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 process_smi(struct kvm_vcpu *vcpu);
109 static void enter_smm(struct kvm_vcpu *vcpu);
110 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
111 static void store_regs(struct kvm_vcpu *vcpu);
112 static int sync_regs(struct kvm_vcpu *vcpu);
114 struct kvm_x86_ops kvm_x86_ops __read_mostly;
115 EXPORT_SYMBOL_GPL(kvm_x86_ops);
117 static bool __read_mostly ignore_msrs = 0;
118 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
120 static bool __read_mostly report_ignored_msrs = true;
121 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
123 unsigned int min_timer_period_us = 200;
124 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
126 static bool __read_mostly kvmclock_periodic_sync = true;
127 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
129 bool __read_mostly kvm_has_tsc_control;
130 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
131 u32 __read_mostly kvm_max_guest_tsc_khz;
132 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
133 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
134 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
135 u64 __read_mostly kvm_max_tsc_scaling_ratio;
136 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
137 u64 __read_mostly kvm_default_tsc_scaling_ratio;
138 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
140 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
141 static u32 __read_mostly tsc_tolerance_ppm = 250;
142 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
145 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
146 * adaptive tuning starting from default advancment of 1000ns. '0' disables
147 * advancement entirely. Any other value is used as-is and disables adaptive
148 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
150 static int __read_mostly lapic_timer_advance_ns = -1;
151 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
153 static bool __read_mostly vector_hashing = true;
154 module_param(vector_hashing, bool, S_IRUGO);
156 bool __read_mostly enable_vmware_backdoor = false;
157 module_param(enable_vmware_backdoor, bool, S_IRUGO);
158 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
160 static bool __read_mostly force_emulation_prefix = false;
161 module_param(force_emulation_prefix, bool, S_IRUGO);
163 int __read_mostly pi_inject_timer = -1;
164 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
167 * Restoring the host value for MSRs that are only consumed when running in
168 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
169 * returns to userspace, i.e. the kernel can run with the guest's value.
171 #define KVM_MAX_NR_USER_RETURN_MSRS 16
173 struct kvm_user_return_msrs_global {
175 u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS];
178 struct kvm_user_return_msrs {
179 struct user_return_notifier urn;
181 struct kvm_user_return_msr_values {
184 } values[KVM_MAX_NR_USER_RETURN_MSRS];
187 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global;
188 static struct kvm_user_return_msrs __percpu *user_return_msrs;
190 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
191 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
192 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
193 | XFEATURE_MASK_PKRU)
195 u64 __read_mostly host_efer;
196 EXPORT_SYMBOL_GPL(host_efer);
198 bool __read_mostly allow_smaller_maxphyaddr = 0;
199 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
201 u64 __read_mostly host_xss;
202 EXPORT_SYMBOL_GPL(host_xss);
203 u64 __read_mostly supported_xss;
204 EXPORT_SYMBOL_GPL(supported_xss);
206 struct kvm_stats_debugfs_item debugfs_entries[] = {
207 VCPU_STAT("pf_fixed", pf_fixed),
208 VCPU_STAT("pf_guest", pf_guest),
209 VCPU_STAT("tlb_flush", tlb_flush),
210 VCPU_STAT("invlpg", invlpg),
211 VCPU_STAT("exits", exits),
212 VCPU_STAT("io_exits", io_exits),
213 VCPU_STAT("mmio_exits", mmio_exits),
214 VCPU_STAT("signal_exits", signal_exits),
215 VCPU_STAT("irq_window", irq_window_exits),
216 VCPU_STAT("nmi_window", nmi_window_exits),
217 VCPU_STAT("halt_exits", halt_exits),
218 VCPU_STAT("halt_successful_poll", halt_successful_poll),
219 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
220 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
221 VCPU_STAT("halt_wakeup", halt_wakeup),
222 VCPU_STAT("hypercalls", hypercalls),
223 VCPU_STAT("request_irq", request_irq_exits),
224 VCPU_STAT("irq_exits", irq_exits),
225 VCPU_STAT("host_state_reload", host_state_reload),
226 VCPU_STAT("fpu_reload", fpu_reload),
227 VCPU_STAT("insn_emulation", insn_emulation),
228 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
229 VCPU_STAT("irq_injections", irq_injections),
230 VCPU_STAT("nmi_injections", nmi_injections),
231 VCPU_STAT("req_event", req_event),
232 VCPU_STAT("l1d_flush", l1d_flush),
233 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
234 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
235 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
236 VM_STAT("mmu_pte_write", mmu_pte_write),
237 VM_STAT("mmu_pte_updated", mmu_pte_updated),
238 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
239 VM_STAT("mmu_flooded", mmu_flooded),
240 VM_STAT("mmu_recycled", mmu_recycled),
241 VM_STAT("mmu_cache_miss", mmu_cache_miss),
242 VM_STAT("mmu_unsync", mmu_unsync),
243 VM_STAT("remote_tlb_flush", remote_tlb_flush),
244 VM_STAT("largepages", lpages, .mode = 0444),
245 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
246 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
250 u64 __read_mostly host_xcr0;
251 u64 __read_mostly supported_xcr0;
252 EXPORT_SYMBOL_GPL(supported_xcr0);
254 static struct kmem_cache *x86_fpu_cache;
256 static struct kmem_cache *x86_emulator_cache;
259 * When called, it means the previous get/set msr reached an invalid msr.
260 * Return true if we want to ignore/silent this failed msr access.
262 static bool kvm_msr_ignored_check(struct kvm_vcpu *vcpu, u32 msr,
263 u64 data, bool write)
265 const char *op = write ? "wrmsr" : "rdmsr";
268 if (report_ignored_msrs)
269 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
274 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
280 static struct kmem_cache *kvm_alloc_emulator_cache(void)
282 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
283 unsigned int size = sizeof(struct x86_emulate_ctxt);
285 return kmem_cache_create_usercopy("x86_emulator", size,
286 __alignof__(struct x86_emulate_ctxt),
287 SLAB_ACCOUNT, useroffset,
288 size - useroffset, NULL);
291 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
293 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
296 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
297 vcpu->arch.apf.gfns[i] = ~0;
300 static void kvm_on_user_return(struct user_return_notifier *urn)
303 struct kvm_user_return_msrs *msrs
304 = container_of(urn, struct kvm_user_return_msrs, urn);
305 struct kvm_user_return_msr_values *values;
309 * Disabling irqs at this point since the following code could be
310 * interrupted and executed through kvm_arch_hardware_disable()
312 local_irq_save(flags);
313 if (msrs->registered) {
314 msrs->registered = false;
315 user_return_notifier_unregister(urn);
317 local_irq_restore(flags);
318 for (slot = 0; slot < user_return_msrs_global.nr; ++slot) {
319 values = &msrs->values[slot];
320 if (values->host != values->curr) {
321 wrmsrl(user_return_msrs_global.msrs[slot], values->host);
322 values->curr = values->host;
327 void kvm_define_user_return_msr(unsigned slot, u32 msr)
329 BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
330 user_return_msrs_global.msrs[slot] = msr;
331 if (slot >= user_return_msrs_global.nr)
332 user_return_msrs_global.nr = slot + 1;
334 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
336 static void kvm_user_return_msr_cpu_online(void)
338 unsigned int cpu = smp_processor_id();
339 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
343 for (i = 0; i < user_return_msrs_global.nr; ++i) {
344 rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
345 msrs->values[i].host = value;
346 msrs->values[i].curr = value;
350 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
352 unsigned int cpu = smp_processor_id();
353 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
356 value = (value & mask) | (msrs->values[slot].host & ~mask);
357 if (value == msrs->values[slot].curr)
359 err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
363 msrs->values[slot].curr = value;
364 if (!msrs->registered) {
365 msrs->urn.on_user_return = kvm_on_user_return;
366 user_return_notifier_register(&msrs->urn);
367 msrs->registered = true;
371 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
373 static void drop_user_return_notifiers(void)
375 unsigned int cpu = smp_processor_id();
376 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
378 if (msrs->registered)
379 kvm_on_user_return(&msrs->urn);
382 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
384 return vcpu->arch.apic_base;
386 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
388 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
390 return kvm_apic_mode(kvm_get_apic_base(vcpu));
392 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
394 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
396 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
397 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
398 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
399 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
401 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
403 if (!msr_info->host_initiated) {
404 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
406 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
410 kvm_lapic_set_base(vcpu, msr_info->data);
411 kvm_recalculate_apic_map(vcpu->kvm);
414 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
416 asmlinkage __visible noinstr void kvm_spurious_fault(void)
418 /* Fault while not rebooting. We want the trace. */
419 BUG_ON(!kvm_rebooting);
421 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
423 #define EXCPT_BENIGN 0
424 #define EXCPT_CONTRIBUTORY 1
427 static int exception_class(int vector)
437 return EXCPT_CONTRIBUTORY;
444 #define EXCPT_FAULT 0
446 #define EXCPT_ABORT 2
447 #define EXCPT_INTERRUPT 3
449 static int exception_type(int vector)
453 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
454 return EXCPT_INTERRUPT;
458 /* #DB is trap, as instruction watchpoints are handled elsewhere */
459 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
462 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
465 /* Reserved exceptions will result in fault */
469 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
471 unsigned nr = vcpu->arch.exception.nr;
472 bool has_payload = vcpu->arch.exception.has_payload;
473 unsigned long payload = vcpu->arch.exception.payload;
481 * "Certain debug exceptions may clear bit 0-3. The
482 * remaining contents of the DR6 register are never
483 * cleared by the processor".
485 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
487 * DR6.RTM is set by all #DB exceptions that don't clear it.
489 vcpu->arch.dr6 |= DR6_RTM;
490 vcpu->arch.dr6 |= payload;
492 * Bit 16 should be set in the payload whenever the #DB
493 * exception should clear DR6.RTM. This makes the payload
494 * compatible with the pending debug exceptions under VMX.
495 * Though not currently documented in the SDM, this also
496 * makes the payload compatible with the exit qualification
497 * for #DB exceptions under VMX.
499 vcpu->arch.dr6 ^= payload & DR6_RTM;
502 * The #DB payload is defined as compatible with the 'pending
503 * debug exceptions' field under VMX, not DR6. While bit 12 is
504 * defined in the 'pending debug exceptions' field (enabled
505 * breakpoint), it is reserved and must be zero in DR6.
507 vcpu->arch.dr6 &= ~BIT(12);
510 vcpu->arch.cr2 = payload;
514 vcpu->arch.exception.has_payload = false;
515 vcpu->arch.exception.payload = 0;
517 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
519 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
520 unsigned nr, bool has_error, u32 error_code,
521 bool has_payload, unsigned long payload, bool reinject)
526 kvm_make_request(KVM_REQ_EVENT, vcpu);
528 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
530 if (has_error && !is_protmode(vcpu))
534 * On vmentry, vcpu->arch.exception.pending is only
535 * true if an event injection was blocked by
536 * nested_run_pending. In that case, however,
537 * vcpu_enter_guest requests an immediate exit,
538 * and the guest shouldn't proceed far enough to
541 WARN_ON_ONCE(vcpu->arch.exception.pending);
542 vcpu->arch.exception.injected = true;
543 if (WARN_ON_ONCE(has_payload)) {
545 * A reinjected event has already
546 * delivered its payload.
552 vcpu->arch.exception.pending = true;
553 vcpu->arch.exception.injected = false;
555 vcpu->arch.exception.has_error_code = has_error;
556 vcpu->arch.exception.nr = nr;
557 vcpu->arch.exception.error_code = error_code;
558 vcpu->arch.exception.has_payload = has_payload;
559 vcpu->arch.exception.payload = payload;
560 if (!is_guest_mode(vcpu))
561 kvm_deliver_exception_payload(vcpu);
565 /* to check exception */
566 prev_nr = vcpu->arch.exception.nr;
567 if (prev_nr == DF_VECTOR) {
568 /* triple fault -> shutdown */
569 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
572 class1 = exception_class(prev_nr);
573 class2 = exception_class(nr);
574 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
575 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
577 * Generate double fault per SDM Table 5-5. Set
578 * exception.pending = true so that the double fault
579 * can trigger a nested vmexit.
581 vcpu->arch.exception.pending = true;
582 vcpu->arch.exception.injected = false;
583 vcpu->arch.exception.has_error_code = true;
584 vcpu->arch.exception.nr = DF_VECTOR;
585 vcpu->arch.exception.error_code = 0;
586 vcpu->arch.exception.has_payload = false;
587 vcpu->arch.exception.payload = 0;
589 /* replace previous exception with a new one in a hope
590 that instruction re-execution will regenerate lost
595 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
597 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
599 EXPORT_SYMBOL_GPL(kvm_queue_exception);
601 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
603 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
605 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
607 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
608 unsigned long payload)
610 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
612 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
614 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
615 u32 error_code, unsigned long payload)
617 kvm_multiple_exception(vcpu, nr, true, error_code,
618 true, payload, false);
621 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
624 kvm_inject_gp(vcpu, 0);
626 return kvm_skip_emulated_instruction(vcpu);
630 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
632 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
634 ++vcpu->stat.pf_guest;
635 vcpu->arch.exception.nested_apf =
636 is_guest_mode(vcpu) && fault->async_page_fault;
637 if (vcpu->arch.exception.nested_apf) {
638 vcpu->arch.apf.nested_apf_token = fault->address;
639 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
641 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
645 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
647 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
648 struct x86_exception *fault)
650 struct kvm_mmu *fault_mmu;
651 WARN_ON_ONCE(fault->vector != PF_VECTOR);
653 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
657 * Invalidate the TLB entry for the faulting address, if it exists,
658 * else the access will fault indefinitely (and to emulate hardware).
660 if ((fault->error_code & PFERR_PRESENT_MASK) &&
661 !(fault->error_code & PFERR_RSVD_MASK))
662 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
663 fault_mmu->root_hpa);
665 fault_mmu->inject_page_fault(vcpu, fault);
666 return fault->nested_page_fault;
668 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
670 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
672 atomic_inc(&vcpu->arch.nmi_queued);
673 kvm_make_request(KVM_REQ_NMI, vcpu);
675 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
677 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
679 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
681 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
683 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
685 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
687 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
690 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
691 * a #GP and return false.
693 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
695 if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
697 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
700 EXPORT_SYMBOL_GPL(kvm_require_cpl);
702 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
704 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
707 kvm_queue_exception(vcpu, UD_VECTOR);
710 EXPORT_SYMBOL_GPL(kvm_require_dr);
713 * This function will be used to read from the physical memory of the currently
714 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
715 * can read from guest physical or from the guest's guest physical memory.
717 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
718 gfn_t ngfn, void *data, int offset, int len,
721 struct x86_exception exception;
725 ngpa = gfn_to_gpa(ngfn);
726 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
727 if (real_gfn == UNMAPPED_GVA)
730 real_gfn = gpa_to_gfn(real_gfn);
732 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
734 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
736 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
737 void *data, int offset, int len, u32 access)
739 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
740 data, offset, len, access);
743 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
745 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
750 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
752 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
754 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
755 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
758 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
760 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
761 offset * sizeof(u64), sizeof(pdpte),
762 PFERR_USER_MASK|PFERR_WRITE_MASK);
767 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
768 if ((pdpte[i] & PT_PRESENT_MASK) &&
769 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
776 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
777 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
783 EXPORT_SYMBOL_GPL(load_pdptrs);
785 bool pdptrs_changed(struct kvm_vcpu *vcpu)
787 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
792 if (!is_pae_paging(vcpu))
795 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
798 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
799 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
800 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
801 PFERR_USER_MASK | PFERR_WRITE_MASK);
805 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
807 EXPORT_SYMBOL_GPL(pdptrs_changed);
809 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
811 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
813 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
814 kvm_clear_async_pf_completion_queue(vcpu);
815 kvm_async_pf_hash_reset(vcpu);
818 if ((cr0 ^ old_cr0) & update_bits)
819 kvm_mmu_reset_context(vcpu);
821 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
822 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
823 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
824 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
826 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
828 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
830 unsigned long old_cr0 = kvm_read_cr0(vcpu);
831 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
836 if (cr0 & 0xffffffff00000000UL)
840 cr0 &= ~CR0_RESERVED_BITS;
842 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
845 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
849 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
850 (cr0 & X86_CR0_PG)) {
855 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
860 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
861 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
862 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
865 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
868 kvm_x86_ops.set_cr0(vcpu, cr0);
870 kvm_post_set_cr0(vcpu, old_cr0, cr0);
874 EXPORT_SYMBOL_GPL(kvm_set_cr0);
876 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
878 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
880 EXPORT_SYMBOL_GPL(kvm_lmsw);
882 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
884 if (vcpu->arch.guest_state_protected)
887 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
889 if (vcpu->arch.xcr0 != host_xcr0)
890 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
892 if (vcpu->arch.xsaves_enabled &&
893 vcpu->arch.ia32_xss != host_xss)
894 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
897 if (static_cpu_has(X86_FEATURE_PKU) &&
898 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
899 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
900 vcpu->arch.pkru != vcpu->arch.host_pkru)
901 __write_pkru(vcpu->arch.pkru);
903 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
905 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
907 if (vcpu->arch.guest_state_protected)
910 if (static_cpu_has(X86_FEATURE_PKU) &&
911 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
912 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
913 vcpu->arch.pkru = rdpkru();
914 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
915 __write_pkru(vcpu->arch.host_pkru);
918 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
920 if (vcpu->arch.xcr0 != host_xcr0)
921 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
923 if (vcpu->arch.xsaves_enabled &&
924 vcpu->arch.ia32_xss != host_xss)
925 wrmsrl(MSR_IA32_XSS, host_xss);
929 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
931 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
934 u64 old_xcr0 = vcpu->arch.xcr0;
937 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
938 if (index != XCR_XFEATURE_ENABLED_MASK)
940 if (!(xcr0 & XFEATURE_MASK_FP))
942 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
946 * Do not allow the guest to set bits that we do not support
947 * saving. However, xcr0 bit 0 is always set, even if the
948 * emulated CPU does not support XSAVE (see fx_init).
950 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
951 if (xcr0 & ~valid_bits)
954 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
955 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
958 if (xcr0 & XFEATURE_MASK_AVX512) {
959 if (!(xcr0 & XFEATURE_MASK_YMM))
961 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
964 vcpu->arch.xcr0 = xcr0;
966 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
967 kvm_update_cpuid_runtime(vcpu);
971 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
973 if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
974 __kvm_set_xcr(vcpu, index, xcr)) {
975 kvm_inject_gp(vcpu, 0);
980 EXPORT_SYMBOL_GPL(kvm_set_xcr);
982 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
984 if (cr4 & cr4_reserved_bits)
987 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
990 return kvm_x86_ops.is_valid_cr4(vcpu, cr4);
992 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
994 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
996 unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
997 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
999 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1000 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1001 kvm_mmu_reset_context(vcpu);
1003 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1005 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1007 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1008 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1011 if (!kvm_is_valid_cr4(vcpu, cr4))
1014 if (is_long_mode(vcpu)) {
1015 if (!(cr4 & X86_CR4_PAE))
1017 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1019 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1020 && ((cr4 ^ old_cr4) & pdptr_bits)
1021 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1022 kvm_read_cr3(vcpu)))
1025 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1026 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1029 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1030 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1034 kvm_x86_ops.set_cr4(vcpu, cr4);
1036 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1040 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1042 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1044 bool skip_tlb_flush = false;
1045 #ifdef CONFIG_X86_64
1046 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1049 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1050 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1054 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1055 if (!skip_tlb_flush) {
1056 kvm_mmu_sync_roots(vcpu);
1057 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1062 if (is_long_mode(vcpu) &&
1063 (cr3 & vcpu->arch.cr3_lm_rsvd_bits))
1065 else if (is_pae_paging(vcpu) &&
1066 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1069 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1070 vcpu->arch.cr3 = cr3;
1071 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1075 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1077 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1079 if (cr8 & CR8_RESERVED_BITS)
1081 if (lapic_in_kernel(vcpu))
1082 kvm_lapic_set_tpr(vcpu, cr8);
1084 vcpu->arch.cr8 = cr8;
1087 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1089 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1091 if (lapic_in_kernel(vcpu))
1092 return kvm_lapic_get_cr8(vcpu);
1094 return vcpu->arch.cr8;
1096 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1098 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1102 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1103 for (i = 0; i < KVM_NR_DB_REGS; i++)
1104 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1105 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1109 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1113 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1114 dr7 = vcpu->arch.guest_debug_dr7;
1116 dr7 = vcpu->arch.dr7;
1117 kvm_x86_ops.set_dr7(vcpu, dr7);
1118 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1119 if (dr7 & DR7_BP_EN_MASK)
1120 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1122 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1124 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1126 u64 fixed = DR6_FIXED_1;
1128 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1133 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1135 size_t size = ARRAY_SIZE(vcpu->arch.db);
1139 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1140 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1141 vcpu->arch.eff_db[dr] = val;
1145 if (!kvm_dr6_valid(val))
1146 return -1; /* #GP */
1147 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1151 if (!kvm_dr7_valid(val))
1152 return -1; /* #GP */
1153 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1154 kvm_update_dr7(vcpu);
1161 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1163 if (__kvm_set_dr(vcpu, dr, val)) {
1164 kvm_inject_gp(vcpu, 0);
1169 EXPORT_SYMBOL_GPL(kvm_set_dr);
1171 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1173 size_t size = ARRAY_SIZE(vcpu->arch.db);
1177 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1181 *val = vcpu->arch.dr6;
1185 *val = vcpu->arch.dr7;
1190 EXPORT_SYMBOL_GPL(kvm_get_dr);
1192 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1194 u32 ecx = kvm_rcx_read(vcpu);
1198 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1201 kvm_rax_write(vcpu, (u32)data);
1202 kvm_rdx_write(vcpu, data >> 32);
1205 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1208 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1209 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1211 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1212 * extract the supported MSRs from the related const lists.
1213 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1214 * capabilities of the host cpu. This capabilities test skips MSRs that are
1215 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1216 * may depend on host virtualization features rather than host cpu features.
1219 static const u32 msrs_to_save_all[] = {
1220 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1222 #ifdef CONFIG_X86_64
1223 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1225 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1226 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1228 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1229 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1230 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1231 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1232 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1233 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1234 MSR_IA32_UMWAIT_CONTROL,
1236 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1237 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1238 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1239 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1240 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1241 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1242 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1243 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1244 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1245 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1246 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1247 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1248 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1249 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1250 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1251 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1252 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1253 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1254 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1255 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1256 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1257 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1260 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1261 static unsigned num_msrs_to_save;
1263 static const u32 emulated_msrs_all[] = {
1264 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1265 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1266 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1267 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1268 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1269 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1270 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1272 HV_X64_MSR_VP_INDEX,
1273 HV_X64_MSR_VP_RUNTIME,
1274 HV_X64_MSR_SCONTROL,
1275 HV_X64_MSR_STIMER0_CONFIG,
1276 HV_X64_MSR_VP_ASSIST_PAGE,
1277 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1278 HV_X64_MSR_TSC_EMULATION_STATUS,
1279 HV_X64_MSR_SYNDBG_OPTIONS,
1280 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1281 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1282 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1284 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1285 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1287 MSR_IA32_TSC_ADJUST,
1288 MSR_IA32_TSCDEADLINE,
1289 MSR_IA32_ARCH_CAPABILITIES,
1290 MSR_IA32_PERF_CAPABILITIES,
1291 MSR_IA32_MISC_ENABLE,
1292 MSR_IA32_MCG_STATUS,
1294 MSR_IA32_MCG_EXT_CTL,
1298 MSR_MISC_FEATURES_ENABLES,
1299 MSR_AMD64_VIRT_SPEC_CTRL,
1304 * The following list leaves out MSRs whose values are determined
1305 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1306 * We always support the "true" VMX control MSRs, even if the host
1307 * processor does not, so I am putting these registers here rather
1308 * than in msrs_to_save_all.
1311 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1312 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1313 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1314 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1316 MSR_IA32_VMX_CR0_FIXED0,
1317 MSR_IA32_VMX_CR4_FIXED0,
1318 MSR_IA32_VMX_VMCS_ENUM,
1319 MSR_IA32_VMX_PROCBASED_CTLS2,
1320 MSR_IA32_VMX_EPT_VPID_CAP,
1321 MSR_IA32_VMX_VMFUNC,
1324 MSR_KVM_POLL_CONTROL,
1327 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1328 static unsigned num_emulated_msrs;
1331 * List of msr numbers which are used to expose MSR-based features that
1332 * can be used by a hypervisor to validate requested CPU features.
1334 static const u32 msr_based_features_all[] = {
1336 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1337 MSR_IA32_VMX_PINBASED_CTLS,
1338 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1339 MSR_IA32_VMX_PROCBASED_CTLS,
1340 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1341 MSR_IA32_VMX_EXIT_CTLS,
1342 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1343 MSR_IA32_VMX_ENTRY_CTLS,
1345 MSR_IA32_VMX_CR0_FIXED0,
1346 MSR_IA32_VMX_CR0_FIXED1,
1347 MSR_IA32_VMX_CR4_FIXED0,
1348 MSR_IA32_VMX_CR4_FIXED1,
1349 MSR_IA32_VMX_VMCS_ENUM,
1350 MSR_IA32_VMX_PROCBASED_CTLS2,
1351 MSR_IA32_VMX_EPT_VPID_CAP,
1352 MSR_IA32_VMX_VMFUNC,
1356 MSR_IA32_ARCH_CAPABILITIES,
1357 MSR_IA32_PERF_CAPABILITIES,
1360 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1361 static unsigned int num_msr_based_features;
1363 static u64 kvm_get_arch_capabilities(void)
1367 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1368 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1371 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1372 * the nested hypervisor runs with NX huge pages. If it is not,
1373 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1374 * L1 guests, so it need not worry about its own (L2) guests.
1376 data |= ARCH_CAP_PSCHANGE_MC_NO;
1379 * If we're doing cache flushes (either "always" or "cond")
1380 * we will do one whenever the guest does a vmlaunch/vmresume.
1381 * If an outer hypervisor is doing the cache flush for us
1382 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1383 * capability to the guest too, and if EPT is disabled we're not
1384 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1385 * require a nested hypervisor to do a flush of its own.
1387 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1388 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1390 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1391 data |= ARCH_CAP_RDCL_NO;
1392 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1393 data |= ARCH_CAP_SSB_NO;
1394 if (!boot_cpu_has_bug(X86_BUG_MDS))
1395 data |= ARCH_CAP_MDS_NO;
1397 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1399 * If RTM=0 because the kernel has disabled TSX, the host might
1400 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1401 * and therefore knows that there cannot be TAA) but keep
1402 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1403 * and we want to allow migrating those guests to tsx=off hosts.
1405 data &= ~ARCH_CAP_TAA_NO;
1406 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1407 data |= ARCH_CAP_TAA_NO;
1410 * Nothing to do here; we emulate TSX_CTRL if present on the
1411 * host so the guest can choose between disabling TSX or
1412 * using VERW to clear CPU buffers.
1419 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1421 switch (msr->index) {
1422 case MSR_IA32_ARCH_CAPABILITIES:
1423 msr->data = kvm_get_arch_capabilities();
1425 case MSR_IA32_UCODE_REV:
1426 rdmsrl_safe(msr->index, &msr->data);
1429 return kvm_x86_ops.get_msr_feature(msr);
1434 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1436 struct kvm_msr_entry msr;
1440 r = kvm_get_msr_feature(&msr);
1442 if (r == KVM_MSR_RET_INVALID) {
1443 /* Unconditionally clear the output for simplicity */
1445 if (kvm_msr_ignored_check(vcpu, index, 0, false))
1457 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1459 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1462 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1465 if (efer & (EFER_LME | EFER_LMA) &&
1466 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1469 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1475 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1477 if (efer & efer_reserved_bits)
1480 return __kvm_valid_efer(vcpu, efer);
1482 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1484 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1486 u64 old_efer = vcpu->arch.efer;
1487 u64 efer = msr_info->data;
1490 if (efer & efer_reserved_bits)
1493 if (!msr_info->host_initiated) {
1494 if (!__kvm_valid_efer(vcpu, efer))
1497 if (is_paging(vcpu) &&
1498 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1503 efer |= vcpu->arch.efer & EFER_LMA;
1505 r = kvm_x86_ops.set_efer(vcpu, efer);
1511 /* Update reserved bits */
1512 if ((efer ^ old_efer) & EFER_NX)
1513 kvm_mmu_reset_context(vcpu);
1518 void kvm_enable_efer_bits(u64 mask)
1520 efer_reserved_bits &= ~mask;
1522 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1524 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1526 struct kvm *kvm = vcpu->kvm;
1527 struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
1528 u32 count = kvm->arch.msr_filter.count;
1530 bool r = kvm->arch.msr_filter.default_allow;
1533 /* MSR filtering not set up or x2APIC enabled, allow everything */
1534 if (!count || (index >= 0x800 && index <= 0x8ff))
1537 /* Prevent collision with set_msr_filter */
1538 idx = srcu_read_lock(&kvm->srcu);
1540 for (i = 0; i < count; i++) {
1541 u32 start = ranges[i].base;
1542 u32 end = start + ranges[i].nmsrs;
1543 u32 flags = ranges[i].flags;
1544 unsigned long *bitmap = ranges[i].bitmap;
1546 if ((index >= start) && (index < end) && (flags & type)) {
1547 r = !!test_bit(index - start, bitmap);
1552 srcu_read_unlock(&kvm->srcu, idx);
1556 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1559 * Write @data into the MSR specified by @index. Select MSR specific fault
1560 * checks are bypassed if @host_initiated is %true.
