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"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/isolation.h>
58 #include <linux/mem_encrypt.h>
59 #include <linux/entry-kvm.h>
61 #include <trace/events/kvm.h>
63 #include <asm/debugreg.h>
67 #include <linux/kernel_stat.h>
68 #include <asm/fpu/internal.h> /* Ugh! */
69 #include <asm/pvclock.h>
70 #include <asm/div64.h>
71 #include <asm/irq_remapping.h>
72 #include <asm/mshyperv.h>
73 #include <asm/hypervisor.h>
74 #include <asm/tlbflush.h>
75 #include <asm/intel_pt.h>
76 #include <asm/emulate_prefix.h>
77 #include <clocksource/hyperv_timer.h>
79 #define CREATE_TRACE_POINTS
82 #define MAX_IO_MSRS 256
83 #define KVM_MAX_MCE_BANKS 32
84 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
85 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
87 #define emul_to_vcpu(ctxt) \
88 ((struct kvm_vcpu *)(ctxt)->vcpu)
91 * - enable syscall per default because its emulated by KVM
92 * - enable LME and LMA per default on 64 bit KVM
96 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
98 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
101 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
103 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
104 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
106 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
107 static void process_nmi(struct kvm_vcpu *vcpu);
108 static void enter_smm(struct kvm_vcpu *vcpu);
109 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
110 static void store_regs(struct kvm_vcpu *vcpu);
111 static int sync_regs(struct kvm_vcpu *vcpu);
113 struct kvm_x86_ops kvm_x86_ops __read_mostly;
114 EXPORT_SYMBOL_GPL(kvm_x86_ops);
116 static bool __read_mostly ignore_msrs = 0;
117 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
119 static bool __read_mostly report_ignored_msrs = true;
120 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
122 unsigned int min_timer_period_us = 200;
123 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
125 static bool __read_mostly kvmclock_periodic_sync = true;
126 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
128 bool __read_mostly kvm_has_tsc_control;
129 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
130 u32 __read_mostly kvm_max_guest_tsc_khz;
131 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
132 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
133 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
134 u64 __read_mostly kvm_max_tsc_scaling_ratio;
135 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
136 u64 __read_mostly kvm_default_tsc_scaling_ratio;
137 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
139 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
140 static u32 __read_mostly tsc_tolerance_ppm = 250;
141 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
144 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
145 * adaptive tuning starting from default advancment of 1000ns. '0' disables
146 * advancement entirely. Any other value is used as-is and disables adaptive
147 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
149 static int __read_mostly lapic_timer_advance_ns = -1;
150 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
152 static bool __read_mostly vector_hashing = true;
153 module_param(vector_hashing, bool, S_IRUGO);
155 bool __read_mostly enable_vmware_backdoor = false;
156 module_param(enable_vmware_backdoor, bool, S_IRUGO);
157 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
159 static bool __read_mostly force_emulation_prefix = false;
160 module_param(force_emulation_prefix, bool, S_IRUGO);
162 int __read_mostly pi_inject_timer = -1;
163 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
166 * Restoring the host value for MSRs that are only consumed when running in
167 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
168 * returns to userspace, i.e. the kernel can run with the guest's value.
170 #define KVM_MAX_NR_USER_RETURN_MSRS 16
172 struct kvm_user_return_msrs_global {
174 u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS];
177 struct kvm_user_return_msrs {
178 struct user_return_notifier urn;
180 struct kvm_user_return_msr_values {
183 } values[KVM_MAX_NR_USER_RETURN_MSRS];
186 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global;
187 static struct kvm_user_return_msrs __percpu *user_return_msrs;
189 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
190 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
191 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
192 | XFEATURE_MASK_PKRU)
194 u64 __read_mostly host_efer;
195 EXPORT_SYMBOL_GPL(host_efer);
197 bool __read_mostly allow_smaller_maxphyaddr;
198 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
200 static u64 __read_mostly host_xss;
201 u64 __read_mostly supported_xss;
202 EXPORT_SYMBOL_GPL(supported_xss);
204 struct kvm_stats_debugfs_item debugfs_entries[] = {
205 VCPU_STAT("pf_fixed", pf_fixed),
206 VCPU_STAT("pf_guest", pf_guest),
207 VCPU_STAT("tlb_flush", tlb_flush),
208 VCPU_STAT("invlpg", invlpg),
209 VCPU_STAT("exits", exits),
210 VCPU_STAT("io_exits", io_exits),
211 VCPU_STAT("mmio_exits", mmio_exits),
212 VCPU_STAT("signal_exits", signal_exits),
213 VCPU_STAT("irq_window", irq_window_exits),
214 VCPU_STAT("nmi_window", nmi_window_exits),
215 VCPU_STAT("halt_exits", halt_exits),
216 VCPU_STAT("halt_successful_poll", halt_successful_poll),
217 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
218 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
219 VCPU_STAT("halt_wakeup", halt_wakeup),
220 VCPU_STAT("hypercalls", hypercalls),
221 VCPU_STAT("request_irq", request_irq_exits),
222 VCPU_STAT("irq_exits", irq_exits),
223 VCPU_STAT("host_state_reload", host_state_reload),
224 VCPU_STAT("fpu_reload", fpu_reload),
225 VCPU_STAT("insn_emulation", insn_emulation),
226 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
227 VCPU_STAT("irq_injections", irq_injections),
228 VCPU_STAT("nmi_injections", nmi_injections),
229 VCPU_STAT("req_event", req_event),
230 VCPU_STAT("l1d_flush", l1d_flush),
231 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
232 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
233 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
234 VM_STAT("mmu_pte_write", mmu_pte_write),
235 VM_STAT("mmu_pte_updated", mmu_pte_updated),
236 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
237 VM_STAT("mmu_flooded", mmu_flooded),
238 VM_STAT("mmu_recycled", mmu_recycled),
239 VM_STAT("mmu_cache_miss", mmu_cache_miss),
240 VM_STAT("mmu_unsync", mmu_unsync),
241 VM_STAT("remote_tlb_flush", remote_tlb_flush),
242 VM_STAT("largepages", lpages, .mode = 0444),
243 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
244 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
248 u64 __read_mostly host_xcr0;
249 u64 __read_mostly supported_xcr0;
250 EXPORT_SYMBOL_GPL(supported_xcr0);
252 static struct kmem_cache *x86_fpu_cache;
254 static struct kmem_cache *x86_emulator_cache;
257 * When called, it means the previous get/set msr reached an invalid msr.
258 * Return 0 if we want to ignore/silent this failed msr access, or 1 if we want
259 * to fail the caller.
261 static int kvm_msr_ignored_check(struct kvm_vcpu *vcpu, u32 msr,
262 u64 data, bool write)
264 const char *op = write ? "wrmsr" : "rdmsr";
267 if (report_ignored_msrs)
268 vcpu_unimpl(vcpu, "ignored %s: 0x%x data 0x%llx\n",
273 vcpu_debug_ratelimited(vcpu, "unhandled %s: 0x%x data 0x%llx\n",
279 static struct kmem_cache *kvm_alloc_emulator_cache(void)
281 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
282 unsigned int size = sizeof(struct x86_emulate_ctxt);
284 return kmem_cache_create_usercopy("x86_emulator", size,
285 __alignof__(struct x86_emulate_ctxt),
286 SLAB_ACCOUNT, useroffset,
287 size - useroffset, NULL);
290 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
292 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
295 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
296 vcpu->arch.apf.gfns[i] = ~0;
299 static void kvm_on_user_return(struct user_return_notifier *urn)
302 struct kvm_user_return_msrs *msrs
303 = container_of(urn, struct kvm_user_return_msrs, urn);
304 struct kvm_user_return_msr_values *values;
308 * Disabling irqs at this point since the following code could be
309 * interrupted and executed through kvm_arch_hardware_disable()
311 local_irq_save(flags);
312 if (msrs->registered) {
313 msrs->registered = false;
314 user_return_notifier_unregister(urn);
316 local_irq_restore(flags);
317 for (slot = 0; slot < user_return_msrs_global.nr; ++slot) {
318 values = &msrs->values[slot];
319 if (values->host != values->curr) {
320 wrmsrl(user_return_msrs_global.msrs[slot], values->host);
321 values->curr = values->host;
326 void kvm_define_user_return_msr(unsigned slot, u32 msr)
328 BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
329 user_return_msrs_global.msrs[slot] = msr;
330 if (slot >= user_return_msrs_global.nr)
331 user_return_msrs_global.nr = slot + 1;
333 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
335 static void kvm_user_return_msr_cpu_online(void)
337 unsigned int cpu = smp_processor_id();
338 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
342 for (i = 0; i < user_return_msrs_global.nr; ++i) {
343 rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
344 msrs->values[i].host = value;
345 msrs->values[i].curr = value;
349 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
351 unsigned int cpu = smp_processor_id();
352 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
355 value = (value & mask) | (msrs->values[slot].host & ~mask);
356 if (value == msrs->values[slot].curr)
358 err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
362 msrs->values[slot].curr = value;
363 if (!msrs->registered) {
364 msrs->urn.on_user_return = kvm_on_user_return;
365 user_return_notifier_register(&msrs->urn);
366 msrs->registered = true;
370 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
372 static void drop_user_return_notifiers(void)
374 unsigned int cpu = smp_processor_id();
375 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
377 if (msrs->registered)
378 kvm_on_user_return(&msrs->urn);
381 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
383 return vcpu->arch.apic_base;
385 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
387 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
389 return kvm_apic_mode(kvm_get_apic_base(vcpu));
391 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
393 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
395 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
396 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
397 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
398 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
400 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
402 if (!msr_info->host_initiated) {
403 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
405 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
409 kvm_lapic_set_base(vcpu, msr_info->data);
410 kvm_recalculate_apic_map(vcpu->kvm);
413 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
415 asmlinkage __visible noinstr void kvm_spurious_fault(void)
417 /* Fault while not rebooting. We want the trace. */
418 BUG_ON(!kvm_rebooting);
420 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
422 #define EXCPT_BENIGN 0
423 #define EXCPT_CONTRIBUTORY 1
426 static int exception_class(int vector)
436 return EXCPT_CONTRIBUTORY;
443 #define EXCPT_FAULT 0
445 #define EXCPT_ABORT 2
446 #define EXCPT_INTERRUPT 3
448 static int exception_type(int vector)
452 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
453 return EXCPT_INTERRUPT;
457 /* #DB is trap, as instruction watchpoints are handled elsewhere */
458 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
461 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
464 /* Reserved exceptions will result in fault */
468 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
470 unsigned nr = vcpu->arch.exception.nr;
471 bool has_payload = vcpu->arch.exception.has_payload;
472 unsigned long payload = vcpu->arch.exception.payload;
480 * "Certain debug exceptions may clear bit 0-3. The
481 * remaining contents of the DR6 register are never
482 * cleared by the processor".
484 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
486 * DR6.RTM is set by all #DB exceptions that don't clear it.
488 vcpu->arch.dr6 |= DR6_RTM;
489 vcpu->arch.dr6 |= payload;
491 * Bit 16 should be set in the payload whenever the #DB
492 * exception should clear DR6.RTM. This makes the payload
493 * compatible with the pending debug exceptions under VMX.
494 * Though not currently documented in the SDM, this also
495 * makes the payload compatible with the exit qualification
496 * for #DB exceptions under VMX.
498 vcpu->arch.dr6 ^= payload & DR6_RTM;
501 * The #DB payload is defined as compatible with the 'pending
502 * debug exceptions' field under VMX, not DR6. While bit 12 is
503 * defined in the 'pending debug exceptions' field (enabled
504 * breakpoint), it is reserved and must be zero in DR6.
506 vcpu->arch.dr6 &= ~BIT(12);
509 vcpu->arch.cr2 = payload;
513 vcpu->arch.exception.has_payload = false;
514 vcpu->arch.exception.payload = 0;
516 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
518 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
519 unsigned nr, bool has_error, u32 error_code,
520 bool has_payload, unsigned long payload, bool reinject)
525 kvm_make_request(KVM_REQ_EVENT, vcpu);
527 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
529 if (has_error && !is_protmode(vcpu))
533 * On vmentry, vcpu->arch.exception.pending is only
534 * true if an event injection was blocked by
535 * nested_run_pending. In that case, however,
536 * vcpu_enter_guest requests an immediate exit,
537 * and the guest shouldn't proceed far enough to
540 WARN_ON_ONCE(vcpu->arch.exception.pending);
541 vcpu->arch.exception.injected = true;
542 if (WARN_ON_ONCE(has_payload)) {
544 * A reinjected event has already
545 * delivered its payload.
551 vcpu->arch.exception.pending = true;
552 vcpu->arch.exception.injected = false;
554 vcpu->arch.exception.has_error_code = has_error;
555 vcpu->arch.exception.nr = nr;
556 vcpu->arch.exception.error_code = error_code;
557 vcpu->arch.exception.has_payload = has_payload;
558 vcpu->arch.exception.payload = payload;
559 if (!is_guest_mode(vcpu))
560 kvm_deliver_exception_payload(vcpu);
564 /* to check exception */
565 prev_nr = vcpu->arch.exception.nr;
566 if (prev_nr == DF_VECTOR) {
567 /* triple fault -> shutdown */
568 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
571 class1 = exception_class(prev_nr);
572 class2 = exception_class(nr);
573 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
574 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
576 * Generate double fault per SDM Table 5-5. Set
577 * exception.pending = true so that the double fault
578 * can trigger a nested vmexit.
580 vcpu->arch.exception.pending = true;
581 vcpu->arch.exception.injected = false;
582 vcpu->arch.exception.has_error_code = true;
583 vcpu->arch.exception.nr = DF_VECTOR;
584 vcpu->arch.exception.error_code = 0;
585 vcpu->arch.exception.has_payload = false;
586 vcpu->arch.exception.payload = 0;
588 /* replace previous exception with a new one in a hope
589 that instruction re-execution will regenerate lost
594 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
596 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
598 EXPORT_SYMBOL_GPL(kvm_queue_exception);
600 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
602 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
604 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
606 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
607 unsigned long payload)
609 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
611 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
613 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
614 u32 error_code, unsigned long payload)
616 kvm_multiple_exception(vcpu, nr, true, error_code,
617 true, payload, false);
620 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
623 kvm_inject_gp(vcpu, 0);
625 return kvm_skip_emulated_instruction(vcpu);
629 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
631 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
633 ++vcpu->stat.pf_guest;
634 vcpu->arch.exception.nested_apf =
635 is_guest_mode(vcpu) && fault->async_page_fault;
636 if (vcpu->arch.exception.nested_apf) {
637 vcpu->arch.apf.nested_apf_token = fault->address;
638 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
640 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
644 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
646 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
647 struct x86_exception *fault)
649 struct kvm_mmu *fault_mmu;
650 WARN_ON_ONCE(fault->vector != PF_VECTOR);
652 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
656 * Invalidate the TLB entry for the faulting address, if it exists,
657 * else the access will fault indefinitely (and to emulate hardware).
659 if ((fault->error_code & PFERR_PRESENT_MASK) &&
660 !(fault->error_code & PFERR_RSVD_MASK))
661 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
662 fault_mmu->root_hpa);
664 fault_mmu->inject_page_fault(vcpu, fault);
665 return fault->nested_page_fault;
667 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
669 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
671 atomic_inc(&vcpu->arch.nmi_queued);
672 kvm_make_request(KVM_REQ_NMI, vcpu);
674 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
676 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
678 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
680 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
682 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
684 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
686 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
689 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
690 * a #GP and return false.
692 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
694 if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
696 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
699 EXPORT_SYMBOL_GPL(kvm_require_cpl);
701 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
703 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
706 kvm_queue_exception(vcpu, UD_VECTOR);
709 EXPORT_SYMBOL_GPL(kvm_require_dr);
712 * This function will be used to read from the physical memory of the currently
713 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
714 * can read from guest physical or from the guest's guest physical memory.
716 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
717 gfn_t ngfn, void *data, int offset, int len,
720 struct x86_exception exception;
724 ngpa = gfn_to_gpa(ngfn);
725 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
726 if (real_gfn == UNMAPPED_GVA)
729 real_gfn = gpa_to_gfn(real_gfn);
731 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
733 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
735 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
736 void *data, int offset, int len, u32 access)
738 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
739 data, offset, len, access);
742 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
744 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
749 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
751 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
753 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
754 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
757 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
759 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
760 offset * sizeof(u64), sizeof(pdpte),
761 PFERR_USER_MASK|PFERR_WRITE_MASK);
766 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
767 if ((pdpte[i] & PT_PRESENT_MASK) &&
768 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
775 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
776 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
782 EXPORT_SYMBOL_GPL(load_pdptrs);
784 bool pdptrs_changed(struct kvm_vcpu *vcpu)
786 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
791 if (!is_pae_paging(vcpu))
794 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
797 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
798 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
799 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
800 PFERR_USER_MASK | PFERR_WRITE_MASK);
804 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
806 EXPORT_SYMBOL_GPL(pdptrs_changed);
808 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
810 unsigned long old_cr0 = kvm_read_cr0(vcpu);
811 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
812 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
817 if (cr0 & 0xffffffff00000000UL)
821 cr0 &= ~CR0_RESERVED_BITS;
823 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
826 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
830 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
831 (cr0 & X86_CR0_PG)) {
836 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
841 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
842 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
843 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
846 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
849 kvm_x86_ops.set_cr0(vcpu, cr0);
851 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
852 kvm_clear_async_pf_completion_queue(vcpu);
853 kvm_async_pf_hash_reset(vcpu);
856 if ((cr0 ^ old_cr0) & update_bits)
857 kvm_mmu_reset_context(vcpu);
859 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
860 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
861 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
862 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
866 EXPORT_SYMBOL_GPL(kvm_set_cr0);
868 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
870 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
872 EXPORT_SYMBOL_GPL(kvm_lmsw);
874 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
876 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
878 if (vcpu->arch.xcr0 != host_xcr0)
879 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
881 if (vcpu->arch.xsaves_enabled &&
882 vcpu->arch.ia32_xss != host_xss)
883 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
886 if (static_cpu_has(X86_FEATURE_PKU) &&
887 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
888 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
889 vcpu->arch.pkru != vcpu->arch.host_pkru)
890 __write_pkru(vcpu->arch.pkru);
892 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
894 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
896 if (static_cpu_has(X86_FEATURE_PKU) &&
897 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
898 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
899 vcpu->arch.pkru = rdpkru();
900 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
901 __write_pkru(vcpu->arch.host_pkru);
904 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
906 if (vcpu->arch.xcr0 != host_xcr0)
907 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
909 if (vcpu->arch.xsaves_enabled &&
910 vcpu->arch.ia32_xss != host_xss)
911 wrmsrl(MSR_IA32_XSS, host_xss);
915 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
917 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
920 u64 old_xcr0 = vcpu->arch.xcr0;
923 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
924 if (index != XCR_XFEATURE_ENABLED_MASK)
926 if (!(xcr0 & XFEATURE_MASK_FP))
928 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
932 * Do not allow the guest to set bits that we do not support
933 * saving. However, xcr0 bit 0 is always set, even if the
934 * emulated CPU does not support XSAVE (see fx_init).
936 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
937 if (xcr0 & ~valid_bits)
940 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
941 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
944 if (xcr0 & XFEATURE_MASK_AVX512) {
945 if (!(xcr0 & XFEATURE_MASK_YMM))
947 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
950 vcpu->arch.xcr0 = xcr0;
952 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
953 kvm_update_cpuid_runtime(vcpu);
957 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
959 if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
960 __kvm_set_xcr(vcpu, index, xcr)) {
961 kvm_inject_gp(vcpu, 0);
966 EXPORT_SYMBOL_GPL(kvm_set_xcr);
968 int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
970 if (cr4 & cr4_reserved_bits)
973 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
978 EXPORT_SYMBOL_GPL(kvm_valid_cr4);
980 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
982 unsigned long old_cr4 = kvm_read_cr4(vcpu);
983 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
986 if (kvm_valid_cr4(vcpu, cr4))
989 if (is_long_mode(vcpu)) {
990 if (!(cr4 & X86_CR4_PAE))
992 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
994 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
995 && ((cr4 ^ old_cr4) & pdptr_bits)
996 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1000 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1001 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1004 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1005 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1009 if (kvm_x86_ops.set_cr4(vcpu, cr4))
1012 if (((cr4 ^ old_cr4) & pdptr_bits) ||
1013 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1014 kvm_mmu_reset_context(vcpu);
1016 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1017 kvm_update_cpuid_runtime(vcpu);
1021 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1023 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1025 bool skip_tlb_flush = false;
1026 #ifdef CONFIG_X86_64
1027 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1030 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1031 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1035 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1036 if (!skip_tlb_flush) {
1037 kvm_mmu_sync_roots(vcpu);
1038 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1043 if (is_long_mode(vcpu) &&
1044 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
1046 else if (is_pae_paging(vcpu) &&
1047 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1050 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1051 vcpu->arch.cr3 = cr3;
1052 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1056 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1058 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1060 if (cr8 & CR8_RESERVED_BITS)
1062 if (lapic_in_kernel(vcpu))
1063 kvm_lapic_set_tpr(vcpu, cr8);
1065 vcpu->arch.cr8 = cr8;
1068 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1070 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1072 if (lapic_in_kernel(vcpu))
1073 return kvm_lapic_get_cr8(vcpu);
1075 return vcpu->arch.cr8;
1077 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1079 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1083 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1084 for (i = 0; i < KVM_NR_DB_REGS; i++)
1085 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1086 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1090 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1094 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1095 dr7 = vcpu->arch.guest_debug_dr7;
1097 dr7 = vcpu->arch.dr7;
1098 kvm_x86_ops.set_dr7(vcpu, dr7);
1099 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1100 if (dr7 & DR7_BP_EN_MASK)
1101 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1103 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1105 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1107 u64 fixed = DR6_FIXED_1;
1109 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1114 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1116 size_t size = ARRAY_SIZE(vcpu->arch.db);
1120 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1121 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1122 vcpu->arch.eff_db[dr] = val;
1126 if (!kvm_dr6_valid(val))
1127 return -1; /* #GP */
1128 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1132 if (!kvm_dr7_valid(val))
1133 return -1; /* #GP */
1134 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1135 kvm_update_dr7(vcpu);
1142 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1144 if (__kvm_set_dr(vcpu, dr, val)) {
1145 kvm_inject_gp(vcpu, 0);
1150 EXPORT_SYMBOL_GPL(kvm_set_dr);
1152 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1154 size_t size = ARRAY_SIZE(vcpu->arch.db);
1158 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1162 *val = vcpu->arch.dr6;
1166 *val = vcpu->arch.dr7;
1171 EXPORT_SYMBOL_GPL(kvm_get_dr);
1173 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1175 u32 ecx = kvm_rcx_read(vcpu);
1179 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1182 kvm_rax_write(vcpu, (u32)data);
1183 kvm_rdx_write(vcpu, data >> 32);
1186 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1189 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1190 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1192 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1193 * extract the supported MSRs from the related const lists.
1194 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1195 * capabilities of the host cpu. This capabilities test skips MSRs that are
1196 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1197 * may depend on host virtualization features rather than host cpu features.
1200 static const u32 msrs_to_save_all[] = {
1201 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1203 #ifdef CONFIG_X86_64
1204 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1206 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1207 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1209 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1210 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1211 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1212 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1213 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1214 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1215 MSR_IA32_UMWAIT_CONTROL,
1217 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1218 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1219 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1220 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1221 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1222 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1223 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1224 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1225 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1226 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1227 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1228 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1229 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1230 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1231 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1232 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1233 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1234 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1235 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1236 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1237 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1238 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1241 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1242 static unsigned num_msrs_to_save;
1244 static const u32 emulated_msrs_all[] = {
1245 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1246 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1247 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1248 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1249 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1250 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1251 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1253 HV_X64_MSR_VP_INDEX,
1254 HV_X64_MSR_VP_RUNTIME,
1255 HV_X64_MSR_SCONTROL,
1256 HV_X64_MSR_STIMER0_CONFIG,
1257 HV_X64_MSR_VP_ASSIST_PAGE,
1258 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1259 HV_X64_MSR_TSC_EMULATION_STATUS,
1260 HV_X64_MSR_SYNDBG_OPTIONS,
1261 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1262 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1263 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1265 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1266 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1268 MSR_IA32_TSC_ADJUST,
1269 MSR_IA32_TSCDEADLINE,
1270 MSR_IA32_ARCH_CAPABILITIES,
1271 MSR_IA32_PERF_CAPABILITIES,
1272 MSR_IA32_MISC_ENABLE,
1273 MSR_IA32_MCG_STATUS,
1275 MSR_IA32_MCG_EXT_CTL,
1279 MSR_MISC_FEATURES_ENABLES,
1280 MSR_AMD64_VIRT_SPEC_CTRL,
1285 * The following list leaves out MSRs whose values are determined
1286 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1287 * We always support the "true" VMX control MSRs, even if the host
1288 * processor does not, so I am putting these registers here rather
1289 * than in msrs_to_save_all.
1292 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1293 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1294 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1295 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1297 MSR_IA32_VMX_CR0_FIXED0,
1298 MSR_IA32_VMX_CR4_FIXED0,
1299 MSR_IA32_VMX_VMCS_ENUM,
1300 MSR_IA32_VMX_PROCBASED_CTLS2,
1301 MSR_IA32_VMX_EPT_VPID_CAP,
1302 MSR_IA32_VMX_VMFUNC,
1305 MSR_KVM_POLL_CONTROL,
1308 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1309 static unsigned num_emulated_msrs;
1312 * List of msr numbers which are used to expose MSR-based features that
1313 * can be used by a hypervisor to validate requested CPU features.
1315 static const u32 msr_based_features_all[] = {
1317 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1318 MSR_IA32_VMX_PINBASED_CTLS,
1319 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1320 MSR_IA32_VMX_PROCBASED_CTLS,
1321 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1322 MSR_IA32_VMX_EXIT_CTLS,
1323 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1324 MSR_IA32_VMX_ENTRY_CTLS,
1326 MSR_IA32_VMX_CR0_FIXED0,
1327 MSR_IA32_VMX_CR0_FIXED1,
1328 MSR_IA32_VMX_CR4_FIXED0,
1329 MSR_IA32_VMX_CR4_FIXED1,
1330 MSR_IA32_VMX_VMCS_ENUM,
1331 MSR_IA32_VMX_PROCBASED_CTLS2,
1332 MSR_IA32_VMX_EPT_VPID_CAP,
1333 MSR_IA32_VMX_VMFUNC,
1337 MSR_IA32_ARCH_CAPABILITIES,
1338 MSR_IA32_PERF_CAPABILITIES,
1341 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1342 static unsigned int num_msr_based_features;
1344 static u64 kvm_get_arch_capabilities(void)
1348 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1349 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1352 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1353 * the nested hypervisor runs with NX huge pages. If it is not,
1354 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1355 * L1 guests, so it need not worry about its own (L2) guests.
1357 data |= ARCH_CAP_PSCHANGE_MC_NO;
1360 * If we're doing cache flushes (either "always" or "cond")
1361 * we will do one whenever the guest does a vmlaunch/vmresume.
1362 * If an outer hypervisor is doing the cache flush for us
1363 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1364 * capability to the guest too, and if EPT is disabled we're not
1365 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1366 * require a nested hypervisor to do a flush of its own.
1368 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1369 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1371 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1372 data |= ARCH_CAP_RDCL_NO;
1373 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1374 data |= ARCH_CAP_SSB_NO;
1375 if (!boot_cpu_has_bug(X86_BUG_MDS))
1376 data |= ARCH_CAP_MDS_NO;
1379 * On TAA affected systems:
1380 * - nothing to do if TSX is disabled on the host.
1381 * - we emulate TSX_CTRL if present on the host.
1382 * This lets the guest use VERW to clear CPU buffers.
