2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
59 #include <trace/events/kvm.h>
61 #include <asm/debugreg.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
73 #define CREATE_TRACE_POINTS
76 #define MAX_IO_MSRS 256
77 #define KVM_MAX_MCE_BANKS 32
78 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
81 #define emul_to_vcpu(ctxt) \
82 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
85 * - enable syscall per default because its emulated by KVM
86 * - enable LME and LMA per default on 64 bit KVM
90 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
92 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
95 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
96 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
98 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
101 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
102 static void process_nmi(struct kvm_vcpu *vcpu);
103 static void enter_smm(struct kvm_vcpu *vcpu);
104 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
105 static void store_regs(struct kvm_vcpu *vcpu);
106 static int sync_regs(struct kvm_vcpu *vcpu);
108 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
109 EXPORT_SYMBOL_GPL(kvm_x86_ops);
111 static bool __read_mostly ignore_msrs = 0;
112 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
114 static bool __read_mostly report_ignored_msrs = true;
115 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
117 unsigned int min_timer_period_us = 200;
118 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
120 static bool __read_mostly kvmclock_periodic_sync = true;
121 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
123 bool __read_mostly kvm_has_tsc_control;
124 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
125 u32 __read_mostly kvm_max_guest_tsc_khz;
126 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
127 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
128 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
129 u64 __read_mostly kvm_max_tsc_scaling_ratio;
130 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
131 u64 __read_mostly kvm_default_tsc_scaling_ratio;
132 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
134 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
135 static u32 __read_mostly tsc_tolerance_ppm = 250;
136 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
138 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
139 unsigned int __read_mostly lapic_timer_advance_ns = 1000;
140 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
141 EXPORT_SYMBOL_GPL(lapic_timer_advance_ns);
143 static bool __read_mostly vector_hashing = true;
144 module_param(vector_hashing, bool, S_IRUGO);
146 bool __read_mostly enable_vmware_backdoor = false;
147 module_param(enable_vmware_backdoor, bool, S_IRUGO);
148 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
150 static bool __read_mostly force_emulation_prefix = false;
151 module_param(force_emulation_prefix, bool, S_IRUGO);
153 #define KVM_NR_SHARED_MSRS 16
155 struct kvm_shared_msrs_global {
157 u32 msrs[KVM_NR_SHARED_MSRS];
160 struct kvm_shared_msrs {
161 struct user_return_notifier urn;
163 struct kvm_shared_msr_values {
166 } values[KVM_NR_SHARED_MSRS];
169 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
170 static struct kvm_shared_msrs __percpu *shared_msrs;
172 struct kvm_stats_debugfs_item debugfs_entries[] = {
173 { "pf_fixed", VCPU_STAT(pf_fixed) },
174 { "pf_guest", VCPU_STAT(pf_guest) },
175 { "tlb_flush", VCPU_STAT(tlb_flush) },
176 { "invlpg", VCPU_STAT(invlpg) },
177 { "exits", VCPU_STAT(exits) },
178 { "io_exits", VCPU_STAT(io_exits) },
179 { "mmio_exits", VCPU_STAT(mmio_exits) },
180 { "signal_exits", VCPU_STAT(signal_exits) },
181 { "irq_window", VCPU_STAT(irq_window_exits) },
182 { "nmi_window", VCPU_STAT(nmi_window_exits) },
183 { "halt_exits", VCPU_STAT(halt_exits) },
184 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
185 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
186 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
187 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
188 { "hypercalls", VCPU_STAT(hypercalls) },
189 { "request_irq", VCPU_STAT(request_irq_exits) },
190 { "irq_exits", VCPU_STAT(irq_exits) },
191 { "host_state_reload", VCPU_STAT(host_state_reload) },
192 { "fpu_reload", VCPU_STAT(fpu_reload) },
193 { "insn_emulation", VCPU_STAT(insn_emulation) },
194 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
195 { "irq_injections", VCPU_STAT(irq_injections) },
196 { "nmi_injections", VCPU_STAT(nmi_injections) },
197 { "req_event", VCPU_STAT(req_event) },
198 { "l1d_flush", VCPU_STAT(l1d_flush) },
199 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
200 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
201 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
202 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
203 { "mmu_flooded", VM_STAT(mmu_flooded) },
204 { "mmu_recycled", VM_STAT(mmu_recycled) },
205 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
206 { "mmu_unsync", VM_STAT(mmu_unsync) },
207 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
208 { "largepages", VM_STAT(lpages) },
209 { "max_mmu_page_hash_collisions",
210 VM_STAT(max_mmu_page_hash_collisions) },
214 u64 __read_mostly host_xcr0;
216 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
218 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
221 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
222 vcpu->arch.apf.gfns[i] = ~0;
225 static void kvm_on_user_return(struct user_return_notifier *urn)
228 struct kvm_shared_msrs *locals
229 = container_of(urn, struct kvm_shared_msrs, urn);
230 struct kvm_shared_msr_values *values;
234 * Disabling irqs at this point since the following code could be
235 * interrupted and executed through kvm_arch_hardware_disable()
237 local_irq_save(flags);
238 if (locals->registered) {
239 locals->registered = false;
240 user_return_notifier_unregister(urn);
242 local_irq_restore(flags);
243 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
244 values = &locals->values[slot];
245 if (values->host != values->curr) {
246 wrmsrl(shared_msrs_global.msrs[slot], values->host);
247 values->curr = values->host;
252 static void shared_msr_update(unsigned slot, u32 msr)
255 unsigned int cpu = smp_processor_id();
256 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
258 /* only read, and nobody should modify it at this time,
259 * so don't need lock */
260 if (slot >= shared_msrs_global.nr) {
261 printk(KERN_ERR "kvm: invalid MSR slot!");
264 rdmsrl_safe(msr, &value);
265 smsr->values[slot].host = value;
266 smsr->values[slot].curr = value;
269 void kvm_define_shared_msr(unsigned slot, u32 msr)
271 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
272 shared_msrs_global.msrs[slot] = msr;
273 if (slot >= shared_msrs_global.nr)
274 shared_msrs_global.nr = slot + 1;
276 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
278 static void kvm_shared_msr_cpu_online(void)
282 for (i = 0; i < shared_msrs_global.nr; ++i)
283 shared_msr_update(i, shared_msrs_global.msrs[i]);
286 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
288 unsigned int cpu = smp_processor_id();
289 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
292 if (((value ^ smsr->values[slot].curr) & mask) == 0)
294 smsr->values[slot].curr = value;
295 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
299 if (!smsr->registered) {
300 smsr->urn.on_user_return = kvm_on_user_return;
301 user_return_notifier_register(&smsr->urn);
302 smsr->registered = true;
306 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
308 static void drop_user_return_notifiers(void)
310 unsigned int cpu = smp_processor_id();
311 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
313 if (smsr->registered)
314 kvm_on_user_return(&smsr->urn);
317 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
319 return vcpu->arch.apic_base;
321 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
323 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
325 return kvm_apic_mode(kvm_get_apic_base(vcpu));
327 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
329 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
331 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
332 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
333 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
334 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
336 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
338 if (!msr_info->host_initiated) {
339 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
341 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
345 kvm_lapic_set_base(vcpu, msr_info->data);
348 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
350 asmlinkage __visible void kvm_spurious_fault(void)
352 /* Fault while not rebooting. We want the trace. */
355 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
357 #define EXCPT_BENIGN 0
358 #define EXCPT_CONTRIBUTORY 1
361 static int exception_class(int vector)
371 return EXCPT_CONTRIBUTORY;
378 #define EXCPT_FAULT 0
380 #define EXCPT_ABORT 2
381 #define EXCPT_INTERRUPT 3
383 static int exception_type(int vector)
387 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
388 return EXCPT_INTERRUPT;
392 /* #DB is trap, as instruction watchpoints are handled elsewhere */
393 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
396 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
399 /* Reserved exceptions will result in fault */
403 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
405 unsigned nr = vcpu->arch.exception.nr;
406 bool has_payload = vcpu->arch.exception.has_payload;
407 unsigned long payload = vcpu->arch.exception.payload;
415 * "Certain debug exceptions may clear bit 0-3. The
416 * remaining contents of the DR6 register are never
417 * cleared by the processor".
419 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
421 * DR6.RTM is set by all #DB exceptions that don't clear it.
423 vcpu->arch.dr6 |= DR6_RTM;
424 vcpu->arch.dr6 |= payload;
426 * Bit 16 should be set in the payload whenever the #DB
427 * exception should clear DR6.RTM. This makes the payload
428 * compatible with the pending debug exceptions under VMX.
429 * Though not currently documented in the SDM, this also
430 * makes the payload compatible with the exit qualification
431 * for #DB exceptions under VMX.
433 vcpu->arch.dr6 ^= payload & DR6_RTM;
436 vcpu->arch.cr2 = payload;
440 vcpu->arch.exception.has_payload = false;
441 vcpu->arch.exception.payload = 0;
443 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
445 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
446 unsigned nr, bool has_error, u32 error_code,
447 bool has_payload, unsigned long payload, bool reinject)
452 kvm_make_request(KVM_REQ_EVENT, vcpu);
454 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
456 if (has_error && !is_protmode(vcpu))
460 * On vmentry, vcpu->arch.exception.pending is only
461 * true if an event injection was blocked by
462 * nested_run_pending. In that case, however,
463 * vcpu_enter_guest requests an immediate exit,
464 * and the guest shouldn't proceed far enough to
467 WARN_ON_ONCE(vcpu->arch.exception.pending);
468 vcpu->arch.exception.injected = true;
469 if (WARN_ON_ONCE(has_payload)) {
471 * A reinjected event has already
472 * delivered its payload.
478 vcpu->arch.exception.pending = true;
479 vcpu->arch.exception.injected = false;
481 vcpu->arch.exception.has_error_code = has_error;
482 vcpu->arch.exception.nr = nr;
483 vcpu->arch.exception.error_code = error_code;
484 vcpu->arch.exception.has_payload = has_payload;
485 vcpu->arch.exception.payload = payload;
487 * In guest mode, payload delivery should be deferred,
488 * so that the L1 hypervisor can intercept #PF before
489 * CR2 is modified (or intercept #DB before DR6 is
490 * modified under nVMX). However, for ABI
491 * compatibility with KVM_GET_VCPU_EVENTS and
492 * KVM_SET_VCPU_EVENTS, we can't delay payload
493 * delivery unless userspace has enabled this
494 * functionality via the per-VM capability,
495 * KVM_CAP_EXCEPTION_PAYLOAD.
497 if (!vcpu->kvm->arch.exception_payload_enabled ||
498 !is_guest_mode(vcpu))
499 kvm_deliver_exception_payload(vcpu);
503 /* to check exception */
504 prev_nr = vcpu->arch.exception.nr;
505 if (prev_nr == DF_VECTOR) {
506 /* triple fault -> shutdown */
507 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
510 class1 = exception_class(prev_nr);
511 class2 = exception_class(nr);
512 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
513 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
515 * Generate double fault per SDM Table 5-5. Set
516 * exception.pending = true so that the double fault
517 * can trigger a nested vmexit.
519 vcpu->arch.exception.pending = true;
520 vcpu->arch.exception.injected = false;
521 vcpu->arch.exception.has_error_code = true;
522 vcpu->arch.exception.nr = DF_VECTOR;
523 vcpu->arch.exception.error_code = 0;
524 vcpu->arch.exception.has_payload = false;
525 vcpu->arch.exception.payload = 0;
527 /* replace previous exception with a new one in a hope
528 that instruction re-execution will regenerate lost
533 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
535 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
537 EXPORT_SYMBOL_GPL(kvm_queue_exception);
539 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
541 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
543 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
545 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
546 unsigned long payload)
548 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
551 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
552 u32 error_code, unsigned long payload)
554 kvm_multiple_exception(vcpu, nr, true, error_code,
555 true, payload, false);
558 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
561 kvm_inject_gp(vcpu, 0);
563 return kvm_skip_emulated_instruction(vcpu);
567 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
569 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
571 ++vcpu->stat.pf_guest;
572 vcpu->arch.exception.nested_apf =
573 is_guest_mode(vcpu) && fault->async_page_fault;
574 if (vcpu->arch.exception.nested_apf) {
575 vcpu->arch.apf.nested_apf_token = fault->address;
576 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
578 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
582 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
584 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
586 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
587 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
589 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
591 return fault->nested_page_fault;
594 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
596 atomic_inc(&vcpu->arch.nmi_queued);
597 kvm_make_request(KVM_REQ_NMI, vcpu);
599 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
601 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
603 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
605 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
607 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
609 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
611 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
614 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
615 * a #GP and return false.
617 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
619 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
621 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
624 EXPORT_SYMBOL_GPL(kvm_require_cpl);
626 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
628 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
631 kvm_queue_exception(vcpu, UD_VECTOR);
634 EXPORT_SYMBOL_GPL(kvm_require_dr);
637 * This function will be used to read from the physical memory of the currently
638 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
639 * can read from guest physical or from the guest's guest physical memory.
641 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
642 gfn_t ngfn, void *data, int offset, int len,
645 struct x86_exception exception;
649 ngpa = gfn_to_gpa(ngfn);
650 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
651 if (real_gfn == UNMAPPED_GVA)
654 real_gfn = gpa_to_gfn(real_gfn);
656 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
658 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
660 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
661 void *data, int offset, int len, u32 access)
663 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
664 data, offset, len, access);
668 * Load the pae pdptrs. Return true is they are all valid.
670 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
672 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
673 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
676 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
678 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
679 offset * sizeof(u64), sizeof(pdpte),
680 PFERR_USER_MASK|PFERR_WRITE_MASK);
685 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
686 if ((pdpte[i] & PT_PRESENT_MASK) &&
688 vcpu->arch.mmu->guest_rsvd_check.rsvd_bits_mask[0][2])) {
695 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
696 __set_bit(VCPU_EXREG_PDPTR,
697 (unsigned long *)&vcpu->arch.regs_avail);
698 __set_bit(VCPU_EXREG_PDPTR,
699 (unsigned long *)&vcpu->arch.regs_dirty);
704 EXPORT_SYMBOL_GPL(load_pdptrs);
706 bool pdptrs_changed(struct kvm_vcpu *vcpu)
708 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
714 if (is_long_mode(vcpu) || !is_pae(vcpu) || !is_paging(vcpu))
717 if (!test_bit(VCPU_EXREG_PDPTR,
718 (unsigned long *)&vcpu->arch.regs_avail))
721 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
722 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
723 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
724 PFERR_USER_MASK | PFERR_WRITE_MASK);
727 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
732 EXPORT_SYMBOL_GPL(pdptrs_changed);
734 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
736 unsigned long old_cr0 = kvm_read_cr0(vcpu);
737 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
742 if (cr0 & 0xffffffff00000000UL)
746 cr0 &= ~CR0_RESERVED_BITS;
748 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
751 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
754 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
756 if ((vcpu->arch.efer & EFER_LME)) {
761 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
766 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
771 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
774 kvm_x86_ops->set_cr0(vcpu, cr0);
776 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
777 kvm_clear_async_pf_completion_queue(vcpu);
778 kvm_async_pf_hash_reset(vcpu);
781 if ((cr0 ^ old_cr0) & update_bits)
782 kvm_mmu_reset_context(vcpu);
784 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
785 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
786 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
787 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
791 EXPORT_SYMBOL_GPL(kvm_set_cr0);
793 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
795 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
797 EXPORT_SYMBOL_GPL(kvm_lmsw);
799 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
801 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
802 !vcpu->guest_xcr0_loaded) {
803 /* kvm_set_xcr() also depends on this */
804 if (vcpu->arch.xcr0 != host_xcr0)
805 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
806 vcpu->guest_xcr0_loaded = 1;
810 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
812 if (vcpu->guest_xcr0_loaded) {
813 if (vcpu->arch.xcr0 != host_xcr0)
814 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
815 vcpu->guest_xcr0_loaded = 0;
819 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
822 u64 old_xcr0 = vcpu->arch.xcr0;
825 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
826 if (index != XCR_XFEATURE_ENABLED_MASK)
828 if (!(xcr0 & XFEATURE_MASK_FP))
830 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
834 * Do not allow the guest to set bits that we do not support
835 * saving. However, xcr0 bit 0 is always set, even if the
836 * emulated CPU does not support XSAVE (see fx_init).
838 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
839 if (xcr0 & ~valid_bits)
842 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
843 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
846 if (xcr0 & XFEATURE_MASK_AVX512) {
847 if (!(xcr0 & XFEATURE_MASK_YMM))
849 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
852 vcpu->arch.xcr0 = xcr0;
854 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
855 kvm_update_cpuid(vcpu);
859 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
861 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
862 __kvm_set_xcr(vcpu, index, xcr)) {
863 kvm_inject_gp(vcpu, 0);
868 EXPORT_SYMBOL_GPL(kvm_set_xcr);
870 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
872 unsigned long old_cr4 = kvm_read_cr4(vcpu);
873 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
874 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
876 if (cr4 & CR4_RESERVED_BITS)
879 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
882 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
885 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
888 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
891 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
894 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
897 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
900 if (is_long_mode(vcpu)) {
901 if (!(cr4 & X86_CR4_PAE))
903 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
904 && ((cr4 ^ old_cr4) & pdptr_bits)
905 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
909 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
910 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
913 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
914 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
918 if (kvm_x86_ops->set_cr4(vcpu, cr4))
921 if (((cr4 ^ old_cr4) & pdptr_bits) ||
922 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
923 kvm_mmu_reset_context(vcpu);
925 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
926 kvm_update_cpuid(vcpu);
930 EXPORT_SYMBOL_GPL(kvm_set_cr4);
932 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
934 bool skip_tlb_flush = false;
936 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
939 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
940 cr3 &= ~X86_CR3_PCID_NOFLUSH;
944 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
945 if (!skip_tlb_flush) {
946 kvm_mmu_sync_roots(vcpu);
947 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
952 if (is_long_mode(vcpu) &&
953 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
955 else if (is_pae(vcpu) && is_paging(vcpu) &&
956 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
959 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
960 vcpu->arch.cr3 = cr3;
961 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
965 EXPORT_SYMBOL_GPL(kvm_set_cr3);
967 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
969 if (cr8 & CR8_RESERVED_BITS)
971 if (lapic_in_kernel(vcpu))
972 kvm_lapic_set_tpr(vcpu, cr8);
974 vcpu->arch.cr8 = cr8;
977 EXPORT_SYMBOL_GPL(kvm_set_cr8);
979 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
981 if (lapic_in_kernel(vcpu))
982 return kvm_lapic_get_cr8(vcpu);
984 return vcpu->arch.cr8;
986 EXPORT_SYMBOL_GPL(kvm_get_cr8);
988 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
992 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
993 for (i = 0; i < KVM_NR_DB_REGS; i++)
994 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
995 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
999 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1001 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1002 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1005 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1009 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1010 dr7 = vcpu->arch.guest_debug_dr7;
1012 dr7 = vcpu->arch.dr7;
1013 kvm_x86_ops->set_dr7(vcpu, dr7);
1014 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1015 if (dr7 & DR7_BP_EN_MASK)
1016 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1019 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1021 u64 fixed = DR6_FIXED_1;
1023 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1028 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1032 vcpu->arch.db[dr] = val;
1033 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1034 vcpu->arch.eff_db[dr] = val;
1039 if (val & 0xffffffff00000000ULL)
1040 return -1; /* #GP */
1041 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1042 kvm_update_dr6(vcpu);
1047 if (val & 0xffffffff00000000ULL)
1048 return -1; /* #GP */
1049 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1050 kvm_update_dr7(vcpu);
1057 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1059 if (__kvm_set_dr(vcpu, dr, val)) {
1060 kvm_inject_gp(vcpu, 0);
1065 EXPORT_SYMBOL_GPL(kvm_set_dr);
1067 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1071 *val = vcpu->arch.db[dr];
1076 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1077 *val = vcpu->arch.dr6;
1079 *val = kvm_x86_ops->get_dr6(vcpu);
1084 *val = vcpu->arch.dr7;
1089 EXPORT_SYMBOL_GPL(kvm_get_dr);
1091 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1093 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
1097 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1100 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1101 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1104 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1107 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1108 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1110 * This list is modified at module load time to reflect the
1111 * capabilities of the host cpu. This capabilities test skips MSRs that are
1112 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1113 * may depend on host virtualization features rather than host cpu features.
1116 static u32 msrs_to_save[] = {
1117 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1119 #ifdef CONFIG_X86_64
1120 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1122 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1123 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1124 MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
1127 static unsigned num_msrs_to_save;
1129 static u32 emulated_msrs[] = {
1130 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1131 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1132 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1133 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1134 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1135 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1136 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1138 HV_X64_MSR_VP_INDEX,
1139 HV_X64_MSR_VP_RUNTIME,
1140 HV_X64_MSR_SCONTROL,
1141 HV_X64_MSR_STIMER0_CONFIG,
1142 HV_X64_MSR_VP_ASSIST_PAGE,
1143 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1144 HV_X64_MSR_TSC_EMULATION_STATUS,
1146 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1149 MSR_IA32_TSC_ADJUST,
1150 MSR_IA32_TSCDEADLINE,
1151 MSR_IA32_MISC_ENABLE,
1152 MSR_IA32_MCG_STATUS,
1154 MSR_IA32_MCG_EXT_CTL,
1158 MSR_MISC_FEATURES_ENABLES,
1159 MSR_AMD64_VIRT_SPEC_CTRL,
1162 static unsigned num_emulated_msrs;
1165 * List of msr numbers which are used to expose MSR-based features that
1166 * can be used by a hypervisor to validate requested CPU features.
1168 static u32 msr_based_features[] = {
1170 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1171 MSR_IA32_VMX_PINBASED_CTLS,
1172 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1173 MSR_IA32_VMX_PROCBASED_CTLS,
1174 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1175 MSR_IA32_VMX_EXIT_CTLS,
1176 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1177 MSR_IA32_VMX_ENTRY_CTLS,
1179 MSR_IA32_VMX_CR0_FIXED0,
1180 MSR_IA32_VMX_CR0_FIXED1,
1181 MSR_IA32_VMX_CR4_FIXED0,
1182 MSR_IA32_VMX_CR4_FIXED1,
1183 MSR_IA32_VMX_VMCS_ENUM,
1184 MSR_IA32_VMX_PROCBASED_CTLS2,
1185 MSR_IA32_VMX_EPT_VPID_CAP,
1186 MSR_IA32_VMX_VMFUNC,
1190 MSR_IA32_ARCH_CAPABILITIES,
1193 static unsigned int num_msr_based_features;
1195 u64 kvm_get_arch_capabilities(void)
1199 rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1202 * If we're doing cache flushes (either "always" or "cond")
1203 * we will do one whenever the guest does a vmlaunch/vmresume.
1204 * If an outer hypervisor is doing the cache flush for us
1205 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1206 * capability to the guest too, and if EPT is disabled we're not
1207 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1208 * require a nested hypervisor to do a flush of its own.
1210 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1211 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1215 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1217 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1219 switch (msr->index) {
1220 case MSR_IA32_ARCH_CAPABILITIES:
1221 msr->data = kvm_get_arch_capabilities();
1223 case MSR_IA32_UCODE_REV:
1224 rdmsrl_safe(msr->index, &msr->data);
1227 if (kvm_x86_ops->get_msr_feature(msr))
1233 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1235 struct kvm_msr_entry msr;
1239 r = kvm_get_msr_feature(&msr);
1248 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1250 if (efer & efer_reserved_bits)
1253 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1256 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1261 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1263 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1265 u64 old_efer = vcpu->arch.efer;
1267 if (!kvm_valid_efer(vcpu, efer))
1271 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1275 efer |= vcpu->arch.efer & EFER_LMA;
1277 kvm_x86_ops->set_efer(vcpu, efer);
1279 /* Update reserved bits */
1280 if ((efer ^ old_efer) & EFER_NX)
1281 kvm_mmu_reset_context(vcpu);
1286 void kvm_enable_efer_bits(u64 mask)
1288 efer_reserved_bits &= ~mask;
1290 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1293 * Writes msr value into into the appropriate "register".
1294 * Returns 0 on success, non-0 otherwise.
1295 * Assumes vcpu_load() was already called.
1297 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1299 switch (msr->index) {
1302 case MSR_KERNEL_GS_BASE:
1305 if (is_noncanonical_address(msr->data, vcpu))
1308 case MSR_IA32_SYSENTER_EIP:
1309 case MSR_IA32_SYSENTER_ESP:
1311 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1312 * non-canonical address is written on Intel but not on
1313 * AMD (which ignores the top 32-bits, because it does
1314 * not implement 64-bit SYSENTER).
