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 = 0;
140 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
142 static bool __read_mostly vector_hashing = true;
143 module_param(vector_hashing, bool, S_IRUGO);
145 bool __read_mostly enable_vmware_backdoor = false;
146 module_param(enable_vmware_backdoor, bool, S_IRUGO);
147 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
149 static bool __read_mostly force_emulation_prefix = false;
150 module_param(force_emulation_prefix, bool, S_IRUGO);
152 #define KVM_NR_SHARED_MSRS 16
154 struct kvm_shared_msrs_global {
156 u32 msrs[KVM_NR_SHARED_MSRS];
159 struct kvm_shared_msrs {
160 struct user_return_notifier urn;
162 struct kvm_shared_msr_values {
165 } values[KVM_NR_SHARED_MSRS];
168 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
169 static struct kvm_shared_msrs __percpu *shared_msrs;
171 struct kvm_stats_debugfs_item debugfs_entries[] = {
172 { "pf_fixed", VCPU_STAT(pf_fixed) },
173 { "pf_guest", VCPU_STAT(pf_guest) },
174 { "tlb_flush", VCPU_STAT(tlb_flush) },
175 { "invlpg", VCPU_STAT(invlpg) },
176 { "exits", VCPU_STAT(exits) },
177 { "io_exits", VCPU_STAT(io_exits) },
178 { "mmio_exits", VCPU_STAT(mmio_exits) },
179 { "signal_exits", VCPU_STAT(signal_exits) },
180 { "irq_window", VCPU_STAT(irq_window_exits) },
181 { "nmi_window", VCPU_STAT(nmi_window_exits) },
182 { "halt_exits", VCPU_STAT(halt_exits) },
183 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
184 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
185 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
186 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
187 { "hypercalls", VCPU_STAT(hypercalls) },
188 { "request_irq", VCPU_STAT(request_irq_exits) },
189 { "irq_exits", VCPU_STAT(irq_exits) },
190 { "host_state_reload", VCPU_STAT(host_state_reload) },
191 { "fpu_reload", VCPU_STAT(fpu_reload) },
192 { "insn_emulation", VCPU_STAT(insn_emulation) },
193 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
194 { "irq_injections", VCPU_STAT(irq_injections) },
195 { "nmi_injections", VCPU_STAT(nmi_injections) },
196 { "req_event", VCPU_STAT(req_event) },
197 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
198 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
199 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
200 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
201 { "mmu_flooded", VM_STAT(mmu_flooded) },
202 { "mmu_recycled", VM_STAT(mmu_recycled) },
203 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
204 { "mmu_unsync", VM_STAT(mmu_unsync) },
205 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
206 { "largepages", VM_STAT(lpages) },
207 { "max_mmu_page_hash_collisions",
208 VM_STAT(max_mmu_page_hash_collisions) },
212 u64 __read_mostly host_xcr0;
214 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
216 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
219 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
220 vcpu->arch.apf.gfns[i] = ~0;
223 static void kvm_on_user_return(struct user_return_notifier *urn)
226 struct kvm_shared_msrs *locals
227 = container_of(urn, struct kvm_shared_msrs, urn);
228 struct kvm_shared_msr_values *values;
232 * Disabling irqs at this point since the following code could be
233 * interrupted and executed through kvm_arch_hardware_disable()
235 local_irq_save(flags);
236 if (locals->registered) {
237 locals->registered = false;
238 user_return_notifier_unregister(urn);
240 local_irq_restore(flags);
241 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
242 values = &locals->values[slot];
243 if (values->host != values->curr) {
244 wrmsrl(shared_msrs_global.msrs[slot], values->host);
245 values->curr = values->host;
250 static void shared_msr_update(unsigned slot, u32 msr)
253 unsigned int cpu = smp_processor_id();
254 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
256 /* only read, and nobody should modify it at this time,
257 * so don't need lock */
258 if (slot >= shared_msrs_global.nr) {
259 printk(KERN_ERR "kvm: invalid MSR slot!");
262 rdmsrl_safe(msr, &value);
263 smsr->values[slot].host = value;
264 smsr->values[slot].curr = value;
267 void kvm_define_shared_msr(unsigned slot, u32 msr)
269 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
270 shared_msrs_global.msrs[slot] = msr;
271 if (slot >= shared_msrs_global.nr)
272 shared_msrs_global.nr = slot + 1;
274 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
276 static void kvm_shared_msr_cpu_online(void)
280 for (i = 0; i < shared_msrs_global.nr; ++i)
281 shared_msr_update(i, shared_msrs_global.msrs[i]);
284 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
286 unsigned int cpu = smp_processor_id();
287 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
290 if (((value ^ smsr->values[slot].curr) & mask) == 0)
292 smsr->values[slot].curr = value;
293 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
297 if (!smsr->registered) {
298 smsr->urn.on_user_return = kvm_on_user_return;
299 user_return_notifier_register(&smsr->urn);
300 smsr->registered = true;
304 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
306 static void drop_user_return_notifiers(void)
308 unsigned int cpu = smp_processor_id();
309 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
311 if (smsr->registered)
312 kvm_on_user_return(&smsr->urn);
315 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
317 return vcpu->arch.apic_base;
319 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
321 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
323 u64 old_state = vcpu->arch.apic_base &
324 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
325 u64 new_state = msr_info->data &
326 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
327 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
328 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
330 if ((msr_info->data & reserved_bits) || new_state == X2APIC_ENABLE)
332 if (!msr_info->host_initiated &&
333 ((new_state == MSR_IA32_APICBASE_ENABLE &&
334 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
335 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
339 kvm_lapic_set_base(vcpu, msr_info->data);
342 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
344 asmlinkage __visible void kvm_spurious_fault(void)
346 /* Fault while not rebooting. We want the trace. */
349 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
351 #define EXCPT_BENIGN 0
352 #define EXCPT_CONTRIBUTORY 1
355 static int exception_class(int vector)
365 return EXCPT_CONTRIBUTORY;
372 #define EXCPT_FAULT 0
374 #define EXCPT_ABORT 2
375 #define EXCPT_INTERRUPT 3
377 static int exception_type(int vector)
381 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
382 return EXCPT_INTERRUPT;
386 /* #DB is trap, as instruction watchpoints are handled elsewhere */
387 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
390 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
393 /* Reserved exceptions will result in fault */
397 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
398 unsigned nr, bool has_error, u32 error_code,
404 kvm_make_request(KVM_REQ_EVENT, vcpu);
406 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
408 if (has_error && !is_protmode(vcpu))
412 * On vmentry, vcpu->arch.exception.pending is only
413 * true if an event injection was blocked by
414 * nested_run_pending. In that case, however,
415 * vcpu_enter_guest requests an immediate exit,
416 * and the guest shouldn't proceed far enough to
419 WARN_ON_ONCE(vcpu->arch.exception.pending);
420 vcpu->arch.exception.injected = true;
422 vcpu->arch.exception.pending = true;
423 vcpu->arch.exception.injected = false;
425 vcpu->arch.exception.has_error_code = has_error;
426 vcpu->arch.exception.nr = nr;
427 vcpu->arch.exception.error_code = error_code;
431 /* to check exception */
432 prev_nr = vcpu->arch.exception.nr;
433 if (prev_nr == DF_VECTOR) {
434 /* triple fault -> shutdown */
435 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
438 class1 = exception_class(prev_nr);
439 class2 = exception_class(nr);
440 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
441 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
443 * Generate double fault per SDM Table 5-5. Set
444 * exception.pending = true so that the double fault
445 * can trigger a nested vmexit.
447 vcpu->arch.exception.pending = true;
448 vcpu->arch.exception.injected = false;
449 vcpu->arch.exception.has_error_code = true;
450 vcpu->arch.exception.nr = DF_VECTOR;
451 vcpu->arch.exception.error_code = 0;
453 /* replace previous exception with a new one in a hope
454 that instruction re-execution will regenerate lost
459 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
461 kvm_multiple_exception(vcpu, nr, false, 0, false);
463 EXPORT_SYMBOL_GPL(kvm_queue_exception);
465 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
467 kvm_multiple_exception(vcpu, nr, false, 0, true);
469 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
471 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
474 kvm_inject_gp(vcpu, 0);
476 return kvm_skip_emulated_instruction(vcpu);
480 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
482 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
484 ++vcpu->stat.pf_guest;
485 vcpu->arch.exception.nested_apf =
486 is_guest_mode(vcpu) && fault->async_page_fault;
487 if (vcpu->arch.exception.nested_apf)
488 vcpu->arch.apf.nested_apf_token = fault->address;
490 vcpu->arch.cr2 = fault->address;
491 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
493 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
495 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
497 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
498 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
500 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
502 return fault->nested_page_fault;
505 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
507 atomic_inc(&vcpu->arch.nmi_queued);
508 kvm_make_request(KVM_REQ_NMI, vcpu);
510 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
512 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
514 kvm_multiple_exception(vcpu, nr, true, error_code, false);
516 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
518 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
520 kvm_multiple_exception(vcpu, nr, true, error_code, true);
522 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
525 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
526 * a #GP and return false.
528 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
530 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
532 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
535 EXPORT_SYMBOL_GPL(kvm_require_cpl);
537 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
539 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
542 kvm_queue_exception(vcpu, UD_VECTOR);
545 EXPORT_SYMBOL_GPL(kvm_require_dr);
548 * This function will be used to read from the physical memory of the currently
549 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
550 * can read from guest physical or from the guest's guest physical memory.
552 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
553 gfn_t ngfn, void *data, int offset, int len,
556 struct x86_exception exception;
560 ngpa = gfn_to_gpa(ngfn);
561 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
562 if (real_gfn == UNMAPPED_GVA)
565 real_gfn = gpa_to_gfn(real_gfn);
567 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
569 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
571 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
572 void *data, int offset, int len, u32 access)
574 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
575 data, offset, len, access);
579 * Load the pae pdptrs. Return true is they are all valid.
581 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
583 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
584 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
587 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
589 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
590 offset * sizeof(u64), sizeof(pdpte),
591 PFERR_USER_MASK|PFERR_WRITE_MASK);
596 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
597 if ((pdpte[i] & PT_PRESENT_MASK) &&
599 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
606 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
607 __set_bit(VCPU_EXREG_PDPTR,
608 (unsigned long *)&vcpu->arch.regs_avail);
609 __set_bit(VCPU_EXREG_PDPTR,
610 (unsigned long *)&vcpu->arch.regs_dirty);
615 EXPORT_SYMBOL_GPL(load_pdptrs);
617 bool pdptrs_changed(struct kvm_vcpu *vcpu)
619 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
625 if (is_long_mode(vcpu) || !is_pae(vcpu))
628 if (!test_bit(VCPU_EXREG_PDPTR,
629 (unsigned long *)&vcpu->arch.regs_avail))
632 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
633 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
634 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
635 PFERR_USER_MASK | PFERR_WRITE_MASK);
638 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
643 EXPORT_SYMBOL_GPL(pdptrs_changed);
645 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
647 unsigned long old_cr0 = kvm_read_cr0(vcpu);
648 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
653 if (cr0 & 0xffffffff00000000UL)
657 cr0 &= ~CR0_RESERVED_BITS;
659 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
662 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
665 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
667 if ((vcpu->arch.efer & EFER_LME)) {
672 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
677 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
682 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
685 kvm_x86_ops->set_cr0(vcpu, cr0);
687 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
688 kvm_clear_async_pf_completion_queue(vcpu);
689 kvm_async_pf_hash_reset(vcpu);
692 if ((cr0 ^ old_cr0) & update_bits)
693 kvm_mmu_reset_context(vcpu);
695 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
696 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
697 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
698 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
702 EXPORT_SYMBOL_GPL(kvm_set_cr0);
704 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
706 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
708 EXPORT_SYMBOL_GPL(kvm_lmsw);
710 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
712 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
713 !vcpu->guest_xcr0_loaded) {
714 /* kvm_set_xcr() also depends on this */
715 if (vcpu->arch.xcr0 != host_xcr0)
716 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
717 vcpu->guest_xcr0_loaded = 1;
721 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
723 if (vcpu->guest_xcr0_loaded) {
724 if (vcpu->arch.xcr0 != host_xcr0)
725 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
726 vcpu->guest_xcr0_loaded = 0;
730 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
733 u64 old_xcr0 = vcpu->arch.xcr0;
736 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
737 if (index != XCR_XFEATURE_ENABLED_MASK)
739 if (!(xcr0 & XFEATURE_MASK_FP))
741 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
745 * Do not allow the guest to set bits that we do not support
746 * saving. However, xcr0 bit 0 is always set, even if the
747 * emulated CPU does not support XSAVE (see fx_init).
749 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
750 if (xcr0 & ~valid_bits)
753 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
754 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
757 if (xcr0 & XFEATURE_MASK_AVX512) {
758 if (!(xcr0 & XFEATURE_MASK_YMM))
760 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
763 vcpu->arch.xcr0 = xcr0;
765 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
766 kvm_update_cpuid(vcpu);
770 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
772 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
773 __kvm_set_xcr(vcpu, index, xcr)) {
774 kvm_inject_gp(vcpu, 0);
779 EXPORT_SYMBOL_GPL(kvm_set_xcr);
781 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
783 unsigned long old_cr4 = kvm_read_cr4(vcpu);
784 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
785 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
787 if (cr4 & CR4_RESERVED_BITS)
790 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
793 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
796 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
799 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
802 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
805 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
808 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
811 if (is_long_mode(vcpu)) {
812 if (!(cr4 & X86_CR4_PAE))
814 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
815 && ((cr4 ^ old_cr4) & pdptr_bits)
816 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
820 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
821 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
824 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
825 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
829 if (kvm_x86_ops->set_cr4(vcpu, cr4))
832 if (((cr4 ^ old_cr4) & pdptr_bits) ||
833 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
834 kvm_mmu_reset_context(vcpu);
836 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
837 kvm_update_cpuid(vcpu);
841 EXPORT_SYMBOL_GPL(kvm_set_cr4);
843 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
846 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
849 cr3 &= ~CR3_PCID_INVD;
852 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
853 kvm_mmu_sync_roots(vcpu);
854 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
858 if (is_long_mode(vcpu) &&
859 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 62)))
861 else if (is_pae(vcpu) && is_paging(vcpu) &&
862 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
865 vcpu->arch.cr3 = cr3;
866 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
867 kvm_mmu_new_cr3(vcpu);
870 EXPORT_SYMBOL_GPL(kvm_set_cr3);
872 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
874 if (cr8 & CR8_RESERVED_BITS)
876 if (lapic_in_kernel(vcpu))
877 kvm_lapic_set_tpr(vcpu, cr8);
879 vcpu->arch.cr8 = cr8;
882 EXPORT_SYMBOL_GPL(kvm_set_cr8);
884 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
886 if (lapic_in_kernel(vcpu))
887 return kvm_lapic_get_cr8(vcpu);
889 return vcpu->arch.cr8;
891 EXPORT_SYMBOL_GPL(kvm_get_cr8);
893 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
897 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
898 for (i = 0; i < KVM_NR_DB_REGS; i++)
899 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
900 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
904 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
906 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
907 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
910 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
914 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
915 dr7 = vcpu->arch.guest_debug_dr7;
917 dr7 = vcpu->arch.dr7;
918 kvm_x86_ops->set_dr7(vcpu, dr7);
919 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
920 if (dr7 & DR7_BP_EN_MASK)
921 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
924 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
926 u64 fixed = DR6_FIXED_1;
928 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
933 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
937 vcpu->arch.db[dr] = val;
938 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
939 vcpu->arch.eff_db[dr] = val;
944 if (val & 0xffffffff00000000ULL)
946 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
947 kvm_update_dr6(vcpu);
952 if (val & 0xffffffff00000000ULL)
954 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
955 kvm_update_dr7(vcpu);
962 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
964 if (__kvm_set_dr(vcpu, dr, val)) {
965 kvm_inject_gp(vcpu, 0);
970 EXPORT_SYMBOL_GPL(kvm_set_dr);
972 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
976 *val = vcpu->arch.db[dr];
981 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
982 *val = vcpu->arch.dr6;
984 *val = kvm_x86_ops->get_dr6(vcpu);
989 *val = vcpu->arch.dr7;
994 EXPORT_SYMBOL_GPL(kvm_get_dr);
996 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
998 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
1002 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1005 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1006 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1009 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1012 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1013 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1015 * This list is modified at module load time to reflect the
1016 * capabilities of the host cpu. This capabilities test skips MSRs that are
1017 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1018 * may depend on host virtualization features rather than host cpu features.