1384 if (!boot_cpu_has(X86_FEATURE_RTM))
1385 data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1386 else if (!boot_cpu_has_bug(X86_BUG_TAA))
1387 data |= ARCH_CAP_TAA_NO;
1392 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1394 switch (msr->index) {
1395 case MSR_IA32_ARCH_CAPABILITIES:
1396 msr->data = kvm_get_arch_capabilities();
1398 case MSR_IA32_UCODE_REV:
1399 rdmsrl_safe(msr->index, &msr->data);
1402 return kvm_x86_ops.get_msr_feature(msr);
1407 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1409 struct kvm_msr_entry msr;
1413 r = kvm_get_msr_feature(&msr);
1415 if (r == KVM_MSR_RET_INVALID) {
1416 /* Unconditionally clear the output for simplicity */
1418 r = kvm_msr_ignored_check(vcpu, index, 0, false);
1429 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1431 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1434 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1437 if (efer & (EFER_LME | EFER_LMA) &&
1438 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1441 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1447 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1449 if (efer & efer_reserved_bits)
1452 return __kvm_valid_efer(vcpu, efer);
1454 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1456 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1458 u64 old_efer = vcpu->arch.efer;
1459 u64 efer = msr_info->data;
1461 if (efer & efer_reserved_bits)
1464 if (!msr_info->host_initiated) {
1465 if (!__kvm_valid_efer(vcpu, efer))
1468 if (is_paging(vcpu) &&
1469 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1474 efer |= vcpu->arch.efer & EFER_LMA;
1476 kvm_x86_ops.set_efer(vcpu, efer);
1478 /* Update reserved bits */
1479 if ((efer ^ old_efer) & EFER_NX)
1480 kvm_mmu_reset_context(vcpu);
1485 void kvm_enable_efer_bits(u64 mask)
1487 efer_reserved_bits &= ~mask;
1489 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1491 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1493 struct kvm *kvm = vcpu->kvm;
1494 struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
1495 u32 count = kvm->arch.msr_filter.count;
1497 bool r = kvm->arch.msr_filter.default_allow;
1500 /* MSR filtering not set up or x2APIC enabled, allow everything */
1501 if (!count || (index >= 0x800 && index <= 0x8ff))
1504 /* Prevent collision with set_msr_filter */
1505 idx = srcu_read_lock(&kvm->srcu);
1507 for (i = 0; i < count; i++) {
1508 u32 start = ranges[i].base;
1509 u32 end = start + ranges[i].nmsrs;
1510 u32 flags = ranges[i].flags;
1511 unsigned long *bitmap = ranges[i].bitmap;
1513 if ((index >= start) && (index < end) && (flags & type)) {
1514 r = !!test_bit(index - start, bitmap);
1519 srcu_read_unlock(&kvm->srcu, idx);
1523 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1526 * Write @data into the MSR specified by @index. Select MSR specific fault
1527 * checks are bypassed if @host_initiated is %true.
1528 * Returns 0 on success, non-0 otherwise.
1529 * Assumes vcpu_load() was already called.
1531 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1532 bool host_initiated)
1534 struct msr_data msr;
1536 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1542 case MSR_KERNEL_GS_BASE:
1545 if (is_noncanonical_address(data, vcpu))
1548 case MSR_IA32_SYSENTER_EIP:
1549 case MSR_IA32_SYSENTER_ESP:
1551 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1552 * non-canonical address is written on Intel but not on
1553 * AMD (which ignores the top 32-bits, because it does
1554 * not implement 64-bit SYSENTER).
1556 * 64-bit code should hence be able to write a non-canonical
1557 * value on AMD. Making the address canonical ensures that
1558 * vmentry does not fail on Intel after writing a non-canonical
1559 * value, and that something deterministic happens if the guest
1560 * invokes 64-bit SYSENTER.
1562 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1567 msr.host_initiated = host_initiated;
1569 return kvm_x86_ops.set_msr(vcpu, &msr);
1572 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1573 u32 index, u64 data, bool host_initiated)
1575 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1577 if (ret == KVM_MSR_RET_INVALID)
1578 ret = kvm_msr_ignored_check(vcpu, index, data, true);
1584 * Read the MSR specified by @index into @data. Select MSR specific fault
1585 * checks are bypassed if @host_initiated is %true.
1586 * Returns 0 on success, non-0 otherwise.
1587 * Assumes vcpu_load() was already called.
1589 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1590 bool host_initiated)
1592 struct msr_data msr;
1595 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1599 msr.host_initiated = host_initiated;
1601 ret = kvm_x86_ops.get_msr(vcpu, &msr);
1607 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1608 u32 index, u64 *data, bool host_initiated)
1610 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1612 if (ret == KVM_MSR_RET_INVALID) {
1613 /* Unconditionally clear *data for simplicity */
1615 ret = kvm_msr_ignored_check(vcpu, index, 0, false);
1621 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1623 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1625 EXPORT_SYMBOL_GPL(kvm_get_msr);
1627 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1629 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1631 EXPORT_SYMBOL_GPL(kvm_set_msr);
1633 static int complete_emulated_msr(struct kvm_vcpu *vcpu, bool is_read)
1635 if (vcpu->run->msr.error) {
1636 kvm_inject_gp(vcpu, 0);
1638 } else if (is_read) {
1639 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1640 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1643 return kvm_skip_emulated_instruction(vcpu);
1646 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1648 return complete_emulated_msr(vcpu, true);
1651 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1653 return complete_emulated_msr(vcpu, false);
1656 static u64 kvm_msr_reason(int r)
1660 return KVM_MSR_EXIT_REASON_UNKNOWN;
1662 return KVM_MSR_EXIT_REASON_FILTER;
1664 return KVM_MSR_EXIT_REASON_INVAL;
1668 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1669 u32 exit_reason, u64 data,
1670 int (*completion)(struct kvm_vcpu *vcpu),
1673 u64 msr_reason = kvm_msr_reason(r);
1675 /* Check if the user wanted to know about this MSR fault */
1676 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1679 vcpu->run->exit_reason = exit_reason;
1680 vcpu->run->msr.error = 0;
1681 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1682 vcpu->run->msr.reason = msr_reason;
1683 vcpu->run->msr.index = index;
1684 vcpu->run->msr.data = data;
1685 vcpu->arch.complete_userspace_io = completion;
1690 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1692 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1693 complete_emulated_rdmsr, r);
1696 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1698 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1699 complete_emulated_wrmsr, r);
1702 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1704 u32 ecx = kvm_rcx_read(vcpu);
1708 r = kvm_get_msr(vcpu, ecx, &data);
1710 /* MSR read failed? See if we should ask user space */
1711 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1712 /* Bounce to user space */
1716 /* MSR read failed? Inject a #GP */
1718 trace_kvm_msr_read_ex(ecx);
1719 kvm_inject_gp(vcpu, 0);
1723 trace_kvm_msr_read(ecx, data);
1725 kvm_rax_write(vcpu, data & -1u);
1726 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1727 return kvm_skip_emulated_instruction(vcpu);
1729 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1731 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1733 u32 ecx = kvm_rcx_read(vcpu);
1734 u64 data = kvm_read_edx_eax(vcpu);
1737 r = kvm_set_msr(vcpu, ecx, data);
1739 /* MSR write failed? See if we should ask user space */
1740 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r)) {
1741 /* Bounce to user space */
1745 /* MSR write failed? Inject a #GP */
1747 trace_kvm_msr_write_ex(ecx, data);
1748 kvm_inject_gp(vcpu, 0);
1752 trace_kvm_msr_write(ecx, data);
1753 return kvm_skip_emulated_instruction(vcpu);
1755 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1757 bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1759 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1760 xfer_to_guest_mode_work_pending();
1762 EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request);
1765 * The fast path for frequent and performance sensitive wrmsr emulation,
1766 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1767 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1768 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1769 * other cases which must be called after interrupts are enabled on the host.
1771 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1773 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1776 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1777 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1778 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1779 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1782 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1783 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1784 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1785 trace_kvm_apic_write(APIC_ICR, (u32)data);
1792 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1794 if (!kvm_can_use_hv_timer(vcpu))
1797 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1801 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1803 u32 msr = kvm_rcx_read(vcpu);
1805 fastpath_t ret = EXIT_FASTPATH_NONE;
1808 case APIC_BASE_MSR + (APIC_ICR >> 4):
1809 data = kvm_read_edx_eax(vcpu);
1810 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1811 kvm_skip_emulated_instruction(vcpu);
1812 ret = EXIT_FASTPATH_EXIT_HANDLED;
1815 case MSR_IA32_TSCDEADLINE:
1816 data = kvm_read_edx_eax(vcpu);
1817 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1818 kvm_skip_emulated_instruction(vcpu);
1819 ret = EXIT_FASTPATH_REENTER_GUEST;
1826 if (ret != EXIT_FASTPATH_NONE)
1827 trace_kvm_msr_write(msr, data);
1831 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1834 * Adapt set_msr() to msr_io()'s calling convention
1836 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1838 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1841 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1843 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1846 #ifdef CONFIG_X86_64
1847 struct pvclock_clock {
1857 struct pvclock_gtod_data {
1860 struct pvclock_clock clock; /* extract of a clocksource struct */
1861 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1867 static struct pvclock_gtod_data pvclock_gtod_data;
1869 static void update_pvclock_gtod(struct timekeeper *tk)
1871 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1873 write_seqcount_begin(&vdata->seq);
1875 /* copy pvclock gtod data */
1876 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1877 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1878 vdata->clock.mask = tk->tkr_mono.mask;
1879 vdata->clock.mult = tk->tkr_mono.mult;
1880 vdata->clock.shift = tk->tkr_mono.shift;
1881 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1882 vdata->clock.offset = tk->tkr_mono.base;
1884 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1885 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1886 vdata->raw_clock.mask = tk->tkr_raw.mask;
1887 vdata->raw_clock.mult = tk->tkr_raw.mult;
1888 vdata->raw_clock.shift = tk->tkr_raw.shift;
1889 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1890 vdata->raw_clock.offset = tk->tkr_raw.base;
1892 vdata->wall_time_sec = tk->xtime_sec;
1894 vdata->offs_boot = tk->offs_boot;
1896 write_seqcount_end(&vdata->seq);
1899 static s64 get_kvmclock_base_ns(void)
1901 /* Count up from boot time, but with the frequency of the raw clock. */
1902 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1905 static s64 get_kvmclock_base_ns(void)
1907 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1908 return ktime_get_boottime_ns();
1912 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1916 struct pvclock_wall_clock wc;
1922 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1927 ++version; /* first time write, random junk */
1931 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1935 * The guest calculates current wall clock time by adding
1936 * system time (updated by kvm_guest_time_update below) to the
1937 * wall clock specified here. We do the reverse here.
1939 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1941 wc.nsec = do_div(wall_nsec, 1000000000);
1942 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1943 wc.version = version;
1945 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1948 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1951 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
1952 bool old_msr, bool host_initiated)
1954 struct kvm_arch *ka = &vcpu->kvm->arch;
1956 if (vcpu->vcpu_id == 0 && !host_initiated) {
1957 if (ka->boot_vcpu_runs_old_kvmclock && old_msr)
1958 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1960 ka->boot_vcpu_runs_old_kvmclock = old_msr;
1963 vcpu->arch.time = system_time;
1964 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
1966 /* we verify if the enable bit is set... */
1967 vcpu->arch.pv_time_enabled = false;
1968 if (!(system_time & 1))
1971 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
1972 &vcpu->arch.pv_time, system_time & ~1ULL,
1973 sizeof(struct pvclock_vcpu_time_info)))
1974 vcpu->arch.pv_time_enabled = true;
1979 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1981 do_shl32_div32(dividend, divisor);
1985 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1986 s8 *pshift, u32 *pmultiplier)
1994 scaled64 = scaled_hz;
1995 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2000 tps32 = (uint32_t)tps64;
2001 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2002 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2010 *pmultiplier = div_frac(scaled64, tps32);
2013 #ifdef CONFIG_X86_64
2014 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2017 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2018 static unsigned long max_tsc_khz;
2020 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2022 u64 v = (u64)khz * (1000000 + ppm);
2027 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2031 /* Guest TSC same frequency as host TSC? */
2033 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2037 /* TSC scaling supported? */
2038 if (!kvm_has_tsc_control) {
2039 if (user_tsc_khz > tsc_khz) {
2040 vcpu->arch.tsc_catchup = 1;
2041 vcpu->arch.tsc_always_catchup = 1;
2044 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2049 /* TSC scaling required - calculate ratio */
2050 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2051 user_tsc_khz, tsc_khz);
2053 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2054 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2059 vcpu->arch.tsc_scaling_ratio = ratio;
2063 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2065 u32 thresh_lo, thresh_hi;
2066 int use_scaling = 0;
2068 /* tsc_khz can be zero if TSC calibration fails */
2069 if (user_tsc_khz == 0) {
2070 /* set tsc_scaling_ratio to a safe value */
2071 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2075 /* Compute a scale to convert nanoseconds in TSC cycles */
2076 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2077 &vcpu->arch.virtual_tsc_shift,
2078 &vcpu->arch.virtual_tsc_mult);
2079 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2082 * Compute the variation in TSC rate which is acceptable
2083 * within the range of tolerance and decide if the
2084 * rate being applied is within that bounds of the hardware
2085 * rate. If so, no scaling or compensation need be done.
2087 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2088 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2089 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2090 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2093 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2096 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2098 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2099 vcpu->arch.virtual_tsc_mult,
2100 vcpu->arch.virtual_tsc_shift);
2101 tsc += vcpu->arch.this_tsc_write;
2105 static inline int gtod_is_based_on_tsc(int mode)
2107 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2110 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2112 #ifdef CONFIG_X86_64
2114 struct kvm_arch *ka = &vcpu->kvm->arch;
2115 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2117 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2118 atomic_read(&vcpu->kvm->online_vcpus));
2121 * Once the masterclock is enabled, always perform request in
2122 * order to update it.
2124 * In order to enable masterclock, the host clocksource must be TSC
2125 * and the vcpus need to have matched TSCs. When that happens,
2126 * perform request to enable masterclock.
2128 if (ka->use_master_clock ||
2129 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2130 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2132 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2133 atomic_read(&vcpu->kvm->online_vcpus),
2134 ka->use_master_clock, gtod->clock.vclock_mode);
2139 * Multiply tsc by a fixed point number represented by ratio.
2141 * The most significant 64-N bits (mult) of ratio represent the
2142 * integral part of the fixed point number; the remaining N bits
2143 * (frac) represent the fractional part, ie. ratio represents a fixed
2144 * point number (mult + frac * 2^(-N)).
2146 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2148 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2150 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2153 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2156 u64 ratio = vcpu->arch.tsc_scaling_ratio;
2158 if (ratio != kvm_default_tsc_scaling_ratio)
2159 _tsc = __scale_tsc(ratio, tsc);
2163 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2165 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2169 tsc = kvm_scale_tsc(vcpu, rdtsc());
2171 return target_tsc - tsc;
2174 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2176 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2178 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2180 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2182 vcpu->arch.l1_tsc_offset = offset;
2183 vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
2186 static inline bool kvm_check_tsc_unstable(void)
2188 #ifdef CONFIG_X86_64
2190 * TSC is marked unstable when we're running on Hyper-V,
2191 * 'TSC page' clocksource is good.
2193 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2196 return check_tsc_unstable();
2199 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2201 struct kvm *kvm = vcpu->kvm;
2202 u64 offset, ns, elapsed;
2203 unsigned long flags;
2205 bool already_matched;
2206 bool synchronizing = false;
2208 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2209 offset = kvm_compute_tsc_offset(vcpu, data);
2210 ns = get_kvmclock_base_ns();
2211 elapsed = ns - kvm->arch.last_tsc_nsec;
2213 if (vcpu->arch.virtual_tsc_khz) {
2216 * detection of vcpu initialization -- need to sync
2217 * with other vCPUs. This particularly helps to keep
2218 * kvm_clock stable after CPU hotplug
2220 synchronizing = true;
2222 u64 tsc_exp = kvm->arch.last_tsc_write +
2223 nsec_to_cycles(vcpu, elapsed);
2224 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2226 * Special case: TSC write with a small delta (1 second)
2227 * of virtual cycle time against real time is
2228 * interpreted as an attempt to synchronize the CPU.
2230 synchronizing = data < tsc_exp + tsc_hz &&
2231 data + tsc_hz > tsc_exp;
2236 * For a reliable TSC, we can match TSC offsets, and for an unstable
2237 * TSC, we add elapsed time in this computation. We could let the
2238 * compensation code attempt to catch up if we fall behind, but
2239 * it's better to try to match offsets from the beginning.
2241 if (synchronizing &&
2242 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2243 if (!kvm_check_tsc_unstable()) {
2244 offset = kvm->arch.cur_tsc_offset;
2246 u64 delta = nsec_to_cycles(vcpu, elapsed);
2248 offset = kvm_compute_tsc_offset(vcpu, data);
2251 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2254 * We split periods of matched TSC writes into generations.
2255 * For each generation, we track the original measured
2256 * nanosecond time, offset, and write, so if TSCs are in
2257 * sync, we can match exact offset, and if not, we can match
2258 * exact software computation in compute_guest_tsc()
2260 * These values are tracked in kvm->arch.cur_xxx variables.
2262 kvm->arch.cur_tsc_generation++;
2263 kvm->arch.cur_tsc_nsec = ns;
2264 kvm->arch.cur_tsc_write = data;
2265 kvm->arch.cur_tsc_offset = offset;
2270 * We also track th most recent recorded KHZ, write and time to
2271 * allow the matching interval to be extended at each write.
2273 kvm->arch.last_tsc_nsec = ns;
2274 kvm->arch.last_tsc_write = data;
2275 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2277 vcpu->arch.last_guest_tsc = data;
2279 /* Keep track of which generation this VCPU has synchronized to */
2280 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2281 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2282 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2284 kvm_vcpu_write_tsc_offset(vcpu, offset);
2285 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2287 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2289 kvm->arch.nr_vcpus_matched_tsc = 0;
2290 } else if (!already_matched) {
2291 kvm->arch.nr_vcpus_matched_tsc++;
2294 kvm_track_tsc_matching(vcpu);
2295 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2298 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2301 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2302 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2305 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2307 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2308 WARN_ON(adjustment < 0);
2309 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2310 adjust_tsc_offset_guest(vcpu, adjustment);
2313 #ifdef CONFIG_X86_64
2315 static u64 read_tsc(void)
2317 u64 ret = (u64)rdtsc_ordered();
2318 u64 last = pvclock_gtod_data.clock.cycle_last;
2320 if (likely(ret >= last))
2324 * GCC likes to generate cmov here, but this branch is extremely
2325 * predictable (it's just a function of time and the likely is
2326 * very likely) and there's a data dependence, so force GCC
2327 * to generate a branch instead. I don't barrier() because
2328 * we don't actually need a barrier, and if this function
2329 * ever gets inlined it will generate worse code.
2335 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2341 switch (clock->vclock_mode) {
2342 case VDSO_CLOCKMODE_HVCLOCK:
2343 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2345 if (tsc_pg_val != U64_MAX) {
2346 /* TSC page valid */
2347 *mode = VDSO_CLOCKMODE_HVCLOCK;
2348 v = (tsc_pg_val - clock->cycle_last) &
2351 /* TSC page invalid */
2352 *mode = VDSO_CLOCKMODE_NONE;
2355 case VDSO_CLOCKMODE_TSC:
2356 *mode = VDSO_CLOCKMODE_TSC;
2357 *tsc_timestamp = read_tsc();
2358 v = (*tsc_timestamp - clock->cycle_last) &
2362 *mode = VDSO_CLOCKMODE_NONE;
2365 if (*mode == VDSO_CLOCKMODE_NONE)
2366 *tsc_timestamp = v = 0;
2368 return v * clock->mult;
2371 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2373 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2379 seq = read_seqcount_begin(>od->seq);
2380 ns = gtod->raw_clock.base_cycles;
2381 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2382 ns >>= gtod->raw_clock.shift;
2383 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2384 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2390 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2392 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2398 seq = read_seqcount_begin(>od->seq);
2399 ts->tv_sec = gtod->wall_time_sec;
2400 ns = gtod->clock.base_cycles;
2401 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2402 ns >>= gtod->clock.shift;
2403 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2405 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2411 /* returns true if host is using TSC based clocksource */
2412 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2414 /* checked again under seqlock below */
2415 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2418 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2422 /* returns true if host is using TSC based clocksource */
2423 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2426 /* checked again under seqlock below */
2427 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2430 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2436 * Assuming a stable TSC across physical CPUS, and a stable TSC
2437 * across virtual CPUs, the following condition is possible.
2438 * Each numbered line represents an event visible to both
2439 * CPUs at the next numbered event.
2441 * "timespecX" represents host monotonic time. "tscX" represents
2444 * VCPU0 on CPU0 | VCPU1 on CPU1
2446 * 1. read timespec0,tsc0
2447 * 2. | timespec1 = timespec0 + N
2449 * 3. transition to guest | transition to guest
2450 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2451 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2452 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2454 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2457 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2459 * - 0 < N - M => M < N
2461 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2462 * always the case (the difference between two distinct xtime instances
2463 * might be smaller then the difference between corresponding TSC reads,
2464 * when updating guest vcpus pvclock areas).
2466 * To avoid that problem, do not allow visibility of distinct
2467 * system_timestamp/tsc_timestamp values simultaneously: use a master
2468 * copy of host monotonic time values. Update that master copy
2471 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2475 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2477 #ifdef CONFIG_X86_64
2478 struct kvm_arch *ka = &kvm->arch;
2480 bool host_tsc_clocksource, vcpus_matched;
2482 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2483 atomic_read(&kvm->online_vcpus));
2486 * If the host uses TSC clock, then passthrough TSC as stable
2489 host_tsc_clocksource = kvm_get_time_and_clockread(
2490 &ka->master_kernel_ns,
2491 &ka->master_cycle_now);
2493 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2494 && !ka->backwards_tsc_observed
2495 && !ka->boot_vcpu_runs_old_kvmclock;
2497 if (ka->use_master_clock)
2498 atomic_set(&kvm_guest_has_master_clock, 1);
2500 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2501 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2506 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2508 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2511 static void kvm_gen_update_masterclock(struct kvm *kvm)
2513 #ifdef CONFIG_X86_64
2515 struct kvm_vcpu *vcpu;
2516 struct kvm_arch *ka = &kvm->arch;
2518 spin_lock(&ka->pvclock_gtod_sync_lock);
2519 kvm_make_mclock_inprogress_request(kvm);
2520 /* no guest entries from this point */
2521 pvclock_update_vm_gtod_copy(kvm);
2523 kvm_for_each_vcpu(i, vcpu, kvm)
2524 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2526 /* guest entries allowed */
2527 kvm_for_each_vcpu(i, vcpu, kvm)
2528 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2530 spin_unlock(&ka->pvclock_gtod_sync_lock);
2534 u64 get_kvmclock_ns(struct kvm *kvm)
2536 struct kvm_arch *ka = &kvm->arch;
2537 struct pvclock_vcpu_time_info hv_clock;
2540 spin_lock(&ka->pvclock_gtod_sync_lock);
2541 if (!ka->use_master_clock) {
2542 spin_unlock(&ka->pvclock_gtod_sync_lock);
2543 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2546 hv_clock.tsc_timestamp = ka->master_cycle_now;
2547 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2548 spin_unlock(&ka->pvclock_gtod_sync_lock);
2550 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2553 if (__this_cpu_read(cpu_tsc_khz)) {
2554 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2555 &hv_clock.tsc_shift,
2556 &hv_clock.tsc_to_system_mul);
2557 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2559 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2566 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2568 struct kvm_vcpu_arch *vcpu = &v->arch;
2569 struct pvclock_vcpu_time_info guest_hv_clock;
2571 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2572 &guest_hv_clock, sizeof(guest_hv_clock))))
2575 /* This VCPU is paused, but it's legal for a guest to read another
2576 * VCPU's kvmclock, so we really have to follow the specification where
2577 * it says that version is odd if data is being modified, and even after
2580 * Version field updates must be kept separate. This is because
2581 * kvm_write_guest_cached might use a "rep movs" instruction, and
2582 * writes within a string instruction are weakly ordered. So there
2583 * are three writes overall.
2585 * As a small optimization, only write the version field in the first
2586 * and third write. The vcpu->pv_time cache is still valid, because the
2587 * version field is the first in the struct.
2589 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2591 if (guest_hv_clock.version & 1)
2592 ++guest_hv_clock.version; /* first time write, random junk */
2594 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2595 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2597 sizeof(vcpu->hv_clock.version));
2601 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2602 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2604 if (vcpu->pvclock_set_guest_stopped_request) {
2605 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2606 vcpu->pvclock_set_guest_stopped_request = false;
2609 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2611 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2613 sizeof(vcpu->hv_clock));
2617 vcpu->hv_clock.version++;
2618 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2620 sizeof(vcpu->hv_clock.version));
2623 static int kvm_guest_time_update(struct kvm_vcpu *v)
2625 unsigned long flags, tgt_tsc_khz;
2626 struct kvm_vcpu_arch *vcpu = &v->arch;
2627 struct kvm_arch *ka = &v->kvm->arch;
2629 u64 tsc_timestamp, host_tsc;
2631 bool use_master_clock;
2637 * If the host uses TSC clock, then passthrough TSC as stable
2640 spin_lock(&ka->pvclock_gtod_sync_lock);
2641 use_master_clock = ka->use_master_clock;
2642 if (use_master_clock) {
2643 host_tsc = ka->master_cycle_now;
2644 kernel_ns = ka->master_kernel_ns;
2646 spin_unlock(&ka->pvclock_gtod_sync_lock);
2648 /* Keep irq disabled to prevent changes to the clock */
2649 local_irq_save(flags);
2650 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2651 if (unlikely(tgt_tsc_khz == 0)) {
2652 local_irq_restore(flags);
2653 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2656 if (!use_master_clock) {
2658 kernel_ns = get_kvmclock_base_ns();
2661 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2664 * We may have to catch up the TSC to match elapsed wall clock
2665 * time for two reasons, even if kvmclock is used.
2666 * 1) CPU could have been running below the maximum TSC rate
2667 * 2) Broken TSC compensation resets the base at each VCPU
2668 * entry to avoid unknown leaps of TSC even when running
2669 * again on the same CPU. This may cause apparent elapsed
2670 * time to disappear, and the guest to stand still or run
2673 if (vcpu->tsc_catchup) {
2674 u64 tsc = compute_guest_tsc(v, kernel_ns);
2675 if (tsc > tsc_timestamp) {
2676 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2677 tsc_timestamp = tsc;
2681 local_irq_restore(flags);
2683 /* With all the info we got, fill in the values */
2685 if (kvm_has_tsc_control)
2686 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2688 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2689 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2690 &vcpu->hv_clock.tsc_shift,
2691 &vcpu->hv_clock.tsc_to_system_mul);
2692 vcpu->hw_tsc_khz = tgt_tsc_khz;
2695 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2696 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2697 vcpu->last_guest_tsc = tsc_timestamp;
2699 /* If the host uses TSC clocksource, then it is stable */
2701 if (use_master_clock)
2702 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2704 vcpu->hv_clock.flags = pvclock_flags;
2706 if (vcpu->pv_time_enabled)
2707 kvm_setup_pvclock_page(v);
2708 if (v == kvm_get_vcpu(v->kvm, 0))
2709 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2714 * kvmclock updates which are isolated to a given vcpu, such as
2715 * vcpu->cpu migration, should not allow system_timestamp from
2716 * the rest of the vcpus to remain static. Otherwise ntp frequency
2717 * correction applies to one vcpu's system_timestamp but not
2720 * So in those cases, request a kvmclock update for all vcpus.
2721 * We need to rate-limit these requests though, as they can
2722 * considerably slow guests that have a large number of vcpus.
2723 * The time for a remote vcpu to update its kvmclock is bound
2724 * by the delay we use to rate-limit the updates.
