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>
24 #include "kvm_cache_regs.h"
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
34 #include <linux/vmalloc.h>
35 #include <linux/export.h>
36 #include <linux/moduleparam.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <linux/kvm_irqfd.h>
52 #include <linux/irqbypass.h>
53 #include <linux/sched/stat.h>
54 #include <linux/sched/isolation.h>
55 #include <linux/mem_encrypt.h>
57 #include <trace/events/kvm.h>
59 #include <asm/debugreg.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
67 #include <asm/irq_remapping.h>
68 #include <asm/mshyperv.h>
69 #include <asm/hypervisor.h>
70 #include <asm/intel_pt.h>
71 #include <asm/emulate_prefix.h>
72 #include <clocksource/hyperv_timer.h>
74 #define CREATE_TRACE_POINTS
77 #define MAX_IO_MSRS 256
78 #define KVM_MAX_MCE_BANKS 32
79 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
80 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
82 #define emul_to_vcpu(ctxt) \
83 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
86 * - enable syscall per default because its emulated by KVM
87 * - enable LME and LMA per default on 64 bit KVM
91 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
93 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
96 #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__
97 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
99 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
100 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
102 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
103 static void process_nmi(struct kvm_vcpu *vcpu);
104 static void enter_smm(struct kvm_vcpu *vcpu);
105 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
106 static void store_regs(struct kvm_vcpu *vcpu);
107 static int sync_regs(struct kvm_vcpu *vcpu);
109 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
110 EXPORT_SYMBOL_GPL(kvm_x86_ops);
112 static bool __read_mostly ignore_msrs = 0;
113 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
115 static bool __read_mostly report_ignored_msrs = true;
116 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
118 unsigned int min_timer_period_us = 200;
119 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
121 static bool __read_mostly kvmclock_periodic_sync = true;
122 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
124 bool __read_mostly kvm_has_tsc_control;
125 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
126 u32 __read_mostly kvm_max_guest_tsc_khz;
127 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
128 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
129 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
130 u64 __read_mostly kvm_max_tsc_scaling_ratio;
131 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
132 u64 __read_mostly kvm_default_tsc_scaling_ratio;
133 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
135 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
136 static u32 __read_mostly tsc_tolerance_ppm = 250;
137 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
140 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
141 * adaptive tuning starting from default advancment of 1000ns. '0' disables
142 * advancement entirely. Any other value is used as-is and disables adaptive
143 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
145 static int __read_mostly lapic_timer_advance_ns = -1;
146 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
148 static bool __read_mostly vector_hashing = true;
149 module_param(vector_hashing, bool, S_IRUGO);
151 bool __read_mostly enable_vmware_backdoor = false;
152 module_param(enable_vmware_backdoor, bool, S_IRUGO);
153 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
155 static bool __read_mostly force_emulation_prefix = false;
156 module_param(force_emulation_prefix, bool, S_IRUGO);
158 int __read_mostly pi_inject_timer = -1;
159 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
161 #define KVM_NR_SHARED_MSRS 16
163 struct kvm_shared_msrs_global {
165 u32 msrs[KVM_NR_SHARED_MSRS];
168 struct kvm_shared_msrs {
169 struct user_return_notifier urn;
171 struct kvm_shared_msr_values {
174 } values[KVM_NR_SHARED_MSRS];
177 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
178 static struct kvm_shared_msrs __percpu *shared_msrs;
180 static u64 __read_mostly host_xss;
182 struct kvm_stats_debugfs_item debugfs_entries[] = {
183 { "pf_fixed", VCPU_STAT(pf_fixed) },
184 { "pf_guest", VCPU_STAT(pf_guest) },
185 { "tlb_flush", VCPU_STAT(tlb_flush) },
186 { "invlpg", VCPU_STAT(invlpg) },
187 { "exits", VCPU_STAT(exits) },
188 { "io_exits", VCPU_STAT(io_exits) },
189 { "mmio_exits", VCPU_STAT(mmio_exits) },
190 { "signal_exits", VCPU_STAT(signal_exits) },
191 { "irq_window", VCPU_STAT(irq_window_exits) },
192 { "nmi_window", VCPU_STAT(nmi_window_exits) },
193 { "halt_exits", VCPU_STAT(halt_exits) },
194 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
195 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
196 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
197 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
198 { "hypercalls", VCPU_STAT(hypercalls) },
199 { "request_irq", VCPU_STAT(request_irq_exits) },
200 { "irq_exits", VCPU_STAT(irq_exits) },
201 { "host_state_reload", VCPU_STAT(host_state_reload) },
202 { "fpu_reload", VCPU_STAT(fpu_reload) },
203 { "insn_emulation", VCPU_STAT(insn_emulation) },
204 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
205 { "irq_injections", VCPU_STAT(irq_injections) },
206 { "nmi_injections", VCPU_STAT(nmi_injections) },
207 { "req_event", VCPU_STAT(req_event) },
208 { "l1d_flush", VCPU_STAT(l1d_flush) },
209 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
210 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
211 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
212 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
213 { "mmu_flooded", VM_STAT(mmu_flooded) },
214 { "mmu_recycled", VM_STAT(mmu_recycled) },
215 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
216 { "mmu_unsync", VM_STAT(mmu_unsync) },
217 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
218 { "largepages", VM_STAT(lpages, .mode = 0444) },
219 { "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) },
220 { "max_mmu_page_hash_collisions",
221 VM_STAT(max_mmu_page_hash_collisions) },
225 u64 __read_mostly host_xcr0;
227 struct kmem_cache *x86_fpu_cache;
228 EXPORT_SYMBOL_GPL(x86_fpu_cache);
230 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
232 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
235 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
236 vcpu->arch.apf.gfns[i] = ~0;
239 static void kvm_on_user_return(struct user_return_notifier *urn)
242 struct kvm_shared_msrs *locals
243 = container_of(urn, struct kvm_shared_msrs, urn);
244 struct kvm_shared_msr_values *values;
248 * Disabling irqs at this point since the following code could be
249 * interrupted and executed through kvm_arch_hardware_disable()
251 local_irq_save(flags);
252 if (locals->registered) {
253 locals->registered = false;
254 user_return_notifier_unregister(urn);
256 local_irq_restore(flags);
257 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
258 values = &locals->values[slot];
259 if (values->host != values->curr) {
260 wrmsrl(shared_msrs_global.msrs[slot], values->host);
261 values->curr = values->host;
266 void kvm_define_shared_msr(unsigned slot, u32 msr)
268 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
269 shared_msrs_global.msrs[slot] = msr;
270 if (slot >= shared_msrs_global.nr)
271 shared_msrs_global.nr = slot + 1;
273 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
275 static void kvm_shared_msr_cpu_online(void)
277 unsigned int cpu = smp_processor_id();
278 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
282 for (i = 0; i < shared_msrs_global.nr; ++i) {
283 rdmsrl_safe(shared_msrs_global.msrs[i], &value);
284 smsr->values[i].host = value;
285 smsr->values[i].curr = value;
289 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
291 unsigned int cpu = smp_processor_id();
292 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
295 value = (value & mask) | (smsr->values[slot].host & ~mask);
296 if (value == smsr->values[slot].curr)
298 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
302 smsr->values[slot].curr = value;
303 if (!smsr->registered) {
304 smsr->urn.on_user_return = kvm_on_user_return;
305 user_return_notifier_register(&smsr->urn);
306 smsr->registered = true;
310 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
312 static void drop_user_return_notifiers(void)
314 unsigned int cpu = smp_processor_id();
315 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
317 if (smsr->registered)
318 kvm_on_user_return(&smsr->urn);
321 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
323 return vcpu->arch.apic_base;
325 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
327 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
329 return kvm_apic_mode(kvm_get_apic_base(vcpu));
331 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
333 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
335 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
336 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
337 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
338 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
340 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
342 if (!msr_info->host_initiated) {
343 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
345 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
349 kvm_lapic_set_base(vcpu, msr_info->data);
352 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
354 asmlinkage __visible void kvm_spurious_fault(void)
356 /* Fault while not rebooting. We want the trace. */
357 BUG_ON(!kvm_rebooting);
359 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
361 #define EXCPT_BENIGN 0
362 #define EXCPT_CONTRIBUTORY 1
365 static int exception_class(int vector)
375 return EXCPT_CONTRIBUTORY;
382 #define EXCPT_FAULT 0
384 #define EXCPT_ABORT 2
385 #define EXCPT_INTERRUPT 3
387 static int exception_type(int vector)
391 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
392 return EXCPT_INTERRUPT;
396 /* #DB is trap, as instruction watchpoints are handled elsewhere */
397 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
400 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
403 /* Reserved exceptions will result in fault */
407 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
409 unsigned nr = vcpu->arch.exception.nr;
410 bool has_payload = vcpu->arch.exception.has_payload;
411 unsigned long payload = vcpu->arch.exception.payload;
419 * "Certain debug exceptions may clear bit 0-3. The
420 * remaining contents of the DR6 register are never
421 * cleared by the processor".
423 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
425 * DR6.RTM is set by all #DB exceptions that don't clear it.
427 vcpu->arch.dr6 |= DR6_RTM;
428 vcpu->arch.dr6 |= payload;
430 * Bit 16 should be set in the payload whenever the #DB
431 * exception should clear DR6.RTM. This makes the payload
432 * compatible with the pending debug exceptions under VMX.
433 * Though not currently documented in the SDM, this also
434 * makes the payload compatible with the exit qualification
435 * for #DB exceptions under VMX.
437 vcpu->arch.dr6 ^= payload & DR6_RTM;
440 vcpu->arch.cr2 = payload;
444 vcpu->arch.exception.has_payload = false;
445 vcpu->arch.exception.payload = 0;
447 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
449 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
450 unsigned nr, bool has_error, u32 error_code,
451 bool has_payload, unsigned long payload, bool reinject)
456 kvm_make_request(KVM_REQ_EVENT, vcpu);
458 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
460 if (has_error && !is_protmode(vcpu))
464 * On vmentry, vcpu->arch.exception.pending is only
465 * true if an event injection was blocked by
466 * nested_run_pending. In that case, however,
467 * vcpu_enter_guest requests an immediate exit,
468 * and the guest shouldn't proceed far enough to
471 WARN_ON_ONCE(vcpu->arch.exception.pending);
472 vcpu->arch.exception.injected = true;
473 if (WARN_ON_ONCE(has_payload)) {
475 * A reinjected event has already
476 * delivered its payload.
482 vcpu->arch.exception.pending = true;
483 vcpu->arch.exception.injected = false;
485 vcpu->arch.exception.has_error_code = has_error;
486 vcpu->arch.exception.nr = nr;
487 vcpu->arch.exception.error_code = error_code;
488 vcpu->arch.exception.has_payload = has_payload;
489 vcpu->arch.exception.payload = payload;
491 * In guest mode, payload delivery should be deferred,
492 * so that the L1 hypervisor can intercept #PF before
493 * CR2 is modified (or intercept #DB before DR6 is
494 * modified under nVMX). However, for ABI
495 * compatibility with KVM_GET_VCPU_EVENTS and
496 * KVM_SET_VCPU_EVENTS, we can't delay payload
497 * delivery unless userspace has enabled this
498 * functionality via the per-VM capability,
499 * KVM_CAP_EXCEPTION_PAYLOAD.
501 if (!vcpu->kvm->arch.exception_payload_enabled ||
502 !is_guest_mode(vcpu))
503 kvm_deliver_exception_payload(vcpu);
507 /* to check exception */
508 prev_nr = vcpu->arch.exception.nr;
509 if (prev_nr == DF_VECTOR) {
510 /* triple fault -> shutdown */
511 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
514 class1 = exception_class(prev_nr);
515 class2 = exception_class(nr);
516 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
517 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
519 * Generate double fault per SDM Table 5-5. Set
520 * exception.pending = true so that the double fault
521 * can trigger a nested vmexit.
523 vcpu->arch.exception.pending = true;
524 vcpu->arch.exception.injected = false;
525 vcpu->arch.exception.has_error_code = true;
526 vcpu->arch.exception.nr = DF_VECTOR;
527 vcpu->arch.exception.error_code = 0;
528 vcpu->arch.exception.has_payload = false;
529 vcpu->arch.exception.payload = 0;
531 /* replace previous exception with a new one in a hope
532 that instruction re-execution will regenerate lost
537 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
539 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
541 EXPORT_SYMBOL_GPL(kvm_queue_exception);
543 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
545 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
547 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
549 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
550 unsigned long payload)
552 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
555 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
556 u32 error_code, unsigned long payload)
558 kvm_multiple_exception(vcpu, nr, true, error_code,
559 true, payload, false);
562 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
565 kvm_inject_gp(vcpu, 0);
567 return kvm_skip_emulated_instruction(vcpu);
571 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
573 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
575 ++vcpu->stat.pf_guest;
576 vcpu->arch.exception.nested_apf =
577 is_guest_mode(vcpu) && fault->async_page_fault;
578 if (vcpu->arch.exception.nested_apf) {
579 vcpu->arch.apf.nested_apf_token = fault->address;
580 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
582 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
586 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
588 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
590 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
591 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
593 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
595 return fault->nested_page_fault;
598 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
600 atomic_inc(&vcpu->arch.nmi_queued);
601 kvm_make_request(KVM_REQ_NMI, vcpu);
603 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
605 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
607 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
609 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
611 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
613 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
615 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
618 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
619 * a #GP and return false.
621 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
623 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
625 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
628 EXPORT_SYMBOL_GPL(kvm_require_cpl);
630 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
632 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
635 kvm_queue_exception(vcpu, UD_VECTOR);
638 EXPORT_SYMBOL_GPL(kvm_require_dr);
641 * This function will be used to read from the physical memory of the currently
642 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
643 * can read from guest physical or from the guest's guest physical memory.
645 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
646 gfn_t ngfn, void *data, int offset, int len,
649 struct x86_exception exception;
653 ngpa = gfn_to_gpa(ngfn);
654 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
655 if (real_gfn == UNMAPPED_GVA)
658 real_gfn = gpa_to_gfn(real_gfn);
660 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
662 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
664 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
665 void *data, int offset, int len, u32 access)
667 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
668 data, offset, len, access);
671 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
673 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
678 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
680 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
682 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
683 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
686 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
688 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
689 offset * sizeof(u64), sizeof(pdpte),
690 PFERR_USER_MASK|PFERR_WRITE_MASK);
695 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
696 if ((pdpte[i] & PT_PRESENT_MASK) &&
697 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
704 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
705 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
711 EXPORT_SYMBOL_GPL(load_pdptrs);
713 bool pdptrs_changed(struct kvm_vcpu *vcpu)
715 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
720 if (!is_pae_paging(vcpu))
723 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
726 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
727 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
728 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
729 PFERR_USER_MASK | PFERR_WRITE_MASK);
733 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
735 EXPORT_SYMBOL_GPL(pdptrs_changed);
737 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
739 unsigned long old_cr0 = kvm_read_cr0(vcpu);
740 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
745 if (cr0 & 0xffffffff00000000UL)
749 cr0 &= ~CR0_RESERVED_BITS;
751 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
754 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
757 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
759 if ((vcpu->arch.efer & EFER_LME)) {
764 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
769 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
774 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
777 kvm_x86_ops->set_cr0(vcpu, cr0);
779 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
780 kvm_clear_async_pf_completion_queue(vcpu);
781 kvm_async_pf_hash_reset(vcpu);
784 if ((cr0 ^ old_cr0) & update_bits)
785 kvm_mmu_reset_context(vcpu);
787 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
788 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
789 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
790 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
794 EXPORT_SYMBOL_GPL(kvm_set_cr0);
796 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
798 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
800 EXPORT_SYMBOL_GPL(kvm_lmsw);
802 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
804 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
806 if (vcpu->arch.xcr0 != host_xcr0)
807 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
809 if (vcpu->arch.xsaves_enabled &&
810 vcpu->arch.ia32_xss != host_xss)
811 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
814 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
816 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
818 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
820 if (vcpu->arch.xcr0 != host_xcr0)
821 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
823 if (vcpu->arch.xsaves_enabled &&
824 vcpu->arch.ia32_xss != host_xss)
825 wrmsrl(MSR_IA32_XSS, host_xss);
829 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
831 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
834 u64 old_xcr0 = vcpu->arch.xcr0;
837 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
838 if (index != XCR_XFEATURE_ENABLED_MASK)
840 if (!(xcr0 & XFEATURE_MASK_FP))
842 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
846 * Do not allow the guest to set bits that we do not support
847 * saving. However, xcr0 bit 0 is always set, even if the
848 * emulated CPU does not support XSAVE (see fx_init).
850 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
851 if (xcr0 & ~valid_bits)
854 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
855 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
858 if (xcr0 & XFEATURE_MASK_AVX512) {
859 if (!(xcr0 & XFEATURE_MASK_YMM))
861 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
864 vcpu->arch.xcr0 = xcr0;
866 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
867 kvm_update_cpuid(vcpu);
871 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
873 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
874 __kvm_set_xcr(vcpu, index, xcr)) {
875 kvm_inject_gp(vcpu, 0);
880 EXPORT_SYMBOL_GPL(kvm_set_xcr);
882 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
884 if (cr4 & CR4_RESERVED_BITS)
887 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
890 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
893 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
896 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
899 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
902 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
905 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
911 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
913 unsigned long old_cr4 = kvm_read_cr4(vcpu);
914 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
915 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
917 if (kvm_valid_cr4(vcpu, cr4))
920 if (is_long_mode(vcpu)) {
921 if (!(cr4 & X86_CR4_PAE))
923 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
924 && ((cr4 ^ old_cr4) & pdptr_bits)
925 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
929 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
930 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
933 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
934 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
938 if (kvm_x86_ops->set_cr4(vcpu, cr4))
941 if (((cr4 ^ old_cr4) & pdptr_bits) ||
942 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
943 kvm_mmu_reset_context(vcpu);
945 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
946 kvm_update_cpuid(vcpu);
950 EXPORT_SYMBOL_GPL(kvm_set_cr4);
952 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
954 bool skip_tlb_flush = false;
956 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
959 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
960 cr3 &= ~X86_CR3_PCID_NOFLUSH;
964 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
965 if (!skip_tlb_flush) {
966 kvm_mmu_sync_roots(vcpu);
967 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
972 if (is_long_mode(vcpu) &&
973 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
975 else if (is_pae_paging(vcpu) &&
976 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
979 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
980 vcpu->arch.cr3 = cr3;
981 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
985 EXPORT_SYMBOL_GPL(kvm_set_cr3);
987 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
989 if (cr8 & CR8_RESERVED_BITS)
991 if (lapic_in_kernel(vcpu))
992 kvm_lapic_set_tpr(vcpu, cr8);
994 vcpu->arch.cr8 = cr8;
997 EXPORT_SYMBOL_GPL(kvm_set_cr8);
999 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1001 if (lapic_in_kernel(vcpu))
1002 return kvm_lapic_get_cr8(vcpu);
1004 return vcpu->arch.cr8;
1006 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1008 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1012 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1013 for (i = 0; i < KVM_NR_DB_REGS; i++)
1014 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1015 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1019 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1021 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1022 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1025 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1029 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1030 dr7 = vcpu->arch.guest_debug_dr7;
1032 dr7 = vcpu->arch.dr7;
1033 kvm_x86_ops->set_dr7(vcpu, dr7);
1034 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1035 if (dr7 & DR7_BP_EN_MASK)
1036 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1039 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1041 u64 fixed = DR6_FIXED_1;
1043 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1048 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1052 vcpu->arch.db[dr] = val;
1053 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1054 vcpu->arch.eff_db[dr] = val;
1059 if (val & 0xffffffff00000000ULL)
1060 return -1; /* #GP */
1061 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1062 kvm_update_dr6(vcpu);
1067 if (val & 0xffffffff00000000ULL)
1068 return -1; /* #GP */
1069 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1070 kvm_update_dr7(vcpu);
1077 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1079 if (__kvm_set_dr(vcpu, dr, val)) {
1080 kvm_inject_gp(vcpu, 0);
1085 EXPORT_SYMBOL_GPL(kvm_set_dr);
1087 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1091 *val = vcpu->arch.db[dr];
1096 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1097 *val = vcpu->arch.dr6;
1099 *val = kvm_x86_ops->get_dr6(vcpu);
1104 *val = vcpu->arch.dr7;
1109 EXPORT_SYMBOL_GPL(kvm_get_dr);
1111 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1113 u32 ecx = kvm_rcx_read(vcpu);
1117 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1120 kvm_rax_write(vcpu, (u32)data);
1121 kvm_rdx_write(vcpu, data >> 32);
1124 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1127 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1128 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1130 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1131 * extract the supported MSRs from the related const lists.
1132 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1133 * capabilities of the host cpu. This capabilities test skips MSRs that are
1134 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1135 * may depend on host virtualization features rather than host cpu features.
1138 static const u32 msrs_to_save_all[] = {
1139 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1141 #ifdef CONFIG_X86_64
1142 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1144 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1145 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1147 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1148 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1149 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1150 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1151 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1152 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1153 MSR_IA32_UMWAIT_CONTROL,
1155 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1156 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1157 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1158 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1159 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1160 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1161 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1162 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1163 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1164 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1165 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1166 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1167 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1168 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1169 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1170 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1171 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1172 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1173 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1174 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1175 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1176 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1179 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1180 static unsigned num_msrs_to_save;
1182 static const u32 emulated_msrs_all[] = {
1183 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1184 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1185 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1186 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1187 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1188 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1189 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1191 HV_X64_MSR_VP_INDEX,
1192 HV_X64_MSR_VP_RUNTIME,
1193 HV_X64_MSR_SCONTROL,
1194 HV_X64_MSR_STIMER0_CONFIG,
1195 HV_X64_MSR_VP_ASSIST_PAGE,
1196 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1197 HV_X64_MSR_TSC_EMULATION_STATUS,
1199 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1202 MSR_IA32_TSC_ADJUST,
1203 MSR_IA32_TSCDEADLINE,
1204 MSR_IA32_ARCH_CAPABILITIES,
1205 MSR_IA32_MISC_ENABLE,
1206 MSR_IA32_MCG_STATUS,
1208 MSR_IA32_MCG_EXT_CTL,
1212 MSR_MISC_FEATURES_ENABLES,
1213 MSR_AMD64_VIRT_SPEC_CTRL,
1217 * The following list leaves out MSRs whose values are determined
1218 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1219 * We always support the "true" VMX control MSRs, even if the host
1220 * processor does not, so I am putting these registers here rather
1221 * than in msrs_to_save_all.
1224 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1225 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1226 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1227 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1229 MSR_IA32_VMX_CR0_FIXED0,
1230 MSR_IA32_VMX_CR4_FIXED0,
1231 MSR_IA32_VMX_VMCS_ENUM,
1232 MSR_IA32_VMX_PROCBASED_CTLS2,
1233 MSR_IA32_VMX_EPT_VPID_CAP,
1234 MSR_IA32_VMX_VMFUNC,
1237 MSR_KVM_POLL_CONTROL,
1240 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1241 static unsigned num_emulated_msrs;
1244 * List of msr numbers which are used to expose MSR-based features that
1245 * can be used by a hypervisor to validate requested CPU features.
1247 static const u32 msr_based_features_all[] = {
1249 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1250 MSR_IA32_VMX_PINBASED_CTLS,
1251 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1252 MSR_IA32_VMX_PROCBASED_CTLS,
1253 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1254 MSR_IA32_VMX_EXIT_CTLS,
1255 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1256 MSR_IA32_VMX_ENTRY_CTLS,
1258 MSR_IA32_VMX_CR0_FIXED0,
1259 MSR_IA32_VMX_CR0_FIXED1,
1260 MSR_IA32_VMX_CR4_FIXED0,
1261 MSR_IA32_VMX_CR4_FIXED1,
1262 MSR_IA32_VMX_VMCS_ENUM,
1263 MSR_IA32_VMX_PROCBASED_CTLS2,
1264 MSR_IA32_VMX_EPT_VPID_CAP,
1265 MSR_IA32_VMX_VMFUNC,
1269 MSR_IA32_ARCH_CAPABILITIES,
1272 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1273 static unsigned int num_msr_based_features;
1275 static u64 kvm_get_arch_capabilities(void)
1279 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1280 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1283 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1284 * the nested hypervisor runs with NX huge pages. If it is not,
1285 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1286 * L1 guests, so it need not worry about its own (L2) guests.
1288 data |= ARCH_CAP_PSCHANGE_MC_NO;
1291 * If we're doing cache flushes (either "always" or "cond")
1292 * we will do one whenever the guest does a vmlaunch/vmresume.
1293 * If an outer hypervisor is doing the cache flush for us
1294 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1295 * capability to the guest too, and if EPT is disabled we're not
1296 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1297 * require a nested hypervisor to do a flush of its own.
1299 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1300 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1302 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1303 data |= ARCH_CAP_RDCL_NO;
1304 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1305 data |= ARCH_CAP_SSB_NO;
1306 if (!boot_cpu_has_bug(X86_BUG_MDS))
1307 data |= ARCH_CAP_MDS_NO;
1310 * On TAA affected systems:
1311 * - nothing to do if TSX is disabled on the host.
1312 * - we emulate TSX_CTRL if present on the host.
1313 * This lets the guest use VERW to clear CPU buffers.
1315 if (!boot_cpu_has(X86_FEATURE_RTM))
1316 data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1317 else if (!boot_cpu_has_bug(X86_BUG_TAA))
1318 data |= ARCH_CAP_TAA_NO;
1323 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1325 switch (msr->index) {
1326 case MSR_IA32_ARCH_CAPABILITIES:
1327 msr->data = kvm_get_arch_capabilities();
1329 case MSR_IA32_UCODE_REV:
1330 rdmsrl_safe(msr->index, &msr->data);
1333 if (kvm_x86_ops->get_msr_feature(msr))
1339 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1341 struct kvm_msr_entry msr;
1345 r = kvm_get_msr_feature(&msr);
1354 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1356 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1359 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1362 if (efer & (EFER_LME | EFER_LMA) &&
1363 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1366 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1372 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1374 if (efer & efer_reserved_bits)
1377 return __kvm_valid_efer(vcpu, efer);
1379 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1381 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1383 u64 old_efer = vcpu->arch.efer;
1384 u64 efer = msr_info->data;
1386 if (efer & efer_reserved_bits)
1389 if (!msr_info->host_initiated) {
1390 if (!__kvm_valid_efer(vcpu, efer))
1393 if (is_paging(vcpu) &&
1394 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1399 efer |= vcpu->arch.efer & EFER_LMA;
1401 kvm_x86_ops->set_efer(vcpu, efer);
1403 /* Update reserved bits */
1404 if ((efer ^ old_efer) & EFER_NX)
1405 kvm_mmu_reset_context(vcpu);
1410 void kvm_enable_efer_bits(u64 mask)
1412 efer_reserved_bits &= ~mask;
1414 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1417 * Write @data into the MSR specified by @index. Select MSR specific fault
1418 * checks are bypassed if @host_initiated is %true.