1021 static u32 msrs_to_save[] = {
1022 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1024 #ifdef CONFIG_X86_64
1025 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1027 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1028 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1029 MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
1032 static unsigned num_msrs_to_save;
1034 static u32 emulated_msrs[] = {
1035 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1036 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1037 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1038 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1039 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1040 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1041 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1043 HV_X64_MSR_VP_INDEX,
1044 HV_X64_MSR_VP_RUNTIME,
1045 HV_X64_MSR_SCONTROL,
1046 HV_X64_MSR_STIMER0_CONFIG,
1047 HV_X64_MSR_VP_ASSIST_PAGE,
1048 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1049 HV_X64_MSR_TSC_EMULATION_STATUS,
1051 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1054 MSR_IA32_TSC_ADJUST,
1055 MSR_IA32_TSCDEADLINE,
1056 MSR_IA32_MISC_ENABLE,
1057 MSR_IA32_MCG_STATUS,
1059 MSR_IA32_MCG_EXT_CTL,
1063 MSR_MISC_FEATURES_ENABLES,
1064 MSR_AMD64_VIRT_SPEC_CTRL,
1067 static unsigned num_emulated_msrs;
1070 * List of msr numbers which are used to expose MSR-based features that
1071 * can be used by a hypervisor to validate requested CPU features.
1073 static u32 msr_based_features[] = {
1075 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1076 MSR_IA32_VMX_PINBASED_CTLS,
1077 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1078 MSR_IA32_VMX_PROCBASED_CTLS,
1079 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1080 MSR_IA32_VMX_EXIT_CTLS,
1081 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1082 MSR_IA32_VMX_ENTRY_CTLS,
1084 MSR_IA32_VMX_CR0_FIXED0,
1085 MSR_IA32_VMX_CR0_FIXED1,
1086 MSR_IA32_VMX_CR4_FIXED0,
1087 MSR_IA32_VMX_CR4_FIXED1,
1088 MSR_IA32_VMX_VMCS_ENUM,
1089 MSR_IA32_VMX_PROCBASED_CTLS2,
1090 MSR_IA32_VMX_EPT_VPID_CAP,
1091 MSR_IA32_VMX_VMFUNC,
1095 MSR_IA32_ARCH_CAPABILITIES,
1098 static unsigned int num_msr_based_features;
1100 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1102 switch (msr->index) {
1103 case MSR_IA32_UCODE_REV:
1104 case MSR_IA32_ARCH_CAPABILITIES:
1105 rdmsrl_safe(msr->index, &msr->data);
1108 if (kvm_x86_ops->get_msr_feature(msr))
1114 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1116 struct kvm_msr_entry msr;
1120 r = kvm_get_msr_feature(&msr);
1129 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1131 if (efer & efer_reserved_bits)
1134 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1137 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1142 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1144 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1146 u64 old_efer = vcpu->arch.efer;
1148 if (!kvm_valid_efer(vcpu, efer))
1152 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1156 efer |= vcpu->arch.efer & EFER_LMA;
1158 kvm_x86_ops->set_efer(vcpu, efer);
1160 /* Update reserved bits */
1161 if ((efer ^ old_efer) & EFER_NX)
1162 kvm_mmu_reset_context(vcpu);
1167 void kvm_enable_efer_bits(u64 mask)
1169 efer_reserved_bits &= ~mask;
1171 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1174 * Writes msr value into into the appropriate "register".
1175 * Returns 0 on success, non-0 otherwise.
1176 * Assumes vcpu_load() was already called.
1178 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1180 switch (msr->index) {
1183 case MSR_KERNEL_GS_BASE:
1186 if (is_noncanonical_address(msr->data, vcpu))
1189 case MSR_IA32_SYSENTER_EIP:
1190 case MSR_IA32_SYSENTER_ESP:
1192 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1193 * non-canonical address is written on Intel but not on
1194 * AMD (which ignores the top 32-bits, because it does
1195 * not implement 64-bit SYSENTER).
1197 * 64-bit code should hence be able to write a non-canonical
1198 * value on AMD. Making the address canonical ensures that
1199 * vmentry does not fail on Intel after writing a non-canonical
1200 * value, and that something deterministic happens if the guest
1201 * invokes 64-bit SYSENTER.
1203 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1205 return kvm_x86_ops->set_msr(vcpu, msr);
1207 EXPORT_SYMBOL_GPL(kvm_set_msr);
1210 * Adapt set_msr() to msr_io()'s calling convention
1212 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1214 struct msr_data msr;
1218 msr.host_initiated = true;
1219 r = kvm_get_msr(vcpu, &msr);
1227 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1229 struct msr_data msr;
1233 msr.host_initiated = true;
1234 return kvm_set_msr(vcpu, &msr);
1237 #ifdef CONFIG_X86_64
1238 struct pvclock_gtod_data {
1241 struct { /* extract of a clocksource struct */
1254 static struct pvclock_gtod_data pvclock_gtod_data;
1256 static void update_pvclock_gtod(struct timekeeper *tk)
1258 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1261 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1263 write_seqcount_begin(&vdata->seq);
1265 /* copy pvclock gtod data */
1266 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1267 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1268 vdata->clock.mask = tk->tkr_mono.mask;
1269 vdata->clock.mult = tk->tkr_mono.mult;
1270 vdata->clock.shift = tk->tkr_mono.shift;
1272 vdata->boot_ns = boot_ns;
1273 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1275 vdata->wall_time_sec = tk->xtime_sec;
1277 write_seqcount_end(&vdata->seq);
1281 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1284 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1285 * vcpu_enter_guest. This function is only called from
1286 * the physical CPU that is running vcpu.
1288 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1291 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1295 struct pvclock_wall_clock wc;
1296 struct timespec64 boot;
1301 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1306 ++version; /* first time write, random junk */
1310 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1314 * The guest calculates current wall clock time by adding
1315 * system time (updated by kvm_guest_time_update below) to the
1316 * wall clock specified here. guest system time equals host
1317 * system time for us, thus we must fill in host boot time here.
1319 getboottime64(&boot);
1321 if (kvm->arch.kvmclock_offset) {
1322 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1323 boot = timespec64_sub(boot, ts);
1325 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1326 wc.nsec = boot.tv_nsec;
1327 wc.version = version;
1329 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1332 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1335 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1337 do_shl32_div32(dividend, divisor);
1341 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1342 s8 *pshift, u32 *pmultiplier)
1350 scaled64 = scaled_hz;
1351 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1356 tps32 = (uint32_t)tps64;
1357 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1358 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1366 *pmultiplier = div_frac(scaled64, tps32);
1368 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1369 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1372 #ifdef CONFIG_X86_64
1373 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1376 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1377 static unsigned long max_tsc_khz;
1379 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1381 u64 v = (u64)khz * (1000000 + ppm);
1386 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1390 /* Guest TSC same frequency as host TSC? */
1392 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1396 /* TSC scaling supported? */
1397 if (!kvm_has_tsc_control) {
1398 if (user_tsc_khz > tsc_khz) {
1399 vcpu->arch.tsc_catchup = 1;
1400 vcpu->arch.tsc_always_catchup = 1;
1403 WARN(1, "user requested TSC rate below hardware speed\n");
1408 /* TSC scaling required - calculate ratio */
1409 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1410 user_tsc_khz, tsc_khz);
1412 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1413 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1418 vcpu->arch.tsc_scaling_ratio = ratio;
1422 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1424 u32 thresh_lo, thresh_hi;
1425 int use_scaling = 0;
1427 /* tsc_khz can be zero if TSC calibration fails */
1428 if (user_tsc_khz == 0) {
1429 /* set tsc_scaling_ratio to a safe value */
1430 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1434 /* Compute a scale to convert nanoseconds in TSC cycles */
1435 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1436 &vcpu->arch.virtual_tsc_shift,
1437 &vcpu->arch.virtual_tsc_mult);
1438 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1441 * Compute the variation in TSC rate which is acceptable
1442 * within the range of tolerance and decide if the
1443 * rate being applied is within that bounds of the hardware
1444 * rate. If so, no scaling or compensation need be done.
1446 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1447 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1448 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1449 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1452 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1455 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1457 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1458 vcpu->arch.virtual_tsc_mult,
1459 vcpu->arch.virtual_tsc_shift);
1460 tsc += vcpu->arch.this_tsc_write;
1464 static inline int gtod_is_based_on_tsc(int mode)
1466 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1469 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1471 #ifdef CONFIG_X86_64
1473 struct kvm_arch *ka = &vcpu->kvm->arch;
1474 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1476 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1477 atomic_read(&vcpu->kvm->online_vcpus));
1480 * Once the masterclock is enabled, always perform request in
1481 * order to update it.
1483 * In order to enable masterclock, the host clocksource must be TSC
1484 * and the vcpus need to have matched TSCs. When that happens,
1485 * perform request to enable masterclock.
1487 if (ka->use_master_clock ||
1488 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1489 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1491 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1492 atomic_read(&vcpu->kvm->online_vcpus),
1493 ka->use_master_clock, gtod->clock.vclock_mode);
1497 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1499 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1500 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1504 * Multiply tsc by a fixed point number represented by ratio.
1506 * The most significant 64-N bits (mult) of ratio represent the
1507 * integral part of the fixed point number; the remaining N bits
1508 * (frac) represent the fractional part, ie. ratio represents a fixed
1509 * point number (mult + frac * 2^(-N)).
1511 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1513 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1515 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1518 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1521 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1523 if (ratio != kvm_default_tsc_scaling_ratio)
1524 _tsc = __scale_tsc(ratio, tsc);
1528 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1530 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1534 tsc = kvm_scale_tsc(vcpu, rdtsc());
1536 return target_tsc - tsc;
1539 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1541 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1543 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1545 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1547 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1549 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1550 vcpu->arch.tsc_offset = offset;
1553 static inline bool kvm_check_tsc_unstable(void)
1555 #ifdef CONFIG_X86_64
1557 * TSC is marked unstable when we're running on Hyper-V,
1558 * 'TSC page' clocksource is good.
1560 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1563 return check_tsc_unstable();
1566 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1568 struct kvm *kvm = vcpu->kvm;
1569 u64 offset, ns, elapsed;
1570 unsigned long flags;
1572 bool already_matched;
1573 u64 data = msr->data;
1574 bool synchronizing = false;
1576 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1577 offset = kvm_compute_tsc_offset(vcpu, data);
1578 ns = ktime_get_boot_ns();
1579 elapsed = ns - kvm->arch.last_tsc_nsec;
1581 if (vcpu->arch.virtual_tsc_khz) {
1582 if (data == 0 && msr->host_initiated) {
1584 * detection of vcpu initialization -- need to sync
1585 * with other vCPUs. This particularly helps to keep
1586 * kvm_clock stable after CPU hotplug
1588 synchronizing = true;
1590 u64 tsc_exp = kvm->arch.last_tsc_write +
1591 nsec_to_cycles(vcpu, elapsed);
1592 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1594 * Special case: TSC write with a small delta (1 second)
1595 * of virtual cycle time against real time is
1596 * interpreted as an attempt to synchronize the CPU.
1598 synchronizing = data < tsc_exp + tsc_hz &&
1599 data + tsc_hz > tsc_exp;
1604 * For a reliable TSC, we can match TSC offsets, and for an unstable
1605 * TSC, we add elapsed time in this computation. We could let the
1606 * compensation code attempt to catch up if we fall behind, but
1607 * it's better to try to match offsets from the beginning.
1609 if (synchronizing &&
1610 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1611 if (!kvm_check_tsc_unstable()) {
1612 offset = kvm->arch.cur_tsc_offset;
1613 pr_debug("kvm: matched tsc offset for %llu\n", data);
1615 u64 delta = nsec_to_cycles(vcpu, elapsed);
1617 offset = kvm_compute_tsc_offset(vcpu, data);
1618 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1621 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1624 * We split periods of matched TSC writes into generations.
1625 * For each generation, we track the original measured
1626 * nanosecond time, offset, and write, so if TSCs are in
1627 * sync, we can match exact offset, and if not, we can match
1628 * exact software computation in compute_guest_tsc()
1630 * These values are tracked in kvm->arch.cur_xxx variables.
1632 kvm->arch.cur_tsc_generation++;
1633 kvm->arch.cur_tsc_nsec = ns;
1634 kvm->arch.cur_tsc_write = data;
1635 kvm->arch.cur_tsc_offset = offset;
1637 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1638 kvm->arch.cur_tsc_generation, data);
1642 * We also track th most recent recorded KHZ, write and time to
1643 * allow the matching interval to be extended at each write.
1645 kvm->arch.last_tsc_nsec = ns;
1646 kvm->arch.last_tsc_write = data;
1647 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1649 vcpu->arch.last_guest_tsc = data;
1651 /* Keep track of which generation this VCPU has synchronized to */
1652 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1653 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1654 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1656 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1657 update_ia32_tsc_adjust_msr(vcpu, offset);
1659 kvm_vcpu_write_tsc_offset(vcpu, offset);
1660 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1662 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1664 kvm->arch.nr_vcpus_matched_tsc = 0;
1665 } else if (!already_matched) {
1666 kvm->arch.nr_vcpus_matched_tsc++;
1669 kvm_track_tsc_matching(vcpu);
1670 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1673 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1675 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1678 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1681 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1683 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1684 WARN_ON(adjustment < 0);
1685 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1686 adjust_tsc_offset_guest(vcpu, adjustment);
1689 #ifdef CONFIG_X86_64
1691 static u64 read_tsc(void)
1693 u64 ret = (u64)rdtsc_ordered();
1694 u64 last = pvclock_gtod_data.clock.cycle_last;
1696 if (likely(ret >= last))
1700 * GCC likes to generate cmov here, but this branch is extremely
1701 * predictable (it's just a function of time and the likely is
1702 * very likely) and there's a data dependence, so force GCC
1703 * to generate a branch instead. I don't barrier() because
1704 * we don't actually need a barrier, and if this function
1705 * ever gets inlined it will generate worse code.