2727 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2729 static void kvmclock_update_fn(struct work_struct *work)
2732 struct delayed_work *dwork = to_delayed_work(work);
2733 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2734 kvmclock_update_work);
2735 struct kvm *kvm = container_of(ka, struct kvm, arch);
2736 struct kvm_vcpu *vcpu;
2738 kvm_for_each_vcpu(i, vcpu, kvm) {
2739 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2740 kvm_vcpu_kick(vcpu);
2744 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2746 struct kvm *kvm = v->kvm;
2748 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2749 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2750 KVMCLOCK_UPDATE_DELAY);
2753 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2755 static void kvmclock_sync_fn(struct work_struct *work)
2757 struct delayed_work *dwork = to_delayed_work(work);
2758 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2759 kvmclock_sync_work);
2760 struct kvm *kvm = container_of(ka, struct kvm, arch);
2762 if (!kvmclock_periodic_sync)
2765 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2766 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2767 KVMCLOCK_SYNC_PERIOD);
2771 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2773 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2775 /* McStatusWrEn enabled? */
2776 if (guest_cpuid_is_amd_or_hygon(vcpu))
2777 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2782 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2784 u64 mcg_cap = vcpu->arch.mcg_cap;
2785 unsigned bank_num = mcg_cap & 0xff;
2786 u32 msr = msr_info->index;
2787 u64 data = msr_info->data;
2790 case MSR_IA32_MCG_STATUS:
2791 vcpu->arch.mcg_status = data;
2793 case MSR_IA32_MCG_CTL:
2794 if (!(mcg_cap & MCG_CTL_P) &&
2795 (data || !msr_info->host_initiated))
2797 if (data != 0 && data != ~(u64)0)
2799 vcpu->arch.mcg_ctl = data;
2802 if (msr >= MSR_IA32_MC0_CTL &&
2803 msr < MSR_IA32_MCx_CTL(bank_num)) {
2804 u32 offset = array_index_nospec(
2805 msr - MSR_IA32_MC0_CTL,
2806 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2808 /* only 0 or all 1s can be written to IA32_MCi_CTL
2809 * some Linux kernels though clear bit 10 in bank 4 to
2810 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2811 * this to avoid an uncatched #GP in the guest
2813 if ((offset & 0x3) == 0 &&
2814 data != 0 && (data | (1 << 10)) != ~(u64)0)
2818 if (!msr_info->host_initiated &&
2819 (offset & 0x3) == 1 && data != 0) {
2820 if (!can_set_mci_status(vcpu))
2824 vcpu->arch.mce_banks[offset] = data;
2832 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2834 struct kvm *kvm = vcpu->kvm;
2835 int lm = is_long_mode(vcpu);
2836 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2837 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2838 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2839 : kvm->arch.xen_hvm_config.blob_size_32;
2840 u32 page_num = data & ~PAGE_MASK;
2841 u64 page_addr = data & PAGE_MASK;
2846 if (page_num >= blob_size)
2849 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2854 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2863 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2865 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2867 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2870 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2872 gpa_t gpa = data & ~0x3f;
2874 /* Bits 4:5 are reserved, Should be zero */
2878 if (!lapic_in_kernel(vcpu))
2879 return data ? 1 : 0;
2881 vcpu->arch.apf.msr_en_val = data;
2883 if (!kvm_pv_async_pf_enabled(vcpu)) {
2884 kvm_clear_async_pf_completion_queue(vcpu);
2885 kvm_async_pf_hash_reset(vcpu);
2889 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2893 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2894 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2896 kvm_async_pf_wakeup_all(vcpu);
2901 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2903 /* Bits 8-63 are reserved */
2907 if (!lapic_in_kernel(vcpu))
2910 vcpu->arch.apf.msr_int_val = data;
2912 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2917 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2919 vcpu->arch.pv_time_enabled = false;
2920 vcpu->arch.time = 0;
2923 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2925 ++vcpu->stat.tlb_flush;
2926 kvm_x86_ops.tlb_flush_all(vcpu);
2929 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2931 ++vcpu->stat.tlb_flush;
2932 kvm_x86_ops.tlb_flush_guest(vcpu);
2935 static void record_steal_time(struct kvm_vcpu *vcpu)
2937 struct kvm_host_map map;
2938 struct kvm_steal_time *st;
2940 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2943 /* -EAGAIN is returned in atomic context so we can just return. */
2944 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2945 &map, &vcpu->arch.st.cache, false))
2949 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2952 * Doing a TLB flush here, on the guest's behalf, can avoid
2955 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2956 st->preempted & KVM_VCPU_FLUSH_TLB);
2957 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2958 kvm_vcpu_flush_tlb_guest(vcpu);
2960 vcpu->arch.st.preempted = 0;
2962 if (st->version & 1)
2963 st->version += 1; /* first time write, random junk */
2969 st->steal += current->sched_info.run_delay -
2970 vcpu->arch.st.last_steal;
2971 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2977 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2980 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2983 u32 msr = msr_info->index;
2984 u64 data = msr_info->data;
2987 case MSR_AMD64_NB_CFG:
2988 case MSR_IA32_UCODE_WRITE:
2989 case MSR_VM_HSAVE_PA:
2990 case MSR_AMD64_PATCH_LOADER:
2991 case MSR_AMD64_BU_CFG2:
2992 case MSR_AMD64_DC_CFG:
2993 case MSR_F15H_EX_CFG:
2996 case MSR_IA32_UCODE_REV:
2997 if (msr_info->host_initiated)
2998 vcpu->arch.microcode_version = data;
3000 case MSR_IA32_ARCH_CAPABILITIES:
3001 if (!msr_info->host_initiated)
3003 vcpu->arch.arch_capabilities = data;
3005 case MSR_IA32_PERF_CAPABILITIES: {
3006 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3008 if (!msr_info->host_initiated)
3010 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3012 if (data & ~msr_ent.data)
3015 vcpu->arch.perf_capabilities = data;
3020 return set_efer(vcpu, msr_info);
3022 data &= ~(u64)0x40; /* ignore flush filter disable */
3023 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3024 data &= ~(u64)0x8; /* ignore TLB cache disable */
3026 /* Handle McStatusWrEn */
3027 if (data == BIT_ULL(18)) {
3028 vcpu->arch.msr_hwcr = data;
3029 } else if (data != 0) {
3030 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3035 case MSR_FAM10H_MMIO_CONF_BASE:
3037 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3042 case MSR_IA32_DEBUGCTLMSR:
3044 /* We support the non-activated case already */
3046 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
3047 /* Values other than LBR and BTF are vendor-specific,
3048 thus reserved and should throw a #GP */
3051 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
3054 case 0x200 ... 0x2ff:
3055 return kvm_mtrr_set_msr(vcpu, msr, data);
3056 case MSR_IA32_APICBASE:
3057 return kvm_set_apic_base(vcpu, msr_info);
3058 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3059 return kvm_x2apic_msr_write(vcpu, msr, data);
3060 case MSR_IA32_TSCDEADLINE:
3061 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3063 case MSR_IA32_TSC_ADJUST:
3064 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3065 if (!msr_info->host_initiated) {
3066 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3067 adjust_tsc_offset_guest(vcpu, adj);
3069 vcpu->arch.ia32_tsc_adjust_msr = data;
3072 case MSR_IA32_MISC_ENABLE:
3073 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3074 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3075 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3077 vcpu->arch.ia32_misc_enable_msr = data;
3078 kvm_update_cpuid_runtime(vcpu);
3080 vcpu->arch.ia32_misc_enable_msr = data;
3083 case MSR_IA32_SMBASE:
3084 if (!msr_info->host_initiated)
3086 vcpu->arch.smbase = data;
3088 case MSR_IA32_POWER_CTL:
3089 vcpu->arch.msr_ia32_power_ctl = data;
3092 if (msr_info->host_initiated) {
3093 kvm_synchronize_tsc(vcpu, data);
3095 u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3096 adjust_tsc_offset_guest(vcpu, adj);
3097 vcpu->arch.ia32_tsc_adjust_msr += adj;
3101 if (!msr_info->host_initiated &&
3102 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3105 * KVM supports exposing PT to the guest, but does not support
3106 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3107 * XSAVES/XRSTORS to save/restore PT MSRs.
3109 if (data & ~supported_xss)
3111 vcpu->arch.ia32_xss = data;
3114 if (!msr_info->host_initiated)
3116 vcpu->arch.smi_count = data;
3118 case MSR_KVM_WALL_CLOCK_NEW:
3119 case MSR_KVM_WALL_CLOCK:
3120 vcpu->kvm->arch.wall_clock = data;
3121 kvm_write_wall_clock(vcpu->kvm, data);
3123 case MSR_KVM_SYSTEM_TIME_NEW:
3124 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3126 case MSR_KVM_SYSTEM_TIME:
3127 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3129 case MSR_KVM_ASYNC_PF_EN:
3130 if (kvm_pv_enable_async_pf(vcpu, data))
3133 case MSR_KVM_ASYNC_PF_INT:
3134 if (kvm_pv_enable_async_pf_int(vcpu, data))
3137 case MSR_KVM_ASYNC_PF_ACK:
3139 vcpu->arch.apf.pageready_pending = false;
3140 kvm_check_async_pf_completion(vcpu);
3143 case MSR_KVM_STEAL_TIME:
3145 if (unlikely(!sched_info_on()))
3148 if (data & KVM_STEAL_RESERVED_MASK)
3151 vcpu->arch.st.msr_val = data;
3153 if (!(data & KVM_MSR_ENABLED))
3156 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3159 case MSR_KVM_PV_EOI_EN:
3160 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3164 case MSR_KVM_POLL_CONTROL:
3165 /* only enable bit supported */
3166 if (data & (-1ULL << 1))
3169 vcpu->arch.msr_kvm_poll_control = data;
3172 case MSR_IA32_MCG_CTL:
3173 case MSR_IA32_MCG_STATUS:
3174 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3175 return set_msr_mce(vcpu, msr_info);
3177 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3178 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3181 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3182 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3183 if (kvm_pmu_is_valid_msr(vcpu, msr))
3184 return kvm_pmu_set_msr(vcpu, msr_info);
3186 if (pr || data != 0)
3187 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3188 "0x%x data 0x%llx\n", msr, data);
3190 case MSR_K7_CLK_CTL:
3192 * Ignore all writes to this no longer documented MSR.
3193 * Writes are only relevant for old K7 processors,
3194 * all pre-dating SVM, but a recommended workaround from
3195 * AMD for these chips. It is possible to specify the
3196 * affected processor models on the command line, hence
3197 * the need to ignore the workaround.
3200 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3201 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3202 case HV_X64_MSR_SYNDBG_OPTIONS:
3203 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3204 case HV_X64_MSR_CRASH_CTL:
3205 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3206 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3207 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3208 case HV_X64_MSR_TSC_EMULATION_STATUS:
3209 return kvm_hv_set_msr_common(vcpu, msr, data,
3210 msr_info->host_initiated);
3211 case MSR_IA32_BBL_CR_CTL3:
3212 /* Drop writes to this legacy MSR -- see rdmsr
3213 * counterpart for further detail.
3215 if (report_ignored_msrs)
3216 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3219 case MSR_AMD64_OSVW_ID_LENGTH:
3220 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3222 vcpu->arch.osvw.length = data;
3224 case MSR_AMD64_OSVW_STATUS:
3225 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3227 vcpu->arch.osvw.status = data;
3229 case MSR_PLATFORM_INFO:
3230 if (!msr_info->host_initiated ||
3231 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3232 cpuid_fault_enabled(vcpu)))
3234 vcpu->arch.msr_platform_info = data;
3236 case MSR_MISC_FEATURES_ENABLES:
3237 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3238 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3239 !supports_cpuid_fault(vcpu)))
3241 vcpu->arch.msr_misc_features_enables = data;
3244 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
3245 return xen_hvm_config(vcpu, data);
3246 if (kvm_pmu_is_valid_msr(vcpu, msr))
3247 return kvm_pmu_set_msr(vcpu, msr_info);
3248 return KVM_MSR_RET_INVALID;
3252 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3254 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3257 u64 mcg_cap = vcpu->arch.mcg_cap;
3258 unsigned bank_num = mcg_cap & 0xff;
3261 case MSR_IA32_P5_MC_ADDR:
3262 case MSR_IA32_P5_MC_TYPE:
3265 case MSR_IA32_MCG_CAP:
3266 data = vcpu->arch.mcg_cap;
3268 case MSR_IA32_MCG_CTL:
3269 if (!(mcg_cap & MCG_CTL_P) && !host)
3271 data = vcpu->arch.mcg_ctl;
3273 case MSR_IA32_MCG_STATUS:
3274 data = vcpu->arch.mcg_status;
3277 if (msr >= MSR_IA32_MC0_CTL &&
3278 msr < MSR_IA32_MCx_CTL(bank_num)) {
3279 u32 offset = array_index_nospec(
3280 msr - MSR_IA32_MC0_CTL,
3281 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3283 data = vcpu->arch.mce_banks[offset];
3292 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3294 switch (msr_info->index) {
3295 case MSR_IA32_PLATFORM_ID:
3296 case MSR_IA32_EBL_CR_POWERON:
3297 case MSR_IA32_DEBUGCTLMSR:
3298 case MSR_IA32_LASTBRANCHFROMIP:
3299 case MSR_IA32_LASTBRANCHTOIP:
3300 case MSR_IA32_LASTINTFROMIP:
3301 case MSR_IA32_LASTINTTOIP:
3303 case MSR_K8_TSEG_ADDR:
3304 case MSR_K8_TSEG_MASK:
3305 case MSR_VM_HSAVE_PA:
3306 case MSR_K8_INT_PENDING_MSG:
3307 case MSR_AMD64_NB_CFG:
3308 case MSR_FAM10H_MMIO_CONF_BASE:
3309 case MSR_AMD64_BU_CFG2:
3310 case MSR_IA32_PERF_CTL:
3311 case MSR_AMD64_DC_CFG:
3312 case MSR_F15H_EX_CFG:
3314 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3315 * limit) MSRs. Just return 0, as we do not want to expose the host
3316 * data here. Do not conditionalize this on CPUID, as KVM does not do
3317 * so for existing CPU-specific MSRs.
3319 case MSR_RAPL_POWER_UNIT:
3320 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3321 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3322 case MSR_PKG_ENERGY_STATUS: /* Total package */
3323 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3326 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3327 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3328 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3329 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3330 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3331 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3332 return kvm_pmu_get_msr(vcpu, msr_info);
3335 case MSR_IA32_UCODE_REV:
3336 msr_info->data = vcpu->arch.microcode_version;
3338 case MSR_IA32_ARCH_CAPABILITIES:
3339 if (!msr_info->host_initiated &&
3340 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3342 msr_info->data = vcpu->arch.arch_capabilities;
3344 case MSR_IA32_PERF_CAPABILITIES:
3345 if (!msr_info->host_initiated &&
3346 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3348 msr_info->data = vcpu->arch.perf_capabilities;
3350 case MSR_IA32_POWER_CTL:
3351 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3353 case MSR_IA32_TSC: {
3355 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3356 * even when not intercepted. AMD manual doesn't explicitly
3357 * state this but appears to behave the same.
3359 * Unconditionally return L1's TSC offset on userspace reads
3360 * so that userspace reads and writes always operate on L1's
3361 * offset, e.g. to ensure deterministic behavior for migration.
3363 u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3364 vcpu->arch.tsc_offset;
3366 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3370 case 0x200 ... 0x2ff:
3371 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3372 case 0xcd: /* fsb frequency */
3376 * MSR_EBC_FREQUENCY_ID
3377 * Conservative value valid for even the basic CPU models.
3378 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3379 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3380 * and 266MHz for model 3, or 4. Set Core Clock
3381 * Frequency to System Bus Frequency Ratio to 1 (bits
3382 * 31:24) even though these are only valid for CPU
3383 * models > 2, however guests may end up dividing or
3384 * multiplying by zero otherwise.
3386 case MSR_EBC_FREQUENCY_ID:
3387 msr_info->data = 1 << 24;
3389 case MSR_IA32_APICBASE:
3390 msr_info->data = kvm_get_apic_base(vcpu);
3392 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3393 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3394 case MSR_IA32_TSCDEADLINE:
3395 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3397 case MSR_IA32_TSC_ADJUST:
3398 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3400 case MSR_IA32_MISC_ENABLE:
3401 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3403 case MSR_IA32_SMBASE:
3404 if (!msr_info->host_initiated)
3406 msr_info->data = vcpu->arch.smbase;
3409 msr_info->data = vcpu->arch.smi_count;
3411 case MSR_IA32_PERF_STATUS:
3412 /* TSC increment by tick */
3413 msr_info->data = 1000ULL;
3414 /* CPU multiplier */
3415 msr_info->data |= (((uint64_t)4ULL) << 40);
3418 msr_info->data = vcpu->arch.efer;
3420 case MSR_KVM_WALL_CLOCK:
3421 case MSR_KVM_WALL_CLOCK_NEW:
3422 msr_info->data = vcpu->kvm->arch.wall_clock;
3424 case MSR_KVM_SYSTEM_TIME:
3425 case MSR_KVM_SYSTEM_TIME_NEW:
3426 msr_info->data = vcpu->arch.time;
3428 case MSR_KVM_ASYNC_PF_EN:
3429 msr_info->data = vcpu->arch.apf.msr_en_val;
3431 case MSR_KVM_ASYNC_PF_INT:
3432 msr_info->data = vcpu->arch.apf.msr_int_val;
3434 case MSR_KVM_ASYNC_PF_ACK:
3437 case MSR_KVM_STEAL_TIME:
3438 msr_info->data = vcpu->arch.st.msr_val;
3440 case MSR_KVM_PV_EOI_EN:
3441 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3443 case MSR_KVM_POLL_CONTROL:
3444 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3446 case MSR_IA32_P5_MC_ADDR:
3447 case MSR_IA32_P5_MC_TYPE:
3448 case MSR_IA32_MCG_CAP:
3449 case MSR_IA32_MCG_CTL:
3450 case MSR_IA32_MCG_STATUS:
3451 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3452 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3453 msr_info->host_initiated);
3455 if (!msr_info->host_initiated &&
3456 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3458 msr_info->data = vcpu->arch.ia32_xss;
3460 case MSR_K7_CLK_CTL:
3462 * Provide expected ramp-up count for K7. All other
3463 * are set to zero, indicating minimum divisors for
3466 * This prevents guest kernels on AMD host with CPU
3467 * type 6, model 8 and higher from exploding due to
3468 * the rdmsr failing.
3470 msr_info->data = 0x20000000;
3472 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3473 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3474 case HV_X64_MSR_SYNDBG_OPTIONS:
3475 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3476 case HV_X64_MSR_CRASH_CTL:
3477 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3478 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3479 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3480 case HV_X64_MSR_TSC_EMULATION_STATUS:
3481 return kvm_hv_get_msr_common(vcpu,
3482 msr_info->index, &msr_info->data,
3483 msr_info->host_initiated);
3484 case MSR_IA32_BBL_CR_CTL3:
3485 /* This legacy MSR exists but isn't fully documented in current
3486 * silicon. It is however accessed by winxp in very narrow
3487 * scenarios where it sets bit #19, itself documented as
3488 * a "reserved" bit. Best effort attempt to source coherent
3489 * read data here should the balance of the register be
3490 * interpreted by the guest:
3492 * L2 cache control register 3: 64GB range, 256KB size,
3493 * enabled, latency 0x1, configured
3495 msr_info->data = 0xbe702111;
3497 case MSR_AMD64_OSVW_ID_LENGTH:
3498 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3500 msr_info->data = vcpu->arch.osvw.length;
3502 case MSR_AMD64_OSVW_STATUS:
3503 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3505 msr_info->data = vcpu->arch.osvw.status;
3507 case MSR_PLATFORM_INFO:
3508 if (!msr_info->host_initiated &&
3509 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3511 msr_info->data = vcpu->arch.msr_platform_info;
3513 case MSR_MISC_FEATURES_ENABLES:
3514 msr_info->data = vcpu->arch.msr_misc_features_enables;
3517 msr_info->data = vcpu->arch.msr_hwcr;
3520 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3521 return kvm_pmu_get_msr(vcpu, msr_info);
3522 return KVM_MSR_RET_INVALID;
3526 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3529 * Read or write a bunch of msrs. All parameters are kernel addresses.
3531 * @return number of msrs set successfully.
3533 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3534 struct kvm_msr_entry *entries,
3535 int (*do_msr)(struct kvm_vcpu *vcpu,
3536 unsigned index, u64 *data))
3540 for (i = 0; i < msrs->nmsrs; ++i)
3541 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3548 * Read or write a bunch of msrs. Parameters are user addresses.
3550 * @return number of msrs set successfully.
3552 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3553 int (*do_msr)(struct kvm_vcpu *vcpu,
3554 unsigned index, u64 *data),
3557 struct kvm_msrs msrs;
3558 struct kvm_msr_entry *entries;
3563 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3567 if (msrs.nmsrs >= MAX_IO_MSRS)
3570 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3571 entries = memdup_user(user_msrs->entries, size);
3572 if (IS_ERR(entries)) {
3573 r = PTR_ERR(entries);
3577 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3582 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3593 static inline bool kvm_can_mwait_in_guest(void)
3595 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3596 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3597 boot_cpu_has(X86_FEATURE_ARAT);
3600 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3605 case KVM_CAP_IRQCHIP:
3607 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3608 case KVM_CAP_SET_TSS_ADDR:
3609 case KVM_CAP_EXT_CPUID:
3610 case KVM_CAP_EXT_EMUL_CPUID:
3611 case KVM_CAP_CLOCKSOURCE:
3613 case KVM_CAP_NOP_IO_DELAY:
3614 case KVM_CAP_MP_STATE:
3615 case KVM_CAP_SYNC_MMU:
3616 case KVM_CAP_USER_NMI:
3617 case KVM_CAP_REINJECT_CONTROL:
3618 case KVM_CAP_IRQ_INJECT_STATUS:
3619 case KVM_CAP_IOEVENTFD:
3620 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3622 case KVM_CAP_PIT_STATE2:
3623 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3624 case KVM_CAP_XEN_HVM:
3625 case KVM_CAP_VCPU_EVENTS:
3626 case KVM_CAP_HYPERV:
3627 case KVM_CAP_HYPERV_VAPIC:
3628 case KVM_CAP_HYPERV_SPIN:
3629 case KVM_CAP_HYPERV_SYNIC:
3630 case KVM_CAP_HYPERV_SYNIC2:
3631 case KVM_CAP_HYPERV_VP_INDEX:
3632 case KVM_CAP_HYPERV_EVENTFD:
3633 case KVM_CAP_HYPERV_TLBFLUSH:
3634 case KVM_CAP_HYPERV_SEND_IPI:
3635 case KVM_CAP_HYPERV_CPUID:
3636 case KVM_CAP_PCI_SEGMENT:
3637 case KVM_CAP_DEBUGREGS:
3638 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3640 case KVM_CAP_ASYNC_PF:
3641 case KVM_CAP_ASYNC_PF_INT:
3642 case KVM_CAP_GET_TSC_KHZ:
3643 case KVM_CAP_KVMCLOCK_CTRL:
3644 case KVM_CAP_READONLY_MEM:
3645 case KVM_CAP_HYPERV_TIME:
3646 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3647 case KVM_CAP_TSC_DEADLINE_TIMER:
3648 case KVM_CAP_DISABLE_QUIRKS:
3649 case KVM_CAP_SET_BOOT_CPU_ID:
3650 case KVM_CAP_SPLIT_IRQCHIP:
3651 case KVM_CAP_IMMEDIATE_EXIT:
3652 case KVM_CAP_PMU_EVENT_FILTER:
3653 case KVM_CAP_GET_MSR_FEATURES:
3654 case KVM_CAP_MSR_PLATFORM_INFO:
3655 case KVM_CAP_EXCEPTION_PAYLOAD:
3656 case KVM_CAP_SET_GUEST_DEBUG:
3657 case KVM_CAP_LAST_CPU:
3658 case KVM_CAP_X86_USER_SPACE_MSR:
3659 case KVM_CAP_X86_MSR_FILTER:
3662 case KVM_CAP_SYNC_REGS:
3663 r = KVM_SYNC_X86_VALID_FIELDS;
3665 case KVM_CAP_ADJUST_CLOCK:
3666 r = KVM_CLOCK_TSC_STABLE;
3668 case KVM_CAP_X86_DISABLE_EXITS:
3669 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3670 KVM_X86_DISABLE_EXITS_CSTATE;
3671 if(kvm_can_mwait_in_guest())
3672 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3674 case KVM_CAP_X86_SMM:
3675 /* SMBASE is usually relocated above 1M on modern chipsets,
3676 * and SMM handlers might indeed rely on 4G segment limits,
3677 * so do not report SMM to be available if real mode is
3678 * emulated via vm86 mode. Still, do not go to great lengths
3679 * to avoid userspace's usage of the feature, because it is a
3680 * fringe case that is not enabled except via specific settings
3681 * of the module parameters.
3683 r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3686 r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3688 case KVM_CAP_NR_VCPUS:
3689 r = KVM_SOFT_MAX_VCPUS;
3691 case KVM_CAP_MAX_VCPUS:
3694 case KVM_CAP_MAX_VCPU_ID:
3695 r = KVM_MAX_VCPU_ID;
3697 case KVM_CAP_PV_MMU: /* obsolete */
3701 r = KVM_MAX_MCE_BANKS;
3704 r = boot_cpu_has(X86_FEATURE_XSAVE);
3706 case KVM_CAP_TSC_CONTROL:
3707 r = kvm_has_tsc_control;
3709 case KVM_CAP_X2APIC_API:
3710 r = KVM_X2APIC_API_VALID_FLAGS;
3712 case KVM_CAP_NESTED_STATE:
3713 r = kvm_x86_ops.nested_ops->get_state ?
3714 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3716 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3717 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3719 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3720 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3722 case KVM_CAP_SMALLER_MAXPHYADDR:
3723 r = (int) allow_smaller_maxphyaddr;
3725 case KVM_CAP_STEAL_TIME:
3726 r = sched_info_on();
3735 long kvm_arch_dev_ioctl(struct file *filp,
3736 unsigned int ioctl, unsigned long arg)
3738 void __user *argp = (void __user *)arg;
3742 case KVM_GET_MSR_INDEX_LIST: {
3743 struct kvm_msr_list __user *user_msr_list = argp;
3744 struct kvm_msr_list msr_list;
3748 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3751 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3752 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3755 if (n < msr_list.nmsrs)
3758 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3759 num_msrs_to_save * sizeof(u32)))
3761 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3763 num_emulated_msrs * sizeof(u32)))
3768 case KVM_GET_SUPPORTED_CPUID:
3769 case KVM_GET_EMULATED_CPUID: {
3770 struct kvm_cpuid2 __user *cpuid_arg = argp;
3771 struct kvm_cpuid2 cpuid;
3774 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3777 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3783 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3788 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3790 if (copy_to_user(argp, &kvm_mce_cap_supported,
3791 sizeof(kvm_mce_cap_supported)))
3795 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3796 struct kvm_msr_list __user *user_msr_list = argp;
3797 struct kvm_msr_list msr_list;
3801 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3804 msr_list.nmsrs = num_msr_based_features;
3805 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3808 if (n < msr_list.nmsrs)
3811 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3812 num_msr_based_features * sizeof(u32)))
3818 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3828 static void wbinvd_ipi(void *garbage)
3833 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3835 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3838 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3840 /* Address WBINVD may be executed by guest */
3841 if (need_emulate_wbinvd(vcpu)) {
3842 if (kvm_x86_ops.has_wbinvd_exit())
3843 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3844 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3845 smp_call_function_single(vcpu->cpu,
3846 wbinvd_ipi, NULL, 1);
3849 kvm_x86_ops.vcpu_load(vcpu, cpu);
3851 /* Save host pkru register if supported */
3852 vcpu->arch.host_pkru = read_pkru();
3854 /* Apply any externally detected TSC adjustments (due to suspend) */
3855 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3856 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3857 vcpu->arch.tsc_offset_adjustment = 0;
3858 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3861 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3862 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3863 rdtsc() - vcpu->arch.last_host_tsc;
3865 mark_tsc_unstable("KVM discovered backwards TSC");
3867 if (kvm_check_tsc_unstable()) {
3868 u64 offset = kvm_compute_tsc_offset(vcpu,
3869 vcpu->arch.last_guest_tsc);
3870 kvm_vcpu_write_tsc_offset(vcpu, offset);
3871 vcpu->arch.tsc_catchup = 1;
3874 if (kvm_lapic_hv_timer_in_use(vcpu))
3875 kvm_lapic_restart_hv_timer(vcpu);
3878 * On a host with synchronized TSC, there is no need to update
3879 * kvmclock on vcpu->cpu migration
3881 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3882 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3883 if (vcpu->cpu != cpu)
3884 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3888 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3891 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3893 struct kvm_host_map map;
3894 struct kvm_steal_time *st;
3896 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3899 if (vcpu->arch.st.preempted)
3902 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3903 &vcpu->arch.st.cache, true))
3907 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3909 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3911 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3914 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3918 if (vcpu->preempted)
3919 vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
3922 * Disable page faults because we're in atomic context here.
3923 * kvm_write_guest_offset_cached() would call might_fault()
3924 * that relies on pagefault_disable() to tell if there's a
3925 * bug. NOTE: the write to guest memory may not go through if
3926 * during postcopy live migration or if there's heavy guest
3929 pagefault_disable();
3931 * kvm_memslots() will be called by
3932 * kvm_write_guest_offset_cached() so take the srcu lock.
3934 idx = srcu_read_lock(&vcpu->kvm->srcu);
3935 kvm_steal_time_set_preempted(vcpu);
3936 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3938 kvm_x86_ops.vcpu_put(vcpu);
3939 vcpu->arch.last_host_tsc = rdtsc();
3941 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3942 * on every vmexit, but if not, we might have a stale dr6 from the
3943 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3948 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3949 struct kvm_lapic_state *s)
3951 if (vcpu->arch.apicv_active)
3952 kvm_x86_ops.sync_pir_to_irr(vcpu);
3954 return kvm_apic_get_state(vcpu, s);
3957 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3958 struct kvm_lapic_state *s)
3962 r = kvm_apic_set_state(vcpu, s);
3965 update_cr8_intercept(vcpu);
3970 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3972 return (!lapic_in_kernel(vcpu) ||
3973 kvm_apic_accept_pic_intr(vcpu));
3977 * if userspace requested an interrupt window, check that the
3978 * interrupt window is open.
3980 * No need to exit to userspace if we already have an interrupt queued.