1316 * 64-bit code should hence be able to write a non-canonical
1317 * value on AMD. Making the address canonical ensures that
1318 * vmentry does not fail on Intel after writing a non-canonical
1319 * value, and that something deterministic happens if the guest
1320 * invokes 64-bit SYSENTER.
1322 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1324 return kvm_x86_ops->set_msr(vcpu, msr);
1326 EXPORT_SYMBOL_GPL(kvm_set_msr);
1329 * Adapt set_msr() to msr_io()'s calling convention
1331 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1333 struct msr_data msr;
1337 msr.host_initiated = true;
1338 r = kvm_get_msr(vcpu, &msr);
1346 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1348 struct msr_data msr;
1352 msr.host_initiated = true;
1353 return kvm_set_msr(vcpu, &msr);
1356 #ifdef CONFIG_X86_64
1357 struct pvclock_gtod_data {
1360 struct { /* extract of a clocksource struct */
1373 static struct pvclock_gtod_data pvclock_gtod_data;
1375 static void update_pvclock_gtod(struct timekeeper *tk)
1377 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1380 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1382 write_seqcount_begin(&vdata->seq);
1384 /* copy pvclock gtod data */
1385 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1386 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1387 vdata->clock.mask = tk->tkr_mono.mask;
1388 vdata->clock.mult = tk->tkr_mono.mult;
1389 vdata->clock.shift = tk->tkr_mono.shift;
1391 vdata->boot_ns = boot_ns;
1392 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1394 vdata->wall_time_sec = tk->xtime_sec;
1396 write_seqcount_end(&vdata->seq);
1400 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1403 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1404 * vcpu_enter_guest. This function is only called from
1405 * the physical CPU that is running vcpu.
1407 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1410 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1414 struct pvclock_wall_clock wc;
1415 struct timespec64 boot;
1420 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1425 ++version; /* first time write, random junk */
1429 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1433 * The guest calculates current wall clock time by adding
1434 * system time (updated by kvm_guest_time_update below) to the
1435 * wall clock specified here. guest system time equals host
1436 * system time for us, thus we must fill in host boot time here.
1438 getboottime64(&boot);
1440 if (kvm->arch.kvmclock_offset) {
1441 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1442 boot = timespec64_sub(boot, ts);
1444 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1445 wc.nsec = boot.tv_nsec;
1446 wc.version = version;
1448 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1451 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1454 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1456 do_shl32_div32(dividend, divisor);
1460 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1461 s8 *pshift, u32 *pmultiplier)
1469 scaled64 = scaled_hz;
1470 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1475 tps32 = (uint32_t)tps64;
1476 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1477 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1485 *pmultiplier = div_frac(scaled64, tps32);
1487 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1488 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1491 #ifdef CONFIG_X86_64
1492 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1495 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1496 static unsigned long max_tsc_khz;
1498 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1500 u64 v = (u64)khz * (1000000 + ppm);
1505 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1509 /* Guest TSC same frequency as host TSC? */
1511 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1515 /* TSC scaling supported? */
1516 if (!kvm_has_tsc_control) {
1517 if (user_tsc_khz > tsc_khz) {
1518 vcpu->arch.tsc_catchup = 1;
1519 vcpu->arch.tsc_always_catchup = 1;
1522 WARN(1, "user requested TSC rate below hardware speed\n");
1527 /* TSC scaling required - calculate ratio */
1528 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1529 user_tsc_khz, tsc_khz);
1531 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1532 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1537 vcpu->arch.tsc_scaling_ratio = ratio;
1541 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1543 u32 thresh_lo, thresh_hi;
1544 int use_scaling = 0;
1546 /* tsc_khz can be zero if TSC calibration fails */
1547 if (user_tsc_khz == 0) {
1548 /* set tsc_scaling_ratio to a safe value */
1549 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1553 /* Compute a scale to convert nanoseconds in TSC cycles */
1554 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1555 &vcpu->arch.virtual_tsc_shift,
1556 &vcpu->arch.virtual_tsc_mult);
1557 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1560 * Compute the variation in TSC rate which is acceptable
1561 * within the range of tolerance and decide if the
1562 * rate being applied is within that bounds of the hardware
1563 * rate. If so, no scaling or compensation need be done.
1565 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1566 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1567 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1568 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1571 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1574 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1576 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1577 vcpu->arch.virtual_tsc_mult,
1578 vcpu->arch.virtual_tsc_shift);
1579 tsc += vcpu->arch.this_tsc_write;
1583 static inline int gtod_is_based_on_tsc(int mode)
1585 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1588 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1590 #ifdef CONFIG_X86_64
1592 struct kvm_arch *ka = &vcpu->kvm->arch;
1593 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1595 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1596 atomic_read(&vcpu->kvm->online_vcpus));
1599 * Once the masterclock is enabled, always perform request in
1600 * order to update it.
1602 * In order to enable masterclock, the host clocksource must be TSC
1603 * and the vcpus need to have matched TSCs. When that happens,
1604 * perform request to enable masterclock.
1606 if (ka->use_master_clock ||
1607 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1608 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1610 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1611 atomic_read(&vcpu->kvm->online_vcpus),
1612 ka->use_master_clock, gtod->clock.vclock_mode);
1616 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1618 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1619 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1623 * Multiply tsc by a fixed point number represented by ratio.
1625 * The most significant 64-N bits (mult) of ratio represent the
1626 * integral part of the fixed point number; the remaining N bits
1627 * (frac) represent the fractional part, ie. ratio represents a fixed
1628 * point number (mult + frac * 2^(-N)).
1630 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1632 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1634 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1637 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1640 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1642 if (ratio != kvm_default_tsc_scaling_ratio)
1643 _tsc = __scale_tsc(ratio, tsc);
1647 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1649 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1653 tsc = kvm_scale_tsc(vcpu, rdtsc());
1655 return target_tsc - tsc;
1658 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1660 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1662 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1664 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1666 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1668 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1671 static inline bool kvm_check_tsc_unstable(void)
1673 #ifdef CONFIG_X86_64
1675 * TSC is marked unstable when we're running on Hyper-V,
1676 * 'TSC page' clocksource is good.
1678 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1681 return check_tsc_unstable();
1684 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1686 struct kvm *kvm = vcpu->kvm;
1687 u64 offset, ns, elapsed;
1688 unsigned long flags;
1690 bool already_matched;
1691 u64 data = msr->data;
1692 bool synchronizing = false;
1694 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1695 offset = kvm_compute_tsc_offset(vcpu, data);
1696 ns = ktime_get_boot_ns();
1697 elapsed = ns - kvm->arch.last_tsc_nsec;
1699 if (vcpu->arch.virtual_tsc_khz) {
1700 if (data == 0 && msr->host_initiated) {
1702 * detection of vcpu initialization -- need to sync
1703 * with other vCPUs. This particularly helps to keep
1704 * kvm_clock stable after CPU hotplug
1706 synchronizing = true;
1708 u64 tsc_exp = kvm->arch.last_tsc_write +
1709 nsec_to_cycles(vcpu, elapsed);
1710 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1712 * Special case: TSC write with a small delta (1 second)
1713 * of virtual cycle time against real time is
1714 * interpreted as an attempt to synchronize the CPU.
1716 synchronizing = data < tsc_exp + tsc_hz &&
1717 data + tsc_hz > tsc_exp;
1722 * For a reliable TSC, we can match TSC offsets, and for an unstable
1723 * TSC, we add elapsed time in this computation. We could let the
1724 * compensation code attempt to catch up if we fall behind, but
1725 * it's better to try to match offsets from the beginning.
1727 if (synchronizing &&
1728 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1729 if (!kvm_check_tsc_unstable()) {
1730 offset = kvm->arch.cur_tsc_offset;
1731 pr_debug("kvm: matched tsc offset for %llu\n", data);
1733 u64 delta = nsec_to_cycles(vcpu, elapsed);
1735 offset = kvm_compute_tsc_offset(vcpu, data);
1736 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1739 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1742 * We split periods of matched TSC writes into generations.
1743 * For each generation, we track the original measured
1744 * nanosecond time, offset, and write, so if TSCs are in
1745 * sync, we can match exact offset, and if not, we can match
1746 * exact software computation in compute_guest_tsc()
1748 * These values are tracked in kvm->arch.cur_xxx variables.
1750 kvm->arch.cur_tsc_generation++;
1751 kvm->arch.cur_tsc_nsec = ns;
1752 kvm->arch.cur_tsc_write = data;
1753 kvm->arch.cur_tsc_offset = offset;
1755 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1756 kvm->arch.cur_tsc_generation, data);
1760 * We also track th most recent recorded KHZ, write and time to
1761 * allow the matching interval to be extended at each write.
1763 kvm->arch.last_tsc_nsec = ns;
1764 kvm->arch.last_tsc_write = data;
1765 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1767 vcpu->arch.last_guest_tsc = data;
1769 /* Keep track of which generation this VCPU has synchronized to */
1770 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1771 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1772 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1774 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1775 update_ia32_tsc_adjust_msr(vcpu, offset);
1777 kvm_vcpu_write_tsc_offset(vcpu, offset);
1778 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1780 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1782 kvm->arch.nr_vcpus_matched_tsc = 0;
1783 } else if (!already_matched) {
1784 kvm->arch.nr_vcpus_matched_tsc++;
1787 kvm_track_tsc_matching(vcpu);
1788 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1791 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1793 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1796 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1797 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1800 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1802 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1803 WARN_ON(adjustment < 0);
1804 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1805 adjust_tsc_offset_guest(vcpu, adjustment);
1808 #ifdef CONFIG_X86_64
1810 static u64 read_tsc(void)
1812 u64 ret = (u64)rdtsc_ordered();
1813 u64 last = pvclock_gtod_data.clock.cycle_last;
1815 if (likely(ret >= last))
1819 * GCC likes to generate cmov here, but this branch is extremely
1820 * predictable (it's just a function of time and the likely is
1821 * very likely) and there's a data dependence, so force GCC
1822 * to generate a branch instead. I don't barrier() because
1823 * we don't actually need a barrier, and if this function
1824 * ever gets inlined it will generate worse code.
1830 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1833 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1836 switch (gtod->clock.vclock_mode) {
1837 case VCLOCK_HVCLOCK:
1838 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1840 if (tsc_pg_val != U64_MAX) {
1841 /* TSC page valid */
1842 *mode = VCLOCK_HVCLOCK;
1843 v = (tsc_pg_val - gtod->clock.cycle_last) &
1846 /* TSC page invalid */
1847 *mode = VCLOCK_NONE;
1852 *tsc_timestamp = read_tsc();
1853 v = (*tsc_timestamp - gtod->clock.cycle_last) &
1857 *mode = VCLOCK_NONE;
1860 if (*mode == VCLOCK_NONE)
1861 *tsc_timestamp = v = 0;
1863 return v * gtod->clock.mult;
1866 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1868 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1874 seq = read_seqcount_begin(>od->seq);
1875 ns = gtod->nsec_base;
1876 ns += vgettsc(tsc_timestamp, &mode);
1877 ns >>= gtod->clock.shift;
1878 ns += gtod->boot_ns;
1879 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1885 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1887 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1893 seq = read_seqcount_begin(>od->seq);
1894 ts->tv_sec = gtod->wall_time_sec;
1895 ns = gtod->nsec_base;
1896 ns += vgettsc(tsc_timestamp, &mode);
1897 ns >>= gtod->clock.shift;
1898 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1900 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1906 /* returns true if host is using TSC based clocksource */
1907 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1909 /* checked again under seqlock below */
1910 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1913 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1917 /* returns true if host is using TSC based clocksource */
1918 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1921 /* checked again under seqlock below */
1922 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1925 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1931 * Assuming a stable TSC across physical CPUS, and a stable TSC
1932 * across virtual CPUs, the following condition is possible.
1933 * Each numbered line represents an event visible to both
1934 * CPUs at the next numbered event.
1936 * "timespecX" represents host monotonic time. "tscX" represents
1939 * VCPU0 on CPU0 | VCPU1 on CPU1
1941 * 1. read timespec0,tsc0
1942 * 2. | timespec1 = timespec0 + N
1944 * 3. transition to guest | transition to guest
1945 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1946 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1947 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1949 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1952 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1954 * - 0 < N - M => M < N
1956 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1957 * always the case (the difference between two distinct xtime instances
1958 * might be smaller then the difference between corresponding TSC reads,
1959 * when updating guest vcpus pvclock areas).
1961 * To avoid that problem, do not allow visibility of distinct
1962 * system_timestamp/tsc_timestamp values simultaneously: use a master
1963 * copy of host monotonic time values. Update that master copy
1966 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1970 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1972 #ifdef CONFIG_X86_64
1973 struct kvm_arch *ka = &kvm->arch;
1975 bool host_tsc_clocksource, vcpus_matched;
1977 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1978 atomic_read(&kvm->online_vcpus));
1981 * If the host uses TSC clock, then passthrough TSC as stable
1984 host_tsc_clocksource = kvm_get_time_and_clockread(
1985 &ka->master_kernel_ns,
1986 &ka->master_cycle_now);
1988 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1989 && !ka->backwards_tsc_observed
1990 && !ka->boot_vcpu_runs_old_kvmclock;
1992 if (ka->use_master_clock)
1993 atomic_set(&kvm_guest_has_master_clock, 1);
1995 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1996 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2001 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2003 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2006 static void kvm_gen_update_masterclock(struct kvm *kvm)
2008 #ifdef CONFIG_X86_64
2010 struct kvm_vcpu *vcpu;
2011 struct kvm_arch *ka = &kvm->arch;
2013 spin_lock(&ka->pvclock_gtod_sync_lock);
2014 kvm_make_mclock_inprogress_request(kvm);
2015 /* no guest entries from this point */
2016 pvclock_update_vm_gtod_copy(kvm);
2018 kvm_for_each_vcpu(i, vcpu, kvm)
2019 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2021 /* guest entries allowed */
2022 kvm_for_each_vcpu(i, vcpu, kvm)
2023 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2025 spin_unlock(&ka->pvclock_gtod_sync_lock);
2029 u64 get_kvmclock_ns(struct kvm *kvm)
2031 struct kvm_arch *ka = &kvm->arch;
2032 struct pvclock_vcpu_time_info hv_clock;
2035 spin_lock(&ka->pvclock_gtod_sync_lock);
2036 if (!ka->use_master_clock) {
2037 spin_unlock(&ka->pvclock_gtod_sync_lock);
2038 return ktime_get_boot_ns() + ka->kvmclock_offset;
2041 hv_clock.tsc_timestamp = ka->master_cycle_now;
2042 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2043 spin_unlock(&ka->pvclock_gtod_sync_lock);
2045 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2048 if (__this_cpu_read(cpu_tsc_khz)) {
2049 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2050 &hv_clock.tsc_shift,
2051 &hv_clock.tsc_to_system_mul);
2052 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2054 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2061 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2063 struct kvm_vcpu_arch *vcpu = &v->arch;
2064 struct pvclock_vcpu_time_info guest_hv_clock;
2066 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2067 &guest_hv_clock, sizeof(guest_hv_clock))))
2070 /* This VCPU is paused, but it's legal for a guest to read another
2071 * VCPU's kvmclock, so we really have to follow the specification where
2072 * it says that version is odd if data is being modified, and even after
2075 * Version field updates must be kept separate. This is because
2076 * kvm_write_guest_cached might use a "rep movs" instruction, and
2077 * writes within a string instruction are weakly ordered. So there
2078 * are three writes overall.
2080 * As a small optimization, only write the version field in the first
2081 * and third write. The vcpu->pv_time cache is still valid, because the
2082 * version field is the first in the struct.
2084 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2086 if (guest_hv_clock.version & 1)
2087 ++guest_hv_clock.version; /* first time write, random junk */
2089 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2090 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2092 sizeof(vcpu->hv_clock.version));
2096 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2097 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2099 if (vcpu->pvclock_set_guest_stopped_request) {
2100 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2101 vcpu->pvclock_set_guest_stopped_request = false;
2104 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2106 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2108 sizeof(vcpu->hv_clock));
2112 vcpu->hv_clock.version++;
2113 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2115 sizeof(vcpu->hv_clock.version));
2118 static int kvm_guest_time_update(struct kvm_vcpu *v)
2120 unsigned long flags, tgt_tsc_khz;
2121 struct kvm_vcpu_arch *vcpu = &v->arch;
2122 struct kvm_arch *ka = &v->kvm->arch;
2124 u64 tsc_timestamp, host_tsc;
2126 bool use_master_clock;
2132 * If the host uses TSC clock, then passthrough TSC as stable
2135 spin_lock(&ka->pvclock_gtod_sync_lock);
2136 use_master_clock = ka->use_master_clock;
2137 if (use_master_clock) {
2138 host_tsc = ka->master_cycle_now;
2139 kernel_ns = ka->master_kernel_ns;
2141 spin_unlock(&ka->pvclock_gtod_sync_lock);
2143 /* Keep irq disabled to prevent changes to the clock */
2144 local_irq_save(flags);
2145 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2146 if (unlikely(tgt_tsc_khz == 0)) {
2147 local_irq_restore(flags);
2148 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2151 if (!use_master_clock) {
2153 kernel_ns = ktime_get_boot_ns();
2156 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2159 * We may have to catch up the TSC to match elapsed wall clock
2160 * time for two reasons, even if kvmclock is used.
2161 * 1) CPU could have been running below the maximum TSC rate
2162 * 2) Broken TSC compensation resets the base at each VCPU
2163 * entry to avoid unknown leaps of TSC even when running
2164 * again on the same CPU. This may cause apparent elapsed
2165 * time to disappear, and the guest to stand still or run
2168 if (vcpu->tsc_catchup) {
2169 u64 tsc = compute_guest_tsc(v, kernel_ns);
2170 if (tsc > tsc_timestamp) {
2171 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2172 tsc_timestamp = tsc;
2176 local_irq_restore(flags);
2178 /* With all the info we got, fill in the values */
2180 if (kvm_has_tsc_control)
2181 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2183 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2184 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2185 &vcpu->hv_clock.tsc_shift,
2186 &vcpu->hv_clock.tsc_to_system_mul);
2187 vcpu->hw_tsc_khz = tgt_tsc_khz;
2190 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2191 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2192 vcpu->last_guest_tsc = tsc_timestamp;
2194 /* If the host uses TSC clocksource, then it is stable */
2196 if (use_master_clock)
2197 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2199 vcpu->hv_clock.flags = pvclock_flags;
2201 if (vcpu->pv_time_enabled)
2202 kvm_setup_pvclock_page(v);
2203 if (v == kvm_get_vcpu(v->kvm, 0))
2204 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2209 * kvmclock updates which are isolated to a given vcpu, such as
2210 * vcpu->cpu migration, should not allow system_timestamp from
2211 * the rest of the vcpus to remain static. Otherwise ntp frequency
2212 * correction applies to one vcpu's system_timestamp but not
2215 * So in those cases, request a kvmclock update for all vcpus.
2216 * We need to rate-limit these requests though, as they can
2217 * considerably slow guests that have a large number of vcpus.
2218 * The time for a remote vcpu to update its kvmclock is bound
2219 * by the delay we use to rate-limit the updates.
2222 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2224 static void kvmclock_update_fn(struct work_struct *work)
2227 struct delayed_work *dwork = to_delayed_work(work);
2228 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2229 kvmclock_update_work);
2230 struct kvm *kvm = container_of(ka, struct kvm, arch);
2231 struct kvm_vcpu *vcpu;
2233 kvm_for_each_vcpu(i, vcpu, kvm) {
2234 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2235 kvm_vcpu_kick(vcpu);
2239 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2241 struct kvm *kvm = v->kvm;
2243 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2244 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2245 KVMCLOCK_UPDATE_DELAY);
2248 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2250 static void kvmclock_sync_fn(struct work_struct *work)
2252 struct delayed_work *dwork = to_delayed_work(work);
2253 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2254 kvmclock_sync_work);
2255 struct kvm *kvm = container_of(ka, struct kvm, arch);
2257 if (!kvmclock_periodic_sync)
2260 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2261 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2262 KVMCLOCK_SYNC_PERIOD);
2265 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2267 u64 mcg_cap = vcpu->arch.mcg_cap;
2268 unsigned bank_num = mcg_cap & 0xff;
2269 u32 msr = msr_info->index;
2270 u64 data = msr_info->data;
2273 case MSR_IA32_MCG_STATUS:
2274 vcpu->arch.mcg_status = data;
2276 case MSR_IA32_MCG_CTL:
2277 if (!(mcg_cap & MCG_CTL_P) &&
2278 (data || !msr_info->host_initiated))
2280 if (data != 0 && data != ~(u64)0)
2282 vcpu->arch.mcg_ctl = data;
2285 if (msr >= MSR_IA32_MC0_CTL &&
2286 msr < MSR_IA32_MCx_CTL(bank_num)) {
2287 u32 offset = msr - MSR_IA32_MC0_CTL;
2288 /* only 0 or all 1s can be written to IA32_MCi_CTL
2289 * some Linux kernels though clear bit 10 in bank 4 to
2290 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2291 * this to avoid an uncatched #GP in the guest
2293 if ((offset & 0x3) == 0 &&
2294 data != 0 && (data | (1 << 10)) != ~(u64)0)
2296 if (!msr_info->host_initiated &&
2297 (offset & 0x3) == 1 && data != 0)
2299 vcpu->arch.mce_banks[offset] = data;
2307 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2309 struct kvm *kvm = vcpu->kvm;
2310 int lm = is_long_mode(vcpu);
2311 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2312 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2313 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2314 : kvm->arch.xen_hvm_config.blob_size_32;
2315 u32 page_num = data & ~PAGE_MASK;
2316 u64 page_addr = data & PAGE_MASK;
2321 if (page_num >= blob_size)
2324 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2329 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2338 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2340 gpa_t gpa = data & ~0x3f;
2342 /* Bits 3:5 are reserved, Should be zero */
2346 vcpu->arch.apf.msr_val = data;
2348 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2349 kvm_clear_async_pf_completion_queue(vcpu);
2350 kvm_async_pf_hash_reset(vcpu);
2354 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2358 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2359 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2360 kvm_async_pf_wakeup_all(vcpu);
2364 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2366 vcpu->arch.pv_time_enabled = false;
2369 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2371 ++vcpu->stat.tlb_flush;
2372 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2375 static void record_steal_time(struct kvm_vcpu *vcpu)
2377 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2380 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2381 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2385 * Doing a TLB flush here, on the guest's behalf, can avoid
2388 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2389 kvm_vcpu_flush_tlb(vcpu, false);
2391 if (vcpu->arch.st.steal.version & 1)
2392 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2394 vcpu->arch.st.steal.version += 1;
2396 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2397 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2401 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2402 vcpu->arch.st.last_steal;
2403 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2405 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2406 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2410 vcpu->arch.st.steal.version += 1;
2412 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2413 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2416 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2419 u32 msr = msr_info->index;
2420 u64 data = msr_info->data;
2423 case MSR_AMD64_NB_CFG:
2424 case MSR_IA32_UCODE_WRITE:
2425 case MSR_VM_HSAVE_PA:
2426 case MSR_AMD64_PATCH_LOADER:
2427 case MSR_AMD64_BU_CFG2:
2428 case MSR_AMD64_DC_CFG:
2429 case MSR_F15H_EX_CFG:
2432 case MSR_IA32_UCODE_REV:
2433 if (msr_info->host_initiated)
2434 vcpu->arch.microcode_version = data;
2437 return set_efer(vcpu, data);
2439 data &= ~(u64)0x40; /* ignore flush filter disable */
2440 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2441 data &= ~(u64)0x8; /* ignore TLB cache disable */
2442 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2444 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2449 case MSR_FAM10H_MMIO_CONF_BASE:
2451 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2456 case MSR_IA32_DEBUGCTLMSR:
2458 /* We support the non-activated case already */
2460 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2461 /* Values other than LBR and BTF are vendor-specific,
2462 thus reserved and should throw a #GP */
2465 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2468 case 0x200 ... 0x2ff:
2469 return kvm_mtrr_set_msr(vcpu, msr, data);
2470 case MSR_IA32_APICBASE:
2471 return kvm_set_apic_base(vcpu, msr_info);
2472 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2473 return kvm_x2apic_msr_write(vcpu, msr, data);
2474 case MSR_IA32_TSCDEADLINE:
2475 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2477 case MSR_IA32_TSC_ADJUST:
2478 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2479 if (!msr_info->host_initiated) {
2480 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2481 adjust_tsc_offset_guest(vcpu, adj);
2483 vcpu->arch.ia32_tsc_adjust_msr = data;
2486 case MSR_IA32_MISC_ENABLE:
2487 vcpu->arch.ia32_misc_enable_msr = data;
2489 case MSR_IA32_SMBASE:
2490 if (!msr_info->host_initiated)
2492 vcpu->arch.smbase = data;
2495 kvm_write_tsc(vcpu, msr_info);
2498 if (!msr_info->host_initiated)
2500 vcpu->arch.smi_count = data;
2502 case MSR_KVM_WALL_CLOCK_NEW:
2503 case MSR_KVM_WALL_CLOCK:
2504 vcpu->kvm->arch.wall_clock = data;
2505 kvm_write_wall_clock(vcpu->kvm, data);
2507 case MSR_KVM_SYSTEM_TIME_NEW:
2508 case MSR_KVM_SYSTEM_TIME: {
2509 struct kvm_arch *ka = &vcpu->kvm->arch;
2511 kvmclock_reset(vcpu);
2513 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2514 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2516 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2517 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2519 ka->boot_vcpu_runs_old_kvmclock = tmp;
2522 vcpu->arch.time = data;
2523 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2525 /* we verify if the enable bit is set... */
2529 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2530 &vcpu->arch.pv_time, data & ~1ULL,
2531 sizeof(struct pvclock_vcpu_time_info)))
2532 vcpu->arch.pv_time_enabled = false;
2534 vcpu->arch.pv_time_enabled = true;
2538 case MSR_KVM_ASYNC_PF_EN:
2539 if (kvm_pv_enable_async_pf(vcpu, data))
2542 case MSR_KVM_STEAL_TIME:
2544 if (unlikely(!sched_info_on()))
2547 if (data & KVM_STEAL_RESERVED_MASK)
2550 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2551 data & KVM_STEAL_VALID_BITS,
2552 sizeof(struct kvm_steal_time)))
2555 vcpu->arch.st.msr_val = data;
2557 if (!(data & KVM_MSR_ENABLED))
2560 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2563 case MSR_KVM_PV_EOI_EN:
2564 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2568 case MSR_IA32_MCG_CTL:
2569 case MSR_IA32_MCG_STATUS:
2570 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2571 return set_msr_mce(vcpu, msr_info);
2573 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2574 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2575 pr = true; /* fall through */
2576 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2577 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2578 if (kvm_pmu_is_valid_msr(vcpu, msr))
2579 return kvm_pmu_set_msr(vcpu, msr_info);
2581 if (pr || data != 0)
2582 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2583 "0x%x data 0x%llx\n", msr, data);
2585 case MSR_K7_CLK_CTL:
2587 * Ignore all writes to this no longer documented MSR.
