu->kvm->srcu);
5033 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5034 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5037 case KVM_GET_SUPPORTED_HV_CPUID:
5038 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5040 #ifdef CONFIG_KVM_XEN
5041 case KVM_XEN_VCPU_GET_ATTR: {
5042 struct kvm_xen_vcpu_attr xva;
5045 if (copy_from_user(&xva, argp, sizeof(xva)))
5047 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5048 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5052 case KVM_XEN_VCPU_SET_ATTR: {
5053 struct kvm_xen_vcpu_attr xva;
5056 if (copy_from_user(&xva, argp, sizeof(xva)))
5058 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5072 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5074 return VM_FAULT_SIGBUS;
5077 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5081 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5083 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5087 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5090 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5093 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5094 unsigned long kvm_nr_mmu_pages)
5096 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5099 mutex_lock(&kvm->slots_lock);
5101 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5102 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5104 mutex_unlock(&kvm->slots_lock);
5108 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5110 return kvm->arch.n_max_mmu_pages;
5113 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5115 struct kvm_pic *pic = kvm->arch.vpic;
5119 switch (chip->chip_id) {
5120 case KVM_IRQCHIP_PIC_MASTER:
5121 memcpy(&chip->chip.pic, &pic->pics[0],
5122 sizeof(struct kvm_pic_state));
5124 case KVM_IRQCHIP_PIC_SLAVE:
5125 memcpy(&chip->chip.pic, &pic->pics[1],
5126 sizeof(struct kvm_pic_state));
5128 case KVM_IRQCHIP_IOAPIC:
5129 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5138 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5140 struct kvm_pic *pic = kvm->arch.vpic;
5144 switch (chip->chip_id) {
5145 case KVM_IRQCHIP_PIC_MASTER:
5146 spin_lock(&pic->lock);
5147 memcpy(&pic->pics[0], &chip->chip.pic,
5148 sizeof(struct kvm_pic_state));
5149 spin_unlock(&pic->lock);
5151 case KVM_IRQCHIP_PIC_SLAVE:
5152 spin_lock(&pic->lock);
5153 memcpy(&pic->pics[1], &chip->chip.pic,
5154 sizeof(struct kvm_pic_state));
5155 spin_unlock(&pic->lock);
5157 case KVM_IRQCHIP_IOAPIC:
5158 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5164 kvm_pic_update_irq(pic);
5168 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5170 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5172 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5174 mutex_lock(&kps->lock);
5175 memcpy(ps, &kps->channels, sizeof(*ps));
5176 mutex_unlock(&kps->lock);
5180 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5183 struct kvm_pit *pit = kvm->arch.vpit;
5185 mutex_lock(&pit->pit_state.lock);
5186 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5187 for (i = 0; i < 3; i++)
5188 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5189 mutex_unlock(&pit->pit_state.lock);
5193 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5195 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5196 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5197 sizeof(ps->channels));
5198 ps->flags = kvm->arch.vpit->pit_state.flags;
5199 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5200 memset(&ps->reserved, 0, sizeof(ps->reserved));
5204 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5208 u32 prev_legacy, cur_legacy;
5209 struct kvm_pit *pit = kvm->arch.vpit;
5211 mutex_lock(&pit->pit_state.lock);
5212 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5213 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5214 if (!prev_legacy && cur_legacy)
5216 memcpy(&pit->pit_state.channels, &ps->channels,
5217 sizeof(pit->pit_state.channels));
5218 pit->pit_state.flags = ps->flags;
5219 for (i = 0; i < 3; i++)
5220 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5222 mutex_unlock(&pit->pit_state.lock);
5226 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5227 struct kvm_reinject_control *control)
5229 struct kvm_pit *pit = kvm->arch.vpit;
5231 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5232 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5233 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5235 mutex_lock(&pit->pit_state.lock);
5236 kvm_pit_set_reinject(pit, control->pit_reinject);
5237 mutex_unlock(&pit->pit_state.lock);
5242 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5246 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5247 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5248 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5251 struct kvm_vcpu *vcpu;
5254 kvm_for_each_vcpu(i, vcpu, kvm)
5255 kvm_vcpu_kick(vcpu);
5258 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5261 if (!irqchip_in_kernel(kvm))
5264 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5265 irq_event->irq, irq_event->level,
5270 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5271 struct kvm_enable_cap *cap)
5279 case KVM_CAP_DISABLE_QUIRKS:
5280 kvm->arch.disabled_quirks = cap->args[0];
5283 case KVM_CAP_SPLIT_IRQCHIP: {
5284 mutex_lock(&kvm->lock);
5286 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5287 goto split_irqchip_unlock;
5289 if (irqchip_in_kernel(kvm))
5290 goto split_irqchip_unlock;
5291 if (kvm->created_vcpus)
5292 goto split_irqchip_unlock;
5293 r = kvm_setup_empty_irq_routing(kvm);
5295 goto split_irqchip_unlock;
5296 /* Pairs with irqchip_in_kernel. */
5298 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5299 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5301 split_irqchip_unlock:
5302 mutex_unlock(&kvm->lock);
5305 case KVM_CAP_X2APIC_API:
5307 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5310 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5311 kvm->arch.x2apic_format = true;
5312 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5313 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5317 case KVM_CAP_X86_DISABLE_EXITS:
5319 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5322 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5323 kvm_can_mwait_in_guest())
5324 kvm->arch.mwait_in_guest = true;
5325 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5326 kvm->arch.hlt_in_guest = true;
5327 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5328 kvm->arch.pause_in_guest = true;
5329 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5330 kvm->arch.cstate_in_guest = true;
5333 case KVM_CAP_MSR_PLATFORM_INFO:
5334 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5337 case KVM_CAP_EXCEPTION_PAYLOAD:
5338 kvm->arch.exception_payload_enabled = cap->args[0];
5341 case KVM_CAP_X86_USER_SPACE_MSR:
5342 kvm->arch.user_space_msr_mask = cap->args[0];
5345 case KVM_CAP_X86_BUS_LOCK_EXIT:
5347 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5350 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5351 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5354 if (kvm_has_bus_lock_exit &&
5355 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5356 kvm->arch.bus_lock_detection_enabled = true;
5366 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5368 struct kvm_x86_msr_filter *msr_filter;
5370 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5374 msr_filter->default_allow = default_allow;
5378 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5385 for (i = 0; i < msr_filter->count; i++)
5386 kfree(msr_filter->ranges[i].bitmap);
5391 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5392 struct kvm_msr_filter_range *user_range)
5394 struct msr_bitmap_range range;
5395 unsigned long *bitmap = NULL;
5399 if (!user_range->nmsrs)
5402 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5403 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5406 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5408 return PTR_ERR(bitmap);
5410 range = (struct msr_bitmap_range) {
5411 .flags = user_range->flags,
5412 .base = user_range->base,
5413 .nmsrs = user_range->nmsrs,
5417 if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) {
5427 /* Everything ok, add this range identifier. */
5428 msr_filter->ranges[msr_filter->count] = range;
5429 msr_filter->count++;
5437 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5439 struct kvm_msr_filter __user *user_msr_filter = argp;
5440 struct kvm_x86_msr_filter *new_filter, *old_filter;
5441 struct kvm_msr_filter filter;
5447 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5450 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5451 empty &= !filter.ranges[i].nmsrs;
5453 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5454 if (empty && !default_allow)
5457 new_filter = kvm_alloc_msr_filter(default_allow);
5461 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5462 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5464 kvm_free_msr_filter(new_filter);
5469 mutex_lock(&kvm->lock);
5471 /* The per-VM filter is protected by kvm->lock... */
5472 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5474 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5475 synchronize_srcu(&kvm->srcu);
5477 kvm_free_msr_filter(old_filter);
5479 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5480 mutex_unlock(&kvm->lock);
5485 long kvm_arch_vm_ioctl(struct file *filp,
5486 unsigned int ioctl, unsigned long arg)
5488 struct kvm *kvm = filp->private_data;
5489 void __user *argp = (void __user *)arg;
5492 * This union makes it completely explicit to gcc-3.x
5493 * that these two variables' stack usage should be
5494 * combined, not added together.
5497 struct kvm_pit_state ps;
5498 struct kvm_pit_state2 ps2;
5499 struct kvm_pit_config pit_config;
5503 case KVM_SET_TSS_ADDR:
5504 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5506 case KVM_SET_IDENTITY_MAP_ADDR: {
5509 mutex_lock(&kvm->lock);
5511 if (kvm->created_vcpus)
5512 goto set_identity_unlock;
5514 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5515 goto set_identity_unlock;
5516 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5517 set_identity_unlock:
5518 mutex_unlock(&kvm->lock);
5521 case KVM_SET_NR_MMU_PAGES:
5522 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5524 case KVM_GET_NR_MMU_PAGES:
5525 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5527 case KVM_CREATE_IRQCHIP: {
5528 mutex_lock(&kvm->lock);
5531 if (irqchip_in_kernel(kvm))
5532 goto create_irqchip_unlock;
5535 if (kvm->created_vcpus)
5536 goto create_irqchip_unlock;
5538 r = kvm_pic_init(kvm);
5540 goto create_irqchip_unlock;
5542 r = kvm_ioapic_init(kvm);
5544 kvm_pic_destroy(kvm);
5545 goto create_irqchip_unlock;
5548 r = kvm_setup_default_irq_routing(kvm);
5550 kvm_ioapic_destroy(kvm);
5551 kvm_pic_destroy(kvm);
5552 goto create_irqchip_unlock;
5554 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5556 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5557 create_irqchip_unlock:
5558 mutex_unlock(&kvm->lock);
5561 case KVM_CREATE_PIT:
5562 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5564 case KVM_CREATE_PIT2:
5566 if (copy_from_user(&u.pit_config, argp,
5567 sizeof(struct kvm_pit_config)))
5570 mutex_lock(&kvm->lock);
5573 goto create_pit_unlock;
5575 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5579 mutex_unlock(&kvm->lock);
5581 case KVM_GET_IRQCHIP: {
5582 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5583 struct kvm_irqchip *chip;
5585 chip = memdup_user(argp, sizeof(*chip));
5592 if (!irqchip_kernel(kvm))
5593 goto get_irqchip_out;
5594 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5596 goto get_irqchip_out;
5598 if (copy_to_user(argp, chip, sizeof(*chip)))
5599 goto get_irqchip_out;
5605 case KVM_SET_IRQCHIP: {
5606 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5607 struct kvm_irqchip *chip;
5609 chip = memdup_user(argp, sizeof(*chip));
5616 if (!irqchip_kernel(kvm))
5617 goto set_irqchip_out;
5618 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5625 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5628 if (!kvm->arch.vpit)
5630 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5634 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5641 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5643 mutex_lock(&kvm->lock);
5645 if (!kvm->arch.vpit)
5647 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5649 mutex_unlock(&kvm->lock);
5652 case KVM_GET_PIT2: {
5654 if (!kvm->arch.vpit)
5656 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5660 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5665 case KVM_SET_PIT2: {
5667 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5669 mutex_lock(&kvm->lock);
5671 if (!kvm->arch.vpit)
5673 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5675 mutex_unlock(&kvm->lock);
5678 case KVM_REINJECT_CONTROL: {
5679 struct kvm_reinject_control control;
5681 if (copy_from_user(&control, argp, sizeof(control)))
5684 if (!kvm->arch.vpit)
5686 r = kvm_vm_ioctl_reinject(kvm, &control);
5689 case KVM_SET_BOOT_CPU_ID:
5691 mutex_lock(&kvm->lock);
5692 if (kvm->created_vcpus)
5695 kvm->arch.bsp_vcpu_id = arg;
5696 mutex_unlock(&kvm->lock);
5698 #ifdef CONFIG_KVM_XEN
5699 case KVM_XEN_HVM_CONFIG: {
5700 struct kvm_xen_hvm_config xhc;
5702 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5704 r = kvm_xen_hvm_config(kvm, &xhc);
5707 case KVM_XEN_HVM_GET_ATTR: {
5708 struct kvm_xen_hvm_attr xha;
5711 if (copy_from_user(&xha, argp, sizeof(xha)))
5713 r = kvm_xen_hvm_get_attr(kvm, &xha);
5714 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5718 case KVM_XEN_HVM_SET_ATTR: {
5719 struct kvm_xen_hvm_attr xha;
5722 if (copy_from_user(&xha, argp, sizeof(xha)))
5724 r = kvm_xen_hvm_set_attr(kvm, &xha);
5728 case KVM_SET_CLOCK: {
5729 struct kvm_clock_data user_ns;
5733 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5742 * TODO: userspace has to take care of races with VCPU_RUN, so
5743 * kvm_gen_update_masterclock() can be cut down to locked
5744 * pvclock_update_vm_gtod_copy().
5746 kvm_gen_update_masterclock(kvm);
5747 now_ns = get_kvmclock_ns(kvm);
5748 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5749 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5752 case KVM_GET_CLOCK: {
5753 struct kvm_clock_data user_ns;
5756 now_ns = get_kvmclock_ns(kvm);
5757 user_ns.clock = now_ns;
5758 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5759 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5762 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5767 case KVM_MEMORY_ENCRYPT_OP: {
5769 if (kvm_x86_ops.mem_enc_op)
5770 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
5773 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5774 struct kvm_enc_region region;
5777 if (copy_from_user(®ion, argp, sizeof(region)))
5781 if (kvm_x86_ops.mem_enc_reg_region)
5782 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
5785 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5786 struct kvm_enc_region region;
5789 if (copy_from_user(®ion, argp, sizeof(region)))
5793 if (kvm_x86_ops.mem_enc_unreg_region)
5794 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
5797 case KVM_HYPERV_EVENTFD: {
5798 struct kvm_hyperv_eventfd hvevfd;
5801 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5803 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5806 case KVM_SET_PMU_EVENT_FILTER:
5807 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5809 case KVM_X86_SET_MSR_FILTER:
5810 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5819 static void kvm_init_msr_list(void)
5821 struct x86_pmu_capability x86_pmu;
5825 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5826 "Please update the fixed PMCs in msrs_to_saved_all[]");
5828 perf_get_x86_pmu_capability(&x86_pmu);
5830 num_msrs_to_save = 0;
5831 num_emulated_msrs = 0;
5832 num_msr_based_features = 0;
5834 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5835 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5839 * Even MSRs that are valid in the host may not be exposed
5840 * to the guests in some cases.
5842 switch (msrs_to_save_all[i]) {
5843 case MSR_IA32_BNDCFGS:
5844 if (!kvm_mpx_supported())
5848 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5851 case MSR_IA32_UMWAIT_CONTROL:
5852 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5855 case MSR_IA32_RTIT_CTL:
5856 case MSR_IA32_RTIT_STATUS:
5857 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5860 case MSR_IA32_RTIT_CR3_MATCH:
5861 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5862 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5865 case MSR_IA32_RTIT_OUTPUT_BASE:
5866 case MSR_IA32_RTIT_OUTPUT_MASK:
5867 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5868 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5869 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5872 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5873 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5874 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5875 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5878 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5879 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5880 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5883 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5884 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5885 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5892 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5895 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5896 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
5899 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5902 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5903 struct kvm_msr_entry msr;
5905 msr.index = msr_based_features_all[i];
5906 if (kvm_get_msr_feature(&msr))
5909 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5913 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5921 if (!(lapic_in_kernel(vcpu) &&
5922 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5923 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5934 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5941 if (!(lapic_in_kernel(vcpu) &&
5942 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5944 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5946 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5956 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5957 struct kvm_segment *var, int seg)
5959 static_call(kvm_x86_set_segment)(vcpu, var, seg);
5962 void kvm_get_segment(struct kvm_vcpu *vcpu,
5963 struct kvm_segment *var, int seg)
5965 static_call(kvm_x86_get_segment)(vcpu, var, seg);
5968 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5969 struct x86_exception *exception)
5973 BUG_ON(!mmu_is_nested(vcpu));
5975 /* NPT walks are always user-walks */
5976 access |= PFERR_USER_MASK;
5977 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5982 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5983 struct x86_exception *exception)
5985 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
5986 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5989 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5990 struct x86_exception *exception)
5992 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
5993 access |= PFERR_FETCH_MASK;
5994 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5997 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5998 struct x86_exception *exception)
6000 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6001 access |= PFERR_WRITE_MASK;
6002 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6005 /* uses this to access any guest's mapped memory without checking CPL */
6006 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6007 struct x86_exception *exception)
6009 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6012 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6013 struct kvm_vcpu *vcpu, u32 access,
6014 struct x86_exception *exception)
6017 int r = X86EMUL_CONTINUE;
6020 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6022 unsigned offset = addr & (PAGE_SIZE-1);
6023 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6026 if (gpa == UNMAPPED_GVA)
6027 return X86EMUL_PROPAGATE_FAULT;
6028 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6031 r = X86EMUL_IO_NEEDED;
6043 /* used for instruction fetching */
6044 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6045 gva_t addr, void *val, unsigned int bytes,
6046 struct x86_exception *exception)
6048 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6049 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6053 /* Inline kvm_read_guest_virt_helper for speed. */
6054 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6056 if (unlikely(gpa == UNMAPPED_GVA))
6057 return X86EMUL_PROPAGATE_FAULT;
6059 offset = addr & (PAGE_SIZE-1);
6060 if (WARN_ON(offset + bytes > PAGE_SIZE))
6061 bytes = (unsigned)PAGE_SIZE - offset;
6062 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6064 if (unlikely(ret < 0))
6065 return X86EMUL_IO_NEEDED;
6067 return X86EMUL_CONTINUE;
6070 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6071 gva_t addr, void *val, unsigned int bytes,
6072 struct x86_exception *exception)
6074 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6077 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6078 * is returned, but our callers are not ready for that and they blindly
6079 * call kvm_inject_page_fault. Ensure that they at least do not leak
6080 * uninitialized kernel stack memory into cr2 and error code.