1419 * Returns 0 on success, non-0 otherwise.
1420 * Assumes vcpu_load() was already called.
1422 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1423 bool host_initiated)
1425 struct msr_data msr;
1430 case MSR_KERNEL_GS_BASE:
1433 if (is_noncanonical_address(data, vcpu))
1436 case MSR_IA32_SYSENTER_EIP:
1437 case MSR_IA32_SYSENTER_ESP:
1439 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1440 * non-canonical address is written on Intel but not on
1441 * AMD (which ignores the top 32-bits, because it does
1442 * not implement 64-bit SYSENTER).
1444 * 64-bit code should hence be able to write a non-canonical
1445 * value on AMD. Making the address canonical ensures that
1446 * vmentry does not fail on Intel after writing a non-canonical
1447 * value, and that something deterministic happens if the guest
1448 * invokes 64-bit SYSENTER.
1450 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1455 msr.host_initiated = host_initiated;
1457 return kvm_x86_ops->set_msr(vcpu, &msr);
1461 * Read the MSR specified by @index into @data. Select MSR specific fault
1462 * checks are bypassed if @host_initiated is %true.
1463 * Returns 0 on success, non-0 otherwise.
1464 * Assumes vcpu_load() was already called.
1466 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1467 bool host_initiated)
1469 struct msr_data msr;
1473 msr.host_initiated = host_initiated;
1475 ret = kvm_x86_ops->get_msr(vcpu, &msr);
1481 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1483 return __kvm_get_msr(vcpu, index, data, false);
1485 EXPORT_SYMBOL_GPL(kvm_get_msr);
1487 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1489 return __kvm_set_msr(vcpu, index, data, false);
1491 EXPORT_SYMBOL_GPL(kvm_set_msr);
1493 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1495 u32 ecx = kvm_rcx_read(vcpu);
1498 if (kvm_get_msr(vcpu, ecx, &data)) {
1499 trace_kvm_msr_read_ex(ecx);
1500 kvm_inject_gp(vcpu, 0);
1504 trace_kvm_msr_read(ecx, data);
1506 kvm_rax_write(vcpu, data & -1u);
1507 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1508 return kvm_skip_emulated_instruction(vcpu);
1510 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1512 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1514 u32 ecx = kvm_rcx_read(vcpu);
1515 u64 data = kvm_read_edx_eax(vcpu);
1517 if (kvm_set_msr(vcpu, ecx, data)) {
1518 trace_kvm_msr_write_ex(ecx, data);
1519 kvm_inject_gp(vcpu, 0);
1523 trace_kvm_msr_write(ecx, data);
1524 return kvm_skip_emulated_instruction(vcpu);
1526 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1529 * Adapt set_msr() to msr_io()'s calling convention
1531 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1533 return __kvm_get_msr(vcpu, index, data, true);
1536 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1538 return __kvm_set_msr(vcpu, index, *data, true);
1541 #ifdef CONFIG_X86_64
1542 struct pvclock_clock {
1550 struct pvclock_gtod_data {
1553 struct pvclock_clock clock; /* extract of a clocksource struct */
1554 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1560 u64 monotonic_raw_nsec;
1563 static struct pvclock_gtod_data pvclock_gtod_data;
1565 static void update_pvclock_gtod(struct timekeeper *tk)
1567 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1568 u64 boot_ns, boot_ns_raw;
1570 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1571 boot_ns_raw = ktime_to_ns(ktime_add(tk->tkr_raw.base, tk->offs_boot));
1573 write_seqcount_begin(&vdata->seq);
1575 /* copy pvclock gtod data */
1576 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1577 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1578 vdata->clock.mask = tk->tkr_mono.mask;
1579 vdata->clock.mult = tk->tkr_mono.mult;
1580 vdata->clock.shift = tk->tkr_mono.shift;
1582 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->archdata.vclock_mode;
1583 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1584 vdata->raw_clock.mask = tk->tkr_raw.mask;
1585 vdata->raw_clock.mult = tk->tkr_raw.mult;
1586 vdata->raw_clock.shift = tk->tkr_raw.shift;
1588 vdata->boot_ns = boot_ns;
1589 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1591 vdata->wall_time_sec = tk->xtime_sec;
1593 vdata->boot_ns_raw = boot_ns_raw;
1594 vdata->monotonic_raw_nsec = tk->tkr_raw.xtime_nsec;
1596 write_seqcount_end(&vdata->seq);
1600 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1602 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1603 kvm_vcpu_kick(vcpu);
1606 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1610 struct pvclock_wall_clock wc;
1611 struct timespec64 boot;
1616 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1621 ++version; /* first time write, random junk */
1625 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1629 * The guest calculates current wall clock time by adding
1630 * system time (updated by kvm_guest_time_update below) to the
1631 * wall clock specified here. guest system time equals host
1632 * system time for us, thus we must fill in host boot time here.
1634 getboottime64(&boot);
1636 if (kvm->arch.kvmclock_offset) {
1637 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1638 boot = timespec64_sub(boot, ts);
1640 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1641 wc.nsec = boot.tv_nsec;
1642 wc.version = version;
1644 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1647 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1650 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1652 do_shl32_div32(dividend, divisor);
1656 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1657 s8 *pshift, u32 *pmultiplier)
1665 scaled64 = scaled_hz;
1666 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1671 tps32 = (uint32_t)tps64;
1672 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1673 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1681 *pmultiplier = div_frac(scaled64, tps32);
1684 #ifdef CONFIG_X86_64
1685 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1688 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1689 static unsigned long max_tsc_khz;
1691 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1693 u64 v = (u64)khz * (1000000 + ppm);
1698 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1702 /* Guest TSC same frequency as host TSC? */
1704 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1708 /* TSC scaling supported? */
1709 if (!kvm_has_tsc_control) {
1710 if (user_tsc_khz > tsc_khz) {
1711 vcpu->arch.tsc_catchup = 1;
1712 vcpu->arch.tsc_always_catchup = 1;
1715 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1720 /* TSC scaling required - calculate ratio */
1721 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1722 user_tsc_khz, tsc_khz);
1724 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1725 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1730 vcpu->arch.tsc_scaling_ratio = ratio;
1734 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1736 u32 thresh_lo, thresh_hi;
1737 int use_scaling = 0;
1739 /* tsc_khz can be zero if TSC calibration fails */
1740 if (user_tsc_khz == 0) {
1741 /* set tsc_scaling_ratio to a safe value */
1742 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1746 /* Compute a scale to convert nanoseconds in TSC cycles */
1747 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1748 &vcpu->arch.virtual_tsc_shift,
1749 &vcpu->arch.virtual_tsc_mult);
1750 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1753 * Compute the variation in TSC rate which is acceptable
1754 * within the range of tolerance and decide if the
1755 * rate being applied is within that bounds of the hardware
1756 * rate. If so, no scaling or compensation need be done.
1758 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1759 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1760 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1761 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1764 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1767 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1769 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1770 vcpu->arch.virtual_tsc_mult,
1771 vcpu->arch.virtual_tsc_shift);
1772 tsc += vcpu->arch.this_tsc_write;
1776 static inline int gtod_is_based_on_tsc(int mode)
1778 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1781 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1783 #ifdef CONFIG_X86_64
1785 struct kvm_arch *ka = &vcpu->kvm->arch;
1786 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1788 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1789 atomic_read(&vcpu->kvm->online_vcpus));
1792 * Once the masterclock is enabled, always perform request in
1793 * order to update it.
1795 * In order to enable masterclock, the host clocksource must be TSC
1796 * and the vcpus need to have matched TSCs. When that happens,
1797 * perform request to enable masterclock.
1799 if (ka->use_master_clock ||
1800 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1801 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1803 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1804 atomic_read(&vcpu->kvm->online_vcpus),
1805 ka->use_master_clock, gtod->clock.vclock_mode);
1809 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1811 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1812 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1816 * Multiply tsc by a fixed point number represented by ratio.
1818 * The most significant 64-N bits (mult) of ratio represent the
1819 * integral part of the fixed point number; the remaining N bits
1820 * (frac) represent the fractional part, ie. ratio represents a fixed
1821 * point number (mult + frac * 2^(-N)).
1823 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1825 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1827 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1830 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1833 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1835 if (ratio != kvm_default_tsc_scaling_ratio)
1836 _tsc = __scale_tsc(ratio, tsc);
1840 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1842 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1846 tsc = kvm_scale_tsc(vcpu, rdtsc());
1848 return target_tsc - tsc;
1851 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1853 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1855 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1857 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1859 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1861 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1864 static inline bool kvm_check_tsc_unstable(void)
1866 #ifdef CONFIG_X86_64
1868 * TSC is marked unstable when we're running on Hyper-V,
1869 * 'TSC page' clocksource is good.
1871 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1874 return check_tsc_unstable();
1877 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1879 struct kvm *kvm = vcpu->kvm;
1880 u64 offset, ns, elapsed;
1881 unsigned long flags;
1883 bool already_matched;
1884 u64 data = msr->data;
1885 bool synchronizing = false;
1887 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1888 offset = kvm_compute_tsc_offset(vcpu, data);
1889 ns = ktime_get_boottime_ns();
1890 elapsed = ns - kvm->arch.last_tsc_nsec;
1892 if (vcpu->arch.virtual_tsc_khz) {
1893 if (data == 0 && msr->host_initiated) {
1895 * detection of vcpu initialization -- need to sync
1896 * with other vCPUs. This particularly helps to keep
1897 * kvm_clock stable after CPU hotplug
1899 synchronizing = true;
1901 u64 tsc_exp = kvm->arch.last_tsc_write +
1902 nsec_to_cycles(vcpu, elapsed);
1903 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1905 * Special case: TSC write with a small delta (1 second)
1906 * of virtual cycle time against real time is
1907 * interpreted as an attempt to synchronize the CPU.
1909 synchronizing = data < tsc_exp + tsc_hz &&
1910 data + tsc_hz > tsc_exp;
1915 * For a reliable TSC, we can match TSC offsets, and for an unstable
1916 * TSC, we add elapsed time in this computation. We could let the
1917 * compensation code attempt to catch up if we fall behind, but
1918 * it's better to try to match offsets from the beginning.
1920 if (synchronizing &&
1921 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1922 if (!kvm_check_tsc_unstable()) {
1923 offset = kvm->arch.cur_tsc_offset;
1925 u64 delta = nsec_to_cycles(vcpu, elapsed);
1927 offset = kvm_compute_tsc_offset(vcpu, data);
1930 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1933 * We split periods of matched TSC writes into generations.
1934 * For each generation, we track the original measured
1935 * nanosecond time, offset, and write, so if TSCs are in
1936 * sync, we can match exact offset, and if not, we can match
1937 * exact software computation in compute_guest_tsc()
1939 * These values are tracked in kvm->arch.cur_xxx variables.
1941 kvm->arch.cur_tsc_generation++;
1942 kvm->arch.cur_tsc_nsec = ns;
1943 kvm->arch.cur_tsc_write = data;
1944 kvm->arch.cur_tsc_offset = offset;
1949 * We also track th most recent recorded KHZ, write and time to
1950 * allow the matching interval to be extended at each write.
1952 kvm->arch.last_tsc_nsec = ns;
1953 kvm->arch.last_tsc_write = data;
1954 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1956 vcpu->arch.last_guest_tsc = data;
1958 /* Keep track of which generation this VCPU has synchronized to */
1959 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1960 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1961 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1963 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1964 update_ia32_tsc_adjust_msr(vcpu, offset);
1966 kvm_vcpu_write_tsc_offset(vcpu, offset);
1967 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1969 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1971 kvm->arch.nr_vcpus_matched_tsc = 0;
1972 } else if (!already_matched) {
1973 kvm->arch.nr_vcpus_matched_tsc++;
1976 kvm_track_tsc_matching(vcpu);
1977 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1980 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1982 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1985 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1986 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1989 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1991 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1992 WARN_ON(adjustment < 0);
1993 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1994 adjust_tsc_offset_guest(vcpu, adjustment);
1997 #ifdef CONFIG_X86_64
1999 static u64 read_tsc(void)
2001 u64 ret = (u64)rdtsc_ordered();
2002 u64 last = pvclock_gtod_data.clock.cycle_last;
2004 if (likely(ret >= last))
2008 * GCC likes to generate cmov here, but this branch is extremely
2009 * predictable (it's just a function of time and the likely is
2010 * very likely) and there's a data dependence, so force GCC
2011 * to generate a branch instead. I don't barrier() because
2012 * we don't actually need a barrier, and if this function
2013 * ever gets inlined it will generate worse code.
2019 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2025 switch (clock->vclock_mode) {
2026 case VCLOCK_HVCLOCK:
2027 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2029 if (tsc_pg_val != U64_MAX) {
2030 /* TSC page valid */
2031 *mode = VCLOCK_HVCLOCK;
2032 v = (tsc_pg_val - clock->cycle_last) &
2035 /* TSC page invalid */
2036 *mode = VCLOCK_NONE;
2041 *tsc_timestamp = read_tsc();
2042 v = (*tsc_timestamp - clock->cycle_last) &
2046 *mode = VCLOCK_NONE;
2049 if (*mode == VCLOCK_NONE)
2050 *tsc_timestamp = v = 0;
2052 return v * clock->mult;
2055 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2057 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2063 seq = read_seqcount_begin(>od->seq);
2064 ns = gtod->monotonic_raw_nsec;
2065 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2066 ns >>= gtod->clock.shift;
2067 ns += gtod->boot_ns_raw;
2068 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2074 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2076 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2082 seq = read_seqcount_begin(>od->seq);
2083 ts->tv_sec = gtod->wall_time_sec;
2084 ns = gtod->nsec_base;
2085 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2086 ns >>= gtod->clock.shift;
2087 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2089 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2095 /* returns true if host is using TSC based clocksource */
2096 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2098 /* checked again under seqlock below */
2099 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2102 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2106 /* returns true if host is using TSC based clocksource */
2107 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2110 /* checked again under seqlock below */
2111 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2114 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2120 * Assuming a stable TSC across physical CPUS, and a stable TSC
2121 * across virtual CPUs, the following condition is possible.
2122 * Each numbered line represents an event visible to both
2123 * CPUs at the next numbered event.
2125 * "timespecX" represents host monotonic time. "tscX" represents
2128 * VCPU0 on CPU0 | VCPU1 on CPU1
2130 * 1. read timespec0,tsc0
2131 * 2. | timespec1 = timespec0 + N
2133 * 3. transition to guest | transition to guest
2134 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2135 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2136 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2138 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2141 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2143 * - 0 < N - M => M < N
2145 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2146 * always the case (the difference between two distinct xtime instances
2147 * might be smaller then the difference between corresponding TSC reads,
2148 * when updating guest vcpus pvclock areas).
2150 * To avoid that problem, do not allow visibility of distinct
2151 * system_timestamp/tsc_timestamp values simultaneously: use a master
2152 * copy of host monotonic time values. Update that master copy
2155 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2159 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2161 #ifdef CONFIG_X86_64
2162 struct kvm_arch *ka = &kvm->arch;
2164 bool host_tsc_clocksource, vcpus_matched;
2166 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2167 atomic_read(&kvm->online_vcpus));
2170 * If the host uses TSC clock, then passthrough TSC as stable
2173 host_tsc_clocksource = kvm_get_time_and_clockread(
2174 &ka->master_kernel_ns,
2175 &ka->master_cycle_now);
2177 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2178 && !ka->backwards_tsc_observed
2179 && !ka->boot_vcpu_runs_old_kvmclock;
2181 if (ka->use_master_clock)
2182 atomic_set(&kvm_guest_has_master_clock, 1);
2184 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2185 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2190 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2192 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2195 static void kvm_gen_update_masterclock(struct kvm *kvm)
2197 #ifdef CONFIG_X86_64
2199 struct kvm_vcpu *vcpu;
2200 struct kvm_arch *ka = &kvm->arch;
2202 spin_lock(&ka->pvclock_gtod_sync_lock);
2203 kvm_make_mclock_inprogress_request(kvm);
2204 /* no guest entries from this point */
2205 pvclock_update_vm_gtod_copy(kvm);
2207 kvm_for_each_vcpu(i, vcpu, kvm)
2208 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2210 /* guest entries allowed */
2211 kvm_for_each_vcpu(i, vcpu, kvm)
2212 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2214 spin_unlock(&ka->pvclock_gtod_sync_lock);
2218 u64 get_kvmclock_ns(struct kvm *kvm)
2220 struct kvm_arch *ka = &kvm->arch;
2221 struct pvclock_vcpu_time_info hv_clock;
2224 spin_lock(&ka->pvclock_gtod_sync_lock);
2225 if (!ka->use_master_clock) {
2226 spin_unlock(&ka->pvclock_gtod_sync_lock);
2227 return ktime_get_boottime_ns() + ka->kvmclock_offset;
2230 hv_clock.tsc_timestamp = ka->master_cycle_now;
2231 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2232 spin_unlock(&ka->pvclock_gtod_sync_lock);
2234 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2237 if (__this_cpu_read(cpu_tsc_khz)) {
2238 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2239 &hv_clock.tsc_shift,
2240 &hv_clock.tsc_to_system_mul);
2241 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2243 ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2250 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2252 struct kvm_vcpu_arch *vcpu = &v->arch;
2253 struct pvclock_vcpu_time_info guest_hv_clock;
2255 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2256 &guest_hv_clock, sizeof(guest_hv_clock))))
2259 /* This VCPU is paused, but it's legal for a guest to read another
2260 * VCPU's kvmclock, so we really have to follow the specification where
2261 * it says that version is odd if data is being modified, and even after
2264 * Version field updates must be kept separate. This is because
2265 * kvm_write_guest_cached might use a "rep movs" instruction, and
2266 * writes within a string instruction are weakly ordered. So there
2267 * are three writes overall.
2269 * As a small optimization, only write the version field in the first
2270 * and third write. The vcpu->pv_time cache is still valid, because the
2271 * version field is the first in the struct.
2273 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2275 if (guest_hv_clock.version & 1)
2276 ++guest_hv_clock.version; /* first time write, random junk */
2278 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2279 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2281 sizeof(vcpu->hv_clock.version));
2285 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2286 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2288 if (vcpu->pvclock_set_guest_stopped_request) {
2289 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2290 vcpu->pvclock_set_guest_stopped_request = false;
2293 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2295 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2297 sizeof(vcpu->hv_clock));
2301 vcpu->hv_clock.version++;
2302 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2304 sizeof(vcpu->hv_clock.version));
2307 static int kvm_guest_time_update(struct kvm_vcpu *v)
2309 unsigned long flags, tgt_tsc_khz;
2310 struct kvm_vcpu_arch *vcpu = &v->arch;
2311 struct kvm_arch *ka = &v->kvm->arch;
2313 u64 tsc_timestamp, host_tsc;
2315 bool use_master_clock;
2321 * If the host uses TSC clock, then passthrough TSC as stable
2324 spin_lock(&ka->pvclock_gtod_sync_lock);
2325 use_master_clock = ka->use_master_clock;
2326 if (use_master_clock) {
2327 host_tsc = ka->master_cycle_now;
2328 kernel_ns = ka->master_kernel_ns;
2330 spin_unlock(&ka->pvclock_gtod_sync_lock);
2332 /* Keep irq disabled to prevent changes to the clock */
2333 local_irq_save(flags);
2334 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2335 if (unlikely(tgt_tsc_khz == 0)) {
2336 local_irq_restore(flags);
2337 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2340 if (!use_master_clock) {
2342 kernel_ns = ktime_get_boottime_ns();
2345 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2348 * We may have to catch up the TSC to match elapsed wall clock
2349 * time for two reasons, even if kvmclock is used.
2350 * 1) CPU could have been running below the maximum TSC rate
2351 * 2) Broken TSC compensation resets the base at each VCPU
2352 * entry to avoid unknown leaps of TSC even when running
2353 * again on the same CPU. This may cause apparent elapsed
2354 * time to disappear, and the guest to stand still or run
2357 if (vcpu->tsc_catchup) {
2358 u64 tsc = compute_guest_tsc(v, kernel_ns);
2359 if (tsc > tsc_timestamp) {
2360 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2361 tsc_timestamp = tsc;
2365 local_irq_restore(flags);
2367 /* With all the info we got, fill in the values */
2369 if (kvm_has_tsc_control)
2370 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2372 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2373 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2374 &vcpu->hv_clock.tsc_shift,
2375 &vcpu->hv_clock.tsc_to_system_mul);
2376 vcpu->hw_tsc_khz = tgt_tsc_khz;
2379 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2380 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2381 vcpu->last_guest_tsc = tsc_timestamp;
2383 /* If the host uses TSC clocksource, then it is stable */
2385 if (use_master_clock)
2386 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2388 vcpu->hv_clock.flags = pvclock_flags;
2390 if (vcpu->pv_time_enabled)
2391 kvm_setup_pvclock_page(v);
2392 if (v == kvm_get_vcpu(v->kvm, 0))
2393 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2398 * kvmclock updates which are isolated to a given vcpu, such as
2399 * vcpu->cpu migration, should not allow system_timestamp from
2400 * the rest of the vcpus to remain static. Otherwise ntp frequency
2401 * correction applies to one vcpu's system_timestamp but not
2404 * So in those cases, request a kvmclock update for all vcpus.
2405 * We need to rate-limit these requests though, as they can
2406 * considerably slow guests that have a large number of vcpus.
2407 * The time for a remote vcpu to update its kvmclock is bound
2408 * by the delay we use to rate-limit the updates.