1711 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1714 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1717 switch (gtod->clock.vclock_mode) {
1718 case VCLOCK_HVCLOCK:
1719 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1721 if (tsc_pg_val != U64_MAX) {
1722 /* TSC page valid */
1723 *mode = VCLOCK_HVCLOCK;
1724 v = (tsc_pg_val - gtod->clock.cycle_last) &
1727 /* TSC page invalid */
1728 *mode = VCLOCK_NONE;
1733 *tsc_timestamp = read_tsc();
1734 v = (*tsc_timestamp - gtod->clock.cycle_last) &
1738 *mode = VCLOCK_NONE;
1741 if (*mode == VCLOCK_NONE)
1742 *tsc_timestamp = v = 0;
1744 return v * gtod->clock.mult;
1747 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1749 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1755 seq = read_seqcount_begin(>od->seq);
1756 ns = gtod->nsec_base;
1757 ns += vgettsc(tsc_timestamp, &mode);
1758 ns >>= gtod->clock.shift;
1759 ns += gtod->boot_ns;
1760 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1766 static int do_realtime(struct timespec *ts, u64 *tsc_timestamp)
1768 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1774 seq = read_seqcount_begin(>od->seq);
1775 ts->tv_sec = gtod->wall_time_sec;
1776 ns = gtod->nsec_base;
1777 ns += vgettsc(tsc_timestamp, &mode);
1778 ns >>= gtod->clock.shift;
1779 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1781 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1787 /* returns true if host is using TSC based clocksource */
1788 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1790 /* checked again under seqlock below */
1791 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1794 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1798 /* returns true if host is using TSC based clocksource */
1799 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1802 /* checked again under seqlock below */
1803 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1806 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1812 * Assuming a stable TSC across physical CPUS, and a stable TSC
1813 * across virtual CPUs, the following condition is possible.
1814 * Each numbered line represents an event visible to both
1815 * CPUs at the next numbered event.
1817 * "timespecX" represents host monotonic time. "tscX" represents
1820 * VCPU0 on CPU0 | VCPU1 on CPU1
1822 * 1. read timespec0,tsc0
1823 * 2. | timespec1 = timespec0 + N
1825 * 3. transition to guest | transition to guest
1826 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1827 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1828 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1830 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1833 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1835 * - 0 < N - M => M < N
1837 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1838 * always the case (the difference between two distinct xtime instances
1839 * might be smaller then the difference between corresponding TSC reads,
1840 * when updating guest vcpus pvclock areas).
1842 * To avoid that problem, do not allow visibility of distinct
1843 * system_timestamp/tsc_timestamp values simultaneously: use a master
1844 * copy of host monotonic time values. Update that master copy
1847 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1851 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1853 #ifdef CONFIG_X86_64
1854 struct kvm_arch *ka = &kvm->arch;
1856 bool host_tsc_clocksource, vcpus_matched;
1858 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1859 atomic_read(&kvm->online_vcpus));
1862 * If the host uses TSC clock, then passthrough TSC as stable
1865 host_tsc_clocksource = kvm_get_time_and_clockread(
1866 &ka->master_kernel_ns,
1867 &ka->master_cycle_now);
1869 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1870 && !ka->backwards_tsc_observed
1871 && !ka->boot_vcpu_runs_old_kvmclock;
1873 if (ka->use_master_clock)
1874 atomic_set(&kvm_guest_has_master_clock, 1);
1876 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1877 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1882 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1884 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1887 static void kvm_gen_update_masterclock(struct kvm *kvm)
1889 #ifdef CONFIG_X86_64
1891 struct kvm_vcpu *vcpu;
1892 struct kvm_arch *ka = &kvm->arch;
1894 spin_lock(&ka->pvclock_gtod_sync_lock);
1895 kvm_make_mclock_inprogress_request(kvm);
1896 /* no guest entries from this point */
1897 pvclock_update_vm_gtod_copy(kvm);
1899 kvm_for_each_vcpu(i, vcpu, kvm)
1900 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1902 /* guest entries allowed */
1903 kvm_for_each_vcpu(i, vcpu, kvm)
1904 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1906 spin_unlock(&ka->pvclock_gtod_sync_lock);
1910 u64 get_kvmclock_ns(struct kvm *kvm)
1912 struct kvm_arch *ka = &kvm->arch;
1913 struct pvclock_vcpu_time_info hv_clock;
1916 spin_lock(&ka->pvclock_gtod_sync_lock);
1917 if (!ka->use_master_clock) {
1918 spin_unlock(&ka->pvclock_gtod_sync_lock);
1919 return ktime_get_boot_ns() + ka->kvmclock_offset;
1922 hv_clock.tsc_timestamp = ka->master_cycle_now;
1923 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1924 spin_unlock(&ka->pvclock_gtod_sync_lock);
1926 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1929 if (__this_cpu_read(cpu_tsc_khz)) {
1930 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1931 &hv_clock.tsc_shift,
1932 &hv_clock.tsc_to_system_mul);
1933 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1935 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
1942 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1944 struct kvm_vcpu_arch *vcpu = &v->arch;
1945 struct pvclock_vcpu_time_info guest_hv_clock;
1947 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1948 &guest_hv_clock, sizeof(guest_hv_clock))))
1951 /* This VCPU is paused, but it's legal for a guest to read another
1952 * VCPU's kvmclock, so we really have to follow the specification where
1953 * it says that version is odd if data is being modified, and even after
1956 * Version field updates must be kept separate. This is because
1957 * kvm_write_guest_cached might use a "rep movs" instruction, and
1958 * writes within a string instruction are weakly ordered. So there
1959 * are three writes overall.
1961 * As a small optimization, only write the version field in the first
1962 * and third write. The vcpu->pv_time cache is still valid, because the
1963 * version field is the first in the struct.
1965 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1967 if (guest_hv_clock.version & 1)
1968 ++guest_hv_clock.version; /* first time write, random junk */
1970 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1971 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1973 sizeof(vcpu->hv_clock.version));
1977 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1978 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1980 if (vcpu->pvclock_set_guest_stopped_request) {
1981 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1982 vcpu->pvclock_set_guest_stopped_request = false;
1985 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1987 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1989 sizeof(vcpu->hv_clock));
1993 vcpu->hv_clock.version++;
1994 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1996 sizeof(vcpu->hv_clock.version));
1999 static int kvm_guest_time_update(struct kvm_vcpu *v)
2001 unsigned long flags, tgt_tsc_khz;
2002 struct kvm_vcpu_arch *vcpu = &v->arch;
2003 struct kvm_arch *ka = &v->kvm->arch;
2005 u64 tsc_timestamp, host_tsc;
2007 bool use_master_clock;
2013 * If the host uses TSC clock, then passthrough TSC as stable
2016 spin_lock(&ka->pvclock_gtod_sync_lock);
2017 use_master_clock = ka->use_master_clock;
2018 if (use_master_clock) {
2019 host_tsc = ka->master_cycle_now;
2020 kernel_ns = ka->master_kernel_ns;
2022 spin_unlock(&ka->pvclock_gtod_sync_lock);
2024 /* Keep irq disabled to prevent changes to the clock */
2025 local_irq_save(flags);
2026 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2027 if (unlikely(tgt_tsc_khz == 0)) {
2028 local_irq_restore(flags);
2029 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2032 if (!use_master_clock) {
2034 kernel_ns = ktime_get_boot_ns();
2037 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2040 * We may have to catch up the TSC to match elapsed wall clock
2041 * time for two reasons, even if kvmclock is used.
2042 * 1) CPU could have been running below the maximum TSC rate
2043 * 2) Broken TSC compensation resets the base at each VCPU
2044 * entry to avoid unknown leaps of TSC even when running
2045 * again on the same CPU. This may cause apparent elapsed
2046 * time to disappear, and the guest to stand still or run
2049 if (vcpu->tsc_catchup) {
2050 u64 tsc = compute_guest_tsc(v, kernel_ns);
2051 if (tsc > tsc_timestamp) {
2052 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2053 tsc_timestamp = tsc;
2057 local_irq_restore(flags);
2059 /* With all the info we got, fill in the values */
2061 if (kvm_has_tsc_control)
2062 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2064 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2065 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2066 &vcpu->hv_clock.tsc_shift,
2067 &vcpu->hv_clock.tsc_to_system_mul);
2068 vcpu->hw_tsc_khz = tgt_tsc_khz;
2071 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2072 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2073 vcpu->last_guest_tsc = tsc_timestamp;
2075 /* If the host uses TSC clocksource, then it is stable */
2077 if (use_master_clock)
2078 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2080 vcpu->hv_clock.flags = pvclock_flags;
2082 if (vcpu->pv_time_enabled)
2083 kvm_setup_pvclock_page(v);
2084 if (v == kvm_get_vcpu(v->kvm, 0))
2085 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2090 * kvmclock updates which are isolated to a given vcpu, such as
2091 * vcpu->cpu migration, should not allow system_timestamp from
2092 * the rest of the vcpus to remain static. Otherwise ntp frequency
2093 * correction applies to one vcpu's system_timestamp but not
2096 * So in those cases, request a kvmclock update for all vcpus.
2097 * We need to rate-limit these requests though, as they can
2098 * considerably slow guests that have a large number of vcpus.
2099 * The time for a remote vcpu to update its kvmclock is bound
2100 * by the delay we use to rate-limit the updates.
2103 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2105 static void kvmclock_update_fn(struct work_struct *work)
2108 struct delayed_work *dwork = to_delayed_work(work);
2109 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2110 kvmclock_update_work);
2111 struct kvm *kvm = container_of(ka, struct kvm, arch);
2112 struct kvm_vcpu *vcpu;
2114 kvm_for_each_vcpu(i, vcpu, kvm) {
2115 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2116 kvm_vcpu_kick(vcpu);
2120 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2122 struct kvm *kvm = v->kvm;
2124 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2125 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2126 KVMCLOCK_UPDATE_DELAY);
2129 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2131 static void kvmclock_sync_fn(struct work_struct *work)
2133 struct delayed_work *dwork = to_delayed_work(work);
2134 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2135 kvmclock_sync_work);
2136 struct kvm *kvm = container_of(ka, struct kvm, arch);
2138 if (!kvmclock_periodic_sync)
2141 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2142 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2143 KVMCLOCK_SYNC_PERIOD);
2146 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2148 u64 mcg_cap = vcpu->arch.mcg_cap;
2149 unsigned bank_num = mcg_cap & 0xff;
2150 u32 msr = msr_info->index;
2151 u64 data = msr_info->data;
2154 case MSR_IA32_MCG_STATUS:
2155 vcpu->arch.mcg_status = data;
2157 case MSR_IA32_MCG_CTL:
2158 if (!(mcg_cap & MCG_CTL_P))
2160 if (data != 0 && data != ~(u64)0)
2162 vcpu->arch.mcg_ctl = data;
2165 if (msr >= MSR_IA32_MC0_CTL &&
2166 msr < MSR_IA32_MCx_CTL(bank_num)) {
2167 u32 offset = msr - MSR_IA32_MC0_CTL;
2168 /* only 0 or all 1s can be written to IA32_MCi_CTL
2169 * some Linux kernels though clear bit 10 in bank 4 to
2170 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2171 * this to avoid an uncatched #GP in the guest
2173 if ((offset & 0x3) == 0 &&
2174 data != 0 && (data | (1 << 10)) != ~(u64)0)
2176 if (!msr_info->host_initiated &&
2177 (offset & 0x3) == 1 && data != 0)
2179 vcpu->arch.mce_banks[offset] = data;
2187 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2189 struct kvm *kvm = vcpu->kvm;
2190 int lm = is_long_mode(vcpu);
2191 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2192 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2193 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2194 : kvm->arch.xen_hvm_config.blob_size_32;
2195 u32 page_num = data & ~PAGE_MASK;
2196 u64 page_addr = data & PAGE_MASK;
2201 if (page_num >= blob_size)
2204 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2209 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2218 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2220 gpa_t gpa = data & ~0x3f;
2222 /* Bits 3:5 are reserved, Should be zero */
2226 vcpu->arch.apf.msr_val = data;
2228 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2229 kvm_clear_async_pf_completion_queue(vcpu);
2230 kvm_async_pf_hash_reset(vcpu);
2234 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2238 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2239 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2240 kvm_async_pf_wakeup_all(vcpu);
2244 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2246 vcpu->arch.pv_time_enabled = false;
2249 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2251 ++vcpu->stat.tlb_flush;
2252 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2255 static void record_steal_time(struct kvm_vcpu *vcpu)
2257 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2260 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2261 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2265 * Doing a TLB flush here, on the guest's behalf, can avoid
2268 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2269 kvm_vcpu_flush_tlb(vcpu, false);
2271 if (vcpu->arch.st.steal.version & 1)
2272 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2274 vcpu->arch.st.steal.version += 1;
2276 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2277 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2281 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2282 vcpu->arch.st.last_steal;
2283 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2285 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2286 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2290 vcpu->arch.st.steal.version += 1;
2292 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2293 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2296 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2299 u32 msr = msr_info->index;
2300 u64 data = msr_info->data;
2303 case MSR_AMD64_NB_CFG:
2304 case MSR_IA32_UCODE_WRITE:
2305 case MSR_VM_HSAVE_PA:
2306 case MSR_AMD64_PATCH_LOADER:
2307 case MSR_AMD64_BU_CFG2:
2308 case MSR_AMD64_DC_CFG:
2311 case MSR_IA32_UCODE_REV:
2312 if (msr_info->host_initiated)
2313 vcpu->arch.microcode_version = data;
2316 return set_efer(vcpu, data);
2318 data &= ~(u64)0x40; /* ignore flush filter disable */
2319 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2320 data &= ~(u64)0x8; /* ignore TLB cache disable */
2321 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2323 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2328 case MSR_FAM10H_MMIO_CONF_BASE:
2330 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2335 case MSR_IA32_DEBUGCTLMSR:
2337 /* We support the non-activated case already */
2339 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2340 /* Values other than LBR and BTF are vendor-specific,
2341 thus reserved and should throw a #GP */
2344 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2347 case 0x200 ... 0x2ff:
2348 return kvm_mtrr_set_msr(vcpu, msr, data);
2349 case MSR_IA32_APICBASE:
2350 return kvm_set_apic_base(vcpu, msr_info);
2351 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2352 return kvm_x2apic_msr_write(vcpu, msr, data);
2353 case MSR_IA32_TSCDEADLINE:
2354 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2356 case MSR_IA32_TSC_ADJUST:
2357 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2358 if (!msr_info->host_initiated) {
2359 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2360 adjust_tsc_offset_guest(vcpu, adj);
2362 vcpu->arch.ia32_tsc_adjust_msr = data;
2365 case MSR_IA32_MISC_ENABLE:
2366 vcpu->arch.ia32_misc_enable_msr = data;
2368 case MSR_IA32_SMBASE:
2369 if (!msr_info->host_initiated)
2371 vcpu->arch.smbase = data;
2374 kvm_write_tsc(vcpu, msr_info);
2377 if (!msr_info->host_initiated)
2379 vcpu->arch.smi_count = data;
2381 case MSR_KVM_WALL_CLOCK_NEW:
2382 case MSR_KVM_WALL_CLOCK:
2383 vcpu->kvm->arch.wall_clock = data;
2384 kvm_write_wall_clock(vcpu->kvm, data);
2386 case MSR_KVM_SYSTEM_TIME_NEW:
2387 case MSR_KVM_SYSTEM_TIME: {
2388 struct kvm_arch *ka = &vcpu->kvm->arch;
2390 kvmclock_reset(vcpu);
2392 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2393 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2395 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2396 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2398 ka->boot_vcpu_runs_old_kvmclock = tmp;
2401 vcpu->arch.time = data;
2402 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2404 /* we verify if the enable bit is set... */
2408 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2409 &vcpu->arch.pv_time, data & ~1ULL,
2410 sizeof(struct pvclock_vcpu_time_info)))
2411 vcpu->arch.pv_time_enabled = false;
2413 vcpu->arch.pv_time_enabled = true;
2417 case MSR_KVM_ASYNC_PF_EN:
2418 if (kvm_pv_enable_async_pf(vcpu, data))
2421 case MSR_KVM_STEAL_TIME:
2423 if (unlikely(!sched_info_on()))
2426 if (data & KVM_STEAL_RESERVED_MASK)
2429 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2430 data & KVM_STEAL_VALID_BITS,
2431 sizeof(struct kvm_steal_time)))
2434 vcpu->arch.st.msr_val = data;
2436 if (!(data & KVM_MSR_ENABLED))
2439 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2442 case MSR_KVM_PV_EOI_EN:
2443 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2447 case MSR_IA32_MCG_CTL:
2448 case MSR_IA32_MCG_STATUS:
2449 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2450 return set_msr_mce(vcpu, msr_info);
2452 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2453 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2454 pr = true; /* fall through */
2455 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2456 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2457 if (kvm_pmu_is_valid_msr(vcpu, msr))
2458 return kvm_pmu_set_msr(vcpu, msr_info);
2460 if (pr || data != 0)
2461 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2462 "0x%x data 0x%llx\n", msr, data);
2464 case MSR_K7_CLK_CTL:
2466 * Ignore all writes to this no longer documented MSR.