6082 memset(exception, 0, sizeof(*exception));
6083 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6086 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6088 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6089 gva_t addr, void *val, unsigned int bytes,
6090 struct x86_exception *exception, bool system)
6092 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6095 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6096 access |= PFERR_USER_MASK;
6098 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6101 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6102 unsigned long addr, void *val, unsigned int bytes)
6104 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6105 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6107 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6110 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6111 struct kvm_vcpu *vcpu, u32 access,
6112 struct x86_exception *exception)
6115 int r = X86EMUL_CONTINUE;
6118 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6121 unsigned offset = addr & (PAGE_SIZE-1);
6122 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6125 if (gpa == UNMAPPED_GVA)
6126 return X86EMUL_PROPAGATE_FAULT;
6127 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6129 r = X86EMUL_IO_NEEDED;
6141 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6142 unsigned int bytes, struct x86_exception *exception,
6145 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6146 u32 access = PFERR_WRITE_MASK;
6148 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6149 access |= PFERR_USER_MASK;
6151 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6155 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6156 unsigned int bytes, struct x86_exception *exception)
6158 /* kvm_write_guest_virt_system can pull in tons of pages. */
6159 vcpu->arch.l1tf_flush_l1d = true;
6161 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6162 PFERR_WRITE_MASK, exception);
6164 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6166 int handle_ud(struct kvm_vcpu *vcpu)
6168 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6169 int emul_type = EMULTYPE_TRAP_UD;
6170 char sig[5]; /* ud2; .ascii "kvm" */
6171 struct x86_exception e;
6173 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6176 if (force_emulation_prefix &&
6177 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6178 sig, sizeof(sig), &e) == 0 &&
6179 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6180 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6181 emul_type = EMULTYPE_TRAP_UD_FORCED;
6184 return kvm_emulate_instruction(vcpu, emul_type);
6186 EXPORT_SYMBOL_GPL(handle_ud);
6188 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6189 gpa_t gpa, bool write)
6191 /* For APIC access vmexit */
6192 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6195 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6196 trace_vcpu_match_mmio(gva, gpa, write, true);
6203 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6204 gpa_t *gpa, struct x86_exception *exception,
6207 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6208 | (write ? PFERR_WRITE_MASK : 0);
6211 * currently PKRU is only applied to ept enabled guest so
6212 * there is no pkey in EPT page table for L1 guest or EPT
6213 * shadow page table for L2 guest.
6215 if (vcpu_match_mmio_gva(vcpu, gva)
6216 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6217 vcpu->arch.mmio_access, 0, access)) {
6218 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6219 (gva & (PAGE_SIZE - 1));
6220 trace_vcpu_match_mmio(gva, *gpa, write, false);
6224 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6226 if (*gpa == UNMAPPED_GVA)
6229 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6232 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6233 const void *val, int bytes)
6237 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6240 kvm_page_track_write(vcpu, gpa, val, bytes);
6244 struct read_write_emulator_ops {
6245 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6247 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6248 void *val, int bytes);
6249 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6250 int bytes, void *val);
6251 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6252 void *val, int bytes);
6256 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6258 if (vcpu->mmio_read_completed) {
6259 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6260 vcpu->mmio_fragments[0].gpa, val);
6261 vcpu->mmio_read_completed = 0;
6268 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6269 void *val, int bytes)
6271 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6274 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6275 void *val, int bytes)
6277 return emulator_write_phys(vcpu, gpa, val, bytes);
6280 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6282 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6283 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6286 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6287 void *val, int bytes)
6289 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6290 return X86EMUL_IO_NEEDED;
6293 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6294 void *val, int bytes)
6296 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6298 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6299 return X86EMUL_CONTINUE;
6302 static const struct read_write_emulator_ops read_emultor = {
6303 .read_write_prepare = read_prepare,
6304 .read_write_emulate = read_emulate,
6305 .read_write_mmio = vcpu_mmio_read,
6306 .read_write_exit_mmio = read_exit_mmio,
6309 static const struct read_write_emulator_ops write_emultor = {
6310 .read_write_emulate = write_emulate,
6311 .read_write_mmio = write_mmio,
6312 .read_write_exit_mmio = write_exit_mmio,
6316 static int emulator_read_write_onepage(unsigned long addr, void *val,
6318 struct x86_exception *exception,
6319 struct kvm_vcpu *vcpu,
6320 const struct read_write_emulator_ops *ops)
6324 bool write = ops->write;
6325 struct kvm_mmio_fragment *frag;
6326 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6329 * If the exit was due to a NPF we may already have a GPA.
6330 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6331 * Note, this cannot be used on string operations since string
6332 * operation using rep will only have the initial GPA from the NPF
6335 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6336 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6337 gpa = ctxt->gpa_val;
6338 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6340 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6342 return X86EMUL_PROPAGATE_FAULT;
6345 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6346 return X86EMUL_CONTINUE;
6349 * Is this MMIO handled locally?
6351 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6352 if (handled == bytes)
6353 return X86EMUL_CONTINUE;
6359 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6360 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6364 return X86EMUL_CONTINUE;
6367 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6369 void *val, unsigned int bytes,
6370 struct x86_exception *exception,
6371 const struct read_write_emulator_ops *ops)
6373 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6377 if (ops->read_write_prepare &&
6378 ops->read_write_prepare(vcpu, val, bytes))
6379 return X86EMUL_CONTINUE;
6381 vcpu->mmio_nr_fragments = 0;
6383 /* Crossing a page boundary? */
6384 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6387 now = -addr & ~PAGE_MASK;
6388 rc = emulator_read_write_onepage(addr, val, now, exception,
6391 if (rc != X86EMUL_CONTINUE)
6394 if (ctxt->mode != X86EMUL_MODE_PROT64)
6400 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6402 if (rc != X86EMUL_CONTINUE)
6405 if (!vcpu->mmio_nr_fragments)
6408 gpa = vcpu->mmio_fragments[0].gpa;
6410 vcpu->mmio_needed = 1;
6411 vcpu->mmio_cur_fragment = 0;
6413 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6414 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6415 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6416 vcpu->run->mmio.phys_addr = gpa;
6418 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6421 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6425 struct x86_exception *exception)
6427 return emulator_read_write(ctxt, addr, val, bytes,
6428 exception, &read_emultor);
6431 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6435 struct x86_exception *exception)
6437 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6438 exception, &write_emultor);
6441 #define CMPXCHG_TYPE(t, ptr, old, new) \
6442 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6444 #ifdef CONFIG_X86_64
6445 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6447 # define CMPXCHG64(ptr, old, new) \
6448 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6451 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6456 struct x86_exception *exception)
6458 struct kvm_host_map map;
6459 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6465 /* guests cmpxchg8b have to be emulated atomically */
6466 if (bytes > 8 || (bytes & (bytes - 1)))
6469 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6471 if (gpa == UNMAPPED_GVA ||
6472 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6476 * Emulate the atomic as a straight write to avoid #AC if SLD is
6477 * enabled in the host and the access splits a cache line.
6479 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6480 page_line_mask = ~(cache_line_size() - 1);
6482 page_line_mask = PAGE_MASK;
6484 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6487 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6490 kaddr = map.hva + offset_in_page(gpa);
6494 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6497 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6500 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6503 exchanged = CMPXCHG64(kaddr, old, new);
6509 kvm_vcpu_unmap(vcpu, &map, true);
6512 return X86EMUL_CMPXCHG_FAILED;
6514 kvm_page_track_write(vcpu, gpa, new, bytes);
6516 return X86EMUL_CONTINUE;
6519 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6521 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6524 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6528 for (i = 0; i < vcpu->arch.pio.count; i++) {
6529 if (vcpu->arch.pio.in)
6530 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6531 vcpu->arch.pio.size, pd);
6533 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6534 vcpu->arch.pio.port, vcpu->arch.pio.size,
6538 pd += vcpu->arch.pio.size;
6543 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6544 unsigned short port, void *val,
6545 unsigned int count, bool in)
6547 vcpu->arch.pio.port = port;
6548 vcpu->arch.pio.in = in;
6549 vcpu->arch.pio.count = count;
6550 vcpu->arch.pio.size = size;
6552 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6553 vcpu->arch.pio.count = 0;
6557 vcpu->run->exit_reason = KVM_EXIT_IO;
6558 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6559 vcpu->run->io.size = size;
6560 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6561 vcpu->run->io.count = count;
6562 vcpu->run->io.port = port;
6567 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6568 unsigned short port, void *val, unsigned int count)
6572 if (vcpu->arch.pio.count)
6575 memset(vcpu->arch.pio_data, 0, size * count);
6577 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6580 memcpy(val, vcpu->arch.pio_data, size * count);
6581 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6582 vcpu->arch.pio.count = 0;
6589 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6590 int size, unsigned short port, void *val,
6593 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6597 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6598 unsigned short port, const void *val,
6601 memcpy(vcpu->arch.pio_data, val, size * count);
6602 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6603 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6606 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6607 int size, unsigned short port,
6608 const void *val, unsigned int count)
6610 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6613 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6615 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6618 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6620 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6623 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6625 if (!need_emulate_wbinvd(vcpu))
6626 return X86EMUL_CONTINUE;
6628 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6629 int cpu = get_cpu();
6631 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6632 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6633 wbinvd_ipi, NULL, 1);
6635 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6638 return X86EMUL_CONTINUE;
6641 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6643 kvm_emulate_wbinvd_noskip(vcpu);
6644 return kvm_skip_emulated_instruction(vcpu);
6646 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6650 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6652 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6655 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6656 unsigned long *dest)
6658 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6661 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6662 unsigned long value)
6665 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6668 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6670 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6673 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6675 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6676 unsigned long value;
6680 value = kvm_read_cr0(vcpu);
6683 value = vcpu->arch.cr2;
6686 value = kvm_read_cr3(vcpu);
6689 value = kvm_read_cr4(vcpu);
6692 value = kvm_get_cr8(vcpu);
6695 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6702 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6704 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6709 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6712 vcpu->arch.cr2 = val;
6715 res = kvm_set_cr3(vcpu, val);
6718 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6721 res = kvm_set_cr8(vcpu, val);
6724 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6731 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6733 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
6736 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6738 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
6741 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6743 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
6746 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6748 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
6751 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6753 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
6756 static unsigned long emulator_get_cached_segment_base(
6757 struct x86_emulate_ctxt *ctxt, int seg)
6759 return get_segment_base(emul_to_vcpu(ctxt), seg);
6762 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6763 struct desc_struct *desc, u32 *base3,
6766 struct kvm_segment var;
6768 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6769 *selector = var.selector;
6772 memset(desc, 0, sizeof(*desc));
6780 set_desc_limit(desc, var.limit);
6781 set_desc_base(desc, (unsigned long)var.base);
6782 #ifdef CONFIG_X86_64
6784 *base3 = var.base >> 32;
6786 desc->type = var.type;
6788 desc->dpl = var.dpl;
6789 desc->p = var.present;
6790 desc->avl = var.avl;
6798 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6799 struct desc_struct *desc, u32 base3,
6802 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6803 struct kvm_segment var;
6805 var.selector = selector;
6806 var.base = get_desc_base(desc);
6807 #ifdef CONFIG_X86_64
6808 var.base |= ((u64)base3) << 32;
6810 var.limit = get_desc_limit(desc);
6812 var.limit = (var.limit << 12) | 0xfff;
6813 var.type = desc->type;
6814 var.dpl = desc->dpl;
6819 var.avl = desc->avl;
6820 var.present = desc->p;
6821 var.unusable = !var.present;
6824 kvm_set_segment(vcpu, &var, seg);
6828 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6829 u32 msr_index, u64 *pdata)
6831 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6834 r = kvm_get_msr(vcpu, msr_index, pdata);
6836 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6837 /* Bounce to user space */
6838 return X86EMUL_IO_NEEDED;
6844 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6845 u32 msr_index, u64 data)
6847 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6850 r = kvm_set_msr(vcpu, msr_index, data);
6852 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6853 /* Bounce to user space */
6854 return X86EMUL_IO_NEEDED;
6860 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6862 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6864 return vcpu->arch.smbase;
6867 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6869 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6871 vcpu->arch.smbase = smbase;
6874 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6877 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6880 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6881 u32 pmc, u64 *pdata)
6883 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6886 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6888 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6891 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6892 struct x86_instruction_info *info,
6893 enum x86_intercept_stage stage)
6895 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
6899 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6900 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6903 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6906 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6908 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6911 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6913 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6916 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6918 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6921 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6923 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6926 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6928 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6931 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6933 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
6936 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6938 return emul_to_vcpu(ctxt)->arch.hflags;
6941 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6943 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6946 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6947 const char *smstate)
6949 return static_call(kvm_x86_pre_leave_smm)(emul_to_vcpu(ctxt), smstate);
6952 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6954 kvm_smm_changed(emul_to_vcpu(ctxt));
6957 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6959 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6962 static const struct x86_emulate_ops emulate_ops = {
6963 .read_gpr = emulator_read_gpr,
6964 .write_gpr = emulator_write_gpr,
6965 .read_std = emulator_read_std,
6966 .write_std = emulator_write_std,
6967 .read_phys = kvm_read_guest_phys_system,
6968 .fetch = kvm_fetch_guest_virt,
6969 .read_emulated = emulator_read_emulated,
6970 .write_emulated = emulator_write_emulated,
6971 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6972 .invlpg = emulator_invlpg,
6973 .pio_in_emulated = emulator_pio_in_emulated,
6974 .pio_out_emulated = emulator_pio_out_emulated,
6975 .get_segment = emulator_get_segment,
6976 .set_segment = emulator_set_segment,
6977 .get_cached_segment_base = emulator_get_cached_segment_base,
6978 .get_gdt = emulator_get_gdt,
6979 .get_idt = emulator_get_idt,
6980 .set_gdt = emulator_set_gdt,
6981 .set_idt = emulator_set_idt,
6982 .get_cr = emulator_get_cr,
6983 .set_cr = emulator_set_cr,
6984 .cpl = emulator_get_cpl,
6985 .get_dr = emulator_get_dr,
6986 .set_dr = emulator_set_dr,
6987 .get_smbase = emulator_get_smbase,
6988 .set_smbase = emulator_set_smbase,
6989 .set_msr = emulator_set_msr,
6990 .get_msr = emulator_get_msr,
6991 .check_pmc = emulator_check_pmc,
6992 .read_pmc = emulator_read_pmc,
6993 .halt = emulator_halt,
6994 .wbinvd = emulator_wbinvd,
6995 .fix_hypercall = emulator_fix_hypercall,
6996 .intercept = emulator_intercept,
6997 .get_cpuid = emulator_get_cpuid,
6998 .guest_has_long_mode = emulator_guest_has_long_mode,
6999 .guest_has_movbe = emulator_guest_has_movbe,
7000 .guest_has_fxsr = emulator_guest_has_fxsr,
7001 .set_nmi_mask = emulator_set_nmi_mask,
7002 .get_hflags = emulator_get_hflags,
7003 .set_hflags = emulator_set_hflags,
7004 .pre_leave_smm = emulator_pre_leave_smm,
7005 .post_leave_smm = emulator_post_leave_smm,
7006 .set_xcr = emulator_set_xcr,
7009 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7011 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7013 * an sti; sti; sequence only disable interrupts for the first
7014 * instruction. So, if the last instruction, be it emulated or
7015 * not, left the system with the INT_STI flag enabled, it
7016 * means that the last instruction is an sti. We should not
7017 * leave the flag on in this case. The same goes for mov ss
7019 if (int_shadow & mask)
7021 if (unlikely(int_shadow || mask)) {
7022 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7024 kvm_make_request(KVM_REQ_EVENT, vcpu);
7028 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7030 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7031 if (ctxt->exception.vector == PF_VECTOR)
7032 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7034 if (ctxt->exception.error_code_valid)
7035 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7036 ctxt->exception.error_code);
7038 kvm_queue_exception(vcpu, ctxt->exception.vector);
7042 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7044 struct x86_emulate_ctxt *ctxt;
7046 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7048 pr_err("kvm: failed to allocate vcpu's emulator\n");
7053 ctxt->ops = &emulate_ops;
7054 vcpu->arch.emulate_ctxt = ctxt;
7059 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7061 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7064 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7066 ctxt->gpa_available = false;
7067 ctxt->eflags = kvm_get_rflags(vcpu);
7068 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7070 ctxt->eip = kvm_rip_read(vcpu);
7071 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7072 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7073 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7074 cs_db ? X86EMUL_MODE_PROT32 :
7075 X86EMUL_MODE_PROT16;
7076 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7077 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7078 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7080 init_decode_cache(ctxt);
7081 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7084 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7086 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7089 init_emulate_ctxt(vcpu);
7093 ctxt->_eip = ctxt->eip + inc_eip;
7094 ret = emulate_int_real(ctxt, irq);
7096 if (ret != X86EMUL_CONTINUE) {
7097 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7099 ctxt->eip = ctxt->_eip;
7100 kvm_rip_write(vcpu, ctxt->eip);
7101 kvm_set_rflags(vcpu, ctxt->eflags);
7104 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7106 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7108 ++vcpu->stat.insn_emulation_fail;
7109 trace_kvm_emulate_insn_failed(vcpu);
7111 if (emulation_type & EMULTYPE_VMWARE_GP) {
7112 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7116 if (emulation_type & EMULTYPE_SKIP) {
7117 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7118 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7119 vcpu->run->internal.ndata = 0;
7123 kvm_queue_exception(vcpu, UD_VECTOR);
7125 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7126 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7127 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7128 vcpu->run->internal.ndata = 0;
7135 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7136 bool write_fault_to_shadow_pgtable,
7139 gpa_t gpa = cr2_or_gpa;
7142 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7145 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7146 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7149 if (!vcpu->arch.mmu->direct_map) {
7151 * Write permission should be allowed since only
7152 * write access need to be emulated.