2411 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2413 static void kvmclock_update_fn(struct work_struct *work)
2416 struct delayed_work *dwork = to_delayed_work(work);
2417 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2418 kvmclock_update_work);
2419 struct kvm *kvm = container_of(ka, struct kvm, arch);
2420 struct kvm_vcpu *vcpu;
2422 kvm_for_each_vcpu(i, vcpu, kvm) {
2423 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2424 kvm_vcpu_kick(vcpu);
2428 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2430 struct kvm *kvm = v->kvm;
2432 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2433 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2434 KVMCLOCK_UPDATE_DELAY);
2437 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2439 static void kvmclock_sync_fn(struct work_struct *work)
2441 struct delayed_work *dwork = to_delayed_work(work);
2442 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2443 kvmclock_sync_work);
2444 struct kvm *kvm = container_of(ka, struct kvm, arch);
2446 if (!kvmclock_periodic_sync)
2449 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2450 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2451 KVMCLOCK_SYNC_PERIOD);
2455 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2457 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2459 /* McStatusWrEn enabled? */
2460 if (guest_cpuid_is_amd(vcpu))
2461 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2466 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2468 u64 mcg_cap = vcpu->arch.mcg_cap;
2469 unsigned bank_num = mcg_cap & 0xff;
2470 u32 msr = msr_info->index;
2471 u64 data = msr_info->data;
2474 case MSR_IA32_MCG_STATUS:
2475 vcpu->arch.mcg_status = data;
2477 case MSR_IA32_MCG_CTL:
2478 if (!(mcg_cap & MCG_CTL_P) &&
2479 (data || !msr_info->host_initiated))
2481 if (data != 0 && data != ~(u64)0)
2483 vcpu->arch.mcg_ctl = data;
2486 if (msr >= MSR_IA32_MC0_CTL &&
2487 msr < MSR_IA32_MCx_CTL(bank_num)) {
2488 u32 offset = msr - MSR_IA32_MC0_CTL;
2489 /* only 0 or all 1s can be written to IA32_MCi_CTL
2490 * some Linux kernels though clear bit 10 in bank 4 to
2491 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2492 * this to avoid an uncatched #GP in the guest
2494 if ((offset & 0x3) == 0 &&
2495 data != 0 && (data | (1 << 10)) != ~(u64)0)
2499 if (!msr_info->host_initiated &&
2500 (offset & 0x3) == 1 && data != 0) {
2501 if (!can_set_mci_status(vcpu))
2505 vcpu->arch.mce_banks[offset] = data;
2513 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2515 struct kvm *kvm = vcpu->kvm;
2516 int lm = is_long_mode(vcpu);
2517 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2518 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2519 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2520 : kvm->arch.xen_hvm_config.blob_size_32;
2521 u32 page_num = data & ~PAGE_MASK;
2522 u64 page_addr = data & PAGE_MASK;
2527 if (page_num >= blob_size)
2530 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2535 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2544 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2546 gpa_t gpa = data & ~0x3f;
2548 /* Bits 3:5 are reserved, Should be zero */
2552 vcpu->arch.apf.msr_val = data;
2554 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2555 kvm_clear_async_pf_completion_queue(vcpu);
2556 kvm_async_pf_hash_reset(vcpu);
2560 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2564 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2565 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2566 kvm_async_pf_wakeup_all(vcpu);
2570 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2572 vcpu->arch.pv_time_enabled = false;
2573 vcpu->arch.time = 0;
2576 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2578 ++vcpu->stat.tlb_flush;
2579 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2582 static void record_steal_time(struct kvm_vcpu *vcpu)
2584 struct kvm_host_map map;
2585 struct kvm_steal_time *st;
2587 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2590 /* -EAGAIN is returned in atomic context so we can just return. */
2591 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2592 &map, &vcpu->arch.st.cache, false))
2596 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2599 * Doing a TLB flush here, on the guest's behalf, can avoid
2602 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2603 st->preempted & KVM_VCPU_FLUSH_TLB);
2604 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2605 kvm_vcpu_flush_tlb(vcpu, false);
2607 vcpu->arch.st.steal.preempted = 0;
2609 if (st->version & 1)
2610 st->version += 1; /* first time write, random junk */
2616 st->steal += current->sched_info.run_delay -
2617 vcpu->arch.st.last_steal;
2618 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2624 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2627 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2630 u32 msr = msr_info->index;
2631 u64 data = msr_info->data;
2634 case MSR_AMD64_NB_CFG:
2635 case MSR_IA32_UCODE_WRITE:
2636 case MSR_VM_HSAVE_PA:
2637 case MSR_AMD64_PATCH_LOADER:
2638 case MSR_AMD64_BU_CFG2:
2639 case MSR_AMD64_DC_CFG:
2640 case MSR_F15H_EX_CFG:
2643 case MSR_IA32_UCODE_REV:
2644 if (msr_info->host_initiated)
2645 vcpu->arch.microcode_version = data;
2647 case MSR_IA32_ARCH_CAPABILITIES:
2648 if (!msr_info->host_initiated)
2650 vcpu->arch.arch_capabilities = data;
2653 return set_efer(vcpu, msr_info);
2655 data &= ~(u64)0x40; /* ignore flush filter disable */
2656 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2657 data &= ~(u64)0x8; /* ignore TLB cache disable */
2659 /* Handle McStatusWrEn */
2660 if (data == BIT_ULL(18)) {
2661 vcpu->arch.msr_hwcr = data;
2662 } else if (data != 0) {
2663 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2668 case MSR_FAM10H_MMIO_CONF_BASE:
2670 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2675 case MSR_IA32_DEBUGCTLMSR:
2677 /* We support the non-activated case already */
2679 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2680 /* Values other than LBR and BTF are vendor-specific,
2681 thus reserved and should throw a #GP */
2684 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2687 case 0x200 ... 0x2ff:
2688 return kvm_mtrr_set_msr(vcpu, msr, data);
2689 case MSR_IA32_APICBASE:
2690 return kvm_set_apic_base(vcpu, msr_info);
2691 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2692 return kvm_x2apic_msr_write(vcpu, msr, data);
2693 case MSR_IA32_TSCDEADLINE:
2694 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2696 case MSR_IA32_TSC_ADJUST:
2697 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2698 if (!msr_info->host_initiated) {
2699 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2700 adjust_tsc_offset_guest(vcpu, adj);
2702 vcpu->arch.ia32_tsc_adjust_msr = data;
2705 case MSR_IA32_MISC_ENABLE:
2706 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2707 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2708 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2710 vcpu->arch.ia32_misc_enable_msr = data;
2711 kvm_update_cpuid(vcpu);
2713 vcpu->arch.ia32_misc_enable_msr = data;
2716 case MSR_IA32_SMBASE:
2717 if (!msr_info->host_initiated)
2719 vcpu->arch.smbase = data;
2721 case MSR_IA32_POWER_CTL:
2722 vcpu->arch.msr_ia32_power_ctl = data;
2725 kvm_write_tsc(vcpu, msr_info);
2728 if (!msr_info->host_initiated &&
2729 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2732 * We do support PT if kvm_x86_ops->pt_supported(), but we do
2733 * not support IA32_XSS[bit 8]. Guests will have to use
2734 * RDMSR/WRMSR rather than XSAVES/XRSTORS to save/restore PT
2739 vcpu->arch.ia32_xss = data;
2742 if (!msr_info->host_initiated)
2744 vcpu->arch.smi_count = data;
2746 case MSR_KVM_WALL_CLOCK_NEW:
2747 case MSR_KVM_WALL_CLOCK:
2748 vcpu->kvm->arch.wall_clock = data;
2749 kvm_write_wall_clock(vcpu->kvm, data);
2751 case MSR_KVM_SYSTEM_TIME_NEW:
2752 case MSR_KVM_SYSTEM_TIME: {
2753 struct kvm_arch *ka = &vcpu->kvm->arch;
2755 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2756 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2758 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2759 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2761 ka->boot_vcpu_runs_old_kvmclock = tmp;
2764 vcpu->arch.time = data;
2765 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2767 /* we verify if the enable bit is set... */
2768 vcpu->arch.pv_time_enabled = false;
2772 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2773 &vcpu->arch.pv_time, data & ~1ULL,
2774 sizeof(struct pvclock_vcpu_time_info)))
2775 vcpu->arch.pv_time_enabled = true;
2779 case MSR_KVM_ASYNC_PF_EN:
2780 if (kvm_pv_enable_async_pf(vcpu, data))
2783 case MSR_KVM_STEAL_TIME:
2785 if (unlikely(!sched_info_on()))
2788 if (data & KVM_STEAL_RESERVED_MASK)
2791 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2792 data & KVM_STEAL_VALID_BITS,
2793 sizeof(struct kvm_steal_time)))
2796 vcpu->arch.st.msr_val = data;
2798 if (!(data & KVM_MSR_ENABLED))
2801 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2804 case MSR_KVM_PV_EOI_EN:
2805 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2809 case MSR_KVM_POLL_CONTROL:
2810 /* only enable bit supported */
2811 if (data & (-1ULL << 1))
2814 vcpu->arch.msr_kvm_poll_control = data;
2817 case MSR_IA32_MCG_CTL:
2818 case MSR_IA32_MCG_STATUS:
2819 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2820 return set_msr_mce(vcpu, msr_info);
2822 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2823 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2824 pr = true; /* fall through */
2825 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2826 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2827 if (kvm_pmu_is_valid_msr(vcpu, msr))
2828 return kvm_pmu_set_msr(vcpu, msr_info);
2830 if (pr || data != 0)
2831 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2832 "0x%x data 0x%llx\n", msr, data);
2834 case MSR_K7_CLK_CTL:
2836 * Ignore all writes to this no longer documented MSR.
2837 * Writes are only relevant for old K7 processors,
2838 * all pre-dating SVM, but a recommended workaround from
2839 * AMD for these chips. It is possible to specify the
2840 * affected processor models on the command line, hence
2841 * the need to ignore the workaround.
2844 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2845 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2846 case HV_X64_MSR_CRASH_CTL:
2847 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2848 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2849 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2850 case HV_X64_MSR_TSC_EMULATION_STATUS:
2851 return kvm_hv_set_msr_common(vcpu, msr, data,
2852 msr_info->host_initiated);
2853 case MSR_IA32_BBL_CR_CTL3:
2854 /* Drop writes to this legacy MSR -- see rdmsr
2855 * counterpart for further detail.
2857 if (report_ignored_msrs)
2858 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2861 case MSR_AMD64_OSVW_ID_LENGTH:
2862 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2864 vcpu->arch.osvw.length = data;
2866 case MSR_AMD64_OSVW_STATUS:
2867 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2869 vcpu->arch.osvw.status = data;
2871 case MSR_PLATFORM_INFO:
2872 if (!msr_info->host_initiated ||
2873 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2874 cpuid_fault_enabled(vcpu)))
2876 vcpu->arch.msr_platform_info = data;
2878 case MSR_MISC_FEATURES_ENABLES:
2879 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2880 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2881 !supports_cpuid_fault(vcpu)))
2883 vcpu->arch.msr_misc_features_enables = data;
2886 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2887 return xen_hvm_config(vcpu, data);
2888 if (kvm_pmu_is_valid_msr(vcpu, msr))
2889 return kvm_pmu_set_msr(vcpu, msr_info);
2891 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2895 if (report_ignored_msrs)
2897 "ignored wrmsr: 0x%x data 0x%llx\n",
2904 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2906 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2909 u64 mcg_cap = vcpu->arch.mcg_cap;
2910 unsigned bank_num = mcg_cap & 0xff;
2913 case MSR_IA32_P5_MC_ADDR:
2914 case MSR_IA32_P5_MC_TYPE:
2917 case MSR_IA32_MCG_CAP:
2918 data = vcpu->arch.mcg_cap;
2920 case MSR_IA32_MCG_CTL:
2921 if (!(mcg_cap & MCG_CTL_P) && !host)
2923 data = vcpu->arch.mcg_ctl;
2925 case MSR_IA32_MCG_STATUS:
2926 data = vcpu->arch.mcg_status;
2929 if (msr >= MSR_IA32_MC0_CTL &&
2930 msr < MSR_IA32_MCx_CTL(bank_num)) {
2931 u32 offset = msr - MSR_IA32_MC0_CTL;
2932 data = vcpu->arch.mce_banks[offset];
2941 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2943 switch (msr_info->index) {
2944 case MSR_IA32_PLATFORM_ID:
2945 case MSR_IA32_EBL_CR_POWERON:
2946 case MSR_IA32_DEBUGCTLMSR:
2947 case MSR_IA32_LASTBRANCHFROMIP:
2948 case MSR_IA32_LASTBRANCHTOIP:
2949 case MSR_IA32_LASTINTFROMIP:
2950 case MSR_IA32_LASTINTTOIP:
2952 case MSR_K8_TSEG_ADDR:
2953 case MSR_K8_TSEG_MASK:
2954 case MSR_VM_HSAVE_PA:
2955 case MSR_K8_INT_PENDING_MSG:
2956 case MSR_AMD64_NB_CFG:
2957 case MSR_FAM10H_MMIO_CONF_BASE:
2958 case MSR_AMD64_BU_CFG2:
2959 case MSR_IA32_PERF_CTL:
2960 case MSR_AMD64_DC_CFG:
2961 case MSR_F15H_EX_CFG:
2964 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2965 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2966 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2967 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2968 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2969 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2970 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2973 case MSR_IA32_UCODE_REV:
2974 msr_info->data = vcpu->arch.microcode_version;
2976 case MSR_IA32_ARCH_CAPABILITIES:
2977 if (!msr_info->host_initiated &&
2978 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2980 msr_info->data = vcpu->arch.arch_capabilities;
2982 case MSR_IA32_POWER_CTL:
2983 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2986 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2989 case 0x200 ... 0x2ff:
2990 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2991 case 0xcd: /* fsb frequency */
2995 * MSR_EBC_FREQUENCY_ID
2996 * Conservative value valid for even the basic CPU models.
2997 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2998 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2999 * and 266MHz for model 3, or 4. Set Core Clock
3000 * Frequency to System Bus Frequency Ratio to 1 (bits
3001 * 31:24) even though these are only valid for CPU
3002 * models > 2, however guests may end up dividing or
3003 * multiplying by zero otherwise.
3005 case MSR_EBC_FREQUENCY_ID:
3006 msr_info->data = 1 << 24;
3008 case MSR_IA32_APICBASE:
3009 msr_info->data = kvm_get_apic_base(vcpu);
3011 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
3012 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3014 case MSR_IA32_TSCDEADLINE:
3015 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3017 case MSR_IA32_TSC_ADJUST:
3018 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3020 case MSR_IA32_MISC_ENABLE:
3021 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3023 case MSR_IA32_SMBASE:
3024 if (!msr_info->host_initiated)
3026 msr_info->data = vcpu->arch.smbase;
3029 msr_info->data = vcpu->arch.smi_count;
3031 case MSR_IA32_PERF_STATUS:
3032 /* TSC increment by tick */
3033 msr_info->data = 1000ULL;
3034 /* CPU multiplier */
3035 msr_info->data |= (((uint64_t)4ULL) << 40);
3038 msr_info->data = vcpu->arch.efer;
3040 case MSR_KVM_WALL_CLOCK:
3041 case MSR_KVM_WALL_CLOCK_NEW:
3042 msr_info->data = vcpu->kvm->arch.wall_clock;
3044 case MSR_KVM_SYSTEM_TIME:
3045 case MSR_KVM_SYSTEM_TIME_NEW:
3046 msr_info->data = vcpu->arch.time;
3048 case MSR_KVM_ASYNC_PF_EN:
3049 msr_info->data = vcpu->arch.apf.msr_val;
3051 case MSR_KVM_STEAL_TIME:
3052 msr_info->data = vcpu->arch.st.msr_val;
3054 case MSR_KVM_PV_EOI_EN:
3055 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3057 case MSR_KVM_POLL_CONTROL:
3058 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3060 case MSR_IA32_P5_MC_ADDR:
3061 case MSR_IA32_P5_MC_TYPE:
3062 case MSR_IA32_MCG_CAP:
3063 case MSR_IA32_MCG_CTL:
3064 case MSR_IA32_MCG_STATUS:
3065 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3066 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3067 msr_info->host_initiated);
3069 if (!msr_info->host_initiated &&
3070 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3072 msr_info->data = vcpu->arch.ia32_xss;
3074 case MSR_K7_CLK_CTL:
3076 * Provide expected ramp-up count for K7. All other
3077 * are set to zero, indicating minimum divisors for
3080 * This prevents guest kernels on AMD host with CPU
3081 * type 6, model 8 and higher from exploding due to
3082 * the rdmsr failing.
3084 msr_info->data = 0x20000000;
3086 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3087 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3088 case HV_X64_MSR_CRASH_CTL:
3089 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3090 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3091 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3092 case HV_X64_MSR_TSC_EMULATION_STATUS:
3093 return kvm_hv_get_msr_common(vcpu,
3094 msr_info->index, &msr_info->data,
3095 msr_info->host_initiated);
3097 case MSR_IA32_BBL_CR_CTL3:
3098 /* This legacy MSR exists but isn't fully documented in current
3099 * silicon. It is however accessed by winxp in very narrow
3100 * scenarios where it sets bit #19, itself documented as
3101 * a "reserved" bit. Best effort attempt to source coherent
3102 * read data here should the balance of the register be
3103 * interpreted by the guest:
3105 * L2 cache control register 3: 64GB range, 256KB size,
3106 * enabled, latency 0x1, configured
3108 msr_info->data = 0xbe702111;
3110 case MSR_AMD64_OSVW_ID_LENGTH:
3111 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3113 msr_info->data = vcpu->arch.osvw.length;
3115 case MSR_AMD64_OSVW_STATUS:
3116 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3118 msr_info->data = vcpu->arch.osvw.status;
3120 case MSR_PLATFORM_INFO:
3121 if (!msr_info->host_initiated &&
3122 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3124 msr_info->data = vcpu->arch.msr_platform_info;
3126 case MSR_MISC_FEATURES_ENABLES:
3127 msr_info->data = vcpu->arch.msr_misc_features_enables;
3130 msr_info->data = vcpu->arch.msr_hwcr;
3133 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3134 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3136 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3140 if (report_ignored_msrs)
3141 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3149 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3152 * Read or write a bunch of msrs. All parameters are kernel addresses.
3154 * @return number of msrs set successfully.
3156 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3157 struct kvm_msr_entry *entries,
3158 int (*do_msr)(struct kvm_vcpu *vcpu,
3159 unsigned index, u64 *data))
3163 for (i = 0; i < msrs->nmsrs; ++i)
3164 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3171 * Read or write a bunch of msrs. Parameters are user addresses.
3173 * @return number of msrs set successfully.
3175 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3176 int (*do_msr)(struct kvm_vcpu *vcpu,
3177 unsigned index, u64 *data),
3180 struct kvm_msrs msrs;
3181 struct kvm_msr_entry *entries;
3186 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3190 if (msrs.nmsrs >= MAX_IO_MSRS)
3193 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3194 entries = memdup_user(user_msrs->entries, size);
3195 if (IS_ERR(entries)) {
3196 r = PTR_ERR(entries);
3200 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3205 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3216 static inline bool kvm_can_mwait_in_guest(void)
3218 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3219 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3220 boot_cpu_has(X86_FEATURE_ARAT);
3223 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3228 case KVM_CAP_IRQCHIP:
3230 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3231 case KVM_CAP_SET_TSS_ADDR:
3232 case KVM_CAP_EXT_CPUID:
3233 case KVM_CAP_EXT_EMUL_CPUID:
3234 case KVM_CAP_CLOCKSOURCE:
3236 case KVM_CAP_NOP_IO_DELAY:
3237 case KVM_CAP_MP_STATE:
3238 case KVM_CAP_SYNC_MMU:
3239 case KVM_CAP_USER_NMI:
3240 case KVM_CAP_REINJECT_CONTROL:
3241 case KVM_CAP_IRQ_INJECT_STATUS:
3242 case KVM_CAP_IOEVENTFD:
3243 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3245 case KVM_CAP_PIT_STATE2:
3246 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3247 case KVM_CAP_XEN_HVM:
3248 case KVM_CAP_VCPU_EVENTS:
3249 case KVM_CAP_HYPERV:
3250 case KVM_CAP_HYPERV_VAPIC:
3251 case KVM_CAP_HYPERV_SPIN:
3252 case KVM_CAP_HYPERV_SYNIC:
3253 case KVM_CAP_HYPERV_SYNIC2:
3254 case KVM_CAP_HYPERV_VP_INDEX:
3255 case KVM_CAP_HYPERV_EVENTFD:
3256 case KVM_CAP_HYPERV_TLBFLUSH:
3257 case KVM_CAP_HYPERV_SEND_IPI:
3258 case KVM_CAP_HYPERV_CPUID:
3259 case KVM_CAP_PCI_SEGMENT:
3260 case KVM_CAP_DEBUGREGS:
3261 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3263 case KVM_CAP_ASYNC_PF:
3264 case KVM_CAP_GET_TSC_KHZ:
3265 case KVM_CAP_KVMCLOCK_CTRL:
3266 case KVM_CAP_READONLY_MEM:
3267 case KVM_CAP_HYPERV_TIME:
3268 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3269 case KVM_CAP_TSC_DEADLINE_TIMER:
3270 case KVM_CAP_DISABLE_QUIRKS:
3271 case KVM_CAP_SET_BOOT_CPU_ID:
3272 case KVM_CAP_SPLIT_IRQCHIP:
3273 case KVM_CAP_IMMEDIATE_EXIT:
3274 case KVM_CAP_PMU_EVENT_FILTER:
3275 case KVM_CAP_GET_MSR_FEATURES:
3276 case KVM_CAP_MSR_PLATFORM_INFO:
3277 case KVM_CAP_EXCEPTION_PAYLOAD:
3280 case KVM_CAP_SYNC_REGS:
3281 r = KVM_SYNC_X86_VALID_FIELDS;
3283 case KVM_CAP_ADJUST_CLOCK:
3284 r = KVM_CLOCK_TSC_STABLE;
3286 case KVM_CAP_X86_DISABLE_EXITS:
3287 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3288 KVM_X86_DISABLE_EXITS_CSTATE;
3289 if(kvm_can_mwait_in_guest())
3290 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3292 case KVM_CAP_X86_SMM:
3293 /* SMBASE is usually relocated above 1M on modern chipsets,
3294 * and SMM handlers might indeed rely on 4G segment limits,
3295 * so do not report SMM to be available if real mode is
3296 * emulated via vm86 mode. Still, do not go to great lengths
3297 * to avoid userspace's usage of the feature, because it is a
3298 * fringe case that is not enabled except via specific settings
3299 * of the module parameters.
3301 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3304 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3306 case KVM_CAP_NR_VCPUS:
3307 r = KVM_SOFT_MAX_VCPUS;
3309 case KVM_CAP_MAX_VCPUS:
3312 case KVM_CAP_MAX_VCPU_ID:
3313 r = KVM_MAX_VCPU_ID;
3315 case KVM_CAP_PV_MMU: /* obsolete */
3319 r = KVM_MAX_MCE_BANKS;
3322 r = boot_cpu_has(X86_FEATURE_XSAVE);
3324 case KVM_CAP_TSC_CONTROL:
3325 r = kvm_has_tsc_control;
3327 case KVM_CAP_X2APIC_API:
3328 r = KVM_X2APIC_API_VALID_FLAGS;
3330 case KVM_CAP_NESTED_STATE:
3331 r = kvm_x86_ops->get_nested_state ?
3332 kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3334 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3335 r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3337 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3338 r = kvm_x86_ops->nested_enable_evmcs != NULL;
3347 long kvm_arch_dev_ioctl(struct file *filp,
3348 unsigned int ioctl, unsigned long arg)
3350 void __user *argp = (void __user *)arg;
3354 case KVM_GET_MSR_INDEX_LIST: {
3355 struct kvm_msr_list __user *user_msr_list = argp;
3356 struct kvm_msr_list msr_list;
3360 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3363 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3364 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3367 if (n < msr_list.nmsrs)
3370 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3371 num_msrs_to_save * sizeof(u32)))
3373 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3375 num_emulated_msrs * sizeof(u32)))
3380 case KVM_GET_SUPPORTED_CPUID:
3381 case KVM_GET_EMULATED_CPUID: {
3382 struct kvm_cpuid2 __user *cpuid_arg = argp;
3383 struct kvm_cpuid2 cpuid;
3386 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3389 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3395 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3400 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3402 if (copy_to_user(argp, &kvm_mce_cap_supported,
3403 sizeof(kvm_mce_cap_supported)))
3407 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3408 struct kvm_msr_list __user *user_msr_list = argp;
3409 struct kvm_msr_list msr_list;
3413 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3416 msr_list.nmsrs = num_msr_based_features;
3417 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3420 if (n < msr_list.nmsrs)
3423 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3424 num_msr_based_features * sizeof(u32)))
3430 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3440 static void wbinvd_ipi(void *garbage)
3445 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3447 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3450 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3452 /* Address WBINVD may be executed by guest */
3453 if (need_emulate_wbinvd(vcpu)) {
3454 if (kvm_x86_ops->has_wbinvd_exit())
3455 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3456 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3457 smp_call_function_single(vcpu->cpu,
3458 wbinvd_ipi, NULL, 1);
3461 kvm_x86_ops->vcpu_load(vcpu, cpu);
3463 fpregs_assert_state_consistent();
3464 if (test_thread_flag(TIF_NEED_FPU_LOAD))
3465 switch_fpu_return();
3467 /* Apply any externally detected TSC adjustments (due to suspend) */
3468 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3469 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3470 vcpu->arch.tsc_offset_adjustment = 0;
3471 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3474 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3475 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3476 rdtsc() - vcpu->arch.last_host_tsc;
3478 mark_tsc_unstable("KVM discovered backwards TSC");
3480 if (kvm_check_tsc_unstable()) {
3481 u64 offset = kvm_compute_tsc_offset(vcpu,
3482 vcpu->arch.last_guest_tsc);
3483 kvm_vcpu_write_tsc_offset(vcpu, offset);
3484 vcpu->arch.tsc_catchup = 1;
3487 if (kvm_lapic_hv_timer_in_use(vcpu))
3488 kvm_lapic_restart_hv_timer(vcpu);
3491 * On a host with synchronized TSC, there is no need to update
3492 * kvmclock on vcpu->cpu migration
3494 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3495 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3496 if (vcpu->cpu != cpu)
3497 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3501 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3504 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3506 struct kvm_host_map map;
3507 struct kvm_steal_time *st;
3509 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3512 if (vcpu->arch.st.steal.preempted)
3515 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3516 &vcpu->arch.st.cache, true))
3520 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3522 st->preempted = vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3524 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3527 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3531 if (vcpu->preempted)
3532 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3535 * Disable page faults because we're in atomic context here.
3536 * kvm_write_guest_offset_cached() would call might_fault()
3537 * that relies on pagefault_disable() to tell if there's a
3538 * bug. NOTE: the write to guest memory may not go through if
3539 * during postcopy live migration or if there's heavy guest
3542 pagefault_disable();
3544 * kvm_memslots() will be called by
3545 * kvm_write_guest_offset_cached() so take the srcu lock.
3547 idx = srcu_read_lock(&vcpu->kvm->srcu);
3548 kvm_steal_time_set_preempted(vcpu);
3549 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3551 kvm_x86_ops->vcpu_put(vcpu);
3552 vcpu->arch.last_host_tsc = rdtsc();
3554 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3555 * on every vmexit, but if not, we might have a stale dr6 from the
3556 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3561 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3562 struct kvm_lapic_state *s)
3564 if (vcpu->arch.apicv_active)
3565 kvm_x86_ops->sync_pir_to_irr(vcpu);
3567 return kvm_apic_get_state(vcpu, s);
3570 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3571 struct kvm_lapic_state *s)
3575 r = kvm_apic_set_state(vcpu, s);
3578 update_cr8_intercept(vcpu);
3583 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3585 return (!lapic_in_kernel(vcpu) ||
3586 kvm_apic_accept_pic_intr(vcpu));
3590 * if userspace requested an interrupt window, check that the
3591 * interrupt window is open.
3593 * No need to exit to userspace if we already have an interrupt queued.
3595 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3597 return kvm_arch_interrupt_allowed(vcpu) &&
3598 !kvm_cpu_has_interrupt(vcpu) &&
3599 !kvm_event_needs_reinjection(vcpu) &&
3600 kvm_cpu_accept_dm_intr(vcpu);
3603 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3604 struct kvm_interrupt *irq)
3606 if (irq->irq >= KVM_NR_INTERRUPTS)
3609 if (!irqchip_in_kernel(vcpu->kvm)) {
3610 kvm_queue_interrupt(vcpu, irq->irq, false);
3611 kvm_make_request(KVM_REQ_EVENT, vcpu);
3616 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3617 * fail for in-kernel 8259.