2467 * Writes are only relevant for old K7 processors,
2468 * all pre-dating SVM, but a recommended workaround from
2469 * AMD for these chips. It is possible to specify the
2470 * affected processor models on the command line, hence
2471 * the need to ignore the workaround.
2474 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2475 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2476 case HV_X64_MSR_CRASH_CTL:
2477 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2478 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2479 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2480 case HV_X64_MSR_TSC_EMULATION_STATUS:
2481 return kvm_hv_set_msr_common(vcpu, msr, data,
2482 msr_info->host_initiated);
2483 case MSR_IA32_BBL_CR_CTL3:
2484 /* Drop writes to this legacy MSR -- see rdmsr
2485 * counterpart for further detail.
2487 if (report_ignored_msrs)
2488 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2491 case MSR_AMD64_OSVW_ID_LENGTH:
2492 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2494 vcpu->arch.osvw.length = data;
2496 case MSR_AMD64_OSVW_STATUS:
2497 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2499 vcpu->arch.osvw.status = data;
2501 case MSR_PLATFORM_INFO:
2502 if (!msr_info->host_initiated ||
2503 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2504 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2505 cpuid_fault_enabled(vcpu)))
2507 vcpu->arch.msr_platform_info = data;
2509 case MSR_MISC_FEATURES_ENABLES:
2510 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2511 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2512 !supports_cpuid_fault(vcpu)))
2514 vcpu->arch.msr_misc_features_enables = data;
2517 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2518 return xen_hvm_config(vcpu, data);
2519 if (kvm_pmu_is_valid_msr(vcpu, msr))
2520 return kvm_pmu_set_msr(vcpu, msr_info);
2522 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2526 if (report_ignored_msrs)
2528 "ignored wrmsr: 0x%x data 0x%llx\n",
2535 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2539 * Reads an msr value (of 'msr_index') into 'pdata'.
2540 * Returns 0 on success, non-0 otherwise.
2541 * Assumes vcpu_load() was already called.
2543 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2545 return kvm_x86_ops->get_msr(vcpu, msr);
2547 EXPORT_SYMBOL_GPL(kvm_get_msr);
2549 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2552 u64 mcg_cap = vcpu->arch.mcg_cap;
2553 unsigned bank_num = mcg_cap & 0xff;
2556 case MSR_IA32_P5_MC_ADDR:
2557 case MSR_IA32_P5_MC_TYPE:
2560 case MSR_IA32_MCG_CAP:
2561 data = vcpu->arch.mcg_cap;
2563 case MSR_IA32_MCG_CTL:
2564 if (!(mcg_cap & MCG_CTL_P))
2566 data = vcpu->arch.mcg_ctl;
2568 case MSR_IA32_MCG_STATUS:
2569 data = vcpu->arch.mcg_status;
2572 if (msr >= MSR_IA32_MC0_CTL &&
2573 msr < MSR_IA32_MCx_CTL(bank_num)) {
2574 u32 offset = msr - MSR_IA32_MC0_CTL;
2575 data = vcpu->arch.mce_banks[offset];
2584 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2586 switch (msr_info->index) {
2587 case MSR_IA32_PLATFORM_ID:
2588 case MSR_IA32_EBL_CR_POWERON:
2589 case MSR_IA32_DEBUGCTLMSR:
2590 case MSR_IA32_LASTBRANCHFROMIP:
2591 case MSR_IA32_LASTBRANCHTOIP:
2592 case MSR_IA32_LASTINTFROMIP:
2593 case MSR_IA32_LASTINTTOIP:
2595 case MSR_K8_TSEG_ADDR:
2596 case MSR_K8_TSEG_MASK:
2598 case MSR_VM_HSAVE_PA:
2599 case MSR_K8_INT_PENDING_MSG:
2600 case MSR_AMD64_NB_CFG:
2601 case MSR_FAM10H_MMIO_CONF_BASE:
2602 case MSR_AMD64_BU_CFG2:
2603 case MSR_IA32_PERF_CTL:
2604 case MSR_AMD64_DC_CFG:
2607 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2608 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2609 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2610 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2611 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2612 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2613 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2616 case MSR_IA32_UCODE_REV:
2617 msr_info->data = vcpu->arch.microcode_version;
2620 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2623 case 0x200 ... 0x2ff:
2624 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2625 case 0xcd: /* fsb frequency */
2629 * MSR_EBC_FREQUENCY_ID
2630 * Conservative value valid for even the basic CPU models.
2631 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2632 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2633 * and 266MHz for model 3, or 4. Set Core Clock
2634 * Frequency to System Bus Frequency Ratio to 1 (bits
2635 * 31:24) even though these are only valid for CPU
2636 * models > 2, however guests may end up dividing or
2637 * multiplying by zero otherwise.
2639 case MSR_EBC_FREQUENCY_ID:
2640 msr_info->data = 1 << 24;
2642 case MSR_IA32_APICBASE:
2643 msr_info->data = kvm_get_apic_base(vcpu);
2645 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2646 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2648 case MSR_IA32_TSCDEADLINE:
2649 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2651 case MSR_IA32_TSC_ADJUST:
2652 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2654 case MSR_IA32_MISC_ENABLE:
2655 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2657 case MSR_IA32_SMBASE:
2658 if (!msr_info->host_initiated)
2660 msr_info->data = vcpu->arch.smbase;
2663 msr_info->data = vcpu->arch.smi_count;
2665 case MSR_IA32_PERF_STATUS:
2666 /* TSC increment by tick */
2667 msr_info->data = 1000ULL;
2668 /* CPU multiplier */
2669 msr_info->data |= (((uint64_t)4ULL) << 40);
2672 msr_info->data = vcpu->arch.efer;
2674 case MSR_KVM_WALL_CLOCK:
2675 case MSR_KVM_WALL_CLOCK_NEW:
2676 msr_info->data = vcpu->kvm->arch.wall_clock;
2678 case MSR_KVM_SYSTEM_TIME:
2679 case MSR_KVM_SYSTEM_TIME_NEW:
2680 msr_info->data = vcpu->arch.time;
2682 case MSR_KVM_ASYNC_PF_EN:
2683 msr_info->data = vcpu->arch.apf.msr_val;
2685 case MSR_KVM_STEAL_TIME:
2686 msr_info->data = vcpu->arch.st.msr_val;
2688 case MSR_KVM_PV_EOI_EN:
2689 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2691 case MSR_IA32_P5_MC_ADDR:
2692 case MSR_IA32_P5_MC_TYPE:
2693 case MSR_IA32_MCG_CAP:
2694 case MSR_IA32_MCG_CTL:
2695 case MSR_IA32_MCG_STATUS:
2696 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2697 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2698 case MSR_K7_CLK_CTL:
2700 * Provide expected ramp-up count for K7. All other
2701 * are set to zero, indicating minimum divisors for
2704 * This prevents guest kernels on AMD host with CPU
2705 * type 6, model 8 and higher from exploding due to
2706 * the rdmsr failing.
2708 msr_info->data = 0x20000000;
2710 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2711 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2712 case HV_X64_MSR_CRASH_CTL:
2713 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2714 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2715 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2716 case HV_X64_MSR_TSC_EMULATION_STATUS:
2717 return kvm_hv_get_msr_common(vcpu,
2718 msr_info->index, &msr_info->data);
2720 case MSR_IA32_BBL_CR_CTL3:
2721 /* This legacy MSR exists but isn't fully documented in current
2722 * silicon. It is however accessed by winxp in very narrow
2723 * scenarios where it sets bit #19, itself documented as
2724 * a "reserved" bit. Best effort attempt to source coherent
2725 * read data here should the balance of the register be
2726 * interpreted by the guest:
2728 * L2 cache control register 3: 64GB range, 256KB size,
2729 * enabled, latency 0x1, configured
2731 msr_info->data = 0xbe702111;
2733 case MSR_AMD64_OSVW_ID_LENGTH:
2734 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2736 msr_info->data = vcpu->arch.osvw.length;
2738 case MSR_AMD64_OSVW_STATUS:
2739 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2741 msr_info->data = vcpu->arch.osvw.status;
2743 case MSR_PLATFORM_INFO:
2744 msr_info->data = vcpu->arch.msr_platform_info;
2746 case MSR_MISC_FEATURES_ENABLES:
2747 msr_info->data = vcpu->arch.msr_misc_features_enables;
2750 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2751 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2753 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2757 if (report_ignored_msrs)
2758 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2766 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2769 * Read or write a bunch of msrs. All parameters are kernel addresses.
2771 * @return number of msrs set successfully.
2773 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2774 struct kvm_msr_entry *entries,
2775 int (*do_msr)(struct kvm_vcpu *vcpu,
2776 unsigned index, u64 *data))
2780 for (i = 0; i < msrs->nmsrs; ++i)
2781 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2788 * Read or write a bunch of msrs. Parameters are user addresses.
2790 * @return number of msrs set successfully.
2792 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2793 int (*do_msr)(struct kvm_vcpu *vcpu,
2794 unsigned index, u64 *data),
2797 struct kvm_msrs msrs;
2798 struct kvm_msr_entry *entries;
2803 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2807 if (msrs.nmsrs >= MAX_IO_MSRS)
2810 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2811 entries = memdup_user(user_msrs->entries, size);
2812 if (IS_ERR(entries)) {
2813 r = PTR_ERR(entries);
2817 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2822 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2833 static inline bool kvm_can_mwait_in_guest(void)
2835 return boot_cpu_has(X86_FEATURE_MWAIT) &&
2836 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
2837 boot_cpu_has(X86_FEATURE_ARAT);
2840 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2845 case KVM_CAP_IRQCHIP:
2847 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2848 case KVM_CAP_SET_TSS_ADDR:
2849 case KVM_CAP_EXT_CPUID:
2850 case KVM_CAP_EXT_EMUL_CPUID:
2851 case KVM_CAP_CLOCKSOURCE:
2853 case KVM_CAP_NOP_IO_DELAY:
2854 case KVM_CAP_MP_STATE:
2855 case KVM_CAP_SYNC_MMU:
2856 case KVM_CAP_USER_NMI:
2857 case KVM_CAP_REINJECT_CONTROL:
2858 case KVM_CAP_IRQ_INJECT_STATUS:
2859 case KVM_CAP_IOEVENTFD:
2860 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2862 case KVM_CAP_PIT_STATE2:
2863 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2864 case KVM_CAP_XEN_HVM:
2865 case KVM_CAP_VCPU_EVENTS:
2866 case KVM_CAP_HYPERV:
2867 case KVM_CAP_HYPERV_VAPIC:
2868 case KVM_CAP_HYPERV_SPIN:
2869 case KVM_CAP_HYPERV_SYNIC:
2870 case KVM_CAP_HYPERV_SYNIC2:
2871 case KVM_CAP_HYPERV_VP_INDEX:
2872 case KVM_CAP_HYPERV_EVENTFD:
2873 case KVM_CAP_PCI_SEGMENT:
2874 case KVM_CAP_DEBUGREGS:
2875 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2877 case KVM_CAP_ASYNC_PF:
2878 case KVM_CAP_GET_TSC_KHZ:
2879 case KVM_CAP_KVMCLOCK_CTRL:
2880 case KVM_CAP_READONLY_MEM:
2881 case KVM_CAP_HYPERV_TIME:
2882 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2883 case KVM_CAP_TSC_DEADLINE_TIMER:
2884 case KVM_CAP_ENABLE_CAP_VM:
2885 case KVM_CAP_DISABLE_QUIRKS:
2886 case KVM_CAP_SET_BOOT_CPU_ID:
2887 case KVM_CAP_SPLIT_IRQCHIP:
2888 case KVM_CAP_IMMEDIATE_EXIT:
2889 case KVM_CAP_GET_MSR_FEATURES:
2892 case KVM_CAP_SYNC_REGS:
2893 r = KVM_SYNC_X86_VALID_FIELDS;
2895 case KVM_CAP_ADJUST_CLOCK:
2896 r = KVM_CLOCK_TSC_STABLE;
2898 case KVM_CAP_X86_DISABLE_EXITS:
2899 r |= KVM_X86_DISABLE_EXITS_HTL | KVM_X86_DISABLE_EXITS_PAUSE;
2900 if(kvm_can_mwait_in_guest())
2901 r |= KVM_X86_DISABLE_EXITS_MWAIT;
2903 case KVM_CAP_X86_SMM:
2904 /* SMBASE is usually relocated above 1M on modern chipsets,
2905 * and SMM handlers might indeed rely on 4G segment limits,
2906 * so do not report SMM to be available if real mode is
2907 * emulated via vm86 mode. Still, do not go to great lengths
2908 * to avoid userspace's usage of the feature, because it is a
2909 * fringe case that is not enabled except via specific settings
2910 * of the module parameters.