2588 * Writes are only relevant for old K7 processors,
2589 * all pre-dating SVM, but a recommended workaround from
2590 * AMD for these chips. It is possible to specify the
2591 * affected processor models on the command line, hence
2592 * the need to ignore the workaround.
2595 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2596 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2597 case HV_X64_MSR_CRASH_CTL:
2598 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2599 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2600 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2601 case HV_X64_MSR_TSC_EMULATION_STATUS:
2602 return kvm_hv_set_msr_common(vcpu, msr, data,
2603 msr_info->host_initiated);
2604 case MSR_IA32_BBL_CR_CTL3:
2605 /* Drop writes to this legacy MSR -- see rdmsr
2606 * counterpart for further detail.
2608 if (report_ignored_msrs)
2609 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2612 case MSR_AMD64_OSVW_ID_LENGTH:
2613 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2615 vcpu->arch.osvw.length = data;
2617 case MSR_AMD64_OSVW_STATUS:
2618 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2620 vcpu->arch.osvw.status = data;
2622 case MSR_PLATFORM_INFO:
2623 if (!msr_info->host_initiated ||
2624 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2625 cpuid_fault_enabled(vcpu)))
2627 vcpu->arch.msr_platform_info = data;
2629 case MSR_MISC_FEATURES_ENABLES:
2630 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2631 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2632 !supports_cpuid_fault(vcpu)))
2634 vcpu->arch.msr_misc_features_enables = data;
2637 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2638 return xen_hvm_config(vcpu, data);
2639 if (kvm_pmu_is_valid_msr(vcpu, msr))
2640 return kvm_pmu_set_msr(vcpu, msr_info);
2642 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2646 if (report_ignored_msrs)
2648 "ignored wrmsr: 0x%x data 0x%llx\n",
2655 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2659 * Reads an msr value (of 'msr_index') into 'pdata'.
2660 * Returns 0 on success, non-0 otherwise.
2661 * Assumes vcpu_load() was already called.
2663 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2665 return kvm_x86_ops->get_msr(vcpu, msr);
2667 EXPORT_SYMBOL_GPL(kvm_get_msr);
2669 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2672 u64 mcg_cap = vcpu->arch.mcg_cap;
2673 unsigned bank_num = mcg_cap & 0xff;
2676 case MSR_IA32_P5_MC_ADDR:
2677 case MSR_IA32_P5_MC_TYPE:
2680 case MSR_IA32_MCG_CAP:
2681 data = vcpu->arch.mcg_cap;
2683 case MSR_IA32_MCG_CTL:
2684 if (!(mcg_cap & MCG_CTL_P) && !host)
2686 data = vcpu->arch.mcg_ctl;
2688 case MSR_IA32_MCG_STATUS:
2689 data = vcpu->arch.mcg_status;
2692 if (msr >= MSR_IA32_MC0_CTL &&
2693 msr < MSR_IA32_MCx_CTL(bank_num)) {
2694 u32 offset = msr - MSR_IA32_MC0_CTL;
2695 data = vcpu->arch.mce_banks[offset];
2704 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2706 switch (msr_info->index) {
2707 case MSR_IA32_PLATFORM_ID:
2708 case MSR_IA32_EBL_CR_POWERON:
2709 case MSR_IA32_DEBUGCTLMSR:
2710 case MSR_IA32_LASTBRANCHFROMIP:
2711 case MSR_IA32_LASTBRANCHTOIP:
2712 case MSR_IA32_LASTINTFROMIP:
2713 case MSR_IA32_LASTINTTOIP:
2715 case MSR_K8_TSEG_ADDR:
2716 case MSR_K8_TSEG_MASK:
2718 case MSR_VM_HSAVE_PA:
2719 case MSR_K8_INT_PENDING_MSG:
2720 case MSR_AMD64_NB_CFG:
2721 case MSR_FAM10H_MMIO_CONF_BASE:
2722 case MSR_AMD64_BU_CFG2:
2723 case MSR_IA32_PERF_CTL:
2724 case MSR_AMD64_DC_CFG:
2725 case MSR_F15H_EX_CFG:
2728 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2729 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2730 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2731 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2732 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2733 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2734 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2737 case MSR_IA32_UCODE_REV:
2738 msr_info->data = vcpu->arch.microcode_version;
2741 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2744 case 0x200 ... 0x2ff:
2745 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2746 case 0xcd: /* fsb frequency */
2750 * MSR_EBC_FREQUENCY_ID
2751 * Conservative value valid for even the basic CPU models.
2752 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2753 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2754 * and 266MHz for model 3, or 4. Set Core Clock
2755 * Frequency to System Bus Frequency Ratio to 1 (bits
2756 * 31:24) even though these are only valid for CPU
2757 * models > 2, however guests may end up dividing or
2758 * multiplying by zero otherwise.
2760 case MSR_EBC_FREQUENCY_ID:
2761 msr_info->data = 1 << 24;
2763 case MSR_IA32_APICBASE:
2764 msr_info->data = kvm_get_apic_base(vcpu);
2766 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2767 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2769 case MSR_IA32_TSCDEADLINE:
2770 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2772 case MSR_IA32_TSC_ADJUST:
2773 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2775 case MSR_IA32_MISC_ENABLE:
2776 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2778 case MSR_IA32_SMBASE:
2779 if (!msr_info->host_initiated)
2781 msr_info->data = vcpu->arch.smbase;
2784 msr_info->data = vcpu->arch.smi_count;
2786 case MSR_IA32_PERF_STATUS:
2787 /* TSC increment by tick */
2788 msr_info->data = 1000ULL;
2789 /* CPU multiplier */
2790 msr_info->data |= (((uint64_t)4ULL) << 40);
2793 msr_info->data = vcpu->arch.efer;
2795 case MSR_KVM_WALL_CLOCK:
2796 case MSR_KVM_WALL_CLOCK_NEW:
2797 msr_info->data = vcpu->kvm->arch.wall_clock;
2799 case MSR_KVM_SYSTEM_TIME:
2800 case MSR_KVM_SYSTEM_TIME_NEW:
2801 msr_info->data = vcpu->arch.time;
2803 case MSR_KVM_ASYNC_PF_EN:
2804 msr_info->data = vcpu->arch.apf.msr_val;
2806 case MSR_KVM_STEAL_TIME:
2807 msr_info->data = vcpu->arch.st.msr_val;
2809 case MSR_KVM_PV_EOI_EN:
2810 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2812 case MSR_IA32_P5_MC_ADDR:
2813 case MSR_IA32_P5_MC_TYPE:
2814 case MSR_IA32_MCG_CAP:
2815 case MSR_IA32_MCG_CTL:
2816 case MSR_IA32_MCG_STATUS:
2817 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2818 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2819 msr_info->host_initiated);
2820 case MSR_K7_CLK_CTL:
2822 * Provide expected ramp-up count for K7. All other
2823 * are set to zero, indicating minimum divisors for
2826 * This prevents guest kernels on AMD host with CPU
2827 * type 6, model 8 and higher from exploding due to
2828 * the rdmsr failing.
2830 msr_info->data = 0x20000000;
2832 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2833 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2834 case HV_X64_MSR_CRASH_CTL:
2835 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2836 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2837 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2838 case HV_X64_MSR_TSC_EMULATION_STATUS:
2839 return kvm_hv_get_msr_common(vcpu,
2840 msr_info->index, &msr_info->data,
2841 msr_info->host_initiated);
2843 case MSR_IA32_BBL_CR_CTL3:
2844 /* This legacy MSR exists but isn't fully documented in current
2845 * silicon. It is however accessed by winxp in very narrow
2846 * scenarios where it sets bit #19, itself documented as
2847 * a "reserved" bit. Best effort attempt to source coherent
2848 * read data here should the balance of the register be
2849 * interpreted by the guest:
2851 * L2 cache control register 3: 64GB range, 256KB size,
2852 * enabled, latency 0x1, configured
2854 msr_info->data = 0xbe702111;
2856 case MSR_AMD64_OSVW_ID_LENGTH:
2857 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2859 msr_info->data = vcpu->arch.osvw.length;
2861 case MSR_AMD64_OSVW_STATUS:
2862 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2864 msr_info->data = vcpu->arch.osvw.status;
2866 case MSR_PLATFORM_INFO:
2867 if (!msr_info->host_initiated &&
2868 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2870 msr_info->data = vcpu->arch.msr_platform_info;
2872 case MSR_MISC_FEATURES_ENABLES:
2873 msr_info->data = vcpu->arch.msr_misc_features_enables;
2876 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2877 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2879 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2883 if (report_ignored_msrs)
2884 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2892 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2895 * Read or write a bunch of msrs. All parameters are kernel addresses.
2897 * @return number of msrs set successfully.
2899 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2900 struct kvm_msr_entry *entries,
2901 int (*do_msr)(struct kvm_vcpu *vcpu,
2902 unsigned index, u64 *data))
2906 for (i = 0; i < msrs->nmsrs; ++i)
2907 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2914 * Read or write a bunch of msrs. Parameters are user addresses.
2916 * @return number of msrs set successfully.
2918 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2919 int (*do_msr)(struct kvm_vcpu *vcpu,
2920 unsigned index, u64 *data),
2923 struct kvm_msrs msrs;
2924 struct kvm_msr_entry *entries;
2929 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
2933 if (msrs.nmsrs >= MAX_IO_MSRS)
2936 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2937 entries = memdup_user(user_msrs->entries, size);
2938 if (IS_ERR(entries)) {
2939 r = PTR_ERR(entries);
2943 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2948 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2959 static inline bool kvm_can_mwait_in_guest(void)
2961 return boot_cpu_has(X86_FEATURE_MWAIT) &&
2962 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
2963 boot_cpu_has(X86_FEATURE_ARAT);
2966 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2971 case KVM_CAP_IRQCHIP:
2973 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2974 case KVM_CAP_SET_TSS_ADDR:
2975 case KVM_CAP_EXT_CPUID:
2976 case KVM_CAP_EXT_EMUL_CPUID:
2977 case KVM_CAP_CLOCKSOURCE:
2979 case KVM_CAP_NOP_IO_DELAY:
2980 case KVM_CAP_MP_STATE:
2981 case KVM_CAP_SYNC_MMU:
2982 case KVM_CAP_USER_NMI:
2983 case KVM_CAP_REINJECT_CONTROL:
2984 case KVM_CAP_IRQ_INJECT_STATUS:
2985 case KVM_CAP_IOEVENTFD:
2986 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2988 case KVM_CAP_PIT_STATE2:
2989 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2990 case KVM_CAP_XEN_HVM:
2991 case KVM_CAP_VCPU_EVENTS:
2992 case KVM_CAP_HYPERV:
2993 case KVM_CAP_HYPERV_VAPIC:
2994 case KVM_CAP_HYPERV_SPIN:
2995 case KVM_CAP_HYPERV_SYNIC:
2996 case KVM_CAP_HYPERV_SYNIC2:
2997 case KVM_CAP_HYPERV_VP_INDEX:
2998 case KVM_CAP_HYPERV_EVENTFD:
2999 case KVM_CAP_HYPERV_TLBFLUSH:
3000 case KVM_CAP_HYPERV_SEND_IPI:
3001 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3002 case KVM_CAP_PCI_SEGMENT:
3003 case KVM_CAP_DEBUGREGS:
3004 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3006 case KVM_CAP_ASYNC_PF:
3007 case KVM_CAP_GET_TSC_KHZ:
3008 case KVM_CAP_KVMCLOCK_CTRL:
3009 case KVM_CAP_READONLY_MEM:
3010 case KVM_CAP_HYPERV_TIME:
3011 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3012 case KVM_CAP_TSC_DEADLINE_TIMER:
3013 case KVM_CAP_ENABLE_CAP_VM:
3014 case KVM_CAP_DISABLE_QUIRKS:
3015 case KVM_CAP_SET_BOOT_CPU_ID:
3016 case KVM_CAP_SPLIT_IRQCHIP:
3017 case KVM_CAP_IMMEDIATE_EXIT:
3018 case KVM_CAP_GET_MSR_FEATURES:
3019 case KVM_CAP_MSR_PLATFORM_INFO:
3020 case KVM_CAP_EXCEPTION_PAYLOAD:
3023 case KVM_CAP_SYNC_REGS:
3024 r = KVM_SYNC_X86_VALID_FIELDS;
3026 case KVM_CAP_ADJUST_CLOCK:
3027 r = KVM_CLOCK_TSC_STABLE;
3029 case KVM_CAP_X86_DISABLE_EXITS:
3030 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE;
3031 if(kvm_can_mwait_in_guest())
3032 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3034 case KVM_CAP_X86_SMM:
3035 /* SMBASE is usually relocated above 1M on modern chipsets,
3036 * and SMM handlers might indeed rely on 4G segment limits,
3037 * so do not report SMM to be available if real mode is
3038 * emulated via vm86 mode. Still, do not go to great lengths
3039 * to avoid userspace's usage of the feature, because it is a
3040 * fringe case that is not enabled except via specific settings
3041 * of the module parameters.
3043 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3046 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3048 case KVM_CAP_NR_VCPUS:
3049 r = KVM_SOFT_MAX_VCPUS;
3051 case KVM_CAP_MAX_VCPUS:
3054 case KVM_CAP_NR_MEMSLOTS:
3055 r = KVM_USER_MEM_SLOTS;
3057 case KVM_CAP_PV_MMU: /* obsolete */
3061 r = KVM_MAX_MCE_BANKS;
3064 r = boot_cpu_has(X86_FEATURE_XSAVE);
3066 case KVM_CAP_TSC_CONTROL:
3067 r = kvm_has_tsc_control;
3069 case KVM_CAP_X2APIC_API:
3070 r = KVM_X2APIC_API_VALID_FLAGS;
3072 case KVM_CAP_NESTED_STATE:
3073 r = kvm_x86_ops->get_nested_state ?
3074 kvm_x86_ops->get_nested_state(NULL, 0, 0) : 0;
3083 long kvm_arch_dev_ioctl(struct file *filp,
3084 unsigned int ioctl, unsigned long arg)
3086 void __user *argp = (void __user *)arg;
3090 case KVM_GET_MSR_INDEX_LIST: {
3091 struct kvm_msr_list __user *user_msr_list = argp;
3092 struct kvm_msr_list msr_list;
3096 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3099 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3100 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3103 if (n < msr_list.nmsrs)
3106 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3107 num_msrs_to_save * sizeof(u32)))
3109 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3111 num_emulated_msrs * sizeof(u32)))
3116 case KVM_GET_SUPPORTED_CPUID:
3117 case KVM_GET_EMULATED_CPUID: {
3118 struct kvm_cpuid2 __user *cpuid_arg = argp;
3119 struct kvm_cpuid2 cpuid;
3122 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3125 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3131 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3136 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3138 if (copy_to_user(argp, &kvm_mce_cap_supported,
3139 sizeof(kvm_mce_cap_supported)))
3143 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3144 struct kvm_msr_list __user *user_msr_list = argp;
3145 struct kvm_msr_list msr_list;
3149 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3152 msr_list.nmsrs = num_msr_based_features;
3153 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3156 if (n < msr_list.nmsrs)
3159 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3160 num_msr_based_features * sizeof(u32)))
3166 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3176 static void wbinvd_ipi(void *garbage)
3181 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3183 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3186 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3188 /* Address WBINVD may be executed by guest */
3189 if (need_emulate_wbinvd(vcpu)) {
3190 if (kvm_x86_ops->has_wbinvd_exit())
3191 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3192 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3193 smp_call_function_single(vcpu->cpu,
3194 wbinvd_ipi, NULL, 1);
3197 kvm_x86_ops->vcpu_load(vcpu, cpu);
3199 /* Apply any externally detected TSC adjustments (due to suspend) */
3200 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3201 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3202 vcpu->arch.tsc_offset_adjustment = 0;
3203 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3206 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3207 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3208 rdtsc() - vcpu->arch.last_host_tsc;
3210 mark_tsc_unstable("KVM discovered backwards TSC");
3212 if (kvm_check_tsc_unstable()) {
3213 u64 offset = kvm_compute_tsc_offset(vcpu,
3214 vcpu->arch.last_guest_tsc);
3215 kvm_vcpu_write_tsc_offset(vcpu, offset);
3216 vcpu->arch.tsc_catchup = 1;
3219 if (kvm_lapic_hv_timer_in_use(vcpu))
3220 kvm_lapic_restart_hv_timer(vcpu);
3223 * On a host with synchronized TSC, there is no need to update
3224 * kvmclock on vcpu->cpu migration
3226 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3227 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3228 if (vcpu->cpu != cpu)
3229 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3233 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3236 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3238 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3241 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3243 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3244 &vcpu->arch.st.steal.preempted,
3245 offsetof(struct kvm_steal_time, preempted),
3246 sizeof(vcpu->arch.st.steal.preempted));
3249 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3253 if (vcpu->preempted)
3254 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3257 * Disable page faults because we're in atomic context here.
3258 * kvm_write_guest_offset_cached() would call might_fault()
3259 * that relies on pagefault_disable() to tell if there's a
3260 * bug. NOTE: the write to guest memory may not go through if
3261 * during postcopy live migration or if there's heavy guest
3264 pagefault_disable();
3266 * kvm_memslots() will be called by
3267 * kvm_write_guest_offset_cached() so take the srcu lock.
3269 idx = srcu_read_lock(&vcpu->kvm->srcu);
3270 kvm_steal_time_set_preempted(vcpu);
3271 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3273 kvm_x86_ops->vcpu_put(vcpu);
3274 vcpu->arch.last_host_tsc = rdtsc();
3276 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3277 * on every vmexit, but if not, we might have a stale dr6 from the
3278 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3283 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3284 struct kvm_lapic_state *s)
3286 if (vcpu->arch.apicv_active)
3287 kvm_x86_ops->sync_pir_to_irr(vcpu);
3289 return kvm_apic_get_state(vcpu, s);
3292 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3293 struct kvm_lapic_state *s)
3297 r = kvm_apic_set_state(vcpu, s);
3300 update_cr8_intercept(vcpu);
3305 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3307 return (!lapic_in_kernel(vcpu) ||
3308 kvm_apic_accept_pic_intr(vcpu));
3312 * if userspace requested an interrupt window, check that the
3313 * interrupt window is open.
3315 * No need to exit to userspace if we already have an interrupt queued.
3317 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3319 return kvm_arch_interrupt_allowed(vcpu) &&
3320 !kvm_cpu_has_interrupt(vcpu) &&
3321 !kvm_event_needs_reinjection(vcpu) &&
3322 kvm_cpu_accept_dm_intr(vcpu);
3325 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3326 struct kvm_interrupt *irq)
3328 if (irq->irq >= KVM_NR_INTERRUPTS)
3331 if (!irqchip_in_kernel(vcpu->kvm)) {
3332 kvm_queue_interrupt(vcpu, irq->irq, false);
3333 kvm_make_request(KVM_REQ_EVENT, vcpu);
3338 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3339 * fail for in-kernel 8259.