1561 * Returns 0 on success, non-0 otherwise.
1562 * Assumes vcpu_load() was already called.
1564 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1565 bool host_initiated)
1567 struct msr_data msr;
1569 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1570 return KVM_MSR_RET_FILTERED;
1575 case MSR_KERNEL_GS_BASE:
1578 if (is_noncanonical_address(data, vcpu))
1581 case MSR_IA32_SYSENTER_EIP:
1582 case MSR_IA32_SYSENTER_ESP:
1584 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1585 * non-canonical address is written on Intel but not on
1586 * AMD (which ignores the top 32-bits, because it does
1587 * not implement 64-bit SYSENTER).
1589 * 64-bit code should hence be able to write a non-canonical
1590 * value on AMD. Making the address canonical ensures that
1591 * vmentry does not fail on Intel after writing a non-canonical
1592 * value, and that something deterministic happens if the guest
1593 * invokes 64-bit SYSENTER.
1595 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1600 msr.host_initiated = host_initiated;
1602 return kvm_x86_ops.set_msr(vcpu, &msr);
1605 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1606 u32 index, u64 data, bool host_initiated)
1608 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1610 if (ret == KVM_MSR_RET_INVALID)
1611 if (kvm_msr_ignored_check(vcpu, index, data, true))
1618 * Read the MSR specified by @index into @data. Select MSR specific fault
1619 * checks are bypassed if @host_initiated is %true.
1620 * Returns 0 on success, non-0 otherwise.
1621 * Assumes vcpu_load() was already called.
1623 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1624 bool host_initiated)
1626 struct msr_data msr;
1629 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1630 return KVM_MSR_RET_FILTERED;
1633 msr.host_initiated = host_initiated;
1635 ret = kvm_x86_ops.get_msr(vcpu, &msr);
1641 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1642 u32 index, u64 *data, bool host_initiated)
1644 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1646 if (ret == KVM_MSR_RET_INVALID) {
1647 /* Unconditionally clear *data for simplicity */
1649 if (kvm_msr_ignored_check(vcpu, index, 0, false))
1656 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1658 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1660 EXPORT_SYMBOL_GPL(kvm_get_msr);
1662 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1664 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1666 EXPORT_SYMBOL_GPL(kvm_set_msr);
1668 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1670 int err = vcpu->run->msr.error;
1672 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1673 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1676 return kvm_x86_ops.complete_emulated_msr(vcpu, err);
1679 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1681 return kvm_x86_ops.complete_emulated_msr(vcpu, vcpu->run->msr.error);
1684 static u64 kvm_msr_reason(int r)
1687 case KVM_MSR_RET_INVALID:
1688 return KVM_MSR_EXIT_REASON_UNKNOWN;
1689 case KVM_MSR_RET_FILTERED:
1690 return KVM_MSR_EXIT_REASON_FILTER;
1692 return KVM_MSR_EXIT_REASON_INVAL;
1696 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1697 u32 exit_reason, u64 data,
1698 int (*completion)(struct kvm_vcpu *vcpu),
1701 u64 msr_reason = kvm_msr_reason(r);
1703 /* Check if the user wanted to know about this MSR fault */
1704 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1707 vcpu->run->exit_reason = exit_reason;
1708 vcpu->run->msr.error = 0;
1709 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1710 vcpu->run->msr.reason = msr_reason;
1711 vcpu->run->msr.index = index;
1712 vcpu->run->msr.data = data;
1713 vcpu->arch.complete_userspace_io = completion;
1718 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1720 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1721 complete_emulated_rdmsr, r);
1724 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1726 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1727 complete_emulated_wrmsr, r);
1730 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1732 u32 ecx = kvm_rcx_read(vcpu);
1736 r = kvm_get_msr(vcpu, ecx, &data);
1738 /* MSR read failed? See if we should ask user space */
1739 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1740 /* Bounce to user space */
1745 trace_kvm_msr_read(ecx, data);
1747 kvm_rax_write(vcpu, data & -1u);
1748 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1750 trace_kvm_msr_read_ex(ecx);
1753 return kvm_x86_ops.complete_emulated_msr(vcpu, r);
1755 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1757 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1759 u32 ecx = kvm_rcx_read(vcpu);
1760 u64 data = kvm_read_edx_eax(vcpu);
1763 r = kvm_set_msr(vcpu, ecx, data);
1765 /* MSR write failed? See if we should ask user space */
1766 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1767 /* Bounce to user space */
1770 /* Signal all other negative errors to userspace */
1775 trace_kvm_msr_write(ecx, data);
1777 trace_kvm_msr_write_ex(ecx, data);
1779 return kvm_x86_ops.complete_emulated_msr(vcpu, r);
1781 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1783 bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1785 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1786 xfer_to_guest_mode_work_pending();
1788 EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request);
1791 * The fast path for frequent and performance sensitive wrmsr emulation,
1792 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1793 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1794 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1795 * other cases which must be called after interrupts are enabled on the host.
1797 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1799 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1802 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1803 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1804 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1805 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1808 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1809 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1810 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1811 trace_kvm_apic_write(APIC_ICR, (u32)data);
1818 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1820 if (!kvm_can_use_hv_timer(vcpu))
1823 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1827 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1829 u32 msr = kvm_rcx_read(vcpu);
1831 fastpath_t ret = EXIT_FASTPATH_NONE;
1834 case APIC_BASE_MSR + (APIC_ICR >> 4):
1835 data = kvm_read_edx_eax(vcpu);
1836 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1837 kvm_skip_emulated_instruction(vcpu);
1838 ret = EXIT_FASTPATH_EXIT_HANDLED;
1841 case MSR_IA32_TSCDEADLINE:
1842 data = kvm_read_edx_eax(vcpu);
1843 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1844 kvm_skip_emulated_instruction(vcpu);
1845 ret = EXIT_FASTPATH_REENTER_GUEST;
1852 if (ret != EXIT_FASTPATH_NONE)
1853 trace_kvm_msr_write(msr, data);
1857 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1860 * Adapt set_msr() to msr_io()'s calling convention
1862 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1864 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1867 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1869 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1872 #ifdef CONFIG_X86_64
1873 struct pvclock_clock {
1883 struct pvclock_gtod_data {
1886 struct pvclock_clock clock; /* extract of a clocksource struct */
1887 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1893 static struct pvclock_gtod_data pvclock_gtod_data;
1895 static void update_pvclock_gtod(struct timekeeper *tk)
1897 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1899 write_seqcount_begin(&vdata->seq);
1901 /* copy pvclock gtod data */
1902 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1903 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1904 vdata->clock.mask = tk->tkr_mono.mask;
1905 vdata->clock.mult = tk->tkr_mono.mult;
1906 vdata->clock.shift = tk->tkr_mono.shift;
1907 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1908 vdata->clock.offset = tk->tkr_mono.base;
1910 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1911 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1912 vdata->raw_clock.mask = tk->tkr_raw.mask;
1913 vdata->raw_clock.mult = tk->tkr_raw.mult;
1914 vdata->raw_clock.shift = tk->tkr_raw.shift;
1915 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1916 vdata->raw_clock.offset = tk->tkr_raw.base;
1918 vdata->wall_time_sec = tk->xtime_sec;
1920 vdata->offs_boot = tk->offs_boot;
1922 write_seqcount_end(&vdata->seq);
1925 static s64 get_kvmclock_base_ns(void)
1927 /* Count up from boot time, but with the frequency of the raw clock. */
1928 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1931 static s64 get_kvmclock_base_ns(void)
1933 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1934 return ktime_get_boottime_ns();
1938 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1942 struct pvclock_wall_clock wc;
1945 kvm->arch.wall_clock = wall_clock;
1950 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1955 ++version; /* first time write, random junk */
1959 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1963 * The guest calculates current wall clock time by adding
1964 * system time (updated by kvm_guest_time_update below) to the
1965 * wall clock specified here. We do the reverse here.
1967 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1969 wc.nsec = do_div(wall_nsec, 1000000000);
1970 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1971 wc.version = version;
1973 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1976 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1979 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
1980 bool old_msr, bool host_initiated)
1982 struct kvm_arch *ka = &vcpu->kvm->arch;
1984 if (vcpu->vcpu_id == 0 && !host_initiated) {
1985 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
1986 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1988 ka->boot_vcpu_runs_old_kvmclock = old_msr;
1991 vcpu->arch.time = system_time;
1992 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
1994 /* we verify if the enable bit is set... */
1995 vcpu->arch.pv_time_enabled = false;
1996 if (!(system_time & 1))
1999 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2000 &vcpu->arch.pv_time, system_time & ~1ULL,
2001 sizeof(struct pvclock_vcpu_time_info)))
2002 vcpu->arch.pv_time_enabled = true;
2007 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2009 do_shl32_div32(dividend, divisor);
2013 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2014 s8 *pshift, u32 *pmultiplier)
2022 scaled64 = scaled_hz;
2023 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2028 tps32 = (uint32_t)tps64;
2029 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2030 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2038 *pmultiplier = div_frac(scaled64, tps32);
2041 #ifdef CONFIG_X86_64
2042 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2045 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2046 static unsigned long max_tsc_khz;
2048 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2050 u64 v = (u64)khz * (1000000 + ppm);
2055 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2059 /* Guest TSC same frequency as host TSC? */
2061 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2065 /* TSC scaling supported? */
2066 if (!kvm_has_tsc_control) {
2067 if (user_tsc_khz > tsc_khz) {
2068 vcpu->arch.tsc_catchup = 1;
2069 vcpu->arch.tsc_always_catchup = 1;
2072 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2077 /* TSC scaling required - calculate ratio */
2078 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2079 user_tsc_khz, tsc_khz);
2081 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2082 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2087 vcpu->arch.tsc_scaling_ratio = ratio;
2091 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2093 u32 thresh_lo, thresh_hi;
2094 int use_scaling = 0;
2096 /* tsc_khz can be zero if TSC calibration fails */
2097 if (user_tsc_khz == 0) {
2098 /* set tsc_scaling_ratio to a safe value */
2099 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2103 /* Compute a scale to convert nanoseconds in TSC cycles */
2104 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2105 &vcpu->arch.virtual_tsc_shift,
2106 &vcpu->arch.virtual_tsc_mult);
2107 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2110 * Compute the variation in TSC rate which is acceptable
2111 * within the range of tolerance and decide if the
2112 * rate being applied is within that bounds of the hardware
2113 * rate. If so, no scaling or compensation need be done.
2115 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2116 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2117 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2118 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2121 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2124 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2126 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2127 vcpu->arch.virtual_tsc_mult,
2128 vcpu->arch.virtual_tsc_shift);
2129 tsc += vcpu->arch.this_tsc_write;
2133 static inline int gtod_is_based_on_tsc(int mode)
2135 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2138 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2140 #ifdef CONFIG_X86_64
2142 struct kvm_arch *ka = &vcpu->kvm->arch;
2143 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2145 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2146 atomic_read(&vcpu->kvm->online_vcpus));
2149 * Once the masterclock is enabled, always perform request in
2150 * order to update it.
2152 * In order to enable masterclock, the host clocksource must be TSC
2153 * and the vcpus need to have matched TSCs. When that happens,
2154 * perform request to enable masterclock.
2156 if (ka->use_master_clock ||
2157 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2158 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2160 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2161 atomic_read(&vcpu->kvm->online_vcpus),
2162 ka->use_master_clock, gtod->clock.vclock_mode);
2167 * Multiply tsc by a fixed point number represented by ratio.
2169 * The most significant 64-N bits (mult) of ratio represent the
2170 * integral part of the fixed point number; the remaining N bits
2171 * (frac) represent the fractional part, ie. ratio represents a fixed
2172 * point number (mult + frac * 2^(-N)).
2174 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2176 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2178 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2181 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2184 u64 ratio = vcpu->arch.tsc_scaling_ratio;
2186 if (ratio != kvm_default_tsc_scaling_ratio)
2187 _tsc = __scale_tsc(ratio, tsc);
2191 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2193 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2197 tsc = kvm_scale_tsc(vcpu, rdtsc());
2199 return target_tsc - tsc;
2202 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2204 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2206 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2208 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2210 vcpu->arch.l1_tsc_offset = offset;
2211 vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
2214 static inline bool kvm_check_tsc_unstable(void)
2216 #ifdef CONFIG_X86_64
2218 * TSC is marked unstable when we're running on Hyper-V,
2219 * 'TSC page' clocksource is good.
2221 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2224 return check_tsc_unstable();
2227 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2229 struct kvm *kvm = vcpu->kvm;
2230 u64 offset, ns, elapsed;
2231 unsigned long flags;
2233 bool already_matched;
2234 bool synchronizing = false;
2236 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2237 offset = kvm_compute_tsc_offset(vcpu, data);
2238 ns = get_kvmclock_base_ns();
2239 elapsed = ns - kvm->arch.last_tsc_nsec;
2241 if (vcpu->arch.virtual_tsc_khz) {
2244 * detection of vcpu initialization -- need to sync
2245 * with other vCPUs. This particularly helps to keep
2246 * kvm_clock stable after CPU hotplug
2248 synchronizing = true;
2250 u64 tsc_exp = kvm->arch.last_tsc_write +
2251 nsec_to_cycles(vcpu, elapsed);
2252 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2254 * Special case: TSC write with a small delta (1 second)
2255 * of virtual cycle time against real time is
2256 * interpreted as an attempt to synchronize the CPU.
2258 synchronizing = data < tsc_exp + tsc_hz &&
2259 data + tsc_hz > tsc_exp;
2264 * For a reliable TSC, we can match TSC offsets, and for an unstable
2265 * TSC, we add elapsed time in this computation. We could let the
2266 * compensation code attempt to catch up if we fall behind, but
2267 * it's better to try to match offsets from the beginning.
2269 if (synchronizing &&
2270 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2271 if (!kvm_check_tsc_unstable()) {
2272 offset = kvm->arch.cur_tsc_offset;
2274 u64 delta = nsec_to_cycles(vcpu, elapsed);
2276 offset = kvm_compute_tsc_offset(vcpu, data);
2279 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2282 * We split periods of matched TSC writes into generations.
2283 * For each generation, we track the original measured
2284 * nanosecond time, offset, and write, so if TSCs are in
2285 * sync, we can match exact offset, and if not, we can match
2286 * exact software computation in compute_guest_tsc()
2288 * These values are tracked in kvm->arch.cur_xxx variables.
2290 kvm->arch.cur_tsc_generation++;
2291 kvm->arch.cur_tsc_nsec = ns;
2292 kvm->arch.cur_tsc_write = data;
2293 kvm->arch.cur_tsc_offset = offset;
2298 * We also track th most recent recorded KHZ, write and time to
2299 * allow the matching interval to be extended at each write.
2301 kvm->arch.last_tsc_nsec = ns;
2302 kvm->arch.last_tsc_write = data;
2303 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2305 vcpu->arch.last_guest_tsc = data;
2307 /* Keep track of which generation this VCPU has synchronized to */
2308 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2309 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2310 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2312 kvm_vcpu_write_tsc_offset(vcpu, offset);
2313 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2315 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2317 kvm->arch.nr_vcpus_matched_tsc = 0;
2318 } else if (!already_matched) {
2319 kvm->arch.nr_vcpus_matched_tsc++;
2322 kvm_track_tsc_matching(vcpu);
2323 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2326 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2329 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2330 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2333 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2335 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2336 WARN_ON(adjustment < 0);
2337 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2338 adjust_tsc_offset_guest(vcpu, adjustment);
2341 #ifdef CONFIG_X86_64
2343 static u64 read_tsc(void)
2345 u64 ret = (u64)rdtsc_ordered();
2346 u64 last = pvclock_gtod_data.clock.cycle_last;
2348 if (likely(ret >= last))
2352 * GCC likes to generate cmov here, but this branch is extremely
2353 * predictable (it's just a function of time and the likely is
2354 * very likely) and there's a data dependence, so force GCC
2355 * to generate a branch instead. I don't barrier() because
2356 * we don't actually need a barrier, and if this function
2357 * ever gets inlined it will generate worse code.
2363 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2369 switch (clock->vclock_mode) {
2370 case VDSO_CLOCKMODE_HVCLOCK:
2371 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2373 if (tsc_pg_val != U64_MAX) {
2374 /* TSC page valid */
2375 *mode = VDSO_CLOCKMODE_HVCLOCK;
2376 v = (tsc_pg_val - clock->cycle_last) &
2379 /* TSC page invalid */
2380 *mode = VDSO_CLOCKMODE_NONE;
2383 case VDSO_CLOCKMODE_TSC:
2384 *mode = VDSO_CLOCKMODE_TSC;
2385 *tsc_timestamp = read_tsc();
2386 v = (*tsc_timestamp - clock->cycle_last) &
2390 *mode = VDSO_CLOCKMODE_NONE;
2393 if (*mode == VDSO_CLOCKMODE_NONE)
2394 *tsc_timestamp = v = 0;
2396 return v * clock->mult;
2399 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2401 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2407 seq = read_seqcount_begin(>od->seq);
2408 ns = gtod->raw_clock.base_cycles;
2409 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2410 ns >>= gtod->raw_clock.shift;
2411 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2412 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2418 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2420 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2426 seq = read_seqcount_begin(>od->seq);
2427 ts->tv_sec = gtod->wall_time_sec;
2428 ns = gtod->clock.base_cycles;
2429 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2430 ns >>= gtod->clock.shift;
2431 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2433 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2439 /* returns true if host is using TSC based clocksource */
2440 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2442 /* checked again under seqlock below */
2443 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2446 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2450 /* returns true if host is using TSC based clocksource */
2451 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2454 /* checked again under seqlock below */
2455 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2458 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2464 * Assuming a stable TSC across physical CPUS, and a stable TSC
2465 * across virtual CPUs, the following condition is possible.
2466 * Each numbered line represents an event visible to both
2467 * CPUs at the next numbered event.
2469 * "timespecX" represents host monotonic time. "tscX" represents
2472 * VCPU0 on CPU0 | VCPU1 on CPU1
2474 * 1. read timespec0,tsc0
2475 * 2. | timespec1 = timespec0 + N
2477 * 3. transition to guest | transition to guest
2478 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2479 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2480 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2482 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2485 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2487 * - 0 < N - M => M < N
2489 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2490 * always the case (the difference between two distinct xtime instances
2491 * might be smaller then the difference between corresponding TSC reads,
2492 * when updating guest vcpus pvclock areas).
2494 * To avoid that problem, do not allow visibility of distinct
2495 * system_timestamp/tsc_timestamp values simultaneously: use a master
2496 * copy of host monotonic time values. Update that master copy
2499 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2503 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2505 #ifdef CONFIG_X86_64
2506 struct kvm_arch *ka = &kvm->arch;
2508 bool host_tsc_clocksource, vcpus_matched;
2510 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2511 atomic_read(&kvm->online_vcpus));
2514 * If the host uses TSC clock, then passthrough TSC as stable
2517 host_tsc_clocksource = kvm_get_time_and_clockread(
2518 &ka->master_kernel_ns,
2519 &ka->master_cycle_now);
2521 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2522 && !ka->backwards_tsc_observed
2523 && !ka->boot_vcpu_runs_old_kvmclock;
2525 if (ka->use_master_clock)
2526 atomic_set(&kvm_guest_has_master_clock, 1);
2528 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2529 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2534 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2536 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2539 static void kvm_gen_update_masterclock(struct kvm *kvm)
2541 #ifdef CONFIG_X86_64
2543 struct kvm_vcpu *vcpu;
2544 struct kvm_arch *ka = &kvm->arch;
2546 spin_lock(&ka->pvclock_gtod_sync_lock);
2547 kvm_make_mclock_inprogress_request(kvm);
2548 /* no guest entries from this point */
2549 pvclock_update_vm_gtod_copy(kvm);
2551 kvm_for_each_vcpu(i, vcpu, kvm)
2552 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2554 /* guest entries allowed */
2555 kvm_for_each_vcpu(i, vcpu, kvm)
2556 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2558 spin_unlock(&ka->pvclock_gtod_sync_lock);
2562 u64 get_kvmclock_ns(struct kvm *kvm)
2564 struct kvm_arch *ka = &kvm->arch;
2565 struct pvclock_vcpu_time_info hv_clock;
2568 spin_lock(&ka->pvclock_gtod_sync_lock);
2569 if (!ka->use_master_clock) {
2570 spin_unlock(&ka->pvclock_gtod_sync_lock);
2571 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2574 hv_clock.tsc_timestamp = ka->master_cycle_now;
2575 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2576 spin_unlock(&ka->pvclock_gtod_sync_lock);
2578 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2581 if (__this_cpu_read(cpu_tsc_khz)) {
2582 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2583 &hv_clock.tsc_shift,
2584 &hv_clock.tsc_to_system_mul);
2585 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2587 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2594 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2596 struct kvm_vcpu_arch *vcpu = &v->arch;
2597 struct pvclock_vcpu_time_info guest_hv_clock;
2599 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2600 &guest_hv_clock, sizeof(guest_hv_clock))))
2603 /* This VCPU is paused, but it's legal for a guest to read another
2604 * VCPU's kvmclock, so we really have to follow the specification where
2605 * it says that version is odd if data is being modified, and even after
2608 * Version field updates must be kept separate. This is because
2609 * kvm_write_guest_cached might use a "rep movs" instruction, and
2610 * writes within a string instruction are weakly ordered. So there
2611 * are three writes overall.
2613 * As a small optimization, only write the version field in the first
2614 * and third write. The vcpu->pv_time cache is still valid, because the
2615 * version field is the first in the struct.
2617 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2619 if (guest_hv_clock.version & 1)
2620 ++guest_hv_clock.version; /* first time write, random junk */
2622 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2623 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2625 sizeof(vcpu->hv_clock.version));
2629 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2630 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2632 if (vcpu->pvclock_set_guest_stopped_request) {
2633 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2634 vcpu->pvclock_set_guest_stopped_request = false;
2637 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2639 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2641 sizeof(vcpu->hv_clock));
2645 vcpu->hv_clock.version++;
2646 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2648 sizeof(vcpu->hv_clock.version));
2651 static int kvm_guest_time_update(struct kvm_vcpu *v)
2653 unsigned long flags, tgt_tsc_khz;
2654 struct kvm_vcpu_arch *vcpu = &v->arch;
2655 struct kvm_arch *ka = &v->kvm->arch;
2657 u64 tsc_timestamp, host_tsc;
2659 bool use_master_clock;
2665 * If the host uses TSC clock, then passthrough TSC as stable
2668 spin_lock(&ka->pvclock_gtod_sync_lock);
2669 use_master_clock = ka->use_master_clock;
2670 if (use_master_clock) {
2671 host_tsc = ka->master_cycle_now;
2672 kernel_ns = ka->master_kernel_ns;
2674 spin_unlock(&ka->pvclock_gtod_sync_lock);
2676 /* Keep irq disabled to prevent changes to the clock */
2677 local_irq_save(flags);
2678 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2679 if (unlikely(tgt_tsc_khz == 0)) {
2680 local_irq_restore(flags);
2681 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2684 if (!use_master_clock) {
2686 kernel_ns = get_kvmclock_base_ns();
2689 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2692 * We may have to catch up the TSC to match elapsed wall clock
2693 * time for two reasons, even if kvmclock is used.
2694 * 1) CPU could have been running below the maximum TSC rate
2695 * 2) Broken TSC compensation resets the base at each VCPU
2696 * entry to avoid unknown leaps of TSC even when running
2697 * again on the same CPU. This may cause apparent elapsed
2698 * time to disappear, and the guest to stand still or run
2701 if (vcpu->tsc_catchup) {
2702 u64 tsc = compute_guest_tsc(v, kernel_ns);
2703 if (tsc > tsc_timestamp) {
2704 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2705 tsc_timestamp = tsc;
2709 local_irq_restore(flags);
2711 /* With all the info we got, fill in the values */
2713 if (kvm_has_tsc_control)
2714 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2716 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2717 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2718 &vcpu->hv_clock.tsc_shift,
2719 &vcpu->hv_clock.tsc_to_system_mul);
2720 vcpu->hw_tsc_khz = tgt_tsc_khz;
2723 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2724 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2725 vcpu->last_guest_tsc = tsc_timestamp;
2727 /* If the host uses TSC clocksource, then it is stable */
2729 if (use_master_clock)
2730 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2732 vcpu->hv_clock.flags = pvclock_flags;
2734 if (vcpu->pv_time_enabled)
2735 kvm_setup_pvclock_page(v);
2736 if (v == kvm_get_vcpu(v->kvm, 0))
2737 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2742 * kvmclock updates which are isolated to a given vcpu, such as
2743 * vcpu->cpu migration, should not allow system_timestamp from
2744 * the rest of the vcpus to remain static. Otherwise ntp frequency
2745 * correction applies to one vcpu's system_timestamp but not
2748 * So in those cases, request a kvmclock update for all vcpus.
2749 * We need to rate-limit these requests though, as they can
2750 * considerably slow guests that have a large number of vcpus.
2751 * The time for a remote vcpu to update its kvmclock is bound
2752 * by the delay we use to rate-limit the updates.
2755 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2757 static void kvmclock_update_fn(struct work_struct *work)
2760 struct delayed_work *dwork = to_delayed_work(work);
2761 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2762 kvmclock_update_work);
2763 struct kvm *kvm = container_of(ka, struct kvm, arch);
2764 struct kvm_vcpu *vcpu;
2766 kvm_for_each_vcpu(i, vcpu, kvm) {
2767 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2768 kvm_vcpu_kick(vcpu);
2772 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2774 struct kvm *kvm = v->kvm;
2776 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2777 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2778 KVMCLOCK_UPDATE_DELAY);
2781 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2783 static void kvmclock_sync_fn(struct work_struct *work)
2785 struct delayed_work *dwork = to_delayed_work(work);
2786 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2787 kvmclock_sync_work);
2788 struct kvm *kvm = container_of(ka, struct kvm, arch);
2790 if (!kvmclock_periodic_sync)
2793 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2794 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2795 KVMCLOCK_SYNC_PERIOD);
2799 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2801 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2803 /* McStatusWrEn enabled? */
2804 if (guest_cpuid_is_amd_or_hygon(vcpu))
2805 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2810 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2812 u64 mcg_cap = vcpu->arch.mcg_cap;
2813 unsigned bank_num = mcg_cap & 0xff;
2814 u32 msr = msr_info->index;
2815 u64 data = msr_info->data;
2818 case MSR_IA32_MCG_STATUS:
2819 vcpu->arch.mcg_status = data;
2821 case MSR_IA32_MCG_CTL:
2822 if (!(mcg_cap & MCG_CTL_P) &&
2823 (data || !msr_info->host_initiated))
2825 if (data != 0 && data != ~(u64)0)
2827 vcpu->arch.mcg_ctl = data;
2830 if (msr >= MSR_IA32_MC0_CTL &&
2831 msr < MSR_IA32_MCx_CTL(bank_num)) {
2832 u32 offset = array_index_nospec(
2833 msr - MSR_IA32_MC0_CTL,
2834 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2836 /* only 0 or all 1s can be written to IA32_MCi_CTL
2837 * some Linux kernels though clear bit 10 in bank 4 to
2838 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2839 * this to avoid an uncatched #GP in the guest
2841 if ((offset & 0x3) == 0 &&
2842 data != 0 && (data | (1 << 10)) != ~(u64)0)
2846 if (!msr_info->host_initiated &&
2847 (offset & 0x3) == 1 && data != 0) {
2848 if (!can_set_mci_status(vcpu))
2852 vcpu->arch.mce_banks[offset] = data;
2860 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2862 struct kvm *kvm = vcpu->kvm;
2863 int lm = is_long_mode(vcpu);
2864 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2865 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2866 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2867 : kvm->arch.xen_hvm_config.blob_size_32;
2868 u32 page_num = data & ~PAGE_MASK;
2869 u64 page_addr = data & PAGE_MASK;
2872 if (page_num >= blob_size)
2875 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2877 return PTR_ERR(page);
2879 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) {
2886 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2888 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2890 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2893 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2895 gpa_t gpa = data & ~0x3f;
2897 /* Bits 4:5 are reserved, Should be zero */
2901 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2902 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2905 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2906 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2909 if (!lapic_in_kernel(vcpu))
2910 return data ? 1 : 0;
2912 vcpu->arch.apf.msr_en_val = data;
2914 if (!kvm_pv_async_pf_enabled(vcpu)) {
2915 kvm_clear_async_pf_completion_queue(vcpu);
2916 kvm_async_pf_hash_reset(vcpu);
2920 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2924 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2925 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2927 kvm_async_pf_wakeup_all(vcpu);
2932 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2934 /* Bits 8-63 are reserved */
2938 if (!lapic_in_kernel(vcpu))
2941 vcpu->arch.apf.msr_int_val = data;
2943 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2948 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2950 vcpu->arch.pv_time_enabled = false;
2951 vcpu->arch.time = 0;
2954 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2956 ++vcpu->stat.tlb_flush;
2957 kvm_x86_ops.tlb_flush_all(vcpu);
2960 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2962 ++vcpu->stat.tlb_flush;
2963 kvm_x86_ops.tlb_flush_guest(vcpu);
2966 static void record_steal_time(struct kvm_vcpu *vcpu)
2968 struct kvm_host_map map;
2969 struct kvm_steal_time *st;
2971 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2974 /* -EAGAIN is returned in atomic context so we can just return. */
2975 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2976 &map, &vcpu->arch.st.cache, false))
2980 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2983 * Doing a TLB flush here, on the guest's behalf, can avoid
2986 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
2987 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2988 st->preempted & KVM_VCPU_FLUSH_TLB);
2989 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2990 kvm_vcpu_flush_tlb_guest(vcpu);
2993 vcpu->arch.st.preempted = 0;
2995 if (st->version & 1)
2996 st->version += 1; /* first time write, random junk */
3002 st->steal += current->sched_info.run_delay -
3003 vcpu->arch.st.last_steal;
3004 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3010 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3013 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3016 u32 msr = msr_info->index;
3017 u64 data = msr_info->data;
3020 case MSR_AMD64_NB_CFG:
3021 case MSR_IA32_UCODE_WRITE:
3022 case MSR_VM_HSAVE_PA:
3023 case MSR_AMD64_PATCH_LOADER:
3024 case MSR_AMD64_BU_CFG2:
3025 case MSR_AMD64_DC_CFG:
3026 case MSR_F15H_EX_CFG:
3029 case MSR_IA32_UCODE_REV:
3030 if (msr_info->host_initiated)
3031 vcpu->arch.microcode_version = data;
3033 case MSR_IA32_ARCH_CAPABILITIES:
3034 if (!msr_info->host_initiated)
3036 vcpu->arch.arch_capabilities = data;
3038 case MSR_IA32_PERF_CAPABILITIES: {
3039 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3041 if (!msr_info->host_initiated)
3043 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3045 if (data & ~msr_ent.data)
3048 vcpu->arch.perf_capabilities = data;
3053 return set_efer(vcpu, msr_info);
3055 data &= ~(u64)0x40; /* ignore flush filter disable */
3056 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3057 data &= ~(u64)0x8; /* ignore TLB cache disable */
3059 /* Handle McStatusWrEn */
3060 if (data == BIT_ULL(18)) {
3061 vcpu->arch.msr_hwcr = data;
3062 } else if (data != 0) {
3063 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3068 case MSR_FAM10H_MMIO_CONF_BASE:
3070 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3075 case MSR_IA32_DEBUGCTLMSR:
3077 /* We support the non-activated case already */
3079 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
3080 /* Values other than LBR and BTF are vendor-specific,
3081 thus reserved and should throw a #GP */
3083 } else if (report_ignored_msrs)
3084 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
3087 case 0x200 ... 0x2ff:
3088 return kvm_mtrr_set_msr(vcpu, msr, data);
3089 case MSR_IA32_APICBASE:
3090 return kvm_set_apic_base(vcpu, msr_info);
3091 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3092 return kvm_x2apic_msr_write(vcpu, msr, data);
3093 case MSR_IA32_TSCDEADLINE:
3094 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3096 case MSR_IA32_TSC_ADJUST:
3097 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3098 if (!msr_info->host_initiated) {
3099 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3100 adjust_tsc_offset_guest(vcpu, adj);
3102 vcpu->arch.ia32_tsc_adjust_msr = data;
3105 case MSR_IA32_MISC_ENABLE:
3106 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3107 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3108 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3110 vcpu->arch.ia32_misc_enable_msr = data;
3111 kvm_update_cpuid_runtime(vcpu);
3113 vcpu->arch.ia32_misc_enable_msr = data;
3116 case MSR_IA32_SMBASE:
3117 if (!msr_info->host_initiated)
3119 vcpu->arch.smbase = data;
3121 case MSR_IA32_POWER_CTL:
3122 vcpu->arch.msr_ia32_power_ctl = data;
3125 if (msr_info->host_initiated) {
3126 kvm_synchronize_tsc(vcpu, data);
3128 u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3129 adjust_tsc_offset_guest(vcpu, adj);
3130 vcpu->arch.ia32_tsc_adjust_msr += adj;
3134 if (!msr_info->host_initiated &&
3135 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3138 * KVM supports exposing PT to the guest, but does not support
3139 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3140 * XSAVES/XRSTORS to save/restore PT MSRs.