2912 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
2915 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2917 case KVM_CAP_NR_VCPUS:
2918 r = KVM_SOFT_MAX_VCPUS;
2920 case KVM_CAP_MAX_VCPUS:
2923 case KVM_CAP_NR_MEMSLOTS:
2924 r = KVM_USER_MEM_SLOTS;
2926 case KVM_CAP_PV_MMU: /* obsolete */
2930 r = KVM_MAX_MCE_BANKS;
2933 r = boot_cpu_has(X86_FEATURE_XSAVE);
2935 case KVM_CAP_TSC_CONTROL:
2936 r = kvm_has_tsc_control;
2938 case KVM_CAP_X2APIC_API:
2939 r = KVM_X2APIC_API_VALID_FLAGS;
2948 long kvm_arch_dev_ioctl(struct file *filp,
2949 unsigned int ioctl, unsigned long arg)
2951 void __user *argp = (void __user *)arg;
2955 case KVM_GET_MSR_INDEX_LIST: {
2956 struct kvm_msr_list __user *user_msr_list = argp;
2957 struct kvm_msr_list msr_list;
2961 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2964 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2965 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2968 if (n < msr_list.nmsrs)
2971 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2972 num_msrs_to_save * sizeof(u32)))
2974 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2976 num_emulated_msrs * sizeof(u32)))
2981 case KVM_GET_SUPPORTED_CPUID:
2982 case KVM_GET_EMULATED_CPUID: {
2983 struct kvm_cpuid2 __user *cpuid_arg = argp;
2984 struct kvm_cpuid2 cpuid;
2987 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2990 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2996 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3001 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3003 if (copy_to_user(argp, &kvm_mce_cap_supported,
3004 sizeof(kvm_mce_cap_supported)))
3008 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3009 struct kvm_msr_list __user *user_msr_list = argp;
3010 struct kvm_msr_list msr_list;
3014 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3017 msr_list.nmsrs = num_msr_based_features;
3018 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3021 if (n < msr_list.nmsrs)
3024 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3025 num_msr_based_features * sizeof(u32)))
3031 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3041 static void wbinvd_ipi(void *garbage)
3046 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3048 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3051 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3053 /* Address WBINVD may be executed by guest */
3054 if (need_emulate_wbinvd(vcpu)) {
3055 if (kvm_x86_ops->has_wbinvd_exit())
3056 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3057 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3058 smp_call_function_single(vcpu->cpu,
3059 wbinvd_ipi, NULL, 1);
3062 kvm_x86_ops->vcpu_load(vcpu, cpu);
3064 /* Apply any externally detected TSC adjustments (due to suspend) */
3065 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3066 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3067 vcpu->arch.tsc_offset_adjustment = 0;
3068 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3071 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3072 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3073 rdtsc() - vcpu->arch.last_host_tsc;
3075 mark_tsc_unstable("KVM discovered backwards TSC");
3077 if (kvm_check_tsc_unstable()) {
3078 u64 offset = kvm_compute_tsc_offset(vcpu,
3079 vcpu->arch.last_guest_tsc);
3080 kvm_vcpu_write_tsc_offset(vcpu, offset);
3081 vcpu->arch.tsc_catchup = 1;
3084 if (kvm_lapic_hv_timer_in_use(vcpu))
3085 kvm_lapic_restart_hv_timer(vcpu);
3088 * On a host with synchronized TSC, there is no need to update
3089 * kvmclock on vcpu->cpu migration
3091 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3092 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3093 if (vcpu->cpu != cpu)
3094 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3098 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3101 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3103 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3106 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3108 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3109 &vcpu->arch.st.steal.preempted,
3110 offsetof(struct kvm_steal_time, preempted),
3111 sizeof(vcpu->arch.st.steal.preempted));
3114 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3118 if (vcpu->preempted)
3119 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3122 * Disable page faults because we're in atomic context here.
3123 * kvm_write_guest_offset_cached() would call might_fault()
3124 * that relies on pagefault_disable() to tell if there's a
3125 * bug. NOTE: the write to guest memory may not go through if
3126 * during postcopy live migration or if there's heavy guest
3129 pagefault_disable();
3131 * kvm_memslots() will be called by
3132 * kvm_write_guest_offset_cached() so take the srcu lock.
3134 idx = srcu_read_lock(&vcpu->kvm->srcu);
3135 kvm_steal_time_set_preempted(vcpu);
3136 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3138 kvm_x86_ops->vcpu_put(vcpu);
3139 vcpu->arch.last_host_tsc = rdtsc();
3141 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3142 * on every vmexit, but if not, we might have a stale dr6 from the
3143 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3148 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3149 struct kvm_lapic_state *s)
3151 if (vcpu->arch.apicv_active)
3152 kvm_x86_ops->sync_pir_to_irr(vcpu);
3154 return kvm_apic_get_state(vcpu, s);
3157 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3158 struct kvm_lapic_state *s)
3162 r = kvm_apic_set_state(vcpu, s);
3165 update_cr8_intercept(vcpu);
3170 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3172 return (!lapic_in_kernel(vcpu) ||
3173 kvm_apic_accept_pic_intr(vcpu));
3177 * if userspace requested an interrupt window, check that the
3178 * interrupt window is open.
3180 * No need to exit to userspace if we already have an interrupt queued.
3182 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3184 return kvm_arch_interrupt_allowed(vcpu) &&
3185 !kvm_cpu_has_interrupt(vcpu) &&
3186 !kvm_event_needs_reinjection(vcpu) &&
3187 kvm_cpu_accept_dm_intr(vcpu);
3190 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3191 struct kvm_interrupt *irq)
3193 if (irq->irq >= KVM_NR_INTERRUPTS)
3196 if (!irqchip_in_kernel(vcpu->kvm)) {
3197 kvm_queue_interrupt(vcpu, irq->irq, false);
3198 kvm_make_request(KVM_REQ_EVENT, vcpu);
3203 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3204 * fail for in-kernel 8259.
3206 if (pic_in_kernel(vcpu->kvm))
3209 if (vcpu->arch.pending_external_vector != -1)
3212 vcpu->arch.pending_external_vector = irq->irq;
3213 kvm_make_request(KVM_REQ_EVENT, vcpu);
3217 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3219 kvm_inject_nmi(vcpu);
3224 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3226 kvm_make_request(KVM_REQ_SMI, vcpu);
3231 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3232 struct kvm_tpr_access_ctl *tac)
3236 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3240 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3244 unsigned bank_num = mcg_cap & 0xff, bank;
3247 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3249 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3252 vcpu->arch.mcg_cap = mcg_cap;
3253 /* Init IA32_MCG_CTL to all 1s */
3254 if (mcg_cap & MCG_CTL_P)
3255 vcpu->arch.mcg_ctl = ~(u64)0;
3256 /* Init IA32_MCi_CTL to all 1s */
3257 for (bank = 0; bank < bank_num; bank++)
3258 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3260 if (kvm_x86_ops->setup_mce)
3261 kvm_x86_ops->setup_mce(vcpu);
3266 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3267 struct kvm_x86_mce *mce)
3269 u64 mcg_cap = vcpu->arch.mcg_cap;
3270 unsigned bank_num = mcg_cap & 0xff;
3271 u64 *banks = vcpu->arch.mce_banks;
3273 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3276 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3277 * reporting is disabled
3279 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3280 vcpu->arch.mcg_ctl != ~(u64)0)
3282 banks += 4 * mce->bank;
3284 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3285 * reporting is disabled for the bank
3287 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3289 if (mce->status & MCI_STATUS_UC) {
3290 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3291 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3292 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3295 if (banks[1] & MCI_STATUS_VAL)
3296 mce->status |= MCI_STATUS_OVER;
3297 banks[2] = mce->addr;
3298 banks[3] = mce->misc;
3299 vcpu->arch.mcg_status = mce->mcg_status;
3300 banks[1] = mce->status;
3301 kvm_queue_exception(vcpu, MC_VECTOR);
3302 } else if (!(banks[1] & MCI_STATUS_VAL)
3303 || !(banks[1] & MCI_STATUS_UC)) {
3304 if (banks[1] & MCI_STATUS_VAL)
3305 mce->status |= MCI_STATUS_OVER;
3306 banks[2] = mce->addr;
3307 banks[3] = mce->misc;
3308 banks[1] = mce->status;
3310 banks[1] |= MCI_STATUS_OVER;
3314 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3315 struct kvm_vcpu_events *events)
3319 * FIXME: pass injected and pending separately. This is only
3320 * needed for nested virtualization, whose state cannot be
3321 * migrated yet. For now we can combine them.
3323 events->exception.injected =
3324 (vcpu->arch.exception.pending ||
3325 vcpu->arch.exception.injected) &&
3326 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3327 events->exception.nr = vcpu->arch.exception.nr;
3328 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3329 events->exception.pad = 0;
3330 events->exception.error_code = vcpu->arch.exception.error_code;
3332 events->interrupt.injected =
3333 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3334 events->interrupt.nr = vcpu->arch.interrupt.nr;
3335 events->interrupt.soft = 0;
3336 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3338 events->nmi.injected = vcpu->arch.nmi_injected;
3339 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3340 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3341 events->nmi.pad = 0;
3343 events->sipi_vector = 0; /* never valid when reporting to user space */
3345 events->smi.smm = is_smm(vcpu);
3346 events->smi.pending = vcpu->arch.smi_pending;
3347 events->smi.smm_inside_nmi =
3348 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3349 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3351 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3352 | KVM_VCPUEVENT_VALID_SHADOW
3353 | KVM_VCPUEVENT_VALID_SMM);
3354 memset(&events->reserved, 0, sizeof(events->reserved));
3357 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3359 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3360 struct kvm_vcpu_events *events)
3362 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3363 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3364 | KVM_VCPUEVENT_VALID_SHADOW
3365 | KVM_VCPUEVENT_VALID_SMM))
3368 if (events->exception.injected &&
3369 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3370 is_guest_mode(vcpu)))
3373 /* INITs are latched while in SMM */
3374 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3375 (events->smi.smm || events->smi.pending) &&
3376 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3380 vcpu->arch.exception.injected = false;
3381 vcpu->arch.exception.pending = events->exception.injected;
3382 vcpu->arch.exception.nr = events->exception.nr;
3383 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3384 vcpu->arch.exception.error_code = events->exception.error_code;
3386 vcpu->arch.interrupt.injected = events->interrupt.injected;
3387 vcpu->arch.interrupt.nr = events->interrupt.nr;
3388 vcpu->arch.interrupt.soft = events->interrupt.soft;
3389 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3390 kvm_x86_ops->set_interrupt_shadow(vcpu,
3391 events->interrupt.shadow);
3393 vcpu->arch.nmi_injected = events->nmi.injected;
3394 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3395 vcpu->arch.nmi_pending = events->nmi.pending;
3396 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3398 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3399 lapic_in_kernel(vcpu))
3400 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3402 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3403 u32 hflags = vcpu->arch.hflags;
3404 if (events->smi.smm)
3405 hflags |= HF_SMM_MASK;
3407 hflags &= ~HF_SMM_MASK;
3408 kvm_set_hflags(vcpu, hflags);
3410 vcpu->arch.smi_pending = events->smi.pending;
3412 if (events->smi.smm) {
3413 if (events->smi.smm_inside_nmi)
3414 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3416 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3417 if (lapic_in_kernel(vcpu)) {
3418 if (events->smi.latched_init)
3419 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3421 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3426 kvm_make_request(KVM_REQ_EVENT, vcpu);
3431 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3432 struct kvm_debugregs *dbgregs)
3436 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3437 kvm_get_dr(vcpu, 6, &val);
3439 dbgregs->dr7 = vcpu->arch.dr7;
3441 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3444 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3445 struct kvm_debugregs *dbgregs)
3450 if (dbgregs->dr6 & ~0xffffffffull)
3452 if (dbgregs->dr7 & ~0xffffffffull)
3455 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3456 kvm_update_dr0123(vcpu);
3457 vcpu->arch.dr6 = dbgregs->dr6;
3458 kvm_update_dr6(vcpu);
3459 vcpu->arch.dr7 = dbgregs->dr7;
3460 kvm_update_dr7(vcpu);
3465 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3467 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3469 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3470 u64 xstate_bv = xsave->header.xfeatures;
3474 * Copy legacy XSAVE area, to avoid complications with CPUID
3475 * leaves 0 and 1 in the loop below.
3477 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3480 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3481 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3484 * Copy each region from the possibly compacted offset to the
3485 * non-compacted offset.
3487 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3489 u64 feature = valid & -valid;
3490 int index = fls64(feature) - 1;
3491 void *src = get_xsave_addr(xsave, feature);
3494 u32 size, offset, ecx, edx;
3495 cpuid_count(XSTATE_CPUID, index,
3496 &size, &offset, &ecx, &edx);
3497 if (feature == XFEATURE_MASK_PKRU)
3498 memcpy(dest + offset, &vcpu->arch.pkru,
3499 sizeof(vcpu->arch.pkru));
3501 memcpy(dest + offset, src, size);
3509 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3511 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3512 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3516 * Copy legacy XSAVE area, to avoid complications with CPUID
3517 * leaves 0 and 1 in the loop below.
3519 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3521 /* Set XSTATE_BV and possibly XCOMP_BV. */
3522 xsave->header.xfeatures = xstate_bv;
3523 if (boot_cpu_has(X86_FEATURE_XSAVES))
3524 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3527 * Copy each region from the non-compacted offset to the
3528 * possibly compacted offset.
3530 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3532 u64 feature = valid & -valid;
3533 int index = fls64(feature) - 1;
3534 void *dest = get_xsave_addr(xsave, feature);
3537 u32 size, offset, ecx, edx;
3538 cpuid_count(XSTATE_CPUID, index,
3539 &size, &offset, &ecx, &edx);
3540 if (feature == XFEATURE_MASK_PKRU)
3541 memcpy(&vcpu->arch.pkru, src + offset,
3542 sizeof(vcpu->arch.pkru));
3544 memcpy(dest, src + offset, size);
3551 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3552 struct kvm_xsave *guest_xsave)
3554 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3555 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3556 fill_xsave((u8 *) guest_xsave->region, vcpu);
3558 memcpy(guest_xsave->region,
3559 &vcpu->arch.guest_fpu.state.fxsave,
3560 sizeof(struct fxregs_state));
3561 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3562 XFEATURE_MASK_FPSSE;
3566 #define XSAVE_MXCSR_OFFSET 24
3568 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3569 struct kvm_xsave *guest_xsave)
3572 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3573 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3575 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3577 * Here we allow setting states that are not present in
3578 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3579 * with old userspace.
3581 if (xstate_bv & ~kvm_supported_xcr0() ||
3582 mxcsr & ~mxcsr_feature_mask)
3584 load_xsave(vcpu, (u8 *)guest_xsave->region);
3586 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3587 mxcsr & ~mxcsr_feature_mask)
3589 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3590 guest_xsave->region, sizeof(struct fxregs_state));
3595 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3596 struct kvm_xcrs *guest_xcrs)
3598 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3599 guest_xcrs->nr_xcrs = 0;
3603 guest_xcrs->nr_xcrs = 1;
3604 guest_xcrs->flags = 0;
3605 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3606 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3609 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3610 struct kvm_xcrs *guest_xcrs)
3614 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3617 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3620 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3621 /* Only support XCR0 currently */
3622 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3623 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3624 guest_xcrs->xcrs[i].value);
3633 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3634 * stopped by the hypervisor. This function will be called from the host only.
3635 * EINVAL is returned when the host attempts to set the flag for a guest that
3636 * does not support pv clocks.