3341 if (pic_in_kernel(vcpu->kvm))
3344 if (vcpu->arch.pending_external_vector != -1)
3347 vcpu->arch.pending_external_vector = irq->irq;
3348 kvm_make_request(KVM_REQ_EVENT, vcpu);
3352 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3354 kvm_inject_nmi(vcpu);
3359 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3361 kvm_make_request(KVM_REQ_SMI, vcpu);
3366 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3367 struct kvm_tpr_access_ctl *tac)
3371 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3375 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3379 unsigned bank_num = mcg_cap & 0xff, bank;
3382 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3384 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3387 vcpu->arch.mcg_cap = mcg_cap;
3388 /* Init IA32_MCG_CTL to all 1s */
3389 if (mcg_cap & MCG_CTL_P)
3390 vcpu->arch.mcg_ctl = ~(u64)0;
3391 /* Init IA32_MCi_CTL to all 1s */
3392 for (bank = 0; bank < bank_num; bank++)
3393 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3395 if (kvm_x86_ops->setup_mce)
3396 kvm_x86_ops->setup_mce(vcpu);
3401 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3402 struct kvm_x86_mce *mce)
3404 u64 mcg_cap = vcpu->arch.mcg_cap;
3405 unsigned bank_num = mcg_cap & 0xff;
3406 u64 *banks = vcpu->arch.mce_banks;
3408 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3411 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3412 * reporting is disabled
3414 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3415 vcpu->arch.mcg_ctl != ~(u64)0)
3417 banks += 4 * mce->bank;
3419 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3420 * reporting is disabled for the bank
3422 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3424 if (mce->status & MCI_STATUS_UC) {
3425 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3426 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3427 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3430 if (banks[1] & MCI_STATUS_VAL)
3431 mce->status |= MCI_STATUS_OVER;
3432 banks[2] = mce->addr;
3433 banks[3] = mce->misc;
3434 vcpu->arch.mcg_status = mce->mcg_status;
3435 banks[1] = mce->status;
3436 kvm_queue_exception(vcpu, MC_VECTOR);
3437 } else if (!(banks[1] & MCI_STATUS_VAL)
3438 || !(banks[1] & MCI_STATUS_UC)) {
3439 if (banks[1] & MCI_STATUS_VAL)
3440 mce->status |= MCI_STATUS_OVER;
3441 banks[2] = mce->addr;
3442 banks[3] = mce->misc;
3443 banks[1] = mce->status;
3445 banks[1] |= MCI_STATUS_OVER;
3449 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3450 struct kvm_vcpu_events *events)
3455 * The API doesn't provide the instruction length for software
3456 * exceptions, so don't report them. As long as the guest RIP
3457 * isn't advanced, we should expect to encounter the exception
3460 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3461 events->exception.injected = 0;
3462 events->exception.pending = 0;
3464 events->exception.injected = vcpu->arch.exception.injected;
3465 events->exception.pending = vcpu->arch.exception.pending;
3467 * For ABI compatibility, deliberately conflate
3468 * pending and injected exceptions when
3469 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3471 if (!vcpu->kvm->arch.exception_payload_enabled)
3472 events->exception.injected |=
3473 vcpu->arch.exception.pending;
3475 events->exception.nr = vcpu->arch.exception.nr;
3476 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3477 events->exception.error_code = vcpu->arch.exception.error_code;
3478 events->exception_has_payload = vcpu->arch.exception.has_payload;
3479 events->exception_payload = vcpu->arch.exception.payload;
3481 events->interrupt.injected =
3482 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3483 events->interrupt.nr = vcpu->arch.interrupt.nr;
3484 events->interrupt.soft = 0;
3485 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3487 events->nmi.injected = vcpu->arch.nmi_injected;
3488 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3489 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3490 events->nmi.pad = 0;
3492 events->sipi_vector = 0; /* never valid when reporting to user space */
3494 events->smi.smm = is_smm(vcpu);
3495 events->smi.pending = vcpu->arch.smi_pending;
3496 events->smi.smm_inside_nmi =
3497 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3498 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3500 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3501 | KVM_VCPUEVENT_VALID_SHADOW
3502 | KVM_VCPUEVENT_VALID_SMM);
3503 if (vcpu->kvm->arch.exception_payload_enabled)
3504 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3506 memset(&events->reserved, 0, sizeof(events->reserved));
3509 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3511 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3512 struct kvm_vcpu_events *events)
3514 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3515 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3516 | KVM_VCPUEVENT_VALID_SHADOW
3517 | KVM_VCPUEVENT_VALID_SMM
3518 | KVM_VCPUEVENT_VALID_PAYLOAD))
3521 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3522 if (!vcpu->kvm->arch.exception_payload_enabled)
3524 if (events->exception.pending)
3525 events->exception.injected = 0;
3527 events->exception_has_payload = 0;
3529 events->exception.pending = 0;
3530 events->exception_has_payload = 0;
3533 if ((events->exception.injected || events->exception.pending) &&
3534 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3537 /* INITs are latched while in SMM */
3538 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3539 (events->smi.smm || events->smi.pending) &&
3540 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3544 vcpu->arch.exception.injected = events->exception.injected;
3545 vcpu->arch.exception.pending = events->exception.pending;
3546 vcpu->arch.exception.nr = events->exception.nr;
3547 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3548 vcpu->arch.exception.error_code = events->exception.error_code;
3549 vcpu->arch.exception.has_payload = events->exception_has_payload;
3550 vcpu->arch.exception.payload = events->exception_payload;
3552 vcpu->arch.interrupt.injected = events->interrupt.injected;
3553 vcpu->arch.interrupt.nr = events->interrupt.nr;
3554 vcpu->arch.interrupt.soft = events->interrupt.soft;
3555 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3556 kvm_x86_ops->set_interrupt_shadow(vcpu,
3557 events->interrupt.shadow);
3559 vcpu->arch.nmi_injected = events->nmi.injected;
3560 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3561 vcpu->arch.nmi_pending = events->nmi.pending;
3562 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3564 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3565 lapic_in_kernel(vcpu))
3566 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3568 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3569 u32 hflags = vcpu->arch.hflags;
3570 if (events->smi.smm)
3571 hflags |= HF_SMM_MASK;
3573 hflags &= ~HF_SMM_MASK;
3574 kvm_set_hflags(vcpu, hflags);
3576 vcpu->arch.smi_pending = events->smi.pending;
3578 if (events->smi.smm) {
3579 if (events->smi.smm_inside_nmi)
3580 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3582 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3583 if (lapic_in_kernel(vcpu)) {
3584 if (events->smi.latched_init)
3585 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3587 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3592 kvm_make_request(KVM_REQ_EVENT, vcpu);
3597 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3598 struct kvm_debugregs *dbgregs)
3602 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3603 kvm_get_dr(vcpu, 6, &val);
3605 dbgregs->dr7 = vcpu->arch.dr7;
3607 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3610 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3611 struct kvm_debugregs *dbgregs)
3616 if (dbgregs->dr6 & ~0xffffffffull)
3618 if (dbgregs->dr7 & ~0xffffffffull)
3621 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3622 kvm_update_dr0123(vcpu);
3623 vcpu->arch.dr6 = dbgregs->dr6;
3624 kvm_update_dr6(vcpu);
3625 vcpu->arch.dr7 = dbgregs->dr7;
3626 kvm_update_dr7(vcpu);
3631 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3633 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3635 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3636 u64 xstate_bv = xsave->header.xfeatures;
3640 * Copy legacy XSAVE area, to avoid complications with CPUID
3641 * leaves 0 and 1 in the loop below.
3643 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3646 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3647 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3650 * Copy each region from the possibly compacted offset to the
3651 * non-compacted offset.
3653 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3655 u64 feature = valid & -valid;
3656 int index = fls64(feature) - 1;
3657 void *src = get_xsave_addr(xsave, feature);
3660 u32 size, offset, ecx, edx;
3661 cpuid_count(XSTATE_CPUID, index,
3662 &size, &offset, &ecx, &edx);
3663 if (feature == XFEATURE_MASK_PKRU)
3664 memcpy(dest + offset, &vcpu->arch.pkru,
3665 sizeof(vcpu->arch.pkru));
3667 memcpy(dest + offset, src, size);
3675 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3677 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3678 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3682 * Copy legacy XSAVE area, to avoid complications with CPUID
3683 * leaves 0 and 1 in the loop below.
3685 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3687 /* Set XSTATE_BV and possibly XCOMP_BV. */
3688 xsave->header.xfeatures = xstate_bv;
3689 if (boot_cpu_has(X86_FEATURE_XSAVES))
3690 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3693 * Copy each region from the non-compacted offset to the
3694 * possibly compacted offset.
3696 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3698 u64 feature = valid & -valid;
3699 int index = fls64(feature) - 1;
3700 void *dest = get_xsave_addr(xsave, feature);
3703 u32 size, offset, ecx, edx;
3704 cpuid_count(XSTATE_CPUID, index,
3705 &size, &offset, &ecx, &edx);
3706 if (feature == XFEATURE_MASK_PKRU)
3707 memcpy(&vcpu->arch.pkru, src + offset,
3708 sizeof(vcpu->arch.pkru));
3710 memcpy(dest, src + offset, size);
3717 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3718 struct kvm_xsave *guest_xsave)
3720 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3721 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3722 fill_xsave((u8 *) guest_xsave->region, vcpu);
3724 memcpy(guest_xsave->region,
3725 &vcpu->arch.guest_fpu.state.fxsave,
3726 sizeof(struct fxregs_state));
3727 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3728 XFEATURE_MASK_FPSSE;
3732 #define XSAVE_MXCSR_OFFSET 24
3734 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3735 struct kvm_xsave *guest_xsave)
3738 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3739 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3741 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3743 * Here we allow setting states that are not present in
3744 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3745 * with old userspace.
3747 if (xstate_bv & ~kvm_supported_xcr0() ||
3748 mxcsr & ~mxcsr_feature_mask)
3750 load_xsave(vcpu, (u8 *)guest_xsave->region);
3752 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3753 mxcsr & ~mxcsr_feature_mask)
3755 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3756 guest_xsave->region, sizeof(struct fxregs_state));
3761 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3762 struct kvm_xcrs *guest_xcrs)
3764 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3765 guest_xcrs->nr_xcrs = 0;
3769 guest_xcrs->nr_xcrs = 1;
3770 guest_xcrs->flags = 0;
3771 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3772 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3775 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3776 struct kvm_xcrs *guest_xcrs)
3780 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3783 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3786 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3787 /* Only support XCR0 currently */
3788 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3789 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3790 guest_xcrs->xcrs[i].value);
3799 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3800 * stopped by the hypervisor. This function will be called from the host only.
3801 * EINVAL is returned when the host attempts to set the flag for a guest that
3802 * does not support pv clocks.
3804 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3806 if (!vcpu->arch.pv_time_enabled)
3808 vcpu->arch.pvclock_set_guest_stopped_request = true;
3809 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3813 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3814 struct kvm_enable_cap *cap)
3817 uint16_t vmcs_version;
3818 void __user *user_ptr;
3824 case KVM_CAP_HYPERV_SYNIC2:
3827 case KVM_CAP_HYPERV_SYNIC:
3828 if (!irqchip_in_kernel(vcpu->kvm))
3830 return kvm_hv_activate_synic(vcpu, cap->cap ==
3831 KVM_CAP_HYPERV_SYNIC2);
3832 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3833 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3835 user_ptr = (void __user *)(uintptr_t)cap->args[0];
3836 if (copy_to_user(user_ptr, &vmcs_version,
3837 sizeof(vmcs_version)))
3847 long kvm_arch_vcpu_ioctl(struct file *filp,
3848 unsigned int ioctl, unsigned long arg)
3850 struct kvm_vcpu *vcpu = filp->private_data;
3851 void __user *argp = (void __user *)arg;
3854 struct kvm_lapic_state *lapic;
3855 struct kvm_xsave *xsave;
3856 struct kvm_xcrs *xcrs;
3864 case KVM_GET_LAPIC: {
3866 if (!lapic_in_kernel(vcpu))
3868 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3873 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3877 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3882 case KVM_SET_LAPIC: {
3884 if (!lapic_in_kernel(vcpu))
3886 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3887 if (IS_ERR(u.lapic)) {
3888 r = PTR_ERR(u.lapic);
3892 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3895 case KVM_INTERRUPT: {
3896 struct kvm_interrupt irq;
3899 if (copy_from_user(&irq, argp, sizeof(irq)))
3901 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3905 r = kvm_vcpu_ioctl_nmi(vcpu);
3909 r = kvm_vcpu_ioctl_smi(vcpu);
3912 case KVM_SET_CPUID: {
3913 struct kvm_cpuid __user *cpuid_arg = argp;
3914 struct kvm_cpuid cpuid;
3917 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3919 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3922 case KVM_SET_CPUID2: {
3923 struct kvm_cpuid2 __user *cpuid_arg = argp;
3924 struct kvm_cpuid2 cpuid;
3927 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3929 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3930 cpuid_arg->entries);
3933 case KVM_GET_CPUID2: {
3934 struct kvm_cpuid2 __user *cpuid_arg = argp;
3935 struct kvm_cpuid2 cpuid;
3938 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3940 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3941 cpuid_arg->entries);
3945 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3950 case KVM_GET_MSRS: {
3951 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3952 r = msr_io(vcpu, argp, do_get_msr, 1);
3953 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3956 case KVM_SET_MSRS: {
3957 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3958 r = msr_io(vcpu, argp, do_set_msr, 0);
3959 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3962 case KVM_TPR_ACCESS_REPORTING: {
3963 struct kvm_tpr_access_ctl tac;
3966 if (copy_from_user(&tac, argp, sizeof(tac)))
3968 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3972 if (copy_to_user(argp, &tac, sizeof(tac)))
3977 case KVM_SET_VAPIC_ADDR: {
3978 struct kvm_vapic_addr va;
3982 if (!lapic_in_kernel(vcpu))
3985 if (copy_from_user(&va, argp, sizeof(va)))
3987 idx = srcu_read_lock(&vcpu->kvm->srcu);
3988 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3989 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3992 case KVM_X86_SETUP_MCE: {
3996 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
3998 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4001 case KVM_X86_SET_MCE: {
4002 struct kvm_x86_mce mce;
4005 if (copy_from_user(&mce, argp, sizeof(mce)))
4007 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4010 case KVM_GET_VCPU_EVENTS: {
4011 struct kvm_vcpu_events events;
4013 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4016 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4021 case KVM_SET_VCPU_EVENTS: {
4022 struct kvm_vcpu_events events;
4025 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4028 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4031 case KVM_GET_DEBUGREGS: {
4032 struct kvm_debugregs dbgregs;
4034 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4037 if (copy_to_user(argp, &dbgregs,
4038 sizeof(struct kvm_debugregs)))
4043 case KVM_SET_DEBUGREGS: {
4044 struct kvm_debugregs dbgregs;
4047 if (copy_from_user(&dbgregs, argp,
4048 sizeof(struct kvm_debugregs)))
4051 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4054 case KVM_GET_XSAVE: {
4055 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
4060 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4063 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4068 case KVM_SET_XSAVE: {
4069 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4070 if (IS_ERR(u.xsave)) {
4071 r = PTR_ERR(u.xsave);
4075 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4078 case KVM_GET_XCRS: {
4079 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
4084 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4087 if (copy_to_user(argp, u.xcrs,
4088 sizeof(struct kvm_xcrs)))
4093 case KVM_SET_XCRS: {
4094 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4095 if (IS_ERR(u.xcrs)) {
4096 r = PTR_ERR(u.xcrs);
4100 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4103 case KVM_SET_TSC_KHZ: {
4107 user_tsc_khz = (u32)arg;
4109 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4112 if (user_tsc_khz == 0)
4113 user_tsc_khz = tsc_khz;
4115 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4120 case KVM_GET_TSC_KHZ: {
4121 r = vcpu->arch.virtual_tsc_khz;
4124 case KVM_KVMCLOCK_CTRL: {
4125 r = kvm_set_guest_paused(vcpu);
4128 case KVM_ENABLE_CAP: {
4129 struct kvm_enable_cap cap;
4132 if (copy_from_user(&cap, argp, sizeof(cap)))
4134 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4137 case KVM_GET_NESTED_STATE: {
4138 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4142 if (!kvm_x86_ops->get_nested_state)
4145 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4147 if (get_user(user_data_size, &user_kvm_nested_state->size))
4150 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4155 if (r > user_data_size) {
4156 if (put_user(r, &user_kvm_nested_state->size))
4166 case KVM_SET_NESTED_STATE: {
4167 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4168 struct kvm_nested_state kvm_state;
4171 if (!kvm_x86_ops->set_nested_state)
4175 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4179 if (kvm_state.size < sizeof(kvm_state))
4182 if (kvm_state.flags &
4183 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4184 | KVM_STATE_NESTED_EVMCS))
4187 /* nested_run_pending implies guest_mode. */
4188 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4189 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4192 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4205 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4207 return VM_FAULT_SIGBUS;
4210 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4214 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4216 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4220 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4223 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4226 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4227 u32 kvm_nr_mmu_pages)
4229 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4232 mutex_lock(&kvm->slots_lock);
4234 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4235 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4237 mutex_unlock(&kvm->slots_lock);
4241 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4243 return kvm->arch.n_max_mmu_pages;
4246 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4248 struct kvm_pic *pic = kvm->arch.vpic;
4252 switch (chip->chip_id) {
4253 case KVM_IRQCHIP_PIC_MASTER:
4254 memcpy(&chip->chip.pic, &pic->pics[0],
4255 sizeof(struct kvm_pic_state));
4257 case KVM_IRQCHIP_PIC_SLAVE:
4258 memcpy(&chip->chip.pic, &pic->pics[1],
4259 sizeof(struct kvm_pic_state));
4261 case KVM_IRQCHIP_IOAPIC:
4262 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4271 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4273 struct kvm_pic *pic = kvm->arch.vpic;
4277 switch (chip->chip_id) {
4278 case KVM_IRQCHIP_PIC_MASTER:
4279 spin_lock(&pic->lock);
4280 memcpy(&pic->pics[0], &chip->chip.pic,
4281 sizeof(struct kvm_pic_state));
4282 spin_unlock(&pic->lock);
4284 case KVM_IRQCHIP_PIC_SLAVE:
4285 spin_lock(&pic->lock);
4286 memcpy(&pic->pics[1], &chip->chip.pic,
4287 sizeof(struct kvm_pic_state));
4288 spin_unlock(&pic->lock);
4290 case KVM_IRQCHIP_IOAPIC:
4291 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4297 kvm_pic_update_irq(pic);
4301 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4303 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4305 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4307 mutex_lock(&kps->lock);
4308 memcpy(ps, &kps->channels, sizeof(*ps));
4309 mutex_unlock(&kps->lock);
4313 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4316 struct kvm_pit *pit = kvm->arch.vpit;
4318 mutex_lock(&pit->pit_state.lock);
4319 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4320 for (i = 0; i < 3; i++)
4321 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4322 mutex_unlock(&pit->pit_state.lock);
4326 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4328 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4329 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4330 sizeof(ps->channels));
4331 ps->flags = kvm->arch.vpit->pit_state.flags;
4332 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4333 memset(&ps->reserved, 0, sizeof(ps->reserved));
4337 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4341 u32 prev_legacy, cur_legacy;
4342 struct kvm_pit *pit = kvm->arch.vpit;
4344 mutex_lock(&pit->pit_state.lock);
4345 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4346 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4347 if (!prev_legacy && cur_legacy)
4349 memcpy(&pit->pit_state.channels, &ps->channels,
4350 sizeof(pit->pit_state.channels));
4351 pit->pit_state.flags = ps->flags;
4352 for (i = 0; i < 3; i++)
4353 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4355 mutex_unlock(&pit->pit_state.lock);
4359 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4360 struct kvm_reinject_control *control)
4362 struct kvm_pit *pit = kvm->arch.vpit;
4367 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4368 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4369 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4371 mutex_lock(&pit->pit_state.lock);
4372 kvm_pit_set_reinject(pit, control->pit_reinject);
4373 mutex_unlock(&pit->pit_state.lock);
4379 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4380 * @kvm: kvm instance
4381 * @log: slot id and address to which we copy the log
4383 * Steps 1-4 below provide general overview of dirty page logging. See
4384 * kvm_get_dirty_log_protect() function description for additional details.
4386 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4387 * always flush the TLB (step 4) even if previous step failed and the dirty
4388 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4389 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4390 * writes will be marked dirty for next log read.
4392 * 1. Take a snapshot of the bit and clear it if needed.
4393 * 2. Write protect the corresponding page.
4394 * 3. Copy the snapshot to the userspace.
4395 * 4. Flush TLB's if needed.
4397 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4399 bool is_dirty = false;
4402 mutex_lock(&kvm->slots_lock);
4405 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4407 if (kvm_x86_ops->flush_log_dirty)
4408 kvm_x86_ops->flush_log_dirty(kvm);
4410 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
4413 * All the TLBs can be flushed out of mmu lock, see the comments in
4414 * kvm_mmu_slot_remove_write_access().
4416 lockdep_assert_held(&kvm->slots_lock);
4418 kvm_flush_remote_tlbs(kvm);
4420 mutex_unlock(&kvm->slots_lock);
4424 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4427 if (!irqchip_in_kernel(kvm))
4430 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4431 irq_event->irq, irq_event->level,
4436 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4437 struct kvm_enable_cap *cap)
4445 case KVM_CAP_DISABLE_QUIRKS:
4446 kvm->arch.disabled_quirks = cap->args[0];
4449 case KVM_CAP_SPLIT_IRQCHIP: {
4450 mutex_lock(&kvm->lock);
4452 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4453 goto split_irqchip_unlock;
4455 if (irqchip_in_kernel(kvm))
4456 goto split_irqchip_unlock;
4457 if (kvm->created_vcpus)
4458 goto split_irqchip_unlock;
4459 r = kvm_setup_empty_irq_routing(kvm);
4461 goto split_irqchip_unlock;
4462 /* Pairs with irqchip_in_kernel. */
4464 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4465 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4467 split_irqchip_unlock:
4468 mutex_unlock(&kvm->lock);
4471 case KVM_CAP_X2APIC_API:
4473 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4476 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4477 kvm->arch.x2apic_format = true;
4478 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4479 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4483 case KVM_CAP_X86_DISABLE_EXITS:
4485 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4488 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4489 kvm_can_mwait_in_guest())
4490 kvm->arch.mwait_in_guest = true;
4491 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4492 kvm->arch.hlt_in_guest = true;
4493 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4494 kvm->arch.pause_in_guest = true;
4497 case KVM_CAP_MSR_PLATFORM_INFO:
4498 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4501 case KVM_CAP_EXCEPTION_PAYLOAD:
4502 kvm->arch.exception_payload_enabled = cap->args[0];
4512 long kvm_arch_vm_ioctl(struct file *filp,
4513 unsigned int ioctl, unsigned long arg)
4515 struct kvm *kvm = filp->private_data;
4516 void __user *argp = (void __user *)arg;
4519 * This union makes it completely explicit to gcc-3.x
4520 * that these two variables' stack usage should be
4521 * combined, not added together.
4524 struct kvm_pit_state ps;
4525 struct kvm_pit_state2 ps2;
4526 struct kvm_pit_config pit_config;
4530 case KVM_SET_TSS_ADDR:
4531 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4533 case KVM_SET_IDENTITY_MAP_ADDR: {
4536 mutex_lock(&kvm->lock);
4538 if (kvm->created_vcpus)
4539 goto set_identity_unlock;
4541 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4542 goto set_identity_unlock;
4543 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4544 set_identity_unlock:
4545 mutex_unlock(&kvm->lock);
4548 case KVM_SET_NR_MMU_PAGES:
4549 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4551 case KVM_GET_NR_MMU_PAGES:
4552 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4554 case KVM_CREATE_IRQCHIP: {
4555 mutex_lock(&kvm->lock);
4558 if (irqchip_in_kernel(kvm))
4559 goto create_irqchip_unlock;
4562 if (kvm->created_vcpus)
4563 goto create_irqchip_unlock;
4565 r = kvm_pic_init(kvm);
4567 goto create_irqchip_unlock;
4569 r = kvm_ioapic_init(kvm);
4571 kvm_pic_destroy(kvm);
4572 goto create_irqchip_unlock;
4575 r = kvm_setup_default_irq_routing(kvm);
4577 kvm_ioapic_destroy(kvm);
4578 kvm_pic_destroy(kvm);
4579 goto create_irqchip_unlock;
4581 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4583 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4584 create_irqchip_unlock:
4585 mutex_unlock(&kvm->lock);
4588 case KVM_CREATE_PIT:
4589 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4591 case KVM_CREATE_PIT2:
4593 if (copy_from_user(&u.pit_config, argp,
4594 sizeof(struct kvm_pit_config)))
4597 mutex_lock(&kvm->lock);
4600 goto create_pit_unlock;
4602 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4606 mutex_unlock(&kvm->lock);
4608 case KVM_GET_IRQCHIP: {
4609 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4610 struct kvm_irqchip *chip;
4612 chip = memdup_user(argp, sizeof(*chip));
4619 if (!irqchip_kernel(kvm))
4620 goto get_irqchip_out;
4621 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4623 goto get_irqchip_out;
4625 if (copy_to_user(argp, chip, sizeof(*chip)))
4626 goto get_irqchip_out;
4632 case KVM_SET_IRQCHIP: {
4633 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4634 struct kvm_irqchip *chip;
4636 chip = memdup_user(argp, sizeof(*chip));
4643 if (!irqchip_kernel(kvm))
4644 goto set_irqchip_out;
4645 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4647 goto set_irqchip_out;
4655 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4658 if (!kvm->arch.vpit)
4660 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4664 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4671 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4674 if (!kvm->arch.vpit)
4676 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4679 case KVM_GET_PIT2: {
4681 if (!kvm->arch.vpit)
4683 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4687 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4692 case KVM_SET_PIT2: {
4694 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4697 if (!kvm->arch.vpit)
4699 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4702 case KVM_REINJECT_CONTROL: {
4703 struct kvm_reinject_control control;
4705 if (copy_from_user(&control, argp, sizeof(control)))
4707 r = kvm_vm_ioctl_reinject(kvm, &control);
4710 case KVM_SET_BOOT_CPU_ID:
4712 mutex_lock(&kvm->lock);
4713 if (kvm->created_vcpus)
4716 kvm->arch.bsp_vcpu_id = arg;
4717 mutex_unlock(&kvm->lock);
4719 case KVM_XEN_HVM_CONFIG: {
4720 struct kvm_xen_hvm_config xhc;
4722 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4727 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4731 case KVM_SET_CLOCK: {
4732 struct kvm_clock_data user_ns;
4736 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4745 * TODO: userspace has to take care of races with VCPU_RUN, so
4746 * kvm_gen_update_masterclock() can be cut down to locked
4747 * pvclock_update_vm_gtod_copy().
4749 kvm_gen_update_masterclock(kvm);
4750 now_ns = get_kvmclock_ns(kvm);
4751 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4752 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4755 case KVM_GET_CLOCK: {
4756 struct kvm_clock_data user_ns;
4759 now_ns = get_kvmclock_ns(kvm);
4760 user_ns.clock = now_ns;
4761 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4762 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4765 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4770 case KVM_ENABLE_CAP: {
4771 struct kvm_enable_cap cap;
4774 if (copy_from_user(&cap, argp, sizeof(cap)))
4776 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4779 case KVM_MEMORY_ENCRYPT_OP: {
4781 if (kvm_x86_ops->mem_enc_op)
4782 r = kvm_x86_ops->mem_enc_op(kvm, argp);
4785 case KVM_MEMORY_ENCRYPT_REG_REGION: {
4786 struct kvm_enc_region region;
4789 if (copy_from_user(®ion, argp, sizeof(region)))
4793 if (kvm_x86_ops->mem_enc_reg_region)
4794 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
4797 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4798 struct kvm_enc_region region;
4801 if (copy_from_user(®ion, argp, sizeof(region)))
4805 if (kvm_x86_ops->mem_enc_unreg_region)
4806 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
4809 case KVM_HYPERV_EVENTFD: {
4810 struct kvm_hyperv_eventfd hvevfd;
4813 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4815 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4825 static void kvm_init_msr_list(void)
4830 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4831 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4835 * Even MSRs that are valid in the host may not be exposed
4836 * to the guests in some cases.