3142 if (data & ~supported_xss)
3144 vcpu->arch.ia32_xss = data;
3147 if (!msr_info->host_initiated)
3149 vcpu->arch.smi_count = data;
3151 case MSR_KVM_WALL_CLOCK_NEW:
3152 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3155 kvm_write_wall_clock(vcpu->kvm, data);
3157 case MSR_KVM_WALL_CLOCK:
3158 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3161 kvm_write_wall_clock(vcpu->kvm, data);
3163 case MSR_KVM_SYSTEM_TIME_NEW:
3164 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3167 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3169 case MSR_KVM_SYSTEM_TIME:
3170 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3173 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3175 case MSR_KVM_ASYNC_PF_EN:
3176 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3179 if (kvm_pv_enable_async_pf(vcpu, data))
3182 case MSR_KVM_ASYNC_PF_INT:
3183 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3186 if (kvm_pv_enable_async_pf_int(vcpu, data))
3189 case MSR_KVM_ASYNC_PF_ACK:
3190 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3193 vcpu->arch.apf.pageready_pending = false;
3194 kvm_check_async_pf_completion(vcpu);
3197 case MSR_KVM_STEAL_TIME:
3198 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3201 if (unlikely(!sched_info_on()))
3204 if (data & KVM_STEAL_RESERVED_MASK)
3207 vcpu->arch.st.msr_val = data;
3209 if (!(data & KVM_MSR_ENABLED))
3212 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3215 case MSR_KVM_PV_EOI_EN:
3216 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3219 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3223 case MSR_KVM_POLL_CONTROL:
3224 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3227 /* only enable bit supported */
3228 if (data & (-1ULL << 1))
3231 vcpu->arch.msr_kvm_poll_control = data;
3234 case MSR_IA32_MCG_CTL:
3235 case MSR_IA32_MCG_STATUS:
3236 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3237 return set_msr_mce(vcpu, msr_info);
3239 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3240 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3243 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3244 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3245 if (kvm_pmu_is_valid_msr(vcpu, msr))
3246 return kvm_pmu_set_msr(vcpu, msr_info);
3248 if (pr || data != 0)
3249 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3250 "0x%x data 0x%llx\n", msr, data);
3252 case MSR_K7_CLK_CTL:
3254 * Ignore all writes to this no longer documented MSR.
3255 * Writes are only relevant for old K7 processors,
3256 * all pre-dating SVM, but a recommended workaround from
3257 * AMD for these chips. It is possible to specify the
3258 * affected processor models on the command line, hence
3259 * the need to ignore the workaround.
3262 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3263 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3264 case HV_X64_MSR_SYNDBG_OPTIONS:
3265 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3266 case HV_X64_MSR_CRASH_CTL:
3267 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3268 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3269 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3270 case HV_X64_MSR_TSC_EMULATION_STATUS:
3271 return kvm_hv_set_msr_common(vcpu, msr, data,
3272 msr_info->host_initiated);
3273 case MSR_IA32_BBL_CR_CTL3:
3274 /* Drop writes to this legacy MSR -- see rdmsr
3275 * counterpart for further detail.
3277 if (report_ignored_msrs)
3278 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3281 case MSR_AMD64_OSVW_ID_LENGTH:
3282 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3284 vcpu->arch.osvw.length = data;
3286 case MSR_AMD64_OSVW_STATUS:
3287 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3289 vcpu->arch.osvw.status = data;
3291 case MSR_PLATFORM_INFO:
3292 if (!msr_info->host_initiated ||
3293 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3294 cpuid_fault_enabled(vcpu)))
3296 vcpu->arch.msr_platform_info = data;
3298 case MSR_MISC_FEATURES_ENABLES:
3299 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3300 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3301 !supports_cpuid_fault(vcpu)))
3303 vcpu->arch.msr_misc_features_enables = data;
3306 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
3307 return xen_hvm_config(vcpu, data);
3308 if (kvm_pmu_is_valid_msr(vcpu, msr))
3309 return kvm_pmu_set_msr(vcpu, msr_info);
3310 return KVM_MSR_RET_INVALID;
3314 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3316 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3319 u64 mcg_cap = vcpu->arch.mcg_cap;
3320 unsigned bank_num = mcg_cap & 0xff;
3323 case MSR_IA32_P5_MC_ADDR:
3324 case MSR_IA32_P5_MC_TYPE:
3327 case MSR_IA32_MCG_CAP:
3328 data = vcpu->arch.mcg_cap;
3330 case MSR_IA32_MCG_CTL:
3331 if (!(mcg_cap & MCG_CTL_P) && !host)
3333 data = vcpu->arch.mcg_ctl;
3335 case MSR_IA32_MCG_STATUS:
3336 data = vcpu->arch.mcg_status;
3339 if (msr >= MSR_IA32_MC0_CTL &&
3340 msr < MSR_IA32_MCx_CTL(bank_num)) {
3341 u32 offset = array_index_nospec(
3342 msr - MSR_IA32_MC0_CTL,
3343 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3345 data = vcpu->arch.mce_banks[offset];
3354 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3356 switch (msr_info->index) {
3357 case MSR_IA32_PLATFORM_ID:
3358 case MSR_IA32_EBL_CR_POWERON:
3359 case MSR_IA32_DEBUGCTLMSR:
3360 case MSR_IA32_LASTBRANCHFROMIP:
3361 case MSR_IA32_LASTBRANCHTOIP:
3362 case MSR_IA32_LASTINTFROMIP:
3363 case MSR_IA32_LASTINTTOIP:
3365 case MSR_K8_TSEG_ADDR:
3366 case MSR_K8_TSEG_MASK:
3367 case MSR_VM_HSAVE_PA:
3368 case MSR_K8_INT_PENDING_MSG:
3369 case MSR_AMD64_NB_CFG:
3370 case MSR_FAM10H_MMIO_CONF_BASE:
3371 case MSR_AMD64_BU_CFG2:
3372 case MSR_IA32_PERF_CTL:
3373 case MSR_AMD64_DC_CFG:
3374 case MSR_F15H_EX_CFG:
3376 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3377 * limit) MSRs. Just return 0, as we do not want to expose the host
3378 * data here. Do not conditionalize this on CPUID, as KVM does not do
3379 * so for existing CPU-specific MSRs.
3381 case MSR_RAPL_POWER_UNIT:
3382 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3383 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3384 case MSR_PKG_ENERGY_STATUS: /* Total package */
3385 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3388 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3389 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3390 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3391 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3392 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3393 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3394 return kvm_pmu_get_msr(vcpu, msr_info);
3397 case MSR_IA32_UCODE_REV:
3398 msr_info->data = vcpu->arch.microcode_version;
3400 case MSR_IA32_ARCH_CAPABILITIES:
3401 if (!msr_info->host_initiated &&
3402 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3404 msr_info->data = vcpu->arch.arch_capabilities;
3406 case MSR_IA32_PERF_CAPABILITIES:
3407 if (!msr_info->host_initiated &&
3408 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3410 msr_info->data = vcpu->arch.perf_capabilities;
3412 case MSR_IA32_POWER_CTL:
3413 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3415 case MSR_IA32_TSC: {
3417 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3418 * even when not intercepted. AMD manual doesn't explicitly
3419 * state this but appears to behave the same.
3421 * On userspace reads and writes, however, we unconditionally
3422 * return L1's TSC value to ensure backwards-compatible
3423 * behavior for migration.
3425 u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3426 vcpu->arch.tsc_offset;
3428 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3432 case 0x200 ... 0x2ff:
3433 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3434 case 0xcd: /* fsb frequency */
3438 * MSR_EBC_FREQUENCY_ID
3439 * Conservative value valid for even the basic CPU models.
3440 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3441 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3442 * and 266MHz for model 3, or 4. Set Core Clock
3443 * Frequency to System Bus Frequency Ratio to 1 (bits
3444 * 31:24) even though these are only valid for CPU
3445 * models > 2, however guests may end up dividing or
3446 * multiplying by zero otherwise.
3448 case MSR_EBC_FREQUENCY_ID:
3449 msr_info->data = 1 << 24;
3451 case MSR_IA32_APICBASE:
3452 msr_info->data = kvm_get_apic_base(vcpu);
3454 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3455 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3456 case MSR_IA32_TSCDEADLINE:
3457 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3459 case MSR_IA32_TSC_ADJUST:
3460 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3462 case MSR_IA32_MISC_ENABLE:
3463 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3465 case MSR_IA32_SMBASE:
3466 if (!msr_info->host_initiated)
3468 msr_info->data = vcpu->arch.smbase;
3471 msr_info->data = vcpu->arch.smi_count;
3473 case MSR_IA32_PERF_STATUS:
3474 /* TSC increment by tick */
3475 msr_info->data = 1000ULL;
3476 /* CPU multiplier */
3477 msr_info->data |= (((uint64_t)4ULL) << 40);
3480 msr_info->data = vcpu->arch.efer;
3482 case MSR_KVM_WALL_CLOCK:
3483 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3486 msr_info->data = vcpu->kvm->arch.wall_clock;
3488 case MSR_KVM_WALL_CLOCK_NEW:
3489 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3492 msr_info->data = vcpu->kvm->arch.wall_clock;
3494 case MSR_KVM_SYSTEM_TIME:
3495 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3498 msr_info->data = vcpu->arch.time;
3500 case MSR_KVM_SYSTEM_TIME_NEW:
3501 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3504 msr_info->data = vcpu->arch.time;
3506 case MSR_KVM_ASYNC_PF_EN:
3507 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3510 msr_info->data = vcpu->arch.apf.msr_en_val;
3512 case MSR_KVM_ASYNC_PF_INT:
3513 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3516 msr_info->data = vcpu->arch.apf.msr_int_val;
3518 case MSR_KVM_ASYNC_PF_ACK:
3519 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3524 case MSR_KVM_STEAL_TIME:
3525 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3528 msr_info->data = vcpu->arch.st.msr_val;
3530 case MSR_KVM_PV_EOI_EN:
3531 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3534 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3536 case MSR_KVM_POLL_CONTROL:
3537 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3540 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3542 case MSR_IA32_P5_MC_ADDR:
3543 case MSR_IA32_P5_MC_TYPE:
3544 case MSR_IA32_MCG_CAP:
3545 case MSR_IA32_MCG_CTL:
3546 case MSR_IA32_MCG_STATUS:
3547 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3548 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3549 msr_info->host_initiated);
3551 if (!msr_info->host_initiated &&
3552 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3554 msr_info->data = vcpu->arch.ia32_xss;
3556 case MSR_K7_CLK_CTL:
3558 * Provide expected ramp-up count for K7. All other
3559 * are set to zero, indicating minimum divisors for
3562 * This prevents guest kernels on AMD host with CPU
3563 * type 6, model 8 and higher from exploding due to
3564 * the rdmsr failing.
3566 msr_info->data = 0x20000000;
3568 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3569 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3570 case HV_X64_MSR_SYNDBG_OPTIONS:
3571 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3572 case HV_X64_MSR_CRASH_CTL:
3573 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3574 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3575 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3576 case HV_X64_MSR_TSC_EMULATION_STATUS:
3577 return kvm_hv_get_msr_common(vcpu,
3578 msr_info->index, &msr_info->data,
3579 msr_info->host_initiated);
3580 case MSR_IA32_BBL_CR_CTL3:
3581 /* This legacy MSR exists but isn't fully documented in current
3582 * silicon. It is however accessed by winxp in very narrow
3583 * scenarios where it sets bit #19, itself documented as
3584 * a "reserved" bit. Best effort attempt to source coherent
3585 * read data here should the balance of the register be
3586 * interpreted by the guest:
3588 * L2 cache control register 3: 64GB range, 256KB size,
3589 * enabled, latency 0x1, configured
3591 msr_info->data = 0xbe702111;
3593 case MSR_AMD64_OSVW_ID_LENGTH:
3594 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3596 msr_info->data = vcpu->arch.osvw.length;
3598 case MSR_AMD64_OSVW_STATUS:
3599 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3601 msr_info->data = vcpu->arch.osvw.status;
3603 case MSR_PLATFORM_INFO:
3604 if (!msr_info->host_initiated &&
3605 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3607 msr_info->data = vcpu->arch.msr_platform_info;
3609 case MSR_MISC_FEATURES_ENABLES:
3610 msr_info->data = vcpu->arch.msr_misc_features_enables;
3613 msr_info->data = vcpu->arch.msr_hwcr;
3616 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3617 return kvm_pmu_get_msr(vcpu, msr_info);
3618 return KVM_MSR_RET_INVALID;
3622 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3625 * Read or write a bunch of msrs. All parameters are kernel addresses.
3627 * @return number of msrs set successfully.
3629 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3630 struct kvm_msr_entry *entries,
3631 int (*do_msr)(struct kvm_vcpu *vcpu,
3632 unsigned index, u64 *data))
3636 for (i = 0; i < msrs->nmsrs; ++i)
3637 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3644 * Read or write a bunch of msrs. Parameters are user addresses.
3646 * @return number of msrs set successfully.
3648 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3649 int (*do_msr)(struct kvm_vcpu *vcpu,
3650 unsigned index, u64 *data),
3653 struct kvm_msrs msrs;
3654 struct kvm_msr_entry *entries;
3659 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3663 if (msrs.nmsrs >= MAX_IO_MSRS)
3666 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3667 entries = memdup_user(user_msrs->entries, size);
3668 if (IS_ERR(entries)) {
3669 r = PTR_ERR(entries);
3673 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3678 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3689 static inline bool kvm_can_mwait_in_guest(void)
3691 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3692 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3693 boot_cpu_has(X86_FEATURE_ARAT);
3696 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3697 struct kvm_cpuid2 __user *cpuid_arg)
3699 struct kvm_cpuid2 cpuid;
3703 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3706 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3711 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3717 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3722 case KVM_CAP_IRQCHIP:
3724 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3725 case KVM_CAP_SET_TSS_ADDR:
3726 case KVM_CAP_EXT_CPUID:
3727 case KVM_CAP_EXT_EMUL_CPUID:
3728 case KVM_CAP_CLOCKSOURCE:
3730 case KVM_CAP_NOP_IO_DELAY:
3731 case KVM_CAP_MP_STATE:
3732 case KVM_CAP_SYNC_MMU:
3733 case KVM_CAP_USER_NMI:
3734 case KVM_CAP_REINJECT_CONTROL:
3735 case KVM_CAP_IRQ_INJECT_STATUS:
3736 case KVM_CAP_IOEVENTFD:
3737 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3739 case KVM_CAP_PIT_STATE2:
3740 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3741 case KVM_CAP_XEN_HVM:
3742 case KVM_CAP_VCPU_EVENTS:
3743 case KVM_CAP_HYPERV:
3744 case KVM_CAP_HYPERV_VAPIC:
3745 case KVM_CAP_HYPERV_SPIN:
3746 case KVM_CAP_HYPERV_SYNIC:
3747 case KVM_CAP_HYPERV_SYNIC2:
3748 case KVM_CAP_HYPERV_VP_INDEX:
3749 case KVM_CAP_HYPERV_EVENTFD:
3750 case KVM_CAP_HYPERV_TLBFLUSH:
3751 case KVM_CAP_HYPERV_SEND_IPI:
3752 case KVM_CAP_HYPERV_CPUID:
3753 case KVM_CAP_SYS_HYPERV_CPUID:
3754 case KVM_CAP_PCI_SEGMENT:
3755 case KVM_CAP_DEBUGREGS:
3756 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3758 case KVM_CAP_ASYNC_PF:
3759 case KVM_CAP_ASYNC_PF_INT:
3760 case KVM_CAP_GET_TSC_KHZ:
3761 case KVM_CAP_KVMCLOCK_CTRL:
3762 case KVM_CAP_READONLY_MEM:
3763 case KVM_CAP_HYPERV_TIME:
3764 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3765 case KVM_CAP_TSC_DEADLINE_TIMER:
3766 case KVM_CAP_DISABLE_QUIRKS:
3767 case KVM_CAP_SET_BOOT_CPU_ID:
3768 case KVM_CAP_SPLIT_IRQCHIP:
3769 case KVM_CAP_IMMEDIATE_EXIT:
3770 case KVM_CAP_PMU_EVENT_FILTER:
3771 case KVM_CAP_GET_MSR_FEATURES:
3772 case KVM_CAP_MSR_PLATFORM_INFO:
3773 case KVM_CAP_EXCEPTION_PAYLOAD:
3774 case KVM_CAP_SET_GUEST_DEBUG:
3775 case KVM_CAP_LAST_CPU:
3776 case KVM_CAP_X86_USER_SPACE_MSR:
3777 case KVM_CAP_X86_MSR_FILTER:
3778 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3781 case KVM_CAP_SYNC_REGS:
3782 r = KVM_SYNC_X86_VALID_FIELDS;
3784 case KVM_CAP_ADJUST_CLOCK:
3785 r = KVM_CLOCK_TSC_STABLE;
3787 case KVM_CAP_X86_DISABLE_EXITS:
3788 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3789 KVM_X86_DISABLE_EXITS_CSTATE;
3790 if(kvm_can_mwait_in_guest())
3791 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3793 case KVM_CAP_X86_SMM:
3794 /* SMBASE is usually relocated above 1M on modern chipsets,
3795 * and SMM handlers might indeed rely on 4G segment limits,
3796 * so do not report SMM to be available if real mode is
3797 * emulated via vm86 mode. Still, do not go to great lengths
3798 * to avoid userspace's usage of the feature, because it is a
3799 * fringe case that is not enabled except via specific settings
3800 * of the module parameters.
3802 r = kvm_x86_ops.has_emulated_msr(kvm, MSR_IA32_SMBASE);
3805 r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3807 case KVM_CAP_NR_VCPUS:
3808 r = KVM_SOFT_MAX_VCPUS;
3810 case KVM_CAP_MAX_VCPUS:
3813 case KVM_CAP_MAX_VCPU_ID:
3814 r = KVM_MAX_VCPU_ID;
3816 case KVM_CAP_PV_MMU: /* obsolete */
3820 r = KVM_MAX_MCE_BANKS;
3823 r = boot_cpu_has(X86_FEATURE_XSAVE);
3825 case KVM_CAP_TSC_CONTROL:
3826 r = kvm_has_tsc_control;
3828 case KVM_CAP_X2APIC_API:
3829 r = KVM_X2APIC_API_VALID_FLAGS;
3831 case KVM_CAP_NESTED_STATE:
3832 r = kvm_x86_ops.nested_ops->get_state ?
3833 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3835 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3836 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3838 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3839 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3841 case KVM_CAP_SMALLER_MAXPHYADDR:
3842 r = (int) allow_smaller_maxphyaddr;
3844 case KVM_CAP_STEAL_TIME:
3845 r = sched_info_on();
3854 long kvm_arch_dev_ioctl(struct file *filp,
3855 unsigned int ioctl, unsigned long arg)
3857 void __user *argp = (void __user *)arg;
3861 case KVM_GET_MSR_INDEX_LIST: {
3862 struct kvm_msr_list __user *user_msr_list = argp;
3863 struct kvm_msr_list msr_list;
3867 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3870 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3871 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3874 if (n < msr_list.nmsrs)
3877 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3878 num_msrs_to_save * sizeof(u32)))
3880 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3882 num_emulated_msrs * sizeof(u32)))
3887 case KVM_GET_SUPPORTED_CPUID:
3888 case KVM_GET_EMULATED_CPUID: {
3889 struct kvm_cpuid2 __user *cpuid_arg = argp;
3890 struct kvm_cpuid2 cpuid;
3893 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3896 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3902 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3907 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3909 if (copy_to_user(argp, &kvm_mce_cap_supported,
3910 sizeof(kvm_mce_cap_supported)))
3914 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3915 struct kvm_msr_list __user *user_msr_list = argp;
3916 struct kvm_msr_list msr_list;
3920 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3923 msr_list.nmsrs = num_msr_based_features;
3924 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3927 if (n < msr_list.nmsrs)
3930 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3931 num_msr_based_features * sizeof(u32)))
3937 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3939 case KVM_GET_SUPPORTED_HV_CPUID:
3940 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
3950 static void wbinvd_ipi(void *garbage)
3955 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3957 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3960 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3962 /* Address WBINVD may be executed by guest */
3963 if (need_emulate_wbinvd(vcpu)) {
3964 if (kvm_x86_ops.has_wbinvd_exit())
3965 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3966 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3967 smp_call_function_single(vcpu->cpu,
3968 wbinvd_ipi, NULL, 1);
3971 kvm_x86_ops.vcpu_load(vcpu, cpu);
3973 /* Save host pkru register if supported */
3974 vcpu->arch.host_pkru = read_pkru();
3976 /* Apply any externally detected TSC adjustments (due to suspend) */
3977 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3978 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3979 vcpu->arch.tsc_offset_adjustment = 0;
3980 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3983 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3984 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3985 rdtsc() - vcpu->arch.last_host_tsc;
3987 mark_tsc_unstable("KVM discovered backwards TSC");
3989 if (kvm_check_tsc_unstable()) {
3990 u64 offset = kvm_compute_tsc_offset(vcpu,
3991 vcpu->arch.last_guest_tsc);
3992 kvm_vcpu_write_tsc_offset(vcpu, offset);
3993 vcpu->arch.tsc_catchup = 1;
3996 if (kvm_lapic_hv_timer_in_use(vcpu))
3997 kvm_lapic_restart_hv_timer(vcpu);
4000 * On a host with synchronized TSC, there is no need to update
4001 * kvmclock on vcpu->cpu migration
4003 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4004 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4005 if (vcpu->cpu != cpu)
4006 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4010 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4013 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4015 struct kvm_host_map map;
4016 struct kvm_steal_time *st;
4019 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4022 if (vcpu->arch.st.preempted)
4026 * Take the srcu lock as memslots will be accessed to check the gfn
4027 * cache generation against the memslots generation.
4029 idx = srcu_read_lock(&vcpu->kvm->srcu);
4031 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4032 &vcpu->arch.st.cache, true))
4036 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4038 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4040 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4043 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4046 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4048 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4049 vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
4051 kvm_steal_time_set_preempted(vcpu);
4052 kvm_x86_ops.vcpu_put(vcpu);
4053 vcpu->arch.last_host_tsc = rdtsc();
4055 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4056 * on every vmexit, but if not, we might have a stale dr6 from the
4057 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4062 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4063 struct kvm_lapic_state *s)
4065 if (vcpu->arch.apicv_active)
4066 kvm_x86_ops.sync_pir_to_irr(vcpu);
4068 return kvm_apic_get_state(vcpu, s);
4071 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4072 struct kvm_lapic_state *s)
4076 r = kvm_apic_set_state(vcpu, s);
4079 update_cr8_intercept(vcpu);
4084 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4087 * We can accept userspace's request for interrupt injection
4088 * as long as we have a place to store the interrupt number.
4089 * The actual injection will happen when the CPU is able to
4090 * deliver the interrupt.
4092 if (kvm_cpu_has_extint(vcpu))
4095 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4096 return (!lapic_in_kernel(vcpu) ||
4097 kvm_apic_accept_pic_intr(vcpu));
4100 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4102 return kvm_arch_interrupt_allowed(vcpu) &&
4103 kvm_cpu_accept_dm_intr(vcpu);
4106 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4107 struct kvm_interrupt *irq)
4109 if (irq->irq >= KVM_NR_INTERRUPTS)
4112 if (!irqchip_in_kernel(vcpu->kvm)) {
4113 kvm_queue_interrupt(vcpu, irq->irq, false);
4114 kvm_make_request(KVM_REQ_EVENT, vcpu);
4119 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4120 * fail for in-kernel 8259.
4122 if (pic_in_kernel(vcpu->kvm))
4125 if (vcpu->arch.pending_external_vector != -1)
4128 vcpu->arch.pending_external_vector = irq->irq;
4129 kvm_make_request(KVM_REQ_EVENT, vcpu);
4133 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4135 kvm_inject_nmi(vcpu);
4140 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4142 kvm_make_request(KVM_REQ_SMI, vcpu);
4147 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4148 struct kvm_tpr_access_ctl *tac)
4152 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4156 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4160 unsigned bank_num = mcg_cap & 0xff, bank;
4163 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4165 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4168 vcpu->arch.mcg_cap = mcg_cap;
4169 /* Init IA32_MCG_CTL to all 1s */
4170 if (mcg_cap & MCG_CTL_P)
4171 vcpu->arch.mcg_ctl = ~(u64)0;
4172 /* Init IA32_MCi_CTL to all 1s */
4173 for (bank = 0; bank < bank_num; bank++)
4174 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4176 kvm_x86_ops.setup_mce(vcpu);
4181 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4182 struct kvm_x86_mce *mce)
4184 u64 mcg_cap = vcpu->arch.mcg_cap;
4185 unsigned bank_num = mcg_cap & 0xff;
4186 u64 *banks = vcpu->arch.mce_banks;
4188 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4191 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4192 * reporting is disabled
4194 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4195 vcpu->arch.mcg_ctl != ~(u64)0)
4197 banks += 4 * mce->bank;
4199 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4200 * reporting is disabled for the bank
4202 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4204 if (mce->status & MCI_STATUS_UC) {
4205 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4206 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4207 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4210 if (banks[1] & MCI_STATUS_VAL)
4211 mce->status |= MCI_STATUS_OVER;
4212 banks[2] = mce->addr;
4213 banks[3] = mce->misc;
4214 vcpu->arch.mcg_status = mce->mcg_status;
4215 banks[1] = mce->status;
4216 kvm_queue_exception(vcpu, MC_VECTOR);
4217 } else if (!(banks[1] & MCI_STATUS_VAL)
4218 || !(banks[1] & MCI_STATUS_UC)) {
4219 if (banks[1] & MCI_STATUS_VAL)
4220 mce->status |= MCI_STATUS_OVER;
4221 banks[2] = mce->addr;
4222 banks[3] = mce->misc;
4223 banks[1] = mce->status;
4225 banks[1] |= MCI_STATUS_OVER;
4229 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4230 struct kvm_vcpu_events *events)
4234 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4238 * In guest mode, payload delivery should be deferred,
4239 * so that the L1 hypervisor can intercept #PF before
4240 * CR2 is modified (or intercept #DB before DR6 is
4241 * modified under nVMX). Unless the per-VM capability,
4242 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4243 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4244 * opportunistically defer the exception payload, deliver it if the
4245 * capability hasn't been requested before processing a
4246 * KVM_GET_VCPU_EVENTS.