3638 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3640 if (!vcpu->arch.pv_time_enabled)
3642 vcpu->arch.pvclock_set_guest_stopped_request = true;
3643 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3647 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3648 struct kvm_enable_cap *cap)
3654 case KVM_CAP_HYPERV_SYNIC2:
3657 case KVM_CAP_HYPERV_SYNIC:
3658 if (!irqchip_in_kernel(vcpu->kvm))
3660 return kvm_hv_activate_synic(vcpu, cap->cap ==
3661 KVM_CAP_HYPERV_SYNIC2);
3667 long kvm_arch_vcpu_ioctl(struct file *filp,
3668 unsigned int ioctl, unsigned long arg)
3670 struct kvm_vcpu *vcpu = filp->private_data;
3671 void __user *argp = (void __user *)arg;
3674 struct kvm_lapic_state *lapic;
3675 struct kvm_xsave *xsave;
3676 struct kvm_xcrs *xcrs;
3684 case KVM_GET_LAPIC: {
3686 if (!lapic_in_kernel(vcpu))
3688 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3693 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3697 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3702 case KVM_SET_LAPIC: {
3704 if (!lapic_in_kernel(vcpu))
3706 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3707 if (IS_ERR(u.lapic)) {
3708 r = PTR_ERR(u.lapic);
3712 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3715 case KVM_INTERRUPT: {
3716 struct kvm_interrupt irq;
3719 if (copy_from_user(&irq, argp, sizeof irq))
3721 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3725 r = kvm_vcpu_ioctl_nmi(vcpu);
3729 r = kvm_vcpu_ioctl_smi(vcpu);
3732 case KVM_SET_CPUID: {
3733 struct kvm_cpuid __user *cpuid_arg = argp;
3734 struct kvm_cpuid cpuid;
3737 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3739 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3742 case KVM_SET_CPUID2: {
3743 struct kvm_cpuid2 __user *cpuid_arg = argp;
3744 struct kvm_cpuid2 cpuid;
3747 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3749 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3750 cpuid_arg->entries);
3753 case KVM_GET_CPUID2: {
3754 struct kvm_cpuid2 __user *cpuid_arg = argp;
3755 struct kvm_cpuid2 cpuid;
3758 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3760 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3761 cpuid_arg->entries);
3765 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3770 case KVM_GET_MSRS: {
3771 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3772 r = msr_io(vcpu, argp, do_get_msr, 1);
3773 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3776 case KVM_SET_MSRS: {
3777 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3778 r = msr_io(vcpu, argp, do_set_msr, 0);
3779 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3782 case KVM_TPR_ACCESS_REPORTING: {
3783 struct kvm_tpr_access_ctl tac;
3786 if (copy_from_user(&tac, argp, sizeof tac))
3788 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3792 if (copy_to_user(argp, &tac, sizeof tac))
3797 case KVM_SET_VAPIC_ADDR: {
3798 struct kvm_vapic_addr va;
3802 if (!lapic_in_kernel(vcpu))
3805 if (copy_from_user(&va, argp, sizeof va))
3807 idx = srcu_read_lock(&vcpu->kvm->srcu);
3808 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3809 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3812 case KVM_X86_SETUP_MCE: {
3816 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3818 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3821 case KVM_X86_SET_MCE: {
3822 struct kvm_x86_mce mce;
3825 if (copy_from_user(&mce, argp, sizeof mce))
3827 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3830 case KVM_GET_VCPU_EVENTS: {
3831 struct kvm_vcpu_events events;
3833 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3836 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3841 case KVM_SET_VCPU_EVENTS: {
3842 struct kvm_vcpu_events events;
3845 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3848 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3851 case KVM_GET_DEBUGREGS: {
3852 struct kvm_debugregs dbgregs;
3854 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3857 if (copy_to_user(argp, &dbgregs,
3858 sizeof(struct kvm_debugregs)))
3863 case KVM_SET_DEBUGREGS: {
3864 struct kvm_debugregs dbgregs;
3867 if (copy_from_user(&dbgregs, argp,
3868 sizeof(struct kvm_debugregs)))
3871 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3874 case KVM_GET_XSAVE: {
3875 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3880 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3883 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3888 case KVM_SET_XSAVE: {
3889 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3890 if (IS_ERR(u.xsave)) {
3891 r = PTR_ERR(u.xsave);
3895 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3898 case KVM_GET_XCRS: {
3899 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3904 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3907 if (copy_to_user(argp, u.xcrs,
3908 sizeof(struct kvm_xcrs)))
3913 case KVM_SET_XCRS: {
3914 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3915 if (IS_ERR(u.xcrs)) {
3916 r = PTR_ERR(u.xcrs);
3920 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3923 case KVM_SET_TSC_KHZ: {
3927 user_tsc_khz = (u32)arg;
3929 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3932 if (user_tsc_khz == 0)
3933 user_tsc_khz = tsc_khz;
3935 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3940 case KVM_GET_TSC_KHZ: {
3941 r = vcpu->arch.virtual_tsc_khz;
3944 case KVM_KVMCLOCK_CTRL: {
3945 r = kvm_set_guest_paused(vcpu);
3948 case KVM_ENABLE_CAP: {
3949 struct kvm_enable_cap cap;
3952 if (copy_from_user(&cap, argp, sizeof(cap)))
3954 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3967 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3969 return VM_FAULT_SIGBUS;
3972 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3976 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3978 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3982 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3985 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
3988 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3989 u32 kvm_nr_mmu_pages)
3991 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3994 mutex_lock(&kvm->slots_lock);
3996 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3997 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3999 mutex_unlock(&kvm->slots_lock);
4003 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4005 return kvm->arch.n_max_mmu_pages;
4008 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4010 struct kvm_pic *pic = kvm->arch.vpic;
4014 switch (chip->chip_id) {
4015 case KVM_IRQCHIP_PIC_MASTER:
4016 memcpy(&chip->chip.pic, &pic->pics[0],
4017 sizeof(struct kvm_pic_state));
4019 case KVM_IRQCHIP_PIC_SLAVE:
4020 memcpy(&chip->chip.pic, &pic->pics[1],
4021 sizeof(struct kvm_pic_state));
4023 case KVM_IRQCHIP_IOAPIC:
4024 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4033 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4035 struct kvm_pic *pic = kvm->arch.vpic;
4039 switch (chip->chip_id) {
4040 case KVM_IRQCHIP_PIC_MASTER:
4041 spin_lock(&pic->lock);
4042 memcpy(&pic->pics[0], &chip->chip.pic,
4043 sizeof(struct kvm_pic_state));
4044 spin_unlock(&pic->lock);
4046 case KVM_IRQCHIP_PIC_SLAVE:
4047 spin_lock(&pic->lock);
4048 memcpy(&pic->pics[1], &chip->chip.pic,
4049 sizeof(struct kvm_pic_state));
4050 spin_unlock(&pic->lock);
4052 case KVM_IRQCHIP_IOAPIC:
4053 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4059 kvm_pic_update_irq(pic);
4063 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4065 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4067 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4069 mutex_lock(&kps->lock);
4070 memcpy(ps, &kps->channels, sizeof(*ps));
4071 mutex_unlock(&kps->lock);
4075 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4078 struct kvm_pit *pit = kvm->arch.vpit;
4080 mutex_lock(&pit->pit_state.lock);
4081 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4082 for (i = 0; i < 3; i++)
4083 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4084 mutex_unlock(&pit->pit_state.lock);
4088 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4090 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4091 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4092 sizeof(ps->channels));
4093 ps->flags = kvm->arch.vpit->pit_state.flags;
4094 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4095 memset(&ps->reserved, 0, sizeof(ps->reserved));
4099 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4103 u32 prev_legacy, cur_legacy;
4104 struct kvm_pit *pit = kvm->arch.vpit;
4106 mutex_lock(&pit->pit_state.lock);
4107 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4108 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4109 if (!prev_legacy && cur_legacy)
4111 memcpy(&pit->pit_state.channels, &ps->channels,
4112 sizeof(pit->pit_state.channels));
4113 pit->pit_state.flags = ps->flags;
4114 for (i = 0; i < 3; i++)
4115 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4117 mutex_unlock(&pit->pit_state.lock);
4121 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4122 struct kvm_reinject_control *control)
4124 struct kvm_pit *pit = kvm->arch.vpit;
4129 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4130 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4131 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4133 mutex_lock(&pit->pit_state.lock);
4134 kvm_pit_set_reinject(pit, control->pit_reinject);
4135 mutex_unlock(&pit->pit_state.lock);
4141 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4142 * @kvm: kvm instance
4143 * @log: slot id and address to which we copy the log
4145 * Steps 1-4 below provide general overview of dirty page logging. See
4146 * kvm_get_dirty_log_protect() function description for additional details.
4148 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4149 * always flush the TLB (step 4) even if previous step failed and the dirty
4150 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4151 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4152 * writes will be marked dirty for next log read.
4154 * 1. Take a snapshot of the bit and clear it if needed.
4155 * 2. Write protect the corresponding page.
4156 * 3. Copy the snapshot to the userspace.
4157 * 4. Flush TLB's if needed.
4159 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4161 bool is_dirty = false;
4164 mutex_lock(&kvm->slots_lock);
4167 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4169 if (kvm_x86_ops->flush_log_dirty)
4170 kvm_x86_ops->flush_log_dirty(kvm);
4172 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
4175 * All the TLBs can be flushed out of mmu lock, see the comments in
4176 * kvm_mmu_slot_remove_write_access().
4178 lockdep_assert_held(&kvm->slots_lock);
4180 kvm_flush_remote_tlbs(kvm);
4182 mutex_unlock(&kvm->slots_lock);
4186 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4189 if (!irqchip_in_kernel(kvm))
4192 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4193 irq_event->irq, irq_event->level,
4198 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4199 struct kvm_enable_cap *cap)
4207 case KVM_CAP_DISABLE_QUIRKS:
4208 kvm->arch.disabled_quirks = cap->args[0];
4211 case KVM_CAP_SPLIT_IRQCHIP: {
4212 mutex_lock(&kvm->lock);
4214 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4215 goto split_irqchip_unlock;
4217 if (irqchip_in_kernel(kvm))
4218 goto split_irqchip_unlock;
4219 if (kvm->created_vcpus)
4220 goto split_irqchip_unlock;
4221 r = kvm_setup_empty_irq_routing(kvm);
4223 goto split_irqchip_unlock;
4224 /* Pairs with irqchip_in_kernel. */
4226 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4227 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4229 split_irqchip_unlock:
4230 mutex_unlock(&kvm->lock);
4233 case KVM_CAP_X2APIC_API:
4235 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4238 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4239 kvm->arch.x2apic_format = true;
4240 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4241 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4245 case KVM_CAP_X86_DISABLE_EXITS:
4247 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4250 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4251 kvm_can_mwait_in_guest())
4252 kvm->arch.mwait_in_guest = true;
4253 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HTL)
4254 kvm->arch.hlt_in_guest = true;
4255 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4256 kvm->arch.pause_in_guest = true;
4266 long kvm_arch_vm_ioctl(struct file *filp,
4267 unsigned int ioctl, unsigned long arg)
4269 struct kvm *kvm = filp->private_data;
4270 void __user *argp = (void __user *)arg;
4273 * This union makes it completely explicit to gcc-3.x
4274 * that these two variables' stack usage should be
4275 * combined, not added together.
4278 struct kvm_pit_state ps;
4279 struct kvm_pit_state2 ps2;
4280 struct kvm_pit_config pit_config;
4284 case KVM_SET_TSS_ADDR:
4285 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4287 case KVM_SET_IDENTITY_MAP_ADDR: {
4290 mutex_lock(&kvm->lock);
4292 if (kvm->created_vcpus)
4293 goto set_identity_unlock;
4295 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4296 goto set_identity_unlock;
4297 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4298 set_identity_unlock:
4299 mutex_unlock(&kvm->lock);
4302 case KVM_SET_NR_MMU_PAGES:
4303 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4305 case KVM_GET_NR_MMU_PAGES:
4306 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4308 case KVM_CREATE_IRQCHIP: {
4309 mutex_lock(&kvm->lock);
4312 if (irqchip_in_kernel(kvm))
4313 goto create_irqchip_unlock;
4316 if (kvm->created_vcpus)
4317 goto create_irqchip_unlock;
4319 r = kvm_pic_init(kvm);
4321 goto create_irqchip_unlock;
4323 r = kvm_ioapic_init(kvm);
4325 kvm_pic_destroy(kvm);
4326 goto create_irqchip_unlock;
4329 r = kvm_setup_default_irq_routing(kvm);
4331 kvm_ioapic_destroy(kvm);
4332 kvm_pic_destroy(kvm);
4333 goto create_irqchip_unlock;
4335 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4337 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4338 create_irqchip_unlock:
4339 mutex_unlock(&kvm->lock);
4342 case KVM_CREATE_PIT:
4343 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4345 case KVM_CREATE_PIT2:
4347 if (copy_from_user(&u.pit_config, argp,
4348 sizeof(struct kvm_pit_config)))
4351 mutex_lock(&kvm->lock);
4354 goto create_pit_unlock;
4356 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4360 mutex_unlock(&kvm->lock);
4362 case KVM_GET_IRQCHIP: {
4363 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4364 struct kvm_irqchip *chip;
4366 chip = memdup_user(argp, sizeof(*chip));
4373 if (!irqchip_kernel(kvm))
4374 goto get_irqchip_out;
4375 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4377 goto get_irqchip_out;
4379 if (copy_to_user(argp, chip, sizeof *chip))
4380 goto get_irqchip_out;
4386 case KVM_SET_IRQCHIP: {
4387 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4388 struct kvm_irqchip *chip;
4390 chip = memdup_user(argp, sizeof(*chip));
4397 if (!irqchip_kernel(kvm))
4398 goto set_irqchip_out;
4399 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4401 goto set_irqchip_out;
4409 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4412 if (!kvm->arch.vpit)
4414 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4418 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4425 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4428 if (!kvm->arch.vpit)
4430 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4433 case KVM_GET_PIT2: {
4435 if (!kvm->arch.vpit)
4437 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4441 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4446 case KVM_SET_PIT2: {
4448 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4451 if (!kvm->arch.vpit)
4453 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4456 case KVM_REINJECT_CONTROL: {
4457 struct kvm_reinject_control control;
4459 if (copy_from_user(&control, argp, sizeof(control)))
4461 r = kvm_vm_ioctl_reinject(kvm, &control);
4464 case KVM_SET_BOOT_CPU_ID:
4466 mutex_lock(&kvm->lock);
4467 if (kvm->created_vcpus)
4470 kvm->arch.bsp_vcpu_id = arg;
4471 mutex_unlock(&kvm->lock);
4473 case KVM_XEN_HVM_CONFIG: {
4474 struct kvm_xen_hvm_config xhc;
4476 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4481 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4485 case KVM_SET_CLOCK: {
4486 struct kvm_clock_data user_ns;
4490 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4499 * TODO: userspace has to take care of races with VCPU_RUN, so
4500 * kvm_gen_update_masterclock() can be cut down to locked
4501 * pvclock_update_vm_gtod_copy().
4503 kvm_gen_update_masterclock(kvm);
4504 now_ns = get_kvmclock_ns(kvm);
4505 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4506 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4509 case KVM_GET_CLOCK: {
4510 struct kvm_clock_data user_ns;
4513 now_ns = get_kvmclock_ns(kvm);
4514 user_ns.clock = now_ns;
4515 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4516 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4519 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4524 case KVM_ENABLE_CAP: {
4525 struct kvm_enable_cap cap;
4528 if (copy_from_user(&cap, argp, sizeof(cap)))
4530 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4533 case KVM_MEMORY_ENCRYPT_OP: {
4535 if (kvm_x86_ops->mem_enc_op)
4536 r = kvm_x86_ops->mem_enc_op(kvm, argp);
4539 case KVM_MEMORY_ENCRYPT_REG_REGION: {
4540 struct kvm_enc_region region;
4543 if (copy_from_user(®ion, argp, sizeof(region)))
4547 if (kvm_x86_ops->mem_enc_reg_region)
4548 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
4551 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4552 struct kvm_enc_region region;
4555 if (copy_from_user(®ion, argp, sizeof(region)))
4559 if (kvm_x86_ops->mem_enc_unreg_region)
4560 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
4563 case KVM_HYPERV_EVENTFD: {
4564 struct kvm_hyperv_eventfd hvevfd;
4567 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4569 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4579 static void kvm_init_msr_list(void)
4584 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4585 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4589 * Even MSRs that are valid in the host may not be exposed
4590 * to the guests in some cases.