3619 if (pic_in_kernel(vcpu->kvm))
3622 if (vcpu->arch.pending_external_vector != -1)
3625 vcpu->arch.pending_external_vector = irq->irq;
3626 kvm_make_request(KVM_REQ_EVENT, vcpu);
3630 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3632 kvm_inject_nmi(vcpu);
3637 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3639 kvm_make_request(KVM_REQ_SMI, vcpu);
3644 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3645 struct kvm_tpr_access_ctl *tac)
3649 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3653 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3657 unsigned bank_num = mcg_cap & 0xff, bank;
3660 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3662 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3665 vcpu->arch.mcg_cap = mcg_cap;
3666 /* Init IA32_MCG_CTL to all 1s */
3667 if (mcg_cap & MCG_CTL_P)
3668 vcpu->arch.mcg_ctl = ~(u64)0;
3669 /* Init IA32_MCi_CTL to all 1s */
3670 for (bank = 0; bank < bank_num; bank++)
3671 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3673 kvm_x86_ops->setup_mce(vcpu);
3678 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3679 struct kvm_x86_mce *mce)
3681 u64 mcg_cap = vcpu->arch.mcg_cap;
3682 unsigned bank_num = mcg_cap & 0xff;
3683 u64 *banks = vcpu->arch.mce_banks;
3685 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3688 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3689 * reporting is disabled
3691 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3692 vcpu->arch.mcg_ctl != ~(u64)0)
3694 banks += 4 * mce->bank;
3696 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3697 * reporting is disabled for the bank
3699 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3701 if (mce->status & MCI_STATUS_UC) {
3702 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3703 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3704 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3707 if (banks[1] & MCI_STATUS_VAL)
3708 mce->status |= MCI_STATUS_OVER;
3709 banks[2] = mce->addr;
3710 banks[3] = mce->misc;
3711 vcpu->arch.mcg_status = mce->mcg_status;
3712 banks[1] = mce->status;
3713 kvm_queue_exception(vcpu, MC_VECTOR);
3714 } else if (!(banks[1] & MCI_STATUS_VAL)
3715 || !(banks[1] & MCI_STATUS_UC)) {
3716 if (banks[1] & MCI_STATUS_VAL)
3717 mce->status |= MCI_STATUS_OVER;
3718 banks[2] = mce->addr;
3719 banks[3] = mce->misc;
3720 banks[1] = mce->status;
3722 banks[1] |= MCI_STATUS_OVER;
3726 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3727 struct kvm_vcpu_events *events)
3732 * The API doesn't provide the instruction length for software
3733 * exceptions, so don't report them. As long as the guest RIP
3734 * isn't advanced, we should expect to encounter the exception
3737 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3738 events->exception.injected = 0;
3739 events->exception.pending = 0;
3741 events->exception.injected = vcpu->arch.exception.injected;
3742 events->exception.pending = vcpu->arch.exception.pending;
3744 * For ABI compatibility, deliberately conflate
3745 * pending and injected exceptions when
3746 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3748 if (!vcpu->kvm->arch.exception_payload_enabled)
3749 events->exception.injected |=
3750 vcpu->arch.exception.pending;
3752 events->exception.nr = vcpu->arch.exception.nr;
3753 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3754 events->exception.error_code = vcpu->arch.exception.error_code;
3755 events->exception_has_payload = vcpu->arch.exception.has_payload;
3756 events->exception_payload = vcpu->arch.exception.payload;
3758 events->interrupt.injected =
3759 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3760 events->interrupt.nr = vcpu->arch.interrupt.nr;
3761 events->interrupt.soft = 0;
3762 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3764 events->nmi.injected = vcpu->arch.nmi_injected;
3765 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3766 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3767 events->nmi.pad = 0;
3769 events->sipi_vector = 0; /* never valid when reporting to user space */
3771 events->smi.smm = is_smm(vcpu);
3772 events->smi.pending = vcpu->arch.smi_pending;
3773 events->smi.smm_inside_nmi =
3774 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3775 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3777 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3778 | KVM_VCPUEVENT_VALID_SHADOW
3779 | KVM_VCPUEVENT_VALID_SMM);
3780 if (vcpu->kvm->arch.exception_payload_enabled)
3781 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3783 memset(&events->reserved, 0, sizeof(events->reserved));
3786 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3788 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3789 struct kvm_vcpu_events *events)
3791 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3792 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3793 | KVM_VCPUEVENT_VALID_SHADOW
3794 | KVM_VCPUEVENT_VALID_SMM
3795 | KVM_VCPUEVENT_VALID_PAYLOAD))
3798 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3799 if (!vcpu->kvm->arch.exception_payload_enabled)
3801 if (events->exception.pending)
3802 events->exception.injected = 0;
3804 events->exception_has_payload = 0;
3806 events->exception.pending = 0;
3807 events->exception_has_payload = 0;
3810 if ((events->exception.injected || events->exception.pending) &&
3811 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3814 /* INITs are latched while in SMM */
3815 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3816 (events->smi.smm || events->smi.pending) &&
3817 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3821 vcpu->arch.exception.injected = events->exception.injected;
3822 vcpu->arch.exception.pending = events->exception.pending;
3823 vcpu->arch.exception.nr = events->exception.nr;
3824 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3825 vcpu->arch.exception.error_code = events->exception.error_code;
3826 vcpu->arch.exception.has_payload = events->exception_has_payload;
3827 vcpu->arch.exception.payload = events->exception_payload;
3829 vcpu->arch.interrupt.injected = events->interrupt.injected;
3830 vcpu->arch.interrupt.nr = events->interrupt.nr;
3831 vcpu->arch.interrupt.soft = events->interrupt.soft;
3832 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3833 kvm_x86_ops->set_interrupt_shadow(vcpu,
3834 events->interrupt.shadow);
3836 vcpu->arch.nmi_injected = events->nmi.injected;
3837 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3838 vcpu->arch.nmi_pending = events->nmi.pending;
3839 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3841 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3842 lapic_in_kernel(vcpu))
3843 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3845 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3846 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3847 if (events->smi.smm)
3848 vcpu->arch.hflags |= HF_SMM_MASK;
3850 vcpu->arch.hflags &= ~HF_SMM_MASK;
3851 kvm_smm_changed(vcpu);
3854 vcpu->arch.smi_pending = events->smi.pending;
3856 if (events->smi.smm) {
3857 if (events->smi.smm_inside_nmi)
3858 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3860 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3863 if (lapic_in_kernel(vcpu)) {
3864 if (events->smi.latched_init)
3865 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3867 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3871 kvm_make_request(KVM_REQ_EVENT, vcpu);
3876 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3877 struct kvm_debugregs *dbgregs)
3881 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3882 kvm_get_dr(vcpu, 6, &val);
3884 dbgregs->dr7 = vcpu->arch.dr7;
3886 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3889 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3890 struct kvm_debugregs *dbgregs)
3895 if (dbgregs->dr6 & ~0xffffffffull)
3897 if (dbgregs->dr7 & ~0xffffffffull)
3900 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3901 kvm_update_dr0123(vcpu);
3902 vcpu->arch.dr6 = dbgregs->dr6;
3903 kvm_update_dr6(vcpu);
3904 vcpu->arch.dr7 = dbgregs->dr7;
3905 kvm_update_dr7(vcpu);
3910 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3912 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3914 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3915 u64 xstate_bv = xsave->header.xfeatures;
3919 * Copy legacy XSAVE area, to avoid complications with CPUID
3920 * leaves 0 and 1 in the loop below.
3922 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3925 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3926 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3929 * Copy each region from the possibly compacted offset to the
3930 * non-compacted offset.
3932 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3934 u64 xfeature_mask = valid & -valid;
3935 int xfeature_nr = fls64(xfeature_mask) - 1;
3936 void *src = get_xsave_addr(xsave, xfeature_nr);
3939 u32 size, offset, ecx, edx;
3940 cpuid_count(XSTATE_CPUID, xfeature_nr,
3941 &size, &offset, &ecx, &edx);
3942 if (xfeature_nr == XFEATURE_PKRU)
3943 memcpy(dest + offset, &vcpu->arch.pkru,
3944 sizeof(vcpu->arch.pkru));
3946 memcpy(dest + offset, src, size);
3950 valid -= xfeature_mask;
3954 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3956 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3957 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3961 * Copy legacy XSAVE area, to avoid complications with CPUID
3962 * leaves 0 and 1 in the loop below.
3964 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3966 /* Set XSTATE_BV and possibly XCOMP_BV. */
3967 xsave->header.xfeatures = xstate_bv;
3968 if (boot_cpu_has(X86_FEATURE_XSAVES))
3969 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3972 * Copy each region from the non-compacted offset to the
3973 * possibly compacted offset.
3975 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3977 u64 xfeature_mask = valid & -valid;
3978 int xfeature_nr = fls64(xfeature_mask) - 1;
3979 void *dest = get_xsave_addr(xsave, xfeature_nr);
3982 u32 size, offset, ecx, edx;
3983 cpuid_count(XSTATE_CPUID, xfeature_nr,
3984 &size, &offset, &ecx, &edx);
3985 if (xfeature_nr == XFEATURE_PKRU)
3986 memcpy(&vcpu->arch.pkru, src + offset,
3987 sizeof(vcpu->arch.pkru));
3989 memcpy(dest, src + offset, size);
3992 valid -= xfeature_mask;
3996 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3997 struct kvm_xsave *guest_xsave)
3999 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4000 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4001 fill_xsave((u8 *) guest_xsave->region, vcpu);
4003 memcpy(guest_xsave->region,
4004 &vcpu->arch.guest_fpu->state.fxsave,
4005 sizeof(struct fxregs_state));
4006 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4007 XFEATURE_MASK_FPSSE;
4011 #define XSAVE_MXCSR_OFFSET 24
4013 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4014 struct kvm_xsave *guest_xsave)
4017 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4018 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4020 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4022 * Here we allow setting states that are not present in
4023 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4024 * with old userspace.
4026 if (xstate_bv & ~kvm_supported_xcr0() ||
4027 mxcsr & ~mxcsr_feature_mask)
4029 load_xsave(vcpu, (u8 *)guest_xsave->region);
4031 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4032 mxcsr & ~mxcsr_feature_mask)
4034 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4035 guest_xsave->region, sizeof(struct fxregs_state));
4040 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4041 struct kvm_xcrs *guest_xcrs)
4043 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4044 guest_xcrs->nr_xcrs = 0;
4048 guest_xcrs->nr_xcrs = 1;
4049 guest_xcrs->flags = 0;
4050 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4051 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4054 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4055 struct kvm_xcrs *guest_xcrs)
4059 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4062 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4065 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4066 /* Only support XCR0 currently */
4067 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4068 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4069 guest_xcrs->xcrs[i].value);
4078 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4079 * stopped by the hypervisor. This function will be called from the host only.
4080 * EINVAL is returned when the host attempts to set the flag for a guest that
4081 * does not support pv clocks.
4083 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4085 if (!vcpu->arch.pv_time_enabled)
4087 vcpu->arch.pvclock_set_guest_stopped_request = true;
4088 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4092 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4093 struct kvm_enable_cap *cap)
4096 uint16_t vmcs_version;
4097 void __user *user_ptr;
4103 case KVM_CAP_HYPERV_SYNIC2:
4108 case KVM_CAP_HYPERV_SYNIC:
4109 if (!irqchip_in_kernel(vcpu->kvm))
4111 return kvm_hv_activate_synic(vcpu, cap->cap ==
4112 KVM_CAP_HYPERV_SYNIC2);
4113 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4114 if (!kvm_x86_ops->nested_enable_evmcs)
4116 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4118 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4119 if (copy_to_user(user_ptr, &vmcs_version,
4120 sizeof(vmcs_version)))
4124 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4125 if (!kvm_x86_ops->enable_direct_tlbflush)
4128 return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4135 long kvm_arch_vcpu_ioctl(struct file *filp,
4136 unsigned int ioctl, unsigned long arg)
4138 struct kvm_vcpu *vcpu = filp->private_data;
4139 void __user *argp = (void __user *)arg;
4142 struct kvm_lapic_state *lapic;
4143 struct kvm_xsave *xsave;
4144 struct kvm_xcrs *xcrs;
4152 case KVM_GET_LAPIC: {
4154 if (!lapic_in_kernel(vcpu))
4156 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4157 GFP_KERNEL_ACCOUNT);
4162 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4166 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4171 case KVM_SET_LAPIC: {
4173 if (!lapic_in_kernel(vcpu))
4175 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4176 if (IS_ERR(u.lapic)) {
4177 r = PTR_ERR(u.lapic);
4181 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4184 case KVM_INTERRUPT: {
4185 struct kvm_interrupt irq;
4188 if (copy_from_user(&irq, argp, sizeof(irq)))
4190 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4194 r = kvm_vcpu_ioctl_nmi(vcpu);
4198 r = kvm_vcpu_ioctl_smi(vcpu);
4201 case KVM_SET_CPUID: {
4202 struct kvm_cpuid __user *cpuid_arg = argp;
4203 struct kvm_cpuid cpuid;
4206 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4208 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4211 case KVM_SET_CPUID2: {
4212 struct kvm_cpuid2 __user *cpuid_arg = argp;
4213 struct kvm_cpuid2 cpuid;
4216 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4218 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4219 cpuid_arg->entries);
4222 case KVM_GET_CPUID2: {
4223 struct kvm_cpuid2 __user *cpuid_arg = argp;
4224 struct kvm_cpuid2 cpuid;
4227 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4229 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4230 cpuid_arg->entries);
4234 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4239 case KVM_GET_MSRS: {
4240 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4241 r = msr_io(vcpu, argp, do_get_msr, 1);
4242 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4245 case KVM_SET_MSRS: {
4246 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4247 r = msr_io(vcpu, argp, do_set_msr, 0);
4248 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4251 case KVM_TPR_ACCESS_REPORTING: {
4252 struct kvm_tpr_access_ctl tac;
4255 if (copy_from_user(&tac, argp, sizeof(tac)))
4257 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4261 if (copy_to_user(argp, &tac, sizeof(tac)))
4266 case KVM_SET_VAPIC_ADDR: {
4267 struct kvm_vapic_addr va;
4271 if (!lapic_in_kernel(vcpu))
4274 if (copy_from_user(&va, argp, sizeof(va)))
4276 idx = srcu_read_lock(&vcpu->kvm->srcu);
4277 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4278 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4281 case KVM_X86_SETUP_MCE: {
4285 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4287 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4290 case KVM_X86_SET_MCE: {
4291 struct kvm_x86_mce mce;
4294 if (copy_from_user(&mce, argp, sizeof(mce)))
4296 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4299 case KVM_GET_VCPU_EVENTS: {
4300 struct kvm_vcpu_events events;
4302 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4305 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4310 case KVM_SET_VCPU_EVENTS: {
4311 struct kvm_vcpu_events events;
4314 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4317 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4320 case KVM_GET_DEBUGREGS: {
4321 struct kvm_debugregs dbgregs;
4323 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4326 if (copy_to_user(argp, &dbgregs,
4327 sizeof(struct kvm_debugregs)))
4332 case KVM_SET_DEBUGREGS: {
4333 struct kvm_debugregs dbgregs;
4336 if (copy_from_user(&dbgregs, argp,
4337 sizeof(struct kvm_debugregs)))
4340 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4343 case KVM_GET_XSAVE: {
4344 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4349 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4352 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4357 case KVM_SET_XSAVE: {
4358 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4359 if (IS_ERR(u.xsave)) {
4360 r = PTR_ERR(u.xsave);
4364 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4367 case KVM_GET_XCRS: {
4368 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4373 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4376 if (copy_to_user(argp, u.xcrs,
4377 sizeof(struct kvm_xcrs)))
4382 case KVM_SET_XCRS: {
4383 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4384 if (IS_ERR(u.xcrs)) {
4385 r = PTR_ERR(u.xcrs);
4389 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4392 case KVM_SET_TSC_KHZ: {
4396 user_tsc_khz = (u32)arg;
4398 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4401 if (user_tsc_khz == 0)
4402 user_tsc_khz = tsc_khz;
4404 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4409 case KVM_GET_TSC_KHZ: {
4410 r = vcpu->arch.virtual_tsc_khz;
4413 case KVM_KVMCLOCK_CTRL: {
4414 r = kvm_set_guest_paused(vcpu);
4417 case KVM_ENABLE_CAP: {
4418 struct kvm_enable_cap cap;
4421 if (copy_from_user(&cap, argp, sizeof(cap)))
4423 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4426 case KVM_GET_NESTED_STATE: {
4427 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4431 if (!kvm_x86_ops->get_nested_state)
4434 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4436 if (get_user(user_data_size, &user_kvm_nested_state->size))
4439 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4444 if (r > user_data_size) {
4445 if (put_user(r, &user_kvm_nested_state->size))
4455 case KVM_SET_NESTED_STATE: {
4456 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4457 struct kvm_nested_state kvm_state;
4461 if (!kvm_x86_ops->set_nested_state)
4465 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4469 if (kvm_state.size < sizeof(kvm_state))
4472 if (kvm_state.flags &
4473 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4474 | KVM_STATE_NESTED_EVMCS))
4477 /* nested_run_pending implies guest_mode. */
4478 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4479 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4482 idx = srcu_read_lock(&vcpu->kvm->srcu);
4483 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4484 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4487 case KVM_GET_SUPPORTED_HV_CPUID: {
4488 struct kvm_cpuid2 __user *cpuid_arg = argp;
4489 struct kvm_cpuid2 cpuid;
4492 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4495 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4496 cpuid_arg->entries);
4501 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4516 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4518 return VM_FAULT_SIGBUS;
4521 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4525 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4527 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4531 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4534 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4537 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4538 unsigned long kvm_nr_mmu_pages)
4540 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4543 mutex_lock(&kvm->slots_lock);
4545 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4546 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4548 mutex_unlock(&kvm->slots_lock);
4552 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4554 return kvm->arch.n_max_mmu_pages;
4557 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4559 struct kvm_pic *pic = kvm->arch.vpic;
4563 switch (chip->chip_id) {
4564 case KVM_IRQCHIP_PIC_MASTER:
4565 memcpy(&chip->chip.pic, &pic->pics[0],
4566 sizeof(struct kvm_pic_state));
4568 case KVM_IRQCHIP_PIC_SLAVE:
4569 memcpy(&chip->chip.pic, &pic->pics[1],
4570 sizeof(struct kvm_pic_state));
4572 case KVM_IRQCHIP_IOAPIC:
4573 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4582 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4584 struct kvm_pic *pic = kvm->arch.vpic;
4588 switch (chip->chip_id) {
4589 case KVM_IRQCHIP_PIC_MASTER:
4590 spin_lock(&pic->lock);
4591 memcpy(&pic->pics[0], &chip->chip.pic,
4592 sizeof(struct kvm_pic_state));
4593 spin_unlock(&pic->lock);
4595 case KVM_IRQCHIP_PIC_SLAVE:
4596 spin_lock(&pic->lock);
4597 memcpy(&pic->pics[1], &chip->chip.pic,
4598 sizeof(struct kvm_pic_state));
4599 spin_unlock(&pic->lock);
4601 case KVM_IRQCHIP_IOAPIC:
4602 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4608 kvm_pic_update_irq(pic);
4612 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4614 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4616 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4618 mutex_lock(&kps->lock);
4619 memcpy(ps, &kps->channels, sizeof(*ps));
4620 mutex_unlock(&kps->lock);
4624 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4627 struct kvm_pit *pit = kvm->arch.vpit;
4629 mutex_lock(&pit->pit_state.lock);
4630 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4631 for (i = 0; i < 3; i++)
4632 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4633 mutex_unlock(&pit->pit_state.lock);
4637 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4639 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4640 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4641 sizeof(ps->channels));
4642 ps->flags = kvm->arch.vpit->pit_state.flags;
4643 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4644 memset(&ps->reserved, 0, sizeof(ps->reserved));
4648 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4652 u32 prev_legacy, cur_legacy;
4653 struct kvm_pit *pit = kvm->arch.vpit;
4655 mutex_lock(&pit->pit_state.lock);
4656 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4657 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4658 if (!prev_legacy && cur_legacy)
4660 memcpy(&pit->pit_state.channels, &ps->channels,
4661 sizeof(pit->pit_state.channels));
4662 pit->pit_state.flags = ps->flags;
4663 for (i = 0; i < 3; i++)
4664 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4666 mutex_unlock(&pit->pit_state.lock);
4670 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4671 struct kvm_reinject_control *control)
4673 struct kvm_pit *pit = kvm->arch.vpit;
4678 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4679 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4680 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4682 mutex_lock(&pit->pit_state.lock);
4683 kvm_pit_set_reinject(pit, control->pit_reinject);
4684 mutex_unlock(&pit->pit_state.lock);
4690 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4691 * @kvm: kvm instance
4692 * @log: slot id and address to which we copy the log
4694 * Steps 1-4 below provide general overview of dirty page logging. See
4695 * kvm_get_dirty_log_protect() function description for additional details.
4697 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4698 * always flush the TLB (step 4) even if previous step failed and the dirty
4699 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4700 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4701 * writes will be marked dirty for next log read.
4703 * 1. Take a snapshot of the bit and clear it if needed.
4704 * 2. Write protect the corresponding page.
4705 * 3. Copy the snapshot to the userspace.
4706 * 4. Flush TLB's if needed.
4708 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4713 mutex_lock(&kvm->slots_lock);
4716 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4718 if (kvm_x86_ops->flush_log_dirty)
4719 kvm_x86_ops->flush_log_dirty(kvm);
4721 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4724 * All the TLBs can be flushed out of mmu lock, see the comments in
4725 * kvm_mmu_slot_remove_write_access().
4727 lockdep_assert_held(&kvm->slots_lock);
4729 kvm_flush_remote_tlbs(kvm);
4731 mutex_unlock(&kvm->slots_lock);
4735 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4740 mutex_lock(&kvm->slots_lock);
4743 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4745 if (kvm_x86_ops->flush_log_dirty)
4746 kvm_x86_ops->flush_log_dirty(kvm);
4748 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4751 * All the TLBs can be flushed out of mmu lock, see the comments in
4752 * kvm_mmu_slot_remove_write_access().
4754 lockdep_assert_held(&kvm->slots_lock);
4756 kvm_flush_remote_tlbs(kvm);
4758 mutex_unlock(&kvm->slots_lock);
4762 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4765 if (!irqchip_in_kernel(kvm))
4768 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4769 irq_event->irq, irq_event->level,
4774 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4775 struct kvm_enable_cap *cap)
4783 case KVM_CAP_DISABLE_QUIRKS:
4784 kvm->arch.disabled_quirks = cap->args[0];
4787 case KVM_CAP_SPLIT_IRQCHIP: {
4788 mutex_lock(&kvm->lock);
4790 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4791 goto split_irqchip_unlock;
4793 if (irqchip_in_kernel(kvm))
4794 goto split_irqchip_unlock;
4795 if (kvm->created_vcpus)
4796 goto split_irqchip_unlock;
4797 r = kvm_setup_empty_irq_routing(kvm);
4799 goto split_irqchip_unlock;
4800 /* Pairs with irqchip_in_kernel. */
4802 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4803 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4805 split_irqchip_unlock:
4806 mutex_unlock(&kvm->lock);
4809 case KVM_CAP_X2APIC_API:
4811 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4814 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4815 kvm->arch.x2apic_format = true;
4816 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4817 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4821 case KVM_CAP_X86_DISABLE_EXITS:
4823 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4826 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4827 kvm_can_mwait_in_guest())
4828 kvm->arch.mwait_in_guest = true;
4829 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4830 kvm->arch.hlt_in_guest = true;
4831 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4832 kvm->arch.pause_in_guest = true;
4833 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4834 kvm->arch.cstate_in_guest = true;
4837 case KVM_CAP_MSR_PLATFORM_INFO:
4838 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4841 case KVM_CAP_EXCEPTION_PAYLOAD:
4842 kvm->arch.exception_payload_enabled = cap->args[0];
4852 long kvm_arch_vm_ioctl(struct file *filp,
4853 unsigned int ioctl, unsigned long arg)
4855 struct kvm *kvm = filp->private_data;
4856 void __user *argp = (void __user *)arg;
4859 * This union makes it completely explicit to gcc-3.x
4860 * that these two variables' stack usage should be
4861 * combined, not added together.
4864 struct kvm_pit_state ps;
4865 struct kvm_pit_state2 ps2;
4866 struct kvm_pit_config pit_config;
4870 case KVM_SET_TSS_ADDR:
4871 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4873 case KVM_SET_IDENTITY_MAP_ADDR: {
4876 mutex_lock(&kvm->lock);
4878 if (kvm->created_vcpus)
4879 goto set_identity_unlock;
4881 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4882 goto set_identity_unlock;
4883 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4884 set_identity_unlock:
4885 mutex_unlock(&kvm->lock);
4888 case KVM_SET_NR_MMU_PAGES:
4889 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4891 case KVM_GET_NR_MMU_PAGES:
4892 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4894 case KVM_CREATE_IRQCHIP: {
4895 mutex_lock(&kvm->lock);
4898 if (irqchip_in_kernel(kvm))
4899 goto create_irqchip_unlock;
4902 if (kvm->created_vcpus)
4903 goto create_irqchip_unlock;
4905 r = kvm_pic_init(kvm);
4907 goto create_irqchip_unlock;
4909 r = kvm_ioapic_init(kvm);
4911 kvm_pic_destroy(kvm);
4912 goto create_irqchip_unlock;
4915 r = kvm_setup_default_irq_routing(kvm);
4917 kvm_ioapic_destroy(kvm);
4918 kvm_pic_destroy(kvm);
4919 goto create_irqchip_unlock;
4921 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4923 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4924 create_irqchip_unlock:
4925 mutex_unlock(&kvm->lock);
4928 case KVM_CREATE_PIT:
4929 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4931 case KVM_CREATE_PIT2:
4933 if (copy_from_user(&u.pit_config, argp,
4934 sizeof(struct kvm_pit_config)))
4937 mutex_lock(&kvm->lock);
4940 goto create_pit_unlock;
4942 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4946 mutex_unlock(&kvm->lock);
4948 case KVM_GET_IRQCHIP: {
4949 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4950 struct kvm_irqchip *chip;
4952 chip = memdup_user(argp, sizeof(*chip));
4959 if (!irqchip_kernel(kvm))
4960 goto get_irqchip_out;
4961 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4963 goto get_irqchip_out;
4965 if (copy_to_user(argp, chip, sizeof(*chip)))
4966 goto get_irqchip_out;
4972 case KVM_SET_IRQCHIP: {
4973 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4974 struct kvm_irqchip *chip;
4976 chip = memdup_user(argp, sizeof(*chip));
4983 if (!irqchip_kernel(kvm))
4984 goto set_irqchip_out;
4985 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4992 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4995 if (!kvm->arch.vpit)
4997 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5001 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5008 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5011 if (!kvm->arch.vpit)
5013 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5016 case KVM_GET_PIT2: {
5018 if (!kvm->arch.vpit)
5020 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5024 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5029 case KVM_SET_PIT2: {
5031 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5034 if (!kvm->arch.vpit)
5036 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5039 case KVM_REINJECT_CONTROL: {
5040 struct kvm_reinject_control control;
5042 if (copy_from_user(&control, argp, sizeof(control)))
5044 r = kvm_vm_ioctl_reinject(kvm, &control);
5047 case KVM_SET_BOOT_CPU_ID:
5049 mutex_lock(&kvm->lock);
5050 if (kvm->created_vcpus)
5053 kvm->arch.bsp_vcpu_id = arg;
5054 mutex_unlock(&kvm->lock);
5056 case KVM_XEN_HVM_CONFIG: {
5057 struct kvm_xen_hvm_config xhc;
5059 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5064 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5068 case KVM_SET_CLOCK: {
5069 struct kvm_clock_data user_ns;
5073 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5082 * TODO: userspace has to take care of races with VCPU_RUN, so
5083 * kvm_gen_update_masterclock() can be cut down to locked
5084 * pvclock_update_vm_gtod_copy().