4248 if (!vcpu->kvm->arch.exception_payload_enabled &&
4249 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4250 kvm_deliver_exception_payload(vcpu);
4253 * The API doesn't provide the instruction length for software
4254 * exceptions, so don't report them. As long as the guest RIP
4255 * isn't advanced, we should expect to encounter the exception
4258 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4259 events->exception.injected = 0;
4260 events->exception.pending = 0;
4262 events->exception.injected = vcpu->arch.exception.injected;
4263 events->exception.pending = vcpu->arch.exception.pending;
4265 * For ABI compatibility, deliberately conflate
4266 * pending and injected exceptions when
4267 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4269 if (!vcpu->kvm->arch.exception_payload_enabled)
4270 events->exception.injected |=
4271 vcpu->arch.exception.pending;
4273 events->exception.nr = vcpu->arch.exception.nr;
4274 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4275 events->exception.error_code = vcpu->arch.exception.error_code;
4276 events->exception_has_payload = vcpu->arch.exception.has_payload;
4277 events->exception_payload = vcpu->arch.exception.payload;
4279 events->interrupt.injected =
4280 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4281 events->interrupt.nr = vcpu->arch.interrupt.nr;
4282 events->interrupt.soft = 0;
4283 events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
4285 events->nmi.injected = vcpu->arch.nmi_injected;
4286 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4287 events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
4288 events->nmi.pad = 0;
4290 events->sipi_vector = 0; /* never valid when reporting to user space */
4292 events->smi.smm = is_smm(vcpu);
4293 events->smi.pending = vcpu->arch.smi_pending;
4294 events->smi.smm_inside_nmi =
4295 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4296 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4298 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4299 | KVM_VCPUEVENT_VALID_SHADOW
4300 | KVM_VCPUEVENT_VALID_SMM);
4301 if (vcpu->kvm->arch.exception_payload_enabled)
4302 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4304 memset(&events->reserved, 0, sizeof(events->reserved));
4307 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4309 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4310 struct kvm_vcpu_events *events)
4312 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4313 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4314 | KVM_VCPUEVENT_VALID_SHADOW
4315 | KVM_VCPUEVENT_VALID_SMM
4316 | KVM_VCPUEVENT_VALID_PAYLOAD))
4319 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4320 if (!vcpu->kvm->arch.exception_payload_enabled)
4322 if (events->exception.pending)
4323 events->exception.injected = 0;
4325 events->exception_has_payload = 0;
4327 events->exception.pending = 0;
4328 events->exception_has_payload = 0;
4331 if ((events->exception.injected || events->exception.pending) &&
4332 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4335 /* INITs are latched while in SMM */
4336 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4337 (events->smi.smm || events->smi.pending) &&
4338 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4342 vcpu->arch.exception.injected = events->exception.injected;
4343 vcpu->arch.exception.pending = events->exception.pending;
4344 vcpu->arch.exception.nr = events->exception.nr;
4345 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4346 vcpu->arch.exception.error_code = events->exception.error_code;
4347 vcpu->arch.exception.has_payload = events->exception_has_payload;
4348 vcpu->arch.exception.payload = events->exception_payload;
4350 vcpu->arch.interrupt.injected = events->interrupt.injected;
4351 vcpu->arch.interrupt.nr = events->interrupt.nr;
4352 vcpu->arch.interrupt.soft = events->interrupt.soft;
4353 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4354 kvm_x86_ops.set_interrupt_shadow(vcpu,
4355 events->interrupt.shadow);
4357 vcpu->arch.nmi_injected = events->nmi.injected;
4358 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4359 vcpu->arch.nmi_pending = events->nmi.pending;
4360 kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
4362 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4363 lapic_in_kernel(vcpu))
4364 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4366 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4367 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4368 if (events->smi.smm)
4369 vcpu->arch.hflags |= HF_SMM_MASK;
4371 vcpu->arch.hflags &= ~HF_SMM_MASK;
4372 kvm_smm_changed(vcpu);
4375 vcpu->arch.smi_pending = events->smi.pending;
4377 if (events->smi.smm) {
4378 if (events->smi.smm_inside_nmi)
4379 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4381 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4384 if (lapic_in_kernel(vcpu)) {
4385 if (events->smi.latched_init)
4386 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4388 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4392 kvm_make_request(KVM_REQ_EVENT, vcpu);
4397 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4398 struct kvm_debugregs *dbgregs)
4402 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4403 kvm_get_dr(vcpu, 6, &val);
4405 dbgregs->dr7 = vcpu->arch.dr7;
4407 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4410 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4411 struct kvm_debugregs *dbgregs)
4416 if (dbgregs->dr6 & ~0xffffffffull)
4418 if (dbgregs->dr7 & ~0xffffffffull)
4421 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4422 kvm_update_dr0123(vcpu);
4423 vcpu->arch.dr6 = dbgregs->dr6;
4424 vcpu->arch.dr7 = dbgregs->dr7;
4425 kvm_update_dr7(vcpu);
4430 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4432 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4434 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4435 u64 xstate_bv = xsave->header.xfeatures;
4439 * Copy legacy XSAVE area, to avoid complications with CPUID
4440 * leaves 0 and 1 in the loop below.
4442 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4445 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4446 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4449 * Copy each region from the possibly compacted offset to the
4450 * non-compacted offset.
4452 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4454 u64 xfeature_mask = valid & -valid;
4455 int xfeature_nr = fls64(xfeature_mask) - 1;
4456 void *src = get_xsave_addr(xsave, xfeature_nr);
4459 u32 size, offset, ecx, edx;
4460 cpuid_count(XSTATE_CPUID, xfeature_nr,
4461 &size, &offset, &ecx, &edx);
4462 if (xfeature_nr == XFEATURE_PKRU)
4463 memcpy(dest + offset, &vcpu->arch.pkru,
4464 sizeof(vcpu->arch.pkru));
4466 memcpy(dest + offset, src, size);
4470 valid -= xfeature_mask;
4474 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4476 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4477 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4481 * Copy legacy XSAVE area, to avoid complications with CPUID
4482 * leaves 0 and 1 in the loop below.
4484 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4486 /* Set XSTATE_BV and possibly XCOMP_BV. */
4487 xsave->header.xfeatures = xstate_bv;
4488 if (boot_cpu_has(X86_FEATURE_XSAVES))
4489 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4492 * Copy each region from the non-compacted offset to the
4493 * possibly compacted offset.
4495 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4497 u64 xfeature_mask = valid & -valid;
4498 int xfeature_nr = fls64(xfeature_mask) - 1;
4499 void *dest = get_xsave_addr(xsave, xfeature_nr);
4502 u32 size, offset, ecx, edx;
4503 cpuid_count(XSTATE_CPUID, xfeature_nr,
4504 &size, &offset, &ecx, &edx);
4505 if (xfeature_nr == XFEATURE_PKRU)
4506 memcpy(&vcpu->arch.pkru, src + offset,
4507 sizeof(vcpu->arch.pkru));
4509 memcpy(dest, src + offset, size);
4512 valid -= xfeature_mask;
4516 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4517 struct kvm_xsave *guest_xsave)
4519 if (!vcpu->arch.guest_fpu)
4522 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4523 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4524 fill_xsave((u8 *) guest_xsave->region, vcpu);
4526 memcpy(guest_xsave->region,
4527 &vcpu->arch.guest_fpu->state.fxsave,
4528 sizeof(struct fxregs_state));
4529 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4530 XFEATURE_MASK_FPSSE;
4534 #define XSAVE_MXCSR_OFFSET 24
4536 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4537 struct kvm_xsave *guest_xsave)
4542 if (!vcpu->arch.guest_fpu)
4545 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4546 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4548 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4550 * Here we allow setting states that are not present in
4551 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4552 * with old userspace.
4554 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4556 load_xsave(vcpu, (u8 *)guest_xsave->region);
4558 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4559 mxcsr & ~mxcsr_feature_mask)
4561 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4562 guest_xsave->region, sizeof(struct fxregs_state));
4567 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4568 struct kvm_xcrs *guest_xcrs)
4570 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4571 guest_xcrs->nr_xcrs = 0;
4575 guest_xcrs->nr_xcrs = 1;
4576 guest_xcrs->flags = 0;
4577 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4578 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4581 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4582 struct kvm_xcrs *guest_xcrs)
4586 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4589 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4592 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4593 /* Only support XCR0 currently */
4594 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4595 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4596 guest_xcrs->xcrs[i].value);
4605 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4606 * stopped by the hypervisor. This function will be called from the host only.
4607 * EINVAL is returned when the host attempts to set the flag for a guest that
4608 * does not support pv clocks.
4610 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4612 if (!vcpu->arch.pv_time_enabled)
4614 vcpu->arch.pvclock_set_guest_stopped_request = true;
4615 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4619 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4620 struct kvm_enable_cap *cap)
4623 uint16_t vmcs_version;
4624 void __user *user_ptr;
4630 case KVM_CAP_HYPERV_SYNIC2:
4635 case KVM_CAP_HYPERV_SYNIC:
4636 if (!irqchip_in_kernel(vcpu->kvm))
4638 return kvm_hv_activate_synic(vcpu, cap->cap ==
4639 KVM_CAP_HYPERV_SYNIC2);
4640 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4641 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4643 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4645 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4646 if (copy_to_user(user_ptr, &vmcs_version,
4647 sizeof(vmcs_version)))
4651 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4652 if (!kvm_x86_ops.enable_direct_tlbflush)
4655 return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4657 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4658 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4659 if (vcpu->arch.pv_cpuid.enforce)
4660 kvm_update_pv_runtime(vcpu);
4669 long kvm_arch_vcpu_ioctl(struct file *filp,
4670 unsigned int ioctl, unsigned long arg)
4672 struct kvm_vcpu *vcpu = filp->private_data;
4673 void __user *argp = (void __user *)arg;
4676 struct kvm_lapic_state *lapic;
4677 struct kvm_xsave *xsave;
4678 struct kvm_xcrs *xcrs;
4686 case KVM_GET_LAPIC: {
4688 if (!lapic_in_kernel(vcpu))
4690 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4691 GFP_KERNEL_ACCOUNT);
4696 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4700 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4705 case KVM_SET_LAPIC: {
4707 if (!lapic_in_kernel(vcpu))
4709 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4710 if (IS_ERR(u.lapic)) {
4711 r = PTR_ERR(u.lapic);
4715 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4718 case KVM_INTERRUPT: {
4719 struct kvm_interrupt irq;
4722 if (copy_from_user(&irq, argp, sizeof(irq)))
4724 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4728 r = kvm_vcpu_ioctl_nmi(vcpu);
4732 r = kvm_vcpu_ioctl_smi(vcpu);
4735 case KVM_SET_CPUID: {
4736 struct kvm_cpuid __user *cpuid_arg = argp;
4737 struct kvm_cpuid cpuid;
4740 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4742 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4745 case KVM_SET_CPUID2: {
4746 struct kvm_cpuid2 __user *cpuid_arg = argp;
4747 struct kvm_cpuid2 cpuid;
4750 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4752 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4753 cpuid_arg->entries);
4756 case KVM_GET_CPUID2: {
4757 struct kvm_cpuid2 __user *cpuid_arg = argp;
4758 struct kvm_cpuid2 cpuid;
4761 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4763 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4764 cpuid_arg->entries);
4768 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4773 case KVM_GET_MSRS: {
4774 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4775 r = msr_io(vcpu, argp, do_get_msr, 1);
4776 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4779 case KVM_SET_MSRS: {
4780 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4781 r = msr_io(vcpu, argp, do_set_msr, 0);
4782 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4785 case KVM_TPR_ACCESS_REPORTING: {
4786 struct kvm_tpr_access_ctl tac;
4789 if (copy_from_user(&tac, argp, sizeof(tac)))
4791 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4795 if (copy_to_user(argp, &tac, sizeof(tac)))
4800 case KVM_SET_VAPIC_ADDR: {
4801 struct kvm_vapic_addr va;
4805 if (!lapic_in_kernel(vcpu))
4808 if (copy_from_user(&va, argp, sizeof(va)))
4810 idx = srcu_read_lock(&vcpu->kvm->srcu);
4811 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4812 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4815 case KVM_X86_SETUP_MCE: {
4819 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4821 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4824 case KVM_X86_SET_MCE: {
4825 struct kvm_x86_mce mce;
4828 if (copy_from_user(&mce, argp, sizeof(mce)))
4830 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4833 case KVM_GET_VCPU_EVENTS: {
4834 struct kvm_vcpu_events events;
4836 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4839 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4844 case KVM_SET_VCPU_EVENTS: {
4845 struct kvm_vcpu_events events;
4848 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4851 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4854 case KVM_GET_DEBUGREGS: {
4855 struct kvm_debugregs dbgregs;
4857 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4860 if (copy_to_user(argp, &dbgregs,
4861 sizeof(struct kvm_debugregs)))
4866 case KVM_SET_DEBUGREGS: {
4867 struct kvm_debugregs dbgregs;
4870 if (copy_from_user(&dbgregs, argp,
4871 sizeof(struct kvm_debugregs)))
4874 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4877 case KVM_GET_XSAVE: {
4878 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4883 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4886 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4891 case KVM_SET_XSAVE: {
4892 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4893 if (IS_ERR(u.xsave)) {
4894 r = PTR_ERR(u.xsave);
4898 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4901 case KVM_GET_XCRS: {
4902 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4907 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4910 if (copy_to_user(argp, u.xcrs,
4911 sizeof(struct kvm_xcrs)))
4916 case KVM_SET_XCRS: {
4917 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4918 if (IS_ERR(u.xcrs)) {
4919 r = PTR_ERR(u.xcrs);
4923 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4926 case KVM_SET_TSC_KHZ: {
4930 user_tsc_khz = (u32)arg;
4932 if (kvm_has_tsc_control &&
4933 user_tsc_khz >= kvm_max_guest_tsc_khz)
4936 if (user_tsc_khz == 0)
4937 user_tsc_khz = tsc_khz;
4939 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4944 case KVM_GET_TSC_KHZ: {
4945 r = vcpu->arch.virtual_tsc_khz;
4948 case KVM_KVMCLOCK_CTRL: {
4949 r = kvm_set_guest_paused(vcpu);
4952 case KVM_ENABLE_CAP: {
4953 struct kvm_enable_cap cap;
4956 if (copy_from_user(&cap, argp, sizeof(cap)))
4958 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4961 case KVM_GET_NESTED_STATE: {
4962 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4966 if (!kvm_x86_ops.nested_ops->get_state)
4969 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4971 if (get_user(user_data_size, &user_kvm_nested_state->size))
4974 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4979 if (r > user_data_size) {
4980 if (put_user(r, &user_kvm_nested_state->size))
4990 case KVM_SET_NESTED_STATE: {
4991 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4992 struct kvm_nested_state kvm_state;
4996 if (!kvm_x86_ops.nested_ops->set_state)
5000 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5004 if (kvm_state.size < sizeof(kvm_state))
5007 if (kvm_state.flags &
5008 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5009 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5010 | KVM_STATE_NESTED_GIF_SET))
5013 /* nested_run_pending implies guest_mode. */
5014 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5015 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5018 idx = srcu_read_lock(&vcpu->kvm->srcu);
5019 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5020 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5023 case KVM_GET_SUPPORTED_HV_CPUID:
5024 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5036 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5038 return VM_FAULT_SIGBUS;
5041 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5045 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5047 ret = kvm_x86_ops.set_tss_addr(kvm, addr);
5051 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5054 return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
5057 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5058 unsigned long kvm_nr_mmu_pages)
5060 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5063 mutex_lock(&kvm->slots_lock);
5065 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5066 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5068 mutex_unlock(&kvm->slots_lock);
5072 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5074 return kvm->arch.n_max_mmu_pages;
5077 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5079 struct kvm_pic *pic = kvm->arch.vpic;
5083 switch (chip->chip_id) {
5084 case KVM_IRQCHIP_PIC_MASTER:
5085 memcpy(&chip->chip.pic, &pic->pics[0],
5086 sizeof(struct kvm_pic_state));
5088 case KVM_IRQCHIP_PIC_SLAVE:
5089 memcpy(&chip->chip.pic, &pic->pics[1],
5090 sizeof(struct kvm_pic_state));
5092 case KVM_IRQCHIP_IOAPIC:
5093 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5102 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5104 struct kvm_pic *pic = kvm->arch.vpic;
5108 switch (chip->chip_id) {
5109 case KVM_IRQCHIP_PIC_MASTER:
5110 spin_lock(&pic->lock);
5111 memcpy(&pic->pics[0], &chip->chip.pic,
5112 sizeof(struct kvm_pic_state));
5113 spin_unlock(&pic->lock);
5115 case KVM_IRQCHIP_PIC_SLAVE:
5116 spin_lock(&pic->lock);
5117 memcpy(&pic->pics[1], &chip->chip.pic,
5118 sizeof(struct kvm_pic_state));
5119 spin_unlock(&pic->lock);
5121 case KVM_IRQCHIP_IOAPIC:
5122 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5128 kvm_pic_update_irq(pic);
5132 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5134 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5136 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5138 mutex_lock(&kps->lock);
5139 memcpy(ps, &kps->channels, sizeof(*ps));
5140 mutex_unlock(&kps->lock);
5144 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5147 struct kvm_pit *pit = kvm->arch.vpit;
5149 mutex_lock(&pit->pit_state.lock);
5150 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5151 for (i = 0; i < 3; i++)
5152 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5153 mutex_unlock(&pit->pit_state.lock);
5157 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5159 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5160 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5161 sizeof(ps->channels));
5162 ps->flags = kvm->arch.vpit->pit_state.flags;
5163 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5164 memset(&ps->reserved, 0, sizeof(ps->reserved));
5168 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5172 u32 prev_legacy, cur_legacy;
5173 struct kvm_pit *pit = kvm->arch.vpit;
5175 mutex_lock(&pit->pit_state.lock);
5176 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5177 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5178 if (!prev_legacy && cur_legacy)
5180 memcpy(&pit->pit_state.channels, &ps->channels,
5181 sizeof(pit->pit_state.channels));
5182 pit->pit_state.flags = ps->flags;
5183 for (i = 0; i < 3; i++)
5184 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5186 mutex_unlock(&pit->pit_state.lock);
5190 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5191 struct kvm_reinject_control *control)
5193 struct kvm_pit *pit = kvm->arch.vpit;
5195 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5196 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5197 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5199 mutex_lock(&pit->pit_state.lock);
5200 kvm_pit_set_reinject(pit, control->pit_reinject);
5201 mutex_unlock(&pit->pit_state.lock);
5206 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5209 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
5211 if (kvm_x86_ops.flush_log_dirty)
5212 kvm_x86_ops.flush_log_dirty(kvm);
5215 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5218 if (!irqchip_in_kernel(kvm))
5221 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5222 irq_event->irq, irq_event->level,
5227 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5228 struct kvm_enable_cap *cap)
5236 case KVM_CAP_DISABLE_QUIRKS:
5237 kvm->arch.disabled_quirks = cap->args[0];
5240 case KVM_CAP_SPLIT_IRQCHIP: {
5241 mutex_lock(&kvm->lock);
5243 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5244 goto split_irqchip_unlock;
5246 if (irqchip_in_kernel(kvm))
5247 goto split_irqchip_unlock;
5248 if (kvm->created_vcpus)
5249 goto split_irqchip_unlock;
5250 r = kvm_setup_empty_irq_routing(kvm);
5252 goto split_irqchip_unlock;
5253 /* Pairs with irqchip_in_kernel. */
5255 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5256 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5258 split_irqchip_unlock:
5259 mutex_unlock(&kvm->lock);
5262 case KVM_CAP_X2APIC_API:
5264 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5267 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5268 kvm->arch.x2apic_format = true;
5269 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5270 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5274 case KVM_CAP_X86_DISABLE_EXITS:
5276 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5279 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5280 kvm_can_mwait_in_guest())
5281 kvm->arch.mwait_in_guest = true;
5282 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5283 kvm->arch.hlt_in_guest = true;
5284 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5285 kvm->arch.pause_in_guest = true;
5286 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5287 kvm->arch.cstate_in_guest = true;
5290 case KVM_CAP_MSR_PLATFORM_INFO:
5291 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5294 case KVM_CAP_EXCEPTION_PAYLOAD:
5295 kvm->arch.exception_payload_enabled = cap->args[0];
5298 case KVM_CAP_X86_USER_SPACE_MSR:
5299 kvm->arch.user_space_msr_mask = cap->args[0];
5309 static void kvm_clear_msr_filter(struct kvm *kvm)
5312 u32 count = kvm->arch.msr_filter.count;
5313 struct msr_bitmap_range ranges[16];
5315 mutex_lock(&kvm->lock);
5316 kvm->arch.msr_filter.count = 0;
5317 memcpy(ranges, kvm->arch.msr_filter.ranges, count * sizeof(ranges[0]));
5318 mutex_unlock(&kvm->lock);
5319 synchronize_srcu(&kvm->srcu);
5321 for (i = 0; i < count; i++)
5322 kfree(ranges[i].bitmap);
5325 static int kvm_add_msr_filter(struct kvm *kvm, struct kvm_msr_filter_range *user_range)
5327 struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
5328 struct msr_bitmap_range range;
5329 unsigned long *bitmap = NULL;
5333 if (!user_range->nmsrs)
5336 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5337 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5340 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5342 return PTR_ERR(bitmap);
5344 range = (struct msr_bitmap_range) {
5345 .flags = user_range->flags,
5346 .base = user_range->base,
5347 .nmsrs = user_range->nmsrs,
5351 if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) {
5361 /* Everything ok, add this range identifier to our global pool */
5362 ranges[kvm->arch.msr_filter.count] = range;
5363 /* Make sure we filled the array before we tell anyone to walk it */
5365 kvm->arch.msr_filter.count++;
5373 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5375 struct kvm_msr_filter __user *user_msr_filter = argp;
5376 struct kvm_msr_filter filter;
5382 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5385 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5386 empty &= !filter.ranges[i].nmsrs;
5388 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5389 if (empty && !default_allow)
5392 kvm_clear_msr_filter(kvm);
5394 kvm->arch.msr_filter.default_allow = default_allow;
5397 * Protect from concurrent calls to this function that could trigger
5398 * a TOCTOU violation on kvm->arch.msr_filter.count.
5400 mutex_lock(&kvm->lock);
5401 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5402 r = kvm_add_msr_filter(kvm, &filter.ranges[i]);
5407 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5408 mutex_unlock(&kvm->lock);
5413 long kvm_arch_vm_ioctl(struct file *filp,
5414 unsigned int ioctl, unsigned long arg)
5416 struct kvm *kvm = filp->private_data;
5417 void __user *argp = (void __user *)arg;
5420 * This union makes it completely explicit to gcc-3.x
5421 * that these two variables' stack usage should be
5422 * combined, not added together.
5425 struct kvm_pit_state ps;
5426 struct kvm_pit_state2 ps2;
5427 struct kvm_pit_config pit_config;
5431 case KVM_SET_TSS_ADDR:
5432 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5434 case KVM_SET_IDENTITY_MAP_ADDR: {
5437 mutex_lock(&kvm->lock);
5439 if (kvm->created_vcpus)
5440 goto set_identity_unlock;
5442 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5443 goto set_identity_unlock;
5444 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5445 set_identity_unlock:
5446 mutex_unlock(&kvm->lock);
5449 case KVM_SET_NR_MMU_PAGES:
5450 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5452 case KVM_GET_NR_MMU_PAGES:
5453 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5455 case KVM_CREATE_IRQCHIP: {
5456 mutex_lock(&kvm->lock);
5459 if (irqchip_in_kernel(kvm))
5460 goto create_irqchip_unlock;
5463 if (kvm->created_vcpus)
5464 goto create_irqchip_unlock;
5466 r = kvm_pic_init(kvm);
5468 goto create_irqchip_unlock;
5470 r = kvm_ioapic_init(kvm);
5472 kvm_pic_destroy(kvm);
5473 goto create_irqchip_unlock;
5476 r = kvm_setup_default_irq_routing(kvm);
5478 kvm_ioapic_destroy(kvm);
5479 kvm_pic_destroy(kvm);
5480 goto create_irqchip_unlock;
5482 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5484 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5485 create_irqchip_unlock:
5486 mutex_unlock(&kvm->lock);
5489 case KVM_CREATE_PIT:
5490 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5492 case KVM_CREATE_PIT2:
5494 if (copy_from_user(&u.pit_config, argp,
5495 sizeof(struct kvm_pit_config)))
5498 mutex_lock(&kvm->lock);
5501 goto create_pit_unlock;
5503 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5507 mutex_unlock(&kvm->lock);
5509 case KVM_GET_IRQCHIP: {
5510 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5511 struct kvm_irqchip *chip;
5513 chip = memdup_user(argp, sizeof(*chip));
5520 if (!irqchip_kernel(kvm))
5521 goto get_irqchip_out;
5522 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5524 goto get_irqchip_out;
5526 if (copy_to_user(argp, chip, sizeof(*chip)))
5527 goto get_irqchip_out;
5533 case KVM_SET_IRQCHIP: {
5534 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5535 struct kvm_irqchip *chip;
5537 chip = memdup_user(argp, sizeof(*chip));
5544 if (!irqchip_kernel(kvm))
5545 goto set_irqchip_out;
5546 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5553 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5556 if (!kvm->arch.vpit)
5558 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5562 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5569 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5571 mutex_lock(&kvm->lock);
5573 if (!kvm->arch.vpit)
5575 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5577 mutex_unlock(&kvm->lock);
5580 case KVM_GET_PIT2: {
5582 if (!kvm->arch.vpit)
5584 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5588 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5593 case KVM_SET_PIT2: {
5595 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5597 mutex_lock(&kvm->lock);
5599 if (!kvm->arch.vpit)
5601 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5603 mutex_unlock(&kvm->lock);
5606 case KVM_REINJECT_CONTROL: {
5607 struct kvm_reinject_control control;
5609 if (copy_from_user(&control, argp, sizeof(control)))
5612 if (!kvm->arch.vpit)
5614 r = kvm_vm_ioctl_reinject(kvm, &control);
5617 case KVM_SET_BOOT_CPU_ID:
5619 mutex_lock(&kvm->lock);
5620 if (kvm->created_vcpus)
5623 kvm->arch.bsp_vcpu_id = arg;
5624 mutex_unlock(&kvm->lock);
5626 case KVM_XEN_HVM_CONFIG: {
5627 struct kvm_xen_hvm_config xhc;
5629 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5634 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5638 case KVM_SET_CLOCK: {
5639 struct kvm_clock_data user_ns;
5643 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5652 * TODO: userspace has to take care of races with VCPU_RUN, so
5653 * kvm_gen_update_masterclock() can be cut down to locked
5654 * pvclock_update_vm_gtod_copy().
5656 kvm_gen_update_masterclock(kvm);
5657 now_ns = get_kvmclock_ns(kvm);
5658 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5659 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5662 case KVM_GET_CLOCK: {
5663 struct kvm_clock_data user_ns;
5666 now_ns = get_kvmclock_ns(kvm);
5667 user_ns.clock = now_ns;
5668 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5669 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5672 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5677 case KVM_MEMORY_ENCRYPT_OP: {
5679 if (kvm_x86_ops.mem_enc_op)
5680 r = kvm_x86_ops.mem_enc_op(kvm, argp);
5683 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5684 struct kvm_enc_region region;
5687 if (copy_from_user(®ion, argp, sizeof(region)))
5691 if (kvm_x86_ops.mem_enc_reg_region)
5692 r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion);
5695 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5696 struct kvm_enc_region region;
5699 if (copy_from_user(®ion, argp, sizeof(region)))
5703 if (kvm_x86_ops.mem_enc_unreg_region)
5704 r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion);
5707 case KVM_HYPERV_EVENTFD: {
5708 struct kvm_hyperv_eventfd hvevfd;
5711 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5713 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5716 case KVM_SET_PMU_EVENT_FILTER:
5717 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5719 case KVM_X86_SET_MSR_FILTER:
5720 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5729 static void kvm_init_msr_list(void)
5731 struct x86_pmu_capability x86_pmu;
5735 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5736 "Please update the fixed PMCs in msrs_to_saved_all[]");
5738 perf_get_x86_pmu_capability(&x86_pmu);
5740 num_msrs_to_save = 0;
5741 num_emulated_msrs = 0;
5742 num_msr_based_features = 0;
5744 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5745 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5749 * Even MSRs that are valid in the host may not be exposed
5750 * to the guests in some cases.