4838 switch (msrs_to_save[i]) {
4839 case MSR_IA32_BNDCFGS:
4840 if (!kvm_mpx_supported())
4844 if (!kvm_x86_ops->rdtscp_supported())
4852 msrs_to_save[j] = msrs_to_save[i];
4855 num_msrs_to_save = j;
4857 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4858 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4862 emulated_msrs[j] = emulated_msrs[i];
4865 num_emulated_msrs = j;
4867 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4868 struct kvm_msr_entry msr;
4870 msr.index = msr_based_features[i];
4871 if (kvm_get_msr_feature(&msr))
4875 msr_based_features[j] = msr_based_features[i];
4878 num_msr_based_features = j;
4881 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4889 if (!(lapic_in_kernel(vcpu) &&
4890 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4891 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4902 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4909 if (!(lapic_in_kernel(vcpu) &&
4910 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4912 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4914 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4924 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4925 struct kvm_segment *var, int seg)
4927 kvm_x86_ops->set_segment(vcpu, var, seg);
4930 void kvm_get_segment(struct kvm_vcpu *vcpu,
4931 struct kvm_segment *var, int seg)
4933 kvm_x86_ops->get_segment(vcpu, var, seg);
4936 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4937 struct x86_exception *exception)
4941 BUG_ON(!mmu_is_nested(vcpu));
4943 /* NPT walks are always user-walks */
4944 access |= PFERR_USER_MASK;
4945 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
4950 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4951 struct x86_exception *exception)
4953 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4954 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4957 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4958 struct x86_exception *exception)
4960 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4961 access |= PFERR_FETCH_MASK;
4962 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4965 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4966 struct x86_exception *exception)
4968 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4969 access |= PFERR_WRITE_MASK;
4970 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4973 /* uses this to access any guest's mapped memory without checking CPL */
4974 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4975 struct x86_exception *exception)
4977 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4980 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4981 struct kvm_vcpu *vcpu, u32 access,
4982 struct x86_exception *exception)
4985 int r = X86EMUL_CONTINUE;
4988 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4990 unsigned offset = addr & (PAGE_SIZE-1);
4991 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4994 if (gpa == UNMAPPED_GVA)
4995 return X86EMUL_PROPAGATE_FAULT;
4996 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4999 r = X86EMUL_IO_NEEDED;
5011 /* used for instruction fetching */
5012 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5013 gva_t addr, void *val, unsigned int bytes,
5014 struct x86_exception *exception)
5016 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5017 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5021 /* Inline kvm_read_guest_virt_helper for speed. */
5022 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5024 if (unlikely(gpa == UNMAPPED_GVA))
5025 return X86EMUL_PROPAGATE_FAULT;
5027 offset = addr & (PAGE_SIZE-1);
5028 if (WARN_ON(offset + bytes > PAGE_SIZE))
5029 bytes = (unsigned)PAGE_SIZE - offset;
5030 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5032 if (unlikely(ret < 0))
5033 return X86EMUL_IO_NEEDED;
5035 return X86EMUL_CONTINUE;
5038 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5039 gva_t addr, void *val, unsigned int bytes,
5040 struct x86_exception *exception)
5042 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5044 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5047 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5049 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5050 gva_t addr, void *val, unsigned int bytes,
5051 struct x86_exception *exception, bool system)
5053 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5056 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5057 access |= PFERR_USER_MASK;
5059 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5062 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5063 unsigned long addr, void *val, unsigned int bytes)
5065 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5066 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5068 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5071 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5072 struct kvm_vcpu *vcpu, u32 access,
5073 struct x86_exception *exception)
5076 int r = X86EMUL_CONTINUE;
5079 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5082 unsigned offset = addr & (PAGE_SIZE-1);
5083 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5086 if (gpa == UNMAPPED_GVA)
5087 return X86EMUL_PROPAGATE_FAULT;
5088 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5090 r = X86EMUL_IO_NEEDED;
5102 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5103 unsigned int bytes, struct x86_exception *exception,
5106 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5107 u32 access = PFERR_WRITE_MASK;
5109 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5110 access |= PFERR_USER_MASK;
5112 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5116 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5117 unsigned int bytes, struct x86_exception *exception)
5119 /* kvm_write_guest_virt_system can pull in tons of pages. */
5120 vcpu->arch.l1tf_flush_l1d = true;
5122 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5123 PFERR_WRITE_MASK, exception);
5125 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5127 int handle_ud(struct kvm_vcpu *vcpu)
5129 int emul_type = EMULTYPE_TRAP_UD;
5130 enum emulation_result er;
5131 char sig[5]; /* ud2; .ascii "kvm" */
5132 struct x86_exception e;
5134 if (force_emulation_prefix &&
5135 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5136 sig, sizeof(sig), &e) == 0 &&
5137 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5138 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5142 er = kvm_emulate_instruction(vcpu, emul_type);
5143 if (er == EMULATE_USER_EXIT)
5145 if (er != EMULATE_DONE)
5146 kvm_queue_exception(vcpu, UD_VECTOR);
5149 EXPORT_SYMBOL_GPL(handle_ud);
5151 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5152 gpa_t gpa, bool write)
5154 /* For APIC access vmexit */
5155 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5158 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5159 trace_vcpu_match_mmio(gva, gpa, write, true);
5166 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5167 gpa_t *gpa, struct x86_exception *exception,
5170 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5171 | (write ? PFERR_WRITE_MASK : 0);
5174 * currently PKRU is only applied to ept enabled guest so
5175 * there is no pkey in EPT page table for L1 guest or EPT
5176 * shadow page table for L2 guest.
5178 if (vcpu_match_mmio_gva(vcpu, gva)
5179 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5180 vcpu->arch.access, 0, access)) {
5181 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5182 (gva & (PAGE_SIZE - 1));
5183 trace_vcpu_match_mmio(gva, *gpa, write, false);
5187 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5189 if (*gpa == UNMAPPED_GVA)
5192 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5195 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5196 const void *val, int bytes)
5200 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5203 kvm_page_track_write(vcpu, gpa, val, bytes);
5207 struct read_write_emulator_ops {
5208 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5210 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5211 void *val, int bytes);
5212 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5213 int bytes, void *val);
5214 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5215 void *val, int bytes);
5219 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5221 if (vcpu->mmio_read_completed) {
5222 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5223 vcpu->mmio_fragments[0].gpa, val);
5224 vcpu->mmio_read_completed = 0;
5231 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5232 void *val, int bytes)
5234 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5237 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5238 void *val, int bytes)
5240 return emulator_write_phys(vcpu, gpa, val, bytes);
5243 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5245 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5246 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5249 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5250 void *val, int bytes)
5252 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5253 return X86EMUL_IO_NEEDED;
5256 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5257 void *val, int bytes)
5259 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5261 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5262 return X86EMUL_CONTINUE;
5265 static const struct read_write_emulator_ops read_emultor = {
5266 .read_write_prepare = read_prepare,
5267 .read_write_emulate = read_emulate,
5268 .read_write_mmio = vcpu_mmio_read,
5269 .read_write_exit_mmio = read_exit_mmio,
5272 static const struct read_write_emulator_ops write_emultor = {
5273 .read_write_emulate = write_emulate,
5274 .read_write_mmio = write_mmio,
5275 .read_write_exit_mmio = write_exit_mmio,
5279 static int emulator_read_write_onepage(unsigned long addr, void *val,
5281 struct x86_exception *exception,
5282 struct kvm_vcpu *vcpu,
5283 const struct read_write_emulator_ops *ops)
5287 bool write = ops->write;
5288 struct kvm_mmio_fragment *frag;
5289 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5292 * If the exit was due to a NPF we may already have a GPA.
5293 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5294 * Note, this cannot be used on string operations since string
5295 * operation using rep will only have the initial GPA from the NPF
5298 if (vcpu->arch.gpa_available &&
5299 emulator_can_use_gpa(ctxt) &&
5300 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5301 gpa = vcpu->arch.gpa_val;
5302 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5304 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5306 return X86EMUL_PROPAGATE_FAULT;
5309 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5310 return X86EMUL_CONTINUE;
5313 * Is this MMIO handled locally?
5315 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5316 if (handled == bytes)
5317 return X86EMUL_CONTINUE;
5323 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5324 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5328 return X86EMUL_CONTINUE;
5331 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5333 void *val, unsigned int bytes,
5334 struct x86_exception *exception,
5335 const struct read_write_emulator_ops *ops)
5337 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5341 if (ops->read_write_prepare &&
5342 ops->read_write_prepare(vcpu, val, bytes))
5343 return X86EMUL_CONTINUE;
5345 vcpu->mmio_nr_fragments = 0;
5347 /* Crossing a page boundary? */
5348 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5351 now = -addr & ~PAGE_MASK;
5352 rc = emulator_read_write_onepage(addr, val, now, exception,
5355 if (rc != X86EMUL_CONTINUE)
5358 if (ctxt->mode != X86EMUL_MODE_PROT64)
5364 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5366 if (rc != X86EMUL_CONTINUE)
5369 if (!vcpu->mmio_nr_fragments)
5372 gpa = vcpu->mmio_fragments[0].gpa;
5374 vcpu->mmio_needed = 1;
5375 vcpu->mmio_cur_fragment = 0;
5377 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5378 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5379 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5380 vcpu->run->mmio.phys_addr = gpa;
5382 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5385 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5389 struct x86_exception *exception)
5391 return emulator_read_write(ctxt, addr, val, bytes,
5392 exception, &read_emultor);
5395 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5399 struct x86_exception *exception)
5401 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5402 exception, &write_emultor);
5405 #define CMPXCHG_TYPE(t, ptr, old, new) \
5406 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5408 #ifdef CONFIG_X86_64
5409 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5411 # define CMPXCHG64(ptr, old, new) \
5412 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5415 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5420 struct x86_exception *exception)
5422 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5428 /* guests cmpxchg8b have to be emulated atomically */
5429 if (bytes > 8 || (bytes & (bytes - 1)))
5432 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5434 if (gpa == UNMAPPED_GVA ||
5435 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5438 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5441 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5442 if (is_error_page(page))
5445 kaddr = kmap_atomic(page);
5446 kaddr += offset_in_page(gpa);
5449 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5452 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5455 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5458 exchanged = CMPXCHG64(kaddr, old, new);
5463 kunmap_atomic(kaddr);
5464 kvm_release_page_dirty(page);
5467 return X86EMUL_CMPXCHG_FAILED;
5469 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5470 kvm_page_track_write(vcpu, gpa, new, bytes);
5472 return X86EMUL_CONTINUE;
5475 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5477 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5480 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5484 for (i = 0; i < vcpu->arch.pio.count; i++) {
5485 if (vcpu->arch.pio.in)
5486 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5487 vcpu->arch.pio.size, pd);
5489 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5490 vcpu->arch.pio.port, vcpu->arch.pio.size,
5494 pd += vcpu->arch.pio.size;
5499 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5500 unsigned short port, void *val,
5501 unsigned int count, bool in)
5503 vcpu->arch.pio.port = port;
5504 vcpu->arch.pio.in = in;
5505 vcpu->arch.pio.count = count;
5506 vcpu->arch.pio.size = size;
5508 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5509 vcpu->arch.pio.count = 0;
5513 vcpu->run->exit_reason = KVM_EXIT_IO;
5514 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5515 vcpu->run->io.size = size;
5516 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5517 vcpu->run->io.count = count;
5518 vcpu->run->io.port = port;
5523 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5524 int size, unsigned short port, void *val,
5527 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5530 if (vcpu->arch.pio.count)
5533 memset(vcpu->arch.pio_data, 0, size * count);
5535 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5538 memcpy(val, vcpu->arch.pio_data, size * count);
5539 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5540 vcpu->arch.pio.count = 0;
5547 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5548 int size, unsigned short port,
5549 const void *val, unsigned int count)
5551 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5553 memcpy(vcpu->arch.pio_data, val, size * count);
5554 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5555 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5558 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5560 return kvm_x86_ops->get_segment_base(vcpu, seg);
5563 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5565 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5568 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5570 if (!need_emulate_wbinvd(vcpu))
5571 return X86EMUL_CONTINUE;
5573 if (kvm_x86_ops->has_wbinvd_exit()) {
5574 int cpu = get_cpu();
5576 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5577 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5578 wbinvd_ipi, NULL, 1);
5580 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5583 return X86EMUL_CONTINUE;
5586 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5588 kvm_emulate_wbinvd_noskip(vcpu);
5589 return kvm_skip_emulated_instruction(vcpu);
5591 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5595 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5597 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5600 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5601 unsigned long *dest)
5603 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5606 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5607 unsigned long value)
5610 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5613 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5615 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5618 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5620 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5621 unsigned long value;
5625 value = kvm_read_cr0(vcpu);
5628 value = vcpu->arch.cr2;
5631 value = kvm_read_cr3(vcpu);
5634 value = kvm_read_cr4(vcpu);
5637 value = kvm_get_cr8(vcpu);
5640 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5647 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5649 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5654 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5657 vcpu->arch.cr2 = val;
5660 res = kvm_set_cr3(vcpu, val);
5663 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5666 res = kvm_set_cr8(vcpu, val);
5669 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5676 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5678 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5681 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5683 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5686 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5688 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5691 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5693 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5696 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5698 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5701 static unsigned long emulator_get_cached_segment_base(
5702 struct x86_emulate_ctxt *ctxt, int seg)
5704 return get_segment_base(emul_to_vcpu(ctxt), seg);
5707 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5708 struct desc_struct *desc, u32 *base3,
5711 struct kvm_segment var;
5713 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5714 *selector = var.selector;
5717 memset(desc, 0, sizeof(*desc));
5725 set_desc_limit(desc, var.limit);
5726 set_desc_base(desc, (unsigned long)var.base);
5727 #ifdef CONFIG_X86_64
5729 *base3 = var.base >> 32;
5731 desc->type = var.type;
5733 desc->dpl = var.dpl;
5734 desc->p = var.present;
5735 desc->avl = var.avl;
5743 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5744 struct desc_struct *desc, u32 base3,
5747 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5748 struct kvm_segment var;
5750 var.selector = selector;
5751 var.base = get_desc_base(desc);
5752 #ifdef CONFIG_X86_64
5753 var.base |= ((u64)base3) << 32;
5755 var.limit = get_desc_limit(desc);
5757 var.limit = (var.limit << 12) | 0xfff;
5758 var.type = desc->type;
5759 var.dpl = desc->dpl;
5764 var.avl = desc->avl;
5765 var.present = desc->p;
5766 var.unusable = !var.present;
5769 kvm_set_segment(vcpu, &var, seg);
5773 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5774 u32 msr_index, u64 *pdata)
5776 struct msr_data msr;
5779 msr.index = msr_index;
5780 msr.host_initiated = false;
5781 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5789 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5790 u32 msr_index, u64 data)
5792 struct msr_data msr;
5795 msr.index = msr_index;
5796 msr.host_initiated = false;
5797 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5800 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5802 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5804 return vcpu->arch.smbase;
5807 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5809 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5811 vcpu->arch.smbase = smbase;
5814 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5817 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5820 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5821 u32 pmc, u64 *pdata)
5823 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5826 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5828 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5831 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5832 struct x86_instruction_info *info,
5833 enum x86_intercept_stage stage)
5835 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5838 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5839 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5841 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5844 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5846 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5849 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5851 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5854 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5856 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5859 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5861 return emul_to_vcpu(ctxt)->arch.hflags;
5864 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5866 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5869 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5871 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5874 static const struct x86_emulate_ops emulate_ops = {
5875 .read_gpr = emulator_read_gpr,
5876 .write_gpr = emulator_write_gpr,
5877 .read_std = emulator_read_std,
5878 .write_std = emulator_write_std,
5879 .read_phys = kvm_read_guest_phys_system,
5880 .fetch = kvm_fetch_guest_virt,
5881 .read_emulated = emulator_read_emulated,
5882 .write_emulated = emulator_write_emulated,
5883 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5884 .invlpg = emulator_invlpg,
5885 .pio_in_emulated = emulator_pio_in_emulated,
5886 .pio_out_emulated = emulator_pio_out_emulated,
5887 .get_segment = emulator_get_segment,
5888 .set_segment = emulator_set_segment,
5889 .get_cached_segment_base = emulator_get_cached_segment_base,
5890 .get_gdt = emulator_get_gdt,
5891 .get_idt = emulator_get_idt,
5892 .set_gdt = emulator_set_gdt,
5893 .set_idt = emulator_set_idt,
5894 .get_cr = emulator_get_cr,
5895 .set_cr = emulator_set_cr,
5896 .cpl = emulator_get_cpl,
5897 .get_dr = emulator_get_dr,
5898 .set_dr = emulator_set_dr,
5899 .get_smbase = emulator_get_smbase,
5900 .set_smbase = emulator_set_smbase,
5901 .set_msr = emulator_set_msr,
5902 .get_msr = emulator_get_msr,
5903 .check_pmc = emulator_check_pmc,
5904 .read_pmc = emulator_read_pmc,
5905 .halt = emulator_halt,
5906 .wbinvd = emulator_wbinvd,
5907 .fix_hypercall = emulator_fix_hypercall,
5908 .intercept = emulator_intercept,
5909 .get_cpuid = emulator_get_cpuid,
5910 .set_nmi_mask = emulator_set_nmi_mask,
5911 .get_hflags = emulator_get_hflags,
5912 .set_hflags = emulator_set_hflags,
5913 .pre_leave_smm = emulator_pre_leave_smm,
5916 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5918 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5920 * an sti; sti; sequence only disable interrupts for the first
5921 * instruction. So, if the last instruction, be it emulated or
5922 * not, left the system with the INT_STI flag enabled, it
5923 * means that the last instruction is an sti. We should not
5924 * leave the flag on in this case. The same goes for mov ss
5926 if (int_shadow & mask)
5928 if (unlikely(int_shadow || mask)) {
5929 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5931 kvm_make_request(KVM_REQ_EVENT, vcpu);
5935 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5937 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5938 if (ctxt->exception.vector == PF_VECTOR)
5939 return kvm_propagate_fault(vcpu, &ctxt->exception);
5941 if (ctxt->exception.error_code_valid)
5942 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5943 ctxt->exception.error_code);
5945 kvm_queue_exception(vcpu, ctxt->exception.vector);
5949 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5951 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5954 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5956 ctxt->eflags = kvm_get_rflags(vcpu);
5957 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5959 ctxt->eip = kvm_rip_read(vcpu);
5960 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5961 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5962 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5963 cs_db ? X86EMUL_MODE_PROT32 :
5964 X86EMUL_MODE_PROT16;
5965 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5966 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5967 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5969 init_decode_cache(ctxt);
5970 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5973 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5975 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5978 init_emulate_ctxt(vcpu);
5982 ctxt->_eip = ctxt->eip + inc_eip;
5983 ret = emulate_int_real(ctxt, irq);
5985 if (ret != X86EMUL_CONTINUE)
5986 return EMULATE_FAIL;
5988 ctxt->eip = ctxt->_eip;
5989 kvm_rip_write(vcpu, ctxt->eip);
5990 kvm_set_rflags(vcpu, ctxt->eflags);
5992 return EMULATE_DONE;
5994 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5996 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
5998 int r = EMULATE_DONE;
6000 ++vcpu->stat.insn_emulation_fail;
6001 trace_kvm_emulate_insn_failed(vcpu);
6003 if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6004 return EMULATE_FAIL;
6006 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6007 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6008 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6009 vcpu->run->internal.ndata = 0;
6010 r = EMULATE_USER_EXIT;
6013 kvm_queue_exception(vcpu, UD_VECTOR);
6018 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6019 bool write_fault_to_shadow_pgtable,
6025 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6028 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6031 if (!vcpu->arch.mmu->direct_map) {
6033 * Write permission should be allowed since only
6034 * write access need to be emulated.
6036 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6039 * If the mapping is invalid in guest, let cpu retry
6040 * it to generate fault.
6042 if (gpa == UNMAPPED_GVA)
6047 * Do not retry the unhandleable instruction if it faults on the
6048 * readonly host memory, otherwise it will goto a infinite loop:
6049 * retry instruction -> write #PF -> emulation fail -> retry
6050 * instruction -> ...
6052 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6055 * If the instruction failed on the error pfn, it can not be fixed,
6056 * report the error to userspace.
6058 if (is_error_noslot_pfn(pfn))
6061 kvm_release_pfn_clean(pfn);
6063 /* The instructions are well-emulated on direct mmu. */
6064 if (vcpu->arch.mmu->direct_map) {
6065 unsigned int indirect_shadow_pages;
6067 spin_lock(&vcpu->kvm->mmu_lock);
6068 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6069 spin_unlock(&vcpu->kvm->mmu_lock);
6071 if (indirect_shadow_pages)
6072 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6078 * if emulation was due to access to shadowed page table
6079 * and it failed try to unshadow page and re-enter the
6080 * guest to let CPU execute the instruction.
6082 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6085 * If the access faults on its page table, it can not
6086 * be fixed by unprotecting shadow page and it should
6087 * be reported to userspace.
6089 return !write_fault_to_shadow_pgtable;
6092 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6093 unsigned long cr2, int emulation_type)
6095 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6096 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6098 last_retry_eip = vcpu->arch.last_retry_eip;
6099 last_retry_addr = vcpu->arch.last_retry_addr;
6102 * If the emulation is caused by #PF and it is non-page_table
6103 * writing instruction, it means the VM-EXIT is caused by shadow
6104 * page protected, we can zap the shadow page and retry this
6105 * instruction directly.
6107 * Note: if the guest uses a non-page-table modifying instruction
6108 * on the PDE that points to the instruction, then we will unmap
6109 * the instruction and go to an infinite loop. So, we cache the
6110 * last retried eip and the last fault address, if we meet the eip
6111 * and the address again, we can break out of the potential infinite
6114 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6116 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6119 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6122 if (x86_page_table_writing_insn(ctxt))
6125 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6128 vcpu->arch.last_retry_eip = ctxt->eip;
6129 vcpu->arch.last_retry_addr = cr2;
6131 if (!vcpu->arch.mmu->direct_map)
6132 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6134 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6139 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6140 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6142 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6144 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6145 /* This is a good place to trace that we are exiting SMM. */
6146 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6148 /* Process a latched INIT or SMI, if any. */
6149 kvm_make_request(KVM_REQ_EVENT, vcpu);
6152 kvm_mmu_reset_context(vcpu);
6155 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
6157 unsigned changed = vcpu->arch.hflags ^ emul_flags;
6159 vcpu->arch.hflags = emul_flags;
6161 if (changed & HF_SMM_MASK)
6162 kvm_smm_changed(vcpu);
6165 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6174 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6175 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6180 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6182 struct kvm_run *kvm_run = vcpu->run;
6184 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6185 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6186 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6187 kvm_run->debug.arch.exception = DB_VECTOR;
6188 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6189 *r = EMULATE_USER_EXIT;
6191 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6195 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6197 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6198 int r = EMULATE_DONE;
6200 kvm_x86_ops->skip_emulated_instruction(vcpu);
6203 * rflags is the old, "raw" value of the flags. The new value has
6204 * not been saved yet.
6206 * This is correct even for TF set by the guest, because "the
6207 * processor will not generate this exception after the instruction
6208 * that sets the TF flag".