5086 kvm_gen_update_masterclock(kvm);
5087 now_ns = get_kvmclock_ns(kvm);
5088 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5089 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5092 case KVM_GET_CLOCK: {
5093 struct kvm_clock_data user_ns;
5096 now_ns = get_kvmclock_ns(kvm);
5097 user_ns.clock = now_ns;
5098 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5099 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5102 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5107 case KVM_MEMORY_ENCRYPT_OP: {
5109 if (kvm_x86_ops->mem_enc_op)
5110 r = kvm_x86_ops->mem_enc_op(kvm, argp);
5113 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5114 struct kvm_enc_region region;
5117 if (copy_from_user(®ion, argp, sizeof(region)))
5121 if (kvm_x86_ops->mem_enc_reg_region)
5122 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
5125 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5126 struct kvm_enc_region region;
5129 if (copy_from_user(®ion, argp, sizeof(region)))
5133 if (kvm_x86_ops->mem_enc_unreg_region)
5134 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
5137 case KVM_HYPERV_EVENTFD: {
5138 struct kvm_hyperv_eventfd hvevfd;
5141 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5143 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5146 case KVM_SET_PMU_EVENT_FILTER:
5147 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5156 static void kvm_init_msr_list(void)
5158 struct x86_pmu_capability x86_pmu;
5162 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5163 "Please update the fixed PMCs in msrs_to_saved_all[]");
5165 perf_get_x86_pmu_capability(&x86_pmu);
5167 num_msrs_to_save = 0;
5168 num_emulated_msrs = 0;
5169 num_msr_based_features = 0;
5171 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5172 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5176 * Even MSRs that are valid in the host may not be exposed
5177 * to the guests in some cases.
5179 switch (msrs_to_save_all[i]) {
5180 case MSR_IA32_BNDCFGS:
5181 if (!kvm_mpx_supported())
5185 if (!kvm_x86_ops->rdtscp_supported())
5188 case MSR_IA32_RTIT_CTL:
5189 case MSR_IA32_RTIT_STATUS:
5190 if (!kvm_x86_ops->pt_supported())
5193 case MSR_IA32_RTIT_CR3_MATCH:
5194 if (!kvm_x86_ops->pt_supported() ||
5195 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5198 case MSR_IA32_RTIT_OUTPUT_BASE:
5199 case MSR_IA32_RTIT_OUTPUT_MASK:
5200 if (!kvm_x86_ops->pt_supported() ||
5201 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5202 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5205 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5206 if (!kvm_x86_ops->pt_supported() ||
5207 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5208 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5211 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5212 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5213 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5216 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5217 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5218 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5225 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5228 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5229 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs_all[i]))
5232 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5235 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5236 struct kvm_msr_entry msr;
5238 msr.index = msr_based_features_all[i];
5239 if (kvm_get_msr_feature(&msr))
5242 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5246 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5254 if (!(lapic_in_kernel(vcpu) &&
5255 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5256 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5267 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5274 if (!(lapic_in_kernel(vcpu) &&
5275 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5277 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5279 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5289 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5290 struct kvm_segment *var, int seg)
5292 kvm_x86_ops->set_segment(vcpu, var, seg);
5295 void kvm_get_segment(struct kvm_vcpu *vcpu,
5296 struct kvm_segment *var, int seg)
5298 kvm_x86_ops->get_segment(vcpu, var, seg);
5301 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5302 struct x86_exception *exception)
5306 BUG_ON(!mmu_is_nested(vcpu));
5308 /* NPT walks are always user-walks */
5309 access |= PFERR_USER_MASK;
5310 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5315 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5316 struct x86_exception *exception)
5318 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5319 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5322 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5323 struct x86_exception *exception)
5325 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5326 access |= PFERR_FETCH_MASK;
5327 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5330 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5331 struct x86_exception *exception)
5333 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5334 access |= PFERR_WRITE_MASK;
5335 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5338 /* uses this to access any guest's mapped memory without checking CPL */
5339 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5340 struct x86_exception *exception)
5342 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5345 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5346 struct kvm_vcpu *vcpu, u32 access,
5347 struct x86_exception *exception)
5350 int r = X86EMUL_CONTINUE;
5353 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5355 unsigned offset = addr & (PAGE_SIZE-1);
5356 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5359 if (gpa == UNMAPPED_GVA)
5360 return X86EMUL_PROPAGATE_FAULT;
5361 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5364 r = X86EMUL_IO_NEEDED;
5376 /* used for instruction fetching */
5377 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5378 gva_t addr, void *val, unsigned int bytes,
5379 struct x86_exception *exception)
5381 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5382 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5386 /* Inline kvm_read_guest_virt_helper for speed. */
5387 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5389 if (unlikely(gpa == UNMAPPED_GVA))
5390 return X86EMUL_PROPAGATE_FAULT;
5392 offset = addr & (PAGE_SIZE-1);
5393 if (WARN_ON(offset + bytes > PAGE_SIZE))
5394 bytes = (unsigned)PAGE_SIZE - offset;
5395 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5397 if (unlikely(ret < 0))
5398 return X86EMUL_IO_NEEDED;
5400 return X86EMUL_CONTINUE;
5403 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5404 gva_t addr, void *val, unsigned int bytes,
5405 struct x86_exception *exception)
5407 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5410 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5411 * is returned, but our callers are not ready for that and they blindly
5412 * call kvm_inject_page_fault. Ensure that they at least do not leak
5413 * uninitialized kernel stack memory into cr2 and error code.
5415 memset(exception, 0, sizeof(*exception));
5416 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5419 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5421 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5422 gva_t addr, void *val, unsigned int bytes,
5423 struct x86_exception *exception, bool system)
5425 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5428 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5429 access |= PFERR_USER_MASK;
5431 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5434 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5435 unsigned long addr, void *val, unsigned int bytes)
5437 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5438 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5440 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5443 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5444 struct kvm_vcpu *vcpu, u32 access,
5445 struct x86_exception *exception)
5448 int r = X86EMUL_CONTINUE;
5451 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5454 unsigned offset = addr & (PAGE_SIZE-1);
5455 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5458 if (gpa == UNMAPPED_GVA)
5459 return X86EMUL_PROPAGATE_FAULT;
5460 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5462 r = X86EMUL_IO_NEEDED;
5474 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5475 unsigned int bytes, struct x86_exception *exception,
5478 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5479 u32 access = PFERR_WRITE_MASK;
5481 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5482 access |= PFERR_USER_MASK;
5484 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5488 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5489 unsigned int bytes, struct x86_exception *exception)
5491 /* kvm_write_guest_virt_system can pull in tons of pages. */
5492 vcpu->arch.l1tf_flush_l1d = true;
5495 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5496 * is returned, but our callers are not ready for that and they blindly
5497 * call kvm_inject_page_fault. Ensure that they at least do not leak
5498 * uninitialized kernel stack memory into cr2 and error code.
5500 memset(exception, 0, sizeof(*exception));
5501 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5502 PFERR_WRITE_MASK, exception);
5504 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5506 int handle_ud(struct kvm_vcpu *vcpu)
5508 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5509 int emul_type = EMULTYPE_TRAP_UD;
5510 char sig[5]; /* ud2; .ascii "kvm" */
5511 struct x86_exception e;
5513 if (force_emulation_prefix &&
5514 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5515 sig, sizeof(sig), &e) == 0 &&
5516 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5517 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5518 emul_type = EMULTYPE_TRAP_UD_FORCED;
5521 return kvm_emulate_instruction(vcpu, emul_type);
5523 EXPORT_SYMBOL_GPL(handle_ud);
5525 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5526 gpa_t gpa, bool write)
5528 /* For APIC access vmexit */
5529 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5532 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5533 trace_vcpu_match_mmio(gva, gpa, write, true);
5540 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5541 gpa_t *gpa, struct x86_exception *exception,
5544 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5545 | (write ? PFERR_WRITE_MASK : 0);
5548 * currently PKRU is only applied to ept enabled guest so
5549 * there is no pkey in EPT page table for L1 guest or EPT
5550 * shadow page table for L2 guest.
5552 if (vcpu_match_mmio_gva(vcpu, gva)
5553 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5554 vcpu->arch.mmio_access, 0, access)) {
5555 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5556 (gva & (PAGE_SIZE - 1));
5557 trace_vcpu_match_mmio(gva, *gpa, write, false);
5561 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5563 if (*gpa == UNMAPPED_GVA)
5566 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5569 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5570 const void *val, int bytes)
5574 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5577 kvm_page_track_write(vcpu, gpa, val, bytes);
5581 struct read_write_emulator_ops {
5582 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5584 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5585 void *val, int bytes);
5586 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5587 int bytes, void *val);
5588 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5589 void *val, int bytes);
5593 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5595 if (vcpu->mmio_read_completed) {
5596 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5597 vcpu->mmio_fragments[0].gpa, val);
5598 vcpu->mmio_read_completed = 0;
5605 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5606 void *val, int bytes)
5608 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5611 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5612 void *val, int bytes)
5614 return emulator_write_phys(vcpu, gpa, val, bytes);
5617 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5619 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5620 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5623 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5624 void *val, int bytes)
5626 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5627 return X86EMUL_IO_NEEDED;
5630 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5631 void *val, int bytes)
5633 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5635 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5636 return X86EMUL_CONTINUE;
5639 static const struct read_write_emulator_ops read_emultor = {
5640 .read_write_prepare = read_prepare,
5641 .read_write_emulate = read_emulate,
5642 .read_write_mmio = vcpu_mmio_read,
5643 .read_write_exit_mmio = read_exit_mmio,
5646 static const struct read_write_emulator_ops write_emultor = {
5647 .read_write_emulate = write_emulate,
5648 .read_write_mmio = write_mmio,
5649 .read_write_exit_mmio = write_exit_mmio,
5653 static int emulator_read_write_onepage(unsigned long addr, void *val,
5655 struct x86_exception *exception,
5656 struct kvm_vcpu *vcpu,
5657 const struct read_write_emulator_ops *ops)
5661 bool write = ops->write;
5662 struct kvm_mmio_fragment *frag;
5663 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5666 * If the exit was due to a NPF we may already have a GPA.
5667 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5668 * Note, this cannot be used on string operations since string
5669 * operation using rep will only have the initial GPA from the NPF
5672 if (vcpu->arch.gpa_available &&
5673 emulator_can_use_gpa(ctxt) &&
5674 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5675 gpa = vcpu->arch.gpa_val;
5676 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5678 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5680 return X86EMUL_PROPAGATE_FAULT;
5683 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5684 return X86EMUL_CONTINUE;
5687 * Is this MMIO handled locally?
5689 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5690 if (handled == bytes)
5691 return X86EMUL_CONTINUE;
5697 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5698 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5702 return X86EMUL_CONTINUE;
5705 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5707 void *val, unsigned int bytes,
5708 struct x86_exception *exception,
5709 const struct read_write_emulator_ops *ops)
5711 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5715 if (ops->read_write_prepare &&
5716 ops->read_write_prepare(vcpu, val, bytes))
5717 return X86EMUL_CONTINUE;
5719 vcpu->mmio_nr_fragments = 0;
5721 /* Crossing a page boundary? */
5722 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5725 now = -addr & ~PAGE_MASK;
5726 rc = emulator_read_write_onepage(addr, val, now, exception,
5729 if (rc != X86EMUL_CONTINUE)
5732 if (ctxt->mode != X86EMUL_MODE_PROT64)
5738 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5740 if (rc != X86EMUL_CONTINUE)
5743 if (!vcpu->mmio_nr_fragments)
5746 gpa = vcpu->mmio_fragments[0].gpa;
5748 vcpu->mmio_needed = 1;
5749 vcpu->mmio_cur_fragment = 0;
5751 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5752 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5753 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5754 vcpu->run->mmio.phys_addr = gpa;
5756 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5759 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5763 struct x86_exception *exception)
5765 return emulator_read_write(ctxt, addr, val, bytes,
5766 exception, &read_emultor);
5769 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5773 struct x86_exception *exception)
5775 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5776 exception, &write_emultor);
5779 #define CMPXCHG_TYPE(t, ptr, old, new) \
5780 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5782 #ifdef CONFIG_X86_64
5783 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5785 # define CMPXCHG64(ptr, old, new) \
5786 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5789 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5794 struct x86_exception *exception)
5796 struct kvm_host_map map;
5797 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5802 /* guests cmpxchg8b have to be emulated atomically */
5803 if (bytes > 8 || (bytes & (bytes - 1)))
5806 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5808 if (gpa == UNMAPPED_GVA ||
5809 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5812 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5815 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5818 kaddr = map.hva + offset_in_page(gpa);
5822 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5825 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5828 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5831 exchanged = CMPXCHG64(kaddr, old, new);
5837 kvm_vcpu_unmap(vcpu, &map, true);
5840 return X86EMUL_CMPXCHG_FAILED;
5842 kvm_page_track_write(vcpu, gpa, new, bytes);
5844 return X86EMUL_CONTINUE;
5847 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5849 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5852 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5856 for (i = 0; i < vcpu->arch.pio.count; i++) {
5857 if (vcpu->arch.pio.in)
5858 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5859 vcpu->arch.pio.size, pd);
5861 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5862 vcpu->arch.pio.port, vcpu->arch.pio.size,
5866 pd += vcpu->arch.pio.size;
5871 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5872 unsigned short port, void *val,
5873 unsigned int count, bool in)
5875 vcpu->arch.pio.port = port;
5876 vcpu->arch.pio.in = in;
5877 vcpu->arch.pio.count = count;
5878 vcpu->arch.pio.size = size;
5880 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5881 vcpu->arch.pio.count = 0;
5885 vcpu->run->exit_reason = KVM_EXIT_IO;
5886 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5887 vcpu->run->io.size = size;
5888 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5889 vcpu->run->io.count = count;
5890 vcpu->run->io.port = port;
5895 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5896 int size, unsigned short port, void *val,
5899 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5902 if (vcpu->arch.pio.count)
5905 memset(vcpu->arch.pio_data, 0, size * count);
5907 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5910 memcpy(val, vcpu->arch.pio_data, size * count);
5911 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5912 vcpu->arch.pio.count = 0;
5919 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5920 int size, unsigned short port,
5921 const void *val, unsigned int count)
5923 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5925 memcpy(vcpu->arch.pio_data, val, size * count);
5926 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5927 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5930 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5932 return kvm_x86_ops->get_segment_base(vcpu, seg);
5935 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5937 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5940 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5942 if (!need_emulate_wbinvd(vcpu))
5943 return X86EMUL_CONTINUE;
5945 if (kvm_x86_ops->has_wbinvd_exit()) {
5946 int cpu = get_cpu();
5948 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5949 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5950 wbinvd_ipi, NULL, 1);
5952 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5955 return X86EMUL_CONTINUE;
5958 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5960 kvm_emulate_wbinvd_noskip(vcpu);
5961 return kvm_skip_emulated_instruction(vcpu);
5963 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5967 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5969 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5972 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5973 unsigned long *dest)
5975 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5978 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5979 unsigned long value)
5982 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5985 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5987 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5990 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5992 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5993 unsigned long value;
5997 value = kvm_read_cr0(vcpu);
6000 value = vcpu->arch.cr2;
6003 value = kvm_read_cr3(vcpu);
6006 value = kvm_read_cr4(vcpu);
6009 value = kvm_get_cr8(vcpu);
6012 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6019 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6021 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6026 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6029 vcpu->arch.cr2 = val;
6032 res = kvm_set_cr3(vcpu, val);
6035 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6038 res = kvm_set_cr8(vcpu, val);
6041 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6048 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6050 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
6053 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6055 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
6058 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6060 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
6063 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6065 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6068 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6070 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6073 static unsigned long emulator_get_cached_segment_base(
6074 struct x86_emulate_ctxt *ctxt, int seg)
6076 return get_segment_base(emul_to_vcpu(ctxt), seg);
6079 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6080 struct desc_struct *desc, u32 *base3,
6083 struct kvm_segment var;
6085 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6086 *selector = var.selector;
6089 memset(desc, 0, sizeof(*desc));
6097 set_desc_limit(desc, var.limit);
6098 set_desc_base(desc, (unsigned long)var.base);
6099 #ifdef CONFIG_X86_64
6101 *base3 = var.base >> 32;
6103 desc->type = var.type;
6105 desc->dpl = var.dpl;
6106 desc->p = var.present;
6107 desc->avl = var.avl;
6115 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6116 struct desc_struct *desc, u32 base3,
6119 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6120 struct kvm_segment var;
6122 var.selector = selector;
6123 var.base = get_desc_base(desc);
6124 #ifdef CONFIG_X86_64
6125 var.base |= ((u64)base3) << 32;
6127 var.limit = get_desc_limit(desc);
6129 var.limit = (var.limit << 12) | 0xfff;
6130 var.type = desc->type;
6131 var.dpl = desc->dpl;
6136 var.avl = desc->avl;
6137 var.present = desc->p;
6138 var.unusable = !var.present;
6141 kvm_set_segment(vcpu, &var, seg);
6145 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6146 u32 msr_index, u64 *pdata)
6148 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6151 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6152 u32 msr_index, u64 data)
6154 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6157 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6159 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6161 return vcpu->arch.smbase;
6164 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6166 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6168 vcpu->arch.smbase = smbase;
6171 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6174 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6177 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6178 u32 pmc, u64 *pdata)
6180 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6183 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6185 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6188 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6189 struct x86_instruction_info *info,
6190 enum x86_intercept_stage stage)
6192 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6195 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6196 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6198 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6201 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6203 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6206 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6208 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6211 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6213 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6216 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6218 return emul_to_vcpu(ctxt)->arch.hflags;
6221 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6223 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6226 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6227 const char *smstate)
6229 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6232 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6234 kvm_smm_changed(emul_to_vcpu(ctxt));
6237 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6239 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6242 static const struct x86_emulate_ops emulate_ops = {
6243 .read_gpr = emulator_read_gpr,
6244 .write_gpr = emulator_write_gpr,
6245 .read_std = emulator_read_std,
6246 .write_std = emulator_write_std,
6247 .read_phys = kvm_read_guest_phys_system,
6248 .fetch = kvm_fetch_guest_virt,
6249 .read_emulated = emulator_read_emulated,
6250 .write_emulated = emulator_write_emulated,
6251 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6252 .invlpg = emulator_invlpg,
6253 .pio_in_emulated = emulator_pio_in_emulated,
6254 .pio_out_emulated = emulator_pio_out_emulated,
6255 .get_segment = emulator_get_segment,
6256 .set_segment = emulator_set_segment,
6257 .get_cached_segment_base = emulator_get_cached_segment_base,
6258 .get_gdt = emulator_get_gdt,
6259 .get_idt = emulator_get_idt,
6260 .set_gdt = emulator_set_gdt,
6261 .set_idt = emulator_set_idt,
6262 .get_cr = emulator_get_cr,
6263 .set_cr = emulator_set_cr,
6264 .cpl = emulator_get_cpl,
6265 .get_dr = emulator_get_dr,
6266 .set_dr = emulator_set_dr,
6267 .get_smbase = emulator_get_smbase,
6268 .set_smbase = emulator_set_smbase,
6269 .set_msr = emulator_set_msr,
6270 .get_msr = emulator_get_msr,
6271 .check_pmc = emulator_check_pmc,
6272 .read_pmc = emulator_read_pmc,
6273 .halt = emulator_halt,
6274 .wbinvd = emulator_wbinvd,
6275 .fix_hypercall = emulator_fix_hypercall,
6276 .intercept = emulator_intercept,
6277 .get_cpuid = emulator_get_cpuid,
6278 .set_nmi_mask = emulator_set_nmi_mask,
6279 .get_hflags = emulator_get_hflags,
6280 .set_hflags = emulator_set_hflags,
6281 .pre_leave_smm = emulator_pre_leave_smm,
6282 .post_leave_smm = emulator_post_leave_smm,
6283 .set_xcr = emulator_set_xcr,
6286 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6288 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6290 * an sti; sti; sequence only disable interrupts for the first
6291 * instruction. So, if the last instruction, be it emulated or
6292 * not, left the system with the INT_STI flag enabled, it
6293 * means that the last instruction is an sti. We should not
6294 * leave the flag on in this case. The same goes for mov ss
6296 if (int_shadow & mask)
6298 if (unlikely(int_shadow || mask)) {
6299 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6301 kvm_make_request(KVM_REQ_EVENT, vcpu);
6305 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6307 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6308 if (ctxt->exception.vector == PF_VECTOR)
6309 return kvm_propagate_fault(vcpu, &ctxt->exception);
6311 if (ctxt->exception.error_code_valid)
6312 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6313 ctxt->exception.error_code);
6315 kvm_queue_exception(vcpu, ctxt->exception.vector);
6319 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6321 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6324 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6326 ctxt->eflags = kvm_get_rflags(vcpu);
6327 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6329 ctxt->eip = kvm_rip_read(vcpu);
6330 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6331 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6332 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6333 cs_db ? X86EMUL_MODE_PROT32 :
6334 X86EMUL_MODE_PROT16;
6335 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6336 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6337 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6339 init_decode_cache(ctxt);
6340 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6343 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6345 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6348 init_emulate_ctxt(vcpu);
6352 ctxt->_eip = ctxt->eip + inc_eip;
6353 ret = emulate_int_real(ctxt, irq);
6355 if (ret != X86EMUL_CONTINUE) {
6356 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6358 ctxt->eip = ctxt->_eip;
6359 kvm_rip_write(vcpu, ctxt->eip);
6360 kvm_set_rflags(vcpu, ctxt->eflags);
6363 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6365 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6367 ++vcpu->stat.insn_emulation_fail;
6368 trace_kvm_emulate_insn_failed(vcpu);
6370 if (emulation_type & EMULTYPE_VMWARE_GP) {
6371 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6375 if (emulation_type & EMULTYPE_SKIP) {
6376 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6377 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6378 vcpu->run->internal.ndata = 0;
6382 kvm_queue_exception(vcpu, UD_VECTOR);
6384 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6385 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6386 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6387 vcpu->run->internal.ndata = 0;
6394 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6395 bool write_fault_to_shadow_pgtable,
6401 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6404 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6407 if (!vcpu->arch.mmu->direct_map) {
6409 * Write permission should be allowed since only
6410 * write access need to be emulated.
6412 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6415 * If the mapping is invalid in guest, let cpu retry
6416 * it to generate fault.
6418 if (gpa == UNMAPPED_GVA)
6423 * Do not retry the unhandleable instruction if it faults on the
6424 * readonly host memory, otherwise it will goto a infinite loop:
6425 * retry instruction -> write #PF -> emulation fail -> retry
6426 * instruction -> ...
6428 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6431 * If the instruction failed on the error pfn, it can not be fixed,
6432 * report the error to userspace.
6434 if (is_error_noslot_pfn(pfn))
6437 kvm_release_pfn_clean(pfn);
6439 /* The instructions are well-emulated on direct mmu. */
6440 if (vcpu->arch.mmu->direct_map) {
6441 unsigned int indirect_shadow_pages;
6443 spin_lock(&vcpu->kvm->mmu_lock);
6444 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6445 spin_unlock(&vcpu->kvm->mmu_lock);
6447 if (indirect_shadow_pages)
6448 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6454 * if emulation was due to access to shadowed page table
6455 * and it failed try to unshadow page and re-enter the
6456 * guest to let CPU execute the instruction.
6458 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6461 * If the access faults on its page table, it can not
6462 * be fixed by unprotecting shadow page and it should
6463 * be reported to userspace.
6465 return !write_fault_to_shadow_pgtable;
6468 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6469 unsigned long cr2, int emulation_type)
6471 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6472 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6474 last_retry_eip = vcpu->arch.last_retry_eip;
6475 last_retry_addr = vcpu->arch.last_retry_addr;
6478 * If the emulation is caused by #PF and it is non-page_table
6479 * writing instruction, it means the VM-EXIT is caused by shadow
6480 * page protected, we can zap the shadow page and retry this
6481 * instruction directly.
6483 * Note: if the guest uses a non-page-table modifying instruction
6484 * on the PDE that points to the instruction, then we will unmap
6485 * the instruction and go to an infinite loop. So, we cache the
6486 * last retried eip and the last fault address, if we meet the eip
6487 * and the address again, we can break out of the potential infinite
6490 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6492 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6495 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6498 if (x86_page_table_writing_insn(ctxt))
6501 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6504 vcpu->arch.last_retry_eip = ctxt->eip;
6505 vcpu->arch.last_retry_addr = cr2;
6507 if (!vcpu->arch.mmu->direct_map)
6508 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6510 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6515 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6516 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6518 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6520 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6521 /* This is a good place to trace that we are exiting SMM. */
6522 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6524 /* Process a latched INIT or SMI, if any. */
6525 kvm_make_request(KVM_REQ_EVENT, vcpu);
6528 kvm_mmu_reset_context(vcpu);
6531 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6540 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6541 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6546 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6548 struct kvm_run *kvm_run = vcpu->run;
6550 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6551 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6552 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6553 kvm_run->debug.arch.exception = DB_VECTOR;
6554 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6557 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6561 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6563 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6566 r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6571 * rflags is the old, "raw" value of the flags. The new value has
6572 * not been saved yet.
6574 * This is correct even for TF set by the guest, because "the
6575 * processor will not generate this exception after the instruction
6576 * that sets the TF flag".
6578 if (unlikely(rflags & X86_EFLAGS_TF))
6579 r = kvm_vcpu_do_singlestep(vcpu);
6582 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6584 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6586 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6587 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6588 struct kvm_run *kvm_run = vcpu->run;
6589 unsigned long eip = kvm_get_linear_rip(vcpu);
6590 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6591 vcpu->arch.guest_debug_dr7,
6595 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6596 kvm_run->debug.arch.pc = eip;
6597 kvm_run->debug.arch.exception = DB_VECTOR;
6598 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6604 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6605 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6606 unsigned long eip = kvm_get_linear_rip(vcpu);
6607 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6612 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6613 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6614 kvm_queue_exception(vcpu, DB_VECTOR);
6623 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6625 switch (ctxt->opcode_len) {
6632 case 0xe6: /* OUT */
6636 case 0x6c: /* INS */
6638 case 0x6e: /* OUTS */
6645 case 0x33: /* RDPMC */
6654 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6661 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6662 bool writeback = true;
6663 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6665 vcpu->arch.l1tf_flush_l1d = true;
6668 * Clear write_fault_to_shadow_pgtable here to ensure it is
6671 vcpu->arch.write_fault_to_shadow_pgtable = false;
6672 kvm_clear_exception_queue(vcpu);
6674 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6675 init_emulate_ctxt(vcpu);
6678 * We will reenter on the same instruction since
6679 * we do not set complete_userspace_io. This does not
6680 * handle watchpoints yet, those would be handled in
6683 if (!(emulation_type & EMULTYPE_SKIP) &&
6684 kvm_vcpu_check_breakpoint(vcpu, &r))
6687 ctxt->interruptibility = 0;
6688 ctxt->have_exception = false;
6689 ctxt->exception.vector = -1;
6690 ctxt->perm_ok = false;
6692 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6694 r = x86_decode_insn(ctxt, insn, insn_len);
6696 trace_kvm_emulate_insn_start(vcpu);
6697 ++vcpu->stat.insn_emulation;
6698 if (r != EMULATION_OK) {
6699 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6700 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6701 kvm_queue_exception(vcpu, UD_VECTOR);
6704 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6707 if (ctxt->have_exception) {
6709 * #UD should result in just EMULATION_FAILED, and trap-like
6710 * exception should not be encountered during decode.