5752 switch (msrs_to_save_all[i]) {
5753 case MSR_IA32_BNDCFGS:
5754 if (!kvm_mpx_supported())
5758 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5761 case MSR_IA32_UMWAIT_CONTROL:
5762 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5765 case MSR_IA32_RTIT_CTL:
5766 case MSR_IA32_RTIT_STATUS:
5767 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5770 case MSR_IA32_RTIT_CR3_MATCH:
5771 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5772 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5775 case MSR_IA32_RTIT_OUTPUT_BASE:
5776 case MSR_IA32_RTIT_OUTPUT_MASK:
5777 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5778 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5779 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5782 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5783 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5784 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5785 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5788 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5789 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5790 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5793 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5794 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5795 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5802 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5805 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5806 if (!kvm_x86_ops.has_emulated_msr(NULL, emulated_msrs_all[i]))
5809 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5812 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5813 struct kvm_msr_entry msr;
5815 msr.index = msr_based_features_all[i];
5816 if (kvm_get_msr_feature(&msr))
5819 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5823 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5831 if (!(lapic_in_kernel(vcpu) &&
5832 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5833 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5844 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5851 if (!(lapic_in_kernel(vcpu) &&
5852 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5854 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5856 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5866 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5867 struct kvm_segment *var, int seg)
5869 kvm_x86_ops.set_segment(vcpu, var, seg);
5872 void kvm_get_segment(struct kvm_vcpu *vcpu,
5873 struct kvm_segment *var, int seg)
5875 kvm_x86_ops.get_segment(vcpu, var, seg);
5878 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5879 struct x86_exception *exception)
5883 BUG_ON(!mmu_is_nested(vcpu));
5885 /* NPT walks are always user-walks */
5886 access |= PFERR_USER_MASK;
5887 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5892 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5893 struct x86_exception *exception)
5895 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5896 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5899 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5900 struct x86_exception *exception)
5902 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5903 access |= PFERR_FETCH_MASK;
5904 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5907 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5908 struct x86_exception *exception)
5910 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5911 access |= PFERR_WRITE_MASK;
5912 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5915 /* uses this to access any guest's mapped memory without checking CPL */
5916 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5917 struct x86_exception *exception)
5919 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5922 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5923 struct kvm_vcpu *vcpu, u32 access,
5924 struct x86_exception *exception)
5927 int r = X86EMUL_CONTINUE;
5930 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5932 unsigned offset = addr & (PAGE_SIZE-1);
5933 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5936 if (gpa == UNMAPPED_GVA)
5937 return X86EMUL_PROPAGATE_FAULT;
5938 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5941 r = X86EMUL_IO_NEEDED;
5953 /* used for instruction fetching */
5954 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5955 gva_t addr, void *val, unsigned int bytes,
5956 struct x86_exception *exception)
5958 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5959 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5963 /* Inline kvm_read_guest_virt_helper for speed. */
5964 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5966 if (unlikely(gpa == UNMAPPED_GVA))
5967 return X86EMUL_PROPAGATE_FAULT;
5969 offset = addr & (PAGE_SIZE-1);
5970 if (WARN_ON(offset + bytes > PAGE_SIZE))
5971 bytes = (unsigned)PAGE_SIZE - offset;
5972 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5974 if (unlikely(ret < 0))
5975 return X86EMUL_IO_NEEDED;
5977 return X86EMUL_CONTINUE;
5980 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5981 gva_t addr, void *val, unsigned int bytes,
5982 struct x86_exception *exception)
5984 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5987 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5988 * is returned, but our callers are not ready for that and they blindly
5989 * call kvm_inject_page_fault. Ensure that they at least do not leak
5990 * uninitialized kernel stack memory into cr2 and error code.
5992 memset(exception, 0, sizeof(*exception));
5993 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5996 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5998 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5999 gva_t addr, void *val, unsigned int bytes,
6000 struct x86_exception *exception, bool system)
6002 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6005 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
6006 access |= PFERR_USER_MASK;
6008 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6011 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6012 unsigned long addr, void *val, unsigned int bytes)
6014 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6015 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6017 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6020 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6021 struct kvm_vcpu *vcpu, u32 access,
6022 struct x86_exception *exception)
6025 int r = X86EMUL_CONTINUE;
6028 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6031 unsigned offset = addr & (PAGE_SIZE-1);
6032 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6035 if (gpa == UNMAPPED_GVA)
6036 return X86EMUL_PROPAGATE_FAULT;
6037 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6039 r = X86EMUL_IO_NEEDED;
6051 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6052 unsigned int bytes, struct x86_exception *exception,
6055 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6056 u32 access = PFERR_WRITE_MASK;
6058 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
6059 access |= PFERR_USER_MASK;
6061 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6065 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6066 unsigned int bytes, struct x86_exception *exception)
6068 /* kvm_write_guest_virt_system can pull in tons of pages. */
6069 vcpu->arch.l1tf_flush_l1d = true;
6071 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6072 PFERR_WRITE_MASK, exception);
6074 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6076 int handle_ud(struct kvm_vcpu *vcpu)
6078 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6079 int emul_type = EMULTYPE_TRAP_UD;
6080 char sig[5]; /* ud2; .ascii "kvm" */
6081 struct x86_exception e;
6083 if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, NULL, 0)))
6086 if (force_emulation_prefix &&
6087 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6088 sig, sizeof(sig), &e) == 0 &&
6089 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6090 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6091 emul_type = EMULTYPE_TRAP_UD_FORCED;
6094 return kvm_emulate_instruction(vcpu, emul_type);
6096 EXPORT_SYMBOL_GPL(handle_ud);
6098 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6099 gpa_t gpa, bool write)
6101 /* For APIC access vmexit */
6102 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6105 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6106 trace_vcpu_match_mmio(gva, gpa, write, true);
6113 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6114 gpa_t *gpa, struct x86_exception *exception,
6117 u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
6118 | (write ? PFERR_WRITE_MASK : 0);
6121 * currently PKRU is only applied to ept enabled guest so
6122 * there is no pkey in EPT page table for L1 guest or EPT
6123 * shadow page table for L2 guest.
6125 if (vcpu_match_mmio_gva(vcpu, gva)
6126 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6127 vcpu->arch.mmio_access, 0, access)) {
6128 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6129 (gva & (PAGE_SIZE - 1));
6130 trace_vcpu_match_mmio(gva, *gpa, write, false);
6134 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6136 if (*gpa == UNMAPPED_GVA)
6139 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6142 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6143 const void *val, int bytes)
6147 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6150 kvm_page_track_write(vcpu, gpa, val, bytes);
6154 struct read_write_emulator_ops {
6155 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6157 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6158 void *val, int bytes);
6159 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6160 int bytes, void *val);
6161 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6162 void *val, int bytes);
6166 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6168 if (vcpu->mmio_read_completed) {
6169 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6170 vcpu->mmio_fragments[0].gpa, val);
6171 vcpu->mmio_read_completed = 0;
6178 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6179 void *val, int bytes)
6181 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6184 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6185 void *val, int bytes)
6187 return emulator_write_phys(vcpu, gpa, val, bytes);
6190 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6192 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6193 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6196 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6197 void *val, int bytes)
6199 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6200 return X86EMUL_IO_NEEDED;
6203 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6204 void *val, int bytes)
6206 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6208 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6209 return X86EMUL_CONTINUE;
6212 static const struct read_write_emulator_ops read_emultor = {
6213 .read_write_prepare = read_prepare,
6214 .read_write_emulate = read_emulate,
6215 .read_write_mmio = vcpu_mmio_read,
6216 .read_write_exit_mmio = read_exit_mmio,
6219 static const struct read_write_emulator_ops write_emultor = {
6220 .read_write_emulate = write_emulate,
6221 .read_write_mmio = write_mmio,
6222 .read_write_exit_mmio = write_exit_mmio,
6226 static int emulator_read_write_onepage(unsigned long addr, void *val,
6228 struct x86_exception *exception,
6229 struct kvm_vcpu *vcpu,
6230 const struct read_write_emulator_ops *ops)
6234 bool write = ops->write;
6235 struct kvm_mmio_fragment *frag;
6236 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6239 * If the exit was due to a NPF we may already have a GPA.
6240 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6241 * Note, this cannot be used on string operations since string
6242 * operation using rep will only have the initial GPA from the NPF
6245 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6246 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6247 gpa = ctxt->gpa_val;
6248 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6250 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6252 return X86EMUL_PROPAGATE_FAULT;
6255 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6256 return X86EMUL_CONTINUE;
6259 * Is this MMIO handled locally?
6261 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6262 if (handled == bytes)
6263 return X86EMUL_CONTINUE;
6269 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6270 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6274 return X86EMUL_CONTINUE;
6277 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6279 void *val, unsigned int bytes,
6280 struct x86_exception *exception,
6281 const struct read_write_emulator_ops *ops)
6283 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6287 if (ops->read_write_prepare &&
6288 ops->read_write_prepare(vcpu, val, bytes))
6289 return X86EMUL_CONTINUE;
6291 vcpu->mmio_nr_fragments = 0;
6293 /* Crossing a page boundary? */
6294 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6297 now = -addr & ~PAGE_MASK;
6298 rc = emulator_read_write_onepage(addr, val, now, exception,
6301 if (rc != X86EMUL_CONTINUE)
6304 if (ctxt->mode != X86EMUL_MODE_PROT64)
6310 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6312 if (rc != X86EMUL_CONTINUE)
6315 if (!vcpu->mmio_nr_fragments)
6318 gpa = vcpu->mmio_fragments[0].gpa;
6320 vcpu->mmio_needed = 1;
6321 vcpu->mmio_cur_fragment = 0;
6323 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6324 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6325 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6326 vcpu->run->mmio.phys_addr = gpa;
6328 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6331 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6335 struct x86_exception *exception)
6337 return emulator_read_write(ctxt, addr, val, bytes,
6338 exception, &read_emultor);
6341 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6345 struct x86_exception *exception)
6347 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6348 exception, &write_emultor);
6351 #define CMPXCHG_TYPE(t, ptr, old, new) \
6352 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6354 #ifdef CONFIG_X86_64
6355 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6357 # define CMPXCHG64(ptr, old, new) \
6358 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6361 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6366 struct x86_exception *exception)
6368 struct kvm_host_map map;
6369 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6375 /* guests cmpxchg8b have to be emulated atomically */
6376 if (bytes > 8 || (bytes & (bytes - 1)))
6379 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6381 if (gpa == UNMAPPED_GVA ||
6382 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6386 * Emulate the atomic as a straight write to avoid #AC if SLD is
6387 * enabled in the host and the access splits a cache line.
6389 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6390 page_line_mask = ~(cache_line_size() - 1);
6392 page_line_mask = PAGE_MASK;
6394 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6397 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6400 kaddr = map.hva + offset_in_page(gpa);
6404 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6407 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6410 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6413 exchanged = CMPXCHG64(kaddr, old, new);
6419 kvm_vcpu_unmap(vcpu, &map, true);
6422 return X86EMUL_CMPXCHG_FAILED;
6424 kvm_page_track_write(vcpu, gpa, new, bytes);
6426 return X86EMUL_CONTINUE;
6429 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6431 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6434 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6438 for (i = 0; i < vcpu->arch.pio.count; i++) {
6439 if (vcpu->arch.pio.in)
6440 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6441 vcpu->arch.pio.size, pd);
6443 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6444 vcpu->arch.pio.port, vcpu->arch.pio.size,
6448 pd += vcpu->arch.pio.size;
6453 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6454 unsigned short port, void *val,
6455 unsigned int count, bool in)
6457 vcpu->arch.pio.port = port;
6458 vcpu->arch.pio.in = in;
6459 vcpu->arch.pio.count = count;
6460 vcpu->arch.pio.size = size;
6462 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6463 vcpu->arch.pio.count = 0;
6467 vcpu->run->exit_reason = KVM_EXIT_IO;
6468 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6469 vcpu->run->io.size = size;
6470 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6471 vcpu->run->io.count = count;
6472 vcpu->run->io.port = port;
6477 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6478 unsigned short port, void *val, unsigned int count)
6482 if (vcpu->arch.pio.count)
6485 memset(vcpu->arch.pio_data, 0, size * count);
6487 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6490 memcpy(val, vcpu->arch.pio_data, size * count);
6491 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6492 vcpu->arch.pio.count = 0;
6499 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6500 int size, unsigned short port, void *val,
6503 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6507 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6508 unsigned short port, const void *val,
6511 memcpy(vcpu->arch.pio_data, val, size * count);
6512 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6513 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6516 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6517 int size, unsigned short port,
6518 const void *val, unsigned int count)
6520 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6523 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6525 return kvm_x86_ops.get_segment_base(vcpu, seg);
6528 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6530 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6533 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6535 if (!need_emulate_wbinvd(vcpu))
6536 return X86EMUL_CONTINUE;
6538 if (kvm_x86_ops.has_wbinvd_exit()) {
6539 int cpu = get_cpu();
6541 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6542 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6543 wbinvd_ipi, NULL, 1);
6545 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6548 return X86EMUL_CONTINUE;
6551 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6553 kvm_emulate_wbinvd_noskip(vcpu);
6554 return kvm_skip_emulated_instruction(vcpu);
6556 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6560 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6562 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6565 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6566 unsigned long *dest)
6568 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6571 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6572 unsigned long value)
6575 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6578 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6580 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6583 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6585 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6586 unsigned long value;
6590 value = kvm_read_cr0(vcpu);
6593 value = vcpu->arch.cr2;
6596 value = kvm_read_cr3(vcpu);
6599 value = kvm_read_cr4(vcpu);
6602 value = kvm_get_cr8(vcpu);
6605 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6612 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6614 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6619 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6622 vcpu->arch.cr2 = val;
6625 res = kvm_set_cr3(vcpu, val);
6628 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6631 res = kvm_set_cr8(vcpu, val);
6634 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6641 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6643 return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6646 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6648 kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6651 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6653 kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6656 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6658 kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6661 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6663 kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6666 static unsigned long emulator_get_cached_segment_base(
6667 struct x86_emulate_ctxt *ctxt, int seg)
6669 return get_segment_base(emul_to_vcpu(ctxt), seg);
6672 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6673 struct desc_struct *desc, u32 *base3,
6676 struct kvm_segment var;
6678 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6679 *selector = var.selector;
6682 memset(desc, 0, sizeof(*desc));
6690 set_desc_limit(desc, var.limit);
6691 set_desc_base(desc, (unsigned long)var.base);
6692 #ifdef CONFIG_X86_64
6694 *base3 = var.base >> 32;
6696 desc->type = var.type;
6698 desc->dpl = var.dpl;
6699 desc->p = var.present;
6700 desc->avl = var.avl;
6708 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6709 struct desc_struct *desc, u32 base3,
6712 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6713 struct kvm_segment var;
6715 var.selector = selector;
6716 var.base = get_desc_base(desc);
6717 #ifdef CONFIG_X86_64
6718 var.base |= ((u64)base3) << 32;
6720 var.limit = get_desc_limit(desc);
6722 var.limit = (var.limit << 12) | 0xfff;
6723 var.type = desc->type;
6724 var.dpl = desc->dpl;
6729 var.avl = desc->avl;
6730 var.present = desc->p;
6731 var.unusable = !var.present;
6734 kvm_set_segment(vcpu, &var, seg);
6738 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6739 u32 msr_index, u64 *pdata)
6741 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6744 r = kvm_get_msr(vcpu, msr_index, pdata);
6746 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6747 /* Bounce to user space */
6748 return X86EMUL_IO_NEEDED;
6754 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6755 u32 msr_index, u64 data)
6757 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6760 r = kvm_set_msr(vcpu, msr_index, data);
6762 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6763 /* Bounce to user space */
6764 return X86EMUL_IO_NEEDED;
6770 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6772 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6774 return vcpu->arch.smbase;
6777 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6779 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6781 vcpu->arch.smbase = smbase;
6784 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6787 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6790 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6791 u32 pmc, u64 *pdata)
6793 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6796 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6798 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6801 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6802 struct x86_instruction_info *info,
6803 enum x86_intercept_stage stage)
6805 return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6809 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6810 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6813 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6816 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6818 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6821 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6823 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6826 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6828 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6831 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6833 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6836 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6838 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6841 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6843 kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6846 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6848 return emul_to_vcpu(ctxt)->arch.hflags;
6851 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6853 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6856 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6857 const char *smstate)
6859 return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6862 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6864 kvm_smm_changed(emul_to_vcpu(ctxt));
6867 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6869 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6872 static const struct x86_emulate_ops emulate_ops = {
6873 .read_gpr = emulator_read_gpr,
6874 .write_gpr = emulator_write_gpr,
6875 .read_std = emulator_read_std,
6876 .write_std = emulator_write_std,
6877 .read_phys = kvm_read_guest_phys_system,
6878 .fetch = kvm_fetch_guest_virt,
6879 .read_emulated = emulator_read_emulated,
6880 .write_emulated = emulator_write_emulated,
6881 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6882 .invlpg = emulator_invlpg,
6883 .pio_in_emulated = emulator_pio_in_emulated,
6884 .pio_out_emulated = emulator_pio_out_emulated,
6885 .get_segment = emulator_get_segment,
6886 .set_segment = emulator_set_segment,
6887 .get_cached_segment_base = emulator_get_cached_segment_base,
6888 .get_gdt = emulator_get_gdt,
6889 .get_idt = emulator_get_idt,
6890 .set_gdt = emulator_set_gdt,
6891 .set_idt = emulator_set_idt,
6892 .get_cr = emulator_get_cr,
6893 .set_cr = emulator_set_cr,
6894 .cpl = emulator_get_cpl,
6895 .get_dr = emulator_get_dr,
6896 .set_dr = emulator_set_dr,
6897 .get_smbase = emulator_get_smbase,
6898 .set_smbase = emulator_set_smbase,
6899 .set_msr = emulator_set_msr,
6900 .get_msr = emulator_get_msr,
6901 .check_pmc = emulator_check_pmc,
6902 .read_pmc = emulator_read_pmc,
6903 .halt = emulator_halt,
6904 .wbinvd = emulator_wbinvd,
6905 .fix_hypercall = emulator_fix_hypercall,
6906 .intercept = emulator_intercept,
6907 .get_cpuid = emulator_get_cpuid,
6908 .guest_has_long_mode = emulator_guest_has_long_mode,
6909 .guest_has_movbe = emulator_guest_has_movbe,
6910 .guest_has_fxsr = emulator_guest_has_fxsr,
6911 .set_nmi_mask = emulator_set_nmi_mask,
6912 .get_hflags = emulator_get_hflags,
6913 .set_hflags = emulator_set_hflags,
6914 .pre_leave_smm = emulator_pre_leave_smm,
6915 .post_leave_smm = emulator_post_leave_smm,
6916 .set_xcr = emulator_set_xcr,
6919 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6921 u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6923 * an sti; sti; sequence only disable interrupts for the first
6924 * instruction. So, if the last instruction, be it emulated or
6925 * not, left the system with the INT_STI flag enabled, it
6926 * means that the last instruction is an sti. We should not
6927 * leave the flag on in this case. The same goes for mov ss
6929 if (int_shadow & mask)
6931 if (unlikely(int_shadow || mask)) {
6932 kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6934 kvm_make_request(KVM_REQ_EVENT, vcpu);
6938 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6940 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6941 if (ctxt->exception.vector == PF_VECTOR)
6942 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
6944 if (ctxt->exception.error_code_valid)
6945 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6946 ctxt->exception.error_code);
6948 kvm_queue_exception(vcpu, ctxt->exception.vector);
6952 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6954 struct x86_emulate_ctxt *ctxt;
6956 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6958 pr_err("kvm: failed to allocate vcpu's emulator\n");
6963 ctxt->ops = &emulate_ops;
6964 vcpu->arch.emulate_ctxt = ctxt;
6969 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6971 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6974 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6976 ctxt->gpa_available = false;
6977 ctxt->eflags = kvm_get_rflags(vcpu);
6978 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6980 ctxt->eip = kvm_rip_read(vcpu);
6981 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6982 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6983 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6984 cs_db ? X86EMUL_MODE_PROT32 :
6985 X86EMUL_MODE_PROT16;
6986 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6987 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6988 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6990 init_decode_cache(ctxt);
6991 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6994 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6996 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6999 init_emulate_ctxt(vcpu);
7003 ctxt->_eip = ctxt->eip + inc_eip;
7004 ret = emulate_int_real(ctxt, irq);
7006 if (ret != X86EMUL_CONTINUE) {
7007 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7009 ctxt->eip = ctxt->_eip;
7010 kvm_rip_write(vcpu, ctxt->eip);
7011 kvm_set_rflags(vcpu, ctxt->eflags);
7014 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7016 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7018 ++vcpu->stat.insn_emulation_fail;
7019 trace_kvm_emulate_insn_failed(vcpu);
7021 if (emulation_type & EMULTYPE_VMWARE_GP) {
7022 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7026 if (emulation_type & EMULTYPE_SKIP) {
7027 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7028 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7029 vcpu->run->internal.ndata = 0;
7033 kvm_queue_exception(vcpu, UD_VECTOR);
7035 if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
7036 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7037 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7038 vcpu->run->internal.ndata = 0;
7045 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7046 bool write_fault_to_shadow_pgtable,
7049 gpa_t gpa = cr2_or_gpa;
7052 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7055 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7056 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7059 if (!vcpu->arch.mmu->direct_map) {
7061 * Write permission should be allowed since only
7062 * write access need to be emulated.
7064 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7067 * If the mapping is invalid in guest, let cpu retry
7068 * it to generate fault.
7070 if (gpa == UNMAPPED_GVA)
7075 * Do not retry the unhandleable instruction if it faults on the
7076 * readonly host memory, otherwise it will goto a infinite loop:
7077 * retry instruction -> write #PF -> emulation fail -> retry
7078 * instruction -> ...
7080 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7083 * If the instruction failed on the error pfn, it can not be fixed,
7084 * report the error to userspace.
7086 if (is_error_noslot_pfn(pfn))
7089 kvm_release_pfn_clean(pfn);
7091 /* The instructions are well-emulated on direct mmu. */
7092 if (vcpu->arch.mmu->direct_map) {
7093 unsigned int indirect_shadow_pages;
7095 spin_lock(&vcpu->kvm->mmu_lock);
7096 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7097 spin_unlock(&vcpu->kvm->mmu_lock);
7099 if (indirect_shadow_pages)
7100 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7106 * if emulation was due to access to shadowed page table
7107 * and it failed try to unshadow page and re-enter the
7108 * guest to let CPU execute the instruction.
7110 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7113 * If the access faults on its page table, it can not
7114 * be fixed by unprotecting shadow page and it should
7115 * be reported to userspace.
7117 return !write_fault_to_shadow_pgtable;
7120 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7121 gpa_t cr2_or_gpa, int emulation_type)
7123 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7124 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7126 last_retry_eip = vcpu->arch.last_retry_eip;
7127 last_retry_addr = vcpu->arch.last_retry_addr;
7130 * If the emulation is caused by #PF and it is non-page_table
7131 * writing instruction, it means the VM-EXIT is caused by shadow
7132 * page protected, we can zap the shadow page and retry this
7133 * instruction directly.
7135 * Note: if the guest uses a non-page-table modifying instruction
7136 * on the PDE that points to the instruction, then we will unmap
7137 * the instruction and go to an infinite loop. So, we cache the
7138 * last retried eip and the last fault address, if we meet the eip
7139 * and the address again, we can break out of the potential infinite
7142 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7144 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7147 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7148 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7151 if (x86_page_table_writing_insn(ctxt))
7154 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7157 vcpu->arch.last_retry_eip = ctxt->eip;
7158 vcpu->arch.last_retry_addr = cr2_or_gpa;
7160 if (!vcpu->arch.mmu->direct_map)
7161 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7163 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7168 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7169 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7171 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7173 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7174 /* This is a good place to trace that we are exiting SMM. */
7175 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7177 /* Process a latched INIT or SMI, if any. */
7178 kvm_make_request(KVM_REQ_EVENT, vcpu);
7181 kvm_mmu_reset_context(vcpu);
7184 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7193 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7194 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7199 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7201 struct kvm_run *kvm_run = vcpu->run;
7203 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7204 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
7205 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7206 kvm_run->debug.arch.exception = DB_VECTOR;
7207 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7210 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7214 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7216 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7219 r = kvm_x86_ops.skip_emulated_instruction(vcpu);
7224 * rflags is the old, "raw" value of the flags. The new value has
7225 * not been saved yet.
7227 * This is correct even for TF set by the guest, because "the
7228 * processor will not generate this exception after the instruction
7229 * that sets the TF flag".
7231 if (unlikely(rflags & X86_EFLAGS_TF))
7232 r = kvm_vcpu_do_singlestep(vcpu);
7235 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7237 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7239 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7240 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7241 struct kvm_run *kvm_run = vcpu->run;
7242 unsigned long eip = kvm_get_linear_rip(vcpu);
7243 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7244 vcpu->arch.guest_debug_dr7,
7248 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
7249 kvm_run->debug.arch.pc = eip;
7250 kvm_run->debug.arch.exception = DB_VECTOR;
7251 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7257 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7258 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7259 unsigned long eip = kvm_get_linear_rip(vcpu);
7260 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7265 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7274 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7276 switch (ctxt->opcode_len) {
7283 case 0xe6: /* OUT */
7287 case 0x6c: /* INS */
7289 case 0x6e: /* OUTS */
7296 case 0x33: /* RDPMC */
7305 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7306 int emulation_type, void *insn, int insn_len)
7309 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7310 bool writeback = true;
7311 bool write_fault_to_spt;
7313 if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, insn, insn_len)))
7316 vcpu->arch.l1tf_flush_l1d = true;
7319 * Clear write_fault_to_shadow_pgtable here to ensure it is
7322 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7323 vcpu->arch.write_fault_to_shadow_pgtable = false;
7324 kvm_clear_exception_queue(vcpu);
7326 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7327 init_emulate_ctxt(vcpu);
7330 * We will reenter on the same instruction since
7331 * we do not set complete_userspace_io. This does not
7332 * handle watchpoints yet, those would be handled in
7335 if (!(emulation_type & EMULTYPE_SKIP) &&
7336 kvm_vcpu_check_breakpoint(vcpu, &r))
7339 ctxt->interruptibility = 0;
7340 ctxt->have_exception = false;
7341 ctxt->exception.vector = -1;
7342 ctxt->perm_ok = false;
7344 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7346 r = x86_decode_insn(ctxt, insn, insn_len);
7348 trace_kvm_emulate_insn_start(vcpu);
7349 ++vcpu->stat.insn_emulation;
7350 if (r != EMULATION_OK) {
7351 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7352 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7353 kvm_queue_exception(vcpu, UD_VECTOR);
7356 if (reexecute_instruction(vcpu, cr2_or_gpa,
7360 if (ctxt->have_exception) {
7362 * #UD should result in just EMULATION_FAILED, and trap-like
7363 * exception should not be encountered during decode.
7365 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7366 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7367 inject_emulated_exception(vcpu);
7370 return handle_emulation_failure(vcpu, emulation_type);
7374 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7375 !is_vmware_backdoor_opcode(ctxt)) {
7376 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7381 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7382 * for kvm_skip_emulated_instruction(). The caller is responsible for
7383 * updating interruptibility state and injecting single-step #DBs.
7385 if (emulation_type & EMULTYPE_SKIP) {
7386 kvm_rip_write(vcpu, ctxt->_eip);
7387 if (ctxt->eflags & X86_EFLAGS_RF)
7388 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7392 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7395 /* this is needed for vmware backdoor interface to work since it
7396 changes registers values during IO operation */
7397 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7398 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7399 emulator_invalidate_register_cache(ctxt);
7403 if (emulation_type & EMULTYPE_PF) {
7404 /* Save the faulting GPA (cr2) in the address field */
7405 ctxt->exception.address = cr2_or_gpa;
7407 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7408 if (vcpu->arch.mmu->direct_map) {
7409 ctxt->gpa_available = true;
7410 ctxt->gpa_val = cr2_or_gpa;
7413 /* Sanitize the address out of an abundance of paranoia. */
7414 ctxt->exception.address = 0;
7417 r = x86_emulate_insn(ctxt);
7419 if (r == EMULATION_INTERCEPTED)
7422 if (r == EMULATION_FAILED) {
7423 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7427 return handle_emulation_failure(vcpu, emulation_type);
7430 if (ctxt->have_exception) {
7432 if (inject_emulated_exception(vcpu))
7434 } else if (vcpu->arch.pio.count) {
7435 if (!vcpu->arch.pio.in) {
7436 /* FIXME: return into emulator if single-stepping. */
7437 vcpu->arch.pio.count = 0;
7440 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7443 } else if (vcpu->mmio_needed) {
7444 ++vcpu->stat.mmio_exits;
7446 if (!vcpu->mmio_is_write)
7449 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7450 } else if (r == EMULATION_RESTART)
7456 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7457 toggle_interruptibility(vcpu, ctxt->interruptibility);
7458 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7459 if (!ctxt->have_exception ||
7460 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7461 kvm_rip_write(vcpu, ctxt->eip);
7462 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7463 r = kvm_vcpu_do_singlestep(vcpu);
7464 if (kvm_x86_ops.update_emulated_instruction)
7465 kvm_x86_ops.update_emulated_instruction(vcpu);
7466 __kvm_set_rflags(vcpu, ctxt->eflags);
7470 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7471 * do nothing, and it will be requested again as soon as
7472 * the shadow expires. But we still need to check here,
7473 * because POPF has no interrupt shadow.