7154 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7157 * If the mapping is invalid in guest, let cpu retry
7158 * it to generate fault.
7160 if (gpa == UNMAPPED_GVA)
7165 * Do not retry the unhandleable instruction if it faults on the
7166 * readonly host memory, otherwise it will goto a infinite loop:
7167 * retry instruction -> write #PF -> emulation fail -> retry
7168 * instruction -> ...
7170 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7173 * If the instruction failed on the error pfn, it can not be fixed,
7174 * report the error to userspace.
7176 if (is_error_noslot_pfn(pfn))
7179 kvm_release_pfn_clean(pfn);
7181 /* The instructions are well-emulated on direct mmu. */
7182 if (vcpu->arch.mmu->direct_map) {
7183 unsigned int indirect_shadow_pages;
7185 write_lock(&vcpu->kvm->mmu_lock);
7186 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7187 write_unlock(&vcpu->kvm->mmu_lock);
7189 if (indirect_shadow_pages)
7190 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7196 * if emulation was due to access to shadowed page table
7197 * and it failed try to unshadow page and re-enter the
7198 * guest to let CPU execute the instruction.
7200 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7203 * If the access faults on its page table, it can not
7204 * be fixed by unprotecting shadow page and it should
7205 * be reported to userspace.
7207 return !write_fault_to_shadow_pgtable;
7210 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7211 gpa_t cr2_or_gpa, int emulation_type)
7213 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7214 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7216 last_retry_eip = vcpu->arch.last_retry_eip;
7217 last_retry_addr = vcpu->arch.last_retry_addr;
7220 * If the emulation is caused by #PF and it is non-page_table
7221 * writing instruction, it means the VM-EXIT is caused by shadow
7222 * page protected, we can zap the shadow page and retry this
7223 * instruction directly.
7225 * Note: if the guest uses a non-page-table modifying instruction
7226 * on the PDE that points to the instruction, then we will unmap
7227 * the instruction and go to an infinite loop. So, we cache the
7228 * last retried eip and the last fault address, if we meet the eip
7229 * and the address again, we can break out of the potential infinite
7232 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7234 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7237 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7238 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7241 if (x86_page_table_writing_insn(ctxt))
7244 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7247 vcpu->arch.last_retry_eip = ctxt->eip;
7248 vcpu->arch.last_retry_addr = cr2_or_gpa;
7250 if (!vcpu->arch.mmu->direct_map)
7251 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7253 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7258 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7259 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7261 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7263 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7264 /* This is a good place to trace that we are exiting SMM. */
7265 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7267 /* Process a latched INIT or SMI, if any. */
7268 kvm_make_request(KVM_REQ_EVENT, vcpu);
7271 kvm_mmu_reset_context(vcpu);
7274 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7283 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7284 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7289 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7291 struct kvm_run *kvm_run = vcpu->run;
7293 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7294 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7295 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7296 kvm_run->debug.arch.exception = DB_VECTOR;
7297 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7300 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7304 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7306 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7309 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7314 * rflags is the old, "raw" value of the flags. The new value has
7315 * not been saved yet.
7317 * This is correct even for TF set by the guest, because "the
7318 * processor will not generate this exception after the instruction
7319 * that sets the TF flag".
7321 if (unlikely(rflags & X86_EFLAGS_TF))
7322 r = kvm_vcpu_do_singlestep(vcpu);
7325 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7327 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7329 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7330 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7331 struct kvm_run *kvm_run = vcpu->run;
7332 unsigned long eip = kvm_get_linear_rip(vcpu);
7333 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7334 vcpu->arch.guest_debug_dr7,
7338 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7339 kvm_run->debug.arch.pc = eip;
7340 kvm_run->debug.arch.exception = DB_VECTOR;
7341 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7347 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7348 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7349 unsigned long eip = kvm_get_linear_rip(vcpu);
7350 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7355 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7364 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7366 switch (ctxt->opcode_len) {
7373 case 0xe6: /* OUT */
7377 case 0x6c: /* INS */
7379 case 0x6e: /* OUTS */
7386 case 0x33: /* RDPMC */
7396 * Decode to be emulated instruction. Return EMULATION_OK if success.
7398 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7399 void *insn, int insn_len)
7401 int r = EMULATION_OK;
7402 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7404 init_emulate_ctxt(vcpu);
7407 * We will reenter on the same instruction since we do not set
7408 * complete_userspace_io. This does not handle watchpoints yet,
7409 * those would be handled in the emulate_ops.
7411 if (!(emulation_type & EMULTYPE_SKIP) &&
7412 kvm_vcpu_check_breakpoint(vcpu, &r))
7415 ctxt->interruptibility = 0;
7416 ctxt->have_exception = false;
7417 ctxt->exception.vector = -1;
7418 ctxt->perm_ok = false;
7420 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7422 r = x86_decode_insn(ctxt, insn, insn_len);
7424 trace_kvm_emulate_insn_start(vcpu);
7425 ++vcpu->stat.insn_emulation;
7429 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7431 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7432 int emulation_type, void *insn, int insn_len)
7435 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7436 bool writeback = true;
7437 bool write_fault_to_spt;
7439 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7442 vcpu->arch.l1tf_flush_l1d = true;
7445 * Clear write_fault_to_shadow_pgtable here to ensure it is
7448 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7449 vcpu->arch.write_fault_to_shadow_pgtable = false;
7451 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7452 kvm_clear_exception_queue(vcpu);
7454 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7456 if (r != EMULATION_OK) {
7457 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7458 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7459 kvm_queue_exception(vcpu, UD_VECTOR);
7462 if (reexecute_instruction(vcpu, cr2_or_gpa,
7466 if (ctxt->have_exception) {
7468 * #UD should result in just EMULATION_FAILED, and trap-like
7469 * exception should not be encountered during decode.
7471 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7472 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7473 inject_emulated_exception(vcpu);
7476 return handle_emulation_failure(vcpu, emulation_type);
7480 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7481 !is_vmware_backdoor_opcode(ctxt)) {
7482 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7487 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7488 * for kvm_skip_emulated_instruction(). The caller is responsible for
7489 * updating interruptibility state and injecting single-step #DBs.
7491 if (emulation_type & EMULTYPE_SKIP) {
7492 kvm_rip_write(vcpu, ctxt->_eip);
7493 if (ctxt->eflags & X86_EFLAGS_RF)
7494 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7498 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7501 /* this is needed for vmware backdoor interface to work since it
7502 changes registers values during IO operation */
7503 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7504 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7505 emulator_invalidate_register_cache(ctxt);
7509 if (emulation_type & EMULTYPE_PF) {
7510 /* Save the faulting GPA (cr2) in the address field */
7511 ctxt->exception.address = cr2_or_gpa;
7513 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7514 if (vcpu->arch.mmu->direct_map) {
7515 ctxt->gpa_available = true;
7516 ctxt->gpa_val = cr2_or_gpa;
7519 /* Sanitize the address out of an abundance of paranoia. */
7520 ctxt->exception.address = 0;
7523 r = x86_emulate_insn(ctxt);
7525 if (r == EMULATION_INTERCEPTED)
7528 if (r == EMULATION_FAILED) {
7529 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7533 return handle_emulation_failure(vcpu, emulation_type);
7536 if (ctxt->have_exception) {
7538 if (inject_emulated_exception(vcpu))
7540 } else if (vcpu->arch.pio.count) {
7541 if (!vcpu->arch.pio.in) {
7542 /* FIXME: return into emulator if single-stepping. */
7543 vcpu->arch.pio.count = 0;
7546 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7549 } else if (vcpu->mmio_needed) {
7550 ++vcpu->stat.mmio_exits;
7552 if (!vcpu->mmio_is_write)
7555 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7556 } else if (r == EMULATION_RESTART)
7562 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7563 toggle_interruptibility(vcpu, ctxt->interruptibility);
7564 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7565 if (!ctxt->have_exception ||
7566 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7567 kvm_rip_write(vcpu, ctxt->eip);
7568 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7569 r = kvm_vcpu_do_singlestep(vcpu);
7570 if (kvm_x86_ops.update_emulated_instruction)
7571 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7572 __kvm_set_rflags(vcpu, ctxt->eflags);
7576 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7577 * do nothing, and it will be requested again as soon as
7578 * the shadow expires. But we still need to check here,
7579 * because POPF has no interrupt shadow.
7581 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7582 kvm_make_request(KVM_REQ_EVENT, vcpu);
7584 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7589 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7591 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7593 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7595 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7596 void *insn, int insn_len)
7598 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7600 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7602 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7604 vcpu->arch.pio.count = 0;
7608 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7610 vcpu->arch.pio.count = 0;
7612 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7615 return kvm_skip_emulated_instruction(vcpu);
7618 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7619 unsigned short port)
7621 unsigned long val = kvm_rax_read(vcpu);
7622 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7628 * Workaround userspace that relies on old KVM behavior of %rip being
7629 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7632 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7633 vcpu->arch.complete_userspace_io =
7634 complete_fast_pio_out_port_0x7e;
7635 kvm_skip_emulated_instruction(vcpu);
7637 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7638 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7643 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7647 /* We should only ever be called with arch.pio.count equal to 1 */
7648 BUG_ON(vcpu->arch.pio.count != 1);
7650 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7651 vcpu->arch.pio.count = 0;
7655 /* For size less than 4 we merge, else we zero extend */
7656 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7659 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7660 * the copy and tracing
7662 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7663 kvm_rax_write(vcpu, val);
7665 return kvm_skip_emulated_instruction(vcpu);
7668 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7669 unsigned short port)
7674 /* For size less than 4 we merge, else we zero extend */
7675 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7677 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7679 kvm_rax_write(vcpu, val);
7683 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7684 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7689 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7694 ret = kvm_fast_pio_in(vcpu, size, port);
7696 ret = kvm_fast_pio_out(vcpu, size, port);
7697 return ret && kvm_skip_emulated_instruction(vcpu);
7699 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7701 static int kvmclock_cpu_down_prep(unsigned int cpu)
7703 __this_cpu_write(cpu_tsc_khz, 0);
7707 static void tsc_khz_changed(void *data)
7709 struct cpufreq_freqs *freq = data;
7710 unsigned long khz = 0;
7714 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7715 khz = cpufreq_quick_get(raw_smp_processor_id());
7718 __this_cpu_write(cpu_tsc_khz, khz);
7721 #ifdef CONFIG_X86_64
7722 static void kvm_hyperv_tsc_notifier(void)
7725 struct kvm_vcpu *vcpu;
7728 mutex_lock(&kvm_lock);
7729 list_for_each_entry(kvm, &vm_list, vm_list)
7730 kvm_make_mclock_inprogress_request(kvm);
7732 hyperv_stop_tsc_emulation();
7734 /* TSC frequency always matches when on Hyper-V */
7735 for_each_present_cpu(cpu)
7736 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7737 kvm_max_guest_tsc_khz = tsc_khz;
7739 list_for_each_entry(kvm, &vm_list, vm_list) {
7740 struct kvm_arch *ka = &kvm->arch;
7742 spin_lock(&ka->pvclock_gtod_sync_lock);
7744 pvclock_update_vm_gtod_copy(kvm);
7746 kvm_for_each_vcpu(cpu, vcpu, kvm)
7747 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7749 kvm_for_each_vcpu(cpu, vcpu, kvm)
7750 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7752 spin_unlock(&ka->pvclock_gtod_sync_lock);
7754 mutex_unlock(&kvm_lock);
7758 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7761 struct kvm_vcpu *vcpu;
7762 int i, send_ipi = 0;
7765 * We allow guests to temporarily run on slowing clocks,
7766 * provided we notify them after, or to run on accelerating
7767 * clocks, provided we notify them before. Thus time never
7770 * However, we have a problem. We can't atomically update
7771 * the frequency of a given CPU from this function; it is
7772 * merely a notifier, which can be called from any CPU.
7773 * Changing the TSC frequency at arbitrary points in time
7774 * requires a recomputation of local variables related to
7775 * the TSC for each VCPU. We must flag these local variables
7776 * to be updated and be sure the update takes place with the
7777 * new frequency before any guests proceed.
7779 * Unfortunately, the combination of hotplug CPU and frequency
7780 * change creates an intractable locking scenario; the order
7781 * of when these callouts happen is undefined with respect to
7782 * CPU hotplug, and they can race with each other. As such,
7783 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7784 * undefined; you can actually have a CPU frequency change take
7785 * place in between the computation of X and the setting of the
7786 * variable. To protect against this problem, all updates of
7787 * the per_cpu tsc_khz variable are done in an interrupt
7788 * protected IPI, and all callers wishing to update the value
7789 * must wait for a synchronous IPI to complete (which is trivial
7790 * if the caller is on the CPU already). This establishes the
7791 * necessary total order on variable updates.
7793 * Note that because a guest time update may take place
7794 * anytime after the setting of the VCPU's request bit, the
7795 * correct TSC value must be set before the request. However,
7796 * to ensure the update actually makes it to any guest which
7797 * starts running in hardware virtualization between the set
7798 * and the acquisition of the spinlock, we must also ping the
7799 * CPU after setting the request bit.
7803 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7805 mutex_lock(&kvm_lock);
7806 list_for_each_entry(kvm, &vm_list, vm_list) {
7807 kvm_for_each_vcpu(i, vcpu, kvm) {
7808 if (vcpu->cpu != cpu)
7810 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7811 if (vcpu->cpu != raw_smp_processor_id())
7815 mutex_unlock(&kvm_lock);
7817 if (freq->old < freq->new && send_ipi) {
7819 * We upscale the frequency. Must make the guest
7820 * doesn't see old kvmclock values while running with
7821 * the new frequency, otherwise we risk the guest sees
7822 * time go backwards.
7824 * In case we update the frequency for another cpu
7825 * (which might be in guest context) send an interrupt
7826 * to kick the cpu out of guest context. Next time
7827 * guest context is entered kvmclock will be updated,
7828 * so the guest will not see stale values.