6210 if (unlikely(rflags & X86_EFLAGS_TF))
6211 kvm_vcpu_do_singlestep(vcpu, &r);
6212 return r == EMULATE_DONE;
6214 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6216 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6218 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6219 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6220 struct kvm_run *kvm_run = vcpu->run;
6221 unsigned long eip = kvm_get_linear_rip(vcpu);
6222 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6223 vcpu->arch.guest_debug_dr7,
6227 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6228 kvm_run->debug.arch.pc = eip;
6229 kvm_run->debug.arch.exception = DB_VECTOR;
6230 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6231 *r = EMULATE_USER_EXIT;
6236 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6237 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6238 unsigned long eip = kvm_get_linear_rip(vcpu);
6239 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6244 vcpu->arch.dr6 &= ~15;
6245 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6246 kvm_queue_exception(vcpu, DB_VECTOR);
6255 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6257 switch (ctxt->opcode_len) {
6264 case 0xe6: /* OUT */
6268 case 0x6c: /* INS */
6270 case 0x6e: /* OUTS */
6277 case 0x33: /* RDPMC */
6286 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6293 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6294 bool writeback = true;
6295 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6297 vcpu->arch.l1tf_flush_l1d = true;
6300 * Clear write_fault_to_shadow_pgtable here to ensure it is
6303 vcpu->arch.write_fault_to_shadow_pgtable = false;
6304 kvm_clear_exception_queue(vcpu);
6306 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6307 init_emulate_ctxt(vcpu);
6310 * We will reenter on the same instruction since
6311 * we do not set complete_userspace_io. This does not
6312 * handle watchpoints yet, those would be handled in
6315 if (!(emulation_type & EMULTYPE_SKIP) &&
6316 kvm_vcpu_check_breakpoint(vcpu, &r))
6319 ctxt->interruptibility = 0;
6320 ctxt->have_exception = false;
6321 ctxt->exception.vector = -1;
6322 ctxt->perm_ok = false;
6324 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6326 r = x86_decode_insn(ctxt, insn, insn_len);
6328 trace_kvm_emulate_insn_start(vcpu);
6329 ++vcpu->stat.insn_emulation;
6330 if (r != EMULATION_OK) {
6331 if (emulation_type & EMULTYPE_TRAP_UD)
6332 return EMULATE_FAIL;
6333 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6335 return EMULATE_DONE;
6336 if (ctxt->have_exception && inject_emulated_exception(vcpu))
6337 return EMULATE_DONE;
6338 if (emulation_type & EMULTYPE_SKIP)
6339 return EMULATE_FAIL;
6340 return handle_emulation_failure(vcpu, emulation_type);
6344 if ((emulation_type & EMULTYPE_VMWARE) &&
6345 !is_vmware_backdoor_opcode(ctxt))
6346 return EMULATE_FAIL;
6348 if (emulation_type & EMULTYPE_SKIP) {
6349 kvm_rip_write(vcpu, ctxt->_eip);
6350 if (ctxt->eflags & X86_EFLAGS_RF)
6351 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6352 return EMULATE_DONE;
6355 if (retry_instruction(ctxt, cr2, emulation_type))
6356 return EMULATE_DONE;
6358 /* this is needed for vmware backdoor interface to work since it
6359 changes registers values during IO operation */
6360 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6361 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6362 emulator_invalidate_register_cache(ctxt);
6366 /* Save the faulting GPA (cr2) in the address field */
6367 ctxt->exception.address = cr2;
6369 r = x86_emulate_insn(ctxt);
6371 if (r == EMULATION_INTERCEPTED)
6372 return EMULATE_DONE;
6374 if (r == EMULATION_FAILED) {
6375 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6377 return EMULATE_DONE;
6379 return handle_emulation_failure(vcpu, emulation_type);
6382 if (ctxt->have_exception) {
6384 if (inject_emulated_exception(vcpu))
6386 } else if (vcpu->arch.pio.count) {
6387 if (!vcpu->arch.pio.in) {
6388 /* FIXME: return into emulator if single-stepping. */
6389 vcpu->arch.pio.count = 0;
6392 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6394 r = EMULATE_USER_EXIT;
6395 } else if (vcpu->mmio_needed) {
6396 if (!vcpu->mmio_is_write)
6398 r = EMULATE_USER_EXIT;
6399 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6400 } else if (r == EMULATION_RESTART)
6406 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6407 toggle_interruptibility(vcpu, ctxt->interruptibility);
6408 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6409 kvm_rip_write(vcpu, ctxt->eip);
6410 if (r == EMULATE_DONE &&
6411 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
6412 kvm_vcpu_do_singlestep(vcpu, &r);
6413 if (!ctxt->have_exception ||
6414 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6415 __kvm_set_rflags(vcpu, ctxt->eflags);
6418 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6419 * do nothing, and it will be requested again as soon as
6420 * the shadow expires. But we still need to check here,
6421 * because POPF has no interrupt shadow.
6423 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6424 kvm_make_request(KVM_REQ_EVENT, vcpu);
6426 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6431 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6433 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6435 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6437 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6438 void *insn, int insn_len)
6440 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6442 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6444 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6445 unsigned short port)
6447 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6448 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6449 size, port, &val, 1);
6450 /* do not return to emulator after return from userspace */
6451 vcpu->arch.pio.count = 0;
6455 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6459 /* We should only ever be called with arch.pio.count equal to 1 */
6460 BUG_ON(vcpu->arch.pio.count != 1);
6462 /* For size less than 4 we merge, else we zero extend */
6463 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6467 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6468 * the copy and tracing
6470 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6471 vcpu->arch.pio.port, &val, 1);
6472 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6477 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6478 unsigned short port)
6483 /* For size less than 4 we merge, else we zero extend */
6484 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6486 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6489 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6493 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6498 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6500 int ret = kvm_skip_emulated_instruction(vcpu);
6503 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6504 * KVM_EXIT_DEBUG here.
6507 return kvm_fast_pio_in(vcpu, size, port) && ret;
6509 return kvm_fast_pio_out(vcpu, size, port) && ret;
6511 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6513 static int kvmclock_cpu_down_prep(unsigned int cpu)
6515 __this_cpu_write(cpu_tsc_khz, 0);
6519 static void tsc_khz_changed(void *data)
6521 struct cpufreq_freqs *freq = data;
6522 unsigned long khz = 0;
6526 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6527 khz = cpufreq_quick_get(raw_smp_processor_id());
6530 __this_cpu_write(cpu_tsc_khz, khz);
6533 #ifdef CONFIG_X86_64
6534 static void kvm_hyperv_tsc_notifier(void)
6537 struct kvm_vcpu *vcpu;
6540 spin_lock(&kvm_lock);
6541 list_for_each_entry(kvm, &vm_list, vm_list)
6542 kvm_make_mclock_inprogress_request(kvm);
6544 hyperv_stop_tsc_emulation();
6546 /* TSC frequency always matches when on Hyper-V */
6547 for_each_present_cpu(cpu)
6548 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6549 kvm_max_guest_tsc_khz = tsc_khz;
6551 list_for_each_entry(kvm, &vm_list, vm_list) {
6552 struct kvm_arch *ka = &kvm->arch;
6554 spin_lock(&ka->pvclock_gtod_sync_lock);
6556 pvclock_update_vm_gtod_copy(kvm);
6558 kvm_for_each_vcpu(cpu, vcpu, kvm)
6559 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6561 kvm_for_each_vcpu(cpu, vcpu, kvm)
6562 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6564 spin_unlock(&ka->pvclock_gtod_sync_lock);
6566 spin_unlock(&kvm_lock);
6570 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6573 struct cpufreq_freqs *freq = data;
6575 struct kvm_vcpu *vcpu;
6576 int i, send_ipi = 0;
6579 * We allow guests to temporarily run on slowing clocks,
6580 * provided we notify them after, or to run on accelerating
6581 * clocks, provided we notify them before. Thus time never
6584 * However, we have a problem. We can't atomically update
6585 * the frequency of a given CPU from this function; it is
6586 * merely a notifier, which can be called from any CPU.
6587 * Changing the TSC frequency at arbitrary points in time
6588 * requires a recomputation of local variables related to
6589 * the TSC for each VCPU. We must flag these local variables
6590 * to be updated and be sure the update takes place with the
6591 * new frequency before any guests proceed.
6593 * Unfortunately, the combination of hotplug CPU and frequency
6594 * change creates an intractable locking scenario; the order
6595 * of when these callouts happen is undefined with respect to
6596 * CPU hotplug, and they can race with each other. As such,
6597 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6598 * undefined; you can actually have a CPU frequency change take
6599 * place in between the computation of X and the setting of the
6600 * variable. To protect against this problem, all updates of
6601 * the per_cpu tsc_khz variable are done in an interrupt
6602 * protected IPI, and all callers wishing to update the value
6603 * must wait for a synchronous IPI to complete (which is trivial
6604 * if the caller is on the CPU already). This establishes the
6605 * necessary total order on variable updates.
6607 * Note that because a guest time update may take place
6608 * anytime after the setting of the VCPU's request bit, the
6609 * correct TSC value must be set before the request. However,
6610 * to ensure the update actually makes it to any guest which
6611 * starts running in hardware virtualization between the set
6612 * and the acquisition of the spinlock, we must also ping the
6613 * CPU after setting the request bit.
6617 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6619 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6622 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6624 spin_lock(&kvm_lock);
6625 list_for_each_entry(kvm, &vm_list, vm_list) {
6626 kvm_for_each_vcpu(i, vcpu, kvm) {
6627 if (vcpu->cpu != freq->cpu)
6629 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6630 if (vcpu->cpu != smp_processor_id())
6634 spin_unlock(&kvm_lock);
6636 if (freq->old < freq->new && send_ipi) {
6638 * We upscale the frequency. Must make the guest
6639 * doesn't see old kvmclock values while running with
6640 * the new frequency, otherwise we risk the guest sees
6641 * time go backwards.
6643 * In case we update the frequency for another cpu
6644 * (which might be in guest context) send an interrupt
6645 * to kick the cpu out of guest context. Next time
6646 * guest context is entered kvmclock will be updated,
6647 * so the guest will not see stale values.
6649 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6654 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6655 .notifier_call = kvmclock_cpufreq_notifier
6658 static int kvmclock_cpu_online(unsigned int cpu)
6660 tsc_khz_changed(NULL);
6664 static void kvm_timer_init(void)
6666 max_tsc_khz = tsc_khz;
6668 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6669 #ifdef CONFIG_CPU_FREQ
6670 struct cpufreq_policy policy;
6673 memset(&policy, 0, sizeof(policy));
6675 cpufreq_get_policy(&policy, cpu);
6676 if (policy.cpuinfo.max_freq)
6677 max_tsc_khz = policy.cpuinfo.max_freq;
6680 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6681 CPUFREQ_TRANSITION_NOTIFIER);
6683 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6685 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6686 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6689 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6690 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6692 int kvm_is_in_guest(void)
6694 return __this_cpu_read(current_vcpu) != NULL;
6697 static int kvm_is_user_mode(void)
6701 if (__this_cpu_read(current_vcpu))
6702 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6704 return user_mode != 0;
6707 static unsigned long kvm_get_guest_ip(void)
6709 unsigned long ip = 0;
6711 if (__this_cpu_read(current_vcpu))
6712 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6717 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6718 .is_in_guest = kvm_is_in_guest,
6719 .is_user_mode = kvm_is_user_mode,
6720 .get_guest_ip = kvm_get_guest_ip,
6723 static void kvm_set_mmio_spte_mask(void)
6726 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6729 * Set the reserved bits and the present bit of an paging-structure
6730 * entry to generate page fault with PFER.RSV = 1.
6734 * Mask the uppermost physical address bit, which would be reserved as
6735 * long as the supported physical address width is less than 52.
6739 /* Set the present bit. */
6743 * If reserved bit is not supported, clear the present bit to disable
6746 if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6749 kvm_mmu_set_mmio_spte_mask(mask, mask);
6752 #ifdef CONFIG_X86_64
6753 static void pvclock_gtod_update_fn(struct work_struct *work)
6757 struct kvm_vcpu *vcpu;
6760 spin_lock(&kvm_lock);
6761 list_for_each_entry(kvm, &vm_list, vm_list)
6762 kvm_for_each_vcpu(i, vcpu, kvm)
6763 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6764 atomic_set(&kvm_guest_has_master_clock, 0);
6765 spin_unlock(&kvm_lock);
6768 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6771 * Notification about pvclock gtod data update.
6773 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6776 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6777 struct timekeeper *tk = priv;
6779 update_pvclock_gtod(tk);
6781 /* disable master clock if host does not trust, or does not
6782 * use, TSC based clocksource.
6784 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6785 atomic_read(&kvm_guest_has_master_clock) != 0)
6786 queue_work(system_long_wq, &pvclock_gtod_work);
6791 static struct notifier_block pvclock_gtod_notifier = {
6792 .notifier_call = pvclock_gtod_notify,
6796 int kvm_arch_init(void *opaque)
6799 struct kvm_x86_ops *ops = opaque;
6802 printk(KERN_ERR "kvm: already loaded the other module\n");
6807 if (!ops->cpu_has_kvm_support()) {
6808 printk(KERN_ERR "kvm: no hardware support\n");
6812 if (ops->disabled_by_bios()) {
6813 printk(KERN_ERR "kvm: disabled by bios\n");
6819 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6821 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6825 r = kvm_mmu_module_init();
6827 goto out_free_percpu;
6829 kvm_set_mmio_spte_mask();
6833 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6834 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6835 PT_PRESENT_MASK, 0, sme_me_mask);
6838 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6840 if (boot_cpu_has(X86_FEATURE_XSAVE))
6841 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6844 #ifdef CONFIG_X86_64
6845 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6847 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6848 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6854 free_percpu(shared_msrs);
6859 void kvm_arch_exit(void)
6861 #ifdef CONFIG_X86_64
6862 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6863 clear_hv_tscchange_cb();
6866 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6868 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6869 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6870 CPUFREQ_TRANSITION_NOTIFIER);
6871 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6872 #ifdef CONFIG_X86_64
6873 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6876 kvm_mmu_module_exit();
6877 free_percpu(shared_msrs);
6880 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6882 ++vcpu->stat.halt_exits;
6883 if (lapic_in_kernel(vcpu)) {
6884 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6887 vcpu->run->exit_reason = KVM_EXIT_HLT;
6891 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6893 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6895 int ret = kvm_skip_emulated_instruction(vcpu);
6897 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6898 * KVM_EXIT_DEBUG here.
6900 return kvm_vcpu_halt(vcpu) && ret;
6902 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6904 #ifdef CONFIG_X86_64
6905 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6906 unsigned long clock_type)
6908 struct kvm_clock_pairing clock_pairing;
6909 struct timespec64 ts;
6913 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6914 return -KVM_EOPNOTSUPP;
6916 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6917 return -KVM_EOPNOTSUPP;
6919 clock_pairing.sec = ts.tv_sec;
6920 clock_pairing.nsec = ts.tv_nsec;
6921 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6922 clock_pairing.flags = 0;
6923 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
6926 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6927 sizeof(struct kvm_clock_pairing)))
6935 * kvm_pv_kick_cpu_op: Kick a vcpu.
6937 * @apicid - apicid of vcpu to be kicked.
6939 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6941 struct kvm_lapic_irq lapic_irq;
6943 lapic_irq.shorthand = 0;
6944 lapic_irq.dest_mode = 0;
6945 lapic_irq.level = 0;
6946 lapic_irq.dest_id = apicid;
6947 lapic_irq.msi_redir_hint = false;
6949 lapic_irq.delivery_mode = APIC_DM_REMRD;
6950 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6953 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6955 vcpu->arch.apicv_active = false;
6956 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6959 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6961 unsigned long nr, a0, a1, a2, a3, ret;
6964 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6965 return kvm_hv_hypercall(vcpu);
6967 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6968 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6969 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6970 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6971 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6973 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6975 op_64_bit = is_64_bit_mode(vcpu);
6984 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6990 case KVM_HC_VAPIC_POLL_IRQ:
6993 case KVM_HC_KICK_CPU:
6994 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6997 #ifdef CONFIG_X86_64
6998 case KVM_HC_CLOCK_PAIRING:
6999 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7001 case KVM_HC_SEND_IPI:
7002 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7012 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
7014 ++vcpu->stat.hypercalls;
7015 return kvm_skip_emulated_instruction(vcpu);
7017 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7019 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7021 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7022 char instruction[3];
7023 unsigned long rip = kvm_rip_read(vcpu);
7025 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7027 return emulator_write_emulated(ctxt, rip, instruction, 3,
7031 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7033 return vcpu->run->request_interrupt_window &&
7034 likely(!pic_in_kernel(vcpu->kvm));
7037 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7039 struct kvm_run *kvm_run = vcpu->run;
7041 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7042 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7043 kvm_run->cr8 = kvm_get_cr8(vcpu);
7044 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7045 kvm_run->ready_for_interrupt_injection =
7046 pic_in_kernel(vcpu->kvm) ||
7047 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7050 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7054 if (!kvm_x86_ops->update_cr8_intercept)
7057 if (!lapic_in_kernel(vcpu))
7060 if (vcpu->arch.apicv_active)
7063 if (!vcpu->arch.apic->vapic_addr)
7064 max_irr = kvm_lapic_find_highest_irr(vcpu);
7071 tpr = kvm_lapic_get_cr8(vcpu);
7073 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7076 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7080 /* try to reinject previous events if any */
7082 if (vcpu->arch.exception.injected)
7083 kvm_x86_ops->queue_exception(vcpu);
7085 * Do not inject an NMI or interrupt if there is a pending
7086 * exception. Exceptions and interrupts are recognized at
7087 * instruction boundaries, i.e. the start of an instruction.
7088 * Trap-like exceptions, e.g. #DB, have higher priority than
7089 * NMIs and interrupts, i.e. traps are recognized before an
7090 * NMI/interrupt that's pending on the same instruction.
7091 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7092 * priority, but are only generated (pended) during instruction
7093 * execution, i.e. a pending fault-like exception means the
7094 * fault occurred on the *previous* instruction and must be
7095 * serviced prior to recognizing any new events in order to
7096 * fully complete the previous instruction.
7098 else if (!vcpu->arch.exception.pending) {
7099 if (vcpu->arch.nmi_injected)
7100 kvm_x86_ops->set_nmi(vcpu);
7101 else if (vcpu->arch.interrupt.injected)
7102 kvm_x86_ops->set_irq(vcpu);
7106 * Call check_nested_events() even if we reinjected a previous event
7107 * in order for caller to determine if it should require immediate-exit
7108 * from L2 to L1 due to pending L1 events which require exit
7111 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7112 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7117 /* try to inject new event if pending */
7118 if (vcpu->arch.exception.pending) {
7119 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7120 vcpu->arch.exception.has_error_code,
7121 vcpu->arch.exception.error_code);
7123 WARN_ON_ONCE(vcpu->arch.exception.injected);
7124 vcpu->arch.exception.pending = false;
7125 vcpu->arch.exception.injected = true;
7127 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7128 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7131 if (vcpu->arch.exception.nr == DB_VECTOR) {
7133 * This code assumes that nSVM doesn't use
7134 * check_nested_events(). If it does, the
7135 * DR6/DR7 changes should happen before L1
7136 * gets a #VMEXIT for an intercepted #DB in
7137 * L2. (Under VMX, on the other hand, the
7138 * DR6/DR7 changes should not happen in the
7139 * event of a VM-exit to L1 for an intercepted
7142 kvm_deliver_exception_payload(vcpu);
7143 if (vcpu->arch.dr7 & DR7_GD) {
7144 vcpu->arch.dr7 &= ~DR7_GD;
7145 kvm_update_dr7(vcpu);
7149 kvm_x86_ops->queue_exception(vcpu);
7152 /* Don't consider new event if we re-injected an event */
7153 if (kvm_event_needs_reinjection(vcpu))
7156 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7157 kvm_x86_ops->smi_allowed(vcpu)) {
7158 vcpu->arch.smi_pending = false;
7159 ++vcpu->arch.smi_count;
7161 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7162 --vcpu->arch.nmi_pending;
7163 vcpu->arch.nmi_injected = true;
7164 kvm_x86_ops->set_nmi(vcpu);
7165 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7167 * Because interrupts can be injected asynchronously, we are
7168 * calling check_nested_events again here to avoid a race condition.
7169 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7170 * proposal and current concerns. Perhaps we should be setting
7171 * KVM_REQ_EVENT only on certain events and not unconditionally?
7173 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7174 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7178 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7179 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7181 kvm_x86_ops->set_irq(vcpu);
7188 static void process_nmi(struct kvm_vcpu *vcpu)
7193 * x86 is limited to one NMI running, and one NMI pending after it.
7194 * If an NMI is already in progress, limit further NMIs to just one.
7195 * Otherwise, allow two (and we'll inject the first one immediately).
7197 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7200 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7201 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7202 kvm_make_request(KVM_REQ_EVENT, vcpu);
7205 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7208 flags |= seg->g << 23;
7209 flags |= seg->db << 22;
7210 flags |= seg->l << 21;
7211 flags |= seg->avl << 20;
7212 flags |= seg->present << 15;
7213 flags |= seg->dpl << 13;
7214 flags |= seg->s << 12;
7215 flags |= seg->type << 8;
7219 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7221 struct kvm_segment seg;
7224 kvm_get_segment(vcpu, &seg, n);
7225 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7228 offset = 0x7f84 + n * 12;
7230 offset = 0x7f2c + (n - 3) * 12;
7232 put_smstate(u32, buf, offset + 8, seg.base);
7233 put_smstate(u32, buf, offset + 4, seg.limit);
7234 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7237 #ifdef CONFIG_X86_64
7238 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7240 struct kvm_segment seg;
7244 kvm_get_segment(vcpu, &seg, n);
7245 offset = 0x7e00 + n * 16;
7247 flags = enter_smm_get_segment_flags(&seg) >> 8;
7248 put_smstate(u16, buf, offset, seg.selector);
7249 put_smstate(u16, buf, offset + 2, flags);
7250 put_smstate(u32, buf, offset + 4, seg.limit);
7251 put_smstate(u64, buf, offset + 8, seg.base);
7255 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7258 struct kvm_segment seg;
7262 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7263 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7264 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7265 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7267 for (i = 0; i < 8; i++)
7268 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7270 kvm_get_dr(vcpu, 6, &val);
7271 put_smstate(u32, buf, 0x7fcc, (u32)val);
7272 kvm_get_dr(vcpu, 7, &val);
7273 put_smstate(u32, buf, 0x7fc8, (u32)val);
7275 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7276 put_smstate(u32, buf, 0x7fc4, seg.selector);
7277 put_smstate(u32, buf, 0x7f64, seg.base);
7278 put_smstate(u32, buf, 0x7f60, seg.limit);
7279 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7281 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7282 put_smstate(u32, buf, 0x7fc0, seg.selector);
7283 put_smstate(u32, buf, 0x7f80, seg.base);
7284 put_smstate(u32, buf, 0x7f7c, seg.limit);
7285 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7287 kvm_x86_ops->get_gdt(vcpu, &dt);
7288 put_smstate(u32, buf, 0x7f74, dt.address);
7289 put_smstate(u32, buf, 0x7f70, dt.size);
7291 kvm_x86_ops->get_idt(vcpu, &dt);
7292 put_smstate(u32, buf, 0x7f58, dt.address);
7293 put_smstate(u32, buf, 0x7f54, dt.size);
7295 for (i = 0; i < 6; i++)
7296 enter_smm_save_seg_32(vcpu, buf, i);
7298 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7301 put_smstate(u32, buf, 0x7efc, 0x00020000);
7302 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7305 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7307 #ifdef CONFIG_X86_64
7309 struct kvm_segment seg;
7313 for (i = 0; i < 16; i++)
7314 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7316 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7317 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7319 kvm_get_dr(vcpu, 6, &val);
7320 put_smstate(u64, buf, 0x7f68, val);
7321 kvm_get_dr(vcpu, 7, &val);
7322 put_smstate(u64, buf, 0x7f60, val);
7324 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7325 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7326 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7328 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7331 put_smstate(u32, buf, 0x7efc, 0x00020064);
7333 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7335 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7336 put_smstate(u16, buf, 0x7e90, seg.selector);
7337 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7338 put_smstate(u32, buf, 0x7e94, seg.limit);
7339 put_smstate(u64, buf, 0x7e98, seg.base);
7341 kvm_x86_ops->get_idt(vcpu, &dt);
7342 put_smstate(u32, buf, 0x7e84, dt.size);
7343 put_smstate(u64, buf, 0x7e88, dt.address);
7345 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7346 put_smstate(u16, buf, 0x7e70, seg.selector);
7347 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7348 put_smstate(u32, buf, 0x7e74, seg.limit);
7349 put_smstate(u64, buf, 0x7e78, seg.base);
7351 kvm_x86_ops->get_gdt(vcpu, &dt);
7352 put_smstate(u32, buf, 0x7e64, dt.size);
7353 put_smstate(u64, buf, 0x7e68, dt.address);
7355 for (i = 0; i < 6; i++)
7356 enter_smm_save_seg_64(vcpu, buf, i);
7362 static void enter_smm(struct kvm_vcpu *vcpu)
7364 struct kvm_segment cs, ds;
7369 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7370 memset(buf, 0, 512);
7371 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7372 enter_smm_save_state_64(vcpu, buf);
7374 enter_smm_save_state_32(vcpu, buf);
7377 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7378 * vCPU state (e.g. leave guest mode) after we've saved the state into
7379 * the SMM state-save area.