6712 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6713 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6714 inject_emulated_exception(vcpu);
6717 return handle_emulation_failure(vcpu, emulation_type);
6721 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6722 !is_vmware_backdoor_opcode(ctxt)) {
6723 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6728 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6729 * for kvm_skip_emulated_instruction(). The caller is responsible for
6730 * updating interruptibility state and injecting single-step #DBs.
6732 if (emulation_type & EMULTYPE_SKIP) {
6733 kvm_rip_write(vcpu, ctxt->_eip);
6734 if (ctxt->eflags & X86_EFLAGS_RF)
6735 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6739 if (retry_instruction(ctxt, cr2, emulation_type))
6742 /* this is needed for vmware backdoor interface to work since it
6743 changes registers values during IO operation */
6744 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6745 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6746 emulator_invalidate_register_cache(ctxt);
6750 /* Save the faulting GPA (cr2) in the address field */
6751 ctxt->exception.address = cr2;
6753 r = x86_emulate_insn(ctxt);
6755 if (r == EMULATION_INTERCEPTED)
6758 if (r == EMULATION_FAILED) {
6759 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6763 return handle_emulation_failure(vcpu, emulation_type);
6766 if (ctxt->have_exception) {
6768 if (inject_emulated_exception(vcpu))
6770 } else if (vcpu->arch.pio.count) {
6771 if (!vcpu->arch.pio.in) {
6772 /* FIXME: return into emulator if single-stepping. */
6773 vcpu->arch.pio.count = 0;
6776 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6779 } else if (vcpu->mmio_needed) {
6780 ++vcpu->stat.mmio_exits;
6782 if (!vcpu->mmio_is_write)
6785 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6786 } else if (r == EMULATION_RESTART)
6792 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6793 toggle_interruptibility(vcpu, ctxt->interruptibility);
6794 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6795 if (!ctxt->have_exception ||
6796 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6797 kvm_rip_write(vcpu, ctxt->eip);
6799 r = kvm_vcpu_do_singlestep(vcpu);
6800 __kvm_set_rflags(vcpu, ctxt->eflags);
6804 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6805 * do nothing, and it will be requested again as soon as
6806 * the shadow expires. But we still need to check here,
6807 * because POPF has no interrupt shadow.
6809 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6810 kvm_make_request(KVM_REQ_EVENT, vcpu);
6812 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6817 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6819 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6821 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6823 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6824 void *insn, int insn_len)
6826 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6828 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6830 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6832 vcpu->arch.pio.count = 0;
6836 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6838 vcpu->arch.pio.count = 0;
6840 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6843 return kvm_skip_emulated_instruction(vcpu);
6846 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6847 unsigned short port)
6849 unsigned long val = kvm_rax_read(vcpu);
6850 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6851 size, port, &val, 1);
6856 * Workaround userspace that relies on old KVM behavior of %rip being
6857 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6860 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6861 vcpu->arch.complete_userspace_io =
6862 complete_fast_pio_out_port_0x7e;
6863 kvm_skip_emulated_instruction(vcpu);
6865 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6866 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6871 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6875 /* We should only ever be called with arch.pio.count equal to 1 */
6876 BUG_ON(vcpu->arch.pio.count != 1);
6878 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6879 vcpu->arch.pio.count = 0;
6883 /* For size less than 4 we merge, else we zero extend */
6884 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6887 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6888 * the copy and tracing
6890 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6891 vcpu->arch.pio.port, &val, 1);
6892 kvm_rax_write(vcpu, val);
6894 return kvm_skip_emulated_instruction(vcpu);
6897 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6898 unsigned short port)
6903 /* For size less than 4 we merge, else we zero extend */
6904 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6906 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6909 kvm_rax_write(vcpu, val);
6913 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6914 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6919 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6924 ret = kvm_fast_pio_in(vcpu, size, port);
6926 ret = kvm_fast_pio_out(vcpu, size, port);
6927 return ret && kvm_skip_emulated_instruction(vcpu);
6929 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6931 static int kvmclock_cpu_down_prep(unsigned int cpu)
6933 __this_cpu_write(cpu_tsc_khz, 0);
6937 static void tsc_khz_changed(void *data)
6939 struct cpufreq_freqs *freq = data;
6940 unsigned long khz = 0;
6944 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6945 khz = cpufreq_quick_get(raw_smp_processor_id());
6948 __this_cpu_write(cpu_tsc_khz, khz);
6951 #ifdef CONFIG_X86_64
6952 static void kvm_hyperv_tsc_notifier(void)
6955 struct kvm_vcpu *vcpu;
6958 mutex_lock(&kvm_lock);
6959 list_for_each_entry(kvm, &vm_list, vm_list)
6960 kvm_make_mclock_inprogress_request(kvm);
6962 hyperv_stop_tsc_emulation();
6964 /* TSC frequency always matches when on Hyper-V */
6965 for_each_present_cpu(cpu)
6966 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6967 kvm_max_guest_tsc_khz = tsc_khz;
6969 list_for_each_entry(kvm, &vm_list, vm_list) {
6970 struct kvm_arch *ka = &kvm->arch;
6972 spin_lock(&ka->pvclock_gtod_sync_lock);
6974 pvclock_update_vm_gtod_copy(kvm);
6976 kvm_for_each_vcpu(cpu, vcpu, kvm)
6977 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6979 kvm_for_each_vcpu(cpu, vcpu, kvm)
6980 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6982 spin_unlock(&ka->pvclock_gtod_sync_lock);
6984 mutex_unlock(&kvm_lock);
6988 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6991 struct kvm_vcpu *vcpu;
6992 int i, send_ipi = 0;
6995 * We allow guests to temporarily run on slowing clocks,
6996 * provided we notify them after, or to run on accelerating
6997 * clocks, provided we notify them before. Thus time never
7000 * However, we have a problem. We can't atomically update
7001 * the frequency of a given CPU from this function; it is
7002 * merely a notifier, which can be called from any CPU.
7003 * Changing the TSC frequency at arbitrary points in time
7004 * requires a recomputation of local variables related to
7005 * the TSC for each VCPU. We must flag these local variables
7006 * to be updated and be sure the update takes place with the
7007 * new frequency before any guests proceed.
7009 * Unfortunately, the combination of hotplug CPU and frequency
7010 * change creates an intractable locking scenario; the order
7011 * of when these callouts happen is undefined with respect to
7012 * CPU hotplug, and they can race with each other. As such,
7013 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7014 * undefined; you can actually have a CPU frequency change take
7015 * place in between the computation of X and the setting of the
7016 * variable. To protect against this problem, all updates of
7017 * the per_cpu tsc_khz variable are done in an interrupt
7018 * protected IPI, and all callers wishing to update the value
7019 * must wait for a synchronous IPI to complete (which is trivial
7020 * if the caller is on the CPU already). This establishes the
7021 * necessary total order on variable updates.
7023 * Note that because a guest time update may take place
7024 * anytime after the setting of the VCPU's request bit, the
7025 * correct TSC value must be set before the request. However,
7026 * to ensure the update actually makes it to any guest which
7027 * starts running in hardware virtualization between the set
7028 * and the acquisition of the spinlock, we must also ping the
7029 * CPU after setting the request bit.
7033 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7035 mutex_lock(&kvm_lock);
7036 list_for_each_entry(kvm, &vm_list, vm_list) {
7037 kvm_for_each_vcpu(i, vcpu, kvm) {
7038 if (vcpu->cpu != cpu)
7040 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7041 if (vcpu->cpu != raw_smp_processor_id())
7045 mutex_unlock(&kvm_lock);
7047 if (freq->old < freq->new && send_ipi) {
7049 * We upscale the frequency. Must make the guest
7050 * doesn't see old kvmclock values while running with
7051 * the new frequency, otherwise we risk the guest sees
7052 * time go backwards.
7054 * In case we update the frequency for another cpu
7055 * (which might be in guest context) send an interrupt
7056 * to kick the cpu out of guest context. Next time
7057 * guest context is entered kvmclock will be updated,
7058 * so the guest will not see stale values.
7060 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7064 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7067 struct cpufreq_freqs *freq = data;
7070 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7072 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7075 for_each_cpu(cpu, freq->policy->cpus)
7076 __kvmclock_cpufreq_notifier(freq, cpu);
7081 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7082 .notifier_call = kvmclock_cpufreq_notifier
7085 static int kvmclock_cpu_online(unsigned int cpu)
7087 tsc_khz_changed(NULL);
7091 static void kvm_timer_init(void)
7093 max_tsc_khz = tsc_khz;
7095 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7096 #ifdef CONFIG_CPU_FREQ
7097 struct cpufreq_policy policy;
7100 memset(&policy, 0, sizeof(policy));
7102 cpufreq_get_policy(&policy, cpu);
7103 if (policy.cpuinfo.max_freq)
7104 max_tsc_khz = policy.cpuinfo.max_freq;
7107 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7108 CPUFREQ_TRANSITION_NOTIFIER);
7111 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7112 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7115 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7116 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7118 int kvm_is_in_guest(void)
7120 return __this_cpu_read(current_vcpu) != NULL;
7123 static int kvm_is_user_mode(void)
7127 if (__this_cpu_read(current_vcpu))
7128 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7130 return user_mode != 0;
7133 static unsigned long kvm_get_guest_ip(void)
7135 unsigned long ip = 0;
7137 if (__this_cpu_read(current_vcpu))
7138 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7143 static void kvm_handle_intel_pt_intr(void)
7145 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7147 kvm_make_request(KVM_REQ_PMI, vcpu);
7148 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7149 (unsigned long *)&vcpu->arch.pmu.global_status);
7152 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7153 .is_in_guest = kvm_is_in_guest,
7154 .is_user_mode = kvm_is_user_mode,
7155 .get_guest_ip = kvm_get_guest_ip,
7156 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7159 #ifdef CONFIG_X86_64
7160 static void pvclock_gtod_update_fn(struct work_struct *work)
7164 struct kvm_vcpu *vcpu;
7167 mutex_lock(&kvm_lock);
7168 list_for_each_entry(kvm, &vm_list, vm_list)
7169 kvm_for_each_vcpu(i, vcpu, kvm)
7170 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7171 atomic_set(&kvm_guest_has_master_clock, 0);
7172 mutex_unlock(&kvm_lock);
7175 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7178 * Notification about pvclock gtod data update.
7180 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7183 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7184 struct timekeeper *tk = priv;
7186 update_pvclock_gtod(tk);
7188 /* disable master clock if host does not trust, or does not
7189 * use, TSC based clocksource.
7191 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7192 atomic_read(&kvm_guest_has_master_clock) != 0)
7193 queue_work(system_long_wq, &pvclock_gtod_work);
7198 static struct notifier_block pvclock_gtod_notifier = {
7199 .notifier_call = pvclock_gtod_notify,
7203 int kvm_arch_init(void *opaque)
7206 struct kvm_x86_ops *ops = opaque;
7209 printk(KERN_ERR "kvm: already loaded the other module\n");
7214 if (!ops->cpu_has_kvm_support()) {
7215 printk(KERN_ERR "kvm: no hardware support\n");
7219 if (ops->disabled_by_bios()) {
7220 printk(KERN_ERR "kvm: disabled by bios\n");
7226 * KVM explicitly assumes that the guest has an FPU and
7227 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7228 * vCPU's FPU state as a fxregs_state struct.
7230 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7231 printk(KERN_ERR "kvm: inadequate fpu\n");
7237 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7238 __alignof__(struct fpu), SLAB_ACCOUNT,
7240 if (!x86_fpu_cache) {
7241 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7245 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7247 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7248 goto out_free_x86_fpu_cache;
7251 r = kvm_mmu_module_init();
7253 goto out_free_percpu;
7257 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7258 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7259 PT_PRESENT_MASK, 0, sme_me_mask);
7262 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7264 if (boot_cpu_has(X86_FEATURE_XSAVE))
7265 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7268 if (pi_inject_timer == -1)
7269 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7270 #ifdef CONFIG_X86_64
7271 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7273 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7274 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7280 free_percpu(shared_msrs);
7281 out_free_x86_fpu_cache:
7282 kmem_cache_destroy(x86_fpu_cache);
7287 void kvm_arch_exit(void)
7289 #ifdef CONFIG_X86_64
7290 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7291 clear_hv_tscchange_cb();
7294 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7296 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7297 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7298 CPUFREQ_TRANSITION_NOTIFIER);
7299 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7300 #ifdef CONFIG_X86_64
7301 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7304 kvm_mmu_module_exit();
7305 free_percpu(shared_msrs);
7306 kmem_cache_destroy(x86_fpu_cache);
7309 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7311 ++vcpu->stat.halt_exits;
7312 if (lapic_in_kernel(vcpu)) {
7313 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7316 vcpu->run->exit_reason = KVM_EXIT_HLT;
7320 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7322 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7324 int ret = kvm_skip_emulated_instruction(vcpu);
7326 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7327 * KVM_EXIT_DEBUG here.
7329 return kvm_vcpu_halt(vcpu) && ret;
7331 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7333 #ifdef CONFIG_X86_64
7334 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7335 unsigned long clock_type)
7337 struct kvm_clock_pairing clock_pairing;
7338 struct timespec64 ts;
7342 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7343 return -KVM_EOPNOTSUPP;
7345 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7346 return -KVM_EOPNOTSUPP;
7348 clock_pairing.sec = ts.tv_sec;
7349 clock_pairing.nsec = ts.tv_nsec;
7350 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7351 clock_pairing.flags = 0;
7352 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7355 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7356 sizeof(struct kvm_clock_pairing)))
7364 * kvm_pv_kick_cpu_op: Kick a vcpu.
7366 * @apicid - apicid of vcpu to be kicked.
7368 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7370 struct kvm_lapic_irq lapic_irq;
7372 lapic_irq.shorthand = 0;
7373 lapic_irq.dest_mode = 0;
7374 lapic_irq.level = 0;
7375 lapic_irq.dest_id = apicid;
7376 lapic_irq.msi_redir_hint = false;
7378 lapic_irq.delivery_mode = APIC_DM_REMRD;
7379 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7382 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7384 if (!lapic_in_kernel(vcpu)) {
7385 WARN_ON_ONCE(vcpu->arch.apicv_active);
7388 if (!vcpu->arch.apicv_active)
7391 vcpu->arch.apicv_active = false;
7392 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7395 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7397 struct kvm_vcpu *target = NULL;
7398 struct kvm_apic_map *map;
7401 map = rcu_dereference(kvm->arch.apic_map);
7403 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7404 target = map->phys_map[dest_id]->vcpu;
7408 if (target && READ_ONCE(target->ready))
7409 kvm_vcpu_yield_to(target);
7412 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7414 unsigned long nr, a0, a1, a2, a3, ret;
7417 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7418 return kvm_hv_hypercall(vcpu);
7420 nr = kvm_rax_read(vcpu);
7421 a0 = kvm_rbx_read(vcpu);
7422 a1 = kvm_rcx_read(vcpu);
7423 a2 = kvm_rdx_read(vcpu);
7424 a3 = kvm_rsi_read(vcpu);
7426 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7428 op_64_bit = is_64_bit_mode(vcpu);
7437 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7443 case KVM_HC_VAPIC_POLL_IRQ:
7446 case KVM_HC_KICK_CPU:
7447 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7448 kvm_sched_yield(vcpu->kvm, a1);
7451 #ifdef CONFIG_X86_64
7452 case KVM_HC_CLOCK_PAIRING:
7453 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7456 case KVM_HC_SEND_IPI:
7457 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7459 case KVM_HC_SCHED_YIELD:
7460 kvm_sched_yield(vcpu->kvm, a0);
7470 kvm_rax_write(vcpu, ret);
7472 ++vcpu->stat.hypercalls;
7473 return kvm_skip_emulated_instruction(vcpu);
7475 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7477 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7479 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7480 char instruction[3];
7481 unsigned long rip = kvm_rip_read(vcpu);
7483 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7485 return emulator_write_emulated(ctxt, rip, instruction, 3,
7489 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7491 return vcpu->run->request_interrupt_window &&
7492 likely(!pic_in_kernel(vcpu->kvm));
7495 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7497 struct kvm_run *kvm_run = vcpu->run;
7499 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7500 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7501 kvm_run->cr8 = kvm_get_cr8(vcpu);
7502 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7503 kvm_run->ready_for_interrupt_injection =
7504 pic_in_kernel(vcpu->kvm) ||
7505 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7508 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7512 if (!kvm_x86_ops->update_cr8_intercept)
7515 if (!lapic_in_kernel(vcpu))
7518 if (vcpu->arch.apicv_active)
7521 if (!vcpu->arch.apic->vapic_addr)
7522 max_irr = kvm_lapic_find_highest_irr(vcpu);
7529 tpr = kvm_lapic_get_cr8(vcpu);
7531 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7534 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7538 /* try to reinject previous events if any */
7540 if (vcpu->arch.exception.injected)
7541 kvm_x86_ops->queue_exception(vcpu);
7543 * Do not inject an NMI or interrupt if there is a pending
7544 * exception. Exceptions and interrupts are recognized at
7545 * instruction boundaries, i.e. the start of an instruction.
7546 * Trap-like exceptions, e.g. #DB, have higher priority than
7547 * NMIs and interrupts, i.e. traps are recognized before an
7548 * NMI/interrupt that's pending on the same instruction.
7549 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7550 * priority, but are only generated (pended) during instruction
7551 * execution, i.e. a pending fault-like exception means the
7552 * fault occurred on the *previous* instruction and must be
7553 * serviced prior to recognizing any new events in order to
7554 * fully complete the previous instruction.
7556 else if (!vcpu->arch.exception.pending) {
7557 if (vcpu->arch.nmi_injected)
7558 kvm_x86_ops->set_nmi(vcpu);
7559 else if (vcpu->arch.interrupt.injected)
7560 kvm_x86_ops->set_irq(vcpu);
7564 * Call check_nested_events() even if we reinjected a previous event
7565 * in order for caller to determine if it should require immediate-exit
7566 * from L2 to L1 due to pending L1 events which require exit
7569 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7570 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7575 /* try to inject new event if pending */
7576 if (vcpu->arch.exception.pending) {
7577 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7578 vcpu->arch.exception.has_error_code,
7579 vcpu->arch.exception.error_code);
7581 WARN_ON_ONCE(vcpu->arch.exception.injected);
7582 vcpu->arch.exception.pending = false;
7583 vcpu->arch.exception.injected = true;
7585 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7586 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7589 if (vcpu->arch.exception.nr == DB_VECTOR) {
7591 * This code assumes that nSVM doesn't use
7592 * check_nested_events(). If it does, the
7593 * DR6/DR7 changes should happen before L1
7594 * gets a #VMEXIT for an intercepted #DB in
7595 * L2. (Under VMX, on the other hand, the
7596 * DR6/DR7 changes should not happen in the
7597 * event of a VM-exit to L1 for an intercepted
7600 kvm_deliver_exception_payload(vcpu);
7601 if (vcpu->arch.dr7 & DR7_GD) {
7602 vcpu->arch.dr7 &= ~DR7_GD;
7603 kvm_update_dr7(vcpu);
7607 kvm_x86_ops->queue_exception(vcpu);
7610 /* Don't consider new event if we re-injected an event */
7611 if (kvm_event_needs_reinjection(vcpu))
7614 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7615 kvm_x86_ops->smi_allowed(vcpu)) {
7616 vcpu->arch.smi_pending = false;
7617 ++vcpu->arch.smi_count;
7619 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7620 --vcpu->arch.nmi_pending;
7621 vcpu->arch.nmi_injected = true;
7622 kvm_x86_ops->set_nmi(vcpu);
7623 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7625 * Because interrupts can be injected asynchronously, we are
7626 * calling check_nested_events again here to avoid a race condition.
7627 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7628 * proposal and current concerns. Perhaps we should be setting
7629 * KVM_REQ_EVENT only on certain events and not unconditionally?
7631 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7632 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7636 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7637 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7639 kvm_x86_ops->set_irq(vcpu);
7646 static void process_nmi(struct kvm_vcpu *vcpu)
7651 * x86 is limited to one NMI running, and one NMI pending after it.
7652 * If an NMI is already in progress, limit further NMIs to just one.
7653 * Otherwise, allow two (and we'll inject the first one immediately).
7655 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7658 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7659 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7660 kvm_make_request(KVM_REQ_EVENT, vcpu);
7663 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7666 flags |= seg->g << 23;
7667 flags |= seg->db << 22;
7668 flags |= seg->l << 21;
7669 flags |= seg->avl << 20;
7670 flags |= seg->present << 15;
7671 flags |= seg->dpl << 13;
7672 flags |= seg->s << 12;
7673 flags |= seg->type << 8;
7677 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7679 struct kvm_segment seg;
7682 kvm_get_segment(vcpu, &seg, n);
7683 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7686 offset = 0x7f84 + n * 12;
7688 offset = 0x7f2c + (n - 3) * 12;
7690 put_smstate(u32, buf, offset + 8, seg.base);
7691 put_smstate(u32, buf, offset + 4, seg.limit);
7692 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7695 #ifdef CONFIG_X86_64
7696 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7698 struct kvm_segment seg;
7702 kvm_get_segment(vcpu, &seg, n);
7703 offset = 0x7e00 + n * 16;
7705 flags = enter_smm_get_segment_flags(&seg) >> 8;
7706 put_smstate(u16, buf, offset, seg.selector);
7707 put_smstate(u16, buf, offset + 2, flags);
7708 put_smstate(u32, buf, offset + 4, seg.limit);
7709 put_smstate(u64, buf, offset + 8, seg.base);
7713 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7716 struct kvm_segment seg;
7720 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7721 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7722 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7723 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7725 for (i = 0; i < 8; i++)
7726 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7728 kvm_get_dr(vcpu, 6, &val);
7729 put_smstate(u32, buf, 0x7fcc, (u32)val);
7730 kvm_get_dr(vcpu, 7, &val);
7731 put_smstate(u32, buf, 0x7fc8, (u32)val);
7733 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7734 put_smstate(u32, buf, 0x7fc4, seg.selector);
7735 put_smstate(u32, buf, 0x7f64, seg.base);
7736 put_smstate(u32, buf, 0x7f60, seg.limit);
7737 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7739 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7740 put_smstate(u32, buf, 0x7fc0, seg.selector);
7741 put_smstate(u32, buf, 0x7f80, seg.base);
7742 put_smstate(u32, buf, 0x7f7c, seg.limit);
7743 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7745 kvm_x86_ops->get_gdt(vcpu, &dt);
7746 put_smstate(u32, buf, 0x7f74, dt.address);
7747 put_smstate(u32, buf, 0x7f70, dt.size);
7749 kvm_x86_ops->get_idt(vcpu, &dt);
7750 put_smstate(u32, buf, 0x7f58, dt.address);
7751 put_smstate(u32, buf, 0x7f54, dt.size);
7753 for (i = 0; i < 6; i++)
7754 enter_smm_save_seg_32(vcpu, buf, i);
7756 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7759 put_smstate(u32, buf, 0x7efc, 0x00020000);
7760 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7763 #ifdef CONFIG_X86_64
7764 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7767 struct kvm_segment seg;
7771 for (i = 0; i < 16; i++)
7772 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7774 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7775 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7777 kvm_get_dr(vcpu, 6, &val);
7778 put_smstate(u64, buf, 0x7f68, val);
7779 kvm_get_dr(vcpu, 7, &val);
7780 put_smstate(u64, buf, 0x7f60, val);
7782 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7783 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7784 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7786 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7789 put_smstate(u32, buf, 0x7efc, 0x00020064);
7791 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7793 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7794 put_smstate(u16, buf, 0x7e90, seg.selector);
7795 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7796 put_smstate(u32, buf, 0x7e94, seg.limit);
7797 put_smstate(u64, buf, 0x7e98, seg.base);
7799 kvm_x86_ops->get_idt(vcpu, &dt);
7800 put_smstate(u32, buf, 0x7e84, dt.size);
7801 put_smstate(u64, buf, 0x7e88, dt.address);
7803 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7804 put_smstate(u16, buf, 0x7e70, seg.selector);
7805 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7806 put_smstate(u32, buf, 0x7e74, seg.limit);
7807 put_smstate(u64, buf, 0x7e78, seg.base);
7809 kvm_x86_ops->get_gdt(vcpu, &dt);
7810 put_smstate(u32, buf, 0x7e64, dt.size);
7811 put_smstate(u64, buf, 0x7e68, dt.address);
7813 for (i = 0; i < 6; i++)
7814 enter_smm_save_seg_64(vcpu, buf, i);
7818 static void enter_smm(struct kvm_vcpu *vcpu)
7820 struct kvm_segment cs, ds;
7825 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7826 memset(buf, 0, 512);
7827 #ifdef CONFIG_X86_64
7828 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7829 enter_smm_save_state_64(vcpu, buf);
7832 enter_smm_save_state_32(vcpu, buf);
7835 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7836 * vCPU state (e.g. leave guest mode) after we've saved the state into
7837 * the SMM state-save area.