3982 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3984 return kvm_arch_interrupt_allowed(vcpu) &&
3985 !kvm_cpu_has_interrupt(vcpu) &&
3986 !kvm_event_needs_reinjection(vcpu) &&
3987 kvm_cpu_accept_dm_intr(vcpu);
3990 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3991 struct kvm_interrupt *irq)
3993 if (irq->irq >= KVM_NR_INTERRUPTS)
3996 if (!irqchip_in_kernel(vcpu->kvm)) {
3997 kvm_queue_interrupt(vcpu, irq->irq, false);
3998 kvm_make_request(KVM_REQ_EVENT, vcpu);
4003 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4004 * fail for in-kernel 8259.
4006 if (pic_in_kernel(vcpu->kvm))
4009 if (vcpu->arch.pending_external_vector != -1)
4012 vcpu->arch.pending_external_vector = irq->irq;
4013 kvm_make_request(KVM_REQ_EVENT, vcpu);
4017 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4019 kvm_inject_nmi(vcpu);
4024 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4026 kvm_make_request(KVM_REQ_SMI, vcpu);
4031 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4032 struct kvm_tpr_access_ctl *tac)
4036 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4040 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4044 unsigned bank_num = mcg_cap & 0xff, bank;
4047 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4049 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4052 vcpu->arch.mcg_cap = mcg_cap;
4053 /* Init IA32_MCG_CTL to all 1s */
4054 if (mcg_cap & MCG_CTL_P)
4055 vcpu->arch.mcg_ctl = ~(u64)0;
4056 /* Init IA32_MCi_CTL to all 1s */
4057 for (bank = 0; bank < bank_num; bank++)
4058 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4060 kvm_x86_ops.setup_mce(vcpu);
4065 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4066 struct kvm_x86_mce *mce)
4068 u64 mcg_cap = vcpu->arch.mcg_cap;
4069 unsigned bank_num = mcg_cap & 0xff;
4070 u64 *banks = vcpu->arch.mce_banks;
4072 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4075 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4076 * reporting is disabled
4078 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4079 vcpu->arch.mcg_ctl != ~(u64)0)
4081 banks += 4 * mce->bank;
4083 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4084 * reporting is disabled for the bank
4086 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4088 if (mce->status & MCI_STATUS_UC) {
4089 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4090 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4091 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4094 if (banks[1] & MCI_STATUS_VAL)
4095 mce->status |= MCI_STATUS_OVER;
4096 banks[2] = mce->addr;
4097 banks[3] = mce->misc;
4098 vcpu->arch.mcg_status = mce->mcg_status;
4099 banks[1] = mce->status;
4100 kvm_queue_exception(vcpu, MC_VECTOR);
4101 } else if (!(banks[1] & MCI_STATUS_VAL)
4102 || !(banks[1] & MCI_STATUS_UC)) {
4103 if (banks[1] & MCI_STATUS_VAL)
4104 mce->status |= MCI_STATUS_OVER;
4105 banks[2] = mce->addr;
4106 banks[3] = mce->misc;
4107 banks[1] = mce->status;
4109 banks[1] |= MCI_STATUS_OVER;
4113 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4114 struct kvm_vcpu_events *events)
4119 * In guest mode, payload delivery should be deferred,
4120 * so that the L1 hypervisor can intercept #PF before
4121 * CR2 is modified (or intercept #DB before DR6 is
4122 * modified under nVMX). Unless the per-VM capability,
4123 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4124 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4125 * opportunistically defer the exception payload, deliver it if the
4126 * capability hasn't been requested before processing a
4127 * KVM_GET_VCPU_EVENTS.
4129 if (!vcpu->kvm->arch.exception_payload_enabled &&
4130 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4131 kvm_deliver_exception_payload(vcpu);
4134 * The API doesn't provide the instruction length for software
4135 * exceptions, so don't report them. As long as the guest RIP
4136 * isn't advanced, we should expect to encounter the exception
4139 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4140 events->exception.injected = 0;
4141 events->exception.pending = 0;
4143 events->exception.injected = vcpu->arch.exception.injected;
4144 events->exception.pending = vcpu->arch.exception.pending;
4146 * For ABI compatibility, deliberately conflate
4147 * pending and injected exceptions when
4148 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4150 if (!vcpu->kvm->arch.exception_payload_enabled)
4151 events->exception.injected |=
4152 vcpu->arch.exception.pending;
4154 events->exception.nr = vcpu->arch.exception.nr;
4155 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4156 events->exception.error_code = vcpu->arch.exception.error_code;
4157 events->exception_has_payload = vcpu->arch.exception.has_payload;
4158 events->exception_payload = vcpu->arch.exception.payload;
4160 events->interrupt.injected =
4161 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4162 events->interrupt.nr = vcpu->arch.interrupt.nr;
4163 events->interrupt.soft = 0;
4164 events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
4166 events->nmi.injected = vcpu->arch.nmi_injected;
4167 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4168 events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
4169 events->nmi.pad = 0;
4171 events->sipi_vector = 0; /* never valid when reporting to user space */
4173 events->smi.smm = is_smm(vcpu);
4174 events->smi.pending = vcpu->arch.smi_pending;
4175 events->smi.smm_inside_nmi =
4176 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4177 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4179 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4180 | KVM_VCPUEVENT_VALID_SHADOW
4181 | KVM_VCPUEVENT_VALID_SMM);
4182 if (vcpu->kvm->arch.exception_payload_enabled)
4183 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4185 memset(&events->reserved, 0, sizeof(events->reserved));
4188 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4190 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4191 struct kvm_vcpu_events *events)
4193 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4194 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4195 | KVM_VCPUEVENT_VALID_SHADOW
4196 | KVM_VCPUEVENT_VALID_SMM
4197 | KVM_VCPUEVENT_VALID_PAYLOAD))
4200 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4201 if (!vcpu->kvm->arch.exception_payload_enabled)
4203 if (events->exception.pending)
4204 events->exception.injected = 0;
4206 events->exception_has_payload = 0;
4208 events->exception.pending = 0;
4209 events->exception_has_payload = 0;
4212 if ((events->exception.injected || events->exception.pending) &&
4213 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4216 /* INITs are latched while in SMM */
4217 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4218 (events->smi.smm || events->smi.pending) &&
4219 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4223 vcpu->arch.exception.injected = events->exception.injected;
4224 vcpu->arch.exception.pending = events->exception.pending;
4225 vcpu->arch.exception.nr = events->exception.nr;
4226 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4227 vcpu->arch.exception.error_code = events->exception.error_code;
4228 vcpu->arch.exception.has_payload = events->exception_has_payload;
4229 vcpu->arch.exception.payload = events->exception_payload;
4231 vcpu->arch.interrupt.injected = events->interrupt.injected;
4232 vcpu->arch.interrupt.nr = events->interrupt.nr;
4233 vcpu->arch.interrupt.soft = events->interrupt.soft;
4234 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4235 kvm_x86_ops.set_interrupt_shadow(vcpu,
4236 events->interrupt.shadow);
4238 vcpu->arch.nmi_injected = events->nmi.injected;
4239 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4240 vcpu->arch.nmi_pending = events->nmi.pending;
4241 kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
4243 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4244 lapic_in_kernel(vcpu))
4245 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4247 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4248 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4249 if (events->smi.smm)
4250 vcpu->arch.hflags |= HF_SMM_MASK;
4252 vcpu->arch.hflags &= ~HF_SMM_MASK;
4253 kvm_smm_changed(vcpu);
4256 vcpu->arch.smi_pending = events->smi.pending;
4258 if (events->smi.smm) {
4259 if (events->smi.smm_inside_nmi)
4260 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4262 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4265 if (lapic_in_kernel(vcpu)) {
4266 if (events->smi.latched_init)
4267 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4269 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4273 kvm_make_request(KVM_REQ_EVENT, vcpu);
4278 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4279 struct kvm_debugregs *dbgregs)
4283 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4284 kvm_get_dr(vcpu, 6, &val);
4286 dbgregs->dr7 = vcpu->arch.dr7;
4288 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4291 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4292 struct kvm_debugregs *dbgregs)
4297 if (dbgregs->dr6 & ~0xffffffffull)
4299 if (dbgregs->dr7 & ~0xffffffffull)
4302 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4303 kvm_update_dr0123(vcpu);
4304 vcpu->arch.dr6 = dbgregs->dr6;
4305 vcpu->arch.dr7 = dbgregs->dr7;
4306 kvm_update_dr7(vcpu);
4311 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4313 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4315 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4316 u64 xstate_bv = xsave->header.xfeatures;
4320 * Copy legacy XSAVE area, to avoid complications with CPUID
4321 * leaves 0 and 1 in the loop below.
4323 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4326 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4327 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4330 * Copy each region from the possibly compacted offset to the
4331 * non-compacted offset.
4333 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4335 u64 xfeature_mask = valid & -valid;
4336 int xfeature_nr = fls64(xfeature_mask) - 1;
4337 void *src = get_xsave_addr(xsave, xfeature_nr);
4340 u32 size, offset, ecx, edx;
4341 cpuid_count(XSTATE_CPUID, xfeature_nr,
4342 &size, &offset, &ecx, &edx);
4343 if (xfeature_nr == XFEATURE_PKRU)
4344 memcpy(dest + offset, &vcpu->arch.pkru,
4345 sizeof(vcpu->arch.pkru));
4347 memcpy(dest + offset, src, size);
4351 valid -= xfeature_mask;
4355 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4357 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4358 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4362 * Copy legacy XSAVE area, to avoid complications with CPUID
4363 * leaves 0 and 1 in the loop below.
4365 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4367 /* Set XSTATE_BV and possibly XCOMP_BV. */
4368 xsave->header.xfeatures = xstate_bv;
4369 if (boot_cpu_has(X86_FEATURE_XSAVES))
4370 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4373 * Copy each region from the non-compacted offset to the
4374 * possibly compacted offset.
4376 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4378 u64 xfeature_mask = valid & -valid;
4379 int xfeature_nr = fls64(xfeature_mask) - 1;
4380 void *dest = get_xsave_addr(xsave, xfeature_nr);
4383 u32 size, offset, ecx, edx;
4384 cpuid_count(XSTATE_CPUID, xfeature_nr,
4385 &size, &offset, &ecx, &edx);
4386 if (xfeature_nr == XFEATURE_PKRU)
4387 memcpy(&vcpu->arch.pkru, src + offset,
4388 sizeof(vcpu->arch.pkru));
4390 memcpy(dest, src + offset, size);
4393 valid -= xfeature_mask;
4397 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4398 struct kvm_xsave *guest_xsave)
4400 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4401 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4402 fill_xsave((u8 *) guest_xsave->region, vcpu);
4404 memcpy(guest_xsave->region,
4405 &vcpu->arch.guest_fpu->state.fxsave,
4406 sizeof(struct fxregs_state));
4407 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4408 XFEATURE_MASK_FPSSE;
4412 #define XSAVE_MXCSR_OFFSET 24
4414 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4415 struct kvm_xsave *guest_xsave)
4418 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4419 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4421 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4423 * Here we allow setting states that are not present in
4424 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4425 * with old userspace.
4427 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4429 load_xsave(vcpu, (u8 *)guest_xsave->region);
4431 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4432 mxcsr & ~mxcsr_feature_mask)
4434 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4435 guest_xsave->region, sizeof(struct fxregs_state));
4440 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4441 struct kvm_xcrs *guest_xcrs)
4443 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4444 guest_xcrs->nr_xcrs = 0;
4448 guest_xcrs->nr_xcrs = 1;
4449 guest_xcrs->flags = 0;
4450 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4451 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4454 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4455 struct kvm_xcrs *guest_xcrs)
4459 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4462 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4465 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4466 /* Only support XCR0 currently */
4467 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4468 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4469 guest_xcrs->xcrs[i].value);
4478 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4479 * stopped by the hypervisor. This function will be called from the host only.
4480 * EINVAL is returned when the host attempts to set the flag for a guest that
4481 * does not support pv clocks.
4483 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4485 if (!vcpu->arch.pv_time_enabled)
4487 vcpu->arch.pvclock_set_guest_stopped_request = true;
4488 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4492 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4493 struct kvm_enable_cap *cap)
4496 uint16_t vmcs_version;
4497 void __user *user_ptr;
4503 case KVM_CAP_HYPERV_SYNIC2:
4508 case KVM_CAP_HYPERV_SYNIC:
4509 if (!irqchip_in_kernel(vcpu->kvm))
4511 return kvm_hv_activate_synic(vcpu, cap->cap ==
4512 KVM_CAP_HYPERV_SYNIC2);
4513 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4514 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4516 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4518 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4519 if (copy_to_user(user_ptr, &vmcs_version,
4520 sizeof(vmcs_version)))
4524 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4525 if (!kvm_x86_ops.enable_direct_tlbflush)
4528 return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4535 long kvm_arch_vcpu_ioctl(struct file *filp,
4536 unsigned int ioctl, unsigned long arg)
4538 struct kvm_vcpu *vcpu = filp->private_data;
4539 void __user *argp = (void __user *)arg;
4542 struct kvm_lapic_state *lapic;
4543 struct kvm_xsave *xsave;
4544 struct kvm_xcrs *xcrs;
4552 case KVM_GET_LAPIC: {
4554 if (!lapic_in_kernel(vcpu))
4556 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4557 GFP_KERNEL_ACCOUNT);
4562 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4566 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4571 case KVM_SET_LAPIC: {
4573 if (!lapic_in_kernel(vcpu))
4575 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4576 if (IS_ERR(u.lapic)) {
4577 r = PTR_ERR(u.lapic);
4581 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4584 case KVM_INTERRUPT: {
4585 struct kvm_interrupt irq;
4588 if (copy_from_user(&irq, argp, sizeof(irq)))
4590 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4594 r = kvm_vcpu_ioctl_nmi(vcpu);
4598 r = kvm_vcpu_ioctl_smi(vcpu);
4601 case KVM_SET_CPUID: {
4602 struct kvm_cpuid __user *cpuid_arg = argp;
4603 struct kvm_cpuid cpuid;
4606 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4608 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4611 case KVM_SET_CPUID2: {
4612 struct kvm_cpuid2 __user *cpuid_arg = argp;
4613 struct kvm_cpuid2 cpuid;
4616 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4618 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4619 cpuid_arg->entries);
4622 case KVM_GET_CPUID2: {
4623 struct kvm_cpuid2 __user *cpuid_arg = argp;
4624 struct kvm_cpuid2 cpuid;
4627 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4629 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4630 cpuid_arg->entries);
4634 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4639 case KVM_GET_MSRS: {
4640 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4641 r = msr_io(vcpu, argp, do_get_msr, 1);
4642 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4645 case KVM_SET_MSRS: {
4646 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4647 r = msr_io(vcpu, argp, do_set_msr, 0);
4648 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4651 case KVM_TPR_ACCESS_REPORTING: {
4652 struct kvm_tpr_access_ctl tac;
4655 if (copy_from_user(&tac, argp, sizeof(tac)))
4657 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4661 if (copy_to_user(argp, &tac, sizeof(tac)))
4666 case KVM_SET_VAPIC_ADDR: {
4667 struct kvm_vapic_addr va;
4671 if (!lapic_in_kernel(vcpu))
4674 if (copy_from_user(&va, argp, sizeof(va)))
4676 idx = srcu_read_lock(&vcpu->kvm->srcu);
4677 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4678 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4681 case KVM_X86_SETUP_MCE: {
4685 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4687 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4690 case KVM_X86_SET_MCE: {
4691 struct kvm_x86_mce mce;
4694 if (copy_from_user(&mce, argp, sizeof(mce)))
4696 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4699 case KVM_GET_VCPU_EVENTS: {
4700 struct kvm_vcpu_events events;
4702 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4705 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4710 case KVM_SET_VCPU_EVENTS: {
4711 struct kvm_vcpu_events events;
4714 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4717 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4720 case KVM_GET_DEBUGREGS: {
4721 struct kvm_debugregs dbgregs;
4723 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4726 if (copy_to_user(argp, &dbgregs,
4727 sizeof(struct kvm_debugregs)))
4732 case KVM_SET_DEBUGREGS: {
4733 struct kvm_debugregs dbgregs;
4736 if (copy_from_user(&dbgregs, argp,
4737 sizeof(struct kvm_debugregs)))
4740 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4743 case KVM_GET_XSAVE: {
4744 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4749 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4752 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4757 case KVM_SET_XSAVE: {
4758 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4759 if (IS_ERR(u.xsave)) {
4760 r = PTR_ERR(u.xsave);
4764 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4767 case KVM_GET_XCRS: {
4768 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4773 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4776 if (copy_to_user(argp, u.xcrs,
4777 sizeof(struct kvm_xcrs)))
4782 case KVM_SET_XCRS: {
4783 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4784 if (IS_ERR(u.xcrs)) {
4785 r = PTR_ERR(u.xcrs);
4789 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4792 case KVM_SET_TSC_KHZ: {
4796 user_tsc_khz = (u32)arg;
4798 if (kvm_has_tsc_control &&
4799 user_tsc_khz >= kvm_max_guest_tsc_khz)
4802 if (user_tsc_khz == 0)
4803 user_tsc_khz = tsc_khz;
4805 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4810 case KVM_GET_TSC_KHZ: {
4811 r = vcpu->arch.virtual_tsc_khz;
4814 case KVM_KVMCLOCK_CTRL: {
4815 r = kvm_set_guest_paused(vcpu);
4818 case KVM_ENABLE_CAP: {
4819 struct kvm_enable_cap cap;
4822 if (copy_from_user(&cap, argp, sizeof(cap)))
4824 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4827 case KVM_GET_NESTED_STATE: {
4828 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4832 if (!kvm_x86_ops.nested_ops->get_state)
4835 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4837 if (get_user(user_data_size, &user_kvm_nested_state->size))
4840 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4845 if (r > user_data_size) {
4846 if (put_user(r, &user_kvm_nested_state->size))
4856 case KVM_SET_NESTED_STATE: {
4857 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4858 struct kvm_nested_state kvm_state;
4862 if (!kvm_x86_ops.nested_ops->set_state)
4866 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4870 if (kvm_state.size < sizeof(kvm_state))
4873 if (kvm_state.flags &
4874 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4875 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
4876 | KVM_STATE_NESTED_GIF_SET))
4879 /* nested_run_pending implies guest_mode. */
4880 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4881 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4884 idx = srcu_read_lock(&vcpu->kvm->srcu);
4885 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
4886 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4889 case KVM_GET_SUPPORTED_HV_CPUID: {
4890 struct kvm_cpuid2 __user *cpuid_arg = argp;
4891 struct kvm_cpuid2 cpuid;
4894 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4897 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4898 cpuid_arg->entries);
4903 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4918 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4920 return VM_FAULT_SIGBUS;
4923 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4927 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4929 ret = kvm_x86_ops.set_tss_addr(kvm, addr);
4933 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4936 return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
4939 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4940 unsigned long kvm_nr_mmu_pages)
4942 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4945 mutex_lock(&kvm->slots_lock);
4947 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4948 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4950 mutex_unlock(&kvm->slots_lock);
4954 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4956 return kvm->arch.n_max_mmu_pages;
4959 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4961 struct kvm_pic *pic = kvm->arch.vpic;
4965 switch (chip->chip_id) {
4966 case KVM_IRQCHIP_PIC_MASTER:
4967 memcpy(&chip->chip.pic, &pic->pics[0],
4968 sizeof(struct kvm_pic_state));
4970 case KVM_IRQCHIP_PIC_SLAVE:
4971 memcpy(&chip->chip.pic, &pic->pics[1],
4972 sizeof(struct kvm_pic_state));
4974 case KVM_IRQCHIP_IOAPIC:
4975 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4984 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4986 struct kvm_pic *pic = kvm->arch.vpic;
4990 switch (chip->chip_id) {
4991 case KVM_IRQCHIP_PIC_MASTER:
4992 spin_lock(&pic->lock);
4993 memcpy(&pic->pics[0], &chip->chip.pic,
4994 sizeof(struct kvm_pic_state));
4995 spin_unlock(&pic->lock);
4997 case KVM_IRQCHIP_PIC_SLAVE:
4998 spin_lock(&pic->lock);
4999 memcpy(&pic->pics[1], &chip->chip.pic,
5000 sizeof(struct kvm_pic_state));
5001 spin_unlock(&pic->lock);
5003 case KVM_IRQCHIP_IOAPIC:
5004 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5010 kvm_pic_update_irq(pic);
5014 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5016 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5018 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5020 mutex_lock(&kps->lock);
5021 memcpy(ps, &kps->channels, sizeof(*ps));
5022 mutex_unlock(&kps->lock);
5026 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5029 struct kvm_pit *pit = kvm->arch.vpit;
5031 mutex_lock(&pit->pit_state.lock);
5032 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5033 for (i = 0; i < 3; i++)
5034 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5035 mutex_unlock(&pit->pit_state.lock);
5039 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5041 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5042 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5043 sizeof(ps->channels));
5044 ps->flags = kvm->arch.vpit->pit_state.flags;
5045 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5046 memset(&ps->reserved, 0, sizeof(ps->reserved));
5050 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5054 u32 prev_legacy, cur_legacy;
5055 struct kvm_pit *pit = kvm->arch.vpit;
5057 mutex_lock(&pit->pit_state.lock);
5058 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5059 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5060 if (!prev_legacy && cur_legacy)
5062 memcpy(&pit->pit_state.channels, &ps->channels,
5063 sizeof(pit->pit_state.channels));
5064 pit->pit_state.flags = ps->flags;
5065 for (i = 0; i < 3; i++)
5066 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5068 mutex_unlock(&pit->pit_state.lock);
5072 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5073 struct kvm_reinject_control *control)
5075 struct kvm_pit *pit = kvm->arch.vpit;
5077 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5078 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5079 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5081 mutex_lock(&pit->pit_state.lock);
5082 kvm_pit_set_reinject(pit, control->pit_reinject);
5083 mutex_unlock(&pit->pit_state.lock);
5088 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5091 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
5093 if (kvm_x86_ops.flush_log_dirty)
5094 kvm_x86_ops.flush_log_dirty(kvm);
5097 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5100 if (!irqchip_in_kernel(kvm))
5103 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5104 irq_event->irq, irq_event->level,
5109 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5110 struct kvm_enable_cap *cap)
5118 case KVM_CAP_DISABLE_QUIRKS:
5119 kvm->arch.disabled_quirks = cap->args[0];
5122 case KVM_CAP_SPLIT_IRQCHIP: {
5123 mutex_lock(&kvm->lock);
5125 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5126 goto split_irqchip_unlock;
5128 if (irqchip_in_kernel(kvm))
5129 goto split_irqchip_unlock;
5130 if (kvm->created_vcpus)
5131 goto split_irqchip_unlock;
5132 r = kvm_setup_empty_irq_routing(kvm);
5134 goto split_irqchip_unlock;
5135 /* Pairs with irqchip_in_kernel. */
5137 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5138 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5140 split_irqchip_unlock:
5141 mutex_unlock(&kvm->lock);
5144 case KVM_CAP_X2APIC_API:
5146 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5149 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5150 kvm->arch.x2apic_format = true;
5151 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5152 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5156 case KVM_CAP_X86_DISABLE_EXITS:
5158 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5161 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5162 kvm_can_mwait_in_guest())
5163 kvm->arch.mwait_in_guest = true;
5164 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5165 kvm->arch.hlt_in_guest = true;
5166 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5167 kvm->arch.pause_in_guest = true;
5168 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5169 kvm->arch.cstate_in_guest = true;
5172 case KVM_CAP_MSR_PLATFORM_INFO:
5173 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5176 case KVM_CAP_EXCEPTION_PAYLOAD:
5177 kvm->arch.exception_payload_enabled = cap->args[0];
5180 case KVM_CAP_X86_USER_SPACE_MSR:
5181 kvm->arch.user_space_msr_mask = cap->args[0];
5191 static void kvm_clear_msr_filter(struct kvm *kvm)
5194 u32 count = kvm->arch.msr_filter.count;
5195 struct msr_bitmap_range ranges[16];
5197 mutex_lock(&kvm->lock);
5198 kvm->arch.msr_filter.count = 0;
5199 memcpy(ranges, kvm->arch.msr_filter.ranges, count * sizeof(ranges[0]));
5200 mutex_unlock(&kvm->lock);
5201 synchronize_srcu(&kvm->srcu);
5203 for (i = 0; i < count; i++)
5204 kfree(ranges[i].bitmap);
5207 static int kvm_add_msr_filter(struct kvm *kvm, struct kvm_msr_filter_range *user_range)
5209 struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
5210 struct msr_bitmap_range range;
5211 unsigned long *bitmap = NULL;
5215 if (!user_range->nmsrs)
5218 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5219 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5222 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5224 return PTR_ERR(bitmap);
5226 range = (struct msr_bitmap_range) {
5227 .flags = user_range->flags,
5228 .base = user_range->base,
5229 .nmsrs = user_range->nmsrs,
5233 if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) {
5243 /* Everything ok, add this range identifier to our global pool */
5244 ranges[kvm->arch.msr_filter.count] = range;
5245 /* Make sure we filled the array before we tell anyone to walk it */
5247 kvm->arch.msr_filter.count++;
5255 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5257 struct kvm_msr_filter __user *user_msr_filter = argp;
5258 struct kvm_msr_filter filter;
5264 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5267 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5268 empty &= !filter.ranges[i].nmsrs;
5270 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5271 if (empty && !default_allow)
5274 kvm_clear_msr_filter(kvm);
5276 kvm->arch.msr_filter.default_allow = default_allow;
5279 * Protect from concurrent calls to this function that could trigger
5280 * a TOCTOU violation on kvm->arch.msr_filter.count.
5282 mutex_lock(&kvm->lock);
5283 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5284 r = kvm_add_msr_filter(kvm, &filter.ranges[i]);
5289 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5290 mutex_unlock(&kvm->lock);
5295 long kvm_arch_vm_ioctl(struct file *filp,
5296 unsigned int ioctl, unsigned long arg)
5298 struct kvm *kvm = filp->private_data;
5299 void __user *argp = (void __user *)arg;
5302 * This union makes it completely explicit to gcc-3.x
5303 * that these two variables' stack usage should be
5304 * combined, not added together.
5307 struct kvm_pit_state ps;
5308 struct kvm_pit_state2 ps2;
5309 struct kvm_pit_config pit_config;
5313 case KVM_SET_TSS_ADDR:
5314 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5316 case KVM_SET_IDENTITY_MAP_ADDR: {
5319 mutex_lock(&kvm->lock);
5321 if (kvm->created_vcpus)
5322 goto set_identity_unlock;
5324 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5325 goto set_identity_unlock;
5326 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5327 set_identity_unlock:
5328 mutex_unlock(&kvm->lock);
5331 case KVM_SET_NR_MMU_PAGES:
5332 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5334 case KVM_GET_NR_MMU_PAGES:
5335 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5337 case KVM_CREATE_IRQCHIP: {
5338 mutex_lock(&kvm->lock);
5341 if (irqchip_in_kernel(kvm))
5342 goto create_irqchip_unlock;
5345 if (kvm->created_vcpus)
5346 goto create_irqchip_unlock;
5348 r = kvm_pic_init(kvm);
5350 goto create_irqchip_unlock;
5352 r = kvm_ioapic_init(kvm);
5354 kvm_pic_destroy(kvm);
5355 goto create_irqchip_unlock;
5358 r = kvm_setup_default_irq_routing(kvm);
5360 kvm_ioapic_destroy(kvm);
5361 kvm_pic_destroy(kvm);
5362 goto create_irqchip_unlock;
5364 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5366 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5367 create_irqchip_unlock:
5368 mutex_unlock(&kvm->lock);
5371 case KVM_CREATE_PIT:
5372 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5374 case KVM_CREATE_PIT2:
5376 if (copy_from_user(&u.pit_config, argp,
5377 sizeof(struct kvm_pit_config)))
5380 mutex_lock(&kvm->lock);
5383 goto create_pit_unlock;
5385 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5389 mutex_unlock(&kvm->lock);
5391 case KVM_GET_IRQCHIP: {
5392 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5393 struct kvm_irqchip *chip;
5395 chip = memdup_user(argp, sizeof(*chip));
5402 if (!irqchip_kernel(kvm))
5403 goto get_irqchip_out;
5404 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5406 goto get_irqchip_out;
5408 if (copy_to_user(argp, chip, sizeof(*chip)))
5409 goto get_irqchip_out;
5415 case KVM_SET_IRQCHIP: {
5416 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5417 struct kvm_irqchip *chip;
5419 chip = memdup_user(argp, sizeof(*chip));
5426 if (!irqchip_kernel(kvm))
5427 goto set_irqchip_out;
5428 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5435 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5438 if (!kvm->arch.vpit)
5440 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5444 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5451 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5453 mutex_lock(&kvm->lock);
5455 if (!kvm->arch.vpit)
5457 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5459 mutex_unlock(&kvm->lock);
5462 case KVM_GET_PIT2: {
5464 if (!kvm->arch.vpit)
5466 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5470 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5475 case KVM_SET_PIT2: {
5477 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5479 mutex_lock(&kvm->lock);
5481 if (!kvm->arch.vpit)
5483 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5485 mutex_unlock(&kvm->lock);
5488 case KVM_REINJECT_CONTROL: {
5489 struct kvm_reinject_control control;
5491 if (copy_from_user(&control, argp, sizeof(control)))
5494 if (!kvm->arch.vpit)
5496 r = kvm_vm_ioctl_reinject(kvm, &control);
5499 case KVM_SET_BOOT_CPU_ID:
5501 mutex_lock(&kvm->lock);
5502 if (kvm->created_vcpus)
5505 kvm->arch.bsp_vcpu_id = arg;
5506 mutex_unlock(&kvm->lock);
5508 case KVM_XEN_HVM_CONFIG: {
5509 struct kvm_xen_hvm_config xhc;
5511 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5516 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5520 case KVM_SET_CLOCK: {
5521 struct kvm_clock_data user_ns;
5525 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5534 * TODO: userspace has to take care of races with VCPU_RUN, so
5535 * kvm_gen_update_masterclock() can be cut down to locked
5536 * pvclock_update_vm_gtod_copy().