7830 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7834 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7837 struct cpufreq_freqs *freq = data;
7840 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7842 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7845 for_each_cpu(cpu, freq->policy->cpus)
7846 __kvmclock_cpufreq_notifier(freq, cpu);
7851 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7852 .notifier_call = kvmclock_cpufreq_notifier
7855 static int kvmclock_cpu_online(unsigned int cpu)
7857 tsc_khz_changed(NULL);
7861 static void kvm_timer_init(void)
7863 max_tsc_khz = tsc_khz;
7865 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7866 #ifdef CONFIG_CPU_FREQ
7867 struct cpufreq_policy *policy;
7871 policy = cpufreq_cpu_get(cpu);
7873 if (policy->cpuinfo.max_freq)
7874 max_tsc_khz = policy->cpuinfo.max_freq;
7875 cpufreq_cpu_put(policy);
7879 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7880 CPUFREQ_TRANSITION_NOTIFIER);
7883 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7884 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7887 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7888 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7890 int kvm_is_in_guest(void)
7892 return __this_cpu_read(current_vcpu) != NULL;
7895 static int kvm_is_user_mode(void)
7899 if (__this_cpu_read(current_vcpu))
7900 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
7902 return user_mode != 0;
7905 static unsigned long kvm_get_guest_ip(void)
7907 unsigned long ip = 0;
7909 if (__this_cpu_read(current_vcpu))
7910 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7915 static void kvm_handle_intel_pt_intr(void)
7917 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7919 kvm_make_request(KVM_REQ_PMI, vcpu);
7920 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7921 (unsigned long *)&vcpu->arch.pmu.global_status);
7924 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7925 .is_in_guest = kvm_is_in_guest,
7926 .is_user_mode = kvm_is_user_mode,
7927 .get_guest_ip = kvm_get_guest_ip,
7928 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7931 #ifdef CONFIG_X86_64
7932 static void pvclock_gtod_update_fn(struct work_struct *work)
7936 struct kvm_vcpu *vcpu;
7939 mutex_lock(&kvm_lock);
7940 list_for_each_entry(kvm, &vm_list, vm_list)
7941 kvm_for_each_vcpu(i, vcpu, kvm)
7942 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7943 atomic_set(&kvm_guest_has_master_clock, 0);
7944 mutex_unlock(&kvm_lock);
7947 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7950 * Notification about pvclock gtod data update.
7952 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7955 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7956 struct timekeeper *tk = priv;
7958 update_pvclock_gtod(tk);
7960 /* disable master clock if host does not trust, or does not
7961 * use, TSC based clocksource.
7963 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7964 atomic_read(&kvm_guest_has_master_clock) != 0)
7965 queue_work(system_long_wq, &pvclock_gtod_work);
7970 static struct notifier_block pvclock_gtod_notifier = {
7971 .notifier_call = pvclock_gtod_notify,
7975 int kvm_arch_init(void *opaque)
7977 struct kvm_x86_init_ops *ops = opaque;
7980 if (kvm_x86_ops.hardware_enable) {
7981 printk(KERN_ERR "kvm: already loaded the other module\n");
7986 if (!ops->cpu_has_kvm_support()) {
7987 pr_err_ratelimited("kvm: no hardware support\n");
7991 if (ops->disabled_by_bios()) {
7992 pr_err_ratelimited("kvm: disabled by bios\n");
7998 * KVM explicitly assumes that the guest has an FPU and
7999 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8000 * vCPU's FPU state as a fxregs_state struct.
8002 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8003 printk(KERN_ERR "kvm: inadequate fpu\n");
8009 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8010 __alignof__(struct fpu), SLAB_ACCOUNT,
8012 if (!x86_fpu_cache) {
8013 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8017 x86_emulator_cache = kvm_alloc_emulator_cache();
8018 if (!x86_emulator_cache) {
8019 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8020 goto out_free_x86_fpu_cache;
8023 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8024 if (!user_return_msrs) {
8025 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8026 goto out_free_x86_emulator_cache;
8029 r = kvm_mmu_module_init();
8031 goto out_free_percpu;
8033 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
8034 PT_DIRTY_MASK, PT64_NX_MASK, 0,
8035 PT_PRESENT_MASK, 0, sme_me_mask);
8038 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8040 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8041 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8042 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8045 if (pi_inject_timer == -1)
8046 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8047 #ifdef CONFIG_X86_64
8048 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8050 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8051 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8057 free_percpu(user_return_msrs);
8058 out_free_x86_emulator_cache:
8059 kmem_cache_destroy(x86_emulator_cache);
8060 out_free_x86_fpu_cache:
8061 kmem_cache_destroy(x86_fpu_cache);
8066 void kvm_arch_exit(void)
8068 #ifdef CONFIG_X86_64
8069 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8070 clear_hv_tscchange_cb();
8073 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8075 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8076 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8077 CPUFREQ_TRANSITION_NOTIFIER);
8078 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8079 #ifdef CONFIG_X86_64
8080 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8082 kvm_x86_ops.hardware_enable = NULL;
8083 kvm_mmu_module_exit();
8084 free_percpu(user_return_msrs);
8085 kmem_cache_destroy(x86_fpu_cache);
8086 #ifdef CONFIG_KVM_XEN
8087 static_key_deferred_flush(&kvm_xen_enabled);
8088 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8092 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8094 ++vcpu->stat.halt_exits;
8095 if (lapic_in_kernel(vcpu)) {
8096 vcpu->arch.mp_state = state;
8099 vcpu->run->exit_reason = reason;
8104 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8106 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8108 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8110 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8112 int ret = kvm_skip_emulated_instruction(vcpu);
8114 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8115 * KVM_EXIT_DEBUG here.
8117 return kvm_vcpu_halt(vcpu) && ret;
8119 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8121 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8123 int ret = kvm_skip_emulated_instruction(vcpu);
8125 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8127 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8129 #ifdef CONFIG_X86_64
8130 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8131 unsigned long clock_type)
8133 struct kvm_clock_pairing clock_pairing;
8134 struct timespec64 ts;
8138 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8139 return -KVM_EOPNOTSUPP;
8141 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8142 return -KVM_EOPNOTSUPP;
8144 clock_pairing.sec = ts.tv_sec;
8145 clock_pairing.nsec = ts.tv_nsec;
8146 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8147 clock_pairing.flags = 0;
8148 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8151 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8152 sizeof(struct kvm_clock_pairing)))
8160 * kvm_pv_kick_cpu_op: Kick a vcpu.
8162 * @apicid - apicid of vcpu to be kicked.
8164 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8166 struct kvm_lapic_irq lapic_irq;
8168 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8169 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8170 lapic_irq.level = 0;
8171 lapic_irq.dest_id = apicid;
8172 lapic_irq.msi_redir_hint = false;
8174 lapic_irq.delivery_mode = APIC_DM_REMRD;
8175 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8178 bool kvm_apicv_activated(struct kvm *kvm)
8180 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8182 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8184 void kvm_apicv_init(struct kvm *kvm, bool enable)
8187 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8188 &kvm->arch.apicv_inhibit_reasons);
8190 set_bit(APICV_INHIBIT_REASON_DISABLE,
8191 &kvm->arch.apicv_inhibit_reasons);
8193 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8195 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
8197 struct kvm_vcpu *target = NULL;
8198 struct kvm_apic_map *map;
8201 map = rcu_dereference(kvm->arch.apic_map);
8203 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8204 target = map->phys_map[dest_id]->vcpu;
8208 if (target && READ_ONCE(target->ready))
8209 kvm_vcpu_yield_to(target);
8212 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8214 unsigned long nr, a0, a1, a2, a3, ret;
8217 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8218 return kvm_xen_hypercall(vcpu);
8220 if (kvm_hv_hypercall_enabled(vcpu))
8221 return kvm_hv_hypercall(vcpu);
8223 nr = kvm_rax_read(vcpu);
8224 a0 = kvm_rbx_read(vcpu);
8225 a1 = kvm_rcx_read(vcpu);
8226 a2 = kvm_rdx_read(vcpu);
8227 a3 = kvm_rsi_read(vcpu);
8229 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8231 op_64_bit = is_64_bit_mode(vcpu);
8240 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8248 case KVM_HC_VAPIC_POLL_IRQ:
8251 case KVM_HC_KICK_CPU:
8252 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8255 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8256 kvm_sched_yield(vcpu->kvm, a1);
8259 #ifdef CONFIG_X86_64
8260 case KVM_HC_CLOCK_PAIRING:
8261 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8264 case KVM_HC_SEND_IPI:
8265 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8268 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8270 case KVM_HC_SCHED_YIELD:
8271 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8274 kvm_sched_yield(vcpu->kvm, a0);
8284 kvm_rax_write(vcpu, ret);
8286 ++vcpu->stat.hypercalls;
8287 return kvm_skip_emulated_instruction(vcpu);
8289 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8291 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8293 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8294 char instruction[3];
8295 unsigned long rip = kvm_rip_read(vcpu);
8297 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8299 return emulator_write_emulated(ctxt, rip, instruction, 3,
8303 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8305 return vcpu->run->request_interrupt_window &&
8306 likely(!pic_in_kernel(vcpu->kvm));
8309 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8311 struct kvm_run *kvm_run = vcpu->run;
8314 * if_flag is obsolete and useless, so do not bother
8315 * setting it for SEV-ES guests. Userspace can just
8316 * use kvm_run->ready_for_interrupt_injection.
8318 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8319 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8321 kvm_run->cr8 = kvm_get_cr8(vcpu);
8322 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8323 kvm_run->ready_for_interrupt_injection =
8324 pic_in_kernel(vcpu->kvm) ||
8325 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8328 kvm_run->flags |= KVM_RUN_X86_SMM;
8331 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8335 if (!kvm_x86_ops.update_cr8_intercept)
8338 if (!lapic_in_kernel(vcpu))
8341 if (vcpu->arch.apicv_active)
8344 if (!vcpu->arch.apic->vapic_addr)
8345 max_irr = kvm_lapic_find_highest_irr(vcpu);
8352 tpr = kvm_lapic_get_cr8(vcpu);
8354 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8357 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8360 bool can_inject = true;
8362 /* try to reinject previous events if any */
8364 if (vcpu->arch.exception.injected) {
8365 static_call(kvm_x86_queue_exception)(vcpu);
8369 * Do not inject an NMI or interrupt if there is a pending
8370 * exception. Exceptions and interrupts are recognized at
8371 * instruction boundaries, i.e. the start of an instruction.
8372 * Trap-like exceptions, e.g. #DB, have higher priority than
8373 * NMIs and interrupts, i.e. traps are recognized before an
8374 * NMI/interrupt that's pending on the same instruction.
8375 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8376 * priority, but are only generated (pended) during instruction
8377 * execution, i.e. a pending fault-like exception means the
8378 * fault occurred on the *previous* instruction and must be
8379 * serviced prior to recognizing any new events in order to
8380 * fully complete the previous instruction.
8382 else if (!vcpu->arch.exception.pending) {
8383 if (vcpu->arch.nmi_injected) {
8384 static_call(kvm_x86_set_nmi)(vcpu);
8386 } else if (vcpu->arch.interrupt.injected) {
8387 static_call(kvm_x86_set_irq)(vcpu);
8392 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8393 vcpu->arch.exception.pending);
8396 * Call check_nested_events() even if we reinjected a previous event
8397 * in order for caller to determine if it should require immediate-exit
8398 * from L2 to L1 due to pending L1 events which require exit
8401 if (is_guest_mode(vcpu)) {
8402 r = kvm_x86_ops.nested_ops->check_events(vcpu);
8407 /* try to inject new event if pending */
8408 if (vcpu->arch.exception.pending) {
8409 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8410 vcpu->arch.exception.has_error_code,
8411 vcpu->arch.exception.error_code);
8413 vcpu->arch.exception.pending = false;
8414 vcpu->arch.exception.injected = true;
8416 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8417 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8420 if (vcpu->arch.exception.nr == DB_VECTOR) {
8421 kvm_deliver_exception_payload(vcpu);
8422 if (vcpu->arch.dr7 & DR7_GD) {
8423 vcpu->arch.dr7 &= ~DR7_GD;
8424 kvm_update_dr7(vcpu);
8428 static_call(kvm_x86_queue_exception)(vcpu);
8433 * Finally, inject interrupt events. If an event cannot be injected
8434 * due to architectural conditions (e.g. IF=0) a window-open exit
8435 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8436 * and can architecturally be injected, but we cannot do it right now:
8437 * an interrupt could have arrived just now and we have to inject it
8438 * as a vmexit, or there could already an event in the queue, which is
8439 * indicated by can_inject. In that case we request an immediate exit
8440 * in order to make progress and get back here for another iteration.
8441 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8443 if (vcpu->arch.smi_pending) {
8444 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8448 vcpu->arch.smi_pending = false;
8449 ++vcpu->arch.smi_count;
8453 static_call(kvm_x86_enable_smi_window)(vcpu);
8456 if (vcpu->arch.nmi_pending) {
8457 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8461 --vcpu->arch.nmi_pending;
8462 vcpu->arch.nmi_injected = true;
8463 static_call(kvm_x86_set_nmi)(vcpu);
8465 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8467 if (vcpu->arch.nmi_pending)
8468 static_call(kvm_x86_enable_nmi_window)(vcpu);
8471 if (kvm_cpu_has_injectable_intr(vcpu)) {
8472 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8476 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8477 static_call(kvm_x86_set_irq)(vcpu);
8478 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8480 if (kvm_cpu_has_injectable_intr(vcpu))
8481 static_call(kvm_x86_enable_irq_window)(vcpu);
8484 if (is_guest_mode(vcpu) &&
8485 kvm_x86_ops.nested_ops->hv_timer_pending &&
8486 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8487 *req_immediate_exit = true;
8489 WARN_ON(vcpu->arch.exception.pending);
8493 *req_immediate_exit = true;
8497 static void process_nmi(struct kvm_vcpu *vcpu)
8502 * x86 is limited to one NMI running, and one NMI pending after it.
8503 * If an NMI is already in progress, limit further NMIs to just one.
8504 * Otherwise, allow two (and we'll inject the first one immediately).
8506 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8509 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8510 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8511 kvm_make_request(KVM_REQ_EVENT, vcpu);
8514 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8517 flags |= seg->g << 23;
8518 flags |= seg->db << 22;
8519 flags |= seg->l << 21;
8520 flags |= seg->avl << 20;
8521 flags |= seg->present << 15;
8522 flags |= seg->dpl << 13;
8523 flags |= seg->s << 12;
8524 flags |= seg->type << 8;
8528 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8530 struct kvm_segment seg;
8533 kvm_get_segment(vcpu, &seg, n);
8534 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8537 offset = 0x7f84 + n * 12;
8539 offset = 0x7f2c + (n - 3) * 12;
8541 put_smstate(u32, buf, offset + 8, seg.base);
8542 put_smstate(u32, buf, offset + 4, seg.limit);
8543 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8546 #ifdef CONFIG_X86_64
8547 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8549 struct kvm_segment seg;
8553 kvm_get_segment(vcpu, &seg, n);
8554 offset = 0x7e00 + n * 16;
8556 flags = enter_smm_get_segment_flags(&seg) >> 8;
8557 put_smstate(u16, buf, offset, seg.selector);
8558 put_smstate(u16, buf, offset + 2, flags);
8559 put_smstate(u32, buf, offset + 4, seg.limit);
8560 put_smstate(u64, buf, offset + 8, seg.base);
8564 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8567 struct kvm_segment seg;
8571 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8572 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8573 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8574 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8576 for (i = 0; i < 8; i++)
8577 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
8579 kvm_get_dr(vcpu, 6, &val);
8580 put_smstate(u32, buf, 0x7fcc, (u32)val);
8581 kvm_get_dr(vcpu, 7, &val);
8582 put_smstate(u32, buf, 0x7fc8, (u32)val);
8584 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8585 put_smstate(u32, buf, 0x7fc4, seg.selector);
8586 put_smstate(u32, buf, 0x7f64, seg.base);
8587 put_smstate(u32, buf, 0x7f60, seg.limit);
8588 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8590 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8591 put_smstate(u32, buf, 0x7fc0, seg.selector);
8592 put_smstate(u32, buf, 0x7f80, seg.base);
8593 put_smstate(u32, buf, 0x7f7c, seg.limit);
8594 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8596 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8597 put_smstate(u32, buf, 0x7f74, dt.address);
8598 put_smstate(u32, buf, 0x7f70, dt.size);
8600 static_call(kvm_x86_get_idt)(vcpu, &dt);
8601 put_smstate(u32, buf, 0x7f58, dt.address);
8602 put_smstate(u32, buf, 0x7f54, dt.size);
8604 for (i = 0; i < 6; i++)
8605 enter_smm_save_seg_32(vcpu, buf, i);
8607 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8610 put_smstate(u32, buf, 0x7efc, 0x00020000);
8611 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8614 #ifdef CONFIG_X86_64
8615 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8618 struct kvm_segment seg;
8622 for (i = 0; i < 16; i++)
8623 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8625 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8626 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8628 kvm_get_dr(vcpu, 6, &val);
8629 put_smstate(u64, buf, 0x7f68, val);
8630 kvm_get_dr(vcpu, 7, &val);
8631 put_smstate(u64, buf, 0x7f60, val);
8633 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8634 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8635 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8637 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8640 put_smstate(u32, buf, 0x7efc, 0x00020064);
8642 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8644 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8645 put_smstate(u16, buf, 0x7e90, seg.selector);
8646 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8647 put_smstate(u32, buf, 0x7e94, seg.limit);
8648 put_smstate(u64, buf, 0x7e98, seg.base);
8650 static_call(kvm_x86_get_idt)(vcpu, &dt);
8651 put_smstate(u32, buf, 0x7e84, dt.size);
8652 put_smstate(u64, buf, 0x7e88, dt.address);
8654 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8655 put_smstate(u16, buf, 0x7e70, seg.selector);
8656 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8657 put_smstate(u32, buf, 0x7e74, seg.limit);
8658 put_smstate(u64, buf, 0x7e78, seg.base);
8660 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8661 put_smstate(u32, buf, 0x7e64, dt.size);
8662 put_smstate(u64, buf, 0x7e68, dt.address);
8664 for (i = 0; i < 6; i++)
8665 enter_smm_save_seg_64(vcpu, buf, i);
8669 static void enter_smm(struct kvm_vcpu *vcpu)
8671 struct kvm_segment cs, ds;
8676 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8677 memset(buf, 0, 512);
8678 #ifdef CONFIG_X86_64
8679 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8680 enter_smm_save_state_64(vcpu, buf);
8683 enter_smm_save_state_32(vcpu, buf);
8686 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8687 * vCPU state (e.g. leave guest mode) after we've saved the state into
8688 * the SMM state-save area.