4592 switch (msrs_to_save[i]) {
4593 case MSR_IA32_BNDCFGS:
4594 if (!kvm_x86_ops->mpx_supported())
4598 if (!kvm_x86_ops->rdtscp_supported())
4606 msrs_to_save[j] = msrs_to_save[i];
4609 num_msrs_to_save = j;
4611 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4612 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4616 emulated_msrs[j] = emulated_msrs[i];
4619 num_emulated_msrs = j;
4621 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4622 struct kvm_msr_entry msr;
4624 msr.index = msr_based_features[i];
4625 if (kvm_get_msr_feature(&msr))
4629 msr_based_features[j] = msr_based_features[i];
4632 num_msr_based_features = j;
4635 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4643 if (!(lapic_in_kernel(vcpu) &&
4644 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4645 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4656 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4663 if (!(lapic_in_kernel(vcpu) &&
4664 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4666 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4668 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4678 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4679 struct kvm_segment *var, int seg)
4681 kvm_x86_ops->set_segment(vcpu, var, seg);
4684 void kvm_get_segment(struct kvm_vcpu *vcpu,
4685 struct kvm_segment *var, int seg)
4687 kvm_x86_ops->get_segment(vcpu, var, seg);
4690 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4691 struct x86_exception *exception)
4695 BUG_ON(!mmu_is_nested(vcpu));
4697 /* NPT walks are always user-walks */
4698 access |= PFERR_USER_MASK;
4699 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4704 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4705 struct x86_exception *exception)
4707 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4708 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4711 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4712 struct x86_exception *exception)
4714 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4715 access |= PFERR_FETCH_MASK;
4716 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4719 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4720 struct x86_exception *exception)
4722 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4723 access |= PFERR_WRITE_MASK;
4724 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4727 /* uses this to access any guest's mapped memory without checking CPL */
4728 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4729 struct x86_exception *exception)
4731 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4734 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4735 struct kvm_vcpu *vcpu, u32 access,
4736 struct x86_exception *exception)
4739 int r = X86EMUL_CONTINUE;
4742 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4744 unsigned offset = addr & (PAGE_SIZE-1);
4745 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4748 if (gpa == UNMAPPED_GVA)
4749 return X86EMUL_PROPAGATE_FAULT;
4750 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4753 r = X86EMUL_IO_NEEDED;
4765 /* used for instruction fetching */
4766 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4767 gva_t addr, void *val, unsigned int bytes,
4768 struct x86_exception *exception)
4770 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4771 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4775 /* Inline kvm_read_guest_virt_helper for speed. */
4776 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4778 if (unlikely(gpa == UNMAPPED_GVA))
4779 return X86EMUL_PROPAGATE_FAULT;
4781 offset = addr & (PAGE_SIZE-1);
4782 if (WARN_ON(offset + bytes > PAGE_SIZE))
4783 bytes = (unsigned)PAGE_SIZE - offset;
4784 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4786 if (unlikely(ret < 0))
4787 return X86EMUL_IO_NEEDED;
4789 return X86EMUL_CONTINUE;
4792 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4793 gva_t addr, void *val, unsigned int bytes,
4794 struct x86_exception *exception)
4796 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4797 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4799 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4802 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4804 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4805 gva_t addr, void *val, unsigned int bytes,
4806 struct x86_exception *exception)
4808 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4809 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4812 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4813 unsigned long addr, void *val, unsigned int bytes)
4815 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4816 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4818 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4821 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4822 gva_t addr, void *val,
4824 struct x86_exception *exception)
4826 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4828 int r = X86EMUL_CONTINUE;
4831 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4834 unsigned offset = addr & (PAGE_SIZE-1);
4835 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4838 if (gpa == UNMAPPED_GVA)
4839 return X86EMUL_PROPAGATE_FAULT;
4840 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4842 r = X86EMUL_IO_NEEDED;
4853 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4855 int handle_ud(struct kvm_vcpu *vcpu)
4857 int emul_type = EMULTYPE_TRAP_UD;
4858 enum emulation_result er;
4859 char sig[5]; /* ud2; .ascii "kvm" */
4860 struct x86_exception e;
4862 if (force_emulation_prefix &&
4863 kvm_read_guest_virt(&vcpu->arch.emulate_ctxt,
4864 kvm_get_linear_rip(vcpu), sig, sizeof(sig), &e) == 0 &&
4865 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
4866 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
4870 er = emulate_instruction(vcpu, emul_type);
4871 if (er == EMULATE_USER_EXIT)
4873 if (er != EMULATE_DONE)
4874 kvm_queue_exception(vcpu, UD_VECTOR);
4877 EXPORT_SYMBOL_GPL(handle_ud);
4879 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4880 gpa_t gpa, bool write)
4882 /* For APIC access vmexit */
4883 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4886 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4887 trace_vcpu_match_mmio(gva, gpa, write, true);
4894 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4895 gpa_t *gpa, struct x86_exception *exception,
4898 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4899 | (write ? PFERR_WRITE_MASK : 0);
4902 * currently PKRU is only applied to ept enabled guest so
4903 * there is no pkey in EPT page table for L1 guest or EPT
4904 * shadow page table for L2 guest.
4906 if (vcpu_match_mmio_gva(vcpu, gva)
4907 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4908 vcpu->arch.access, 0, access)) {
4909 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4910 (gva & (PAGE_SIZE - 1));
4911 trace_vcpu_match_mmio(gva, *gpa, write, false);
4915 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4917 if (*gpa == UNMAPPED_GVA)
4920 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4923 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4924 const void *val, int bytes)
4928 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4931 kvm_page_track_write(vcpu, gpa, val, bytes);
4935 struct read_write_emulator_ops {
4936 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4938 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4939 void *val, int bytes);
4940 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4941 int bytes, void *val);
4942 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4943 void *val, int bytes);
4947 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4949 if (vcpu->mmio_read_completed) {
4950 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4951 vcpu->mmio_fragments[0].gpa, val);
4952 vcpu->mmio_read_completed = 0;
4959 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4960 void *val, int bytes)
4962 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4965 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4966 void *val, int bytes)
4968 return emulator_write_phys(vcpu, gpa, val, bytes);
4971 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4973 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
4974 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4977 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4978 void *val, int bytes)
4980 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
4981 return X86EMUL_IO_NEEDED;
4984 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4985 void *val, int bytes)
4987 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4989 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4990 return X86EMUL_CONTINUE;
4993 static const struct read_write_emulator_ops read_emultor = {
4994 .read_write_prepare = read_prepare,
4995 .read_write_emulate = read_emulate,
4996 .read_write_mmio = vcpu_mmio_read,
4997 .read_write_exit_mmio = read_exit_mmio,
5000 static const struct read_write_emulator_ops write_emultor = {
5001 .read_write_emulate = write_emulate,
5002 .read_write_mmio = write_mmio,
5003 .read_write_exit_mmio = write_exit_mmio,
5007 static int emulator_read_write_onepage(unsigned long addr, void *val,
5009 struct x86_exception *exception,
5010 struct kvm_vcpu *vcpu,
5011 const struct read_write_emulator_ops *ops)
5015 bool write = ops->write;
5016 struct kvm_mmio_fragment *frag;
5017 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5020 * If the exit was due to a NPF we may already have a GPA.
5021 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5022 * Note, this cannot be used on string operations since string
5023 * operation using rep will only have the initial GPA from the NPF
5026 if (vcpu->arch.gpa_available &&
5027 emulator_can_use_gpa(ctxt) &&
5028 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5029 gpa = vcpu->arch.gpa_val;
5030 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5032 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5034 return X86EMUL_PROPAGATE_FAULT;
5037 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5038 return X86EMUL_CONTINUE;
5041 * Is this MMIO handled locally?
5043 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5044 if (handled == bytes)
5045 return X86EMUL_CONTINUE;
5051 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5052 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5056 return X86EMUL_CONTINUE;
5059 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5061 void *val, unsigned int bytes,
5062 struct x86_exception *exception,
5063 const struct read_write_emulator_ops *ops)
5065 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5069 if (ops->read_write_prepare &&
5070 ops->read_write_prepare(vcpu, val, bytes))
5071 return X86EMUL_CONTINUE;
5073 vcpu->mmio_nr_fragments = 0;
5075 /* Crossing a page boundary? */
5076 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5079 now = -addr & ~PAGE_MASK;
5080 rc = emulator_read_write_onepage(addr, val, now, exception,
5083 if (rc != X86EMUL_CONTINUE)
5086 if (ctxt->mode != X86EMUL_MODE_PROT64)
5092 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5094 if (rc != X86EMUL_CONTINUE)
5097 if (!vcpu->mmio_nr_fragments)
5100 gpa = vcpu->mmio_fragments[0].gpa;
5102 vcpu->mmio_needed = 1;
5103 vcpu->mmio_cur_fragment = 0;
5105 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5106 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5107 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5108 vcpu->run->mmio.phys_addr = gpa;
5110 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5113 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5117 struct x86_exception *exception)
5119 return emulator_read_write(ctxt, addr, val, bytes,
5120 exception, &read_emultor);
5123 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5127 struct x86_exception *exception)
5129 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5130 exception, &write_emultor);
5133 #define CMPXCHG_TYPE(t, ptr, old, new) \
5134 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5136 #ifdef CONFIG_X86_64
5137 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5139 # define CMPXCHG64(ptr, old, new) \
5140 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5143 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5148 struct x86_exception *exception)
5150 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5156 /* guests cmpxchg8b have to be emulated atomically */
5157 if (bytes > 8 || (bytes & (bytes - 1)))
5160 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5162 if (gpa == UNMAPPED_GVA ||
5163 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5166 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5169 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5170 if (is_error_page(page))
5173 kaddr = kmap_atomic(page);
5174 kaddr += offset_in_page(gpa);
5177 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5180 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5183 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5186 exchanged = CMPXCHG64(kaddr, old, new);
5191 kunmap_atomic(kaddr);
5192 kvm_release_page_dirty(page);
5195 return X86EMUL_CMPXCHG_FAILED;
5197 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5198 kvm_page_track_write(vcpu, gpa, new, bytes);
5200 return X86EMUL_CONTINUE;
5203 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5205 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5208 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5212 for (i = 0; i < vcpu->arch.pio.count; i++) {
5213 if (vcpu->arch.pio.in)
5214 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5215 vcpu->arch.pio.size, pd);
5217 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5218 vcpu->arch.pio.port, vcpu->arch.pio.size,
5222 pd += vcpu->arch.pio.size;
5227 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5228 unsigned short port, void *val,
5229 unsigned int count, bool in)
5231 vcpu->arch.pio.port = port;
5232 vcpu->arch.pio.in = in;
5233 vcpu->arch.pio.count = count;
5234 vcpu->arch.pio.size = size;
5236 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5237 vcpu->arch.pio.count = 0;
5241 vcpu->run->exit_reason = KVM_EXIT_IO;
5242 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5243 vcpu->run->io.size = size;
5244 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5245 vcpu->run->io.count = count;
5246 vcpu->run->io.port = port;
5251 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5252 int size, unsigned short port, void *val,
5255 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5258 if (vcpu->arch.pio.count)
5261 memset(vcpu->arch.pio_data, 0, size * count);
5263 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5266 memcpy(val, vcpu->arch.pio_data, size * count);
5267 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5268 vcpu->arch.pio.count = 0;
5275 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5276 int size, unsigned short port,
5277 const void *val, unsigned int count)
5279 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5281 memcpy(vcpu->arch.pio_data, val, size * count);
5282 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5283 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5286 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5288 return kvm_x86_ops->get_segment_base(vcpu, seg);
5291 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5293 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5296 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5298 if (!need_emulate_wbinvd(vcpu))
5299 return X86EMUL_CONTINUE;
5301 if (kvm_x86_ops->has_wbinvd_exit()) {
5302 int cpu = get_cpu();
5304 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5305 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5306 wbinvd_ipi, NULL, 1);
5308 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5311 return X86EMUL_CONTINUE;
5314 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5316 kvm_emulate_wbinvd_noskip(vcpu);
5317 return kvm_skip_emulated_instruction(vcpu);
5319 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5323 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5325 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5328 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5329 unsigned long *dest)
5331 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5334 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5335 unsigned long value)
5338 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5341 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5343 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5346 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5348 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5349 unsigned long value;
5353 value = kvm_read_cr0(vcpu);
5356 value = vcpu->arch.cr2;
5359 value = kvm_read_cr3(vcpu);
5362 value = kvm_read_cr4(vcpu);
5365 value = kvm_get_cr8(vcpu);
5368 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5375 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5377 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5382 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5385 vcpu->arch.cr2 = val;
5388 res = kvm_set_cr3(vcpu, val);
5391 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5394 res = kvm_set_cr8(vcpu, val);
5397 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5404 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5406 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5409 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5411 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5414 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5416 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5419 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5421 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5424 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5426 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5429 static unsigned long emulator_get_cached_segment_base(
5430 struct x86_emulate_ctxt *ctxt, int seg)
5432 return get_segment_base(emul_to_vcpu(ctxt), seg);
5435 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5436 struct desc_struct *desc, u32 *base3,
5439 struct kvm_segment var;
5441 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5442 *selector = var.selector;
5445 memset(desc, 0, sizeof(*desc));
5453 set_desc_limit(desc, var.limit);
5454 set_desc_base(desc, (unsigned long)var.base);
5455 #ifdef CONFIG_X86_64
5457 *base3 = var.base >> 32;
5459 desc->type = var.type;
5461 desc->dpl = var.dpl;
5462 desc->p = var.present;
5463 desc->avl = var.avl;
5471 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5472 struct desc_struct *desc, u32 base3,
5475 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5476 struct kvm_segment var;
5478 var.selector = selector;
5479 var.base = get_desc_base(desc);
5480 #ifdef CONFIG_X86_64
5481 var.base |= ((u64)base3) << 32;
5483 var.limit = get_desc_limit(desc);
5485 var.limit = (var.limit << 12) | 0xfff;
5486 var.type = desc->type;
5487 var.dpl = desc->dpl;
5492 var.avl = desc->avl;
5493 var.present = desc->p;
5494 var.unusable = !var.present;
5497 kvm_set_segment(vcpu, &var, seg);
5501 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5502 u32 msr_index, u64 *pdata)
5504 struct msr_data msr;
5507 msr.index = msr_index;
5508 msr.host_initiated = false;
5509 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5517 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5518 u32 msr_index, u64 data)
5520 struct msr_data msr;
5523 msr.index = msr_index;
5524 msr.host_initiated = false;
5525 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5528 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5530 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5532 return vcpu->arch.smbase;
5535 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5537 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5539 vcpu->arch.smbase = smbase;
5542 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5545 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5548 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5549 u32 pmc, u64 *pdata)
5551 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5554 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5556 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5559 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5560 struct x86_instruction_info *info,
5561 enum x86_intercept_stage stage)
5563 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5566 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5567 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5569 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5572 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5574 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5577 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5579 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5582 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5584 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5587 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5589 return emul_to_vcpu(ctxt)->arch.hflags;
5592 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5594 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5597 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5599 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5602 static const struct x86_emulate_ops emulate_ops = {
5603 .read_gpr = emulator_read_gpr,
5604 .write_gpr = emulator_write_gpr,
5605 .read_std = kvm_read_guest_virt_system,
5606 .write_std = kvm_write_guest_virt_system,
5607 .read_phys = kvm_read_guest_phys_system,
5608 .fetch = kvm_fetch_guest_virt,
5609 .read_emulated = emulator_read_emulated,
5610 .write_emulated = emulator_write_emulated,
5611 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5612 .invlpg = emulator_invlpg,
5613 .pio_in_emulated = emulator_pio_in_emulated,
5614 .pio_out_emulated = emulator_pio_out_emulated,
5615 .get_segment = emulator_get_segment,
5616 .set_segment = emulator_set_segment,
5617 .get_cached_segment_base = emulator_get_cached_segment_base,
5618 .get_gdt = emulator_get_gdt,
5619 .get_idt = emulator_get_idt,
5620 .set_gdt = emulator_set_gdt,
5621 .set_idt = emulator_set_idt,
5622 .get_cr = emulator_get_cr,
5623 .set_cr = emulator_set_cr,
5624 .cpl = emulator_get_cpl,
5625 .get_dr = emulator_get_dr,
5626 .set_dr = emulator_set_dr,
5627 .get_smbase = emulator_get_smbase,
5628 .set_smbase = emulator_set_smbase,
5629 .set_msr = emulator_set_msr,
5630 .get_msr = emulator_get_msr,
5631 .check_pmc = emulator_check_pmc,
5632 .read_pmc = emulator_read_pmc,
5633 .halt = emulator_halt,
5634 .wbinvd = emulator_wbinvd,
5635 .fix_hypercall = emulator_fix_hypercall,
5636 .intercept = emulator_intercept,
5637 .get_cpuid = emulator_get_cpuid,
5638 .set_nmi_mask = emulator_set_nmi_mask,
5639 .get_hflags = emulator_get_hflags,
5640 .set_hflags = emulator_set_hflags,
5641 .pre_leave_smm = emulator_pre_leave_smm,
5644 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5646 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5648 * an sti; sti; sequence only disable interrupts for the first
5649 * instruction. So, if the last instruction, be it emulated or
5650 * not, left the system with the INT_STI flag enabled, it
5651 * means that the last instruction is an sti. We should not
5652 * leave the flag on in this case. The same goes for mov ss
5654 if (int_shadow & mask)
5656 if (unlikely(int_shadow || mask)) {
5657 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5659 kvm_make_request(KVM_REQ_EVENT, vcpu);
5663 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5665 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5666 if (ctxt->exception.vector == PF_VECTOR)
5667 return kvm_propagate_fault(vcpu, &ctxt->exception);
5669 if (ctxt->exception.error_code_valid)
5670 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5671 ctxt->exception.error_code);
5673 kvm_queue_exception(vcpu, ctxt->exception.vector);
5677 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5679 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5682 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5684 ctxt->eflags = kvm_get_rflags(vcpu);
5685 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5687 ctxt->eip = kvm_rip_read(vcpu);
5688 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5689 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5690 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5691 cs_db ? X86EMUL_MODE_PROT32 :
5692 X86EMUL_MODE_PROT16;
5693 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5694 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5695 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5697 init_decode_cache(ctxt);
5698 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5701 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5703 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5706 init_emulate_ctxt(vcpu);
5710 ctxt->_eip = ctxt->eip + inc_eip;
5711 ret = emulate_int_real(ctxt, irq);
5713 if (ret != X86EMUL_CONTINUE)
5714 return EMULATE_FAIL;
5716 ctxt->eip = ctxt->_eip;
5717 kvm_rip_write(vcpu, ctxt->eip);
5718 kvm_set_rflags(vcpu, ctxt->eflags);
5720 return EMULATE_DONE;
5722 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5724 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
5726 int r = EMULATE_DONE;
5728 ++vcpu->stat.insn_emulation_fail;
5729 trace_kvm_emulate_insn_failed(vcpu);
5731 if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
5732 return EMULATE_FAIL;
5734 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5735 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5736 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5737 vcpu->run->internal.ndata = 0;
5738 r = EMULATE_USER_EXIT;
5741 kvm_queue_exception(vcpu, UD_VECTOR);
5746 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5747 bool write_fault_to_shadow_pgtable,
5753 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5756 if (!vcpu->arch.mmu.direct_map) {
5758 * Write permission should be allowed since only
5759 * write access need to be emulated.