5538 kvm_gen_update_masterclock(kvm);
5539 now_ns = get_kvmclock_ns(kvm);
5540 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5541 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5544 case KVM_GET_CLOCK: {
5545 struct kvm_clock_data user_ns;
5548 now_ns = get_kvmclock_ns(kvm);
5549 user_ns.clock = now_ns;
5550 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5551 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5554 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5559 case KVM_MEMORY_ENCRYPT_OP: {
5561 if (kvm_x86_ops.mem_enc_op)
5562 r = kvm_x86_ops.mem_enc_op(kvm, argp);
5565 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5566 struct kvm_enc_region region;
5569 if (copy_from_user(®ion, argp, sizeof(region)))
5573 if (kvm_x86_ops.mem_enc_reg_region)
5574 r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion);
5577 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5578 struct kvm_enc_region region;
5581 if (copy_from_user(®ion, argp, sizeof(region)))
5585 if (kvm_x86_ops.mem_enc_unreg_region)
5586 r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion);
5589 case KVM_HYPERV_EVENTFD: {
5590 struct kvm_hyperv_eventfd hvevfd;
5593 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5595 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5598 case KVM_SET_PMU_EVENT_FILTER:
5599 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5601 case KVM_X86_SET_MSR_FILTER:
5602 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5611 static void kvm_init_msr_list(void)
5613 struct x86_pmu_capability x86_pmu;
5617 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5618 "Please update the fixed PMCs in msrs_to_saved_all[]");
5620 perf_get_x86_pmu_capability(&x86_pmu);
5622 num_msrs_to_save = 0;
5623 num_emulated_msrs = 0;
5624 num_msr_based_features = 0;
5626 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5627 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5631 * Even MSRs that are valid in the host may not be exposed
5632 * to the guests in some cases.
5634 switch (msrs_to_save_all[i]) {
5635 case MSR_IA32_BNDCFGS:
5636 if (!kvm_mpx_supported())
5640 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5643 case MSR_IA32_UMWAIT_CONTROL:
5644 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5647 case MSR_IA32_RTIT_CTL:
5648 case MSR_IA32_RTIT_STATUS:
5649 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5652 case MSR_IA32_RTIT_CR3_MATCH:
5653 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5654 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5657 case MSR_IA32_RTIT_OUTPUT_BASE:
5658 case MSR_IA32_RTIT_OUTPUT_MASK:
5659 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5660 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5661 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5664 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5665 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5666 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5667 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5670 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5671 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5672 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5675 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5676 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5677 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5684 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5687 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5688 if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5691 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5694 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5695 struct kvm_msr_entry msr;
5697 msr.index = msr_based_features_all[i];
5698 if (kvm_get_msr_feature(&msr))
5701 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5705 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5713 if (!(lapic_in_kernel(vcpu) &&
5714 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5715 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5726 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5733 if (!(lapic_in_kernel(vcpu) &&
5734 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5736 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5738 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5748 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5749 struct kvm_segment *var, int seg)
5751 kvm_x86_ops.set_segment(vcpu, var, seg);
5754 void kvm_get_segment(struct kvm_vcpu *vcpu,
5755 struct kvm_segment *var, int seg)
5757 kvm_x86_ops.get_segment(vcpu, var, seg);
5760 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5761 struct x86_exception *exception)
5765 BUG_ON(!mmu_is_nested(vcpu));
5767 /* NPT walks are always user-walks */
5768 access |= PFERR_USER_MASK;
5769 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5774 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5775 struct x86_exception *exception)
5777 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5778 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5781 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5782 struct x86_exception *exception)
5784 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5785 access |= PFERR_FETCH_MASK;
5786 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5789 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5790 struct x86_exception *exception)
5792 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5793 access |= PFERR_WRITE_MASK;
5794 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5797 /* uses this to access any guest's mapped memory without checking CPL */
5798 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5799 struct x86_exception *exception)
5801 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5804 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5805 struct kvm_vcpu *vcpu, u32 access,
5806 struct x86_exception *exception)
5809 int r = X86EMUL_CONTINUE;
5812 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5814 unsigned offset = addr & (PAGE_SIZE-1);
5815 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5818 if (gpa == UNMAPPED_GVA)
5819 return X86EMUL_PROPAGATE_FAULT;
5820 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5823 r = X86EMUL_IO_NEEDED;
5835 /* used for instruction fetching */
5836 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5837 gva_t addr, void *val, unsigned int bytes,
5838 struct x86_exception *exception)
5840 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5841 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5845 /* Inline kvm_read_guest_virt_helper for speed. */
5846 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5848 if (unlikely(gpa == UNMAPPED_GVA))
5849 return X86EMUL_PROPAGATE_FAULT;
5851 offset = addr & (PAGE_SIZE-1);
5852 if (WARN_ON(offset + bytes > PAGE_SIZE))
5853 bytes = (unsigned)PAGE_SIZE - offset;
5854 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5856 if (unlikely(ret < 0))
5857 return X86EMUL_IO_NEEDED;
5859 return X86EMUL_CONTINUE;
5862 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5863 gva_t addr, void *val, unsigned int bytes,
5864 struct x86_exception *exception)
5866 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5869 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5870 * is returned, but our callers are not ready for that and they blindly
5871 * call kvm_inject_page_fault. Ensure that they at least do not leak
5872 * uninitialized kernel stack memory into cr2 and error code.
5874 memset(exception, 0, sizeof(*exception));
5875 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5878 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5880 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5881 gva_t addr, void *val, unsigned int bytes,
5882 struct x86_exception *exception, bool system)
5884 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5887 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5888 access |= PFERR_USER_MASK;
5890 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5893 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5894 unsigned long addr, void *val, unsigned int bytes)
5896 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5897 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5899 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5902 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5903 struct kvm_vcpu *vcpu, u32 access,
5904 struct x86_exception *exception)
5907 int r = X86EMUL_CONTINUE;
5910 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5913 unsigned offset = addr & (PAGE_SIZE-1);
5914 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5917 if (gpa == UNMAPPED_GVA)
5918 return X86EMUL_PROPAGATE_FAULT;
5919 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5921 r = X86EMUL_IO_NEEDED;
5933 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5934 unsigned int bytes, struct x86_exception *exception,
5937 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5938 u32 access = PFERR_WRITE_MASK;
5940 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5941 access |= PFERR_USER_MASK;
5943 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5947 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5948 unsigned int bytes, struct x86_exception *exception)
5950 /* kvm_write_guest_virt_system can pull in tons of pages. */
5951 vcpu->arch.l1tf_flush_l1d = true;
5953 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5954 PFERR_WRITE_MASK, exception);
5956 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5958 int handle_ud(struct kvm_vcpu *vcpu)
5960 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5961 int emul_type = EMULTYPE_TRAP_UD;
5962 char sig[5]; /* ud2; .ascii "kvm" */
5963 struct x86_exception e;
5965 if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, NULL, 0)))
5968 if (force_emulation_prefix &&
5969 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5970 sig, sizeof(sig), &e) == 0 &&
5971 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5972 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5973 emul_type = EMULTYPE_TRAP_UD_FORCED;
5976 return kvm_emulate_instruction(vcpu, emul_type);
5978 EXPORT_SYMBOL_GPL(handle_ud);
5980 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5981 gpa_t gpa, bool write)
5983 /* For APIC access vmexit */
5984 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5987 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5988 trace_vcpu_match_mmio(gva, gpa, write, true);
5995 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5996 gpa_t *gpa, struct x86_exception *exception,
5999 u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
6000 | (write ? PFERR_WRITE_MASK : 0);
6003 * currently PKRU is only applied to ept enabled guest so
6004 * there is no pkey in EPT page table for L1 guest or EPT
6005 * shadow page table for L2 guest.
6007 if (vcpu_match_mmio_gva(vcpu, gva)
6008 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6009 vcpu->arch.mmio_access, 0, access)) {
6010 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6011 (gva & (PAGE_SIZE - 1));
6012 trace_vcpu_match_mmio(gva, *gpa, write, false);
6016 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6018 if (*gpa == UNMAPPED_GVA)
6021 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6024 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6025 const void *val, int bytes)
6029 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6032 kvm_page_track_write(vcpu, gpa, val, bytes);
6036 struct read_write_emulator_ops {
6037 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6039 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6040 void *val, int bytes);
6041 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6042 int bytes, void *val);
6043 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6044 void *val, int bytes);
6048 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6050 if (vcpu->mmio_read_completed) {
6051 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6052 vcpu->mmio_fragments[0].gpa, val);
6053 vcpu->mmio_read_completed = 0;
6060 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6061 void *val, int bytes)
6063 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6066 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6067 void *val, int bytes)
6069 return emulator_write_phys(vcpu, gpa, val, bytes);
6072 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6074 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6075 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6078 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6079 void *val, int bytes)
6081 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6082 return X86EMUL_IO_NEEDED;
6085 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6086 void *val, int bytes)
6088 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6090 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6091 return X86EMUL_CONTINUE;
6094 static const struct read_write_emulator_ops read_emultor = {
6095 .read_write_prepare = read_prepare,
6096 .read_write_emulate = read_emulate,
6097 .read_write_mmio = vcpu_mmio_read,
6098 .read_write_exit_mmio = read_exit_mmio,
6101 static const struct read_write_emulator_ops write_emultor = {
6102 .read_write_emulate = write_emulate,
6103 .read_write_mmio = write_mmio,
6104 .read_write_exit_mmio = write_exit_mmio,
6108 static int emulator_read_write_onepage(unsigned long addr, void *val,
6110 struct x86_exception *exception,
6111 struct kvm_vcpu *vcpu,
6112 const struct read_write_emulator_ops *ops)
6116 bool write = ops->write;
6117 struct kvm_mmio_fragment *frag;
6118 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6121 * If the exit was due to a NPF we may already have a GPA.
6122 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6123 * Note, this cannot be used on string operations since string
6124 * operation using rep will only have the initial GPA from the NPF
6127 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6128 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6129 gpa = ctxt->gpa_val;
6130 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6132 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6134 return X86EMUL_PROPAGATE_FAULT;
6137 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6138 return X86EMUL_CONTINUE;
6141 * Is this MMIO handled locally?
6143 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6144 if (handled == bytes)
6145 return X86EMUL_CONTINUE;
6151 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6152 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6156 return X86EMUL_CONTINUE;
6159 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6161 void *val, unsigned int bytes,
6162 struct x86_exception *exception,
6163 const struct read_write_emulator_ops *ops)
6165 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6169 if (ops->read_write_prepare &&
6170 ops->read_write_prepare(vcpu, val, bytes))
6171 return X86EMUL_CONTINUE;
6173 vcpu->mmio_nr_fragments = 0;
6175 /* Crossing a page boundary? */
6176 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6179 now = -addr & ~PAGE_MASK;
6180 rc = emulator_read_write_onepage(addr, val, now, exception,
6183 if (rc != X86EMUL_CONTINUE)
6186 if (ctxt->mode != X86EMUL_MODE_PROT64)
6192 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6194 if (rc != X86EMUL_CONTINUE)
6197 if (!vcpu->mmio_nr_fragments)
6200 gpa = vcpu->mmio_fragments[0].gpa;
6202 vcpu->mmio_needed = 1;
6203 vcpu->mmio_cur_fragment = 0;
6205 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6206 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6207 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6208 vcpu->run->mmio.phys_addr = gpa;
6210 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6213 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6217 struct x86_exception *exception)
6219 return emulator_read_write(ctxt, addr, val, bytes,
6220 exception, &read_emultor);
6223 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6227 struct x86_exception *exception)
6229 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6230 exception, &write_emultor);
6233 #define CMPXCHG_TYPE(t, ptr, old, new) \
6234 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6236 #ifdef CONFIG_X86_64
6237 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6239 # define CMPXCHG64(ptr, old, new) \
6240 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6243 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6248 struct x86_exception *exception)
6250 struct kvm_host_map map;
6251 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6257 /* guests cmpxchg8b have to be emulated atomically */
6258 if (bytes > 8 || (bytes & (bytes - 1)))
6261 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6263 if (gpa == UNMAPPED_GVA ||
6264 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6268 * Emulate the atomic as a straight write to avoid #AC if SLD is
6269 * enabled in the host and the access splits a cache line.
6271 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6272 page_line_mask = ~(cache_line_size() - 1);
6274 page_line_mask = PAGE_MASK;
6276 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6279 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6282 kaddr = map.hva + offset_in_page(gpa);
6286 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6289 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6292 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6295 exchanged = CMPXCHG64(kaddr, old, new);
6301 kvm_vcpu_unmap(vcpu, &map, true);
6304 return X86EMUL_CMPXCHG_FAILED;
6306 kvm_page_track_write(vcpu, gpa, new, bytes);
6308 return X86EMUL_CONTINUE;
6311 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6313 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6316 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6320 for (i = 0; i < vcpu->arch.pio.count; i++) {
6321 if (vcpu->arch.pio.in)
6322 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6323 vcpu->arch.pio.size, pd);
6325 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6326 vcpu->arch.pio.port, vcpu->arch.pio.size,
6330 pd += vcpu->arch.pio.size;
6335 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6336 unsigned short port, void *val,
6337 unsigned int count, bool in)
6339 vcpu->arch.pio.port = port;
6340 vcpu->arch.pio.in = in;
6341 vcpu->arch.pio.count = count;
6342 vcpu->arch.pio.size = size;
6344 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6345 vcpu->arch.pio.count = 0;
6349 vcpu->run->exit_reason = KVM_EXIT_IO;
6350 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6351 vcpu->run->io.size = size;
6352 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6353 vcpu->run->io.count = count;
6354 vcpu->run->io.port = port;
6359 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6360 unsigned short port, void *val, unsigned int count)
6364 if (vcpu->arch.pio.count)
6367 memset(vcpu->arch.pio_data, 0, size * count);
6369 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6372 memcpy(val, vcpu->arch.pio_data, size * count);
6373 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6374 vcpu->arch.pio.count = 0;
6381 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6382 int size, unsigned short port, void *val,
6385 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6389 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6390 unsigned short port, const void *val,
6393 memcpy(vcpu->arch.pio_data, val, size * count);
6394 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6395 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6398 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6399 int size, unsigned short port,
6400 const void *val, unsigned int count)
6402 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6405 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6407 return kvm_x86_ops.get_segment_base(vcpu, seg);
6410 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6412 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6415 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6417 if (!need_emulate_wbinvd(vcpu))
6418 return X86EMUL_CONTINUE;
6420 if (kvm_x86_ops.has_wbinvd_exit()) {
6421 int cpu = get_cpu();
6423 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6424 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6425 wbinvd_ipi, NULL, 1);
6427 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6430 return X86EMUL_CONTINUE;
6433 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6435 kvm_emulate_wbinvd_noskip(vcpu);
6436 return kvm_skip_emulated_instruction(vcpu);
6438 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6442 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6444 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6447 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6448 unsigned long *dest)
6450 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6453 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6454 unsigned long value)
6457 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6460 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6462 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6465 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6467 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6468 unsigned long value;
6472 value = kvm_read_cr0(vcpu);
6475 value = vcpu->arch.cr2;
6478 value = kvm_read_cr3(vcpu);
6481 value = kvm_read_cr4(vcpu);
6484 value = kvm_get_cr8(vcpu);
6487 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6494 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6496 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6501 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6504 vcpu->arch.cr2 = val;
6507 res = kvm_set_cr3(vcpu, val);
6510 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6513 res = kvm_set_cr8(vcpu, val);
6516 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6523 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6525 return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6528 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6530 kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6533 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6535 kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6538 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6540 kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6543 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6545 kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6548 static unsigned long emulator_get_cached_segment_base(
6549 struct x86_emulate_ctxt *ctxt, int seg)
6551 return get_segment_base(emul_to_vcpu(ctxt), seg);
6554 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6555 struct desc_struct *desc, u32 *base3,
6558 struct kvm_segment var;
6560 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6561 *selector = var.selector;
6564 memset(desc, 0, sizeof(*desc));
6572 set_desc_limit(desc, var.limit);
6573 set_desc_base(desc, (unsigned long)var.base);
6574 #ifdef CONFIG_X86_64
6576 *base3 = var.base >> 32;
6578 desc->type = var.type;
6580 desc->dpl = var.dpl;
6581 desc->p = var.present;
6582 desc->avl = var.avl;
6590 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6591 struct desc_struct *desc, u32 base3,
6594 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6595 struct kvm_segment var;
6597 var.selector = selector;
6598 var.base = get_desc_base(desc);
6599 #ifdef CONFIG_X86_64
6600 var.base |= ((u64)base3) << 32;
6602 var.limit = get_desc_limit(desc);
6604 var.limit = (var.limit << 12) | 0xfff;
6605 var.type = desc->type;
6606 var.dpl = desc->dpl;
6611 var.avl = desc->avl;
6612 var.present = desc->p;
6613 var.unusable = !var.present;
6616 kvm_set_segment(vcpu, &var, seg);
6620 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6621 u32 msr_index, u64 *pdata)
6623 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6626 r = kvm_get_msr(vcpu, msr_index, pdata);
6628 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6629 /* Bounce to user space */
6630 return X86EMUL_IO_NEEDED;
6636 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6637 u32 msr_index, u64 data)
6639 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6642 r = kvm_set_msr(vcpu, msr_index, data);
6644 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6645 /* Bounce to user space */
6646 return X86EMUL_IO_NEEDED;
6652 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6654 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6656 return vcpu->arch.smbase;
6659 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6661 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6663 vcpu->arch.smbase = smbase;
6666 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6669 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6672 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6673 u32 pmc, u64 *pdata)
6675 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6678 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6680 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6683 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6684 struct x86_instruction_info *info,
6685 enum x86_intercept_stage stage)
6687 return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6691 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6692 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6695 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6698 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6700 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6703 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6705 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6708 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6710 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6713 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6715 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6718 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6720 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6723 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6725 kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6728 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6730 return emul_to_vcpu(ctxt)->arch.hflags;
6733 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6735 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6738 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6739 const char *smstate)
6741 return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6744 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6746 kvm_smm_changed(emul_to_vcpu(ctxt));
6749 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6751 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6754 static const struct x86_emulate_ops emulate_ops = {
6755 .read_gpr = emulator_read_gpr,
6756 .write_gpr = emulator_write_gpr,
6757 .read_std = emulator_read_std,
6758 .write_std = emulator_write_std,
6759 .read_phys = kvm_read_guest_phys_system,
6760 .fetch = kvm_fetch_guest_virt,
6761 .read_emulated = emulator_read_emulated,
6762 .write_emulated = emulator_write_emulated,
6763 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6764 .invlpg = emulator_invlpg,
6765 .pio_in_emulated = emulator_pio_in_emulated,
6766 .pio_out_emulated = emulator_pio_out_emulated,
6767 .get_segment = emulator_get_segment,
6768 .set_segment = emulator_set_segment,
6769 .get_cached_segment_base = emulator_get_cached_segment_base,
6770 .get_gdt = emulator_get_gdt,
6771 .get_idt = emulator_get_idt,
6772 .set_gdt = emulator_set_gdt,
6773 .set_idt = emulator_set_idt,
6774 .get_cr = emulator_get_cr,
6775 .set_cr = emulator_set_cr,
6776 .cpl = emulator_get_cpl,
6777 .get_dr = emulator_get_dr,
6778 .set_dr = emulator_set_dr,
6779 .get_smbase = emulator_get_smbase,
6780 .set_smbase = emulator_set_smbase,
6781 .set_msr = emulator_set_msr,
6782 .get_msr = emulator_get_msr,
6783 .check_pmc = emulator_check_pmc,
6784 .read_pmc = emulator_read_pmc,
6785 .halt = emulator_halt,
6786 .wbinvd = emulator_wbinvd,
6787 .fix_hypercall = emulator_fix_hypercall,
6788 .intercept = emulator_intercept,
6789 .get_cpuid = emulator_get_cpuid,
6790 .guest_has_long_mode = emulator_guest_has_long_mode,
6791 .guest_has_movbe = emulator_guest_has_movbe,
6792 .guest_has_fxsr = emulator_guest_has_fxsr,
6793 .set_nmi_mask = emulator_set_nmi_mask,
6794 .get_hflags = emulator_get_hflags,
6795 .set_hflags = emulator_set_hflags,
6796 .pre_leave_smm = emulator_pre_leave_smm,
6797 .post_leave_smm = emulator_post_leave_smm,
6798 .set_xcr = emulator_set_xcr,
6801 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6803 u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6805 * an sti; sti; sequence only disable interrupts for the first
6806 * instruction. So, if the last instruction, be it emulated or
6807 * not, left the system with the INT_STI flag enabled, it
6808 * means that the last instruction is an sti. We should not
6809 * leave the flag on in this case. The same goes for mov ss
6811 if (int_shadow & mask)
6813 if (unlikely(int_shadow || mask)) {
6814 kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6816 kvm_make_request(KVM_REQ_EVENT, vcpu);
6820 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6822 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6823 if (ctxt->exception.vector == PF_VECTOR)
6824 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
6826 if (ctxt->exception.error_code_valid)
6827 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6828 ctxt->exception.error_code);
6830 kvm_queue_exception(vcpu, ctxt->exception.vector);
6834 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6836 struct x86_emulate_ctxt *ctxt;
6838 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6840 pr_err("kvm: failed to allocate vcpu's emulator\n");
6845 ctxt->ops = &emulate_ops;
6846 vcpu->arch.emulate_ctxt = ctxt;
6851 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6853 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6856 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6858 ctxt->gpa_available = false;
6859 ctxt->eflags = kvm_get_rflags(vcpu);
6860 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6862 ctxt->eip = kvm_rip_read(vcpu);
6863 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6864 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6865 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6866 cs_db ? X86EMUL_MODE_PROT32 :
6867 X86EMUL_MODE_PROT16;
6868 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6869 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6870 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6872 init_decode_cache(ctxt);
6873 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6876 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6878 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6881 init_emulate_ctxt(vcpu);
6885 ctxt->_eip = ctxt->eip + inc_eip;
6886 ret = emulate_int_real(ctxt, irq);
6888 if (ret != X86EMUL_CONTINUE) {
6889 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6891 ctxt->eip = ctxt->_eip;
6892 kvm_rip_write(vcpu, ctxt->eip);
6893 kvm_set_rflags(vcpu, ctxt->eflags);
6896 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6898 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6900 ++vcpu->stat.insn_emulation_fail;
6901 trace_kvm_emulate_insn_failed(vcpu);
6903 if (emulation_type & EMULTYPE_VMWARE_GP) {
6904 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6908 if (emulation_type & EMULTYPE_SKIP) {
6909 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6910 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6911 vcpu->run->internal.ndata = 0;
6915 kvm_queue_exception(vcpu, UD_VECTOR);
6917 if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
6918 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6919 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6920 vcpu->run->internal.ndata = 0;
6927 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6928 bool write_fault_to_shadow_pgtable,
6931 gpa_t gpa = cr2_or_gpa;
6934 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6937 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6938 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6941 if (!vcpu->arch.mmu->direct_map) {
6943 * Write permission should be allowed since only
6944 * write access need to be emulated.
6946 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6949 * If the mapping is invalid in guest, let cpu retry
6950 * it to generate fault.
6952 if (gpa == UNMAPPED_GVA)
6957 * Do not retry the unhandleable instruction if it faults on the
6958 * readonly host memory, otherwise it will goto a infinite loop:
6959 * retry instruction -> write #PF -> emulation fail -> retry
6960 * instruction -> ...
6962 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6965 * If the instruction failed on the error pfn, it can not be fixed,
6966 * report the error to userspace.
6968 if (is_error_noslot_pfn(pfn))
6971 kvm_release_pfn_clean(pfn);
6973 /* The instructions are well-emulated on direct mmu. */
6974 if (vcpu->arch.mmu->direct_map) {
6975 unsigned int indirect_shadow_pages;
6977 spin_lock(&vcpu->kvm->mmu_lock);
6978 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6979 spin_unlock(&vcpu->kvm->mmu_lock);
6981 if (indirect_shadow_pages)
6982 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6988 * if emulation was due to access to shadowed page table
6989 * and it failed try to unshadow page and re-enter the
6990 * guest to let CPU execute the instruction.
6992 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6995 * If the access faults on its page table, it can not
6996 * be fixed by unprotecting shadow page and it should
6997 * be reported to userspace.
6999 return !write_fault_to_shadow_pgtable;
7002 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7003 gpa_t cr2_or_gpa, int emulation_type)
7005 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7006 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7008 last_retry_eip = vcpu->arch.last_retry_eip;
7009 last_retry_addr = vcpu->arch.last_retry_addr;
7012 * If the emulation is caused by #PF and it is non-page_table
7013 * writing instruction, it means the VM-EXIT is caused by shadow
7014 * page protected, we can zap the shadow page and retry this
7015 * instruction directly.
7017 * Note: if the guest uses a non-page-table modifying instruction
7018 * on the PDE that points to the instruction, then we will unmap
7019 * the instruction and go to an infinite loop. So, we cache the
7020 * last retried eip and the last fault address, if we meet the eip
7021 * and the address again, we can break out of the potential infinite
7024 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7026 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7029 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7030 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7033 if (x86_page_table_writing_insn(ctxt))
7036 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7039 vcpu->arch.last_retry_eip = ctxt->eip;
7040 vcpu->arch.last_retry_addr = cr2_or_gpa;
7042 if (!vcpu->arch.mmu->direct_map)
7043 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7045 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7050 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7051 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7053 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7055 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7056 /* This is a good place to trace that we are exiting SMM. */
7057 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7059 /* Process a latched INIT or SMI, if any. */
7060 kvm_make_request(KVM_REQ_EVENT, vcpu);
7063 kvm_mmu_reset_context(vcpu);
7066 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7075 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7076 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7081 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7083 struct kvm_run *kvm_run = vcpu->run;
7085 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7086 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
7087 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7088 kvm_run->debug.arch.exception = DB_VECTOR;
7089 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7092 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7096 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7098 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7101 r = kvm_x86_ops.skip_emulated_instruction(vcpu);
7106 * rflags is the old, "raw" value of the flags. The new value has
7107 * not been saved yet.
7109 * This is correct even for TF set by the guest, because "the
7110 * processor will not generate this exception after the instruction
7111 * that sets the TF flag".
7113 if (unlikely(rflags & X86_EFLAGS_TF))
7114 r = kvm_vcpu_do_singlestep(vcpu);
7117 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7119 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7121 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7122 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7123 struct kvm_run *kvm_run = vcpu->run;
7124 unsigned long eip = kvm_get_linear_rip(vcpu);
7125 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7126 vcpu->arch.guest_debug_dr7,
7130 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
7131 kvm_run->debug.arch.pc = eip;
7132 kvm_run->debug.arch.exception = DB_VECTOR;
7133 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7139 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7140 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7141 unsigned long eip = kvm_get_linear_rip(vcpu);
7142 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7147 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7156 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7158 switch (ctxt->opcode_len) {
7165 case 0xe6: /* OUT */
7169 case 0x6c: /* INS */
7171 case 0x6e: /* OUTS */
7178 case 0x33: /* RDPMC */
7187 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7188 int emulation_type, void *insn, int insn_len)
7191 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7192 bool writeback = true;
7193 bool write_fault_to_spt;
7195 if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, insn, insn_len)))
7198 vcpu->arch.l1tf_flush_l1d = true;
7201 * Clear write_fault_to_shadow_pgtable here to ensure it is
7204 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7205 vcpu->arch.write_fault_to_shadow_pgtable = false;
7206 kvm_clear_exception_queue(vcpu);
7208 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7209 init_emulate_ctxt(vcpu);
7212 * We will reenter on the same instruction since
7213 * we do not set complete_userspace_io. This does not
7214 * handle watchpoints yet, those would be handled in
7217 if (!(emulation_type & EMULTYPE_SKIP) &&
7218 kvm_vcpu_check_breakpoint(vcpu, &r))
7221 ctxt->interruptibility = 0;
7222 ctxt->have_exception = false;
7223 ctxt->exception.vector = -1;
7224 ctxt->perm_ok = false;
7226 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7228 r = x86_decode_insn(ctxt, insn, insn_len);
7230 trace_kvm_emulate_insn_start(vcpu);
7231 ++vcpu->stat.insn_emulation;
7232 if (r != EMULATION_OK) {
7233 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7234 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7235 kvm_queue_exception(vcpu, UD_VECTOR);
7238 if (reexecute_instruction(vcpu, cr2_or_gpa,
7242 if (ctxt->have_exception) {
7244 * #UD should result in just EMULATION_FAILED, and trap-like
7245 * exception should not be encountered during decode.