8690 static_call(kvm_x86_pre_enter_smm)(vcpu, buf);
8692 vcpu->arch.hflags |= HF_SMM_MASK;
8693 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8695 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
8696 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8698 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
8700 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8701 kvm_rip_write(vcpu, 0x8000);
8703 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8704 static_call(kvm_x86_set_cr0)(vcpu, cr0);
8705 vcpu->arch.cr0 = cr0;
8707 static_call(kvm_x86_set_cr4)(vcpu, 0);
8709 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8710 dt.address = dt.size = 0;
8711 static_call(kvm_x86_set_idt)(vcpu, &dt);
8713 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8715 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8716 cs.base = vcpu->arch.smbase;
8721 cs.limit = ds.limit = 0xffffffff;
8722 cs.type = ds.type = 0x3;
8723 cs.dpl = ds.dpl = 0;
8728 cs.avl = ds.avl = 0;
8729 cs.present = ds.present = 1;
8730 cs.unusable = ds.unusable = 0;
8731 cs.padding = ds.padding = 0;
8733 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8734 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8735 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8736 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8737 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8738 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8740 #ifdef CONFIG_X86_64
8741 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8742 static_call(kvm_x86_set_efer)(vcpu, 0);
8745 kvm_update_cpuid_runtime(vcpu);
8746 kvm_mmu_reset_context(vcpu);
8749 static void process_smi(struct kvm_vcpu *vcpu)
8751 vcpu->arch.smi_pending = true;
8752 kvm_make_request(KVM_REQ_EVENT, vcpu);
8755 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8756 unsigned long *vcpu_bitmap)
8760 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8762 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8763 NULL, vcpu_bitmap, cpus);
8765 free_cpumask_var(cpus);
8768 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8770 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8773 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8775 if (!lapic_in_kernel(vcpu))
8778 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8779 kvm_apic_update_apicv(vcpu);
8780 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
8782 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8785 * NOTE: Do not hold any lock prior to calling this.
8787 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8788 * locked, because it calls __x86_set_memory_region() which does
8789 * synchronize_srcu(&kvm->srcu).
8791 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8793 struct kvm_vcpu *except;
8794 unsigned long old, new, expected;
8796 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8797 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
8800 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8802 expected = new = old;
8804 __clear_bit(bit, &new);
8806 __set_bit(bit, &new);
8809 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8810 } while (old != expected);
8815 trace_kvm_apicv_update_request(activate, bit);
8816 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8817 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
8820 * Sending request to update APICV for all other vcpus,
8821 * while update the calling vcpu immediately instead of
8822 * waiting for another #VMEXIT to handle the request.
8824 except = kvm_get_running_vcpu();
8825 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8828 kvm_vcpu_update_apicv(except);
8830 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8832 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8834 if (!kvm_apic_present(vcpu))
8837 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8839 if (irqchip_split(vcpu->kvm))
8840 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8842 if (vcpu->arch.apicv_active)
8843 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
8844 if (ioapic_in_kernel(vcpu->kvm))
8845 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8848 if (is_guest_mode(vcpu))
8849 vcpu->arch.load_eoi_exitmap_pending = true;
8851 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8854 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8856 u64 eoi_exit_bitmap[4];
8858 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8861 if (to_hv_vcpu(vcpu))
8862 bitmap_or((ulong *)eoi_exit_bitmap,
8863 vcpu->arch.ioapic_handled_vectors,
8864 to_hv_synic(vcpu)->vec_bitmap, 256);
8866 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
8869 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8870 unsigned long start, unsigned long end)
8872 unsigned long apic_address;
8875 * The physical address of apic access page is stored in the VMCS.
8876 * Update it when it becomes invalid.
8878 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8879 if (start <= apic_address && apic_address < end)
8880 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8883 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8885 if (!lapic_in_kernel(vcpu))
8888 if (!kvm_x86_ops.set_apic_access_page_addr)
8891 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
8894 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8896 smp_send_reschedule(vcpu->cpu);
8898 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8901 * Returns 1 to let vcpu_run() continue the guest execution loop without
8902 * exiting to the userspace. Otherwise, the value will be returned to the
8905 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8909 dm_request_for_irq_injection(vcpu) &&
8910 kvm_cpu_accept_dm_intr(vcpu);
8911 fastpath_t exit_fastpath;
8913 bool req_immediate_exit = false;
8915 /* Forbid vmenter if vcpu dirty ring is soft-full */
8916 if (unlikely(vcpu->kvm->dirty_ring_size &&
8917 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
8918 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
8919 trace_kvm_dirty_ring_exit(vcpu);
8924 if (kvm_request_pending(vcpu)) {
8925 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
8926 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
8931 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8932 kvm_mmu_unload(vcpu);
8933 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8934 __kvm_migrate_timers(vcpu);
8935 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8936 kvm_gen_update_masterclock(vcpu->kvm);
8937 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8938 kvm_gen_kvmclock_update(vcpu);
8939 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8940 r = kvm_guest_time_update(vcpu);
8944 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8945 kvm_mmu_sync_roots(vcpu);
8946 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8947 kvm_mmu_load_pgd(vcpu);
8948 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8949 kvm_vcpu_flush_tlb_all(vcpu);
8951 /* Flushing all ASIDs flushes the current ASID... */
8952 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8954 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8955 kvm_vcpu_flush_tlb_current(vcpu);
8956 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8957 kvm_vcpu_flush_tlb_guest(vcpu);
8959 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8960 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8964 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8965 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8966 vcpu->mmio_needed = 0;
8970 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8971 /* Page is swapped out. Do synthetic halt */
8972 vcpu->arch.apf.halted = true;
8976 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8977 record_steal_time(vcpu);
8978 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8980 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8982 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8983 kvm_pmu_handle_event(vcpu);
8984 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8985 kvm_pmu_deliver_pmi(vcpu);
8986 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8987 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8988 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8989 vcpu->arch.ioapic_handled_vectors)) {
8990 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8991 vcpu->run->eoi.vector =
8992 vcpu->arch.pending_ioapic_eoi;
8997 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8998 vcpu_scan_ioapic(vcpu);
8999 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9000 vcpu_load_eoi_exitmap(vcpu);
9001 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9002 kvm_vcpu_reload_apic_access_page(vcpu);
9003 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9004 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9005 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9009 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9010 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9011 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9015 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9016 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9018 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9019 vcpu->run->hyperv = hv_vcpu->exit;
9025 * KVM_REQ_HV_STIMER has to be processed after
9026 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9027 * depend on the guest clock being up-to-date
9029 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9030 kvm_hv_process_stimers(vcpu);
9031 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9032 kvm_vcpu_update_apicv(vcpu);
9033 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9034 kvm_check_async_pf_completion(vcpu);
9035 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9036 static_call(kvm_x86_msr_filter_changed)(vcpu);
9038 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9039 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9042 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9043 kvm_xen_has_interrupt(vcpu)) {
9044 ++vcpu->stat.req_event;
9045 kvm_apic_accept_events(vcpu);
9046 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9051 inject_pending_event(vcpu, &req_immediate_exit);
9053 static_call(kvm_x86_enable_irq_window)(vcpu);
9055 if (kvm_lapic_enabled(vcpu)) {
9056 update_cr8_intercept(vcpu);
9057 kvm_lapic_sync_to_vapic(vcpu);
9061 r = kvm_mmu_reload(vcpu);
9063 goto cancel_injection;
9068 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9071 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9072 * IPI are then delayed after guest entry, which ensures that they
9073 * result in virtual interrupt delivery.
9075 local_irq_disable();
9076 vcpu->mode = IN_GUEST_MODE;
9078 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9081 * 1) We should set ->mode before checking ->requests. Please see
9082 * the comment in kvm_vcpu_exiting_guest_mode().
9084 * 2) For APICv, we should set ->mode before checking PID.ON. This
9085 * pairs with the memory barrier implicit in pi_test_and_set_on
9086 * (see vmx_deliver_posted_interrupt).
9088 * 3) This also orders the write to mode from any reads to the page
9089 * tables done while the VCPU is running. Please see the comment
9090 * in kvm_flush_remote_tlbs.
9092 smp_mb__after_srcu_read_unlock();
9095 * This handles the case where a posted interrupt was
9096 * notified with kvm_vcpu_kick.
9098 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9099 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9101 if (kvm_vcpu_exit_request(vcpu)) {
9102 vcpu->mode = OUTSIDE_GUEST_MODE;
9106 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9108 goto cancel_injection;
9111 if (req_immediate_exit) {
9112 kvm_make_request(KVM_REQ_EVENT, vcpu);
9113 static_call(kvm_x86_request_immediate_exit)(vcpu);
9116 fpregs_assert_state_consistent();
9117 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9118 switch_fpu_return();
9120 if (unlikely(vcpu->arch.switch_db_regs)) {
9122 set_debugreg(vcpu->arch.eff_db[0], 0);
9123 set_debugreg(vcpu->arch.eff_db[1], 1);
9124 set_debugreg(vcpu->arch.eff_db[2], 2);
9125 set_debugreg(vcpu->arch.eff_db[3], 3);
9126 set_debugreg(vcpu->arch.dr6, 6);
9127 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9131 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9132 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9135 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9136 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9140 if (vcpu->arch.apicv_active)
9141 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9145 * Do this here before restoring debug registers on the host. And
9146 * since we do this before handling the vmexit, a DR access vmexit
9147 * can (a) read the correct value of the debug registers, (b) set
9148 * KVM_DEBUGREG_WONT_EXIT again.
9150 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9151 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9152 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9153 kvm_update_dr0123(vcpu);
9154 kvm_update_dr7(vcpu);
9155 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9159 * If the guest has used debug registers, at least dr7
9160 * will be disabled while returning to the host.
9161 * If we don't have active breakpoints in the host, we don't
9162 * care about the messed up debug address registers. But if
9163 * we have some of them active, restore the old state.
9165 if (hw_breakpoint_active())
9166 hw_breakpoint_restore();
9168 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9169 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9171 vcpu->mode = OUTSIDE_GUEST_MODE;
9174 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9177 * Consume any pending interrupts, including the possible source of
9178 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9179 * An instruction is required after local_irq_enable() to fully unblock
9180 * interrupts on processors that implement an interrupt shadow, the
9181 * stat.exits increment will do nicely.
9183 kvm_before_interrupt(vcpu);
9186 local_irq_disable();
9187 kvm_after_interrupt(vcpu);
9189 if (lapic_in_kernel(vcpu)) {
9190 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9191 if (delta != S64_MIN) {
9192 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9193 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9200 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9203 * Profile KVM exit RIPs:
9205 if (unlikely(prof_on == KVM_PROFILING)) {
9206 unsigned long rip = kvm_rip_read(vcpu);
9207 profile_hit(KVM_PROFILING, (void *)rip);
9210 if (unlikely(vcpu->arch.tsc_always_catchup))
9211 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9213 if (vcpu->arch.apic_attention)
9214 kvm_lapic_sync_from_vapic(vcpu);
9216 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9220 if (req_immediate_exit)
9221 kvm_make_request(KVM_REQ_EVENT, vcpu);
9222 static_call(kvm_x86_cancel_injection)(vcpu);
9223 if (unlikely(vcpu->arch.apic_attention))
9224 kvm_lapic_sync_from_vapic(vcpu);
9229 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9231 if (!kvm_arch_vcpu_runnable(vcpu) &&
9232 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9233 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9234 kvm_vcpu_block(vcpu);
9235 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9237 if (kvm_x86_ops.post_block)
9238 static_call(kvm_x86_post_block)(vcpu);
9240 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9244 kvm_apic_accept_events(vcpu);
9245 switch(vcpu->arch.mp_state) {
9246 case KVM_MP_STATE_HALTED:
9247 case KVM_MP_STATE_AP_RESET_HOLD:
9248 vcpu->arch.pv.pv_unhalted = false;
9249 vcpu->arch.mp_state =
9250 KVM_MP_STATE_RUNNABLE;
9252 case KVM_MP_STATE_RUNNABLE:
9253 vcpu->arch.apf.halted = false;
9255 case KVM_MP_STATE_INIT_RECEIVED:
9263 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9265 if (is_guest_mode(vcpu))
9266 kvm_x86_ops.nested_ops->check_events(vcpu);
9268 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9269 !vcpu->arch.apf.halted);
9272 static int vcpu_run(struct kvm_vcpu *vcpu)
9275 struct kvm *kvm = vcpu->kvm;
9277 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9278 vcpu->arch.l1tf_flush_l1d = true;
9281 if (kvm_vcpu_running(vcpu)) {
9282 r = vcpu_enter_guest(vcpu);
9284 r = vcpu_block(kvm, vcpu);
9290 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9291 if (kvm_cpu_has_pending_timer(vcpu))
9292 kvm_inject_pending_timer_irqs(vcpu);
9294 if (dm_request_for_irq_injection(vcpu) &&
9295 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9297 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9298 ++vcpu->stat.request_irq_exits;
9302 if (__xfer_to_guest_mode_work_pending()) {
9303 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9304 r = xfer_to_guest_mode_handle_work(vcpu);
9307 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9311 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9316 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9320 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9321 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9322 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9326 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9328 BUG_ON(!vcpu->arch.pio.count);
9330 return complete_emulated_io(vcpu);
9334 * Implements the following, as a state machine:
9338 * for each mmio piece in the fragment
9346 * for each mmio piece in the fragment
9351 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9353 struct kvm_run *run = vcpu->run;
9354 struct kvm_mmio_fragment *frag;
9357 BUG_ON(!vcpu->mmio_needed);
9359 /* Complete previous fragment */
9360 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9361 len = min(8u, frag->len);
9362 if (!vcpu->mmio_is_write)
9363 memcpy(frag->data, run->mmio.data, len);
9365 if (frag->len <= 8) {
9366 /* Switch to the next fragment. */
9368 vcpu->mmio_cur_fragment++;
9370 /* Go forward to the next mmio piece. */
9376 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9377 vcpu->mmio_needed = 0;
9379 /* FIXME: return into emulator if single-stepping. */
9380 if (vcpu->mmio_is_write)
9382 vcpu->mmio_read_completed = 1;
9383 return complete_emulated_io(vcpu);
9386 run->exit_reason = KVM_EXIT_MMIO;
9387 run->mmio.phys_addr = frag->gpa;
9388 if (vcpu->mmio_is_write)
9389 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9390 run->mmio.len = min(8u, frag->len);
9391 run->mmio.is_write = vcpu->mmio_is_write;
9392 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9396 static void kvm_save_current_fpu(struct fpu *fpu)
9399 * If the target FPU state is not resident in the CPU registers, just
9400 * memcpy() from current, else save CPU state directly to the target.