7475 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7476 kvm_make_request(KVM_REQ_EVENT, vcpu);
7478 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7483 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7485 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7487 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7489 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7490 void *insn, int insn_len)
7492 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7494 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7496 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7498 vcpu->arch.pio.count = 0;
7502 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7504 vcpu->arch.pio.count = 0;
7506 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7509 return kvm_skip_emulated_instruction(vcpu);
7512 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7513 unsigned short port)
7515 unsigned long val = kvm_rax_read(vcpu);
7516 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7522 * Workaround userspace that relies on old KVM behavior of %rip being
7523 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7526 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7527 vcpu->arch.complete_userspace_io =
7528 complete_fast_pio_out_port_0x7e;
7529 kvm_skip_emulated_instruction(vcpu);
7531 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7532 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7537 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7541 /* We should only ever be called with arch.pio.count equal to 1 */
7542 BUG_ON(vcpu->arch.pio.count != 1);
7544 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7545 vcpu->arch.pio.count = 0;
7549 /* For size less than 4 we merge, else we zero extend */
7550 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7553 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7554 * the copy and tracing
7556 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7557 kvm_rax_write(vcpu, val);
7559 return kvm_skip_emulated_instruction(vcpu);
7562 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7563 unsigned short port)
7568 /* For size less than 4 we merge, else we zero extend */
7569 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7571 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7573 kvm_rax_write(vcpu, val);
7577 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7578 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7583 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7588 ret = kvm_fast_pio_in(vcpu, size, port);
7590 ret = kvm_fast_pio_out(vcpu, size, port);
7591 return ret && kvm_skip_emulated_instruction(vcpu);
7593 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7595 static int kvmclock_cpu_down_prep(unsigned int cpu)
7597 __this_cpu_write(cpu_tsc_khz, 0);
7601 static void tsc_khz_changed(void *data)
7603 struct cpufreq_freqs *freq = data;
7604 unsigned long khz = 0;
7608 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7609 khz = cpufreq_quick_get(raw_smp_processor_id());
7612 __this_cpu_write(cpu_tsc_khz, khz);
7615 #ifdef CONFIG_X86_64
7616 static void kvm_hyperv_tsc_notifier(void)
7619 struct kvm_vcpu *vcpu;
7622 mutex_lock(&kvm_lock);
7623 list_for_each_entry(kvm, &vm_list, vm_list)
7624 kvm_make_mclock_inprogress_request(kvm);
7626 hyperv_stop_tsc_emulation();
7628 /* TSC frequency always matches when on Hyper-V */
7629 for_each_present_cpu(cpu)
7630 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7631 kvm_max_guest_tsc_khz = tsc_khz;
7633 list_for_each_entry(kvm, &vm_list, vm_list) {
7634 struct kvm_arch *ka = &kvm->arch;
7636 spin_lock(&ka->pvclock_gtod_sync_lock);
7638 pvclock_update_vm_gtod_copy(kvm);
7640 kvm_for_each_vcpu(cpu, vcpu, kvm)
7641 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7643 kvm_for_each_vcpu(cpu, vcpu, kvm)
7644 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7646 spin_unlock(&ka->pvclock_gtod_sync_lock);
7648 mutex_unlock(&kvm_lock);
7652 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7655 struct kvm_vcpu *vcpu;
7656 int i, send_ipi = 0;
7659 * We allow guests to temporarily run on slowing clocks,
7660 * provided we notify them after, or to run on accelerating
7661 * clocks, provided we notify them before. Thus time never
7664 * However, we have a problem. We can't atomically update
7665 * the frequency of a given CPU from this function; it is
7666 * merely a notifier, which can be called from any CPU.
7667 * Changing the TSC frequency at arbitrary points in time
7668 * requires a recomputation of local variables related to
7669 * the TSC for each VCPU. We must flag these local variables
7670 * to be updated and be sure the update takes place with the
7671 * new frequency before any guests proceed.
7673 * Unfortunately, the combination of hotplug CPU and frequency
7674 * change creates an intractable locking scenario; the order
7675 * of when these callouts happen is undefined with respect to
7676 * CPU hotplug, and they can race with each other. As such,
7677 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7678 * undefined; you can actually have a CPU frequency change take
7679 * place in between the computation of X and the setting of the
7680 * variable. To protect against this problem, all updates of
7681 * the per_cpu tsc_khz variable are done in an interrupt
7682 * protected IPI, and all callers wishing to update the value
7683 * must wait for a synchronous IPI to complete (which is trivial
7684 * if the caller is on the CPU already). This establishes the
7685 * necessary total order on variable updates.
7687 * Note that because a guest time update may take place
7688 * anytime after the setting of the VCPU's request bit, the
7689 * correct TSC value must be set before the request. However,
7690 * to ensure the update actually makes it to any guest which
7691 * starts running in hardware virtualization between the set
7692 * and the acquisition of the spinlock, we must also ping the
7693 * CPU after setting the request bit.
7697 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7699 mutex_lock(&kvm_lock);
7700 list_for_each_entry(kvm, &vm_list, vm_list) {
7701 kvm_for_each_vcpu(i, vcpu, kvm) {
7702 if (vcpu->cpu != cpu)
7704 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7705 if (vcpu->cpu != raw_smp_processor_id())
7709 mutex_unlock(&kvm_lock);
7711 if (freq->old < freq->new && send_ipi) {
7713 * We upscale the frequency. Must make the guest
7714 * doesn't see old kvmclock values while running with
7715 * the new frequency, otherwise we risk the guest sees
7716 * time go backwards.
7718 * In case we update the frequency for another cpu
7719 * (which might be in guest context) send an interrupt
7720 * to kick the cpu out of guest context. Next time
7721 * guest context is entered kvmclock will be updated,
7722 * so the guest will not see stale values.
7724 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7728 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7731 struct cpufreq_freqs *freq = data;
7734 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7736 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7739 for_each_cpu(cpu, freq->policy->cpus)
7740 __kvmclock_cpufreq_notifier(freq, cpu);
7745 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7746 .notifier_call = kvmclock_cpufreq_notifier
7749 static int kvmclock_cpu_online(unsigned int cpu)
7751 tsc_khz_changed(NULL);
7755 static void kvm_timer_init(void)
7757 max_tsc_khz = tsc_khz;
7759 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7760 #ifdef CONFIG_CPU_FREQ
7761 struct cpufreq_policy *policy;
7765 policy = cpufreq_cpu_get(cpu);
7767 if (policy->cpuinfo.max_freq)
7768 max_tsc_khz = policy->cpuinfo.max_freq;
7769 cpufreq_cpu_put(policy);
7773 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7774 CPUFREQ_TRANSITION_NOTIFIER);
7777 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7778 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7781 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7782 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7784 int kvm_is_in_guest(void)
7786 return __this_cpu_read(current_vcpu) != NULL;
7789 static int kvm_is_user_mode(void)
7793 if (__this_cpu_read(current_vcpu))
7794 user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7796 return user_mode != 0;
7799 static unsigned long kvm_get_guest_ip(void)
7801 unsigned long ip = 0;
7803 if (__this_cpu_read(current_vcpu))
7804 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7809 static void kvm_handle_intel_pt_intr(void)
7811 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7813 kvm_make_request(KVM_REQ_PMI, vcpu);
7814 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7815 (unsigned long *)&vcpu->arch.pmu.global_status);
7818 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7819 .is_in_guest = kvm_is_in_guest,
7820 .is_user_mode = kvm_is_user_mode,
7821 .get_guest_ip = kvm_get_guest_ip,
7822 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7825 #ifdef CONFIG_X86_64
7826 static void pvclock_gtod_update_fn(struct work_struct *work)
7830 struct kvm_vcpu *vcpu;
7833 mutex_lock(&kvm_lock);
7834 list_for_each_entry(kvm, &vm_list, vm_list)
7835 kvm_for_each_vcpu(i, vcpu, kvm)
7836 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7837 atomic_set(&kvm_guest_has_master_clock, 0);
7838 mutex_unlock(&kvm_lock);
7841 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7844 * Notification about pvclock gtod data update.
7846 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7849 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7850 struct timekeeper *tk = priv;
7852 update_pvclock_gtod(tk);
7854 /* disable master clock if host does not trust, or does not
7855 * use, TSC based clocksource.
7857 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7858 atomic_read(&kvm_guest_has_master_clock) != 0)
7859 queue_work(system_long_wq, &pvclock_gtod_work);
7864 static struct notifier_block pvclock_gtod_notifier = {
7865 .notifier_call = pvclock_gtod_notify,
7869 int kvm_arch_init(void *opaque)
7871 struct kvm_x86_init_ops *ops = opaque;
7874 if (kvm_x86_ops.hardware_enable) {
7875 printk(KERN_ERR "kvm: already loaded the other module\n");
7880 if (!ops->cpu_has_kvm_support()) {
7881 pr_err_ratelimited("kvm: no hardware support\n");
7885 if (ops->disabled_by_bios()) {
7886 pr_err_ratelimited("kvm: disabled by bios\n");
7892 * KVM explicitly assumes that the guest has an FPU and
7893 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7894 * vCPU's FPU state as a fxregs_state struct.
7896 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7897 printk(KERN_ERR "kvm: inadequate fpu\n");
7903 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7904 __alignof__(struct fpu), SLAB_ACCOUNT,
7906 if (!x86_fpu_cache) {
7907 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7911 x86_emulator_cache = kvm_alloc_emulator_cache();
7912 if (!x86_emulator_cache) {
7913 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7914 goto out_free_x86_fpu_cache;
7917 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
7918 if (!user_return_msrs) {
7919 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
7920 goto out_free_x86_emulator_cache;
7923 r = kvm_mmu_module_init();
7925 goto out_free_percpu;
7927 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7928 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7929 PT_PRESENT_MASK, 0, sme_me_mask);
7932 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7934 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7935 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7936 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7940 if (pi_inject_timer == -1)
7941 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7942 #ifdef CONFIG_X86_64
7943 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7945 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7946 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7952 free_percpu(user_return_msrs);
7953 out_free_x86_emulator_cache:
7954 kmem_cache_destroy(x86_emulator_cache);
7955 out_free_x86_fpu_cache:
7956 kmem_cache_destroy(x86_fpu_cache);
7961 void kvm_arch_exit(void)
7963 #ifdef CONFIG_X86_64
7964 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7965 clear_hv_tscchange_cb();
7968 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7970 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7971 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7972 CPUFREQ_TRANSITION_NOTIFIER);
7973 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7974 #ifdef CONFIG_X86_64
7975 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7977 kvm_x86_ops.hardware_enable = NULL;
7978 kvm_mmu_module_exit();
7979 free_percpu(user_return_msrs);
7980 kmem_cache_destroy(x86_fpu_cache);
7983 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
7985 ++vcpu->stat.halt_exits;
7986 if (lapic_in_kernel(vcpu)) {
7987 vcpu->arch.mp_state = state;
7990 vcpu->run->exit_reason = reason;
7995 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7997 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
7999 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8001 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8003 int ret = kvm_skip_emulated_instruction(vcpu);
8005 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8006 * KVM_EXIT_DEBUG here.
8008 return kvm_vcpu_halt(vcpu) && ret;
8010 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8012 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8014 int ret = kvm_skip_emulated_instruction(vcpu);
8016 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8018 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8020 #ifdef CONFIG_X86_64
8021 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8022 unsigned long clock_type)
8024 struct kvm_clock_pairing clock_pairing;
8025 struct timespec64 ts;
8029 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8030 return -KVM_EOPNOTSUPP;
8032 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8033 return -KVM_EOPNOTSUPP;
8035 clock_pairing.sec = ts.tv_sec;
8036 clock_pairing.nsec = ts.tv_nsec;
8037 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8038 clock_pairing.flags = 0;
8039 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8042 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8043 sizeof(struct kvm_clock_pairing)))
8051 * kvm_pv_kick_cpu_op: Kick a vcpu.
8053 * @apicid - apicid of vcpu to be kicked.
8055 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8057 struct kvm_lapic_irq lapic_irq;
8059 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8060 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8061 lapic_irq.level = 0;
8062 lapic_irq.dest_id = apicid;
8063 lapic_irq.msi_redir_hint = false;
8065 lapic_irq.delivery_mode = APIC_DM_REMRD;
8066 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8069 bool kvm_apicv_activated(struct kvm *kvm)
8071 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8073 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8075 void kvm_apicv_init(struct kvm *kvm, bool enable)
8078 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8079 &kvm->arch.apicv_inhibit_reasons);
8081 set_bit(APICV_INHIBIT_REASON_DISABLE,
8082 &kvm->arch.apicv_inhibit_reasons);
8084 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8086 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
8088 struct kvm_vcpu *target = NULL;
8089 struct kvm_apic_map *map;
8092 map = rcu_dereference(kvm->arch.apic_map);
8094 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8095 target = map->phys_map[dest_id]->vcpu;
8099 if (target && READ_ONCE(target->ready))
8100 kvm_vcpu_yield_to(target);
8103 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8105 unsigned long nr, a0, a1, a2, a3, ret;
8108 if (kvm_hv_hypercall_enabled(vcpu->kvm))
8109 return kvm_hv_hypercall(vcpu);
8111 nr = kvm_rax_read(vcpu);
8112 a0 = kvm_rbx_read(vcpu);
8113 a1 = kvm_rcx_read(vcpu);
8114 a2 = kvm_rdx_read(vcpu);
8115 a3 = kvm_rsi_read(vcpu);
8117 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8119 op_64_bit = is_64_bit_mode(vcpu);
8128 if (kvm_x86_ops.get_cpl(vcpu) != 0) {
8136 case KVM_HC_VAPIC_POLL_IRQ:
8139 case KVM_HC_KICK_CPU:
8140 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8143 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8144 kvm_sched_yield(vcpu->kvm, a1);
8147 #ifdef CONFIG_X86_64
8148 case KVM_HC_CLOCK_PAIRING:
8149 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8152 case KVM_HC_SEND_IPI:
8153 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8156 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8158 case KVM_HC_SCHED_YIELD:
8159 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8162 kvm_sched_yield(vcpu->kvm, a0);
8172 kvm_rax_write(vcpu, ret);
8174 ++vcpu->stat.hypercalls;
8175 return kvm_skip_emulated_instruction(vcpu);
8177 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8179 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8181 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8182 char instruction[3];
8183 unsigned long rip = kvm_rip_read(vcpu);
8185 kvm_x86_ops.patch_hypercall(vcpu, instruction);
8187 return emulator_write_emulated(ctxt, rip, instruction, 3,
8191 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8193 return vcpu->run->request_interrupt_window &&
8194 likely(!pic_in_kernel(vcpu->kvm));
8197 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8199 struct kvm_run *kvm_run = vcpu->run;
8202 * if_flag is obsolete and useless, so do not bother
8203 * setting it for SEV-ES guests. Userspace can just
8204 * use kvm_run->ready_for_interrupt_injection.
8206 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8207 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8209 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
8210 kvm_run->cr8 = kvm_get_cr8(vcpu);
8211 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8212 kvm_run->ready_for_interrupt_injection =
8213 pic_in_kernel(vcpu->kvm) ||
8214 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8217 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8221 if (!kvm_x86_ops.update_cr8_intercept)
8224 if (!lapic_in_kernel(vcpu))
8227 if (vcpu->arch.apicv_active)
8230 if (!vcpu->arch.apic->vapic_addr)
8231 max_irr = kvm_lapic_find_highest_irr(vcpu);
8238 tpr = kvm_lapic_get_cr8(vcpu);
8240 kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
8243 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8246 bool can_inject = true;
8248 /* try to reinject previous events if any */
8250 if (vcpu->arch.exception.injected) {
8251 kvm_x86_ops.queue_exception(vcpu);
8255 * Do not inject an NMI or interrupt if there is a pending
8256 * exception. Exceptions and interrupts are recognized at
8257 * instruction boundaries, i.e. the start of an instruction.
8258 * Trap-like exceptions, e.g. #DB, have higher priority than
8259 * NMIs and interrupts, i.e. traps are recognized before an
8260 * NMI/interrupt that's pending on the same instruction.
8261 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8262 * priority, but are only generated (pended) during instruction
8263 * execution, i.e. a pending fault-like exception means the
8264 * fault occurred on the *previous* instruction and must be
8265 * serviced prior to recognizing any new events in order to
8266 * fully complete the previous instruction.
8268 else if (!vcpu->arch.exception.pending) {
8269 if (vcpu->arch.nmi_injected) {
8270 kvm_x86_ops.set_nmi(vcpu);
8272 } else if (vcpu->arch.interrupt.injected) {
8273 kvm_x86_ops.set_irq(vcpu);
8278 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8279 vcpu->arch.exception.pending);
8282 * Call check_nested_events() even if we reinjected a previous event
8283 * in order for caller to determine if it should require immediate-exit
8284 * from L2 to L1 due to pending L1 events which require exit
8287 if (is_guest_mode(vcpu)) {
8288 r = kvm_x86_ops.nested_ops->check_events(vcpu);
8293 /* try to inject new event if pending */
8294 if (vcpu->arch.exception.pending) {
8295 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8296 vcpu->arch.exception.has_error_code,
8297 vcpu->arch.exception.error_code);
8299 vcpu->arch.exception.pending = false;
8300 vcpu->arch.exception.injected = true;
8302 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8303 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8306 if (vcpu->arch.exception.nr == DB_VECTOR) {
8307 kvm_deliver_exception_payload(vcpu);
8308 if (vcpu->arch.dr7 & DR7_GD) {
8309 vcpu->arch.dr7 &= ~DR7_GD;
8310 kvm_update_dr7(vcpu);
8314 kvm_x86_ops.queue_exception(vcpu);
8319 * Finally, inject interrupt events. If an event cannot be injected
8320 * due to architectural conditions (e.g. IF=0) a window-open exit
8321 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8322 * and can architecturally be injected, but we cannot do it right now:
8323 * an interrupt could have arrived just now and we have to inject it
8324 * as a vmexit, or there could already an event in the queue, which is
8325 * indicated by can_inject. In that case we request an immediate exit
8326 * in order to make progress and get back here for another iteration.
8327 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8329 if (vcpu->arch.smi_pending) {
8330 r = can_inject ? kvm_x86_ops.smi_allowed(vcpu, true) : -EBUSY;
8334 vcpu->arch.smi_pending = false;
8335 ++vcpu->arch.smi_count;
8339 kvm_x86_ops.enable_smi_window(vcpu);
8342 if (vcpu->arch.nmi_pending) {
8343 r = can_inject ? kvm_x86_ops.nmi_allowed(vcpu, true) : -EBUSY;
8347 --vcpu->arch.nmi_pending;
8348 vcpu->arch.nmi_injected = true;
8349 kvm_x86_ops.set_nmi(vcpu);
8351 WARN_ON(kvm_x86_ops.nmi_allowed(vcpu, true) < 0);
8353 if (vcpu->arch.nmi_pending)
8354 kvm_x86_ops.enable_nmi_window(vcpu);
8357 if (kvm_cpu_has_injectable_intr(vcpu)) {
8358 r = can_inject ? kvm_x86_ops.interrupt_allowed(vcpu, true) : -EBUSY;
8362 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8363 kvm_x86_ops.set_irq(vcpu);
8364 WARN_ON(kvm_x86_ops.interrupt_allowed(vcpu, true) < 0);
8366 if (kvm_cpu_has_injectable_intr(vcpu))
8367 kvm_x86_ops.enable_irq_window(vcpu);
8370 if (is_guest_mode(vcpu) &&
8371 kvm_x86_ops.nested_ops->hv_timer_pending &&
8372 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8373 *req_immediate_exit = true;
8375 WARN_ON(vcpu->arch.exception.pending);
8379 *req_immediate_exit = true;
8383 static void process_nmi(struct kvm_vcpu *vcpu)
8388 * x86 is limited to one NMI running, and one NMI pending after it.
8389 * If an NMI is already in progress, limit further NMIs to just one.
8390 * Otherwise, allow two (and we'll inject the first one immediately).
8392 if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
8395 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8396 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8397 kvm_make_request(KVM_REQ_EVENT, vcpu);
8400 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8403 flags |= seg->g << 23;
8404 flags |= seg->db << 22;
8405 flags |= seg->l << 21;
8406 flags |= seg->avl << 20;
8407 flags |= seg->present << 15;
8408 flags |= seg->dpl << 13;
8409 flags |= seg->s << 12;
8410 flags |= seg->type << 8;
8414 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8416 struct kvm_segment seg;
8419 kvm_get_segment(vcpu, &seg, n);
8420 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8423 offset = 0x7f84 + n * 12;
8425 offset = 0x7f2c + (n - 3) * 12;
8427 put_smstate(u32, buf, offset + 8, seg.base);
8428 put_smstate(u32, buf, offset + 4, seg.limit);
8429 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8432 #ifdef CONFIG_X86_64
8433 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8435 struct kvm_segment seg;
8439 kvm_get_segment(vcpu, &seg, n);
8440 offset = 0x7e00 + n * 16;
8442 flags = enter_smm_get_segment_flags(&seg) >> 8;
8443 put_smstate(u16, buf, offset, seg.selector);
8444 put_smstate(u16, buf, offset + 2, flags);
8445 put_smstate(u32, buf, offset + 4, seg.limit);
8446 put_smstate(u64, buf, offset + 8, seg.base);
8450 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8453 struct kvm_segment seg;
8457 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8458 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8459 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8460 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8462 for (i = 0; i < 8; i++)
8463 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
8465 kvm_get_dr(vcpu, 6, &val);
8466 put_smstate(u32, buf, 0x7fcc, (u32)val);
8467 kvm_get_dr(vcpu, 7, &val);
8468 put_smstate(u32, buf, 0x7fc8, (u32)val);
8470 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8471 put_smstate(u32, buf, 0x7fc4, seg.selector);
8472 put_smstate(u32, buf, 0x7f64, seg.base);
8473 put_smstate(u32, buf, 0x7f60, seg.limit);
8474 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8476 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8477 put_smstate(u32, buf, 0x7fc0, seg.selector);
8478 put_smstate(u32, buf, 0x7f80, seg.base);
8479 put_smstate(u32, buf, 0x7f7c, seg.limit);
8480 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8482 kvm_x86_ops.get_gdt(vcpu, &dt);
8483 put_smstate(u32, buf, 0x7f74, dt.address);
8484 put_smstate(u32, buf, 0x7f70, dt.size);
8486 kvm_x86_ops.get_idt(vcpu, &dt);
8487 put_smstate(u32, buf, 0x7f58, dt.address);
8488 put_smstate(u32, buf, 0x7f54, dt.size);
8490 for (i = 0; i < 6; i++)
8491 enter_smm_save_seg_32(vcpu, buf, i);
8493 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8496 put_smstate(u32, buf, 0x7efc, 0x00020000);
8497 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8500 #ifdef CONFIG_X86_64
8501 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8504 struct kvm_segment seg;
8508 for (i = 0; i < 16; i++)
8509 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8511 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8512 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8514 kvm_get_dr(vcpu, 6, &val);
8515 put_smstate(u64, buf, 0x7f68, val);
8516 kvm_get_dr(vcpu, 7, &val);
8517 put_smstate(u64, buf, 0x7f60, val);
8519 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8520 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8521 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8523 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8526 put_smstate(u32, buf, 0x7efc, 0x00020064);
8528 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8530 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8531 put_smstate(u16, buf, 0x7e90, seg.selector);
8532 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8533 put_smstate(u32, buf, 0x7e94, seg.limit);
8534 put_smstate(u64, buf, 0x7e98, seg.base);
8536 kvm_x86_ops.get_idt(vcpu, &dt);
8537 put_smstate(u32, buf, 0x7e84, dt.size);
8538 put_smstate(u64, buf, 0x7e88, dt.address);
8540 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8541 put_smstate(u16, buf, 0x7e70, seg.selector);
8542 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8543 put_smstate(u32, buf, 0x7e74, seg.limit);
8544 put_smstate(u64, buf, 0x7e78, seg.base);
8546 kvm_x86_ops.get_gdt(vcpu, &dt);
8547 put_smstate(u32, buf, 0x7e64, dt.size);
8548 put_smstate(u64, buf, 0x7e68, dt.address);
8550 for (i = 0; i < 6; i++)
8551 enter_smm_save_seg_64(vcpu, buf, i);
8555 static void enter_smm(struct kvm_vcpu *vcpu)
8557 struct kvm_segment cs, ds;
8562 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8563 memset(buf, 0, 512);
8564 #ifdef CONFIG_X86_64
8565 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8566 enter_smm_save_state_64(vcpu, buf);
8569 enter_smm_save_state_32(vcpu, buf);
8572 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8573 * vCPU state (e.g. leave guest mode) after we've saved the state into
8574 * the SMM state-save area.
8576 kvm_x86_ops.pre_enter_smm(vcpu, buf);
8578 vcpu->arch.hflags |= HF_SMM_MASK;
8579 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8581 if (kvm_x86_ops.get_nmi_mask(vcpu))
8582 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8584 kvm_x86_ops.set_nmi_mask(vcpu, true);
8586 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8587 kvm_rip_write(vcpu, 0x8000);
8589 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8590 kvm_x86_ops.set_cr0(vcpu, cr0);
8591 vcpu->arch.cr0 = cr0;
8593 kvm_x86_ops.set_cr4(vcpu, 0);
8595 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8596 dt.address = dt.size = 0;
8597 kvm_x86_ops.set_idt(vcpu, &dt);
8599 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8601 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8602 cs.base = vcpu->arch.smbase;
8607 cs.limit = ds.limit = 0xffffffff;
8608 cs.type = ds.type = 0x3;
8609 cs.dpl = ds.dpl = 0;
8614 cs.avl = ds.avl = 0;
8615 cs.present = ds.present = 1;
8616 cs.unusable = ds.unusable = 0;
8617 cs.padding = ds.padding = 0;
8619 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8620 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8621 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8622 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8623 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8624 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8626 #ifdef CONFIG_X86_64
8627 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8628 kvm_x86_ops.set_efer(vcpu, 0);
8631 kvm_update_cpuid_runtime(vcpu);
8632 kvm_mmu_reset_context(vcpu);
8635 static void process_smi(struct kvm_vcpu *vcpu)
8637 vcpu->arch.smi_pending = true;
8638 kvm_make_request(KVM_REQ_EVENT, vcpu);
8641 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8642 unsigned long *vcpu_bitmap)
8646 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8648 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8649 NULL, vcpu_bitmap, cpus);
8651 free_cpumask_var(cpus);
8654 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8656 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8659 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8661 if (!lapic_in_kernel(vcpu))
8664 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8665 kvm_apic_update_apicv(vcpu);
8666 kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8668 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8671 * NOTE: Do not hold any lock prior to calling this.
8673 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8674 * locked, because it calls __x86_set_memory_region() which does
8675 * synchronize_srcu(&kvm->srcu).
8677 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8679 struct kvm_vcpu *except;
8680 unsigned long old, new, expected;
8682 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8683 !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8686 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8688 expected = new = old;
8690 __clear_bit(bit, &new);
8692 __set_bit(bit, &new);
8695 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8696 } while (old != expected);
8701 trace_kvm_apicv_update_request(activate, bit);
8702 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8703 kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8706 * Sending request to update APICV for all other vcpus,
8707 * while update the calling vcpu immediately instead of
8708 * waiting for another #VMEXIT to handle the request.
8710 except = kvm_get_running_vcpu();
8711 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8714 kvm_vcpu_update_apicv(except);
8716 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8718 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8720 if (!kvm_apic_present(vcpu))
8723 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8725 if (irqchip_split(vcpu->kvm))
8726 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8728 if (vcpu->arch.apicv_active)
8729 kvm_x86_ops.sync_pir_to_irr(vcpu);
8730 if (ioapic_in_kernel(vcpu->kvm))
8731 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8734 if (is_guest_mode(vcpu))
8735 vcpu->arch.load_eoi_exitmap_pending = true;
8737 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8740 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8742 u64 eoi_exit_bitmap[4];
8744 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8747 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8748 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8749 kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8752 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8753 unsigned long start, unsigned long end)
8755 unsigned long apic_address;
8758 * The physical address of apic access page is stored in the VMCS.
8759 * Update it when it becomes invalid.