7839 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7841 vcpu->arch.hflags |= HF_SMM_MASK;
7842 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7844 if (kvm_x86_ops->get_nmi_mask(vcpu))
7845 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7847 kvm_x86_ops->set_nmi_mask(vcpu, true);
7849 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7850 kvm_rip_write(vcpu, 0x8000);
7852 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7853 kvm_x86_ops->set_cr0(vcpu, cr0);
7854 vcpu->arch.cr0 = cr0;
7856 kvm_x86_ops->set_cr4(vcpu, 0);
7858 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7859 dt.address = dt.size = 0;
7860 kvm_x86_ops->set_idt(vcpu, &dt);
7862 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7864 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7865 cs.base = vcpu->arch.smbase;
7870 cs.limit = ds.limit = 0xffffffff;
7871 cs.type = ds.type = 0x3;
7872 cs.dpl = ds.dpl = 0;
7877 cs.avl = ds.avl = 0;
7878 cs.present = ds.present = 1;
7879 cs.unusable = ds.unusable = 0;
7880 cs.padding = ds.padding = 0;
7882 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7883 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7884 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7885 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7886 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7887 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7889 #ifdef CONFIG_X86_64
7890 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7891 kvm_x86_ops->set_efer(vcpu, 0);
7894 kvm_update_cpuid(vcpu);
7895 kvm_mmu_reset_context(vcpu);
7898 static void process_smi(struct kvm_vcpu *vcpu)
7900 vcpu->arch.smi_pending = true;
7901 kvm_make_request(KVM_REQ_EVENT, vcpu);
7904 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
7905 unsigned long *vcpu_bitmap)
7909 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
7911 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
7914 free_cpumask_var(cpus);
7917 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7919 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7922 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7924 if (!kvm_apic_present(vcpu))
7927 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7929 if (irqchip_split(vcpu->kvm))
7930 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7932 if (vcpu->arch.apicv_active)
7933 kvm_x86_ops->sync_pir_to_irr(vcpu);
7934 if (ioapic_in_kernel(vcpu->kvm))
7935 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7938 if (is_guest_mode(vcpu))
7939 vcpu->arch.load_eoi_exitmap_pending = true;
7941 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7944 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7946 u64 eoi_exit_bitmap[4];
7948 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7951 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7952 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7953 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7956 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7957 unsigned long start, unsigned long end,
7960 unsigned long apic_address;
7963 * The physical address of apic access page is stored in the VMCS.
7964 * Update it when it becomes invalid.
7966 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7967 if (start <= apic_address && apic_address < end)
7968 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7973 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7975 struct page *page = NULL;
7977 if (!lapic_in_kernel(vcpu))
7980 if (!kvm_x86_ops->set_apic_access_page_addr)
7983 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7984 if (is_error_page(page))
7986 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7989 * Do not pin apic access page in memory, the MMU notifier
7990 * will call us again if it is migrated or swapped out.
7995 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7997 smp_send_reschedule(vcpu->cpu);
7999 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8002 * Returns 1 to let vcpu_run() continue the guest execution loop without
8003 * exiting to the userspace. Otherwise, the value will be returned to the
8006 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8010 dm_request_for_irq_injection(vcpu) &&
8011 kvm_cpu_accept_dm_intr(vcpu);
8013 bool req_immediate_exit = false;
8015 if (kvm_request_pending(vcpu)) {
8016 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8017 if (unlikely(!kvm_x86_ops->get_vmcs12_pages(vcpu))) {
8022 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8023 kvm_mmu_unload(vcpu);
8024 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8025 __kvm_migrate_timers(vcpu);
8026 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8027 kvm_gen_update_masterclock(vcpu->kvm);
8028 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8029 kvm_gen_kvmclock_update(vcpu);
8030 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8031 r = kvm_guest_time_update(vcpu);
8035 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8036 kvm_mmu_sync_roots(vcpu);
8037 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
8038 kvm_mmu_load_cr3(vcpu);
8039 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
8040 kvm_vcpu_flush_tlb(vcpu, true);
8041 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8042 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8046 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8047 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8048 vcpu->mmio_needed = 0;
8052 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8053 /* Page is swapped out. Do synthetic halt */
8054 vcpu->arch.apf.halted = true;
8058 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8059 record_steal_time(vcpu);
8060 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8062 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8064 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8065 kvm_pmu_handle_event(vcpu);
8066 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8067 kvm_pmu_deliver_pmi(vcpu);
8068 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8069 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8070 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8071 vcpu->arch.ioapic_handled_vectors)) {
8072 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8073 vcpu->run->eoi.vector =
8074 vcpu->arch.pending_ioapic_eoi;
8079 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8080 vcpu_scan_ioapic(vcpu);
8081 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8082 vcpu_load_eoi_exitmap(vcpu);
8083 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8084 kvm_vcpu_reload_apic_access_page(vcpu);
8085 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8086 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8087 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8091 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8092 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8093 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8097 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8098 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8099 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8105 * KVM_REQ_HV_STIMER has to be processed after
8106 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8107 * depend on the guest clock being up-to-date
8109 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8110 kvm_hv_process_stimers(vcpu);
8113 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8114 ++vcpu->stat.req_event;
8115 kvm_apic_accept_events(vcpu);
8116 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8121 if (inject_pending_event(vcpu, req_int_win) != 0)
8122 req_immediate_exit = true;
8124 /* Enable SMI/NMI/IRQ window open exits if needed.
8126 * SMIs have three cases:
8127 * 1) They can be nested, and then there is nothing to
8128 * do here because RSM will cause a vmexit anyway.
8129 * 2) There is an ISA-specific reason why SMI cannot be
8130 * injected, and the moment when this changes can be
8132 * 3) Or the SMI can be pending because
8133 * inject_pending_event has completed the injection
8134 * of an IRQ or NMI from the previous vmexit, and
8135 * then we request an immediate exit to inject the
8138 if (vcpu->arch.smi_pending && !is_smm(vcpu))
8139 if (!kvm_x86_ops->enable_smi_window(vcpu))
8140 req_immediate_exit = true;
8141 if (vcpu->arch.nmi_pending)
8142 kvm_x86_ops->enable_nmi_window(vcpu);
8143 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8144 kvm_x86_ops->enable_irq_window(vcpu);
8145 WARN_ON(vcpu->arch.exception.pending);
8148 if (kvm_lapic_enabled(vcpu)) {
8149 update_cr8_intercept(vcpu);
8150 kvm_lapic_sync_to_vapic(vcpu);
8154 r = kvm_mmu_reload(vcpu);
8156 goto cancel_injection;
8161 kvm_x86_ops->prepare_guest_switch(vcpu);
8164 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8165 * IPI are then delayed after guest entry, which ensures that they
8166 * result in virtual interrupt delivery.
8168 local_irq_disable();
8169 vcpu->mode = IN_GUEST_MODE;
8171 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8174 * 1) We should set ->mode before checking ->requests. Please see
8175 * the comment in kvm_vcpu_exiting_guest_mode().
8177 * 2) For APICv, we should set ->mode before checking PID.ON. This
8178 * pairs with the memory barrier implicit in pi_test_and_set_on
8179 * (see vmx_deliver_posted_interrupt).
8181 * 3) This also orders the write to mode from any reads to the page
8182 * tables done while the VCPU is running. Please see the comment
8183 * in kvm_flush_remote_tlbs.
8185 smp_mb__after_srcu_read_unlock();
8188 * This handles the case where a posted interrupt was
8189 * notified with kvm_vcpu_kick.
8191 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8192 kvm_x86_ops->sync_pir_to_irr(vcpu);
8194 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8195 || need_resched() || signal_pending(current)) {
8196 vcpu->mode = OUTSIDE_GUEST_MODE;
8200 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8202 goto cancel_injection;
8205 if (req_immediate_exit) {
8206 kvm_make_request(KVM_REQ_EVENT, vcpu);
8207 kvm_x86_ops->request_immediate_exit(vcpu);
8210 trace_kvm_entry(vcpu->vcpu_id);
8211 guest_enter_irqoff();
8213 /* The preempt notifier should have taken care of the FPU already. */
8214 WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8216 if (unlikely(vcpu->arch.switch_db_regs)) {
8218 set_debugreg(vcpu->arch.eff_db[0], 0);
8219 set_debugreg(vcpu->arch.eff_db[1], 1);
8220 set_debugreg(vcpu->arch.eff_db[2], 2);
8221 set_debugreg(vcpu->arch.eff_db[3], 3);
8222 set_debugreg(vcpu->arch.dr6, 6);
8223 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8226 kvm_x86_ops->run(vcpu);
8229 * Do this here before restoring debug registers on the host. And
8230 * since we do this before handling the vmexit, a DR access vmexit
8231 * can (a) read the correct value of the debug registers, (b) set
8232 * KVM_DEBUGREG_WONT_EXIT again.
8234 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8235 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8236 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8237 kvm_update_dr0123(vcpu);
8238 kvm_update_dr6(vcpu);
8239 kvm_update_dr7(vcpu);
8240 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8244 * If the guest has used debug registers, at least dr7
8245 * will be disabled while returning to the host.
8246 * If we don't have active breakpoints in the host, we don't
8247 * care about the messed up debug address registers. But if
8248 * we have some of them active, restore the old state.
8250 if (hw_breakpoint_active())
8251 hw_breakpoint_restore();
8253 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8255 vcpu->mode = OUTSIDE_GUEST_MODE;
8258 kvm_x86_ops->handle_exit_irqoff(vcpu);
8261 * Consume any pending interrupts, including the possible source of
8262 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8263 * An instruction is required after local_irq_enable() to fully unblock
8264 * interrupts on processors that implement an interrupt shadow, the
8265 * stat.exits increment will do nicely.
8267 kvm_before_interrupt(vcpu);
8270 local_irq_disable();
8271 kvm_after_interrupt(vcpu);
8273 guest_exit_irqoff();
8274 if (lapic_in_kernel(vcpu)) {
8275 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8276 if (delta != S64_MIN) {
8277 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8278 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8285 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8288 * Profile KVM exit RIPs:
8290 if (unlikely(prof_on == KVM_PROFILING)) {
8291 unsigned long rip = kvm_rip_read(vcpu);
8292 profile_hit(KVM_PROFILING, (void *)rip);
8295 if (unlikely(vcpu->arch.tsc_always_catchup))
8296 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8298 if (vcpu->arch.apic_attention)
8299 kvm_lapic_sync_from_vapic(vcpu);
8301 vcpu->arch.gpa_available = false;
8302 r = kvm_x86_ops->handle_exit(vcpu);
8306 kvm_x86_ops->cancel_injection(vcpu);
8307 if (unlikely(vcpu->arch.apic_attention))
8308 kvm_lapic_sync_from_vapic(vcpu);
8313 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8315 if (!kvm_arch_vcpu_runnable(vcpu) &&
8316 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8317 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8318 kvm_vcpu_block(vcpu);
8319 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8321 if (kvm_x86_ops->post_block)
8322 kvm_x86_ops->post_block(vcpu);
8324 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8328 kvm_apic_accept_events(vcpu);
8329 switch(vcpu->arch.mp_state) {
8330 case KVM_MP_STATE_HALTED:
8331 vcpu->arch.pv.pv_unhalted = false;
8332 vcpu->arch.mp_state =
8333 KVM_MP_STATE_RUNNABLE;
8335 case KVM_MP_STATE_RUNNABLE:
8336 vcpu->arch.apf.halted = false;
8338 case KVM_MP_STATE_INIT_RECEIVED:
8347 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8349 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8350 kvm_x86_ops->check_nested_events(vcpu, false);
8352 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8353 !vcpu->arch.apf.halted);
8356 static int vcpu_run(struct kvm_vcpu *vcpu)
8359 struct kvm *kvm = vcpu->kvm;
8361 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8362 vcpu->arch.l1tf_flush_l1d = true;
8365 if (kvm_vcpu_running(vcpu)) {
8366 r = vcpu_enter_guest(vcpu);
8368 r = vcpu_block(kvm, vcpu);
8374 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8375 if (kvm_cpu_has_pending_timer(vcpu))
8376 kvm_inject_pending_timer_irqs(vcpu);
8378 if (dm_request_for_irq_injection(vcpu) &&
8379 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8381 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8382 ++vcpu->stat.request_irq_exits;
8386 kvm_check_async_pf_completion(vcpu);
8388 if (signal_pending(current)) {
8390 vcpu->run->exit_reason = KVM_EXIT_INTR;
8391 ++vcpu->stat.signal_exits;
8394 if (need_resched()) {
8395 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8397 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8401 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8406 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8410 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8411 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8412 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8416 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8418 BUG_ON(!vcpu->arch.pio.count);
8420 return complete_emulated_io(vcpu);
8424 * Implements the following, as a state machine:
8428 * for each mmio piece in the fragment
8436 * for each mmio piece in the fragment
8441 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8443 struct kvm_run *run = vcpu->run;
8444 struct kvm_mmio_fragment *frag;
8447 BUG_ON(!vcpu->mmio_needed);
8449 /* Complete previous fragment */
8450 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8451 len = min(8u, frag->len);
8452 if (!vcpu->mmio_is_write)
8453 memcpy(frag->data, run->mmio.data, len);
8455 if (frag->len <= 8) {
8456 /* Switch to the next fragment. */
8458 vcpu->mmio_cur_fragment++;
8460 /* Go forward to the next mmio piece. */
8466 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8467 vcpu->mmio_needed = 0;
8469 /* FIXME: return into emulator if single-stepping. */
8470 if (vcpu->mmio_is_write)
8472 vcpu->mmio_read_completed = 1;
8473 return complete_emulated_io(vcpu);
8476 run->exit_reason = KVM_EXIT_MMIO;
8477 run->mmio.phys_addr = frag->gpa;
8478 if (vcpu->mmio_is_write)
8479 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8480 run->mmio.len = min(8u, frag->len);
8481 run->mmio.is_write = vcpu->mmio_is_write;
8482 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8486 /* Swap (qemu) user FPU context for the guest FPU context. */
8487 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8491 copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8492 /* PKRU is separately restored in kvm_x86_ops->run. */
8493 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8494 ~XFEATURE_MASK_PKRU);
8496 fpregs_mark_activate();
8502 /* When vcpu_run ends, restore user space FPU context. */
8503 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8507 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8508 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8510 fpregs_mark_activate();
8513 ++vcpu->stat.fpu_reload;
8517 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8522 kvm_sigset_activate(vcpu);
8523 kvm_load_guest_fpu(vcpu);
8525 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8526 if (kvm_run->immediate_exit) {
8530 kvm_vcpu_block(vcpu);
8531 kvm_apic_accept_events(vcpu);
8532 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8534 if (signal_pending(current)) {
8536 vcpu->run->exit_reason = KVM_EXIT_INTR;
8537 ++vcpu->stat.signal_exits;
8542 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8547 if (vcpu->run->kvm_dirty_regs) {
8548 r = sync_regs(vcpu);
8553 /* re-sync apic's tpr */
8554 if (!lapic_in_kernel(vcpu)) {
8555 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8561 if (unlikely(vcpu->arch.complete_userspace_io)) {
8562 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8563 vcpu->arch.complete_userspace_io = NULL;
8568 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8570 if (kvm_run->immediate_exit)
8576 kvm_put_guest_fpu(vcpu);
8577 if (vcpu->run->kvm_valid_regs)
8579 post_kvm_run_save(vcpu);
8580 kvm_sigset_deactivate(vcpu);
8586 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8588 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8590 * We are here if userspace calls get_regs() in the middle of
8591 * instruction emulation. Registers state needs to be copied
8592 * back from emulation context to vcpu. Userspace shouldn't do
8593 * that usually, but some bad designed PV devices (vmware
8594 * backdoor interface) need this to work
8596 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8597 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8599 regs->rax = kvm_rax_read(vcpu);
8600 regs->rbx = kvm_rbx_read(vcpu);
8601 regs->rcx = kvm_rcx_read(vcpu);
8602 regs->rdx = kvm_rdx_read(vcpu);
8603 regs->rsi = kvm_rsi_read(vcpu);
8604 regs->rdi = kvm_rdi_read(vcpu);
8605 regs->rsp = kvm_rsp_read(vcpu);
8606 regs->rbp = kvm_rbp_read(vcpu);
8607 #ifdef CONFIG_X86_64
8608 regs->r8 = kvm_r8_read(vcpu);
8609 regs->r9 = kvm_r9_read(vcpu);
8610 regs->r10 = kvm_r10_read(vcpu);
8611 regs->r11 = kvm_r11_read(vcpu);
8612 regs->r12 = kvm_r12_read(vcpu);
8613 regs->r13 = kvm_r13_read(vcpu);
8614 regs->r14 = kvm_r14_read(vcpu);
8615 regs->r15 = kvm_r15_read(vcpu);
8618 regs->rip = kvm_rip_read(vcpu);
8619 regs->rflags = kvm_get_rflags(vcpu);
8622 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8625 __get_regs(vcpu, regs);
8630 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8632 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8633 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8635 kvm_rax_write(vcpu, regs->rax);
8636 kvm_rbx_write(vcpu, regs->rbx);
8637 kvm_rcx_write(vcpu, regs->rcx);
8638 kvm_rdx_write(vcpu, regs->rdx);
8639 kvm_rsi_write(vcpu, regs->rsi);
8640 kvm_rdi_write(vcpu, regs->rdi);
8641 kvm_rsp_write(vcpu, regs->rsp);
8642 kvm_rbp_write(vcpu, regs->rbp);
8643 #ifdef CONFIG_X86_64
8644 kvm_r8_write(vcpu, regs->r8);
8645 kvm_r9_write(vcpu, regs->r9);
8646 kvm_r10_write(vcpu, regs->r10);
8647 kvm_r11_write(vcpu, regs->r11);
8648 kvm_r12_write(vcpu, regs->r12);
8649 kvm_r13_write(vcpu, regs->r13);
8650 kvm_r14_write(vcpu, regs->r14);
8651 kvm_r15_write(vcpu, regs->r15);
8654 kvm_rip_write(vcpu, regs->rip);
8655 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8657 vcpu->arch.exception.pending = false;
8659 kvm_make_request(KVM_REQ_EVENT, vcpu);
8662 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8665 __set_regs(vcpu, regs);
8670 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8672 struct kvm_segment cs;
8674 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8678 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8680 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8684 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8685 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8686 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8687 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8688 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8689 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8691 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8692 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8694 kvm_x86_ops->get_idt(vcpu, &dt);
8695 sregs->idt.limit = dt.size;
8696 sregs->idt.base = dt.address;
8697 kvm_x86_ops->get_gdt(vcpu, &dt);
8698 sregs->gdt.limit = dt.size;
8699 sregs->gdt.base = dt.address;
8701 sregs->cr0 = kvm_read_cr0(vcpu);
8702 sregs->cr2 = vcpu->arch.cr2;
8703 sregs->cr3 = kvm_read_cr3(vcpu);
8704 sregs->cr4 = kvm_read_cr4(vcpu);
8705 sregs->cr8 = kvm_get_cr8(vcpu);
8706 sregs->efer = vcpu->arch.efer;
8707 sregs->apic_base = kvm_get_apic_base(vcpu);
8709 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8711 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8712 set_bit(vcpu->arch.interrupt.nr,
8713 (unsigned long *)sregs->interrupt_bitmap);
8716 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8717 struct kvm_sregs *sregs)
8720 __get_sregs(vcpu, sregs);
8725 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8726 struct kvm_mp_state *mp_state)
8730 kvm_apic_accept_events(vcpu);
8731 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8732 vcpu->arch.pv.pv_unhalted)
8733 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8735 mp_state->mp_state = vcpu->arch.mp_state;
8741 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8742 struct kvm_mp_state *mp_state)
8748 if (!lapic_in_kernel(vcpu) &&
8749 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8753 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
8754 * INIT state; latched init should be reported using
8755 * KVM_SET_VCPU_EVENTS, so reject it here.
8757 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
8758 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8759 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8762 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8763 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8764 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8766 vcpu->arch.mp_state = mp_state->mp_state;
8767 kvm_make_request(KVM_REQ_EVENT, vcpu);
8775 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8776 int reason, bool has_error_code, u32 error_code)
8778 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8781 init_emulate_ctxt(vcpu);
8783 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8784 has_error_code, error_code);
8786 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8787 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8788 vcpu->run->internal.ndata = 0;
8792 kvm_rip_write(vcpu, ctxt->eip);
8793 kvm_set_rflags(vcpu, ctxt->eflags);
8794 kvm_make_request(KVM_REQ_EVENT, vcpu);
8797 EXPORT_SYMBOL_GPL(kvm_task_switch);
8799 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8801 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8803 * When EFER.LME and CR0.PG are set, the processor is in
8804 * 64-bit mode (though maybe in a 32-bit code segment).
8805 * CR4.PAE and EFER.LMA must be set.
8807 if (!(sregs->cr4 & X86_CR4_PAE)
8808 || !(sregs->efer & EFER_LMA))
8812 * Not in 64-bit mode: EFER.LMA is clear and the code
8813 * segment cannot be 64-bit.
8815 if (sregs->efer & EFER_LMA || sregs->cs.l)
8819 return kvm_valid_cr4(vcpu, sregs->cr4);
8822 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8824 struct msr_data apic_base_msr;
8825 int mmu_reset_needed = 0;
8826 int cpuid_update_needed = 0;
8827 int pending_vec, max_bits, idx;
8831 if (kvm_valid_sregs(vcpu, sregs))
8834 apic_base_msr.data = sregs->apic_base;
8835 apic_base_msr.host_initiated = true;
8836 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8839 dt.size = sregs->idt.limit;
8840 dt.address = sregs->idt.base;
8841 kvm_x86_ops->set_idt(vcpu, &dt);
8842 dt.size = sregs->gdt.limit;
8843 dt.address = sregs->gdt.base;
8844 kvm_x86_ops->set_gdt(vcpu, &dt);
8846 vcpu->arch.cr2 = sregs->cr2;
8847 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8848 vcpu->arch.cr3 = sregs->cr3;
8849 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
8851 kvm_set_cr8(vcpu, sregs->cr8);
8853 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8854 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8856 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8857 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8858 vcpu->arch.cr0 = sregs->cr0;
8860 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8861 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8862 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8863 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8864 if (cpuid_update_needed)
8865 kvm_update_cpuid(vcpu);
8867 idx = srcu_read_lock(&vcpu->kvm->srcu);
8868 if (is_pae_paging(vcpu)) {
8869 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8870 mmu_reset_needed = 1;
8872 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8874 if (mmu_reset_needed)
8875 kvm_mmu_reset_context(vcpu);
8877 max_bits = KVM_NR_INTERRUPTS;
8878 pending_vec = find_first_bit(
8879 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8880 if (pending_vec < max_bits) {
8881 kvm_queue_interrupt(vcpu, pending_vec, false);
8882 pr_debug("Set back pending irq %d\n", pending_vec);
8885 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8886 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8887 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8888 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8889 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8890 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8892 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8893 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8895 update_cr8_intercept(vcpu);
8897 /* Older userspace won't unhalt the vcpu on reset. */
8898 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8899 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8901 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8903 kvm_make_request(KVM_REQ_EVENT, vcpu);
8910 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8911 struct kvm_sregs *sregs)
8916 ret = __set_sregs(vcpu, sregs);
8921 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8922 struct kvm_guest_debug *dbg)
8924 unsigned long rflags;
8929 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8931 if (vcpu->arch.exception.pending)
8933 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8934 kvm_queue_exception(vcpu, DB_VECTOR);
8936 kvm_queue_exception(vcpu, BP_VECTOR);
8940 * Read rflags as long as potentially injected trace flags are still
8943 rflags = kvm_get_rflags(vcpu);
8945 vcpu->guest_debug = dbg->control;
8946 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8947 vcpu->guest_debug = 0;
8949 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8950 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8951 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8952 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8954 for (i = 0; i < KVM_NR_DB_REGS; i++)
8955 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8957 kvm_update_dr7(vcpu);
8959 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8960 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8961 get_segment_base(vcpu, VCPU_SREG_CS);
8964 * Trigger an rflags update that will inject or remove the trace
8967 kvm_set_rflags(vcpu, rflags);
8969 kvm_x86_ops->update_bp_intercept(vcpu);
8979 * Translate a guest virtual address to a guest physical address.