9402 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9403 memcpy(&fpu->state, ¤t->thread.fpu.state,
9404 fpu_kernel_xstate_size);
9406 copy_fpregs_to_fpstate(fpu);
9409 /* Swap (qemu) user FPU context for the guest FPU context. */
9410 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9414 kvm_save_current_fpu(vcpu->arch.user_fpu);
9417 * Guests with protected state can't have it set by the hypervisor,
9418 * so skip trying to set it.
9420 if (vcpu->arch.guest_fpu)
9421 /* PKRU is separately restored in kvm_x86_ops.run. */
9422 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9423 ~XFEATURE_MASK_PKRU);
9425 fpregs_mark_activate();
9431 /* When vcpu_run ends, restore user space FPU context. */
9432 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9437 * Guests with protected state can't have it read by the hypervisor,
9438 * so skip trying to save it.
9440 if (vcpu->arch.guest_fpu)
9441 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9443 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9445 fpregs_mark_activate();
9448 ++vcpu->stat.fpu_reload;
9452 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9454 struct kvm_run *kvm_run = vcpu->run;
9458 kvm_sigset_activate(vcpu);
9460 kvm_load_guest_fpu(vcpu);
9462 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9463 if (kvm_run->immediate_exit) {
9467 kvm_vcpu_block(vcpu);
9468 kvm_apic_accept_events(vcpu);
9469 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9471 if (signal_pending(current)) {
9473 kvm_run->exit_reason = KVM_EXIT_INTR;
9474 ++vcpu->stat.signal_exits;
9479 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9484 if (kvm_run->kvm_dirty_regs) {
9485 r = sync_regs(vcpu);
9490 /* re-sync apic's tpr */
9491 if (!lapic_in_kernel(vcpu)) {
9492 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9498 if (unlikely(vcpu->arch.complete_userspace_io)) {
9499 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9500 vcpu->arch.complete_userspace_io = NULL;
9505 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9507 if (kvm_run->immediate_exit)
9513 kvm_put_guest_fpu(vcpu);
9514 if (kvm_run->kvm_valid_regs)
9516 post_kvm_run_save(vcpu);
9517 kvm_sigset_deactivate(vcpu);
9523 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9525 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9527 * We are here if userspace calls get_regs() in the middle of
9528 * instruction emulation. Registers state needs to be copied
9529 * back from emulation context to vcpu. Userspace shouldn't do
9530 * that usually, but some bad designed PV devices (vmware
9531 * backdoor interface) need this to work
9533 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9534 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9536 regs->rax = kvm_rax_read(vcpu);
9537 regs->rbx = kvm_rbx_read(vcpu);
9538 regs->rcx = kvm_rcx_read(vcpu);
9539 regs->rdx = kvm_rdx_read(vcpu);
9540 regs->rsi = kvm_rsi_read(vcpu);
9541 regs->rdi = kvm_rdi_read(vcpu);
9542 regs->rsp = kvm_rsp_read(vcpu);
9543 regs->rbp = kvm_rbp_read(vcpu);
9544 #ifdef CONFIG_X86_64
9545 regs->r8 = kvm_r8_read(vcpu);
9546 regs->r9 = kvm_r9_read(vcpu);
9547 regs->r10 = kvm_r10_read(vcpu);
9548 regs->r11 = kvm_r11_read(vcpu);
9549 regs->r12 = kvm_r12_read(vcpu);
9550 regs->r13 = kvm_r13_read(vcpu);
9551 regs->r14 = kvm_r14_read(vcpu);
9552 regs->r15 = kvm_r15_read(vcpu);
9555 regs->rip = kvm_rip_read(vcpu);
9556 regs->rflags = kvm_get_rflags(vcpu);
9559 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9562 __get_regs(vcpu, regs);
9567 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9569 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9570 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9572 kvm_rax_write(vcpu, regs->rax);
9573 kvm_rbx_write(vcpu, regs->rbx);
9574 kvm_rcx_write(vcpu, regs->rcx);
9575 kvm_rdx_write(vcpu, regs->rdx);
9576 kvm_rsi_write(vcpu, regs->rsi);
9577 kvm_rdi_write(vcpu, regs->rdi);
9578 kvm_rsp_write(vcpu, regs->rsp);
9579 kvm_rbp_write(vcpu, regs->rbp);
9580 #ifdef CONFIG_X86_64
9581 kvm_r8_write(vcpu, regs->r8);
9582 kvm_r9_write(vcpu, regs->r9);
9583 kvm_r10_write(vcpu, regs->r10);
9584 kvm_r11_write(vcpu, regs->r11);
9585 kvm_r12_write(vcpu, regs->r12);
9586 kvm_r13_write(vcpu, regs->r13);
9587 kvm_r14_write(vcpu, regs->r14);
9588 kvm_r15_write(vcpu, regs->r15);
9591 kvm_rip_write(vcpu, regs->rip);
9592 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9594 vcpu->arch.exception.pending = false;
9596 kvm_make_request(KVM_REQ_EVENT, vcpu);
9599 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9602 __set_regs(vcpu, regs);
9607 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9609 struct kvm_segment cs;
9611 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9615 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9617 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9621 if (vcpu->arch.guest_state_protected)
9622 goto skip_protected_regs;
9624 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9625 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9626 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9627 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9628 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9629 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9631 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9632 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9634 static_call(kvm_x86_get_idt)(vcpu, &dt);
9635 sregs->idt.limit = dt.size;
9636 sregs->idt.base = dt.address;
9637 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9638 sregs->gdt.limit = dt.size;
9639 sregs->gdt.base = dt.address;
9641 sregs->cr2 = vcpu->arch.cr2;
9642 sregs->cr3 = kvm_read_cr3(vcpu);
9644 skip_protected_regs:
9645 sregs->cr0 = kvm_read_cr0(vcpu);
9646 sregs->cr4 = kvm_read_cr4(vcpu);
9647 sregs->cr8 = kvm_get_cr8(vcpu);
9648 sregs->efer = vcpu->arch.efer;
9649 sregs->apic_base = kvm_get_apic_base(vcpu);
9651 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9653 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9654 set_bit(vcpu->arch.interrupt.nr,
9655 (unsigned long *)sregs->interrupt_bitmap);
9658 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9659 struct kvm_sregs *sregs)
9662 __get_sregs(vcpu, sregs);
9667 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9668 struct kvm_mp_state *mp_state)
9671 if (kvm_mpx_supported())
9672 kvm_load_guest_fpu(vcpu);
9674 kvm_apic_accept_events(vcpu);
9675 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9676 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9677 vcpu->arch.pv.pv_unhalted)
9678 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9680 mp_state->mp_state = vcpu->arch.mp_state;
9682 if (kvm_mpx_supported())
9683 kvm_put_guest_fpu(vcpu);
9688 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9689 struct kvm_mp_state *mp_state)
9695 if (!lapic_in_kernel(vcpu) &&
9696 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9700 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9701 * INIT state; latched init should be reported using
9702 * KVM_SET_VCPU_EVENTS, so reject it here.
9704 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9705 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9706 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9709 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9710 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9711 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9713 vcpu->arch.mp_state = mp_state->mp_state;
9714 kvm_make_request(KVM_REQ_EVENT, vcpu);
9722 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9723 int reason, bool has_error_code, u32 error_code)
9725 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9728 init_emulate_ctxt(vcpu);
9730 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9731 has_error_code, error_code);
9733 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9734 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9735 vcpu->run->internal.ndata = 0;
9739 kvm_rip_write(vcpu, ctxt->eip);
9740 kvm_set_rflags(vcpu, ctxt->eflags);
9743 EXPORT_SYMBOL_GPL(kvm_task_switch);
9745 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9747 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9749 * When EFER.LME and CR0.PG are set, the processor is in
9750 * 64-bit mode (though maybe in a 32-bit code segment).
9751 * CR4.PAE and EFER.LMA must be set.
9753 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9755 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
9759 * Not in 64-bit mode: EFER.LMA is clear and the code
9760 * segment cannot be 64-bit.
9762 if (sregs->efer & EFER_LMA || sregs->cs.l)
9766 return kvm_is_valid_cr4(vcpu, sregs->cr4);
9769 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9771 struct msr_data apic_base_msr;
9772 int mmu_reset_needed = 0;
9773 int pending_vec, max_bits, idx;
9777 if (!kvm_is_valid_sregs(vcpu, sregs))
9780 apic_base_msr.data = sregs->apic_base;
9781 apic_base_msr.host_initiated = true;
9782 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9785 if (vcpu->arch.guest_state_protected)
9786 goto skip_protected_regs;
9788 dt.size = sregs->idt.limit;
9789 dt.address = sregs->idt.base;
9790 static_call(kvm_x86_set_idt)(vcpu, &dt);
9791 dt.size = sregs->gdt.limit;
9792 dt.address = sregs->gdt.base;
9793 static_call(kvm_x86_set_gdt)(vcpu, &dt);
9795 vcpu->arch.cr2 = sregs->cr2;
9796 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9797 vcpu->arch.cr3 = sregs->cr3;
9798 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9800 kvm_set_cr8(vcpu, sregs->cr8);
9802 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9803 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
9805 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9806 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
9807 vcpu->arch.cr0 = sregs->cr0;
9809 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9810 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
9812 idx = srcu_read_lock(&vcpu->kvm->srcu);
9813 if (is_pae_paging(vcpu)) {
9814 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9815 mmu_reset_needed = 1;
9817 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9819 if (mmu_reset_needed)
9820 kvm_mmu_reset_context(vcpu);
9822 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9823 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9824 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9825 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9826 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9827 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9829 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9830 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9832 update_cr8_intercept(vcpu);
9834 /* Older userspace won't unhalt the vcpu on reset. */
9835 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9836 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9838 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9840 skip_protected_regs:
9841 max_bits = KVM_NR_INTERRUPTS;
9842 pending_vec = find_first_bit(
9843 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9844 if (pending_vec < max_bits) {
9845 kvm_queue_interrupt(vcpu, pending_vec, false);
9846 pr_debug("Set back pending irq %d\n", pending_vec);
9849 kvm_make_request(KVM_REQ_EVENT, vcpu);
9856 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9857 struct kvm_sregs *sregs)
9862 ret = __set_sregs(vcpu, sregs);
9867 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9868 struct kvm_guest_debug *dbg)
9870 unsigned long rflags;
9873 if (vcpu->arch.guest_state_protected)
9878 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9880 if (vcpu->arch.exception.pending)
9882 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9883 kvm_queue_exception(vcpu, DB_VECTOR);
9885 kvm_queue_exception(vcpu, BP_VECTOR);
9889 * Read rflags as long as potentially injected trace flags are still
9892 rflags = kvm_get_rflags(vcpu);
9894 vcpu->guest_debug = dbg->control;
9895 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9896 vcpu->guest_debug = 0;
9898 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9899 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9900 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9901 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9903 for (i = 0; i < KVM_NR_DB_REGS; i++)
9904 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9906 kvm_update_dr7(vcpu);
9908 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9909 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9910 get_segment_base(vcpu, VCPU_SREG_CS);
9913 * Trigger an rflags update that will inject or remove the trace
9916 kvm_set_rflags(vcpu, rflags);
9918 static_call(kvm_x86_update_exception_bitmap)(vcpu);
9928 * Translate a guest virtual address to a guest physical address.
9930 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9931 struct kvm_translation *tr)
9933 unsigned long vaddr = tr->linear_address;
9939 idx = srcu_read_lock(&vcpu->kvm->srcu);
9940 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9941 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9942 tr->physical_address = gpa;
9943 tr->valid = gpa != UNMAPPED_GVA;
9951 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9953 struct fxregs_state *fxsave;
9955 if (!vcpu->arch.guest_fpu)
9960 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9961 memcpy(fpu->fpr, fxsave->st_space, 128);
9962 fpu->fcw = fxsave->cwd;
9963 fpu->fsw = fxsave->swd;
9964 fpu->ftwx = fxsave->twd;
9965 fpu->last_opcode = fxsave->fop;
9966 fpu->last_ip = fxsave->rip;
9967 fpu->last_dp = fxsave->rdp;
9968 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9974 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9976 struct fxregs_state *fxsave;
9978 if (!vcpu->arch.guest_fpu)
9983 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9985 memcpy(fxsave->st_space, fpu->fpr, 128);
9986 fxsave->cwd = fpu->fcw;
9987 fxsave->swd = fpu->fsw;
9988 fxsave->twd = fpu->ftwx;
9989 fxsave->fop = fpu->last_opcode;
9990 fxsave->rip = fpu->last_ip;
9991 fxsave->rdp = fpu->last_dp;
9992 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9998 static void store_regs(struct kvm_vcpu *vcpu)
10000 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10002 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10003 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10005 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10006 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10008 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10009 kvm_vcpu_ioctl_x86_get_vcpu_events(
10010 vcpu, &vcpu->run->s.regs.events);
10013 static int sync_regs(struct kvm_vcpu *vcpu)
10015 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10018 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10019 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10020 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10022 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10023 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10025 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10027 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10028 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10029 vcpu, &vcpu->run->s.regs.events))
10031 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10037 static void fx_init(struct kvm_vcpu *vcpu)
10039 if (!vcpu->arch.guest_fpu)
10042 fpstate_init(&vcpu->arch.guest_fpu->state);
10043 if (boot_cpu_has(X86_FEATURE_XSAVES))
10044 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10045 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10048 * Ensure guest xcr0 is valid for loading
10050 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10052 vcpu->arch.cr0 |= X86_CR0_ET;
10055 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10057 if (vcpu->arch.guest_fpu) {
10058 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10059 vcpu->arch.guest_fpu = NULL;
10062 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10064 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10066 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10067 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10068 "guest TSC will not be reliable\n");
10073 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10078 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10079 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10081 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10083 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10085 r = kvm_mmu_create(vcpu);
10089 if (irqchip_in_kernel(vcpu->kvm)) {
10090 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10092 goto fail_mmu_destroy;
10093 if (kvm_apicv_activated(vcpu->kvm))
10094 vcpu->arch.apicv_active = true;
10096 static_branch_inc(&kvm_has_noapic_vcpu);
10100 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10102 goto fail_free_lapic;
10103 vcpu->arch.pio_data = page_address(page);
10105 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10106 GFP_KERNEL_ACCOUNT);
10107 if (!vcpu->arch.mce_banks)
10108 goto fail_free_pio_data;
10109 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10111 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10112 GFP_KERNEL_ACCOUNT))
10113 goto fail_free_mce_banks;
10115 if (!alloc_emulate_ctxt(vcpu))
10116 goto free_wbinvd_dirty_mask;
10118 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10119 GFP_KERNEL_ACCOUNT);
10120 if (!vcpu->arch.user_fpu) {
10121 pr_err("kvm: failed to allocate userspace's fpu\n");
10122 goto free_emulate_ctxt;
10125 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10126 GFP_KERNEL_ACCOUNT);
10127 if (!vcpu->arch.guest_fpu) {
10128 pr_err("kvm: failed to allocate vcpu's fpu\n");
10129 goto free_user_fpu;
10133 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10134 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10136 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10138 kvm_async_pf_hash_reset(vcpu);
10139 kvm_pmu_init(vcpu);
10141 vcpu->arch.pending_external_vector = -1;
10142 vcpu->arch.preempted_in_kernel = false;
10144 r = static_call(kvm_x86_vcpu_create)(vcpu);
10146 goto free_guest_fpu;
10148 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10149 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10150 kvm_vcpu_mtrr_init(vcpu);
10152 kvm_vcpu_reset(vcpu, false);
10153 kvm_init_mmu(vcpu, false);
10158 kvm_free_guest_fpu(vcpu);
10160 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10162 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10163 free_wbinvd_dirty_mask:
10164 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10165 fail_free_mce_banks:
10166 kfree(vcpu->arch.mce_banks);
10167 fail_free_pio_data:
10168 free_page((unsigned long)vcpu->arch.pio_data);
10170 kvm_free_lapic(vcpu);
10172 kvm_mmu_destroy(vcpu);
10176 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10178 struct kvm *kvm = vcpu->kvm;
10180 if (mutex_lock_killable(&vcpu->mutex))
10183 kvm_synchronize_tsc(vcpu, 0);
10186 /* poll control enabled by default */
10187 vcpu->arch.msr_kvm_poll_control = 1;
10189 mutex_unlock(&vcpu->mutex);
10191 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10192 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10193 KVMCLOCK_SYNC_PERIOD);
10196 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10198 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10201 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10203 kvmclock_reset(vcpu);
10205 static_call(kvm_x86_vcpu_free)(vcpu);
10207 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10208 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10209 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10210 kvm_free_guest_fpu(vcpu);
10212 kvm_hv_vcpu_uninit(vcpu);
10213 kvm_pmu_destroy(vcpu);
10214 kfree(vcpu->arch.mce_banks);
10215 kvm_free_lapic(vcpu);
10216 idx = srcu_read_lock(&vcpu->kvm->srcu);
10217 kvm_mmu_destroy(vcpu);
10218 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10219 free_page((unsigned long)vcpu->arch.pio_data);
10220 kvfree(vcpu->arch.cpuid_entries);
10221 if (!lapic_in_kernel(vcpu))
10222 static_branch_dec(&kvm_has_noapic_vcpu);
10225 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10227 kvm_lapic_reset(vcpu, init_event);
10229 vcpu->arch.hflags = 0;
10231 vcpu->arch.smi_pending = 0;
10232 vcpu->arch.smi_count = 0;
10233 atomic_set(&vcpu->arch.nmi_queued, 0);
10234 vcpu->arch.nmi_pending = 0;
10235 vcpu->arch.nmi_injected = false;
10236 kvm_clear_interrupt_queue(vcpu);
10237 kvm_clear_exception_queue(vcpu);
10239 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10240 kvm_update_dr0123(vcpu);
10241 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10242 vcpu->arch.dr7 = DR7_FIXED_1;
10243 kvm_update_dr7(vcpu);
10245 vcpu->arch.cr2 = 0;
10247 kvm_make_request(KVM_REQ_EVENT, vcpu);
10248 vcpu->arch.apf.msr_en_val = 0;
10249 vcpu->arch.apf.msr_int_val = 0;
10250 vcpu->arch.st.msr_val = 0;
10252 kvmclock_reset(vcpu);
10254 kvm_clear_async_pf_completion_queue(vcpu);
10255 kvm_async_pf_hash_reset(vcpu);
10256 vcpu->arch.apf.halted = false;
10258 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10259 void *mpx_state_buffer;
10262 * To avoid have the INIT path from kvm_apic_has_events() that be
10263 * called with loaded FPU and does not let userspace fix the state.