8761 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8762 if (start <= apic_address && apic_address < end)
8763 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8766 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8768 if (!lapic_in_kernel(vcpu))
8771 if (!kvm_x86_ops.set_apic_access_page_addr)
8774 kvm_x86_ops.set_apic_access_page_addr(vcpu);
8777 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8779 smp_send_reschedule(vcpu->cpu);
8781 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8784 * Returns 1 to let vcpu_run() continue the guest execution loop without
8785 * exiting to the userspace. Otherwise, the value will be returned to the
8788 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8792 dm_request_for_irq_injection(vcpu) &&
8793 kvm_cpu_accept_dm_intr(vcpu);
8794 fastpath_t exit_fastpath;
8796 bool req_immediate_exit = false;
8798 /* Forbid vmenter if vcpu dirty ring is soft-full */
8799 if (unlikely(vcpu->kvm->dirty_ring_size &&
8800 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
8801 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
8802 trace_kvm_dirty_ring_exit(vcpu);
8807 if (kvm_request_pending(vcpu)) {
8808 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
8809 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
8814 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8815 kvm_mmu_unload(vcpu);
8816 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8817 __kvm_migrate_timers(vcpu);
8818 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8819 kvm_gen_update_masterclock(vcpu->kvm);
8820 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8821 kvm_gen_kvmclock_update(vcpu);
8822 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8823 r = kvm_guest_time_update(vcpu);
8827 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8828 kvm_mmu_sync_roots(vcpu);
8829 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8830 kvm_mmu_load_pgd(vcpu);
8831 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8832 kvm_vcpu_flush_tlb_all(vcpu);
8834 /* Flushing all ASIDs flushes the current ASID... */
8835 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8837 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8838 kvm_vcpu_flush_tlb_current(vcpu);
8839 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8840 kvm_vcpu_flush_tlb_guest(vcpu);
8842 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8843 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8847 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8848 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8849 vcpu->mmio_needed = 0;
8853 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8854 /* Page is swapped out. Do synthetic halt */
8855 vcpu->arch.apf.halted = true;
8859 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8860 record_steal_time(vcpu);
8861 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8863 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8865 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8866 kvm_pmu_handle_event(vcpu);
8867 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8868 kvm_pmu_deliver_pmi(vcpu);
8869 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8870 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8871 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8872 vcpu->arch.ioapic_handled_vectors)) {
8873 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8874 vcpu->run->eoi.vector =
8875 vcpu->arch.pending_ioapic_eoi;
8880 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8881 vcpu_scan_ioapic(vcpu);
8882 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8883 vcpu_load_eoi_exitmap(vcpu);
8884 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8885 kvm_vcpu_reload_apic_access_page(vcpu);
8886 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8887 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8888 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8892 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8893 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8894 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8898 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8899 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8900 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8906 * KVM_REQ_HV_STIMER has to be processed after
8907 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8908 * depend on the guest clock being up-to-date
8910 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8911 kvm_hv_process_stimers(vcpu);
8912 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8913 kvm_vcpu_update_apicv(vcpu);
8914 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
8915 kvm_check_async_pf_completion(vcpu);
8916 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
8917 kvm_x86_ops.msr_filter_changed(vcpu);
8920 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8921 ++vcpu->stat.req_event;
8922 kvm_apic_accept_events(vcpu);
8923 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8928 inject_pending_event(vcpu, &req_immediate_exit);
8930 kvm_x86_ops.enable_irq_window(vcpu);
8932 if (kvm_lapic_enabled(vcpu)) {
8933 update_cr8_intercept(vcpu);
8934 kvm_lapic_sync_to_vapic(vcpu);
8938 r = kvm_mmu_reload(vcpu);
8940 goto cancel_injection;
8945 kvm_x86_ops.prepare_guest_switch(vcpu);
8948 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8949 * IPI are then delayed after guest entry, which ensures that they
8950 * result in virtual interrupt delivery.
8952 local_irq_disable();
8953 vcpu->mode = IN_GUEST_MODE;
8955 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8958 * 1) We should set ->mode before checking ->requests. Please see
8959 * the comment in kvm_vcpu_exiting_guest_mode().
8961 * 2) For APICv, we should set ->mode before checking PID.ON. This
8962 * pairs with the memory barrier implicit in pi_test_and_set_on
8963 * (see vmx_deliver_posted_interrupt).
8965 * 3) This also orders the write to mode from any reads to the page
8966 * tables done while the VCPU is running. Please see the comment
8967 * in kvm_flush_remote_tlbs.
8969 smp_mb__after_srcu_read_unlock();
8972 * This handles the case where a posted interrupt was
8973 * notified with kvm_vcpu_kick.
8975 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8976 kvm_x86_ops.sync_pir_to_irr(vcpu);
8978 if (kvm_vcpu_exit_request(vcpu)) {
8979 vcpu->mode = OUTSIDE_GUEST_MODE;
8983 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8985 goto cancel_injection;
8988 if (req_immediate_exit) {
8989 kvm_make_request(KVM_REQ_EVENT, vcpu);
8990 kvm_x86_ops.request_immediate_exit(vcpu);
8993 fpregs_assert_state_consistent();
8994 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8995 switch_fpu_return();
8997 if (unlikely(vcpu->arch.switch_db_regs)) {
8999 set_debugreg(vcpu->arch.eff_db[0], 0);
9000 set_debugreg(vcpu->arch.eff_db[1], 1);
9001 set_debugreg(vcpu->arch.eff_db[2], 2);
9002 set_debugreg(vcpu->arch.eff_db[3], 3);
9003 set_debugreg(vcpu->arch.dr6, 6);
9004 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9007 exit_fastpath = kvm_x86_ops.run(vcpu);
9010 * Do this here before restoring debug registers on the host. And
9011 * since we do this before handling the vmexit, a DR access vmexit
9012 * can (a) read the correct value of the debug registers, (b) set
9013 * KVM_DEBUGREG_WONT_EXIT again.
9015 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9016 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9017 kvm_x86_ops.sync_dirty_debug_regs(vcpu);
9018 kvm_update_dr0123(vcpu);
9019 kvm_update_dr7(vcpu);
9020 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9024 * If the guest has used debug registers, at least dr7
9025 * will be disabled while returning to the host.
9026 * If we don't have active breakpoints in the host, we don't
9027 * care about the messed up debug address registers. But if
9028 * we have some of them active, restore the old state.
9030 if (hw_breakpoint_active())
9031 hw_breakpoint_restore();
9033 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9034 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9036 vcpu->mode = OUTSIDE_GUEST_MODE;
9039 kvm_x86_ops.handle_exit_irqoff(vcpu);
9042 * Consume any pending interrupts, including the possible source of
9043 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9044 * An instruction is required after local_irq_enable() to fully unblock
9045 * interrupts on processors that implement an interrupt shadow, the
9046 * stat.exits increment will do nicely.
9048 kvm_before_interrupt(vcpu);
9051 local_irq_disable();
9052 kvm_after_interrupt(vcpu);
9054 if (lapic_in_kernel(vcpu)) {
9055 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9056 if (delta != S64_MIN) {
9057 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9058 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9065 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9068 * Profile KVM exit RIPs:
9070 if (unlikely(prof_on == KVM_PROFILING)) {
9071 unsigned long rip = kvm_rip_read(vcpu);
9072 profile_hit(KVM_PROFILING, (void *)rip);
9075 if (unlikely(vcpu->arch.tsc_always_catchup))
9076 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9078 if (vcpu->arch.apic_attention)
9079 kvm_lapic_sync_from_vapic(vcpu);
9081 r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
9085 if (req_immediate_exit)
9086 kvm_make_request(KVM_REQ_EVENT, vcpu);
9087 kvm_x86_ops.cancel_injection(vcpu);
9088 if (unlikely(vcpu->arch.apic_attention))
9089 kvm_lapic_sync_from_vapic(vcpu);
9094 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9096 if (!kvm_arch_vcpu_runnable(vcpu) &&
9097 (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
9098 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9099 kvm_vcpu_block(vcpu);
9100 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9102 if (kvm_x86_ops.post_block)
9103 kvm_x86_ops.post_block(vcpu);
9105 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9109 kvm_apic_accept_events(vcpu);
9110 switch(vcpu->arch.mp_state) {
9111 case KVM_MP_STATE_HALTED:
9112 case KVM_MP_STATE_AP_RESET_HOLD:
9113 vcpu->arch.pv.pv_unhalted = false;
9114 vcpu->arch.mp_state =
9115 KVM_MP_STATE_RUNNABLE;
9117 case KVM_MP_STATE_RUNNABLE:
9118 vcpu->arch.apf.halted = false;
9120 case KVM_MP_STATE_INIT_RECEIVED:
9128 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9130 if (is_guest_mode(vcpu))
9131 kvm_x86_ops.nested_ops->check_events(vcpu);
9133 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9134 !vcpu->arch.apf.halted);
9137 static int vcpu_run(struct kvm_vcpu *vcpu)
9140 struct kvm *kvm = vcpu->kvm;
9142 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9143 vcpu->arch.l1tf_flush_l1d = true;
9146 if (kvm_vcpu_running(vcpu)) {
9147 r = vcpu_enter_guest(vcpu);
9149 r = vcpu_block(kvm, vcpu);
9155 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9156 if (kvm_cpu_has_pending_timer(vcpu))
9157 kvm_inject_pending_timer_irqs(vcpu);
9159 if (dm_request_for_irq_injection(vcpu) &&
9160 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9162 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9163 ++vcpu->stat.request_irq_exits;
9167 if (__xfer_to_guest_mode_work_pending()) {
9168 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9169 r = xfer_to_guest_mode_handle_work(vcpu);
9172 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9176 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9181 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9185 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9186 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9187 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9191 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9193 BUG_ON(!vcpu->arch.pio.count);
9195 return complete_emulated_io(vcpu);
9199 * Implements the following, as a state machine:
9203 * for each mmio piece in the fragment
9211 * for each mmio piece in the fragment
9216 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9218 struct kvm_run *run = vcpu->run;
9219 struct kvm_mmio_fragment *frag;
9222 BUG_ON(!vcpu->mmio_needed);
9224 /* Complete previous fragment */
9225 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9226 len = min(8u, frag->len);
9227 if (!vcpu->mmio_is_write)
9228 memcpy(frag->data, run->mmio.data, len);
9230 if (frag->len <= 8) {
9231 /* Switch to the next fragment. */
9233 vcpu->mmio_cur_fragment++;
9235 /* Go forward to the next mmio piece. */
9241 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9242 vcpu->mmio_needed = 0;
9244 /* FIXME: return into emulator if single-stepping. */
9245 if (vcpu->mmio_is_write)
9247 vcpu->mmio_read_completed = 1;
9248 return complete_emulated_io(vcpu);
9251 run->exit_reason = KVM_EXIT_MMIO;
9252 run->mmio.phys_addr = frag->gpa;
9253 if (vcpu->mmio_is_write)
9254 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9255 run->mmio.len = min(8u, frag->len);
9256 run->mmio.is_write = vcpu->mmio_is_write;
9257 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9261 static void kvm_save_current_fpu(struct fpu *fpu)
9264 * If the target FPU state is not resident in the CPU registers, just
9265 * memcpy() from current, else save CPU state directly to the target.
9267 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9268 memcpy(&fpu->state, ¤t->thread.fpu.state,
9269 fpu_kernel_xstate_size);
9271 copy_fpregs_to_fpstate(fpu);
9274 /* Swap (qemu) user FPU context for the guest FPU context. */
9275 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9279 kvm_save_current_fpu(vcpu->arch.user_fpu);
9282 * Guests with protected state can't have it set by the hypervisor,
9283 * so skip trying to set it.
9285 if (vcpu->arch.guest_fpu)
9286 /* PKRU is separately restored in kvm_x86_ops.run. */
9287 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9288 ~XFEATURE_MASK_PKRU);
9290 fpregs_mark_activate();
9296 /* When vcpu_run ends, restore user space FPU context. */
9297 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9302 * Guests with protected state can't have it read by the hypervisor,
9303 * so skip trying to save it.
9305 if (vcpu->arch.guest_fpu)
9306 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9308 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9310 fpregs_mark_activate();
9313 ++vcpu->stat.fpu_reload;
9317 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9319 struct kvm_run *kvm_run = vcpu->run;
9323 kvm_sigset_activate(vcpu);
9324 kvm_load_guest_fpu(vcpu);
9326 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9327 if (kvm_run->immediate_exit) {
9331 kvm_vcpu_block(vcpu);
9332 kvm_apic_accept_events(vcpu);
9333 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9335 if (signal_pending(current)) {
9337 kvm_run->exit_reason = KVM_EXIT_INTR;
9338 ++vcpu->stat.signal_exits;
9343 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9348 if (kvm_run->kvm_dirty_regs) {
9349 r = sync_regs(vcpu);
9354 /* re-sync apic's tpr */
9355 if (!lapic_in_kernel(vcpu)) {
9356 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9362 if (unlikely(vcpu->arch.complete_userspace_io)) {
9363 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9364 vcpu->arch.complete_userspace_io = NULL;
9369 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9371 if (kvm_run->immediate_exit)
9377 kvm_put_guest_fpu(vcpu);
9378 if (kvm_run->kvm_valid_regs)
9380 post_kvm_run_save(vcpu);
9381 kvm_sigset_deactivate(vcpu);
9387 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9389 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9391 * We are here if userspace calls get_regs() in the middle of
9392 * instruction emulation. Registers state needs to be copied
9393 * back from emulation context to vcpu. Userspace shouldn't do
9394 * that usually, but some bad designed PV devices (vmware
9395 * backdoor interface) need this to work
9397 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9398 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9400 regs->rax = kvm_rax_read(vcpu);
9401 regs->rbx = kvm_rbx_read(vcpu);
9402 regs->rcx = kvm_rcx_read(vcpu);
9403 regs->rdx = kvm_rdx_read(vcpu);
9404 regs->rsi = kvm_rsi_read(vcpu);
9405 regs->rdi = kvm_rdi_read(vcpu);
9406 regs->rsp = kvm_rsp_read(vcpu);
9407 regs->rbp = kvm_rbp_read(vcpu);
9408 #ifdef CONFIG_X86_64
9409 regs->r8 = kvm_r8_read(vcpu);
9410 regs->r9 = kvm_r9_read(vcpu);
9411 regs->r10 = kvm_r10_read(vcpu);
9412 regs->r11 = kvm_r11_read(vcpu);
9413 regs->r12 = kvm_r12_read(vcpu);
9414 regs->r13 = kvm_r13_read(vcpu);
9415 regs->r14 = kvm_r14_read(vcpu);
9416 regs->r15 = kvm_r15_read(vcpu);
9419 regs->rip = kvm_rip_read(vcpu);
9420 regs->rflags = kvm_get_rflags(vcpu);
9423 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9426 __get_regs(vcpu, regs);
9431 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9433 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9434 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9436 kvm_rax_write(vcpu, regs->rax);
9437 kvm_rbx_write(vcpu, regs->rbx);
9438 kvm_rcx_write(vcpu, regs->rcx);
9439 kvm_rdx_write(vcpu, regs->rdx);
9440 kvm_rsi_write(vcpu, regs->rsi);
9441 kvm_rdi_write(vcpu, regs->rdi);
9442 kvm_rsp_write(vcpu, regs->rsp);
9443 kvm_rbp_write(vcpu, regs->rbp);
9444 #ifdef CONFIG_X86_64
9445 kvm_r8_write(vcpu, regs->r8);
9446 kvm_r9_write(vcpu, regs->r9);
9447 kvm_r10_write(vcpu, regs->r10);
9448 kvm_r11_write(vcpu, regs->r11);
9449 kvm_r12_write(vcpu, regs->r12);
9450 kvm_r13_write(vcpu, regs->r13);
9451 kvm_r14_write(vcpu, regs->r14);
9452 kvm_r15_write(vcpu, regs->r15);
9455 kvm_rip_write(vcpu, regs->rip);
9456 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9458 vcpu->arch.exception.pending = false;
9460 kvm_make_request(KVM_REQ_EVENT, vcpu);
9463 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9466 __set_regs(vcpu, regs);
9471 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9473 struct kvm_segment cs;
9475 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9479 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9481 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9485 if (vcpu->arch.guest_state_protected)
9486 goto skip_protected_regs;
9488 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9489 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9490 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9491 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9492 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9493 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9495 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9496 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9498 kvm_x86_ops.get_idt(vcpu, &dt);
9499 sregs->idt.limit = dt.size;
9500 sregs->idt.base = dt.address;
9501 kvm_x86_ops.get_gdt(vcpu, &dt);
9502 sregs->gdt.limit = dt.size;
9503 sregs->gdt.base = dt.address;
9505 sregs->cr2 = vcpu->arch.cr2;
9506 sregs->cr3 = kvm_read_cr3(vcpu);
9508 skip_protected_regs:
9509 sregs->cr0 = kvm_read_cr0(vcpu);
9510 sregs->cr4 = kvm_read_cr4(vcpu);
9511 sregs->cr8 = kvm_get_cr8(vcpu);
9512 sregs->efer = vcpu->arch.efer;
9513 sregs->apic_base = kvm_get_apic_base(vcpu);
9515 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9517 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9518 set_bit(vcpu->arch.interrupt.nr,
9519 (unsigned long *)sregs->interrupt_bitmap);
9522 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9523 struct kvm_sregs *sregs)
9526 __get_sregs(vcpu, sregs);
9531 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9532 struct kvm_mp_state *mp_state)
9535 if (kvm_mpx_supported())
9536 kvm_load_guest_fpu(vcpu);
9538 kvm_apic_accept_events(vcpu);
9539 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9540 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9541 vcpu->arch.pv.pv_unhalted)
9542 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9544 mp_state->mp_state = vcpu->arch.mp_state;
9546 if (kvm_mpx_supported())
9547 kvm_put_guest_fpu(vcpu);
9552 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9553 struct kvm_mp_state *mp_state)
9559 if (!lapic_in_kernel(vcpu) &&
9560 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9564 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9565 * INIT state; latched init should be reported using
9566 * KVM_SET_VCPU_EVENTS, so reject it here.
9568 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9569 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9570 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9573 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9574 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9575 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9577 vcpu->arch.mp_state = mp_state->mp_state;
9578 kvm_make_request(KVM_REQ_EVENT, vcpu);
9586 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9587 int reason, bool has_error_code, u32 error_code)
9589 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9592 init_emulate_ctxt(vcpu);
9594 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9595 has_error_code, error_code);
9597 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9598 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9599 vcpu->run->internal.ndata = 0;
9603 kvm_rip_write(vcpu, ctxt->eip);
9604 kvm_set_rflags(vcpu, ctxt->eflags);
9607 EXPORT_SYMBOL_GPL(kvm_task_switch);
9609 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9611 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9613 * When EFER.LME and CR0.PG are set, the processor is in
9614 * 64-bit mode (though maybe in a 32-bit code segment).
9615 * CR4.PAE and EFER.LMA must be set.
9617 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9619 if (sregs->cr3 & vcpu->arch.cr3_lm_rsvd_bits)
9623 * Not in 64-bit mode: EFER.LMA is clear and the code
9624 * segment cannot be 64-bit.
9626 if (sregs->efer & EFER_LMA || sregs->cs.l)
9630 return kvm_is_valid_cr4(vcpu, sregs->cr4);
9633 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9635 struct msr_data apic_base_msr;
9636 int mmu_reset_needed = 0;
9637 int pending_vec, max_bits, idx;
9641 if (!kvm_is_valid_sregs(vcpu, sregs))
9644 apic_base_msr.data = sregs->apic_base;
9645 apic_base_msr.host_initiated = true;
9646 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9649 if (vcpu->arch.guest_state_protected)
9650 goto skip_protected_regs;
9652 dt.size = sregs->idt.limit;
9653 dt.address = sregs->idt.base;
9654 kvm_x86_ops.set_idt(vcpu, &dt);
9655 dt.size = sregs->gdt.limit;
9656 dt.address = sregs->gdt.base;
9657 kvm_x86_ops.set_gdt(vcpu, &dt);
9659 vcpu->arch.cr2 = sregs->cr2;
9660 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9661 vcpu->arch.cr3 = sregs->cr3;
9662 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9664 kvm_set_cr8(vcpu, sregs->cr8);
9666 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9667 kvm_x86_ops.set_efer(vcpu, sregs->efer);
9669 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9670 kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9671 vcpu->arch.cr0 = sregs->cr0;
9673 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9674 kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9676 idx = srcu_read_lock(&vcpu->kvm->srcu);
9677 if (is_pae_paging(vcpu)) {
9678 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9679 mmu_reset_needed = 1;
9681 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9683 if (mmu_reset_needed)
9684 kvm_mmu_reset_context(vcpu);
9686 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9687 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9688 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9689 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9690 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9691 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9693 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9694 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9696 update_cr8_intercept(vcpu);
9698 /* Older userspace won't unhalt the vcpu on reset. */
9699 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9700 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9702 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9704 skip_protected_regs:
9705 max_bits = KVM_NR_INTERRUPTS;
9706 pending_vec = find_first_bit(
9707 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9708 if (pending_vec < max_bits) {
9709 kvm_queue_interrupt(vcpu, pending_vec, false);
9710 pr_debug("Set back pending irq %d\n", pending_vec);
9713 kvm_make_request(KVM_REQ_EVENT, vcpu);
9720 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9721 struct kvm_sregs *sregs)
9726 ret = __set_sregs(vcpu, sregs);
9731 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9732 struct kvm_guest_debug *dbg)
9734 unsigned long rflags;
9737 if (vcpu->arch.guest_state_protected)
9742 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9744 if (vcpu->arch.exception.pending)
9746 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9747 kvm_queue_exception(vcpu, DB_VECTOR);
9749 kvm_queue_exception(vcpu, BP_VECTOR);
9753 * Read rflags as long as potentially injected trace flags are still
9756 rflags = kvm_get_rflags(vcpu);
9758 vcpu->guest_debug = dbg->control;
9759 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9760 vcpu->guest_debug = 0;
9762 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9763 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9764 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9765 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9767 for (i = 0; i < KVM_NR_DB_REGS; i++)
9768 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9770 kvm_update_dr7(vcpu);
9772 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9773 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9774 get_segment_base(vcpu, VCPU_SREG_CS);
9777 * Trigger an rflags update that will inject or remove the trace
9780 kvm_set_rflags(vcpu, rflags);
9782 kvm_x86_ops.update_exception_bitmap(vcpu);
9792 * Translate a guest virtual address to a guest physical address.
9794 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9795 struct kvm_translation *tr)
9797 unsigned long vaddr = tr->linear_address;
9803 idx = srcu_read_lock(&vcpu->kvm->srcu);
9804 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9805 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9806 tr->physical_address = gpa;
9807 tr->valid = gpa != UNMAPPED_GVA;
9815 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9817 struct fxregs_state *fxsave;
9819 if (!vcpu->arch.guest_fpu)
9824 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9825 memcpy(fpu->fpr, fxsave->st_space, 128);
9826 fpu->fcw = fxsave->cwd;
9827 fpu->fsw = fxsave->swd;
9828 fpu->ftwx = fxsave->twd;
9829 fpu->last_opcode = fxsave->fop;
9830 fpu->last_ip = fxsave->rip;
9831 fpu->last_dp = fxsave->rdp;
9832 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9838 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9840 struct fxregs_state *fxsave;
9842 if (!vcpu->arch.guest_fpu)
9847 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9849 memcpy(fxsave->st_space, fpu->fpr, 128);
9850 fxsave->cwd = fpu->fcw;
9851 fxsave->swd = fpu->fsw;
9852 fxsave->twd = fpu->ftwx;
9853 fxsave->fop = fpu->last_opcode;
9854 fxsave->rip = fpu->last_ip;
9855 fxsave->rdp = fpu->last_dp;
9856 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9862 static void store_regs(struct kvm_vcpu *vcpu)
9864 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9866 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9867 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9869 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9870 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9872 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9873 kvm_vcpu_ioctl_x86_get_vcpu_events(
9874 vcpu, &vcpu->run->s.regs.events);
9877 static int sync_regs(struct kvm_vcpu *vcpu)
9879 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9882 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9883 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9884 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9886 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9887 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9889 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9891 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9892 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9893 vcpu, &vcpu->run->s.regs.events))
9895 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9901 static void fx_init(struct kvm_vcpu *vcpu)
9903 if (!vcpu->arch.guest_fpu)
9906 fpstate_init(&vcpu->arch.guest_fpu->state);
9907 if (boot_cpu_has(X86_FEATURE_XSAVES))
9908 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9909 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9912 * Ensure guest xcr0 is valid for loading
9914 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9916 vcpu->arch.cr0 |= X86_CR0_ET;
9919 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
9921 if (vcpu->arch.guest_fpu) {
9922 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9923 vcpu->arch.guest_fpu = NULL;
9926 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
9928 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9930 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9931 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9932 "guest TSC will not be reliable\n");
9937 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9942 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9943 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9945 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9947 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9949 r = kvm_mmu_create(vcpu);
9953 if (irqchip_in_kernel(vcpu->kvm)) {
9954 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9956 goto fail_mmu_destroy;
9957 if (kvm_apicv_activated(vcpu->kvm))
9958 vcpu->arch.apicv_active = true;
9960 static_key_slow_inc(&kvm_no_apic_vcpu);
9964 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
9966 goto fail_free_lapic;
9967 vcpu->arch.pio_data = page_address(page);
9969 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9970 GFP_KERNEL_ACCOUNT);
9971 if (!vcpu->arch.mce_banks)
9972 goto fail_free_pio_data;
9973 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9975 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9976 GFP_KERNEL_ACCOUNT))
9977 goto fail_free_mce_banks;
9979 if (!alloc_emulate_ctxt(vcpu))
9980 goto free_wbinvd_dirty_mask;
9982 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9983 GFP_KERNEL_ACCOUNT);
9984 if (!vcpu->arch.user_fpu) {
9985 pr_err("kvm: failed to allocate userspace's fpu\n");
9986 goto free_emulate_ctxt;
9989 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9990 GFP_KERNEL_ACCOUNT);
9991 if (!vcpu->arch.guest_fpu) {
9992 pr_err("kvm: failed to allocate vcpu's fpu\n");
9997 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9999 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10001 kvm_async_pf_hash_reset(vcpu);
10002 kvm_pmu_init(vcpu);
10004 vcpu->arch.pending_external_vector = -1;
10005 vcpu->arch.preempted_in_kernel = false;
10007 kvm_hv_vcpu_init(vcpu);
10009 r = kvm_x86_ops.vcpu_create(vcpu);
10011 goto free_guest_fpu;
10013 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10014 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10015 kvm_vcpu_mtrr_init(vcpu);
10017 kvm_vcpu_reset(vcpu, false);
10018 kvm_init_mmu(vcpu, false);
10023 kvm_free_guest_fpu(vcpu);
10025 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10027 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10028 free_wbinvd_dirty_mask:
10029 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10030 fail_free_mce_banks:
10031 kfree(vcpu->arch.mce_banks);
10032 fail_free_pio_data:
10033 free_page((unsigned long)vcpu->arch.pio_data);
10035 kvm_free_lapic(vcpu);
10037 kvm_mmu_destroy(vcpu);
10041 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10043 struct kvm *kvm = vcpu->kvm;
10045 kvm_hv_vcpu_postcreate(vcpu);
10047 if (mutex_lock_killable(&vcpu->mutex))
10050 kvm_synchronize_tsc(vcpu, 0);
10053 /* poll control enabled by default */
10054 vcpu->arch.msr_kvm_poll_control = 1;
10056 mutex_unlock(&vcpu->mutex);
10058 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10059 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10060 KVMCLOCK_SYNC_PERIOD);
10063 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10065 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10068 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10070 kvmclock_reset(vcpu);
10072 kvm_x86_ops.vcpu_free(vcpu);
10074 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10075 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10076 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10077 kvm_free_guest_fpu(vcpu);
10079 kvm_hv_vcpu_uninit(vcpu);
10080 kvm_pmu_destroy(vcpu);
10081 kfree(vcpu->arch.mce_banks);
10082 kvm_free_lapic(vcpu);
10083 idx = srcu_read_lock(&vcpu->kvm->srcu);
10084 kvm_mmu_destroy(vcpu);
10085 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10086 free_page((unsigned long)vcpu->arch.pio_data);
10087 kvfree(vcpu->arch.cpuid_entries);
10088 if (!lapic_in_kernel(vcpu))
10089 static_key_slow_dec(&kvm_no_apic_vcpu);
10092 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10094 kvm_lapic_reset(vcpu, init_event);
10096 vcpu->arch.hflags = 0;
10098 vcpu->arch.smi_pending = 0;
10099 vcpu->arch.smi_count = 0;
10100 atomic_set(&vcpu->arch.nmi_queued, 0);
10101 vcpu->arch.nmi_pending = 0;
10102 vcpu->arch.nmi_injected = false;
10103 kvm_clear_interrupt_queue(vcpu);
10104 kvm_clear_exception_queue(vcpu);
10106 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10107 kvm_update_dr0123(vcpu);
10108 vcpu->arch.dr6 = DR6_INIT;
10109 vcpu->arch.dr7 = DR7_FIXED_1;
10110 kvm_update_dr7(vcpu);
10112 vcpu->arch.cr2 = 0;
10114 kvm_make_request(KVM_REQ_EVENT, vcpu);
10115 vcpu->arch.apf.msr_en_val = 0;
10116 vcpu->arch.apf.msr_int_val = 0;
10117 vcpu->arch.st.msr_val = 0;
10119 kvmclock_reset(vcpu);
10121 kvm_clear_async_pf_completion_queue(vcpu);
10122 kvm_async_pf_hash_reset(vcpu);
10123 vcpu->arch.apf.halted = false;
10125 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10126 void *mpx_state_buffer;
10129 * To avoid have the INIT path from kvm_apic_has_events() that be
10130 * called with loaded FPU and does not let userspace fix the state.
10133 kvm_put_guest_fpu(vcpu);
10134 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10136 if (mpx_state_buffer)
10137 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10138 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10140 if (mpx_state_buffer)
10141 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10143 kvm_load_guest_fpu(vcpu);
10147 kvm_pmu_reset(vcpu);
10148 vcpu->arch.smbase = 0x30000;
10150 vcpu->arch.msr_misc_features_enables = 0;
10152 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10155 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10156 vcpu->arch.regs_avail = ~0;
10157 vcpu->arch.regs_dirty = ~0;
10159 vcpu->arch.ia32_xss = 0;
10161 kvm_x86_ops.vcpu_reset(vcpu, init_event);
10164 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10166 struct kvm_segment cs;
10168 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10169 cs.selector = vector << 8;
10170 cs.base = vector << 12;
10171 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10172 kvm_rip_write(vcpu, 0);
10174 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10176 int kvm_arch_hardware_enable(void)
10179 struct kvm_vcpu *vcpu;
10184 bool stable, backwards_tsc = false;
10186 kvm_user_return_msr_cpu_online();
10187 ret = kvm_x86_ops.hardware_enable();
10191 local_tsc = rdtsc();
10192 stable = !kvm_check_tsc_unstable();
10193 list_for_each_entry(kvm, &vm_list, vm_list) {
10194 kvm_for_each_vcpu(i, vcpu, kvm) {
10195 if (!stable && vcpu->cpu == smp_processor_id())
10196 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10197 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10198 backwards_tsc = true;
10199 if (vcpu->arch.last_host_tsc > max_tsc)
10200 max_tsc = vcpu->arch.last_host_tsc;
10206 * Sometimes, even reliable TSCs go backwards. This happens on
10207 * platforms that reset TSC during suspend or hibernate actions, but
10208 * maintain synchronization. We must compensate. Fortunately, we can
10209 * detect that condition here, which happens early in CPU bringup,
10210 * before any KVM threads can be running. Unfortunately, we can't
10211 * bring the TSCs fully up to date with real time, as we aren't yet far
10212 * enough into CPU bringup that we know how much real time has actually
10213 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10214 * variables that haven't been updated yet.