8981 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8982 struct kvm_translation *tr)
8984 unsigned long vaddr = tr->linear_address;
8990 idx = srcu_read_lock(&vcpu->kvm->srcu);
8991 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8992 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8993 tr->physical_address = gpa;
8994 tr->valid = gpa != UNMAPPED_GVA;
9002 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9004 struct fxregs_state *fxsave;
9008 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9009 memcpy(fpu->fpr, fxsave->st_space, 128);
9010 fpu->fcw = fxsave->cwd;
9011 fpu->fsw = fxsave->swd;
9012 fpu->ftwx = fxsave->twd;
9013 fpu->last_opcode = fxsave->fop;
9014 fpu->last_ip = fxsave->rip;
9015 fpu->last_dp = fxsave->rdp;
9016 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9022 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9024 struct fxregs_state *fxsave;
9028 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9030 memcpy(fxsave->st_space, fpu->fpr, 128);
9031 fxsave->cwd = fpu->fcw;
9032 fxsave->swd = fpu->fsw;
9033 fxsave->twd = fpu->ftwx;
9034 fxsave->fop = fpu->last_opcode;
9035 fxsave->rip = fpu->last_ip;
9036 fxsave->rdp = fpu->last_dp;
9037 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9043 static void store_regs(struct kvm_vcpu *vcpu)
9045 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9047 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9048 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9050 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9051 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9053 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9054 kvm_vcpu_ioctl_x86_get_vcpu_events(
9055 vcpu, &vcpu->run->s.regs.events);
9058 static int sync_regs(struct kvm_vcpu *vcpu)
9060 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9063 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9064 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9065 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9067 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9068 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9070 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9072 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9073 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9074 vcpu, &vcpu->run->s.regs.events))
9076 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9082 static void fx_init(struct kvm_vcpu *vcpu)
9084 fpstate_init(&vcpu->arch.guest_fpu->state);
9085 if (boot_cpu_has(X86_FEATURE_XSAVES))
9086 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9087 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9090 * Ensure guest xcr0 is valid for loading
9092 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9094 vcpu->arch.cr0 |= X86_CR0_ET;
9097 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
9099 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
9100 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9102 kvm_release_pfn(cache->pfn, cache->dirty, cache);
9104 kvmclock_reset(vcpu);
9106 kvm_x86_ops->vcpu_free(vcpu);
9107 free_cpumask_var(wbinvd_dirty_mask);
9110 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
9113 struct kvm_vcpu *vcpu;
9115 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9116 printk_once(KERN_WARNING
9117 "kvm: SMP vm created on host with unstable TSC; "
9118 "guest TSC will not be reliable\n");
9120 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
9125 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
9127 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9128 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9129 kvm_vcpu_mtrr_init(vcpu);
9131 kvm_vcpu_reset(vcpu, false);
9132 kvm_init_mmu(vcpu, false);
9137 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9139 struct msr_data msr;
9140 struct kvm *kvm = vcpu->kvm;
9142 kvm_hv_vcpu_postcreate(vcpu);
9144 if (mutex_lock_killable(&vcpu->mutex))
9148 msr.index = MSR_IA32_TSC;
9149 msr.host_initiated = true;
9150 kvm_write_tsc(vcpu, &msr);
9153 /* poll control enabled by default */
9154 vcpu->arch.msr_kvm_poll_control = 1;
9156 mutex_unlock(&vcpu->mutex);
9158 if (!kvmclock_periodic_sync)
9161 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9162 KVMCLOCK_SYNC_PERIOD);
9165 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9167 vcpu->arch.apf.msr_val = 0;
9170 kvm_mmu_unload(vcpu);
9173 kvm_x86_ops->vcpu_free(vcpu);
9176 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9178 kvm_lapic_reset(vcpu, init_event);
9180 vcpu->arch.hflags = 0;
9182 vcpu->arch.smi_pending = 0;
9183 vcpu->arch.smi_count = 0;
9184 atomic_set(&vcpu->arch.nmi_queued, 0);
9185 vcpu->arch.nmi_pending = 0;
9186 vcpu->arch.nmi_injected = false;
9187 kvm_clear_interrupt_queue(vcpu);
9188 kvm_clear_exception_queue(vcpu);
9189 vcpu->arch.exception.pending = false;
9191 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9192 kvm_update_dr0123(vcpu);
9193 vcpu->arch.dr6 = DR6_INIT;
9194 kvm_update_dr6(vcpu);
9195 vcpu->arch.dr7 = DR7_FIXED_1;
9196 kvm_update_dr7(vcpu);
9200 kvm_make_request(KVM_REQ_EVENT, vcpu);
9201 vcpu->arch.apf.msr_val = 0;
9202 vcpu->arch.st.msr_val = 0;
9204 kvmclock_reset(vcpu);
9206 kvm_clear_async_pf_completion_queue(vcpu);
9207 kvm_async_pf_hash_reset(vcpu);
9208 vcpu->arch.apf.halted = false;
9210 if (kvm_mpx_supported()) {
9211 void *mpx_state_buffer;
9214 * To avoid have the INIT path from kvm_apic_has_events() that be
9215 * called with loaded FPU and does not let userspace fix the state.
9218 kvm_put_guest_fpu(vcpu);
9219 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9221 if (mpx_state_buffer)
9222 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9223 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9225 if (mpx_state_buffer)
9226 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9228 kvm_load_guest_fpu(vcpu);
9232 kvm_pmu_reset(vcpu);
9233 vcpu->arch.smbase = 0x30000;
9235 vcpu->arch.msr_misc_features_enables = 0;
9237 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9240 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9241 vcpu->arch.regs_avail = ~0;
9242 vcpu->arch.regs_dirty = ~0;
9244 vcpu->arch.ia32_xss = 0;
9246 kvm_x86_ops->vcpu_reset(vcpu, init_event);
9249 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9251 struct kvm_segment cs;
9253 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9254 cs.selector = vector << 8;
9255 cs.base = vector << 12;
9256 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9257 kvm_rip_write(vcpu, 0);
9260 int kvm_arch_hardware_enable(void)
9263 struct kvm_vcpu *vcpu;
9268 bool stable, backwards_tsc = false;
9270 kvm_shared_msr_cpu_online();
9271 ret = kvm_x86_ops->hardware_enable();
9275 local_tsc = rdtsc();
9276 stable = !kvm_check_tsc_unstable();
9277 list_for_each_entry(kvm, &vm_list, vm_list) {
9278 kvm_for_each_vcpu(i, vcpu, kvm) {
9279 if (!stable && vcpu->cpu == smp_processor_id())
9280 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9281 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9282 backwards_tsc = true;
9283 if (vcpu->arch.last_host_tsc > max_tsc)
9284 max_tsc = vcpu->arch.last_host_tsc;
9290 * Sometimes, even reliable TSCs go backwards. This happens on
9291 * platforms that reset TSC during suspend or hibernate actions, but
9292 * maintain synchronization. We must compensate. Fortunately, we can
9293 * detect that condition here, which happens early in CPU bringup,
9294 * before any KVM threads can be running. Unfortunately, we can't
9295 * bring the TSCs fully up to date with real time, as we aren't yet far
9296 * enough into CPU bringup that we know how much real time has actually
9297 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9298 * variables that haven't been updated yet.
9300 * So we simply find the maximum observed TSC above, then record the
9301 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9302 * the adjustment will be applied. Note that we accumulate
9303 * adjustments, in case multiple suspend cycles happen before some VCPU
9304 * gets a chance to run again. In the event that no KVM threads get a
9305 * chance to run, we will miss the entire elapsed period, as we'll have
9306 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9307 * loose cycle time. This isn't too big a deal, since the loss will be
9308 * uniform across all VCPUs (not to mention the scenario is extremely
9309 * unlikely). It is possible that a second hibernate recovery happens
9310 * much faster than a first, causing the observed TSC here to be
9311 * smaller; this would require additional padding adjustment, which is
9312 * why we set last_host_tsc to the local tsc observed here.
9314 * N.B. - this code below runs only on platforms with reliable TSC,
9315 * as that is the only way backwards_tsc is set above. Also note
9316 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9317 * have the same delta_cyc adjustment applied if backwards_tsc
9318 * is detected. Note further, this adjustment is only done once,
9319 * as we reset last_host_tsc on all VCPUs to stop this from being
9320 * called multiple times (one for each physical CPU bringup).
9322 * Platforms with unreliable TSCs don't have to deal with this, they
9323 * will be compensated by the logic in vcpu_load, which sets the TSC to
9324 * catchup mode. This will catchup all VCPUs to real time, but cannot
9325 * guarantee that they stay in perfect synchronization.
9327 if (backwards_tsc) {
9328 u64 delta_cyc = max_tsc - local_tsc;
9329 list_for_each_entry(kvm, &vm_list, vm_list) {
9330 kvm->arch.backwards_tsc_observed = true;
9331 kvm_for_each_vcpu(i, vcpu, kvm) {
9332 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9333 vcpu->arch.last_host_tsc = local_tsc;
9334 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9338 * We have to disable TSC offset matching.. if you were
9339 * booting a VM while issuing an S4 host suspend....
9340 * you may have some problem. Solving this issue is
9341 * left as an exercise to the reader.
9343 kvm->arch.last_tsc_nsec = 0;
9344 kvm->arch.last_tsc_write = 0;
9351 void kvm_arch_hardware_disable(void)
9353 kvm_x86_ops->hardware_disable();
9354 drop_user_return_notifiers();
9357 int kvm_arch_hardware_setup(void)
9361 r = kvm_x86_ops->hardware_setup();
9365 if (kvm_has_tsc_control) {
9367 * Make sure the user can only configure tsc_khz values that
9368 * fit into a signed integer.
9369 * A min value is not calculated because it will always
9370 * be 1 on all machines.
9372 u64 max = min(0x7fffffffULL,
9373 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9374 kvm_max_guest_tsc_khz = max;
9376 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9379 if (boot_cpu_has(X86_FEATURE_XSAVES))
9380 rdmsrl(MSR_IA32_XSS, host_xss);
9382 kvm_init_msr_list();
9386 void kvm_arch_hardware_unsetup(void)
9388 kvm_x86_ops->hardware_unsetup();
9391 int kvm_arch_check_processor_compat(void)
9393 return kvm_x86_ops->check_processor_compatibility();
9396 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9398 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9400 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9402 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9404 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9407 struct static_key kvm_no_apic_vcpu __read_mostly;
9408 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9410 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9415 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9416 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9417 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9419 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9421 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9426 vcpu->arch.pio_data = page_address(page);
9428 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9430 r = kvm_mmu_create(vcpu);
9432 goto fail_free_pio_data;
9434 if (irqchip_in_kernel(vcpu->kvm)) {
9435 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu->kvm);
9436 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9438 goto fail_mmu_destroy;
9440 static_key_slow_inc(&kvm_no_apic_vcpu);
9442 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9443 GFP_KERNEL_ACCOUNT);
9444 if (!vcpu->arch.mce_banks) {
9446 goto fail_free_lapic;
9448 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9450 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9451 GFP_KERNEL_ACCOUNT)) {
9453 goto fail_free_mce_banks;
9458 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9460 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9462 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9464 kvm_async_pf_hash_reset(vcpu);
9467 vcpu->arch.pending_external_vector = -1;
9468 vcpu->arch.preempted_in_kernel = false;
9470 kvm_hv_vcpu_init(vcpu);
9474 fail_free_mce_banks:
9475 kfree(vcpu->arch.mce_banks);
9477 kvm_free_lapic(vcpu);
9479 kvm_mmu_destroy(vcpu);
9481 free_page((unsigned long)vcpu->arch.pio_data);
9486 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9490 kvm_hv_vcpu_uninit(vcpu);
9491 kvm_pmu_destroy(vcpu);
9492 kfree(vcpu->arch.mce_banks);
9493 kvm_free_lapic(vcpu);
9494 idx = srcu_read_lock(&vcpu->kvm->srcu);
9495 kvm_mmu_destroy(vcpu);
9496 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9497 free_page((unsigned long)vcpu->arch.pio_data);
9498 if (!lapic_in_kernel(vcpu))
9499 static_key_slow_dec(&kvm_no_apic_vcpu);
9502 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9504 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9506 vcpu->arch.l1tf_flush_l1d = true;
9507 if (pmu->version && unlikely(pmu->event_count)) {
9508 pmu->need_cleanup = true;
9509 kvm_make_request(KVM_REQ_PMU, vcpu);
9511 kvm_x86_ops->sched_in(vcpu, cpu);
9514 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9519 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9520 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9521 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9522 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9523 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9524 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9526 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9527 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9528 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9529 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9530 &kvm->arch.irq_sources_bitmap);
9532 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9533 mutex_init(&kvm->arch.apic_map_lock);
9534 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9536 kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9537 pvclock_update_vm_gtod_copy(kvm);
9539 kvm->arch.guest_can_read_msr_platform_info = true;
9541 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9542 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9544 kvm_hv_init_vm(kvm);
9545 kvm_page_track_init(kvm);
9546 kvm_mmu_init_vm(kvm);
9548 return kvm_x86_ops->vm_init(kvm);
9551 int kvm_arch_post_init_vm(struct kvm *kvm)
9553 return kvm_mmu_post_init_vm(kvm);
9556 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9559 kvm_mmu_unload(vcpu);
9563 static void kvm_free_vcpus(struct kvm *kvm)
9566 struct kvm_vcpu *vcpu;
9569 * Unpin any mmu pages first.
9571 kvm_for_each_vcpu(i, vcpu, kvm) {
9572 kvm_clear_async_pf_completion_queue(vcpu);
9573 kvm_unload_vcpu_mmu(vcpu);
9575 kvm_for_each_vcpu(i, vcpu, kvm)
9576 kvm_arch_vcpu_free(vcpu);
9578 mutex_lock(&kvm->lock);
9579 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9580 kvm->vcpus[i] = NULL;
9582 atomic_set(&kvm->online_vcpus, 0);
9583 mutex_unlock(&kvm->lock);
9586 void kvm_arch_sync_events(struct kvm *kvm)
9588 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9589 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9593 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9597 struct kvm_memslots *slots = kvm_memslots(kvm);
9598 struct kvm_memory_slot *slot, old;
9600 /* Called with kvm->slots_lock held. */
9601 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9604 slot = id_to_memslot(slots, id);
9610 * MAP_SHARED to prevent internal slot pages from being moved
9613 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9614 MAP_SHARED | MAP_ANONYMOUS, 0);
9615 if (IS_ERR((void *)hva))
9616 return PTR_ERR((void *)hva);
9625 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9626 struct kvm_userspace_memory_region m;
9628 m.slot = id | (i << 16);
9630 m.guest_phys_addr = gpa;
9631 m.userspace_addr = hva;
9632 m.memory_size = size;
9633 r = __kvm_set_memory_region(kvm, &m);
9639 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9643 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9645 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9649 mutex_lock(&kvm->slots_lock);
9650 r = __x86_set_memory_region(kvm, id, gpa, size);
9651 mutex_unlock(&kvm->slots_lock);
9655 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9657 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9659 kvm_mmu_pre_destroy_vm(kvm);
9662 void kvm_arch_destroy_vm(struct kvm *kvm)
9664 if (current->mm == kvm->mm) {
9666 * Free memory regions allocated on behalf of userspace,
9667 * unless the the memory map has changed due to process exit
9670 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9671 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9672 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9674 if (kvm_x86_ops->vm_destroy)
9675 kvm_x86_ops->vm_destroy(kvm);
9676 kvm_pic_destroy(kvm);
9677 kvm_ioapic_destroy(kvm);
9678 kvm_free_vcpus(kvm);
9679 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9680 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9681 kvm_mmu_uninit_vm(kvm);
9682 kvm_page_track_cleanup(kvm);
9683 kvm_hv_destroy_vm(kvm);
9686 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9687 struct kvm_memory_slot *dont)
9691 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9692 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9693 kvfree(free->arch.rmap[i]);
9694 free->arch.rmap[i] = NULL;
9699 if (!dont || free->arch.lpage_info[i - 1] !=
9700 dont->arch.lpage_info[i - 1]) {
9701 kvfree(free->arch.lpage_info[i - 1]);
9702 free->arch.lpage_info[i - 1] = NULL;
9706 kvm_page_track_free_memslot(free, dont);
9709 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9710 unsigned long npages)
9714 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9715 struct kvm_lpage_info *linfo;
9720 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9721 slot->base_gfn, level) + 1;
9723 slot->arch.rmap[i] =
9724 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9725 GFP_KERNEL_ACCOUNT);
9726 if (!slot->arch.rmap[i])
9731 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9735 slot->arch.lpage_info[i - 1] = linfo;
9737 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9738 linfo[0].disallow_lpage = 1;
9739 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9740 linfo[lpages - 1].disallow_lpage = 1;
9741 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9743 * If the gfn and userspace address are not aligned wrt each
9744 * other, or if explicitly asked to, disable large page
9745 * support for this slot
9747 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9748 !kvm_largepages_enabled()) {
9751 for (j = 0; j < lpages; ++j)
9752 linfo[j].disallow_lpage = 1;
9756 if (kvm_page_track_create_memslot(slot, npages))
9762 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9763 kvfree(slot->arch.rmap[i]);
9764 slot->arch.rmap[i] = NULL;
9768 kvfree(slot->arch.lpage_info[i - 1]);
9769 slot->arch.lpage_info[i - 1] = NULL;
9774 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9776 struct kvm_vcpu *vcpu;
9780 * memslots->generation has been incremented.
9781 * mmio generation may have reached its maximum value.
9783 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9785 /* Force re-initialization of steal_time cache */
9786 kvm_for_each_vcpu(i, vcpu, kvm)
9787 kvm_vcpu_kick(vcpu);
9790 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9791 struct kvm_memory_slot *memslot,
9792 const struct kvm_userspace_memory_region *mem,
9793 enum kvm_mr_change change)
9798 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9799 struct kvm_memory_slot *new)
9801 /* Still write protect RO slot */
9802 if (new->flags & KVM_MEM_READONLY) {
9803 kvm_mmu_slot_remove_write_access(kvm, new);
9808 * Call kvm_x86_ops dirty logging hooks when they are valid.
9810 * kvm_x86_ops->slot_disable_log_dirty is called when:
9812 * - KVM_MR_CREATE with dirty logging is disabled
9813 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9815 * The reason is, in case of PML, we need to set D-bit for any slots
9816 * with dirty logging disabled in order to eliminate unnecessary GPA
9817 * logging in PML buffer (and potential PML buffer full VMEXT). This
9818 * guarantees leaving PML enabled during guest's lifetime won't have
9819 * any additional overhead from PML when guest is running with dirty
9820 * logging disabled for memory slots.
9822 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9823 * to dirty logging mode.
9825 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9827 * In case of write protect:
9829 * Write protect all pages for dirty logging.
9831 * All the sptes including the large sptes which point to this
9832 * slot are set to readonly. We can not create any new large
9833 * spte on this slot until the end of the logging.
9835 * See the comments in fast_page_fault().
9837 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9838 if (kvm_x86_ops->slot_enable_log_dirty)
9839 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9841 kvm_mmu_slot_remove_write_access(kvm, new);
9843 if (kvm_x86_ops->slot_disable_log_dirty)
9844 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9848 void kvm_arch_commit_memory_region(struct kvm *kvm,
9849 const struct kvm_userspace_memory_region *mem,
9850 const struct kvm_memory_slot *old,
9851 const struct kvm_memory_slot *new,
9852 enum kvm_mr_change change)
9854 if (!kvm->arch.n_requested_mmu_pages)
9855 kvm_mmu_change_mmu_pages(kvm,
9856 kvm_mmu_calculate_default_mmu_pages(kvm));
9859 * Dirty logging tracks sptes in 4k granularity, meaning that large
9860 * sptes have to be split. If live migration is successful, the guest
9861 * in the source machine will be destroyed and large sptes will be
9862 * created in the destination. However, if the guest continues to run
9863 * in the source machine (for example if live migration fails), small
9864 * sptes will remain around and cause bad performance.
9866 * Scan sptes if dirty logging has been stopped, dropping those
9867 * which can be collapsed into a single large-page spte. Later
9868 * page faults will create the large-page sptes.
9870 * There is no need to do this in any of the following cases:
9871 * CREATE: No dirty mappings will already exist.
9872 * MOVE/DELETE: The old mappings will already have been cleaned up by
9873 * kvm_arch_flush_shadow_memslot()
9875 if (change == KVM_MR_FLAGS_ONLY &&
9876 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9877 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9878 kvm_mmu_zap_collapsible_sptes(kvm, new);
9881 * Set up write protection and/or dirty logging for the new slot.
9883 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9884 * been zapped so no dirty logging staff is needed for old slot. For
9885 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9886 * new and it's also covered when dealing with the new slot.
9888 * FIXME: const-ify all uses of struct kvm_memory_slot.
9890 if (change != KVM_MR_DELETE)
9891 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9894 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9896 kvm_mmu_zap_all(kvm);
9899 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9900 struct kvm_memory_slot *slot)
9902 kvm_page_track_flush_slot(kvm, slot);
9905 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9907 return (is_guest_mode(vcpu) &&
9908 kvm_x86_ops->guest_apic_has_interrupt &&
9909 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9912 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9914 if (!list_empty_careful(&vcpu->async_pf.done))
9917 if (kvm_apic_has_events(vcpu))
9920 if (vcpu->arch.pv.pv_unhalted)
9923 if (vcpu->arch.exception.pending)
9926 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9927 (vcpu->arch.nmi_pending &&
9928 kvm_x86_ops->nmi_allowed(vcpu)))
9931 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9932 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9935 if (kvm_arch_interrupt_allowed(vcpu) &&
9936 (kvm_cpu_has_interrupt(vcpu) ||
9937 kvm_guest_apic_has_interrupt(vcpu)))
9940 if (kvm_hv_has_stimer_pending(vcpu))
9946 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9948 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9951 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9953 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9956 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9957 kvm_test_request(KVM_REQ_SMI, vcpu) ||
9958 kvm_test_request(KVM_REQ_EVENT, vcpu))
9961 if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9967 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9969 return vcpu->arch.preempted_in_kernel;
9972 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9974 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9977 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9979 return kvm_x86_ops->interrupt_allowed(vcpu);
9982 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9984 if (is_64_bit_mode(vcpu))
9985 return kvm_rip_read(vcpu);
9986 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9987 kvm_rip_read(vcpu));
9989 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9991 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9993 return kvm_get_linear_rip(vcpu) == linear_rip;
9995 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9997 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9999 unsigned long rflags;
10001 rflags = kvm_x86_ops->get_rflags(vcpu);
10002 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10003 rflags &= ~X86_EFLAGS_TF;
10006 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10008 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10010 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10011 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10012 rflags |= X86_EFLAGS_TF;
10013 kvm_x86_ops->set_rflags(vcpu, rflags);
10016 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10018 __kvm_set_rflags(vcpu, rflags);
10019 kvm_make_request(KVM_REQ_EVENT, vcpu);
10021 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10023 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10027 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10031 r = kvm_mmu_reload(vcpu);
10035 if (!vcpu->arch.mmu->direct_map &&
10036 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
10039 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
10042 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10044 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10047 static inline u32 kvm_async_pf_next_probe(u32 key)
10049 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
10052 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10054 u32 key = kvm_async_pf_hash_fn(gfn);
10056 while (vcpu->arch.apf.gfns[key] != ~0)
10057 key = kvm_async_pf_next_probe(key);
10059 vcpu->arch.apf.gfns[key] = gfn;
10062 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10065 u32 key = kvm_async_pf_hash_fn(gfn);
10067 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
10068 (vcpu->arch.apf.gfns[key] != gfn &&
10069 vcpu->arch.apf.gfns[key] != ~0); i++)
10070 key = kvm_async_pf_next_probe(key);
10075 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10077 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10080 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10084 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10086 vcpu->arch.apf.gfns[i] = ~0;
10088 j = kvm_async_pf_next_probe(j);
10089 if (vcpu->arch.apf.gfns[j] == ~0)
10091 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10093 * k lies cyclically in ]i,j]
10095 * |....j i.k.| or |.k..j i...|
10097 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10098 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10103 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10106 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10110 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10113 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10117 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10119 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10122 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10123 (vcpu->arch.apf.send_user_only &&
10124 kvm_x86_ops->get_cpl(vcpu) == 0))
10130 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10132 if (unlikely(!lapic_in_kernel(vcpu) ||
10133 kvm_event_needs_reinjection(vcpu) ||
10134 vcpu->arch.exception.pending))
10137 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10141 * If interrupts are off we cannot even use an artificial
10144 return kvm_x86_ops->interrupt_allowed(vcpu);
10147 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10148 struct kvm_async_pf *work)
10150 struct x86_exception fault;
10152 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
10153 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10155 if (kvm_can_deliver_async_pf(vcpu) &&
10156 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10157 fault.vector = PF_VECTOR;
10158 fault.error_code_valid = true;
10159 fault.error_code = 0;
10160 fault.nested_page_fault = false;
10161 fault.address = work->arch.token;
10162 fault.async_page_fault = true;
10163 kvm_inject_page_fault(vcpu, &fault);
10166 * It is not possible to deliver a paravirtualized asynchronous
10167 * page fault, but putting the guest in an artificial halt state
10168 * can be beneficial nevertheless: if an interrupt arrives, we
10169 * can deliver it timely and perhaps the guest will schedule
10170 * another process. When the instruction that triggered a page
10171 * fault is retried, hopefully the page will be ready in the host.
10173 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10177 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10178 struct kvm_async_pf *work)
10180 struct x86_exception fault;
10183 if (work->wakeup_all)
10184 work->arch.token = ~0; /* broadcast wakeup */
10186 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10187 trace_kvm_async_pf_ready(work->arch.token, work->gva);
10189 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10190 !apf_get_user(vcpu, &val)) {
10191 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10192 vcpu->arch.exception.pending &&
10193 vcpu->arch.exception.nr == PF_VECTOR &&
10194 !apf_put_user(vcpu, 0)) {
10195 vcpu->arch.exception.injected = false;
10196 vcpu->arch.exception.pending = false;
10197 vcpu->arch.exception.nr = 0;
10198 vcpu->arch.exception.has_error_code = false;
10199 vcpu->arch.exception.error_code = 0;
10200 vcpu->arch.exception.has_payload = false;
10201 vcpu->arch.exception.payload = 0;
10202 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10203 fault.vector = PF_VECTOR;
10204 fault.error_code_valid = true;
10205 fault.error_code = 0;
10206 fault.nested_page_fault = false;
10207 fault.address = work->arch.token;
10208 fault.async_page_fault = true;
10209 kvm_inject_page_fault(vcpu, &fault);
10212 vcpu->arch.apf.halted = false;
10213 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10216 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10218 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10221 return kvm_can_do_async_pf(vcpu);
10224 void kvm_arch_start_assignment(struct kvm *kvm)
10226 atomic_inc(&kvm->arch.assigned_device_count);
10228 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10230 void kvm_arch_end_assignment(struct kvm *kvm)
10232 atomic_dec(&kvm->arch.assigned_device_count);
10234 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10236 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10238 return atomic_read(&kvm->arch.assigned_device_count);
10240 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10242 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10244 atomic_inc(&kvm->arch.noncoherent_dma_count);
10246 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10248 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10250 atomic_dec(&kvm->arch.noncoherent_dma_count);
10252 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10254 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10256 return atomic_read(&kvm->arch.noncoherent_dma_count);
10258 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10260 bool kvm_arch_has_irq_bypass(void)
10265 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10266 struct irq_bypass_producer *prod)
10268 struct kvm_kernel_irqfd *irqfd =
10269 container_of(cons, struct kvm_kernel_irqfd, consumer);
10271 irqfd->producer = prod;
10273 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10274 prod->irq, irqfd->gsi, 1);
10277 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10278 struct irq_bypass_producer *prod)
10281 struct kvm_kernel_irqfd *irqfd =
10282 container_of(cons, struct kvm_kernel_irqfd, consumer);
10284 WARN_ON(irqfd->producer != prod);
10285 irqfd->producer = NULL;
10288 * When producer of consumer is unregistered, we change back to
10289 * remapped mode, so we can re-use the current implementation
10290 * when the irq is masked/disabled or the consumer side (KVM
10291 * int this case doesn't want to receive the interrupts.
10293 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10295 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10296 " fails: %d\n", irqfd->consumer.token, ret);
10299 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10300 uint32_t guest_irq, bool set)
10302 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10305 bool kvm_vector_hashing_enabled(void)
10307 return vector_hashing;
10309 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10311 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10313 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10315 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10318 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10319 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10320 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10321 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10322 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10323 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10324 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10325 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10326 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10327 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10328 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10329 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10330 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10331 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10332 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10333 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10334 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10335 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10336 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10337 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);