10266 kvm_put_guest_fpu(vcpu);
10267 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10269 if (mpx_state_buffer)
10270 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10271 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10273 if (mpx_state_buffer)
10274 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10276 kvm_load_guest_fpu(vcpu);
10280 kvm_pmu_reset(vcpu);
10281 vcpu->arch.smbase = 0x30000;
10283 vcpu->arch.msr_misc_features_enables = 0;
10285 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10288 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10289 vcpu->arch.regs_avail = ~0;
10290 vcpu->arch.regs_dirty = ~0;
10292 vcpu->arch.ia32_xss = 0;
10294 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10297 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10299 struct kvm_segment cs;
10301 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10302 cs.selector = vector << 8;
10303 cs.base = vector << 12;
10304 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10305 kvm_rip_write(vcpu, 0);
10307 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10309 int kvm_arch_hardware_enable(void)
10312 struct kvm_vcpu *vcpu;
10317 bool stable, backwards_tsc = false;
10319 kvm_user_return_msr_cpu_online();
10320 ret = static_call(kvm_x86_hardware_enable)();
10324 local_tsc = rdtsc();
10325 stable = !kvm_check_tsc_unstable();
10326 list_for_each_entry(kvm, &vm_list, vm_list) {
10327 kvm_for_each_vcpu(i, vcpu, kvm) {
10328 if (!stable && vcpu->cpu == smp_processor_id())
10329 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10330 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10331 backwards_tsc = true;
10332 if (vcpu->arch.last_host_tsc > max_tsc)
10333 max_tsc = vcpu->arch.last_host_tsc;
10339 * Sometimes, even reliable TSCs go backwards. This happens on
10340 * platforms that reset TSC during suspend or hibernate actions, but
10341 * maintain synchronization. We must compensate. Fortunately, we can
10342 * detect that condition here, which happens early in CPU bringup,
10343 * before any KVM threads can be running. Unfortunately, we can't
10344 * bring the TSCs fully up to date with real time, as we aren't yet far
10345 * enough into CPU bringup that we know how much real time has actually
10346 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10347 * variables that haven't been updated yet.
10349 * So we simply find the maximum observed TSC above, then record the
10350 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10351 * the adjustment will be applied. Note that we accumulate
10352 * adjustments, in case multiple suspend cycles happen before some VCPU
10353 * gets a chance to run again. In the event that no KVM threads get a
10354 * chance to run, we will miss the entire elapsed period, as we'll have
10355 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10356 * loose cycle time. This isn't too big a deal, since the loss will be
10357 * uniform across all VCPUs (not to mention the scenario is extremely
10358 * unlikely). It is possible that a second hibernate recovery happens
10359 * much faster than a first, causing the observed TSC here to be
10360 * smaller; this would require additional padding adjustment, which is
10361 * why we set last_host_tsc to the local tsc observed here.
10363 * N.B. - this code below runs only on platforms with reliable TSC,
10364 * as that is the only way backwards_tsc is set above. Also note
10365 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10366 * have the same delta_cyc adjustment applied if backwards_tsc
10367 * is detected. Note further, this adjustment is only done once,
10368 * as we reset last_host_tsc on all VCPUs to stop this from being
10369 * called multiple times (one for each physical CPU bringup).
10371 * Platforms with unreliable TSCs don't have to deal with this, they
10372 * will be compensated by the logic in vcpu_load, which sets the TSC to
10373 * catchup mode. This will catchup all VCPUs to real time, but cannot
10374 * guarantee that they stay in perfect synchronization.
10376 if (backwards_tsc) {
10377 u64 delta_cyc = max_tsc - local_tsc;
10378 list_for_each_entry(kvm, &vm_list, vm_list) {
10379 kvm->arch.backwards_tsc_observed = true;
10380 kvm_for_each_vcpu(i, vcpu, kvm) {
10381 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10382 vcpu->arch.last_host_tsc = local_tsc;
10383 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10387 * We have to disable TSC offset matching.. if you were
10388 * booting a VM while issuing an S4 host suspend....
10389 * you may have some problem. Solving this issue is
10390 * left as an exercise to the reader.
10392 kvm->arch.last_tsc_nsec = 0;
10393 kvm->arch.last_tsc_write = 0;
10400 void kvm_arch_hardware_disable(void)
10402 static_call(kvm_x86_hardware_disable)();
10403 drop_user_return_notifiers();
10406 int kvm_arch_hardware_setup(void *opaque)
10408 struct kvm_x86_init_ops *ops = opaque;
10411 rdmsrl_safe(MSR_EFER, &host_efer);
10413 if (boot_cpu_has(X86_FEATURE_XSAVES))
10414 rdmsrl(MSR_IA32_XSS, host_xss);
10416 r = ops->hardware_setup();
10420 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10421 kvm_ops_static_call_update();
10423 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10426 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10427 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10428 #undef __kvm_cpu_cap_has
10430 if (kvm_has_tsc_control) {
10432 * Make sure the user can only configure tsc_khz values that
10433 * fit into a signed integer.
10434 * A min value is not calculated because it will always
10435 * be 1 on all machines.
10437 u64 max = min(0x7fffffffULL,
10438 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10439 kvm_max_guest_tsc_khz = max;
10441 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10444 kvm_init_msr_list();
10448 void kvm_arch_hardware_unsetup(void)
10450 static_call(kvm_x86_hardware_unsetup)();
10453 int kvm_arch_check_processor_compat(void *opaque)
10455 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10456 struct kvm_x86_init_ops *ops = opaque;
10458 WARN_ON(!irqs_disabled());
10460 if (__cr4_reserved_bits(cpu_has, c) !=
10461 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10464 return ops->check_processor_compatibility();
10467 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10469 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10471 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10473 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10475 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10478 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10479 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10481 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10483 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10485 vcpu->arch.l1tf_flush_l1d = true;
10486 if (pmu->version && unlikely(pmu->event_count)) {
10487 pmu->need_cleanup = true;
10488 kvm_make_request(KVM_REQ_PMU, vcpu);
10490 static_call(kvm_x86_sched_in)(vcpu, cpu);
10493 void kvm_arch_free_vm(struct kvm *kvm)
10495 kfree(to_kvm_hv(kvm)->hv_pa_pg);
10500 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10505 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10506 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10507 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10508 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10509 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10510 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10512 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10513 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10514 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10515 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10516 &kvm->arch.irq_sources_bitmap);
10518 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10519 mutex_init(&kvm->arch.apic_map_lock);
10520 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10522 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10523 pvclock_update_vm_gtod_copy(kvm);
10525 kvm->arch.guest_can_read_msr_platform_info = true;
10527 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10528 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10530 kvm_hv_init_vm(kvm);
10531 kvm_page_track_init(kvm);
10532 kvm_mmu_init_vm(kvm);
10534 return static_call(kvm_x86_vm_init)(kvm);
10537 int kvm_arch_post_init_vm(struct kvm *kvm)
10539 return kvm_mmu_post_init_vm(kvm);
10542 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10545 kvm_mmu_unload(vcpu);
10549 static void kvm_free_vcpus(struct kvm *kvm)
10552 struct kvm_vcpu *vcpu;
10555 * Unpin any mmu pages first.
10557 kvm_for_each_vcpu(i, vcpu, kvm) {
10558 kvm_clear_async_pf_completion_queue(vcpu);
10559 kvm_unload_vcpu_mmu(vcpu);
10561 kvm_for_each_vcpu(i, vcpu, kvm)
10562 kvm_vcpu_destroy(vcpu);
10564 mutex_lock(&kvm->lock);
10565 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10566 kvm->vcpus[i] = NULL;
10568 atomic_set(&kvm->online_vcpus, 0);
10569 mutex_unlock(&kvm->lock);
10572 void kvm_arch_sync_events(struct kvm *kvm)
10574 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10575 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10579 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
10582 * __x86_set_memory_region: Setup KVM internal memory slot
10584 * @kvm: the kvm pointer to the VM.
10585 * @id: the slot ID to setup.
10586 * @gpa: the GPA to install the slot (unused when @size == 0).
10587 * @size: the size of the slot. Set to zero to uninstall a slot.
10589 * This function helps to setup a KVM internal memory slot. Specify
10590 * @size > 0 to install a new slot, while @size == 0 to uninstall a
10591 * slot. The return code can be one of the following:
10593 * HVA: on success (uninstall will return a bogus HVA)
10596 * The caller should always use IS_ERR() to check the return value
10597 * before use. Note, the KVM internal memory slots are guaranteed to
10598 * remain valid and unchanged until the VM is destroyed, i.e., the
10599 * GPA->HVA translation will not change. However, the HVA is a user
10600 * address, i.e. its accessibility is not guaranteed, and must be
10601 * accessed via __copy_{to,from}_user().
10603 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10607 unsigned long hva, old_npages;
10608 struct kvm_memslots *slots = kvm_memslots(kvm);
10609 struct kvm_memory_slot *slot;
10611 /* Called with kvm->slots_lock held. */
10612 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10613 return ERR_PTR_USR(-EINVAL);
10615 slot = id_to_memslot(slots, id);
10617 if (slot && slot->npages)
10618 return ERR_PTR_USR(-EEXIST);
10621 * MAP_SHARED to prevent internal slot pages from being moved
10624 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10625 MAP_SHARED | MAP_ANONYMOUS, 0);
10626 if (IS_ERR((void *)hva))
10627 return (void __user *)hva;
10629 if (!slot || !slot->npages)
10632 old_npages = slot->npages;
10633 hva = slot->userspace_addr;
10636 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10637 struct kvm_userspace_memory_region m;
10639 m.slot = id | (i << 16);
10641 m.guest_phys_addr = gpa;
10642 m.userspace_addr = hva;
10643 m.memory_size = size;
10644 r = __kvm_set_memory_region(kvm, &m);
10646 return ERR_PTR_USR(r);
10650 vm_munmap(hva, old_npages * PAGE_SIZE);
10652 return (void __user *)hva;
10654 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10656 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10658 kvm_mmu_pre_destroy_vm(kvm);
10661 void kvm_arch_destroy_vm(struct kvm *kvm)
10663 if (current->mm == kvm->mm) {
10665 * Free memory regions allocated on behalf of userspace,
10666 * unless the the memory map has changed due to process exit
10669 mutex_lock(&kvm->slots_lock);
10670 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10672 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10674 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10675 mutex_unlock(&kvm->slots_lock);
10677 static_call_cond(kvm_x86_vm_destroy)(kvm);
10678 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
10679 kvm_pic_destroy(kvm);
10680 kvm_ioapic_destroy(kvm);
10681 kvm_free_vcpus(kvm);
10682 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10683 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10684 kvm_mmu_uninit_vm(kvm);
10685 kvm_page_track_cleanup(kvm);
10686 kvm_xen_destroy_vm(kvm);
10687 kvm_hv_destroy_vm(kvm);
10690 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10694 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10695 kvfree(slot->arch.rmap[i]);
10696 slot->arch.rmap[i] = NULL;
10701 kvfree(slot->arch.lpage_info[i - 1]);
10702 slot->arch.lpage_info[i - 1] = NULL;
10705 kvm_page_track_free_memslot(slot);
10708 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10709 unsigned long npages)
10714 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10715 * old arrays will be freed by __kvm_set_memory_region() if installing
10716 * the new memslot is successful.
10718 memset(&slot->arch, 0, sizeof(slot->arch));
10720 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10721 struct kvm_lpage_info *linfo;
10722 unsigned long ugfn;
10726 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10727 slot->base_gfn, level) + 1;
10729 slot->arch.rmap[i] =
10730 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10731 GFP_KERNEL_ACCOUNT);
10732 if (!slot->arch.rmap[i])
10737 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10741 slot->arch.lpage_info[i - 1] = linfo;
10743 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10744 linfo[0].disallow_lpage = 1;
10745 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10746 linfo[lpages - 1].disallow_lpage = 1;
10747 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10749 * If the gfn and userspace address are not aligned wrt each
10750 * other, disable large page support for this slot.
10752 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10755 for (j = 0; j < lpages; ++j)
10756 linfo[j].disallow_lpage = 1;
10760 if (kvm_page_track_create_memslot(slot, npages))
10766 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10767 kvfree(slot->arch.rmap[i]);
10768 slot->arch.rmap[i] = NULL;
10772 kvfree(slot->arch.lpage_info[i - 1]);
10773 slot->arch.lpage_info[i - 1] = NULL;
10778 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10780 struct kvm_vcpu *vcpu;
10784 * memslots->generation has been incremented.
10785 * mmio generation may have reached its maximum value.
10787 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10789 /* Force re-initialization of steal_time cache */
10790 kvm_for_each_vcpu(i, vcpu, kvm)
10791 kvm_vcpu_kick(vcpu);
10794 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10795 struct kvm_memory_slot *memslot,
10796 const struct kvm_userspace_memory_region *mem,
10797 enum kvm_mr_change change)
10799 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10800 return kvm_alloc_memslot_metadata(memslot,
10801 mem->memory_size >> PAGE_SHIFT);
10806 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
10808 struct kvm_arch *ka = &kvm->arch;
10810 if (!kvm_x86_ops.cpu_dirty_log_size)
10813 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
10814 (!enable && --ka->cpu_dirty_logging_count == 0))
10815 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
10817 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
10820 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10821 struct kvm_memory_slot *old,
10822 struct kvm_memory_slot *new,
10823 enum kvm_mr_change change)
10825 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
10828 * Update CPU dirty logging if dirty logging is being toggled. This
10829 * applies to all operations.
10831 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
10832 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
10835 * Nothing more to do for RO slots (which can't be dirtied and can't be
10836 * made writable) or CREATE/MOVE/DELETE of a slot.
10838 * For a memslot with dirty logging disabled:
10839 * CREATE: No dirty mappings will already exist.
10840 * MOVE/DELETE: The old mappings will already have been cleaned up by
10841 * kvm_arch_flush_shadow_memslot()
10843 * For a memslot with dirty logging enabled:
10844 * CREATE: No shadow pages exist, thus nothing to write-protect
10845 * and no dirty bits to clear.