7247 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7248 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7249 inject_emulated_exception(vcpu);
7252 return handle_emulation_failure(vcpu, emulation_type);
7256 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7257 !is_vmware_backdoor_opcode(ctxt)) {
7258 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7263 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7264 * for kvm_skip_emulated_instruction(). The caller is responsible for
7265 * updating interruptibility state and injecting single-step #DBs.
7267 if (emulation_type & EMULTYPE_SKIP) {
7268 kvm_rip_write(vcpu, ctxt->_eip);
7269 if (ctxt->eflags & X86_EFLAGS_RF)
7270 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7274 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7277 /* this is needed for vmware backdoor interface to work since it
7278 changes registers values during IO operation */
7279 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7280 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7281 emulator_invalidate_register_cache(ctxt);
7285 if (emulation_type & EMULTYPE_PF) {
7286 /* Save the faulting GPA (cr2) in the address field */
7287 ctxt->exception.address = cr2_or_gpa;
7289 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7290 if (vcpu->arch.mmu->direct_map) {
7291 ctxt->gpa_available = true;
7292 ctxt->gpa_val = cr2_or_gpa;
7295 /* Sanitize the address out of an abundance of paranoia. */
7296 ctxt->exception.address = 0;
7299 r = x86_emulate_insn(ctxt);
7301 if (r == EMULATION_INTERCEPTED)
7304 if (r == EMULATION_FAILED) {
7305 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7309 return handle_emulation_failure(vcpu, emulation_type);
7312 if (ctxt->have_exception) {
7314 if (inject_emulated_exception(vcpu))
7316 } else if (vcpu->arch.pio.count) {
7317 if (!vcpu->arch.pio.in) {
7318 /* FIXME: return into emulator if single-stepping. */
7319 vcpu->arch.pio.count = 0;
7322 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7325 } else if (vcpu->mmio_needed) {
7326 ++vcpu->stat.mmio_exits;
7328 if (!vcpu->mmio_is_write)
7331 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7332 } else if (r == EMULATION_RESTART)
7338 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7339 toggle_interruptibility(vcpu, ctxt->interruptibility);
7340 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7341 if (!ctxt->have_exception ||
7342 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7343 kvm_rip_write(vcpu, ctxt->eip);
7344 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7345 r = kvm_vcpu_do_singlestep(vcpu);
7346 if (kvm_x86_ops.update_emulated_instruction)
7347 kvm_x86_ops.update_emulated_instruction(vcpu);
7348 __kvm_set_rflags(vcpu, ctxt->eflags);
7352 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7353 * do nothing, and it will be requested again as soon as
7354 * the shadow expires. But we still need to check here,
7355 * because POPF has no interrupt shadow.
7357 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7358 kvm_make_request(KVM_REQ_EVENT, vcpu);
7360 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7365 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7367 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7369 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7371 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7372 void *insn, int insn_len)
7374 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7376 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7378 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7380 vcpu->arch.pio.count = 0;
7384 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7386 vcpu->arch.pio.count = 0;
7388 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7391 return kvm_skip_emulated_instruction(vcpu);
7394 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7395 unsigned short port)
7397 unsigned long val = kvm_rax_read(vcpu);
7398 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7404 * Workaround userspace that relies on old KVM behavior of %rip being
7405 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7408 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7409 vcpu->arch.complete_userspace_io =
7410 complete_fast_pio_out_port_0x7e;
7411 kvm_skip_emulated_instruction(vcpu);
7413 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7414 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7419 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7423 /* We should only ever be called with arch.pio.count equal to 1 */
7424 BUG_ON(vcpu->arch.pio.count != 1);
7426 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7427 vcpu->arch.pio.count = 0;
7431 /* For size less than 4 we merge, else we zero extend */
7432 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7435 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7436 * the copy and tracing
7438 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7439 kvm_rax_write(vcpu, val);
7441 return kvm_skip_emulated_instruction(vcpu);
7444 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7445 unsigned short port)
7450 /* For size less than 4 we merge, else we zero extend */
7451 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7453 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7455 kvm_rax_write(vcpu, val);
7459 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7460 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7465 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7470 ret = kvm_fast_pio_in(vcpu, size, port);
7472 ret = kvm_fast_pio_out(vcpu, size, port);
7473 return ret && kvm_skip_emulated_instruction(vcpu);
7475 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7477 static int kvmclock_cpu_down_prep(unsigned int cpu)
7479 __this_cpu_write(cpu_tsc_khz, 0);
7483 static void tsc_khz_changed(void *data)
7485 struct cpufreq_freqs *freq = data;
7486 unsigned long khz = 0;
7490 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7491 khz = cpufreq_quick_get(raw_smp_processor_id());
7494 __this_cpu_write(cpu_tsc_khz, khz);
7497 #ifdef CONFIG_X86_64
7498 static void kvm_hyperv_tsc_notifier(void)
7501 struct kvm_vcpu *vcpu;
7504 mutex_lock(&kvm_lock);
7505 list_for_each_entry(kvm, &vm_list, vm_list)
7506 kvm_make_mclock_inprogress_request(kvm);
7508 hyperv_stop_tsc_emulation();
7510 /* TSC frequency always matches when on Hyper-V */
7511 for_each_present_cpu(cpu)
7512 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7513 kvm_max_guest_tsc_khz = tsc_khz;
7515 list_for_each_entry(kvm, &vm_list, vm_list) {
7516 struct kvm_arch *ka = &kvm->arch;
7518 spin_lock(&ka->pvclock_gtod_sync_lock);
7520 pvclock_update_vm_gtod_copy(kvm);
7522 kvm_for_each_vcpu(cpu, vcpu, kvm)
7523 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7525 kvm_for_each_vcpu(cpu, vcpu, kvm)
7526 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7528 spin_unlock(&ka->pvclock_gtod_sync_lock);
7530 mutex_unlock(&kvm_lock);
7534 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7537 struct kvm_vcpu *vcpu;
7538 int i, send_ipi = 0;
7541 * We allow guests to temporarily run on slowing clocks,
7542 * provided we notify them after, or to run on accelerating
7543 * clocks, provided we notify them before. Thus time never
7546 * However, we have a problem. We can't atomically update
7547 * the frequency of a given CPU from this function; it is
7548 * merely a notifier, which can be called from any CPU.
7549 * Changing the TSC frequency at arbitrary points in time
7550 * requires a recomputation of local variables related to
7551 * the TSC for each VCPU. We must flag these local variables
7552 * to be updated and be sure the update takes place with the
7553 * new frequency before any guests proceed.
7555 * Unfortunately, the combination of hotplug CPU and frequency
7556 * change creates an intractable locking scenario; the order
7557 * of when these callouts happen is undefined with respect to
7558 * CPU hotplug, and they can race with each other. As such,
7559 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7560 * undefined; you can actually have a CPU frequency change take
7561 * place in between the computation of X and the setting of the
7562 * variable. To protect against this problem, all updates of
7563 * the per_cpu tsc_khz variable are done in an interrupt
7564 * protected IPI, and all callers wishing to update the value
7565 * must wait for a synchronous IPI to complete (which is trivial
7566 * if the caller is on the CPU already). This establishes the
7567 * necessary total order on variable updates.
7569 * Note that because a guest time update may take place
7570 * anytime after the setting of the VCPU's request bit, the
7571 * correct TSC value must be set before the request. However,
7572 * to ensure the update actually makes it to any guest which
7573 * starts running in hardware virtualization between the set
7574 * and the acquisition of the spinlock, we must also ping the
7575 * CPU after setting the request bit.
7579 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7581 mutex_lock(&kvm_lock);
7582 list_for_each_entry(kvm, &vm_list, vm_list) {
7583 kvm_for_each_vcpu(i, vcpu, kvm) {
7584 if (vcpu->cpu != cpu)
7586 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7587 if (vcpu->cpu != raw_smp_processor_id())
7591 mutex_unlock(&kvm_lock);
7593 if (freq->old < freq->new && send_ipi) {
7595 * We upscale the frequency. Must make the guest
7596 * doesn't see old kvmclock values while running with
7597 * the new frequency, otherwise we risk the guest sees
7598 * time go backwards.
7600 * In case we update the frequency for another cpu
7601 * (which might be in guest context) send an interrupt
7602 * to kick the cpu out of guest context. Next time
7603 * guest context is entered kvmclock will be updated,
7604 * so the guest will not see stale values.
7606 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7610 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7613 struct cpufreq_freqs *freq = data;
7616 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7618 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7621 for_each_cpu(cpu, freq->policy->cpus)
7622 __kvmclock_cpufreq_notifier(freq, cpu);
7627 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7628 .notifier_call = kvmclock_cpufreq_notifier
7631 static int kvmclock_cpu_online(unsigned int cpu)
7633 tsc_khz_changed(NULL);
7637 static void kvm_timer_init(void)
7639 max_tsc_khz = tsc_khz;
7641 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7642 #ifdef CONFIG_CPU_FREQ
7643 struct cpufreq_policy *policy;
7647 policy = cpufreq_cpu_get(cpu);
7649 if (policy->cpuinfo.max_freq)
7650 max_tsc_khz = policy->cpuinfo.max_freq;
7651 cpufreq_cpu_put(policy);
7655 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7656 CPUFREQ_TRANSITION_NOTIFIER);
7659 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7660 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7663 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7664 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7666 int kvm_is_in_guest(void)
7668 return __this_cpu_read(current_vcpu) != NULL;
7671 static int kvm_is_user_mode(void)
7675 if (__this_cpu_read(current_vcpu))
7676 user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7678 return user_mode != 0;
7681 static unsigned long kvm_get_guest_ip(void)
7683 unsigned long ip = 0;
7685 if (__this_cpu_read(current_vcpu))
7686 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7691 static void kvm_handle_intel_pt_intr(void)
7693 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7695 kvm_make_request(KVM_REQ_PMI, vcpu);
7696 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7697 (unsigned long *)&vcpu->arch.pmu.global_status);
7700 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7701 .is_in_guest = kvm_is_in_guest,
7702 .is_user_mode = kvm_is_user_mode,
7703 .get_guest_ip = kvm_get_guest_ip,
7704 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7707 #ifdef CONFIG_X86_64
7708 static void pvclock_gtod_update_fn(struct work_struct *work)
7712 struct kvm_vcpu *vcpu;
7715 mutex_lock(&kvm_lock);
7716 list_for_each_entry(kvm, &vm_list, vm_list)
7717 kvm_for_each_vcpu(i, vcpu, kvm)
7718 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7719 atomic_set(&kvm_guest_has_master_clock, 0);
7720 mutex_unlock(&kvm_lock);
7723 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7726 * Notification about pvclock gtod data update.
7728 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7731 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7732 struct timekeeper *tk = priv;
7734 update_pvclock_gtod(tk);
7736 /* disable master clock if host does not trust, or does not
7737 * use, TSC based clocksource.
7739 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7740 atomic_read(&kvm_guest_has_master_clock) != 0)
7741 queue_work(system_long_wq, &pvclock_gtod_work);
7746 static struct notifier_block pvclock_gtod_notifier = {
7747 .notifier_call = pvclock_gtod_notify,
7751 int kvm_arch_init(void *opaque)
7753 struct kvm_x86_init_ops *ops = opaque;
7756 if (kvm_x86_ops.hardware_enable) {
7757 printk(KERN_ERR "kvm: already loaded the other module\n");
7762 if (!ops->cpu_has_kvm_support()) {
7763 pr_err_ratelimited("kvm: no hardware support\n");
7767 if (ops->disabled_by_bios()) {
7768 pr_err_ratelimited("kvm: disabled by bios\n");
7774 * KVM explicitly assumes that the guest has an FPU and
7775 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7776 * vCPU's FPU state as a fxregs_state struct.
7778 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7779 printk(KERN_ERR "kvm: inadequate fpu\n");
7785 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7786 __alignof__(struct fpu), SLAB_ACCOUNT,
7788 if (!x86_fpu_cache) {
7789 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7793 x86_emulator_cache = kvm_alloc_emulator_cache();
7794 if (!x86_emulator_cache) {
7795 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7796 goto out_free_x86_fpu_cache;
7799 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
7800 if (!user_return_msrs) {
7801 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
7802 goto out_free_x86_emulator_cache;
7805 r = kvm_mmu_module_init();
7807 goto out_free_percpu;
7809 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7810 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7811 PT_PRESENT_MASK, 0, sme_me_mask);
7814 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7816 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7817 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7818 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7822 if (pi_inject_timer == -1)
7823 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7824 #ifdef CONFIG_X86_64
7825 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7827 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7828 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7834 free_percpu(user_return_msrs);
7835 out_free_x86_emulator_cache:
7836 kmem_cache_destroy(x86_emulator_cache);
7837 out_free_x86_fpu_cache:
7838 kmem_cache_destroy(x86_fpu_cache);
7843 void kvm_arch_exit(void)
7845 #ifdef CONFIG_X86_64
7846 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7847 clear_hv_tscchange_cb();
7850 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7852 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7853 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7854 CPUFREQ_TRANSITION_NOTIFIER);
7855 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7856 #ifdef CONFIG_X86_64
7857 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7859 kvm_x86_ops.hardware_enable = NULL;
7860 kvm_mmu_module_exit();
7861 free_percpu(user_return_msrs);
7862 kmem_cache_destroy(x86_fpu_cache);
7865 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7867 ++vcpu->stat.halt_exits;
7868 if (lapic_in_kernel(vcpu)) {
7869 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7872 vcpu->run->exit_reason = KVM_EXIT_HLT;
7876 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7878 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7880 int ret = kvm_skip_emulated_instruction(vcpu);
7882 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7883 * KVM_EXIT_DEBUG here.
7885 return kvm_vcpu_halt(vcpu) && ret;
7887 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7889 #ifdef CONFIG_X86_64
7890 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7891 unsigned long clock_type)
7893 struct kvm_clock_pairing clock_pairing;
7894 struct timespec64 ts;
7898 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7899 return -KVM_EOPNOTSUPP;
7901 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7902 return -KVM_EOPNOTSUPP;
7904 clock_pairing.sec = ts.tv_sec;
7905 clock_pairing.nsec = ts.tv_nsec;
7906 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7907 clock_pairing.flags = 0;
7908 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7911 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7912 sizeof(struct kvm_clock_pairing)))
7920 * kvm_pv_kick_cpu_op: Kick a vcpu.
7922 * @apicid - apicid of vcpu to be kicked.
7924 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7926 struct kvm_lapic_irq lapic_irq;
7928 lapic_irq.shorthand = APIC_DEST_NOSHORT;
7929 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7930 lapic_irq.level = 0;
7931 lapic_irq.dest_id = apicid;
7932 lapic_irq.msi_redir_hint = false;
7934 lapic_irq.delivery_mode = APIC_DM_REMRD;
7935 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7938 bool kvm_apicv_activated(struct kvm *kvm)
7940 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
7942 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
7944 void kvm_apicv_init(struct kvm *kvm, bool enable)
7947 clear_bit(APICV_INHIBIT_REASON_DISABLE,
7948 &kvm->arch.apicv_inhibit_reasons);
7950 set_bit(APICV_INHIBIT_REASON_DISABLE,
7951 &kvm->arch.apicv_inhibit_reasons);
7953 EXPORT_SYMBOL_GPL(kvm_apicv_init);
7955 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7957 struct kvm_vcpu *target = NULL;
7958 struct kvm_apic_map *map;
7961 map = rcu_dereference(kvm->arch.apic_map);
7963 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7964 target = map->phys_map[dest_id]->vcpu;
7968 if (target && READ_ONCE(target->ready))
7969 kvm_vcpu_yield_to(target);
7972 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7974 unsigned long nr, a0, a1, a2, a3, ret;
7977 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7978 return kvm_hv_hypercall(vcpu);
7980 nr = kvm_rax_read(vcpu);
7981 a0 = kvm_rbx_read(vcpu);
7982 a1 = kvm_rcx_read(vcpu);
7983 a2 = kvm_rdx_read(vcpu);
7984 a3 = kvm_rsi_read(vcpu);
7986 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7988 op_64_bit = is_64_bit_mode(vcpu);
7997 if (kvm_x86_ops.get_cpl(vcpu) != 0) {
8003 case KVM_HC_VAPIC_POLL_IRQ:
8006 case KVM_HC_KICK_CPU:
8007 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8008 kvm_sched_yield(vcpu->kvm, a1);
8011 #ifdef CONFIG_X86_64
8012 case KVM_HC_CLOCK_PAIRING:
8013 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8016 case KVM_HC_SEND_IPI:
8017 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8019 case KVM_HC_SCHED_YIELD:
8020 kvm_sched_yield(vcpu->kvm, a0);
8030 kvm_rax_write(vcpu, ret);
8032 ++vcpu->stat.hypercalls;
8033 return kvm_skip_emulated_instruction(vcpu);
8035 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8037 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8039 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8040 char instruction[3];
8041 unsigned long rip = kvm_rip_read(vcpu);
8043 kvm_x86_ops.patch_hypercall(vcpu, instruction);
8045 return emulator_write_emulated(ctxt, rip, instruction, 3,
8049 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8051 return vcpu->run->request_interrupt_window &&
8052 likely(!pic_in_kernel(vcpu->kvm));
8055 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8057 struct kvm_run *kvm_run = vcpu->run;
8059 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8060 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
8061 kvm_run->cr8 = kvm_get_cr8(vcpu);
8062 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8063 kvm_run->ready_for_interrupt_injection =
8064 pic_in_kernel(vcpu->kvm) ||
8065 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8068 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8072 if (!kvm_x86_ops.update_cr8_intercept)
8075 if (!lapic_in_kernel(vcpu))
8078 if (vcpu->arch.apicv_active)
8081 if (!vcpu->arch.apic->vapic_addr)
8082 max_irr = kvm_lapic_find_highest_irr(vcpu);
8089 tpr = kvm_lapic_get_cr8(vcpu);
8091 kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
8094 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8097 bool can_inject = true;
8099 /* try to reinject previous events if any */
8101 if (vcpu->arch.exception.injected) {
8102 kvm_x86_ops.queue_exception(vcpu);
8106 * Do not inject an NMI or interrupt if there is a pending
8107 * exception. Exceptions and interrupts are recognized at
8108 * instruction boundaries, i.e. the start of an instruction.
8109 * Trap-like exceptions, e.g. #DB, have higher priority than
8110 * NMIs and interrupts, i.e. traps are recognized before an
8111 * NMI/interrupt that's pending on the same instruction.
8112 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8113 * priority, but are only generated (pended) during instruction
8114 * execution, i.e. a pending fault-like exception means the
8115 * fault occurred on the *previous* instruction and must be
8116 * serviced prior to recognizing any new events in order to
8117 * fully complete the previous instruction.
8119 else if (!vcpu->arch.exception.pending) {
8120 if (vcpu->arch.nmi_injected) {
8121 kvm_x86_ops.set_nmi(vcpu);
8123 } else if (vcpu->arch.interrupt.injected) {
8124 kvm_x86_ops.set_irq(vcpu);
8129 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8130 vcpu->arch.exception.pending);
8133 * Call check_nested_events() even if we reinjected a previous event
8134 * in order for caller to determine if it should require immediate-exit
8135 * from L2 to L1 due to pending L1 events which require exit
8138 if (is_guest_mode(vcpu)) {
8139 r = kvm_x86_ops.nested_ops->check_events(vcpu);
8144 /* try to inject new event if pending */
8145 if (vcpu->arch.exception.pending) {
8146 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8147 vcpu->arch.exception.has_error_code,
8148 vcpu->arch.exception.error_code);
8150 vcpu->arch.exception.pending = false;
8151 vcpu->arch.exception.injected = true;
8153 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8154 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8157 if (vcpu->arch.exception.nr == DB_VECTOR) {
8158 kvm_deliver_exception_payload(vcpu);
8159 if (vcpu->arch.dr7 & DR7_GD) {
8160 vcpu->arch.dr7 &= ~DR7_GD;
8161 kvm_update_dr7(vcpu);
8165 kvm_x86_ops.queue_exception(vcpu);
8170 * Finally, inject interrupt events. If an event cannot be injected
8171 * due to architectural conditions (e.g. IF=0) a window-open exit
8172 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8173 * and can architecturally be injected, but we cannot do it right now:
8174 * an interrupt could have arrived just now and we have to inject it
8175 * as a vmexit, or there could already an event in the queue, which is
8176 * indicated by can_inject. In that case we request an immediate exit
8177 * in order to make progress and get back here for another iteration.
8178 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8180 if (vcpu->arch.smi_pending) {
8181 r = can_inject ? kvm_x86_ops.smi_allowed(vcpu, true) : -EBUSY;
8185 vcpu->arch.smi_pending = false;
8186 ++vcpu->arch.smi_count;
8190 kvm_x86_ops.enable_smi_window(vcpu);
8193 if (vcpu->arch.nmi_pending) {
8194 r = can_inject ? kvm_x86_ops.nmi_allowed(vcpu, true) : -EBUSY;
8198 --vcpu->arch.nmi_pending;
8199 vcpu->arch.nmi_injected = true;
8200 kvm_x86_ops.set_nmi(vcpu);
8202 WARN_ON(kvm_x86_ops.nmi_allowed(vcpu, true) < 0);
8204 if (vcpu->arch.nmi_pending)
8205 kvm_x86_ops.enable_nmi_window(vcpu);
8208 if (kvm_cpu_has_injectable_intr(vcpu)) {
8209 r = can_inject ? kvm_x86_ops.interrupt_allowed(vcpu, true) : -EBUSY;
8213 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8214 kvm_x86_ops.set_irq(vcpu);
8215 WARN_ON(kvm_x86_ops.interrupt_allowed(vcpu, true) < 0);
8217 if (kvm_cpu_has_injectable_intr(vcpu))
8218 kvm_x86_ops.enable_irq_window(vcpu);
8221 if (is_guest_mode(vcpu) &&
8222 kvm_x86_ops.nested_ops->hv_timer_pending &&
8223 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8224 *req_immediate_exit = true;
8226 WARN_ON(vcpu->arch.exception.pending);
8230 *req_immediate_exit = true;
8234 static void process_nmi(struct kvm_vcpu *vcpu)
8239 * x86 is limited to one NMI running, and one NMI pending after it.
8240 * If an NMI is already in progress, limit further NMIs to just one.
8241 * Otherwise, allow two (and we'll inject the first one immediately).
8243 if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
8246 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8247 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8248 kvm_make_request(KVM_REQ_EVENT, vcpu);
8251 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8254 flags |= seg->g << 23;
8255 flags |= seg->db << 22;
8256 flags |= seg->l << 21;
8257 flags |= seg->avl << 20;
8258 flags |= seg->present << 15;
8259 flags |= seg->dpl << 13;
8260 flags |= seg->s << 12;
8261 flags |= seg->type << 8;
8265 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8267 struct kvm_segment seg;
8270 kvm_get_segment(vcpu, &seg, n);
8271 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8274 offset = 0x7f84 + n * 12;
8276 offset = 0x7f2c + (n - 3) * 12;
8278 put_smstate(u32, buf, offset + 8, seg.base);
8279 put_smstate(u32, buf, offset + 4, seg.limit);
8280 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8283 #ifdef CONFIG_X86_64
8284 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8286 struct kvm_segment seg;
8290 kvm_get_segment(vcpu, &seg, n);
8291 offset = 0x7e00 + n * 16;
8293 flags = enter_smm_get_segment_flags(&seg) >> 8;
8294 put_smstate(u16, buf, offset, seg.selector);
8295 put_smstate(u16, buf, offset + 2, flags);
8296 put_smstate(u32, buf, offset + 4, seg.limit);
8297 put_smstate(u64, buf, offset + 8, seg.base);
8301 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8304 struct kvm_segment seg;
8308 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8309 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8310 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8311 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8313 for (i = 0; i < 8; i++)
8314 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
8316 kvm_get_dr(vcpu, 6, &val);
8317 put_smstate(u32, buf, 0x7fcc, (u32)val);
8318 kvm_get_dr(vcpu, 7, &val);
8319 put_smstate(u32, buf, 0x7fc8, (u32)val);
8321 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8322 put_smstate(u32, buf, 0x7fc4, seg.selector);
8323 put_smstate(u32, buf, 0x7f64, seg.base);
8324 put_smstate(u32, buf, 0x7f60, seg.limit);
8325 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8327 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8328 put_smstate(u32, buf, 0x7fc0, seg.selector);
8329 put_smstate(u32, buf, 0x7f80, seg.base);
8330 put_smstate(u32, buf, 0x7f7c, seg.limit);
8331 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8333 kvm_x86_ops.get_gdt(vcpu, &dt);
8334 put_smstate(u32, buf, 0x7f74, dt.address);
8335 put_smstate(u32, buf, 0x7f70, dt.size);
8337 kvm_x86_ops.get_idt(vcpu, &dt);
8338 put_smstate(u32, buf, 0x7f58, dt.address);
8339 put_smstate(u32, buf, 0x7f54, dt.size);
8341 for (i = 0; i < 6; i++)
8342 enter_smm_save_seg_32(vcpu, buf, i);
8344 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8347 put_smstate(u32, buf, 0x7efc, 0x00020000);
8348 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8351 #ifdef CONFIG_X86_64
8352 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8355 struct kvm_segment seg;
8359 for (i = 0; i < 16; i++)
8360 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8362 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8363 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8365 kvm_get_dr(vcpu, 6, &val);
8366 put_smstate(u64, buf, 0x7f68, val);
8367 kvm_get_dr(vcpu, 7, &val);
8368 put_smstate(u64, buf, 0x7f60, val);
8370 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8371 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8372 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8374 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8377 put_smstate(u32, buf, 0x7efc, 0x00020064);
8379 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8381 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8382 put_smstate(u16, buf, 0x7e90, seg.selector);
8383 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8384 put_smstate(u32, buf, 0x7e94, seg.limit);
8385 put_smstate(u64, buf, 0x7e98, seg.base);
8387 kvm_x86_ops.get_idt(vcpu, &dt);
8388 put_smstate(u32, buf, 0x7e84, dt.size);
8389 put_smstate(u64, buf, 0x7e88, dt.address);
8391 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8392 put_smstate(u16, buf, 0x7e70, seg.selector);
8393 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8394 put_smstate(u32, buf, 0x7e74, seg.limit);
8395 put_smstate(u64, buf, 0x7e78, seg.base);
8397 kvm_x86_ops.get_gdt(vcpu, &dt);
8398 put_smstate(u32, buf, 0x7e64, dt.size);
8399 put_smstate(u64, buf, 0x7e68, dt.address);
8401 for (i = 0; i < 6; i++)
8402 enter_smm_save_seg_64(vcpu, buf, i);
8406 static void enter_smm(struct kvm_vcpu *vcpu)
8408 struct kvm_segment cs, ds;
8413 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8414 memset(buf, 0, 512);
8415 #ifdef CONFIG_X86_64
8416 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8417 enter_smm_save_state_64(vcpu, buf);
8420 enter_smm_save_state_32(vcpu, buf);
8423 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8424 * vCPU state (e.g. leave guest mode) after we've saved the state into
8425 * the SMM state-save area.
8427 kvm_x86_ops.pre_enter_smm(vcpu, buf);
8429 vcpu->arch.hflags |= HF_SMM_MASK;
8430 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8432 if (kvm_x86_ops.get_nmi_mask(vcpu))
8433 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8435 kvm_x86_ops.set_nmi_mask(vcpu, true);
8437 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8438 kvm_rip_write(vcpu, 0x8000);
8440 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8441 kvm_x86_ops.set_cr0(vcpu, cr0);
8442 vcpu->arch.cr0 = cr0;
8444 kvm_x86_ops.set_cr4(vcpu, 0);
8446 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8447 dt.address = dt.size = 0;
8448 kvm_x86_ops.set_idt(vcpu, &dt);
8450 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8452 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8453 cs.base = vcpu->arch.smbase;
8458 cs.limit = ds.limit = 0xffffffff;
8459 cs.type = ds.type = 0x3;
8460 cs.dpl = ds.dpl = 0;
8465 cs.avl = ds.avl = 0;
8466 cs.present = ds.present = 1;
8467 cs.unusable = ds.unusable = 0;
8468 cs.padding = ds.padding = 0;
8470 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8471 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8472 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8473 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8474 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8475 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8477 #ifdef CONFIG_X86_64
8478 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8479 kvm_x86_ops.set_efer(vcpu, 0);
8482 kvm_update_cpuid_runtime(vcpu);
8483 kvm_mmu_reset_context(vcpu);
8486 static void process_smi(struct kvm_vcpu *vcpu)
8488 vcpu->arch.smi_pending = true;
8489 kvm_make_request(KVM_REQ_EVENT, vcpu);
8492 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8493 unsigned long *vcpu_bitmap)
8497 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8499 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8500 NULL, vcpu_bitmap, cpus);
8502 free_cpumask_var(cpus);
8505 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8507 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8510 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8512 if (!lapic_in_kernel(vcpu))
8515 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8516 kvm_apic_update_apicv(vcpu);
8517 kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8519 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8522 * NOTE: Do not hold any lock prior to calling this.