7381 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7383 vcpu->arch.hflags |= HF_SMM_MASK;
7384 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7386 if (kvm_x86_ops->get_nmi_mask(vcpu))
7387 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7389 kvm_x86_ops->set_nmi_mask(vcpu, true);
7391 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7392 kvm_rip_write(vcpu, 0x8000);
7394 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7395 kvm_x86_ops->set_cr0(vcpu, cr0);
7396 vcpu->arch.cr0 = cr0;
7398 kvm_x86_ops->set_cr4(vcpu, 0);
7400 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7401 dt.address = dt.size = 0;
7402 kvm_x86_ops->set_idt(vcpu, &dt);
7404 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7406 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7407 cs.base = vcpu->arch.smbase;
7412 cs.limit = ds.limit = 0xffffffff;
7413 cs.type = ds.type = 0x3;
7414 cs.dpl = ds.dpl = 0;
7419 cs.avl = ds.avl = 0;
7420 cs.present = ds.present = 1;
7421 cs.unusable = ds.unusable = 0;
7422 cs.padding = ds.padding = 0;
7424 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7425 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7426 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7427 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7428 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7429 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7431 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7432 kvm_x86_ops->set_efer(vcpu, 0);
7434 kvm_update_cpuid(vcpu);
7435 kvm_mmu_reset_context(vcpu);
7438 static void process_smi(struct kvm_vcpu *vcpu)
7440 vcpu->arch.smi_pending = true;
7441 kvm_make_request(KVM_REQ_EVENT, vcpu);
7444 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7446 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7449 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7451 if (!kvm_apic_present(vcpu))
7454 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7456 if (irqchip_split(vcpu->kvm))
7457 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7459 if (vcpu->arch.apicv_active)
7460 kvm_x86_ops->sync_pir_to_irr(vcpu);
7461 if (ioapic_in_kernel(vcpu->kvm))
7462 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7465 if (is_guest_mode(vcpu))
7466 vcpu->arch.load_eoi_exitmap_pending = true;
7468 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7471 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7473 u64 eoi_exit_bitmap[4];
7475 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7478 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7479 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7480 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7483 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7484 unsigned long start, unsigned long end,
7487 unsigned long apic_address;
7490 * The physical address of apic access page is stored in the VMCS.
7491 * Update it when it becomes invalid.
7493 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7494 if (start <= apic_address && apic_address < end)
7495 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7500 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7502 struct page *page = NULL;
7504 if (!lapic_in_kernel(vcpu))
7507 if (!kvm_x86_ops->set_apic_access_page_addr)
7510 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7511 if (is_error_page(page))
7513 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7516 * Do not pin apic access page in memory, the MMU notifier
7517 * will call us again if it is migrated or swapped out.
7521 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7523 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7525 smp_send_reschedule(vcpu->cpu);
7527 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7530 * Returns 1 to let vcpu_run() continue the guest execution loop without
7531 * exiting to the userspace. Otherwise, the value will be returned to the
7534 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7538 dm_request_for_irq_injection(vcpu) &&
7539 kvm_cpu_accept_dm_intr(vcpu);
7541 bool req_immediate_exit = false;
7543 if (kvm_request_pending(vcpu)) {
7544 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7545 kvm_x86_ops->get_vmcs12_pages(vcpu);
7546 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7547 kvm_mmu_unload(vcpu);
7548 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7549 __kvm_migrate_timers(vcpu);
7550 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7551 kvm_gen_update_masterclock(vcpu->kvm);
7552 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7553 kvm_gen_kvmclock_update(vcpu);
7554 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7555 r = kvm_guest_time_update(vcpu);
7559 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7560 kvm_mmu_sync_roots(vcpu);
7561 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7562 kvm_mmu_load_cr3(vcpu);
7563 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7564 kvm_vcpu_flush_tlb(vcpu, true);
7565 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7566 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7570 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7571 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7572 vcpu->mmio_needed = 0;
7576 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7577 /* Page is swapped out. Do synthetic halt */
7578 vcpu->arch.apf.halted = true;
7582 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7583 record_steal_time(vcpu);
7584 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7586 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7588 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7589 kvm_pmu_handle_event(vcpu);
7590 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7591 kvm_pmu_deliver_pmi(vcpu);
7592 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7593 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7594 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7595 vcpu->arch.ioapic_handled_vectors)) {
7596 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7597 vcpu->run->eoi.vector =
7598 vcpu->arch.pending_ioapic_eoi;
7603 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7604 vcpu_scan_ioapic(vcpu);
7605 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7606 vcpu_load_eoi_exitmap(vcpu);
7607 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7608 kvm_vcpu_reload_apic_access_page(vcpu);
7609 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7610 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7611 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7615 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7616 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7617 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7621 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7622 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7623 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7629 * KVM_REQ_HV_STIMER has to be processed after
7630 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7631 * depend on the guest clock being up-to-date
7633 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7634 kvm_hv_process_stimers(vcpu);
7637 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7638 ++vcpu->stat.req_event;
7639 kvm_apic_accept_events(vcpu);
7640 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7645 if (inject_pending_event(vcpu, req_int_win) != 0)
7646 req_immediate_exit = true;
7648 /* Enable SMI/NMI/IRQ window open exits if needed.
7650 * SMIs have three cases:
7651 * 1) They can be nested, and then there is nothing to
7652 * do here because RSM will cause a vmexit anyway.
7653 * 2) There is an ISA-specific reason why SMI cannot be
7654 * injected, and the moment when this changes can be
7656 * 3) Or the SMI can be pending because
7657 * inject_pending_event has completed the injection
7658 * of an IRQ or NMI from the previous vmexit, and
7659 * then we request an immediate exit to inject the
7662 if (vcpu->arch.smi_pending && !is_smm(vcpu))
7663 if (!kvm_x86_ops->enable_smi_window(vcpu))
7664 req_immediate_exit = true;
7665 if (vcpu->arch.nmi_pending)
7666 kvm_x86_ops->enable_nmi_window(vcpu);
7667 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7668 kvm_x86_ops->enable_irq_window(vcpu);
7669 WARN_ON(vcpu->arch.exception.pending);
7672 if (kvm_lapic_enabled(vcpu)) {
7673 update_cr8_intercept(vcpu);
7674 kvm_lapic_sync_to_vapic(vcpu);
7678 r = kvm_mmu_reload(vcpu);
7680 goto cancel_injection;
7685 kvm_x86_ops->prepare_guest_switch(vcpu);
7688 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7689 * IPI are then delayed after guest entry, which ensures that they
7690 * result in virtual interrupt delivery.
7692 local_irq_disable();
7693 vcpu->mode = IN_GUEST_MODE;
7695 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7698 * 1) We should set ->mode before checking ->requests. Please see
7699 * the comment in kvm_vcpu_exiting_guest_mode().
7701 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7702 * pairs with the memory barrier implicit in pi_test_and_set_on
7703 * (see vmx_deliver_posted_interrupt).
7705 * 3) This also orders the write to mode from any reads to the page
7706 * tables done while the VCPU is running. Please see the comment
7707 * in kvm_flush_remote_tlbs.
7709 smp_mb__after_srcu_read_unlock();
7712 * This handles the case where a posted interrupt was
7713 * notified with kvm_vcpu_kick.
7715 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7716 kvm_x86_ops->sync_pir_to_irr(vcpu);
7718 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7719 || need_resched() || signal_pending(current)) {
7720 vcpu->mode = OUTSIDE_GUEST_MODE;
7724 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7726 goto cancel_injection;
7729 kvm_load_guest_xcr0(vcpu);
7731 if (req_immediate_exit) {
7732 kvm_make_request(KVM_REQ_EVENT, vcpu);
7733 kvm_x86_ops->request_immediate_exit(vcpu);
7736 trace_kvm_entry(vcpu->vcpu_id);
7737 if (lapic_timer_advance_ns)
7738 wait_lapic_expire(vcpu);
7739 guest_enter_irqoff();
7741 if (unlikely(vcpu->arch.switch_db_regs)) {
7743 set_debugreg(vcpu->arch.eff_db[0], 0);
7744 set_debugreg(vcpu->arch.eff_db[1], 1);
7745 set_debugreg(vcpu->arch.eff_db[2], 2);
7746 set_debugreg(vcpu->arch.eff_db[3], 3);
7747 set_debugreg(vcpu->arch.dr6, 6);
7748 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7751 kvm_x86_ops->run(vcpu);
7754 * Do this here before restoring debug registers on the host. And
7755 * since we do this before handling the vmexit, a DR access vmexit
7756 * can (a) read the correct value of the debug registers, (b) set
7757 * KVM_DEBUGREG_WONT_EXIT again.
7759 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7760 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7761 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7762 kvm_update_dr0123(vcpu);
7763 kvm_update_dr6(vcpu);
7764 kvm_update_dr7(vcpu);
7765 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7769 * If the guest has used debug registers, at least dr7
7770 * will be disabled while returning to the host.
7771 * If we don't have active breakpoints in the host, we don't
7772 * care about the messed up debug address registers. But if
7773 * we have some of them active, restore the old state.
7775 if (hw_breakpoint_active())
7776 hw_breakpoint_restore();
7778 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7780 vcpu->mode = OUTSIDE_GUEST_MODE;
7783 kvm_put_guest_xcr0(vcpu);
7785 kvm_before_interrupt(vcpu);
7786 kvm_x86_ops->handle_external_intr(vcpu);
7787 kvm_after_interrupt(vcpu);
7791 guest_exit_irqoff();
7796 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7799 * Profile KVM exit RIPs:
7801 if (unlikely(prof_on == KVM_PROFILING)) {
7802 unsigned long rip = kvm_rip_read(vcpu);
7803 profile_hit(KVM_PROFILING, (void *)rip);
7806 if (unlikely(vcpu->arch.tsc_always_catchup))
7807 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7809 if (vcpu->arch.apic_attention)
7810 kvm_lapic_sync_from_vapic(vcpu);
7812 vcpu->arch.gpa_available = false;
7813 r = kvm_x86_ops->handle_exit(vcpu);
7817 kvm_x86_ops->cancel_injection(vcpu);
7818 if (unlikely(vcpu->arch.apic_attention))
7819 kvm_lapic_sync_from_vapic(vcpu);
7824 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7826 if (!kvm_arch_vcpu_runnable(vcpu) &&
7827 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7828 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7829 kvm_vcpu_block(vcpu);
7830 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7832 if (kvm_x86_ops->post_block)
7833 kvm_x86_ops->post_block(vcpu);
7835 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7839 kvm_apic_accept_events(vcpu);
7840 switch(vcpu->arch.mp_state) {
7841 case KVM_MP_STATE_HALTED:
7842 vcpu->arch.pv.pv_unhalted = false;
7843 vcpu->arch.mp_state =
7844 KVM_MP_STATE_RUNNABLE;
7845 case KVM_MP_STATE_RUNNABLE:
7846 vcpu->arch.apf.halted = false;
7848 case KVM_MP_STATE_INIT_RECEIVED:
7857 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7859 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7860 kvm_x86_ops->check_nested_events(vcpu, false);
7862 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7863 !vcpu->arch.apf.halted);
7866 static int vcpu_run(struct kvm_vcpu *vcpu)
7869 struct kvm *kvm = vcpu->kvm;
7871 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7872 vcpu->arch.l1tf_flush_l1d = true;
7875 if (kvm_vcpu_running(vcpu)) {
7876 r = vcpu_enter_guest(vcpu);
7878 r = vcpu_block(kvm, vcpu);
7884 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7885 if (kvm_cpu_has_pending_timer(vcpu))
7886 kvm_inject_pending_timer_irqs(vcpu);
7888 if (dm_request_for_irq_injection(vcpu) &&
7889 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7891 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7892 ++vcpu->stat.request_irq_exits;
7896 kvm_check_async_pf_completion(vcpu);
7898 if (signal_pending(current)) {
7900 vcpu->run->exit_reason = KVM_EXIT_INTR;
7901 ++vcpu->stat.signal_exits;
7904 if (need_resched()) {
7905 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7907 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7911 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7916 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7919 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7920 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7921 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7922 if (r != EMULATE_DONE)
7927 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7929 BUG_ON(!vcpu->arch.pio.count);
7931 return complete_emulated_io(vcpu);
7935 * Implements the following, as a state machine:
7939 * for each mmio piece in the fragment
7947 * for each mmio piece in the fragment
7952 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7954 struct kvm_run *run = vcpu->run;
7955 struct kvm_mmio_fragment *frag;
7958 BUG_ON(!vcpu->mmio_needed);
7960 /* Complete previous fragment */
7961 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7962 len = min(8u, frag->len);
7963 if (!vcpu->mmio_is_write)
7964 memcpy(frag->data, run->mmio.data, len);
7966 if (frag->len <= 8) {
7967 /* Switch to the next fragment. */
7969 vcpu->mmio_cur_fragment++;
7971 /* Go forward to the next mmio piece. */
7977 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7978 vcpu->mmio_needed = 0;
7980 /* FIXME: return into emulator if single-stepping. */
7981 if (vcpu->mmio_is_write)
7983 vcpu->mmio_read_completed = 1;
7984 return complete_emulated_io(vcpu);
7987 run->exit_reason = KVM_EXIT_MMIO;
7988 run->mmio.phys_addr = frag->gpa;
7989 if (vcpu->mmio_is_write)
7990 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7991 run->mmio.len = min(8u, frag->len);
7992 run->mmio.is_write = vcpu->mmio_is_write;
7993 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7997 /* Swap (qemu) user FPU context for the guest FPU context. */
7998 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8001 copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
8002 /* PKRU is separately restored in kvm_x86_ops->run. */
8003 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
8004 ~XFEATURE_MASK_PKRU);
8009 /* When vcpu_run ends, restore user space FPU context. */
8010 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8013 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
8014 copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
8016 ++vcpu->stat.fpu_reload;
8020 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8025 kvm_sigset_activate(vcpu);
8026 kvm_load_guest_fpu(vcpu);
8028 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8029 if (kvm_run->immediate_exit) {
8033 kvm_vcpu_block(vcpu);
8034 kvm_apic_accept_events(vcpu);
8035 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8037 if (signal_pending(current)) {
8039 vcpu->run->exit_reason = KVM_EXIT_INTR;
8040 ++vcpu->stat.signal_exits;
8045 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8050 if (vcpu->run->kvm_dirty_regs) {
8051 r = sync_regs(vcpu);
8056 /* re-sync apic's tpr */
8057 if (!lapic_in_kernel(vcpu)) {
8058 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8064 if (unlikely(vcpu->arch.complete_userspace_io)) {
8065 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8066 vcpu->arch.complete_userspace_io = NULL;
8071 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8073 if (kvm_run->immediate_exit)
8079 kvm_put_guest_fpu(vcpu);
8080 if (vcpu->run->kvm_valid_regs)
8082 post_kvm_run_save(vcpu);
8083 kvm_sigset_deactivate(vcpu);
8089 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8091 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8093 * We are here if userspace calls get_regs() in the middle of
8094 * instruction emulation. Registers state needs to be copied
8095 * back from emulation context to vcpu. Userspace shouldn't do
8096 * that usually, but some bad designed PV devices (vmware
8097 * backdoor interface) need this to work
8099 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8100 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8102 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
8103 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
8104 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
8105 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
8106 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
8107 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
8108 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8109 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
8110 #ifdef CONFIG_X86_64
8111 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
8112 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
8113 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
8114 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
8115 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
8116 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
8117 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
8118 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
8121 regs->rip = kvm_rip_read(vcpu);
8122 regs->rflags = kvm_get_rflags(vcpu);
8125 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8128 __get_regs(vcpu, regs);
8133 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8135 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8136 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8138 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
8139 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
8140 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
8141 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
8142 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
8143 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
8144 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
8145 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
8146 #ifdef CONFIG_X86_64
8147 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
8148 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
8149 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
8150 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
8151 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
8152 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
8153 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
8154 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
8157 kvm_rip_write(vcpu, regs->rip);
8158 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8160 vcpu->arch.exception.pending = false;
8162 kvm_make_request(KVM_REQ_EVENT, vcpu);
8165 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8168 __set_regs(vcpu, regs);
8173 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8175 struct kvm_segment cs;
8177 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8181 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8183 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8187 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8188 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8189 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8190 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8191 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8192 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8194 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8195 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8197 kvm_x86_ops->get_idt(vcpu, &dt);
8198 sregs->idt.limit = dt.size;
8199 sregs->idt.base = dt.address;
8200 kvm_x86_ops->get_gdt(vcpu, &dt);
8201 sregs->gdt.limit = dt.size;
8202 sregs->gdt.base = dt.address;
8204 sregs->cr0 = kvm_read_cr0(vcpu);
8205 sregs->cr2 = vcpu->arch.cr2;
8206 sregs->cr3 = kvm_read_cr3(vcpu);
8207 sregs->cr4 = kvm_read_cr4(vcpu);
8208 sregs->cr8 = kvm_get_cr8(vcpu);
8209 sregs->efer = vcpu->arch.efer;
8210 sregs->apic_base = kvm_get_apic_base(vcpu);
8212 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8214 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8215 set_bit(vcpu->arch.interrupt.nr,
8216 (unsigned long *)sregs->interrupt_bitmap);
8219 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8220 struct kvm_sregs *sregs)
8223 __get_sregs(vcpu, sregs);
8228 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8229 struct kvm_mp_state *mp_state)
8233 kvm_apic_accept_events(vcpu);
8234 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8235 vcpu->arch.pv.pv_unhalted)
8236 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8238 mp_state->mp_state = vcpu->arch.mp_state;
8244 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8245 struct kvm_mp_state *mp_state)
8251 if (!lapic_in_kernel(vcpu) &&
8252 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8255 /* INITs are latched while in SMM */
8256 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8257 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8258 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8261 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8262 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8263 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8265 vcpu->arch.mp_state = mp_state->mp_state;
8266 kvm_make_request(KVM_REQ_EVENT, vcpu);
8274 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8275 int reason, bool has_error_code, u32 error_code)
8277 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8280 init_emulate_ctxt(vcpu);
8282 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8283 has_error_code, error_code);
8286 return EMULATE_FAIL;
8288 kvm_rip_write(vcpu, ctxt->eip);
8289 kvm_set_rflags(vcpu, ctxt->eflags);
8290 kvm_make_request(KVM_REQ_EVENT, vcpu);
8291 return EMULATE_DONE;
8293 EXPORT_SYMBOL_GPL(kvm_task_switch);
8295 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8297 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8298 (sregs->cr4 & X86_CR4_OSXSAVE))
8301 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8303 * When EFER.LME and CR0.PG are set, the processor is in
8304 * 64-bit mode (though maybe in a 32-bit code segment).
8305 * CR4.PAE and EFER.LMA must be set.
8307 if (!(sregs->cr4 & X86_CR4_PAE)
8308 || !(sregs->efer & EFER_LMA))
8312 * Not in 64-bit mode: EFER.LMA is clear and the code
8313 * segment cannot be 64-bit.
8315 if (sregs->efer & EFER_LMA || sregs->cs.l)
8322 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8324 struct msr_data apic_base_msr;
8325 int mmu_reset_needed = 0;
8326 int cpuid_update_needed = 0;
8327 int pending_vec, max_bits, idx;
8331 if (kvm_valid_sregs(vcpu, sregs))
8334 apic_base_msr.data = sregs->apic_base;
8335 apic_base_msr.host_initiated = true;
8336 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8339 dt.size = sregs->idt.limit;
8340 dt.address = sregs->idt.base;
8341 kvm_x86_ops->set_idt(vcpu, &dt);
8342 dt.size = sregs->gdt.limit;
8343 dt.address = sregs->gdt.base;
8344 kvm_x86_ops->set_gdt(vcpu, &dt);
8346 vcpu->arch.cr2 = sregs->cr2;
8347 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8348 vcpu->arch.cr3 = sregs->cr3;
8349 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8351 kvm_set_cr8(vcpu, sregs->cr8);
8353 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8354 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8356 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8357 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8358 vcpu->arch.cr0 = sregs->cr0;
8360 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8361 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8362 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8363 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8364 if (cpuid_update_needed)
8365 kvm_update_cpuid(vcpu);
8367 idx = srcu_read_lock(&vcpu->kvm->srcu);
8368 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8369 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8370 mmu_reset_needed = 1;
8372 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8374 if (mmu_reset_needed)
8375 kvm_mmu_reset_context(vcpu);
8377 max_bits = KVM_NR_INTERRUPTS;
8378 pending_vec = find_first_bit(
8379 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8380 if (pending_vec < max_bits) {
8381 kvm_queue_interrupt(vcpu, pending_vec, false);
8382 pr_debug("Set back pending irq %d\n", pending_vec);
8385 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8386 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8387 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8388 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8389 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8390 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8392 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8393 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8395 update_cr8_intercept(vcpu);
8397 /* Older userspace won't unhalt the vcpu on reset. */
8398 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8399 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8401 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8403 kvm_make_request(KVM_REQ_EVENT, vcpu);
8410 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8411 struct kvm_sregs *sregs)
8416 ret = __set_sregs(vcpu, sregs);
8421 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8422 struct kvm_guest_debug *dbg)
8424 unsigned long rflags;
8429 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8431 if (vcpu->arch.exception.pending)
8433 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8434 kvm_queue_exception(vcpu, DB_VECTOR);
8436 kvm_queue_exception(vcpu, BP_VECTOR);
8440 * Read rflags as long as potentially injected trace flags are still
8443 rflags = kvm_get_rflags(vcpu);
8445 vcpu->guest_debug = dbg->control;
8446 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8447 vcpu->guest_debug = 0;
8449 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8450 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8451 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8452 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8454 for (i = 0; i < KVM_NR_DB_REGS; i++)
8455 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8457 kvm_update_dr7(vcpu);
8459 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8460 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8461 get_segment_base(vcpu, VCPU_SREG_CS);
8464 * Trigger an rflags update that will inject or remove the trace
8467 kvm_set_rflags(vcpu, rflags);
8469 kvm_x86_ops->update_bp_intercept(vcpu);
8479 * Translate a guest virtual address to a guest physical address.