10216 * So we simply find the maximum observed TSC above, then record the
10217 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10218 * the adjustment will be applied. Note that we accumulate
10219 * adjustments, in case multiple suspend cycles happen before some VCPU
10220 * gets a chance to run again. In the event that no KVM threads get a
10221 * chance to run, we will miss the entire elapsed period, as we'll have
10222 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10223 * loose cycle time. This isn't too big a deal, since the loss will be
10224 * uniform across all VCPUs (not to mention the scenario is extremely
10225 * unlikely). It is possible that a second hibernate recovery happens
10226 * much faster than a first, causing the observed TSC here to be
10227 * smaller; this would require additional padding adjustment, which is
10228 * why we set last_host_tsc to the local tsc observed here.
10230 * N.B. - this code below runs only on platforms with reliable TSC,
10231 * as that is the only way backwards_tsc is set above. Also note
10232 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10233 * have the same delta_cyc adjustment applied if backwards_tsc
10234 * is detected. Note further, this adjustment is only done once,
10235 * as we reset last_host_tsc on all VCPUs to stop this from being
10236 * called multiple times (one for each physical CPU bringup).
10238 * Platforms with unreliable TSCs don't have to deal with this, they
10239 * will be compensated by the logic in vcpu_load, which sets the TSC to
10240 * catchup mode. This will catchup all VCPUs to real time, but cannot
10241 * guarantee that they stay in perfect synchronization.
10243 if (backwards_tsc) {
10244 u64 delta_cyc = max_tsc - local_tsc;
10245 list_for_each_entry(kvm, &vm_list, vm_list) {
10246 kvm->arch.backwards_tsc_observed = true;
10247 kvm_for_each_vcpu(i, vcpu, kvm) {
10248 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10249 vcpu->arch.last_host_tsc = local_tsc;
10250 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10254 * We have to disable TSC offset matching.. if you were
10255 * booting a VM while issuing an S4 host suspend....
10256 * you may have some problem. Solving this issue is
10257 * left as an exercise to the reader.
10259 kvm->arch.last_tsc_nsec = 0;
10260 kvm->arch.last_tsc_write = 0;
10267 void kvm_arch_hardware_disable(void)
10269 kvm_x86_ops.hardware_disable();
10270 drop_user_return_notifiers();
10273 int kvm_arch_hardware_setup(void *opaque)
10275 struct kvm_x86_init_ops *ops = opaque;
10278 rdmsrl_safe(MSR_EFER, &host_efer);
10280 if (boot_cpu_has(X86_FEATURE_XSAVES))
10281 rdmsrl(MSR_IA32_XSS, host_xss);
10283 r = ops->hardware_setup();
10287 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10289 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10292 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10293 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10294 #undef __kvm_cpu_cap_has
10296 if (kvm_has_tsc_control) {
10298 * Make sure the user can only configure tsc_khz values that
10299 * fit into a signed integer.
10300 * A min value is not calculated because it will always
10301 * be 1 on all machines.
10303 u64 max = min(0x7fffffffULL,
10304 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10305 kvm_max_guest_tsc_khz = max;
10307 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10310 kvm_init_msr_list();
10314 void kvm_arch_hardware_unsetup(void)
10316 kvm_x86_ops.hardware_unsetup();
10319 int kvm_arch_check_processor_compat(void *opaque)
10321 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10322 struct kvm_x86_init_ops *ops = opaque;
10324 WARN_ON(!irqs_disabled());
10326 if (__cr4_reserved_bits(cpu_has, c) !=
10327 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10330 return ops->check_processor_compatibility();
10333 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10335 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10337 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10339 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10341 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10344 struct static_key kvm_no_apic_vcpu __read_mostly;
10345 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
10347 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10349 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10351 vcpu->arch.l1tf_flush_l1d = true;
10352 if (pmu->version && unlikely(pmu->event_count)) {
10353 pmu->need_cleanup = true;
10354 kvm_make_request(KVM_REQ_PMU, vcpu);
10356 kvm_x86_ops.sched_in(vcpu, cpu);
10359 void kvm_arch_free_vm(struct kvm *kvm)
10361 kfree(kvm->arch.hyperv.hv_pa_pg);
10366 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10371 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10372 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10373 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10374 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10375 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10376 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10378 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10379 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10380 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10381 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10382 &kvm->arch.irq_sources_bitmap);
10384 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10385 mutex_init(&kvm->arch.apic_map_lock);
10386 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10388 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10389 pvclock_update_vm_gtod_copy(kvm);
10391 kvm->arch.guest_can_read_msr_platform_info = true;
10393 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10394 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10396 kvm_hv_init_vm(kvm);
10397 kvm_page_track_init(kvm);
10398 kvm_mmu_init_vm(kvm);
10400 return kvm_x86_ops.vm_init(kvm);
10403 int kvm_arch_post_init_vm(struct kvm *kvm)
10405 return kvm_mmu_post_init_vm(kvm);
10408 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10411 kvm_mmu_unload(vcpu);
10415 static void kvm_free_vcpus(struct kvm *kvm)
10418 struct kvm_vcpu *vcpu;
10421 * Unpin any mmu pages first.
10423 kvm_for_each_vcpu(i, vcpu, kvm) {
10424 kvm_clear_async_pf_completion_queue(vcpu);
10425 kvm_unload_vcpu_mmu(vcpu);
10427 kvm_for_each_vcpu(i, vcpu, kvm)
10428 kvm_vcpu_destroy(vcpu);
10430 mutex_lock(&kvm->lock);
10431 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10432 kvm->vcpus[i] = NULL;
10434 atomic_set(&kvm->online_vcpus, 0);
10435 mutex_unlock(&kvm->lock);
10438 void kvm_arch_sync_events(struct kvm *kvm)
10440 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10441 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10445 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
10448 * __x86_set_memory_region: Setup KVM internal memory slot
10450 * @kvm: the kvm pointer to the VM.
10451 * @id: the slot ID to setup.
10452 * @gpa: the GPA to install the slot (unused when @size == 0).
10453 * @size: the size of the slot. Set to zero to uninstall a slot.
10455 * This function helps to setup a KVM internal memory slot. Specify
10456 * @size > 0 to install a new slot, while @size == 0 to uninstall a
10457 * slot. The return code can be one of the following:
10459 * HVA: on success (uninstall will return a bogus HVA)
10462 * The caller should always use IS_ERR() to check the return value
10463 * before use. Note, the KVM internal memory slots are guaranteed to
10464 * remain valid and unchanged until the VM is destroyed, i.e., the
10465 * GPA->HVA translation will not change. However, the HVA is a user
10466 * address, i.e. its accessibility is not guaranteed, and must be
10467 * accessed via __copy_{to,from}_user().
10469 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10473 unsigned long hva, old_npages;
10474 struct kvm_memslots *slots = kvm_memslots(kvm);
10475 struct kvm_memory_slot *slot;
10477 /* Called with kvm->slots_lock held. */
10478 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10479 return ERR_PTR_USR(-EINVAL);
10481 slot = id_to_memslot(slots, id);
10483 if (slot && slot->npages)
10484 return ERR_PTR_USR(-EEXIST);
10487 * MAP_SHARED to prevent internal slot pages from being moved
10490 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10491 MAP_SHARED | MAP_ANONYMOUS, 0);
10492 if (IS_ERR((void *)hva))
10493 return (void __user *)hva;
10495 if (!slot || !slot->npages)
10498 old_npages = slot->npages;
10499 hva = slot->userspace_addr;
10502 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10503 struct kvm_userspace_memory_region m;
10505 m.slot = id | (i << 16);
10507 m.guest_phys_addr = gpa;
10508 m.userspace_addr = hva;
10509 m.memory_size = size;
10510 r = __kvm_set_memory_region(kvm, &m);
10512 return ERR_PTR_USR(r);
10516 vm_munmap(hva, old_npages * PAGE_SIZE);
10518 return (void __user *)hva;
10520 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10522 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10524 kvm_mmu_pre_destroy_vm(kvm);
10527 void kvm_arch_destroy_vm(struct kvm *kvm)
10531 if (current->mm == kvm->mm) {
10533 * Free memory regions allocated on behalf of userspace,
10534 * unless the the memory map has changed due to process exit
10537 mutex_lock(&kvm->slots_lock);
10538 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10540 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10542 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10543 mutex_unlock(&kvm->slots_lock);
10545 if (kvm_x86_ops.vm_destroy)
10546 kvm_x86_ops.vm_destroy(kvm);
10547 for (i = 0; i < kvm->arch.msr_filter.count; i++)
10548 kfree(kvm->arch.msr_filter.ranges[i].bitmap);
10549 kvm_pic_destroy(kvm);
10550 kvm_ioapic_destroy(kvm);
10551 kvm_free_vcpus(kvm);
10552 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10553 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10554 kvm_mmu_uninit_vm(kvm);
10555 kvm_page_track_cleanup(kvm);
10556 kvm_hv_destroy_vm(kvm);
10559 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10563 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10564 kvfree(slot->arch.rmap[i]);
10565 slot->arch.rmap[i] = NULL;
10570 kvfree(slot->arch.lpage_info[i - 1]);
10571 slot->arch.lpage_info[i - 1] = NULL;
10574 kvm_page_track_free_memslot(slot);
10577 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10578 unsigned long npages)
10583 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10584 * old arrays will be freed by __kvm_set_memory_region() if installing
10585 * the new memslot is successful.
10587 memset(&slot->arch, 0, sizeof(slot->arch));
10589 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10590 struct kvm_lpage_info *linfo;
10591 unsigned long ugfn;
10595 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10596 slot->base_gfn, level) + 1;
10598 slot->arch.rmap[i] =
10599 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10600 GFP_KERNEL_ACCOUNT);
10601 if (!slot->arch.rmap[i])
10606 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10610 slot->arch.lpage_info[i - 1] = linfo;
10612 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10613 linfo[0].disallow_lpage = 1;
10614 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10615 linfo[lpages - 1].disallow_lpage = 1;
10616 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10618 * If the gfn and userspace address are not aligned wrt each
10619 * other, disable large page support for this slot.
10621 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10624 for (j = 0; j < lpages; ++j)
10625 linfo[j].disallow_lpage = 1;
10629 if (kvm_page_track_create_memslot(slot, npages))
10635 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10636 kvfree(slot->arch.rmap[i]);
10637 slot->arch.rmap[i] = NULL;
10641 kvfree(slot->arch.lpage_info[i - 1]);
10642 slot->arch.lpage_info[i - 1] = NULL;
10647 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10649 struct kvm_vcpu *vcpu;
10653 * memslots->generation has been incremented.
10654 * mmio generation may have reached its maximum value.
10656 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10658 /* Force re-initialization of steal_time cache */
10659 kvm_for_each_vcpu(i, vcpu, kvm)
10660 kvm_vcpu_kick(vcpu);
10663 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10664 struct kvm_memory_slot *memslot,
10665 const struct kvm_userspace_memory_region *mem,
10666 enum kvm_mr_change change)
10668 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10669 return kvm_alloc_memslot_metadata(memslot,
10670 mem->memory_size >> PAGE_SHIFT);
10674 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10675 struct kvm_memory_slot *old,
10676 struct kvm_memory_slot *new,
10677 enum kvm_mr_change change)
10680 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
10681 * See comments below.
10683 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10687 * Dirty logging tracks sptes in 4k granularity, meaning that large
10688 * sptes have to be split. If live migration is successful, the guest
10689 * in the source machine will be destroyed and large sptes will be
10690 * created in the destination. However, if the guest continues to run
10691 * in the source machine (for example if live migration fails), small
10692 * sptes will remain around and cause bad performance.
10694 * Scan sptes if dirty logging has been stopped, dropping those
10695 * which can be collapsed into a single large-page spte. Later
10696 * page faults will create the large-page sptes.
10698 * There is no need to do this in any of the following cases:
10699 * CREATE: No dirty mappings will already exist.
10700 * MOVE/DELETE: The old mappings will already have been cleaned up by
10701 * kvm_arch_flush_shadow_memslot()
10703 if ((old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10704 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10705 kvm_mmu_zap_collapsible_sptes(kvm, new);
10708 * Enable or disable dirty logging for the slot.
10710 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old
10711 * slot have been zapped so no dirty logging updates are needed for
10713 * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible
10714 * any mappings that might be created in it will consume the
10715 * properties of the new slot and do not need to be updated here.
10717 * When PML is enabled, the kvm_x86_ops dirty logging hooks are
10718 * called to enable/disable dirty logging.
10720 * When disabling dirty logging with PML enabled, the D-bit is set
10721 * for sptes in the slot in order to prevent unnecessary GPA
10722 * logging in the PML buffer (and potential PML buffer full VMEXIT).
10723 * This guarantees leaving PML enabled for the guest's lifetime
10724 * won't have any additional overhead from PML when the guest is
10725 * running with dirty logging disabled.
10727 * When enabling dirty logging, large sptes are write-protected
10728 * so they can be split on first write. New large sptes cannot
10729 * be created for this slot until the end of the logging.
10730 * See the comments in fast_page_fault().
10731 * For small sptes, nothing is done if the dirty log is in the
10732 * initial-all-set state. Otherwise, depending on whether pml
10733 * is enabled the D-bit or the W-bit will be cleared.
10735 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10736 if (kvm_x86_ops.slot_enable_log_dirty) {
10737 kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10740 kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10741 PG_LEVEL_2M : PG_LEVEL_4K;
10744 * If we're with initial-all-set, we don't need
10745 * to write protect any small page because
10746 * they're reported as dirty already. However
10747 * we still need to write-protect huge pages
10748 * so that the page split can happen lazily on
10749 * the first write to the huge page.
10751 kvm_mmu_slot_remove_write_access(kvm, new, level);
10754 if (kvm_x86_ops.slot_disable_log_dirty)
10755 kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10759 void kvm_arch_commit_memory_region(struct kvm *kvm,
10760 const struct kvm_userspace_memory_region *mem,
10761 struct kvm_memory_slot *old,
10762 const struct kvm_memory_slot *new,
10763 enum kvm_mr_change change)
10765 if (!kvm->arch.n_requested_mmu_pages)
10766 kvm_mmu_change_mmu_pages(kvm,
10767 kvm_mmu_calculate_default_mmu_pages(kvm));
10770 * FIXME: const-ify all uses of struct kvm_memory_slot.
10772 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
10774 /* Free the arrays associated with the old memslot. */
10775 if (change == KVM_MR_MOVE)
10776 kvm_arch_free_memslot(kvm, old);
10779 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10781 kvm_mmu_zap_all(kvm);
10784 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10785 struct kvm_memory_slot *slot)
10787 kvm_page_track_flush_slot(kvm, slot);
10790 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10792 return (is_guest_mode(vcpu) &&
10793 kvm_x86_ops.guest_apic_has_interrupt &&
10794 kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10797 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10799 if (!list_empty_careful(&vcpu->async_pf.done))
10802 if (kvm_apic_has_events(vcpu))
10805 if (vcpu->arch.pv.pv_unhalted)
10808 if (vcpu->arch.exception.pending)
10811 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10812 (vcpu->arch.nmi_pending &&
10813 kvm_x86_ops.nmi_allowed(vcpu, false)))
10816 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10817 (vcpu->arch.smi_pending &&
10818 kvm_x86_ops.smi_allowed(vcpu, false)))
10821 if (kvm_arch_interrupt_allowed(vcpu) &&
10822 (kvm_cpu_has_interrupt(vcpu) ||
10823 kvm_guest_apic_has_interrupt(vcpu)))
10826 if (kvm_hv_has_stimer_pending(vcpu))
10829 if (is_guest_mode(vcpu) &&
10830 kvm_x86_ops.nested_ops->hv_timer_pending &&
10831 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10837 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10839 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10842 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10844 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10847 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10848 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10849 kvm_test_request(KVM_REQ_EVENT, vcpu))
10852 if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10858 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10860 return vcpu->arch.preempted_in_kernel;
10863 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10865 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10868 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10870 return kvm_x86_ops.interrupt_allowed(vcpu, false);
10873 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10875 /* Can't read the RIP when guest state is protected, just return 0 */
10876 if (vcpu->arch.guest_state_protected)
10879 if (is_64_bit_mode(vcpu))
10880 return kvm_rip_read(vcpu);
10881 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10882 kvm_rip_read(vcpu));
10884 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10886 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10888 return kvm_get_linear_rip(vcpu) == linear_rip;
10890 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10892 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10894 unsigned long rflags;
10896 rflags = kvm_x86_ops.get_rflags(vcpu);
10897 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10898 rflags &= ~X86_EFLAGS_TF;
10901 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10903 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10905 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10906 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10907 rflags |= X86_EFLAGS_TF;
10908 kvm_x86_ops.set_rflags(vcpu, rflags);
10911 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10913 __kvm_set_rflags(vcpu, rflags);
10914 kvm_make_request(KVM_REQ_EVENT, vcpu);
10916 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10918 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10922 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10926 r = kvm_mmu_reload(vcpu);
10930 if (!vcpu->arch.mmu->direct_map &&
10931 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10934 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10937 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10939 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
10941 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10944 static inline u32 kvm_async_pf_next_probe(u32 key)
10946 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
10949 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10951 u32 key = kvm_async_pf_hash_fn(gfn);
10953 while (vcpu->arch.apf.gfns[key] != ~0)
10954 key = kvm_async_pf_next_probe(key);
10956 vcpu->arch.apf.gfns[key] = gfn;
10959 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10962 u32 key = kvm_async_pf_hash_fn(gfn);
10964 for (i = 0; i < ASYNC_PF_PER_VCPU &&
10965 (vcpu->arch.apf.gfns[key] != gfn &&
10966 vcpu->arch.apf.gfns[key] != ~0); i++)
10967 key = kvm_async_pf_next_probe(key);
10972 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10974 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10977 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10981 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10983 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
10987 vcpu->arch.apf.gfns[i] = ~0;
10989 j = kvm_async_pf_next_probe(j);
10990 if (vcpu->arch.apf.gfns[j] == ~0)
10992 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10994 * k lies cyclically in ]i,j]
10996 * |....j i.k.| or |.k..j i...|
10998 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10999 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11004 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11006 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11008 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11012 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11014 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11016 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11017 &token, offset, sizeof(token));
11020 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11022 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11025 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11026 &val, offset, sizeof(val)))
11032 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11034 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11037 if (!kvm_pv_async_pf_enabled(vcpu) ||
11038 (vcpu->arch.apf.send_user_only && kvm_x86_ops.get_cpl(vcpu) == 0))
11044 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11046 if (unlikely(!lapic_in_kernel(vcpu) ||
11047 kvm_event_needs_reinjection(vcpu) ||
11048 vcpu->arch.exception.pending))
11051 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11055 * If interrupts are off we cannot even use an artificial
11058 return kvm_arch_interrupt_allowed(vcpu);
11061 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11062 struct kvm_async_pf *work)
11064 struct x86_exception fault;
11066 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11067 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11069 if (kvm_can_deliver_async_pf(vcpu) &&
11070 !apf_put_user_notpresent(vcpu)) {
11071 fault.vector = PF_VECTOR;
11072 fault.error_code_valid = true;
11073 fault.error_code = 0;
11074 fault.nested_page_fault = false;
11075 fault.address = work->arch.token;
11076 fault.async_page_fault = true;
11077 kvm_inject_page_fault(vcpu, &fault);
11081 * It is not possible to deliver a paravirtualized asynchronous
11082 * page fault, but putting the guest in an artificial halt state
11083 * can be beneficial nevertheless: if an interrupt arrives, we
11084 * can deliver it timely and perhaps the guest will schedule
11085 * another process. When the instruction that triggered a page
11086 * fault is retried, hopefully the page will be ready in the host.
11088 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11093 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11094 struct kvm_async_pf *work)
11096 struct kvm_lapic_irq irq = {
11097 .delivery_mode = APIC_DM_FIXED,
11098 .vector = vcpu->arch.apf.vec
11101 if (work->wakeup_all)
11102 work->arch.token = ~0; /* broadcast wakeup */
11104 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11105 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11107 if ((work->wakeup_all || work->notpresent_injected) &&
11108 kvm_pv_async_pf_enabled(vcpu) &&
11109 !apf_put_user_ready(vcpu, work->arch.token)) {
11110 vcpu->arch.apf.pageready_pending = true;
11111 kvm_apic_set_irq(vcpu, &irq, NULL);
11114 vcpu->arch.apf.halted = false;
11115 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11118 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11120 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11121 if (!vcpu->arch.apf.pageready_pending)
11122 kvm_vcpu_kick(vcpu);
11125 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11127 if (!kvm_pv_async_pf_enabled(vcpu))
11130 return apf_pageready_slot_free(vcpu);
11133 void kvm_arch_start_assignment(struct kvm *kvm)
11135 atomic_inc(&kvm->arch.assigned_device_count);
11137 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11139 void kvm_arch_end_assignment(struct kvm *kvm)
11141 atomic_dec(&kvm->arch.assigned_device_count);
11143 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11145 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11147 return atomic_read(&kvm->arch.assigned_device_count);
11149 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11151 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11153 atomic_inc(&kvm->arch.noncoherent_dma_count);
11155 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11157 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11159 atomic_dec(&kvm->arch.noncoherent_dma_count);
11161 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11163 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11165 return atomic_read(&kvm->arch.noncoherent_dma_count);
11167 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11169 bool kvm_arch_has_irq_bypass(void)
11174 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11175 struct irq_bypass_producer *prod)
11177 struct kvm_kernel_irqfd *irqfd =
11178 container_of(cons, struct kvm_kernel_irqfd, consumer);
11181 irqfd->producer = prod;
11182 kvm_arch_start_assignment(irqfd->kvm);
11183 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm,
11184 prod->irq, irqfd->gsi, 1);
11187 kvm_arch_end_assignment(irqfd->kvm);
11192 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11193 struct irq_bypass_producer *prod)
11196 struct kvm_kernel_irqfd *irqfd =
11197 container_of(cons, struct kvm_kernel_irqfd, consumer);
11199 WARN_ON(irqfd->producer != prod);
11200 irqfd->producer = NULL;
11203 * When producer of consumer is unregistered, we change back to
11204 * remapped mode, so we can re-use the current implementation
11205 * when the irq is masked/disabled or the consumer side (KVM
11206 * int this case doesn't want to receive the interrupts.
11208 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11210 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11211 " fails: %d\n", irqfd->consumer.token, ret);
11213 kvm_arch_end_assignment(irqfd->kvm);
11216 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11217 uint32_t guest_irq, bool set)
11219 return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
11222 bool kvm_vector_hashing_enabled(void)
11224 return vector_hashing;
11227 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11229 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11231 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11234 int kvm_spec_ctrl_test_value(u64 value)
11237 * test that setting IA32_SPEC_CTRL to given value
11238 * is allowed by the host processor
11242 unsigned long flags;
11245 local_irq_save(flags);
11247 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11249 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11252 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11254 local_irq_restore(flags);
11258 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11260 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11262 struct x86_exception fault;
11263 u32 access = error_code &
11264 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11266 if (!(error_code & PFERR_PRESENT_MASK) ||
11267 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11269 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11270 * tables probably do not match the TLB. Just proceed
11271 * with the error code that the processor gave.
11273 fault.vector = PF_VECTOR;
11274 fault.error_code_valid = true;
11275 fault.error_code = error_code;
11276 fault.nested_page_fault = false;
11277 fault.address = gva;
11279 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11281 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11284 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11285 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11286 * indicates whether exit to userspace is needed.
11288 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11289 struct x86_exception *e)
11291 if (r == X86EMUL_PROPAGATE_FAULT) {
11292 kvm_inject_emulated_page_fault(vcpu, e);
11297 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11298 * while handling a VMX instruction KVM could've handled the request
11299 * correctly by exiting to userspace and performing I/O but there
11300 * doesn't seem to be a real use-case behind such requests, just return
11301 * KVM_EXIT_INTERNAL_ERROR for now.
11303 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11304 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11305 vcpu->run->internal.ndata = 0;
11309 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11311 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11314 struct x86_exception e;
11316 unsigned long roots_to_free = 0;
11323 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11324 if (r != X86EMUL_CONTINUE)
11325 return kvm_handle_memory_failure(vcpu, r, &e);
11327 if (operand.pcid >> 12 != 0) {
11328 kvm_inject_gp(vcpu, 0);
11332 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11335 case INVPCID_TYPE_INDIV_ADDR:
11336 if ((!pcid_enabled && (operand.pcid != 0)) ||
11337 is_noncanonical_address(operand.gla, vcpu)) {
11338 kvm_inject_gp(vcpu, 0);
11341 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11342 return kvm_skip_emulated_instruction(vcpu);
11344 case INVPCID_TYPE_SINGLE_CTXT:
11345 if (!pcid_enabled && (operand.pcid != 0)) {
11346 kvm_inject_gp(vcpu, 0);
11350 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11351 kvm_mmu_sync_roots(vcpu);
11352 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11355 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11356 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11358 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11360 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11362 * If neither the current cr3 nor any of the prev_roots use the
11363 * given PCID, then nothing needs to be done here because a
11364 * resync will happen anyway before switching to any other CR3.
11367 return kvm_skip_emulated_instruction(vcpu);
11369 case INVPCID_TYPE_ALL_NON_GLOBAL:
11371 * Currently, KVM doesn't mark global entries in the shadow
11372 * page tables, so a non-global flush just degenerates to a
11373 * global flush. If needed, we could optimize this later by
11374 * keeping track of global entries in shadow page tables.
11378 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11379 kvm_mmu_unload(vcpu);
11380 return kvm_skip_emulated_instruction(vcpu);
11383 BUG(); /* We have already checked above that type <= 3 */
11386 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11388 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11390 struct kvm_run *run = vcpu->run;
11391 struct kvm_mmio_fragment *frag;
11394 BUG_ON(!vcpu->mmio_needed);
11396 /* Complete previous fragment */
11397 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11398 len = min(8u, frag->len);
11399 if (!vcpu->mmio_is_write)
11400 memcpy(frag->data, run->mmio.data, len);
11402 if (frag->len <= 8) {
11403 /* Switch to the next fragment. */
11405 vcpu->mmio_cur_fragment++;
11407 /* Go forward to the next mmio piece. */
11413 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11414 vcpu->mmio_needed = 0;
11416 // VMG change, at this point, we're always done
11417 // RIP has already been advanced
11421 // More MMIO is needed
11422 run->mmio.phys_addr = frag->gpa;
11423 run->mmio.len = min(8u, frag->len);
11424 run->mmio.is_write = vcpu->mmio_is_write;
11425 if (run->mmio.is_write)
11426 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11427 run->exit_reason = KVM_EXIT_MMIO;
11429 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11434 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11438 struct kvm_mmio_fragment *frag;
11443 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11444 if (handled == bytes)
11451 /*TODO: Check if need to increment number of frags */
11452 frag = vcpu->mmio_fragments;
11453 vcpu->mmio_nr_fragments = 1;
11458 vcpu->mmio_needed = 1;
11459 vcpu->mmio_cur_fragment = 0;
11461 vcpu->run->mmio.phys_addr = gpa;
11462 vcpu->run->mmio.len = min(8u, frag->len);
11463 vcpu->run->mmio.is_write = 1;
11464 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11465 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11467 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11471 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11473 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11477 struct kvm_mmio_fragment *frag;
11482 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11483 if (handled == bytes)
11490 /*TODO: Check if need to increment number of frags */
11491 frag = vcpu->mmio_fragments;
11492 vcpu->mmio_nr_fragments = 1;
11497 vcpu->mmio_needed = 1;
11498 vcpu->mmio_cur_fragment = 0;
11500 vcpu->run->mmio.phys_addr = gpa;
11501 vcpu->run->mmio.len = min(8u, frag->len);
11502 vcpu->run->mmio.is_write = 0;
11503 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11505 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11509 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11511 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11513 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11514 vcpu->arch.pio.count * vcpu->arch.pio.size);
11515 vcpu->arch.pio.count = 0;
11520 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11521 unsigned int port, void *data, unsigned int count)
11525 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11530 vcpu->arch.pio.count = 0;
11535 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11536 unsigned int port, void *data, unsigned int count)
11540 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11543 vcpu->arch.pio.count = 0;
11545 vcpu->arch.guest_ins_data = data;
11546 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11552 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11553 unsigned int port, void *data, unsigned int count,
11556 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11557 : kvm_sev_es_outs(vcpu, size, port, data, count);
11559 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11561 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11562 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11563 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11564 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11565 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11566 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11567 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11568 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11569 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11570 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11571 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11572 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11573 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11574 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11575 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11576 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11577 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11578 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11579 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11580 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11581 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11582 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11583 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11584 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11585 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11586 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11587 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);