8524 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8525 * locked, because it calls __x86_set_memory_region() which does
8526 * synchronize_srcu(&kvm->srcu).
8528 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8530 struct kvm_vcpu *except;
8531 unsigned long old, new, expected;
8533 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8534 !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8537 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8539 expected = new = old;
8541 __clear_bit(bit, &new);
8543 __set_bit(bit, &new);
8546 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8547 } while (old != expected);
8552 trace_kvm_apicv_update_request(activate, bit);
8553 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8554 kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8557 * Sending request to update APICV for all other vcpus,
8558 * while update the calling vcpu immediately instead of
8559 * waiting for another #VMEXIT to handle the request.
8561 except = kvm_get_running_vcpu();
8562 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8565 kvm_vcpu_update_apicv(except);
8567 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8569 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8571 if (!kvm_apic_present(vcpu))
8574 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8576 if (irqchip_split(vcpu->kvm))
8577 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8579 if (vcpu->arch.apicv_active)
8580 kvm_x86_ops.sync_pir_to_irr(vcpu);
8581 if (ioapic_in_kernel(vcpu->kvm))
8582 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8585 if (is_guest_mode(vcpu))
8586 vcpu->arch.load_eoi_exitmap_pending = true;
8588 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8591 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8593 u64 eoi_exit_bitmap[4];
8595 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8598 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8599 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8600 kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8603 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8604 unsigned long start, unsigned long end)
8606 unsigned long apic_address;
8609 * The physical address of apic access page is stored in the VMCS.
8610 * Update it when it becomes invalid.
8612 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8613 if (start <= apic_address && apic_address < end)
8614 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8617 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8619 if (!lapic_in_kernel(vcpu))
8622 if (!kvm_x86_ops.set_apic_access_page_addr)
8625 kvm_x86_ops.set_apic_access_page_addr(vcpu);
8628 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8630 smp_send_reschedule(vcpu->cpu);
8632 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8635 * Returns 1 to let vcpu_run() continue the guest execution loop without
8636 * exiting to the userspace. Otherwise, the value will be returned to the
8639 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8643 dm_request_for_irq_injection(vcpu) &&
8644 kvm_cpu_accept_dm_intr(vcpu);
8645 fastpath_t exit_fastpath;
8647 bool req_immediate_exit = false;
8649 if (kvm_request_pending(vcpu)) {
8650 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
8651 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
8656 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8657 kvm_mmu_unload(vcpu);
8658 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8659 __kvm_migrate_timers(vcpu);
8660 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8661 kvm_gen_update_masterclock(vcpu->kvm);
8662 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8663 kvm_gen_kvmclock_update(vcpu);
8664 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8665 r = kvm_guest_time_update(vcpu);
8669 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8670 kvm_mmu_sync_roots(vcpu);
8671 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8672 kvm_mmu_load_pgd(vcpu);
8673 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8674 kvm_vcpu_flush_tlb_all(vcpu);
8676 /* Flushing all ASIDs flushes the current ASID... */
8677 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8679 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8680 kvm_vcpu_flush_tlb_current(vcpu);
8681 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8682 kvm_vcpu_flush_tlb_guest(vcpu);
8684 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8685 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8689 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8690 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8691 vcpu->mmio_needed = 0;
8695 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8696 /* Page is swapped out. Do synthetic halt */
8697 vcpu->arch.apf.halted = true;
8701 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8702 record_steal_time(vcpu);
8703 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8705 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8707 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8708 kvm_pmu_handle_event(vcpu);
8709 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8710 kvm_pmu_deliver_pmi(vcpu);
8711 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8712 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8713 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8714 vcpu->arch.ioapic_handled_vectors)) {
8715 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8716 vcpu->run->eoi.vector =
8717 vcpu->arch.pending_ioapic_eoi;
8722 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8723 vcpu_scan_ioapic(vcpu);
8724 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8725 vcpu_load_eoi_exitmap(vcpu);
8726 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8727 kvm_vcpu_reload_apic_access_page(vcpu);
8728 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8729 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8730 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8734 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8735 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8736 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8740 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8741 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8742 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8748 * KVM_REQ_HV_STIMER has to be processed after
8749 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8750 * depend on the guest clock being up-to-date
8752 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8753 kvm_hv_process_stimers(vcpu);
8754 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8755 kvm_vcpu_update_apicv(vcpu);
8756 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
8757 kvm_check_async_pf_completion(vcpu);
8758 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
8759 kvm_x86_ops.msr_filter_changed(vcpu);
8762 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8763 ++vcpu->stat.req_event;
8764 kvm_apic_accept_events(vcpu);
8765 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8770 inject_pending_event(vcpu, &req_immediate_exit);
8772 kvm_x86_ops.enable_irq_window(vcpu);
8774 if (kvm_lapic_enabled(vcpu)) {
8775 update_cr8_intercept(vcpu);
8776 kvm_lapic_sync_to_vapic(vcpu);
8780 r = kvm_mmu_reload(vcpu);
8782 goto cancel_injection;
8787 kvm_x86_ops.prepare_guest_switch(vcpu);
8790 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8791 * IPI are then delayed after guest entry, which ensures that they
8792 * result in virtual interrupt delivery.
8794 local_irq_disable();
8795 vcpu->mode = IN_GUEST_MODE;
8797 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8800 * 1) We should set ->mode before checking ->requests. Please see
8801 * the comment in kvm_vcpu_exiting_guest_mode().
8803 * 2) For APICv, we should set ->mode before checking PID.ON. This
8804 * pairs with the memory barrier implicit in pi_test_and_set_on
8805 * (see vmx_deliver_posted_interrupt).
8807 * 3) This also orders the write to mode from any reads to the page
8808 * tables done while the VCPU is running. Please see the comment
8809 * in kvm_flush_remote_tlbs.
8811 smp_mb__after_srcu_read_unlock();
8814 * This handles the case where a posted interrupt was
8815 * notified with kvm_vcpu_kick.
8817 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8818 kvm_x86_ops.sync_pir_to_irr(vcpu);
8820 if (kvm_vcpu_exit_request(vcpu)) {
8821 vcpu->mode = OUTSIDE_GUEST_MODE;
8825 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8827 goto cancel_injection;
8830 if (req_immediate_exit) {
8831 kvm_make_request(KVM_REQ_EVENT, vcpu);
8832 kvm_x86_ops.request_immediate_exit(vcpu);
8835 trace_kvm_entry(vcpu);
8837 fpregs_assert_state_consistent();
8838 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8839 switch_fpu_return();
8841 if (unlikely(vcpu->arch.switch_db_regs)) {
8843 set_debugreg(vcpu->arch.eff_db[0], 0);
8844 set_debugreg(vcpu->arch.eff_db[1], 1);
8845 set_debugreg(vcpu->arch.eff_db[2], 2);
8846 set_debugreg(vcpu->arch.eff_db[3], 3);
8847 set_debugreg(vcpu->arch.dr6, 6);
8848 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8851 exit_fastpath = kvm_x86_ops.run(vcpu);
8854 * Do this here before restoring debug registers on the host. And
8855 * since we do this before handling the vmexit, a DR access vmexit
8856 * can (a) read the correct value of the debug registers, (b) set
8857 * KVM_DEBUGREG_WONT_EXIT again.
8859 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8860 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8861 kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8862 kvm_update_dr0123(vcpu);
8863 kvm_update_dr7(vcpu);
8864 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8868 * If the guest has used debug registers, at least dr7
8869 * will be disabled while returning to the host.
8870 * If we don't have active breakpoints in the host, we don't
8871 * care about the messed up debug address registers. But if
8872 * we have some of them active, restore the old state.
8874 if (hw_breakpoint_active())
8875 hw_breakpoint_restore();
8877 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
8878 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8880 vcpu->mode = OUTSIDE_GUEST_MODE;
8883 kvm_x86_ops.handle_exit_irqoff(vcpu);
8886 * Consume any pending interrupts, including the possible source of
8887 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8888 * An instruction is required after local_irq_enable() to fully unblock
8889 * interrupts on processors that implement an interrupt shadow, the
8890 * stat.exits increment will do nicely.
8892 kvm_before_interrupt(vcpu);
8895 local_irq_disable();
8896 kvm_after_interrupt(vcpu);
8898 if (lapic_in_kernel(vcpu)) {
8899 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8900 if (delta != S64_MIN) {
8901 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8902 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8909 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8912 * Profile KVM exit RIPs:
8914 if (unlikely(prof_on == KVM_PROFILING)) {
8915 unsigned long rip = kvm_rip_read(vcpu);
8916 profile_hit(KVM_PROFILING, (void *)rip);
8919 if (unlikely(vcpu->arch.tsc_always_catchup))
8920 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8922 if (vcpu->arch.apic_attention)
8923 kvm_lapic_sync_from_vapic(vcpu);
8925 r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
8929 if (req_immediate_exit)
8930 kvm_make_request(KVM_REQ_EVENT, vcpu);
8931 kvm_x86_ops.cancel_injection(vcpu);
8932 if (unlikely(vcpu->arch.apic_attention))
8933 kvm_lapic_sync_from_vapic(vcpu);
8938 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8940 if (!kvm_arch_vcpu_runnable(vcpu) &&
8941 (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
8942 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8943 kvm_vcpu_block(vcpu);
8944 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8946 if (kvm_x86_ops.post_block)
8947 kvm_x86_ops.post_block(vcpu);
8949 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8953 kvm_apic_accept_events(vcpu);
8954 switch(vcpu->arch.mp_state) {
8955 case KVM_MP_STATE_HALTED:
8956 vcpu->arch.pv.pv_unhalted = false;
8957 vcpu->arch.mp_state =
8958 KVM_MP_STATE_RUNNABLE;
8960 case KVM_MP_STATE_RUNNABLE:
8961 vcpu->arch.apf.halted = false;
8963 case KVM_MP_STATE_INIT_RECEIVED:
8971 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8973 if (is_guest_mode(vcpu))
8974 kvm_x86_ops.nested_ops->check_events(vcpu);
8976 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8977 !vcpu->arch.apf.halted);
8980 static int vcpu_run(struct kvm_vcpu *vcpu)
8983 struct kvm *kvm = vcpu->kvm;
8985 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8986 vcpu->arch.l1tf_flush_l1d = true;
8989 if (kvm_vcpu_running(vcpu)) {
8990 r = vcpu_enter_guest(vcpu);
8992 r = vcpu_block(kvm, vcpu);
8998 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8999 if (kvm_cpu_has_pending_timer(vcpu))
9000 kvm_inject_pending_timer_irqs(vcpu);
9002 if (dm_request_for_irq_injection(vcpu) &&
9003 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9005 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9006 ++vcpu->stat.request_irq_exits;
9010 if (__xfer_to_guest_mode_work_pending()) {
9011 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9012 r = xfer_to_guest_mode_handle_work(vcpu);
9015 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9019 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9024 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9028 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9029 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9030 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9034 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9036 BUG_ON(!vcpu->arch.pio.count);
9038 return complete_emulated_io(vcpu);
9042 * Implements the following, as a state machine:
9046 * for each mmio piece in the fragment
9054 * for each mmio piece in the fragment
9059 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9061 struct kvm_run *run = vcpu->run;
9062 struct kvm_mmio_fragment *frag;
9065 BUG_ON(!vcpu->mmio_needed);
9067 /* Complete previous fragment */
9068 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9069 len = min(8u, frag->len);
9070 if (!vcpu->mmio_is_write)
9071 memcpy(frag->data, run->mmio.data, len);
9073 if (frag->len <= 8) {
9074 /* Switch to the next fragment. */
9076 vcpu->mmio_cur_fragment++;
9078 /* Go forward to the next mmio piece. */
9084 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9085 vcpu->mmio_needed = 0;
9087 /* FIXME: return into emulator if single-stepping. */
9088 if (vcpu->mmio_is_write)
9090 vcpu->mmio_read_completed = 1;
9091 return complete_emulated_io(vcpu);
9094 run->exit_reason = KVM_EXIT_MMIO;
9095 run->mmio.phys_addr = frag->gpa;
9096 if (vcpu->mmio_is_write)
9097 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9098 run->mmio.len = min(8u, frag->len);
9099 run->mmio.is_write = vcpu->mmio_is_write;
9100 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9104 static void kvm_save_current_fpu(struct fpu *fpu)
9107 * If the target FPU state is not resident in the CPU registers, just
9108 * memcpy() from current, else save CPU state directly to the target.
9110 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9111 memcpy(&fpu->state, ¤t->thread.fpu.state,
9112 fpu_kernel_xstate_size);
9114 copy_fpregs_to_fpstate(fpu);
9117 /* Swap (qemu) user FPU context for the guest FPU context. */
9118 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9122 kvm_save_current_fpu(vcpu->arch.user_fpu);
9124 /* PKRU is separately restored in kvm_x86_ops.run. */
9125 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9126 ~XFEATURE_MASK_PKRU);
9128 fpregs_mark_activate();
9134 /* When vcpu_run ends, restore user space FPU context. */
9135 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9139 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9141 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9143 fpregs_mark_activate();
9146 ++vcpu->stat.fpu_reload;
9150 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9152 struct kvm_run *kvm_run = vcpu->run;
9156 kvm_sigset_activate(vcpu);
9157 kvm_load_guest_fpu(vcpu);
9159 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9160 if (kvm_run->immediate_exit) {
9164 kvm_vcpu_block(vcpu);
9165 kvm_apic_accept_events(vcpu);
9166 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9168 if (signal_pending(current)) {
9170 kvm_run->exit_reason = KVM_EXIT_INTR;
9171 ++vcpu->stat.signal_exits;
9176 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9181 if (kvm_run->kvm_dirty_regs) {
9182 r = sync_regs(vcpu);
9187 /* re-sync apic's tpr */
9188 if (!lapic_in_kernel(vcpu)) {
9189 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9195 if (unlikely(vcpu->arch.complete_userspace_io)) {
9196 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9197 vcpu->arch.complete_userspace_io = NULL;
9202 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9204 if (kvm_run->immediate_exit)
9210 kvm_put_guest_fpu(vcpu);
9211 if (kvm_run->kvm_valid_regs)
9213 post_kvm_run_save(vcpu);
9214 kvm_sigset_deactivate(vcpu);
9220 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9222 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9224 * We are here if userspace calls get_regs() in the middle of
9225 * instruction emulation. Registers state needs to be copied
9226 * back from emulation context to vcpu. Userspace shouldn't do
9227 * that usually, but some bad designed PV devices (vmware
9228 * backdoor interface) need this to work
9230 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9231 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9233 regs->rax = kvm_rax_read(vcpu);
9234 regs->rbx = kvm_rbx_read(vcpu);
9235 regs->rcx = kvm_rcx_read(vcpu);
9236 regs->rdx = kvm_rdx_read(vcpu);
9237 regs->rsi = kvm_rsi_read(vcpu);
9238 regs->rdi = kvm_rdi_read(vcpu);
9239 regs->rsp = kvm_rsp_read(vcpu);
9240 regs->rbp = kvm_rbp_read(vcpu);
9241 #ifdef CONFIG_X86_64
9242 regs->r8 = kvm_r8_read(vcpu);
9243 regs->r9 = kvm_r9_read(vcpu);
9244 regs->r10 = kvm_r10_read(vcpu);
9245 regs->r11 = kvm_r11_read(vcpu);
9246 regs->r12 = kvm_r12_read(vcpu);
9247 regs->r13 = kvm_r13_read(vcpu);
9248 regs->r14 = kvm_r14_read(vcpu);
9249 regs->r15 = kvm_r15_read(vcpu);
9252 regs->rip = kvm_rip_read(vcpu);
9253 regs->rflags = kvm_get_rflags(vcpu);
9256 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9259 __get_regs(vcpu, regs);
9264 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9266 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9267 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9269 kvm_rax_write(vcpu, regs->rax);
9270 kvm_rbx_write(vcpu, regs->rbx);
9271 kvm_rcx_write(vcpu, regs->rcx);
9272 kvm_rdx_write(vcpu, regs->rdx);
9273 kvm_rsi_write(vcpu, regs->rsi);
9274 kvm_rdi_write(vcpu, regs->rdi);
9275 kvm_rsp_write(vcpu, regs->rsp);
9276 kvm_rbp_write(vcpu, regs->rbp);
9277 #ifdef CONFIG_X86_64
9278 kvm_r8_write(vcpu, regs->r8);
9279 kvm_r9_write(vcpu, regs->r9);
9280 kvm_r10_write(vcpu, regs->r10);
9281 kvm_r11_write(vcpu, regs->r11);
9282 kvm_r12_write(vcpu, regs->r12);
9283 kvm_r13_write(vcpu, regs->r13);
9284 kvm_r14_write(vcpu, regs->r14);
9285 kvm_r15_write(vcpu, regs->r15);
9288 kvm_rip_write(vcpu, regs->rip);
9289 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9291 vcpu->arch.exception.pending = false;
9293 kvm_make_request(KVM_REQ_EVENT, vcpu);
9296 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9299 __set_regs(vcpu, regs);
9304 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9306 struct kvm_segment cs;
9308 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9312 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9314 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9318 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9319 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9320 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9321 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9322 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9323 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9325 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9326 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9328 kvm_x86_ops.get_idt(vcpu, &dt);
9329 sregs->idt.limit = dt.size;
9330 sregs->idt.base = dt.address;
9331 kvm_x86_ops.get_gdt(vcpu, &dt);
9332 sregs->gdt.limit = dt.size;
9333 sregs->gdt.base = dt.address;
9335 sregs->cr0 = kvm_read_cr0(vcpu);
9336 sregs->cr2 = vcpu->arch.cr2;
9337 sregs->cr3 = kvm_read_cr3(vcpu);
9338 sregs->cr4 = kvm_read_cr4(vcpu);
9339 sregs->cr8 = kvm_get_cr8(vcpu);
9340 sregs->efer = vcpu->arch.efer;
9341 sregs->apic_base = kvm_get_apic_base(vcpu);
9343 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9345 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9346 set_bit(vcpu->arch.interrupt.nr,
9347 (unsigned long *)sregs->interrupt_bitmap);
9350 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9351 struct kvm_sregs *sregs)
9354 __get_sregs(vcpu, sregs);
9359 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9360 struct kvm_mp_state *mp_state)
9363 if (kvm_mpx_supported())
9364 kvm_load_guest_fpu(vcpu);
9366 kvm_apic_accept_events(vcpu);
9367 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
9368 vcpu->arch.pv.pv_unhalted)
9369 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9371 mp_state->mp_state = vcpu->arch.mp_state;
9373 if (kvm_mpx_supported())
9374 kvm_put_guest_fpu(vcpu);
9379 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9380 struct kvm_mp_state *mp_state)
9386 if (!lapic_in_kernel(vcpu) &&
9387 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9391 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9392 * INIT state; latched init should be reported using
9393 * KVM_SET_VCPU_EVENTS, so reject it here.
9395 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9396 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9397 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9400 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9401 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9402 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9404 vcpu->arch.mp_state = mp_state->mp_state;
9405 kvm_make_request(KVM_REQ_EVENT, vcpu);
9413 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9414 int reason, bool has_error_code, u32 error_code)
9416 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9419 init_emulate_ctxt(vcpu);
9421 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9422 has_error_code, error_code);
9424 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9425 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9426 vcpu->run->internal.ndata = 0;
9430 kvm_rip_write(vcpu, ctxt->eip);
9431 kvm_set_rflags(vcpu, ctxt->eflags);
9434 EXPORT_SYMBOL_GPL(kvm_task_switch);
9436 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9438 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9440 * When EFER.LME and CR0.PG are set, the processor is in
9441 * 64-bit mode (though maybe in a 32-bit code segment).
9442 * CR4.PAE and EFER.LMA must be set.
9444 if (!(sregs->cr4 & X86_CR4_PAE)
9445 || !(sregs->efer & EFER_LMA))
9449 * Not in 64-bit mode: EFER.LMA is clear and the code
9450 * segment cannot be 64-bit.
9452 if (sregs->efer & EFER_LMA || sregs->cs.l)
9456 return kvm_valid_cr4(vcpu, sregs->cr4);
9459 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9461 struct msr_data apic_base_msr;
9462 int mmu_reset_needed = 0;
9463 int cpuid_update_needed = 0;
9464 int pending_vec, max_bits, idx;
9468 if (kvm_valid_sregs(vcpu, sregs))
9471 apic_base_msr.data = sregs->apic_base;
9472 apic_base_msr.host_initiated = true;
9473 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9476 dt.size = sregs->idt.limit;
9477 dt.address = sregs->idt.base;
9478 kvm_x86_ops.set_idt(vcpu, &dt);
9479 dt.size = sregs->gdt.limit;
9480 dt.address = sregs->gdt.base;
9481 kvm_x86_ops.set_gdt(vcpu, &dt);
9483 vcpu->arch.cr2 = sregs->cr2;
9484 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9485 vcpu->arch.cr3 = sregs->cr3;
9486 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9488 kvm_set_cr8(vcpu, sregs->cr8);
9490 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9491 kvm_x86_ops.set_efer(vcpu, sregs->efer);
9493 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9494 kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9495 vcpu->arch.cr0 = sregs->cr0;
9497 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9498 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9499 (X86_CR4_OSXSAVE | X86_CR4_PKE));
9500 kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9501 if (cpuid_update_needed)
9502 kvm_update_cpuid_runtime(vcpu);
9504 idx = srcu_read_lock(&vcpu->kvm->srcu);
9505 if (is_pae_paging(vcpu)) {
9506 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9507 mmu_reset_needed = 1;
9509 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9511 if (mmu_reset_needed)
9512 kvm_mmu_reset_context(vcpu);
9514 max_bits = KVM_NR_INTERRUPTS;
9515 pending_vec = find_first_bit(
9516 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9517 if (pending_vec < max_bits) {
9518 kvm_queue_interrupt(vcpu, pending_vec, false);
9519 pr_debug("Set back pending irq %d\n", pending_vec);
9522 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9523 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9524 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9525 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9526 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9527 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9529 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9530 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9532 update_cr8_intercept(vcpu);
9534 /* Older userspace won't unhalt the vcpu on reset. */
9535 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9536 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9538 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9540 kvm_make_request(KVM_REQ_EVENT, vcpu);
9547 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9548 struct kvm_sregs *sregs)
9553 ret = __set_sregs(vcpu, sregs);
9558 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9559 struct kvm_guest_debug *dbg)
9561 unsigned long rflags;
9566 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9568 if (vcpu->arch.exception.pending)
9570 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9571 kvm_queue_exception(vcpu, DB_VECTOR);
9573 kvm_queue_exception(vcpu, BP_VECTOR);
9577 * Read rflags as long as potentially injected trace flags are still
9580 rflags = kvm_get_rflags(vcpu);
9582 vcpu->guest_debug = dbg->control;
9583 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9584 vcpu->guest_debug = 0;
9586 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9587 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9588 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9589 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9591 for (i = 0; i < KVM_NR_DB_REGS; i++)
9592 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9594 kvm_update_dr7(vcpu);
9596 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9597 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9598 get_segment_base(vcpu, VCPU_SREG_CS);
9601 * Trigger an rflags update that will inject or remove the trace
9604 kvm_set_rflags(vcpu, rflags);
9606 kvm_x86_ops.update_exception_bitmap(vcpu);
9616 * Translate a guest virtual address to a guest physical address.
9618 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9619 struct kvm_translation *tr)
9621 unsigned long vaddr = tr->linear_address;
9627 idx = srcu_read_lock(&vcpu->kvm->srcu);
9628 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9629 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9630 tr->physical_address = gpa;
9631 tr->valid = gpa != UNMAPPED_GVA;
9639 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9641 struct fxregs_state *fxsave;
9645 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9646 memcpy(fpu->fpr, fxsave->st_space, 128);
9647 fpu->fcw = fxsave->cwd;
9648 fpu->fsw = fxsave->swd;
9649 fpu->ftwx = fxsave->twd;
9650 fpu->last_opcode = fxsave->fop;
9651 fpu->last_ip = fxsave->rip;
9652 fpu->last_dp = fxsave->rdp;
9653 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9659 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9661 struct fxregs_state *fxsave;
9665 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9667 memcpy(fxsave->st_space, fpu->fpr, 128);
9668 fxsave->cwd = fpu->fcw;
9669 fxsave->swd = fpu->fsw;
9670 fxsave->twd = fpu->ftwx;
9671 fxsave->fop = fpu->last_opcode;
9672 fxsave->rip = fpu->last_ip;
9673 fxsave->rdp = fpu->last_dp;
9674 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9680 static void store_regs(struct kvm_vcpu *vcpu)
9682 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9684 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9685 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9687 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9688 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9690 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9691 kvm_vcpu_ioctl_x86_get_vcpu_events(
9692 vcpu, &vcpu->run->s.regs.events);
9695 static int sync_regs(struct kvm_vcpu *vcpu)
9697 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9700 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9701 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9702 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9704 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9705 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9707 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9709 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9710 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9711 vcpu, &vcpu->run->s.regs.events))
9713 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9719 static void fx_init(struct kvm_vcpu *vcpu)
9721 fpstate_init(&vcpu->arch.guest_fpu->state);
9722 if (boot_cpu_has(X86_FEATURE_XSAVES))
9723 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9724 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9727 * Ensure guest xcr0 is valid for loading
9729 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9731 vcpu->arch.cr0 |= X86_CR0_ET;
9734 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9736 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9737 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9738 "guest TSC will not be reliable\n");
9743 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9748 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9749 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9751 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9753 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9755 r = kvm_mmu_create(vcpu);
9759 if (irqchip_in_kernel(vcpu->kvm)) {
9760 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9762 goto fail_mmu_destroy;
9763 if (kvm_apicv_activated(vcpu->kvm))
9764 vcpu->arch.apicv_active = true;
9766 static_key_slow_inc(&kvm_no_apic_vcpu);
9770 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9772 goto fail_free_lapic;
9773 vcpu->arch.pio_data = page_address(page);
9775 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9776 GFP_KERNEL_ACCOUNT);
9777 if (!vcpu->arch.mce_banks)
9778 goto fail_free_pio_data;
9779 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9781 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9782 GFP_KERNEL_ACCOUNT))
9783 goto fail_free_mce_banks;
9785 if (!alloc_emulate_ctxt(vcpu))
9786 goto free_wbinvd_dirty_mask;
9788 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9789 GFP_KERNEL_ACCOUNT);
9790 if (!vcpu->arch.user_fpu) {
9791 pr_err("kvm: failed to allocate userspace's fpu\n");
9792 goto free_emulate_ctxt;
9795 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9796 GFP_KERNEL_ACCOUNT);
9797 if (!vcpu->arch.guest_fpu) {
9798 pr_err("kvm: failed to allocate vcpu's fpu\n");
9803 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9805 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9807 kvm_async_pf_hash_reset(vcpu);
9810 vcpu->arch.pending_external_vector = -1;
9811 vcpu->arch.preempted_in_kernel = false;
9813 kvm_hv_vcpu_init(vcpu);
9815 r = kvm_x86_ops.vcpu_create(vcpu);
9817 goto free_guest_fpu;
9819 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9820 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9821 kvm_vcpu_mtrr_init(vcpu);
9823 kvm_vcpu_reset(vcpu, false);
9824 kvm_init_mmu(vcpu, false);
9829 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9831 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9833 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9834 free_wbinvd_dirty_mask:
9835 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9836 fail_free_mce_banks:
9837 kfree(vcpu->arch.mce_banks);
9839 free_page((unsigned long)vcpu->arch.pio_data);
9841 kvm_free_lapic(vcpu);
9843 kvm_mmu_destroy(vcpu);
9847 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9849 struct kvm *kvm = vcpu->kvm;
9851 kvm_hv_vcpu_postcreate(vcpu);
9853 if (mutex_lock_killable(&vcpu->mutex))
9856 kvm_synchronize_tsc(vcpu, 0);
9859 /* poll control enabled by default */
9860 vcpu->arch.msr_kvm_poll_control = 1;
9862 mutex_unlock(&vcpu->mutex);
9864 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9865 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9866 KVMCLOCK_SYNC_PERIOD);
9869 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9871 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9874 kvm_release_pfn(cache->pfn, cache->dirty, cache);
9876 kvmclock_reset(vcpu);
9878 kvm_x86_ops.vcpu_free(vcpu);
9880 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9881 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9882 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9883 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9885 kvm_hv_vcpu_uninit(vcpu);
9886 kvm_pmu_destroy(vcpu);
9887 kfree(vcpu->arch.mce_banks);
9888 kvm_free_lapic(vcpu);
9889 idx = srcu_read_lock(&vcpu->kvm->srcu);
9890 kvm_mmu_destroy(vcpu);
9891 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9892 free_page((unsigned long)vcpu->arch.pio_data);
9893 if (!lapic_in_kernel(vcpu))
9894 static_key_slow_dec(&kvm_no_apic_vcpu);
9897 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9899 kvm_lapic_reset(vcpu, init_event);
9901 vcpu->arch.hflags = 0;
9903 vcpu->arch.smi_pending = 0;
9904 vcpu->arch.smi_count = 0;
9905 atomic_set(&vcpu->arch.nmi_queued, 0);
9906 vcpu->arch.nmi_pending = 0;
9907 vcpu->arch.nmi_injected = false;
9908 kvm_clear_interrupt_queue(vcpu);
9909 kvm_clear_exception_queue(vcpu);
9911 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9912 kvm_update_dr0123(vcpu);
9913 vcpu->arch.dr6 = DR6_INIT;
9914 vcpu->arch.dr7 = DR7_FIXED_1;
9915 kvm_update_dr7(vcpu);
9919 kvm_make_request(KVM_REQ_EVENT, vcpu);
9920 vcpu->arch.apf.msr_en_val = 0;
9921 vcpu->arch.apf.msr_int_val = 0;
9922 vcpu->arch.st.msr_val = 0;
9924 kvmclock_reset(vcpu);
9926 kvm_clear_async_pf_completion_queue(vcpu);
9927 kvm_async_pf_hash_reset(vcpu);
9928 vcpu->arch.apf.halted = false;
9930 if (kvm_mpx_supported()) {
9931 void *mpx_state_buffer;
9934 * To avoid have the INIT path from kvm_apic_has_events() that be
9935 * called with loaded FPU and does not let userspace fix the state.