8481 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8482 struct kvm_translation *tr)
8484 unsigned long vaddr = tr->linear_address;
8490 idx = srcu_read_lock(&vcpu->kvm->srcu);
8491 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8492 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8493 tr->physical_address = gpa;
8494 tr->valid = gpa != UNMAPPED_GVA;
8502 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8504 struct fxregs_state *fxsave;
8508 fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8509 memcpy(fpu->fpr, fxsave->st_space, 128);
8510 fpu->fcw = fxsave->cwd;
8511 fpu->fsw = fxsave->swd;
8512 fpu->ftwx = fxsave->twd;
8513 fpu->last_opcode = fxsave->fop;
8514 fpu->last_ip = fxsave->rip;
8515 fpu->last_dp = fxsave->rdp;
8516 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8522 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8524 struct fxregs_state *fxsave;
8528 fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8530 memcpy(fxsave->st_space, fpu->fpr, 128);
8531 fxsave->cwd = fpu->fcw;
8532 fxsave->swd = fpu->fsw;
8533 fxsave->twd = fpu->ftwx;
8534 fxsave->fop = fpu->last_opcode;
8535 fxsave->rip = fpu->last_ip;
8536 fxsave->rdp = fpu->last_dp;
8537 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8543 static void store_regs(struct kvm_vcpu *vcpu)
8545 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8547 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8548 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8550 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8551 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8553 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8554 kvm_vcpu_ioctl_x86_get_vcpu_events(
8555 vcpu, &vcpu->run->s.regs.events);
8558 static int sync_regs(struct kvm_vcpu *vcpu)
8560 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8563 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8564 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8565 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8567 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8568 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8570 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8572 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8573 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8574 vcpu, &vcpu->run->s.regs.events))
8576 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8582 static void fx_init(struct kvm_vcpu *vcpu)
8584 fpstate_init(&vcpu->arch.guest_fpu.state);
8585 if (boot_cpu_has(X86_FEATURE_XSAVES))
8586 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
8587 host_xcr0 | XSTATE_COMPACTION_ENABLED;
8590 * Ensure guest xcr0 is valid for loading
8592 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8594 vcpu->arch.cr0 |= X86_CR0_ET;
8597 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8599 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8601 kvmclock_reset(vcpu);
8603 kvm_x86_ops->vcpu_free(vcpu);
8604 free_cpumask_var(wbinvd_dirty_mask);
8607 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8610 struct kvm_vcpu *vcpu;
8612 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8613 printk_once(KERN_WARNING
8614 "kvm: SMP vm created on host with unstable TSC; "
8615 "guest TSC will not be reliable\n");
8617 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8622 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8624 kvm_vcpu_mtrr_init(vcpu);
8626 kvm_vcpu_reset(vcpu, false);
8627 kvm_init_mmu(vcpu, false);
8632 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8634 struct msr_data msr;
8635 struct kvm *kvm = vcpu->kvm;
8637 kvm_hv_vcpu_postcreate(vcpu);
8639 if (mutex_lock_killable(&vcpu->mutex))
8643 msr.index = MSR_IA32_TSC;
8644 msr.host_initiated = true;
8645 kvm_write_tsc(vcpu, &msr);
8647 mutex_unlock(&vcpu->mutex);
8649 if (!kvmclock_periodic_sync)
8652 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8653 KVMCLOCK_SYNC_PERIOD);
8656 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8658 vcpu->arch.apf.msr_val = 0;
8661 kvm_mmu_unload(vcpu);
8664 kvm_x86_ops->vcpu_free(vcpu);
8667 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8669 kvm_lapic_reset(vcpu, init_event);
8671 vcpu->arch.hflags = 0;
8673 vcpu->arch.smi_pending = 0;
8674 vcpu->arch.smi_count = 0;
8675 atomic_set(&vcpu->arch.nmi_queued, 0);
8676 vcpu->arch.nmi_pending = 0;
8677 vcpu->arch.nmi_injected = false;
8678 kvm_clear_interrupt_queue(vcpu);
8679 kvm_clear_exception_queue(vcpu);
8680 vcpu->arch.exception.pending = false;
8682 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8683 kvm_update_dr0123(vcpu);
8684 vcpu->arch.dr6 = DR6_INIT;
8685 kvm_update_dr6(vcpu);
8686 vcpu->arch.dr7 = DR7_FIXED_1;
8687 kvm_update_dr7(vcpu);
8691 kvm_make_request(KVM_REQ_EVENT, vcpu);
8692 vcpu->arch.apf.msr_val = 0;
8693 vcpu->arch.st.msr_val = 0;
8695 kvmclock_reset(vcpu);
8697 kvm_clear_async_pf_completion_queue(vcpu);
8698 kvm_async_pf_hash_reset(vcpu);
8699 vcpu->arch.apf.halted = false;
8701 if (kvm_mpx_supported()) {
8702 void *mpx_state_buffer;
8705 * To avoid have the INIT path from kvm_apic_has_events() that be
8706 * called with loaded FPU and does not let userspace fix the state.
8709 kvm_put_guest_fpu(vcpu);
8710 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8711 XFEATURE_MASK_BNDREGS);
8712 if (mpx_state_buffer)
8713 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8714 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8715 XFEATURE_MASK_BNDCSR);
8716 if (mpx_state_buffer)
8717 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8719 kvm_load_guest_fpu(vcpu);
8723 kvm_pmu_reset(vcpu);
8724 vcpu->arch.smbase = 0x30000;
8726 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8727 vcpu->arch.msr_misc_features_enables = 0;
8729 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8732 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8733 vcpu->arch.regs_avail = ~0;
8734 vcpu->arch.regs_dirty = ~0;
8736 vcpu->arch.ia32_xss = 0;
8738 kvm_x86_ops->vcpu_reset(vcpu, init_event);
8741 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8743 struct kvm_segment cs;
8745 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8746 cs.selector = vector << 8;
8747 cs.base = vector << 12;
8748 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8749 kvm_rip_write(vcpu, 0);
8752 int kvm_arch_hardware_enable(void)
8755 struct kvm_vcpu *vcpu;
8760 bool stable, backwards_tsc = false;
8762 kvm_shared_msr_cpu_online();
8763 ret = kvm_x86_ops->hardware_enable();
8767 local_tsc = rdtsc();
8768 stable = !kvm_check_tsc_unstable();
8769 list_for_each_entry(kvm, &vm_list, vm_list) {
8770 kvm_for_each_vcpu(i, vcpu, kvm) {
8771 if (!stable && vcpu->cpu == smp_processor_id())
8772 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8773 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8774 backwards_tsc = true;
8775 if (vcpu->arch.last_host_tsc > max_tsc)
8776 max_tsc = vcpu->arch.last_host_tsc;
8782 * Sometimes, even reliable TSCs go backwards. This happens on
8783 * platforms that reset TSC during suspend or hibernate actions, but
8784 * maintain synchronization. We must compensate. Fortunately, we can
8785 * detect that condition here, which happens early in CPU bringup,
8786 * before any KVM threads can be running. Unfortunately, we can't
8787 * bring the TSCs fully up to date with real time, as we aren't yet far
8788 * enough into CPU bringup that we know how much real time has actually
8789 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8790 * variables that haven't been updated yet.
8792 * So we simply find the maximum observed TSC above, then record the
8793 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8794 * the adjustment will be applied. Note that we accumulate
8795 * adjustments, in case multiple suspend cycles happen before some VCPU
8796 * gets a chance to run again. In the event that no KVM threads get a
8797 * chance to run, we will miss the entire elapsed period, as we'll have
8798 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8799 * loose cycle time. This isn't too big a deal, since the loss will be
8800 * uniform across all VCPUs (not to mention the scenario is extremely
8801 * unlikely). It is possible that a second hibernate recovery happens
8802 * much faster than a first, causing the observed TSC here to be
8803 * smaller; this would require additional padding adjustment, which is
8804 * why we set last_host_tsc to the local tsc observed here.
8806 * N.B. - this code below runs only on platforms with reliable TSC,
8807 * as that is the only way backwards_tsc is set above. Also note
8808 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8809 * have the same delta_cyc adjustment applied if backwards_tsc
8810 * is detected. Note further, this adjustment is only done once,
8811 * as we reset last_host_tsc on all VCPUs to stop this from being
8812 * called multiple times (one for each physical CPU bringup).
8814 * Platforms with unreliable TSCs don't have to deal with this, they
8815 * will be compensated by the logic in vcpu_load, which sets the TSC to
8816 * catchup mode. This will catchup all VCPUs to real time, but cannot
8817 * guarantee that they stay in perfect synchronization.
8819 if (backwards_tsc) {
8820 u64 delta_cyc = max_tsc - local_tsc;
8821 list_for_each_entry(kvm, &vm_list, vm_list) {
8822 kvm->arch.backwards_tsc_observed = true;
8823 kvm_for_each_vcpu(i, vcpu, kvm) {
8824 vcpu->arch.tsc_offset_adjustment += delta_cyc;
8825 vcpu->arch.last_host_tsc = local_tsc;
8826 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8830 * We have to disable TSC offset matching.. if you were
8831 * booting a VM while issuing an S4 host suspend....
8832 * you may have some problem. Solving this issue is
8833 * left as an exercise to the reader.
8835 kvm->arch.last_tsc_nsec = 0;
8836 kvm->arch.last_tsc_write = 0;
8843 void kvm_arch_hardware_disable(void)
8845 kvm_x86_ops->hardware_disable();
8846 drop_user_return_notifiers();
8849 int kvm_arch_hardware_setup(void)
8853 r = kvm_x86_ops->hardware_setup();
8857 if (kvm_has_tsc_control) {
8859 * Make sure the user can only configure tsc_khz values that
8860 * fit into a signed integer.
8861 * A min value is not calculated because it will always
8862 * be 1 on all machines.
8864 u64 max = min(0x7fffffffULL,
8865 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8866 kvm_max_guest_tsc_khz = max;
8868 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8871 kvm_init_msr_list();
8875 void kvm_arch_hardware_unsetup(void)
8877 kvm_x86_ops->hardware_unsetup();
8880 void kvm_arch_check_processor_compat(void *rtn)
8882 kvm_x86_ops->check_processor_compatibility(rtn);
8885 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8887 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8889 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8891 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8893 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8896 struct static_key kvm_no_apic_vcpu __read_mostly;
8897 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8899 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8904 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8905 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8906 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8907 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8909 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8911 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8916 vcpu->arch.pio_data = page_address(page);
8918 kvm_set_tsc_khz(vcpu, max_tsc_khz);
8920 r = kvm_mmu_create(vcpu);
8922 goto fail_free_pio_data;
8924 if (irqchip_in_kernel(vcpu->kvm)) {
8925 r = kvm_create_lapic(vcpu);
8927 goto fail_mmu_destroy;
8929 static_key_slow_inc(&kvm_no_apic_vcpu);
8931 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8933 if (!vcpu->arch.mce_banks) {
8935 goto fail_free_lapic;
8937 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8939 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8941 goto fail_free_mce_banks;
8946 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8948 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8950 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8952 kvm_async_pf_hash_reset(vcpu);
8955 vcpu->arch.pending_external_vector = -1;
8956 vcpu->arch.preempted_in_kernel = false;
8958 kvm_hv_vcpu_init(vcpu);
8962 fail_free_mce_banks:
8963 kfree(vcpu->arch.mce_banks);
8965 kvm_free_lapic(vcpu);
8967 kvm_mmu_destroy(vcpu);
8969 free_page((unsigned long)vcpu->arch.pio_data);
8974 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8978 kvm_hv_vcpu_uninit(vcpu);
8979 kvm_pmu_destroy(vcpu);
8980 kfree(vcpu->arch.mce_banks);
8981 kvm_free_lapic(vcpu);
8982 idx = srcu_read_lock(&vcpu->kvm->srcu);
8983 kvm_mmu_destroy(vcpu);
8984 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8985 free_page((unsigned long)vcpu->arch.pio_data);
8986 if (!lapic_in_kernel(vcpu))
8987 static_key_slow_dec(&kvm_no_apic_vcpu);
8990 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8992 vcpu->arch.l1tf_flush_l1d = true;
8993 kvm_x86_ops->sched_in(vcpu, cpu);
8996 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9001 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9002 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9003 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9004 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9005 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9007 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9008 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9009 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9010 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9011 &kvm->arch.irq_sources_bitmap);
9013 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9014 mutex_init(&kvm->arch.apic_map_lock);
9015 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9017 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9018 pvclock_update_vm_gtod_copy(kvm);
9020 kvm->arch.guest_can_read_msr_platform_info = true;
9022 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9023 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9025 kvm_hv_init_vm(kvm);
9026 kvm_page_track_init(kvm);
9027 kvm_mmu_init_vm(kvm);
9029 if (kvm_x86_ops->vm_init)
9030 return kvm_x86_ops->vm_init(kvm);
9035 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9038 kvm_mmu_unload(vcpu);
9042 static void kvm_free_vcpus(struct kvm *kvm)
9045 struct kvm_vcpu *vcpu;
9048 * Unpin any mmu pages first.
9050 kvm_for_each_vcpu(i, vcpu, kvm) {
9051 kvm_clear_async_pf_completion_queue(vcpu);
9052 kvm_unload_vcpu_mmu(vcpu);
9054 kvm_for_each_vcpu(i, vcpu, kvm)
9055 kvm_arch_vcpu_free(vcpu);
9057 mutex_lock(&kvm->lock);
9058 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9059 kvm->vcpus[i] = NULL;
9061 atomic_set(&kvm->online_vcpus, 0);
9062 mutex_unlock(&kvm->lock);
9065 void kvm_arch_sync_events(struct kvm *kvm)
9067 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9068 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9072 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9076 struct kvm_memslots *slots = kvm_memslots(kvm);
9077 struct kvm_memory_slot *slot, old;
9079 /* Called with kvm->slots_lock held. */
9080 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9083 slot = id_to_memslot(slots, id);
9089 * MAP_SHARED to prevent internal slot pages from being moved
9092 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9093 MAP_SHARED | MAP_ANONYMOUS, 0);
9094 if (IS_ERR((void *)hva))
9095 return PTR_ERR((void *)hva);
9104 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9105 struct kvm_userspace_memory_region m;
9107 m.slot = id | (i << 16);
9109 m.guest_phys_addr = gpa;
9110 m.userspace_addr = hva;
9111 m.memory_size = size;
9112 r = __kvm_set_memory_region(kvm, &m);
9118 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9122 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9124 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9128 mutex_lock(&kvm->slots_lock);
9129 r = __x86_set_memory_region(kvm, id, gpa, size);
9130 mutex_unlock(&kvm->slots_lock);
9134 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9136 void kvm_arch_destroy_vm(struct kvm *kvm)
9138 if (current->mm == kvm->mm) {
9140 * Free memory regions allocated on behalf of userspace,
9141 * unless the the memory map has changed due to process exit
9144 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9145 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9146 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9148 if (kvm_x86_ops->vm_destroy)
9149 kvm_x86_ops->vm_destroy(kvm);
9150 kvm_pic_destroy(kvm);
9151 kvm_ioapic_destroy(kvm);
9152 kvm_free_vcpus(kvm);
9153 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9154 kvm_mmu_uninit_vm(kvm);
9155 kvm_page_track_cleanup(kvm);
9156 kvm_hv_destroy_vm(kvm);
9159 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9160 struct kvm_memory_slot *dont)
9164 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9165 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9166 kvfree(free->arch.rmap[i]);
9167 free->arch.rmap[i] = NULL;
9172 if (!dont || free->arch.lpage_info[i - 1] !=
9173 dont->arch.lpage_info[i - 1]) {
9174 kvfree(free->arch.lpage_info[i - 1]);
9175 free->arch.lpage_info[i - 1] = NULL;
9179 kvm_page_track_free_memslot(free, dont);
9182 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9183 unsigned long npages)
9187 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9188 struct kvm_lpage_info *linfo;
9193 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9194 slot->base_gfn, level) + 1;
9196 slot->arch.rmap[i] =
9197 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9199 if (!slot->arch.rmap[i])
9204 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL);
9208 slot->arch.lpage_info[i - 1] = linfo;
9210 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9211 linfo[0].disallow_lpage = 1;
9212 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9213 linfo[lpages - 1].disallow_lpage = 1;
9214 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9216 * If the gfn and userspace address are not aligned wrt each
9217 * other, or if explicitly asked to, disable large page
9218 * support for this slot
9220 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9221 !kvm_largepages_enabled()) {
9224 for (j = 0; j < lpages; ++j)
9225 linfo[j].disallow_lpage = 1;
9229 if (kvm_page_track_create_memslot(slot, npages))
9235 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9236 kvfree(slot->arch.rmap[i]);
9237 slot->arch.rmap[i] = NULL;
9241 kvfree(slot->arch.lpage_info[i - 1]);
9242 slot->arch.lpage_info[i - 1] = NULL;
9247 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
9250 * memslots->generation has been incremented.
9251 * mmio generation may have reached its maximum value.
9253 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
9256 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9257 struct kvm_memory_slot *memslot,
9258 const struct kvm_userspace_memory_region *mem,
9259 enum kvm_mr_change change)
9264 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9265 struct kvm_memory_slot *new)
9267 /* Still write protect RO slot */
9268 if (new->flags & KVM_MEM_READONLY) {
9269 kvm_mmu_slot_remove_write_access(kvm, new);
9274 * Call kvm_x86_ops dirty logging hooks when they are valid.
9276 * kvm_x86_ops->slot_disable_log_dirty is called when:
9278 * - KVM_MR_CREATE with dirty logging is disabled
9279 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9281 * The reason is, in case of PML, we need to set D-bit for any slots
9282 * with dirty logging disabled in order to eliminate unnecessary GPA
9283 * logging in PML buffer (and potential PML buffer full VMEXT). This
9284 * guarantees leaving PML enabled during guest's lifetime won't have
9285 * any additonal overhead from PML when guest is running with dirty
9286 * logging disabled for memory slots.
9288 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9289 * to dirty logging mode.
9291 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9293 * In case of write protect:
9295 * Write protect all pages for dirty logging.
9297 * All the sptes including the large sptes which point to this
9298 * slot are set to readonly. We can not create any new large
9299 * spte on this slot until the end of the logging.
9301 * See the comments in fast_page_fault().
9303 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9304 if (kvm_x86_ops->slot_enable_log_dirty)
9305 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9307 kvm_mmu_slot_remove_write_access(kvm, new);
9309 if (kvm_x86_ops->slot_disable_log_dirty)
9310 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9314 void kvm_arch_commit_memory_region(struct kvm *kvm,
9315 const struct kvm_userspace_memory_region *mem,
9316 const struct kvm_memory_slot *old,
9317 const struct kvm_memory_slot *new,
9318 enum kvm_mr_change change)
9320 int nr_mmu_pages = 0;
9322 if (!kvm->arch.n_requested_mmu_pages)
9323 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9326 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9329 * Dirty logging tracks sptes in 4k granularity, meaning that large
9330 * sptes have to be split. If live migration is successful, the guest
9331 * in the source machine will be destroyed and large sptes will be
9332 * created in the destination. However, if the guest continues to run
9333 * in the source machine (for example if live migration fails), small
9334 * sptes will remain around and cause bad performance.
9336 * Scan sptes if dirty logging has been stopped, dropping those
9337 * which can be collapsed into a single large-page spte. Later
9338 * page faults will create the large-page sptes.
9340 if ((change != KVM_MR_DELETE) &&
9341 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9342 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9343 kvm_mmu_zap_collapsible_sptes(kvm, new);
9346 * Set up write protection and/or dirty logging for the new slot.
9348 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9349 * been zapped so no dirty logging staff is needed for old slot. For
9350 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9351 * new and it's also covered when dealing with the new slot.
9353 * FIXME: const-ify all uses of struct kvm_memory_slot.
9355 if (change != KVM_MR_DELETE)
9356 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9359 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9361 kvm_mmu_invalidate_zap_all_pages(kvm);
9364 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9365 struct kvm_memory_slot *slot)
9367 kvm_page_track_flush_slot(kvm, slot);
9370 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9372 return (is_guest_mode(vcpu) &&
9373 kvm_x86_ops->guest_apic_has_interrupt &&
9374 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9377 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9379 if (!list_empty_careful(&vcpu->async_pf.done))
9382 if (kvm_apic_has_events(vcpu))
9385 if (vcpu->arch.pv.pv_unhalted)
9388 if (vcpu->arch.exception.pending)
9391 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9392 (vcpu->arch.nmi_pending &&
9393 kvm_x86_ops->nmi_allowed(vcpu)))
9396 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9397 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9400 if (kvm_arch_interrupt_allowed(vcpu) &&
9401 (kvm_cpu_has_interrupt(vcpu) ||
9402 kvm_guest_apic_has_interrupt(vcpu)))
9405 if (kvm_hv_has_stimer_pending(vcpu))
9411 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9413 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9416 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9418 return vcpu->arch.preempted_in_kernel;
9421 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9423 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9426 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9428 return kvm_x86_ops->interrupt_allowed(vcpu);
9431 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9433 if (is_64_bit_mode(vcpu))
9434 return kvm_rip_read(vcpu);
9435 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9436 kvm_rip_read(vcpu));
9438 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9440 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9442 return kvm_get_linear_rip(vcpu) == linear_rip;
9444 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9446 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9448 unsigned long rflags;
9450 rflags = kvm_x86_ops->get_rflags(vcpu);
9451 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9452 rflags &= ~X86_EFLAGS_TF;
9455 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9457 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9459 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9460 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9461 rflags |= X86_EFLAGS_TF;
9462 kvm_x86_ops->set_rflags(vcpu, rflags);
9465 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9467 __kvm_set_rflags(vcpu, rflags);
9468 kvm_make_request(KVM_REQ_EVENT, vcpu);
9470 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9472 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9476 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9480 r = kvm_mmu_reload(vcpu);
9484 if (!vcpu->arch.mmu->direct_map &&
9485 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9488 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9491 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9493 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9496 static inline u32 kvm_async_pf_next_probe(u32 key)
9498 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9501 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9503 u32 key = kvm_async_pf_hash_fn(gfn);
9505 while (vcpu->arch.apf.gfns[key] != ~0)
9506 key = kvm_async_pf_next_probe(key);
9508 vcpu->arch.apf.gfns[key] = gfn;
9511 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9514 u32 key = kvm_async_pf_hash_fn(gfn);
9516 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9517 (vcpu->arch.apf.gfns[key] != gfn &&
9518 vcpu->arch.apf.gfns[key] != ~0); i++)
9519 key = kvm_async_pf_next_probe(key);
9524 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9526 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9529 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9533 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9535 vcpu->arch.apf.gfns[i] = ~0;
9537 j = kvm_async_pf_next_probe(j);
9538 if (vcpu->arch.apf.gfns[j] == ~0)
9540 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9542 * k lies cyclically in ]i,j]
9544 * |....j i.k.| or |.k..j i...|
9546 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9547 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9552 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9555 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9559 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9562 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9566 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9567 struct kvm_async_pf *work)
9569 struct x86_exception fault;
9571 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9572 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9574 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9575 (vcpu->arch.apf.send_user_only &&
9576 kvm_x86_ops->get_cpl(vcpu) == 0))
9577 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9578 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9579 fault.vector = PF_VECTOR;
9580 fault.error_code_valid = true;
9581 fault.error_code = 0;
9582 fault.nested_page_fault = false;
9583 fault.address = work->arch.token;
9584 fault.async_page_fault = true;
9585 kvm_inject_page_fault(vcpu, &fault);
9589 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9590 struct kvm_async_pf *work)
9592 struct x86_exception fault;
9595 if (work->wakeup_all)
9596 work->arch.token = ~0; /* broadcast wakeup */
9598 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9599 trace_kvm_async_pf_ready(work->arch.token, work->gva);
9601 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9602 !apf_get_user(vcpu, &val)) {
9603 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9604 vcpu->arch.exception.pending &&
9605 vcpu->arch.exception.nr == PF_VECTOR &&
9606 !apf_put_user(vcpu, 0)) {
9607 vcpu->arch.exception.injected = false;
9608 vcpu->arch.exception.pending = false;
9609 vcpu->arch.exception.nr = 0;
9610 vcpu->arch.exception.has_error_code = false;
9611 vcpu->arch.exception.error_code = 0;
9612 vcpu->arch.exception.has_payload = false;
9613 vcpu->arch.exception.payload = 0;
9614 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9615 fault.vector = PF_VECTOR;
9616 fault.error_code_valid = true;
9617 fault.error_code = 0;
9618 fault.nested_page_fault = false;
9619 fault.address = work->arch.token;
9620 fault.async_page_fault = true;
9621 kvm_inject_page_fault(vcpu, &fault);
9624 vcpu->arch.apf.halted = false;
9625 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9628 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9630 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9633 return kvm_can_do_async_pf(vcpu);
9636 void kvm_arch_start_assignment(struct kvm *kvm)
9638 atomic_inc(&kvm->arch.assigned_device_count);
9640 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9642 void kvm_arch_end_assignment(struct kvm *kvm)
9644 atomic_dec(&kvm->arch.assigned_device_count);
9646 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9648 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9650 return atomic_read(&kvm->arch.assigned_device_count);
9652 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9654 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9656 atomic_inc(&kvm->arch.noncoherent_dma_count);
9658 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9660 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9662 atomic_dec(&kvm->arch.noncoherent_dma_count);
9664 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9666 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9668 return atomic_read(&kvm->arch.noncoherent_dma_count);
9670 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9672 bool kvm_arch_has_irq_bypass(void)
9674 return kvm_x86_ops->update_pi_irte != NULL;
9677 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9678 struct irq_bypass_producer *prod)
9680 struct kvm_kernel_irqfd *irqfd =
9681 container_of(cons, struct kvm_kernel_irqfd, consumer);
9683 irqfd->producer = prod;
9685 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9686 prod->irq, irqfd->gsi, 1);
9689 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9690 struct irq_bypass_producer *prod)
9693 struct kvm_kernel_irqfd *irqfd =
9694 container_of(cons, struct kvm_kernel_irqfd, consumer);
9696 WARN_ON(irqfd->producer != prod);
9697 irqfd->producer = NULL;
9700 * When producer of consumer is unregistered, we change back to
9701 * remapped mode, so we can re-use the current implementation
9702 * when the irq is masked/disabled or the consumer side (KVM
9703 * int this case doesn't want to receive the interrupts.
9705 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9707 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9708 " fails: %d\n", irqfd->consumer.token, ret);
9711 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9712 uint32_t guest_irq, bool set)
9714 if (!kvm_x86_ops->update_pi_irte)
9717 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9720 bool kvm_vector_hashing_enabled(void)
9722 return vector_hashing;
9724 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9726 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9727 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9728 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9729 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9730 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9731 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9732 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9733 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9734 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9735 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9736 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9737 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9738 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9739 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9740 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9741 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9742 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9743 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9744 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);