10846 * MOVE/DELETE: The old mappings will already have been cleaned up by
10847 * kvm_arch_flush_shadow_memslot().
10849 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10853 * READONLY and non-flags changes were filtered out above, and the only
10854 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
10855 * logging isn't being toggled on or off.
10857 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
10860 if (!log_dirty_pages) {
10862 * Dirty logging tracks sptes in 4k granularity, meaning that
10863 * large sptes have to be split. If live migration succeeds,
10864 * the guest in the source machine will be destroyed and large
10865 * sptes will be created in the destination. However, if the
10866 * guest continues to run in the source machine (for example if
10867 * live migration fails), small sptes will remain around and
10868 * cause bad performance.
10870 * Scan sptes if dirty logging has been stopped, dropping those
10871 * which can be collapsed into a single large-page spte. Later
10872 * page faults will create the large-page sptes.
10874 kvm_mmu_zap_collapsible_sptes(kvm, new);
10876 /* By default, write-protect everything to log writes. */
10877 int level = PG_LEVEL_4K;
10879 if (kvm_x86_ops.cpu_dirty_log_size) {
10881 * Clear all dirty bits, unless pages are treated as
10882 * dirty from the get-go.
10884 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
10885 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
10888 * Write-protect large pages on write so that dirty
10889 * logging happens at 4k granularity. No need to
10890 * write-protect small SPTEs since write accesses are
10891 * logged by the CPU via dirty bits.
10893 level = PG_LEVEL_2M;
10894 } else if (kvm_dirty_log_manual_protect_and_init_set(kvm)) {
10896 * If we're with initial-all-set, we don't need
10897 * to write protect any small page because
10898 * they're reported as dirty already. However
10899 * we still need to write-protect huge pages
10900 * so that the page split can happen lazily on
10901 * the first write to the huge page.
10903 level = PG_LEVEL_2M;
10905 kvm_mmu_slot_remove_write_access(kvm, new, level);
10909 void kvm_arch_commit_memory_region(struct kvm *kvm,
10910 const struct kvm_userspace_memory_region *mem,
10911 struct kvm_memory_slot *old,
10912 const struct kvm_memory_slot *new,
10913 enum kvm_mr_change change)
10915 if (!kvm->arch.n_requested_mmu_pages)
10916 kvm_mmu_change_mmu_pages(kvm,
10917 kvm_mmu_calculate_default_mmu_pages(kvm));
10920 * FIXME: const-ify all uses of struct kvm_memory_slot.
10922 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
10924 /* Free the arrays associated with the old memslot. */
10925 if (change == KVM_MR_MOVE)
10926 kvm_arch_free_memslot(kvm, old);
10929 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10931 kvm_mmu_zap_all(kvm);
10934 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10935 struct kvm_memory_slot *slot)
10937 kvm_page_track_flush_slot(kvm, slot);
10940 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10942 return (is_guest_mode(vcpu) &&
10943 kvm_x86_ops.guest_apic_has_interrupt &&
10944 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
10947 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10949 if (!list_empty_careful(&vcpu->async_pf.done))
10952 if (kvm_apic_has_events(vcpu))
10955 if (vcpu->arch.pv.pv_unhalted)
10958 if (vcpu->arch.exception.pending)
10961 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10962 (vcpu->arch.nmi_pending &&
10963 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
10966 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10967 (vcpu->arch.smi_pending &&
10968 static_call(kvm_x86_smi_allowed)(vcpu, false)))
10971 if (kvm_arch_interrupt_allowed(vcpu) &&
10972 (kvm_cpu_has_interrupt(vcpu) ||
10973 kvm_guest_apic_has_interrupt(vcpu)))
10976 if (kvm_hv_has_stimer_pending(vcpu))
10979 if (is_guest_mode(vcpu) &&
10980 kvm_x86_ops.nested_ops->hv_timer_pending &&
10981 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10987 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10989 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10992 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10994 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10997 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10998 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10999 kvm_test_request(KVM_REQ_EVENT, vcpu))
11002 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11008 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11010 return vcpu->arch.preempted_in_kernel;
11013 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11015 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11018 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11020 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11023 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11025 /* Can't read the RIP when guest state is protected, just return 0 */
11026 if (vcpu->arch.guest_state_protected)
11029 if (is_64_bit_mode(vcpu))
11030 return kvm_rip_read(vcpu);
11031 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11032 kvm_rip_read(vcpu));
11034 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11036 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11038 return kvm_get_linear_rip(vcpu) == linear_rip;
11040 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11042 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11044 unsigned long rflags;
11046 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11047 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11048 rflags &= ~X86_EFLAGS_TF;
11051 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11053 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11055 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11056 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11057 rflags |= X86_EFLAGS_TF;
11058 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11061 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11063 __kvm_set_rflags(vcpu, rflags);
11064 kvm_make_request(KVM_REQ_EVENT, vcpu);
11066 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11068 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11072 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11076 r = kvm_mmu_reload(vcpu);
11080 if (!vcpu->arch.mmu->direct_map &&
11081 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11084 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11087 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11089 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11091 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11094 static inline u32 kvm_async_pf_next_probe(u32 key)
11096 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11099 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11101 u32 key = kvm_async_pf_hash_fn(gfn);
11103 while (vcpu->arch.apf.gfns[key] != ~0)
11104 key = kvm_async_pf_next_probe(key);
11106 vcpu->arch.apf.gfns[key] = gfn;
11109 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11112 u32 key = kvm_async_pf_hash_fn(gfn);
11114 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11115 (vcpu->arch.apf.gfns[key] != gfn &&
11116 vcpu->arch.apf.gfns[key] != ~0); i++)
11117 key = kvm_async_pf_next_probe(key);
11122 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11124 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11127 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11131 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11133 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11137 vcpu->arch.apf.gfns[i] = ~0;
11139 j = kvm_async_pf_next_probe(j);
11140 if (vcpu->arch.apf.gfns[j] == ~0)
11142 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11144 * k lies cyclically in ]i,j]
11146 * |....j i.k.| or |.k..j i...|
11148 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11149 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11154 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11156 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11158 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11162 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11164 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11166 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11167 &token, offset, sizeof(token));
11170 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11172 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11175 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11176 &val, offset, sizeof(val)))
11182 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11184 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11187 if (!kvm_pv_async_pf_enabled(vcpu) ||
11188 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11194 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11196 if (unlikely(!lapic_in_kernel(vcpu) ||
11197 kvm_event_needs_reinjection(vcpu) ||
11198 vcpu->arch.exception.pending))
11201 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11205 * If interrupts are off we cannot even use an artificial
11208 return kvm_arch_interrupt_allowed(vcpu);
11211 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11212 struct kvm_async_pf *work)
11214 struct x86_exception fault;
11216 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11217 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11219 if (kvm_can_deliver_async_pf(vcpu) &&
11220 !apf_put_user_notpresent(vcpu)) {
11221 fault.vector = PF_VECTOR;
11222 fault.error_code_valid = true;
11223 fault.error_code = 0;
11224 fault.nested_page_fault = false;
11225 fault.address = work->arch.token;
11226 fault.async_page_fault = true;
11227 kvm_inject_page_fault(vcpu, &fault);
11231 * It is not possible to deliver a paravirtualized asynchronous
11232 * page fault, but putting the guest in an artificial halt state
11233 * can be beneficial nevertheless: if an interrupt arrives, we
11234 * can deliver it timely and perhaps the guest will schedule
11235 * another process. When the instruction that triggered a page
11236 * fault is retried, hopefully the page will be ready in the host.
11238 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11243 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11244 struct kvm_async_pf *work)
11246 struct kvm_lapic_irq irq = {
11247 .delivery_mode = APIC_DM_FIXED,
11248 .vector = vcpu->arch.apf.vec
11251 if (work->wakeup_all)
11252 work->arch.token = ~0; /* broadcast wakeup */
11254 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11255 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11257 if ((work->wakeup_all || work->notpresent_injected) &&
11258 kvm_pv_async_pf_enabled(vcpu) &&
11259 !apf_put_user_ready(vcpu, work->arch.token)) {
11260 vcpu->arch.apf.pageready_pending = true;
11261 kvm_apic_set_irq(vcpu, &irq, NULL);
11264 vcpu->arch.apf.halted = false;
11265 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11268 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11270 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11271 if (!vcpu->arch.apf.pageready_pending)
11272 kvm_vcpu_kick(vcpu);
11275 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11277 if (!kvm_pv_async_pf_enabled(vcpu))
11280 return apf_pageready_slot_free(vcpu);
11283 void kvm_arch_start_assignment(struct kvm *kvm)
11285 atomic_inc(&kvm->arch.assigned_device_count);
11287 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11289 void kvm_arch_end_assignment(struct kvm *kvm)
11291 atomic_dec(&kvm->arch.assigned_device_count);
11293 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11295 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11297 return atomic_read(&kvm->arch.assigned_device_count);
11299 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11301 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11303 atomic_inc(&kvm->arch.noncoherent_dma_count);
11305 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11307 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11309 atomic_dec(&kvm->arch.noncoherent_dma_count);
11311 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11313 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11315 return atomic_read(&kvm->arch.noncoherent_dma_count);
11317 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11319 bool kvm_arch_has_irq_bypass(void)
11324 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11325 struct irq_bypass_producer *prod)
11327 struct kvm_kernel_irqfd *irqfd =
11328 container_of(cons, struct kvm_kernel_irqfd, consumer);
11331 irqfd->producer = prod;
11332 kvm_arch_start_assignment(irqfd->kvm);
11333 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11334 prod->irq, irqfd->gsi, 1);
11337 kvm_arch_end_assignment(irqfd->kvm);
11342 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11343 struct irq_bypass_producer *prod)
11346 struct kvm_kernel_irqfd *irqfd =
11347 container_of(cons, struct kvm_kernel_irqfd, consumer);
11349 WARN_ON(irqfd->producer != prod);
11350 irqfd->producer = NULL;
11353 * When producer of consumer is unregistered, we change back to
11354 * remapped mode, so we can re-use the current implementation
11355 * when the irq is masked/disabled or the consumer side (KVM
11356 * int this case doesn't want to receive the interrupts.
11358 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11360 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11361 " fails: %d\n", irqfd->consumer.token, ret);
11363 kvm_arch_end_assignment(irqfd->kvm);
11366 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11367 uint32_t guest_irq, bool set)
11369 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11372 bool kvm_vector_hashing_enabled(void)
11374 return vector_hashing;
11377 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11379 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11381 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11384 int kvm_spec_ctrl_test_value(u64 value)
11387 * test that setting IA32_SPEC_CTRL to given value
11388 * is allowed by the host processor
11392 unsigned long flags;
11395 local_irq_save(flags);
11397 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11399 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11402 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11404 local_irq_restore(flags);
11408 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11410 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11412 struct x86_exception fault;
11413 u32 access = error_code &
11414 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11416 if (!(error_code & PFERR_PRESENT_MASK) ||
11417 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11419 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11420 * tables probably do not match the TLB. Just proceed
11421 * with the error code that the processor gave.
11423 fault.vector = PF_VECTOR;
11424 fault.error_code_valid = true;
11425 fault.error_code = error_code;
11426 fault.nested_page_fault = false;
11427 fault.address = gva;
11429 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11431 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11434 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11435 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11436 * indicates whether exit to userspace is needed.
11438 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11439 struct x86_exception *e)
11441 if (r == X86EMUL_PROPAGATE_FAULT) {
11442 kvm_inject_emulated_page_fault(vcpu, e);
11447 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11448 * while handling a VMX instruction KVM could've handled the request
11449 * correctly by exiting to userspace and performing I/O but there
11450 * doesn't seem to be a real use-case behind such requests, just return
11451 * KVM_EXIT_INTERNAL_ERROR for now.
11453 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11454 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11455 vcpu->run->internal.ndata = 0;
11459 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11461 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11464 struct x86_exception e;
11466 unsigned long roots_to_free = 0;
11473 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11474 if (r != X86EMUL_CONTINUE)
11475 return kvm_handle_memory_failure(vcpu, r, &e);
11477 if (operand.pcid >> 12 != 0) {
11478 kvm_inject_gp(vcpu, 0);
11482 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11485 case INVPCID_TYPE_INDIV_ADDR:
11486 if ((!pcid_enabled && (operand.pcid != 0)) ||
11487 is_noncanonical_address(operand.gla, vcpu)) {
11488 kvm_inject_gp(vcpu, 0);
11491 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11492 return kvm_skip_emulated_instruction(vcpu);
11494 case INVPCID_TYPE_SINGLE_CTXT:
11495 if (!pcid_enabled && (operand.pcid != 0)) {
11496 kvm_inject_gp(vcpu, 0);
11500 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11501 kvm_mmu_sync_roots(vcpu);
11502 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11505 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11506 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11508 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11510 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11512 * If neither the current cr3 nor any of the prev_roots use the
11513 * given PCID, then nothing needs to be done here because a
11514 * resync will happen anyway before switching to any other CR3.
11517 return kvm_skip_emulated_instruction(vcpu);
11519 case INVPCID_TYPE_ALL_NON_GLOBAL:
11521 * Currently, KVM doesn't mark global entries in the shadow
11522 * page tables, so a non-global flush just degenerates to a
11523 * global flush. If needed, we could optimize this later by
11524 * keeping track of global entries in shadow page tables.
11528 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11529 kvm_mmu_unload(vcpu);
11530 return kvm_skip_emulated_instruction(vcpu);
11533 BUG(); /* We have already checked above that type <= 3 */
11536 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11538 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11540 struct kvm_run *run = vcpu->run;
11541 struct kvm_mmio_fragment *frag;
11544 BUG_ON(!vcpu->mmio_needed);
11546 /* Complete previous fragment */
11547 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11548 len = min(8u, frag->len);
11549 if (!vcpu->mmio_is_write)
11550 memcpy(frag->data, run->mmio.data, len);
11552 if (frag->len <= 8) {
11553 /* Switch to the next fragment. */
11555 vcpu->mmio_cur_fragment++;
11557 /* Go forward to the next mmio piece. */
11563 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11564 vcpu->mmio_needed = 0;
11566 // VMG change, at this point, we're always done
11567 // RIP has already been advanced
11571 // More MMIO is needed
11572 run->mmio.phys_addr = frag->gpa;
11573 run->mmio.len = min(8u, frag->len);
11574 run->mmio.is_write = vcpu->mmio_is_write;
11575 if (run->mmio.is_write)
11576 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11577 run->exit_reason = KVM_EXIT_MMIO;
11579 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11584 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11588 struct kvm_mmio_fragment *frag;
11593 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11594 if (handled == bytes)
11601 /*TODO: Check if need to increment number of frags */
11602 frag = vcpu->mmio_fragments;
11603 vcpu->mmio_nr_fragments = 1;
11608 vcpu->mmio_needed = 1;
11609 vcpu->mmio_cur_fragment = 0;
11611 vcpu->run->mmio.phys_addr = gpa;
11612 vcpu->run->mmio.len = min(8u, frag->len);
11613 vcpu->run->mmio.is_write = 1;
11614 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11615 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11617 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11621 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11623 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11627 struct kvm_mmio_fragment *frag;
11632 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11633 if (handled == bytes)
11640 /*TODO: Check if need to increment number of frags */
11641 frag = vcpu->mmio_fragments;
11642 vcpu->mmio_nr_fragments = 1;
11647 vcpu->mmio_needed = 1;
11648 vcpu->mmio_cur_fragment = 0;
11650 vcpu->run->mmio.phys_addr = gpa;
11651 vcpu->run->mmio.len = min(8u, frag->len);
11652 vcpu->run->mmio.is_write = 0;
11653 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11655 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11659 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11661 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11663 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11664 vcpu->arch.pio.count * vcpu->arch.pio.size);
11665 vcpu->arch.pio.count = 0;
11670 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11671 unsigned int port, void *data, unsigned int count)
11675 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11680 vcpu->arch.pio.count = 0;
11685 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11686 unsigned int port, void *data, unsigned int count)
11690 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11693 vcpu->arch.pio.count = 0;
11695 vcpu->arch.guest_ins_data = data;
11696 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11702 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11703 unsigned int port, void *data, unsigned int count,
11706 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11707 : kvm_sev_es_outs(vcpu, size, port, data, count);
11709 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11711 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11712 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11713 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11714 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11715 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11716 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11717 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11718 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11719 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11720 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11721 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11722 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11723 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11724 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11725 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11726 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11727 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11728 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11729 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11730 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11731 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11732 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11733 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11734 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11735 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11736 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11737 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);