9938 kvm_put_guest_fpu(vcpu);
9939 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9941 if (mpx_state_buffer)
9942 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9943 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9945 if (mpx_state_buffer)
9946 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9948 kvm_load_guest_fpu(vcpu);
9952 kvm_pmu_reset(vcpu);
9953 vcpu->arch.smbase = 0x30000;
9955 vcpu->arch.msr_misc_features_enables = 0;
9957 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9960 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9961 vcpu->arch.regs_avail = ~0;
9962 vcpu->arch.regs_dirty = ~0;
9964 vcpu->arch.ia32_xss = 0;
9966 kvm_x86_ops.vcpu_reset(vcpu, init_event);
9969 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9971 struct kvm_segment cs;
9973 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9974 cs.selector = vector << 8;
9975 cs.base = vector << 12;
9976 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9977 kvm_rip_write(vcpu, 0);
9980 int kvm_arch_hardware_enable(void)
9983 struct kvm_vcpu *vcpu;
9988 bool stable, backwards_tsc = false;
9990 kvm_user_return_msr_cpu_online();
9991 ret = kvm_x86_ops.hardware_enable();
9995 local_tsc = rdtsc();
9996 stable = !kvm_check_tsc_unstable();
9997 list_for_each_entry(kvm, &vm_list, vm_list) {
9998 kvm_for_each_vcpu(i, vcpu, kvm) {
9999 if (!stable && vcpu->cpu == smp_processor_id())
10000 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10001 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10002 backwards_tsc = true;
10003 if (vcpu->arch.last_host_tsc > max_tsc)
10004 max_tsc = vcpu->arch.last_host_tsc;
10010 * Sometimes, even reliable TSCs go backwards. This happens on
10011 * platforms that reset TSC during suspend or hibernate actions, but
10012 * maintain synchronization. We must compensate. Fortunately, we can
10013 * detect that condition here, which happens early in CPU bringup,
10014 * before any KVM threads can be running. Unfortunately, we can't
10015 * bring the TSCs fully up to date with real time, as we aren't yet far
10016 * enough into CPU bringup that we know how much real time has actually
10017 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10018 * variables that haven't been updated yet.
10020 * So we simply find the maximum observed TSC above, then record the
10021 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10022 * the adjustment will be applied. Note that we accumulate
10023 * adjustments, in case multiple suspend cycles happen before some VCPU
10024 * gets a chance to run again. In the event that no KVM threads get a
10025 * chance to run, we will miss the entire elapsed period, as we'll have
10026 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10027 * loose cycle time. This isn't too big a deal, since the loss will be
10028 * uniform across all VCPUs (not to mention the scenario is extremely
10029 * unlikely). It is possible that a second hibernate recovery happens
10030 * much faster than a first, causing the observed TSC here to be
10031 * smaller; this would require additional padding adjustment, which is
10032 * why we set last_host_tsc to the local tsc observed here.
10034 * N.B. - this code below runs only on platforms with reliable TSC,
10035 * as that is the only way backwards_tsc is set above. Also note
10036 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10037 * have the same delta_cyc adjustment applied if backwards_tsc
10038 * is detected. Note further, this adjustment is only done once,
10039 * as we reset last_host_tsc on all VCPUs to stop this from being
10040 * called multiple times (one for each physical CPU bringup).
10042 * Platforms with unreliable TSCs don't have to deal with this, they
10043 * will be compensated by the logic in vcpu_load, which sets the TSC to
10044 * catchup mode. This will catchup all VCPUs to real time, but cannot
10045 * guarantee that they stay in perfect synchronization.
10047 if (backwards_tsc) {
10048 u64 delta_cyc = max_tsc - local_tsc;
10049 list_for_each_entry(kvm, &vm_list, vm_list) {
10050 kvm->arch.backwards_tsc_observed = true;
10051 kvm_for_each_vcpu(i, vcpu, kvm) {
10052 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10053 vcpu->arch.last_host_tsc = local_tsc;
10054 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10058 * We have to disable TSC offset matching.. if you were
10059 * booting a VM while issuing an S4 host suspend....
10060 * you may have some problem. Solving this issue is
10061 * left as an exercise to the reader.
10063 kvm->arch.last_tsc_nsec = 0;
10064 kvm->arch.last_tsc_write = 0;
10071 void kvm_arch_hardware_disable(void)
10073 kvm_x86_ops.hardware_disable();
10074 drop_user_return_notifiers();
10077 int kvm_arch_hardware_setup(void *opaque)
10079 struct kvm_x86_init_ops *ops = opaque;
10082 rdmsrl_safe(MSR_EFER, &host_efer);
10084 if (boot_cpu_has(X86_FEATURE_XSAVES))
10085 rdmsrl(MSR_IA32_XSS, host_xss);
10087 r = ops->hardware_setup();
10091 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10093 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10096 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10097 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10098 #undef __kvm_cpu_cap_has
10100 if (kvm_has_tsc_control) {
10102 * Make sure the user can only configure tsc_khz values that
10103 * fit into a signed integer.
10104 * A min value is not calculated because it will always
10105 * be 1 on all machines.
10107 u64 max = min(0x7fffffffULL,
10108 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10109 kvm_max_guest_tsc_khz = max;
10111 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10114 kvm_init_msr_list();
10118 void kvm_arch_hardware_unsetup(void)
10120 kvm_x86_ops.hardware_unsetup();
10123 int kvm_arch_check_processor_compat(void *opaque)
10125 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10126 struct kvm_x86_init_ops *ops = opaque;
10128 WARN_ON(!irqs_disabled());
10130 if (__cr4_reserved_bits(cpu_has, c) !=
10131 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10134 return ops->check_processor_compatibility();
10137 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10139 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10141 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10143 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10145 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10148 struct static_key kvm_no_apic_vcpu __read_mostly;
10149 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
10151 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10153 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10155 vcpu->arch.l1tf_flush_l1d = true;
10156 if (pmu->version && unlikely(pmu->event_count)) {
10157 pmu->need_cleanup = true;
10158 kvm_make_request(KVM_REQ_PMU, vcpu);
10160 kvm_x86_ops.sched_in(vcpu, cpu);
10163 void kvm_arch_free_vm(struct kvm *kvm)
10165 kfree(kvm->arch.hyperv.hv_pa_pg);
10170 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10175 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10176 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10177 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10178 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10179 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10180 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10182 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10183 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10184 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10185 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10186 &kvm->arch.irq_sources_bitmap);
10188 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10189 mutex_init(&kvm->arch.apic_map_lock);
10190 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10192 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10193 pvclock_update_vm_gtod_copy(kvm);
10195 kvm->arch.guest_can_read_msr_platform_info = true;
10197 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10198 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10200 kvm_hv_init_vm(kvm);
10201 kvm_page_track_init(kvm);
10202 kvm_mmu_init_vm(kvm);
10204 return kvm_x86_ops.vm_init(kvm);
10207 int kvm_arch_post_init_vm(struct kvm *kvm)
10209 return kvm_mmu_post_init_vm(kvm);
10212 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10215 kvm_mmu_unload(vcpu);
10219 static void kvm_free_vcpus(struct kvm *kvm)
10222 struct kvm_vcpu *vcpu;
10225 * Unpin any mmu pages first.
10227 kvm_for_each_vcpu(i, vcpu, kvm) {
10228 kvm_clear_async_pf_completion_queue(vcpu);
10229 kvm_unload_vcpu_mmu(vcpu);
10231 kvm_for_each_vcpu(i, vcpu, kvm)
10232 kvm_vcpu_destroy(vcpu);
10234 mutex_lock(&kvm->lock);
10235 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10236 kvm->vcpus[i] = NULL;
10238 atomic_set(&kvm->online_vcpus, 0);
10239 mutex_unlock(&kvm->lock);
10242 void kvm_arch_sync_events(struct kvm *kvm)
10244 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10245 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10249 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
10252 unsigned long hva, old_npages;
10253 struct kvm_memslots *slots = kvm_memslots(kvm);
10254 struct kvm_memory_slot *slot;
10256 /* Called with kvm->slots_lock held. */
10257 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10260 slot = id_to_memslot(slots, id);
10262 if (slot && slot->npages)
10266 * MAP_SHARED to prevent internal slot pages from being moved
10269 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10270 MAP_SHARED | MAP_ANONYMOUS, 0);
10271 if (IS_ERR((void *)hva))
10272 return PTR_ERR((void *)hva);
10274 if (!slot || !slot->npages)
10277 old_npages = slot->npages;
10281 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10282 struct kvm_userspace_memory_region m;
10284 m.slot = id | (i << 16);
10286 m.guest_phys_addr = gpa;
10287 m.userspace_addr = hva;
10288 m.memory_size = size;
10289 r = __kvm_set_memory_region(kvm, &m);
10295 vm_munmap(hva, old_npages * PAGE_SIZE);
10299 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10301 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10303 kvm_mmu_pre_destroy_vm(kvm);
10306 void kvm_arch_destroy_vm(struct kvm *kvm)
10310 if (current->mm == kvm->mm) {
10312 * Free memory regions allocated on behalf of userspace,
10313 * unless the the memory map has changed due to process exit
10316 mutex_lock(&kvm->slots_lock);
10317 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10319 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10321 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10322 mutex_unlock(&kvm->slots_lock);
10324 if (kvm_x86_ops.vm_destroy)
10325 kvm_x86_ops.vm_destroy(kvm);
10326 for (i = 0; i < kvm->arch.msr_filter.count; i++)
10327 kfree(kvm->arch.msr_filter.ranges[i].bitmap);
10328 kvm_pic_destroy(kvm);
10329 kvm_ioapic_destroy(kvm);
10330 kvm_free_vcpus(kvm);
10331 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10332 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10333 kvm_mmu_uninit_vm(kvm);
10334 kvm_page_track_cleanup(kvm);
10335 kvm_hv_destroy_vm(kvm);
10338 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10342 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10343 kvfree(slot->arch.rmap[i]);
10344 slot->arch.rmap[i] = NULL;
10349 kvfree(slot->arch.lpage_info[i - 1]);
10350 slot->arch.lpage_info[i - 1] = NULL;
10353 kvm_page_track_free_memslot(slot);
10356 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10357 unsigned long npages)
10362 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10363 * old arrays will be freed by __kvm_set_memory_region() if installing
10364 * the new memslot is successful.
10366 memset(&slot->arch, 0, sizeof(slot->arch));
10368 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10369 struct kvm_lpage_info *linfo;
10370 unsigned long ugfn;
10374 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10375 slot->base_gfn, level) + 1;
10377 slot->arch.rmap[i] =
10378 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10379 GFP_KERNEL_ACCOUNT);
10380 if (!slot->arch.rmap[i])
10385 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10389 slot->arch.lpage_info[i - 1] = linfo;
10391 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10392 linfo[0].disallow_lpage = 1;
10393 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10394 linfo[lpages - 1].disallow_lpage = 1;
10395 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10397 * If the gfn and userspace address are not aligned wrt each
10398 * other, disable large page support for this slot.
10400 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10403 for (j = 0; j < lpages; ++j)
10404 linfo[j].disallow_lpage = 1;
10408 if (kvm_page_track_create_memslot(slot, npages))
10414 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10415 kvfree(slot->arch.rmap[i]);
10416 slot->arch.rmap[i] = NULL;
10420 kvfree(slot->arch.lpage_info[i - 1]);
10421 slot->arch.lpage_info[i - 1] = NULL;
10426 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10428 struct kvm_vcpu *vcpu;
10432 * memslots->generation has been incremented.
10433 * mmio generation may have reached its maximum value.
10435 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10437 /* Force re-initialization of steal_time cache */
10438 kvm_for_each_vcpu(i, vcpu, kvm)
10439 kvm_vcpu_kick(vcpu);
10442 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10443 struct kvm_memory_slot *memslot,
10444 const struct kvm_userspace_memory_region *mem,
10445 enum kvm_mr_change change)
10447 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10448 return kvm_alloc_memslot_metadata(memslot,
10449 mem->memory_size >> PAGE_SHIFT);
10453 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10454 struct kvm_memory_slot *old,
10455 struct kvm_memory_slot *new,
10456 enum kvm_mr_change change)
10459 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
10460 * See comments below.
10462 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10466 * Dirty logging tracks sptes in 4k granularity, meaning that large
10467 * sptes have to be split. If live migration is successful, the guest
10468 * in the source machine will be destroyed and large sptes will be
10469 * created in the destination. However, if the guest continues to run
10470 * in the source machine (for example if live migration fails), small
10471 * sptes will remain around and cause bad performance.
10473 * Scan sptes if dirty logging has been stopped, dropping those
10474 * which can be collapsed into a single large-page spte. Later
10475 * page faults will create the large-page sptes.
10477 * There is no need to do this in any of the following cases:
10478 * CREATE: No dirty mappings will already exist.
10479 * MOVE/DELETE: The old mappings will already have been cleaned up by
10480 * kvm_arch_flush_shadow_memslot()
10482 if ((old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10483 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10484 kvm_mmu_zap_collapsible_sptes(kvm, new);
10487 * Enable or disable dirty logging for the slot.
10489 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old
10490 * slot have been zapped so no dirty logging updates are needed for
10492 * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible
10493 * any mappings that might be created in it will consume the
10494 * properties of the new slot and do not need to be updated here.
10496 * When PML is enabled, the kvm_x86_ops dirty logging hooks are
10497 * called to enable/disable dirty logging.
10499 * When disabling dirty logging with PML enabled, the D-bit is set
10500 * for sptes in the slot in order to prevent unnecessary GPA
10501 * logging in the PML buffer (and potential PML buffer full VMEXIT).
10502 * This guarantees leaving PML enabled for the guest's lifetime
10503 * won't have any additional overhead from PML when the guest is
10504 * running with dirty logging disabled.
10506 * When enabling dirty logging, large sptes are write-protected
10507 * so they can be split on first write. New large sptes cannot
10508 * be created for this slot until the end of the logging.
10509 * See the comments in fast_page_fault().
10510 * For small sptes, nothing is done if the dirty log is in the
10511 * initial-all-set state. Otherwise, depending on whether pml
10512 * is enabled the D-bit or the W-bit will be cleared.
10514 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10515 if (kvm_x86_ops.slot_enable_log_dirty) {
10516 kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10519 kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10520 PG_LEVEL_2M : PG_LEVEL_4K;
10523 * If we're with initial-all-set, we don't need
10524 * to write protect any small page because
10525 * they're reported as dirty already. However
10526 * we still need to write-protect huge pages
10527 * so that the page split can happen lazily on
10528 * the first write to the huge page.
10530 kvm_mmu_slot_remove_write_access(kvm, new, level);
10533 if (kvm_x86_ops.slot_disable_log_dirty)
10534 kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10538 void kvm_arch_commit_memory_region(struct kvm *kvm,
10539 const struct kvm_userspace_memory_region *mem,
10540 struct kvm_memory_slot *old,
10541 const struct kvm_memory_slot *new,
10542 enum kvm_mr_change change)
10544 if (!kvm->arch.n_requested_mmu_pages)
10545 kvm_mmu_change_mmu_pages(kvm,
10546 kvm_mmu_calculate_default_mmu_pages(kvm));
10549 * FIXME: const-ify all uses of struct kvm_memory_slot.
10551 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
10553 /* Free the arrays associated with the old memslot. */
10554 if (change == KVM_MR_MOVE)
10555 kvm_arch_free_memslot(kvm, old);
10558 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10560 kvm_mmu_zap_all(kvm);
10563 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10564 struct kvm_memory_slot *slot)
10566 kvm_page_track_flush_slot(kvm, slot);
10569 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10571 return (is_guest_mode(vcpu) &&
10572 kvm_x86_ops.guest_apic_has_interrupt &&
10573 kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10576 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10578 if (!list_empty_careful(&vcpu->async_pf.done))
10581 if (kvm_apic_has_events(vcpu))
10584 if (vcpu->arch.pv.pv_unhalted)
10587 if (vcpu->arch.exception.pending)
10590 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10591 (vcpu->arch.nmi_pending &&
10592 kvm_x86_ops.nmi_allowed(vcpu, false)))
10595 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10596 (vcpu->arch.smi_pending &&
10597 kvm_x86_ops.smi_allowed(vcpu, false)))
10600 if (kvm_arch_interrupt_allowed(vcpu) &&
10601 (kvm_cpu_has_interrupt(vcpu) ||
10602 kvm_guest_apic_has_interrupt(vcpu)))
10605 if (kvm_hv_has_stimer_pending(vcpu))
10608 if (is_guest_mode(vcpu) &&
10609 kvm_x86_ops.nested_ops->hv_timer_pending &&
10610 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10616 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10618 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10621 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10623 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10626 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10627 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10628 kvm_test_request(KVM_REQ_EVENT, vcpu))
10631 if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10637 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10639 return vcpu->arch.preempted_in_kernel;
10642 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10644 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10647 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10649 return kvm_x86_ops.interrupt_allowed(vcpu, false);
10652 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10654 if (is_64_bit_mode(vcpu))
10655 return kvm_rip_read(vcpu);
10656 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10657 kvm_rip_read(vcpu));
10659 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10661 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10663 return kvm_get_linear_rip(vcpu) == linear_rip;
10665 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10667 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10669 unsigned long rflags;
10671 rflags = kvm_x86_ops.get_rflags(vcpu);
10672 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10673 rflags &= ~X86_EFLAGS_TF;
10676 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10678 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10680 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10681 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10682 rflags |= X86_EFLAGS_TF;
10683 kvm_x86_ops.set_rflags(vcpu, rflags);
10686 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10688 __kvm_set_rflags(vcpu, rflags);
10689 kvm_make_request(KVM_REQ_EVENT, vcpu);
10691 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10693 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10697 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10701 r = kvm_mmu_reload(vcpu);
10705 if (!vcpu->arch.mmu->direct_map &&
10706 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10709 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10712 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10714 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
10716 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10719 static inline u32 kvm_async_pf_next_probe(u32 key)
10721 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
10724 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10726 u32 key = kvm_async_pf_hash_fn(gfn);
10728 while (vcpu->arch.apf.gfns[key] != ~0)
10729 key = kvm_async_pf_next_probe(key);
10731 vcpu->arch.apf.gfns[key] = gfn;
10734 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10737 u32 key = kvm_async_pf_hash_fn(gfn);
10739 for (i = 0; i < ASYNC_PF_PER_VCPU &&
10740 (vcpu->arch.apf.gfns[key] != gfn &&
10741 vcpu->arch.apf.gfns[key] != ~0); i++)
10742 key = kvm_async_pf_next_probe(key);
10747 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10749 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10752 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10756 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10758 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
10762 vcpu->arch.apf.gfns[i] = ~0;
10764 j = kvm_async_pf_next_probe(j);
10765 if (vcpu->arch.apf.gfns[j] == ~0)
10767 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10769 * k lies cyclically in ]i,j]
10771 * |....j i.k.| or |.k..j i...|
10773 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10774 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10779 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
10781 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
10783 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
10787 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
10789 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
10791 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
10792 &token, offset, sizeof(token));
10795 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
10797 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
10800 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
10801 &val, offset, sizeof(val)))
10807 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10809 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10812 if (!kvm_pv_async_pf_enabled(vcpu) ||
10813 (vcpu->arch.apf.send_user_only && kvm_x86_ops.get_cpl(vcpu) == 0))
10819 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10821 if (unlikely(!lapic_in_kernel(vcpu) ||
10822 kvm_event_needs_reinjection(vcpu) ||
10823 vcpu->arch.exception.pending))
10826 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10830 * If interrupts are off we cannot even use an artificial
10833 return kvm_arch_interrupt_allowed(vcpu);
10836 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10837 struct kvm_async_pf *work)
10839 struct x86_exception fault;
10841 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10842 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10844 if (kvm_can_deliver_async_pf(vcpu) &&
10845 !apf_put_user_notpresent(vcpu)) {
10846 fault.vector = PF_VECTOR;
10847 fault.error_code_valid = true;
10848 fault.error_code = 0;
10849 fault.nested_page_fault = false;
10850 fault.address = work->arch.token;
10851 fault.async_page_fault = true;
10852 kvm_inject_page_fault(vcpu, &fault);
10856 * It is not possible to deliver a paravirtualized asynchronous
10857 * page fault, but putting the guest in an artificial halt state
10858 * can be beneficial nevertheless: if an interrupt arrives, we
10859 * can deliver it timely and perhaps the guest will schedule
10860 * another process. When the instruction that triggered a page
10861 * fault is retried, hopefully the page will be ready in the host.
10863 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10868 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10869 struct kvm_async_pf *work)
10871 struct kvm_lapic_irq irq = {
10872 .delivery_mode = APIC_DM_FIXED,
10873 .vector = vcpu->arch.apf.vec
10876 if (work->wakeup_all)
10877 work->arch.token = ~0; /* broadcast wakeup */
10879 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10880 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10882 if ((work->wakeup_all || work->notpresent_injected) &&
10883 kvm_pv_async_pf_enabled(vcpu) &&
10884 !apf_put_user_ready(vcpu, work->arch.token)) {
10885 vcpu->arch.apf.pageready_pending = true;
10886 kvm_apic_set_irq(vcpu, &irq, NULL);
10889 vcpu->arch.apf.halted = false;
10890 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10893 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
10895 kvm_make_request(KVM_REQ_APF_READY, vcpu);
10896 if (!vcpu->arch.apf.pageready_pending)
10897 kvm_vcpu_kick(vcpu);
10900 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
10902 if (!kvm_pv_async_pf_enabled(vcpu))
10905 return apf_pageready_slot_free(vcpu);
10908 void kvm_arch_start_assignment(struct kvm *kvm)
10910 atomic_inc(&kvm->arch.assigned_device_count);
10912 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10914 void kvm_arch_end_assignment(struct kvm *kvm)
10916 atomic_dec(&kvm->arch.assigned_device_count);
10918 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10920 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10922 return atomic_read(&kvm->arch.assigned_device_count);
10924 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10926 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10928 atomic_inc(&kvm->arch.noncoherent_dma_count);
10930 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10932 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10934 atomic_dec(&kvm->arch.noncoherent_dma_count);
10936 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10938 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10940 return atomic_read(&kvm->arch.noncoherent_dma_count);
10942 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10944 bool kvm_arch_has_irq_bypass(void)
10949 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10950 struct irq_bypass_producer *prod)
10952 struct kvm_kernel_irqfd *irqfd =
10953 container_of(cons, struct kvm_kernel_irqfd, consumer);
10956 irqfd->producer = prod;
10957 kvm_arch_start_assignment(irqfd->kvm);
10958 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm,
10959 prod->irq, irqfd->gsi, 1);
10962 kvm_arch_end_assignment(irqfd->kvm);
10967 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10968 struct irq_bypass_producer *prod)
10971 struct kvm_kernel_irqfd *irqfd =
10972 container_of(cons, struct kvm_kernel_irqfd, consumer);
10974 WARN_ON(irqfd->producer != prod);
10975 irqfd->producer = NULL;
10978 * When producer of consumer is unregistered, we change back to
10979 * remapped mode, so we can re-use the current implementation
10980 * when the irq is masked/disabled or the consumer side (KVM
10981 * int this case doesn't want to receive the interrupts.
10983 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10985 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10986 " fails: %d\n", irqfd->consumer.token, ret);
10988 kvm_arch_end_assignment(irqfd->kvm);
10991 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10992 uint32_t guest_irq, bool set)
10994 return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
10997 bool kvm_vector_hashing_enabled(void)
10999 return vector_hashing;
11002 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11004 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11006 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11009 int kvm_spec_ctrl_test_value(u64 value)
11012 * test that setting IA32_SPEC_CTRL to given value
11013 * is allowed by the host processor
11017 unsigned long flags;
11020 local_irq_save(flags);
11022 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11024 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11027 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11029 local_irq_restore(flags);
11033 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11035 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11037 struct x86_exception fault;
11038 u32 access = error_code &
11039 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11041 if (!(error_code & PFERR_PRESENT_MASK) ||
11042 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11044 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11045 * tables probably do not match the TLB. Just proceed
11046 * with the error code that the processor gave.
11048 fault.vector = PF_VECTOR;
11049 fault.error_code_valid = true;
11050 fault.error_code = error_code;
11051 fault.nested_page_fault = false;
11052 fault.address = gva;
11054 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11056 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11059 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11060 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11061 * indicates whether exit to userspace is needed.
11063 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11064 struct x86_exception *e)
11066 if (r == X86EMUL_PROPAGATE_FAULT) {
11067 kvm_inject_emulated_page_fault(vcpu, e);
11072 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11073 * while handling a VMX instruction KVM could've handled the request
11074 * correctly by exiting to userspace and performing I/O but there
11075 * doesn't seem to be a real use-case behind such requests, just return
11076 * KVM_EXIT_INTERNAL_ERROR for now.
11078 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11079 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11080 vcpu->run->internal.ndata = 0;
11084 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11086 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11089 struct x86_exception e;
11091 unsigned long roots_to_free = 0;
11098 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11099 if (r != X86EMUL_CONTINUE)
11100 return kvm_handle_memory_failure(vcpu, r, &e);
11102 if (operand.pcid >> 12 != 0) {
11103 kvm_inject_gp(vcpu, 0);
11107 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11110 case INVPCID_TYPE_INDIV_ADDR:
11111 if ((!pcid_enabled && (operand.pcid != 0)) ||
11112 is_noncanonical_address(operand.gla, vcpu)) {
11113 kvm_inject_gp(vcpu, 0);
11116 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11117 return kvm_skip_emulated_instruction(vcpu);
11119 case INVPCID_TYPE_SINGLE_CTXT:
11120 if (!pcid_enabled && (operand.pcid != 0)) {
11121 kvm_inject_gp(vcpu, 0);
11125 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11126 kvm_mmu_sync_roots(vcpu);
11127 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11130 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11131 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11133 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11135 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11137 * If neither the current cr3 nor any of the prev_roots use the
11138 * given PCID, then nothing needs to be done here because a
11139 * resync will happen anyway before switching to any other CR3.
11142 return kvm_skip_emulated_instruction(vcpu);
11144 case INVPCID_TYPE_ALL_NON_GLOBAL:
11146 * Currently, KVM doesn't mark global entries in the shadow
11147 * page tables, so a non-global flush just degenerates to a
11148 * global flush. If needed, we could optimize this later by
11149 * keeping track of global entries in shadow page tables.
11153 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11154 kvm_mmu_unload(vcpu);
11155 return kvm_skip_emulated_instruction(vcpu);
11158 BUG(); /* We have already checked above that type <= 3 */
11161 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11163 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11164 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11165 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11166 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11167 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11168 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11169 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11170 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11171 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11172 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11173 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11174 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11175 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11176 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11177 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11178 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11179 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11180 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11181 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11182 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11183 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11184 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);