5761 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5764 * If the mapping is invalid in guest, let cpu retry
5765 * it to generate fault.
5767 if (gpa == UNMAPPED_GVA)
5772 * Do not retry the unhandleable instruction if it faults on the
5773 * readonly host memory, otherwise it will goto a infinite loop:
5774 * retry instruction -> write #PF -> emulation fail -> retry
5775 * instruction -> ...
5777 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5780 * If the instruction failed on the error pfn, it can not be fixed,
5781 * report the error to userspace.
5783 if (is_error_noslot_pfn(pfn))
5786 kvm_release_pfn_clean(pfn);
5788 /* The instructions are well-emulated on direct mmu. */
5789 if (vcpu->arch.mmu.direct_map) {
5790 unsigned int indirect_shadow_pages;
5792 spin_lock(&vcpu->kvm->mmu_lock);
5793 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5794 spin_unlock(&vcpu->kvm->mmu_lock);
5796 if (indirect_shadow_pages)
5797 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5803 * if emulation was due to access to shadowed page table
5804 * and it failed try to unshadow page and re-enter the
5805 * guest to let CPU execute the instruction.
5807 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5810 * If the access faults on its page table, it can not
5811 * be fixed by unprotecting shadow page and it should
5812 * be reported to userspace.
5814 return !write_fault_to_shadow_pgtable;
5817 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5818 unsigned long cr2, int emulation_type)
5820 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5821 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5823 last_retry_eip = vcpu->arch.last_retry_eip;
5824 last_retry_addr = vcpu->arch.last_retry_addr;
5827 * If the emulation is caused by #PF and it is non-page_table
5828 * writing instruction, it means the VM-EXIT is caused by shadow
5829 * page protected, we can zap the shadow page and retry this
5830 * instruction directly.
5832 * Note: if the guest uses a non-page-table modifying instruction
5833 * on the PDE that points to the instruction, then we will unmap
5834 * the instruction and go to an infinite loop. So, we cache the
5835 * last retried eip and the last fault address, if we meet the eip
5836 * and the address again, we can break out of the potential infinite
5839 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5841 if (!(emulation_type & EMULTYPE_RETRY))
5844 if (x86_page_table_writing_insn(ctxt))
5847 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5850 vcpu->arch.last_retry_eip = ctxt->eip;
5851 vcpu->arch.last_retry_addr = cr2;
5853 if (!vcpu->arch.mmu.direct_map)
5854 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5856 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5861 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5862 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5864 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5866 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5867 /* This is a good place to trace that we are exiting SMM. */
5868 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5870 /* Process a latched INIT or SMI, if any. */
5871 kvm_make_request(KVM_REQ_EVENT, vcpu);
5874 kvm_mmu_reset_context(vcpu);
5877 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5879 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5881 vcpu->arch.hflags = emul_flags;
5883 if (changed & HF_SMM_MASK)
5884 kvm_smm_changed(vcpu);
5887 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5896 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5897 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5902 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5904 struct kvm_run *kvm_run = vcpu->run;
5906 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5907 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5908 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5909 kvm_run->debug.arch.exception = DB_VECTOR;
5910 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5911 *r = EMULATE_USER_EXIT;
5914 * "Certain debug exceptions may clear bit 0-3. The
5915 * remaining contents of the DR6 register are never
5916 * cleared by the processor".
5918 vcpu->arch.dr6 &= ~15;
5919 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5920 kvm_queue_exception(vcpu, DB_VECTOR);
5924 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5926 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5927 int r = EMULATE_DONE;
5929 kvm_x86_ops->skip_emulated_instruction(vcpu);
5932 * rflags is the old, "raw" value of the flags. The new value has
5933 * not been saved yet.
5935 * This is correct even for TF set by the guest, because "the
5936 * processor will not generate this exception after the instruction
5937 * that sets the TF flag".
5939 if (unlikely(rflags & X86_EFLAGS_TF))
5940 kvm_vcpu_do_singlestep(vcpu, &r);
5941 return r == EMULATE_DONE;
5943 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5945 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5947 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5948 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5949 struct kvm_run *kvm_run = vcpu->run;
5950 unsigned long eip = kvm_get_linear_rip(vcpu);
5951 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5952 vcpu->arch.guest_debug_dr7,
5956 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5957 kvm_run->debug.arch.pc = eip;
5958 kvm_run->debug.arch.exception = DB_VECTOR;
5959 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5960 *r = EMULATE_USER_EXIT;
5965 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5966 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5967 unsigned long eip = kvm_get_linear_rip(vcpu);
5968 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5973 vcpu->arch.dr6 &= ~15;
5974 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5975 kvm_queue_exception(vcpu, DB_VECTOR);
5984 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
5986 switch (ctxt->opcode_len) {
5993 case 0xe6: /* OUT */
5997 case 0x6c: /* INS */
5999 case 0x6e: /* OUTS */
6006 case 0x33: /* RDPMC */
6015 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6022 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6023 bool writeback = true;
6024 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6027 * Clear write_fault_to_shadow_pgtable here to ensure it is
6030 vcpu->arch.write_fault_to_shadow_pgtable = false;
6031 kvm_clear_exception_queue(vcpu);
6033 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6034 init_emulate_ctxt(vcpu);
6037 * We will reenter on the same instruction since
6038 * we do not set complete_userspace_io. This does not
6039 * handle watchpoints yet, those would be handled in
6042 if (!(emulation_type & EMULTYPE_SKIP) &&
6043 kvm_vcpu_check_breakpoint(vcpu, &r))
6046 ctxt->interruptibility = 0;
6047 ctxt->have_exception = false;
6048 ctxt->exception.vector = -1;
6049 ctxt->perm_ok = false;
6051 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6053 r = x86_decode_insn(ctxt, insn, insn_len);
6055 trace_kvm_emulate_insn_start(vcpu);
6056 ++vcpu->stat.insn_emulation;
6057 if (r != EMULATION_OK) {
6058 if (emulation_type & EMULTYPE_TRAP_UD)
6059 return EMULATE_FAIL;
6060 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6062 return EMULATE_DONE;
6063 if (ctxt->have_exception && inject_emulated_exception(vcpu))
6064 return EMULATE_DONE;
6065 if (emulation_type & EMULTYPE_SKIP)
6066 return EMULATE_FAIL;
6067 return handle_emulation_failure(vcpu, emulation_type);
6071 if ((emulation_type & EMULTYPE_VMWARE) &&
6072 !is_vmware_backdoor_opcode(ctxt))
6073 return EMULATE_FAIL;
6075 if (emulation_type & EMULTYPE_SKIP) {
6076 kvm_rip_write(vcpu, ctxt->_eip);
6077 if (ctxt->eflags & X86_EFLAGS_RF)
6078 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6079 return EMULATE_DONE;
6082 if (retry_instruction(ctxt, cr2, emulation_type))
6083 return EMULATE_DONE;
6085 /* this is needed for vmware backdoor interface to work since it
6086 changes registers values during IO operation */
6087 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6088 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6089 emulator_invalidate_register_cache(ctxt);
6093 /* Save the faulting GPA (cr2) in the address field */
6094 ctxt->exception.address = cr2;
6096 r = x86_emulate_insn(ctxt);
6098 if (r == EMULATION_INTERCEPTED)
6099 return EMULATE_DONE;
6101 if (r == EMULATION_FAILED) {
6102 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6104 return EMULATE_DONE;
6106 return handle_emulation_failure(vcpu, emulation_type);
6109 if (ctxt->have_exception) {
6111 if (inject_emulated_exception(vcpu))
6113 } else if (vcpu->arch.pio.count) {
6114 if (!vcpu->arch.pio.in) {
6115 /* FIXME: return into emulator if single-stepping. */
6116 vcpu->arch.pio.count = 0;
6119 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6121 r = EMULATE_USER_EXIT;
6122 } else if (vcpu->mmio_needed) {
6123 if (!vcpu->mmio_is_write)
6125 r = EMULATE_USER_EXIT;
6126 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6127 } else if (r == EMULATION_RESTART)
6133 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6134 toggle_interruptibility(vcpu, ctxt->interruptibility);
6135 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6136 kvm_rip_write(vcpu, ctxt->eip);
6137 if (r == EMULATE_DONE &&
6138 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
6139 kvm_vcpu_do_singlestep(vcpu, &r);
6140 if (!ctxt->have_exception ||
6141 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6142 __kvm_set_rflags(vcpu, ctxt->eflags);
6145 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6146 * do nothing, and it will be requested again as soon as
6147 * the shadow expires. But we still need to check here,
6148 * because POPF has no interrupt shadow.
6150 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6151 kvm_make_request(KVM_REQ_EVENT, vcpu);
6153 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6157 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
6159 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6160 unsigned short port)
6162 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6163 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6164 size, port, &val, 1);
6165 /* do not return to emulator after return from userspace */
6166 vcpu->arch.pio.count = 0;
6170 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6174 /* We should only ever be called with arch.pio.count equal to 1 */
6175 BUG_ON(vcpu->arch.pio.count != 1);
6177 /* For size less than 4 we merge, else we zero extend */
6178 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6182 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6183 * the copy and tracing
6185 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6186 vcpu->arch.pio.port, &val, 1);
6187 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6192 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6193 unsigned short port)
6198 /* For size less than 4 we merge, else we zero extend */
6199 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6201 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6204 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6208 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6213 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6215 int ret = kvm_skip_emulated_instruction(vcpu);
6218 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6219 * KVM_EXIT_DEBUG here.
6222 return kvm_fast_pio_in(vcpu, size, port) && ret;
6224 return kvm_fast_pio_out(vcpu, size, port) && ret;
6226 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6228 static int kvmclock_cpu_down_prep(unsigned int cpu)
6230 __this_cpu_write(cpu_tsc_khz, 0);
6234 static void tsc_khz_changed(void *data)
6236 struct cpufreq_freqs *freq = data;
6237 unsigned long khz = 0;
6241 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6242 khz = cpufreq_quick_get(raw_smp_processor_id());
6245 __this_cpu_write(cpu_tsc_khz, khz);
6248 #ifdef CONFIG_X86_64
6249 static void kvm_hyperv_tsc_notifier(void)
6252 struct kvm_vcpu *vcpu;
6255 spin_lock(&kvm_lock);
6256 list_for_each_entry(kvm, &vm_list, vm_list)
6257 kvm_make_mclock_inprogress_request(kvm);
6259 hyperv_stop_tsc_emulation();
6261 /* TSC frequency always matches when on Hyper-V */
6262 for_each_present_cpu(cpu)
6263 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6264 kvm_max_guest_tsc_khz = tsc_khz;
6266 list_for_each_entry(kvm, &vm_list, vm_list) {
6267 struct kvm_arch *ka = &kvm->arch;
6269 spin_lock(&ka->pvclock_gtod_sync_lock);
6271 pvclock_update_vm_gtod_copy(kvm);
6273 kvm_for_each_vcpu(cpu, vcpu, kvm)
6274 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6276 kvm_for_each_vcpu(cpu, vcpu, kvm)
6277 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6279 spin_unlock(&ka->pvclock_gtod_sync_lock);
6281 spin_unlock(&kvm_lock);
6285 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6288 struct cpufreq_freqs *freq = data;
6290 struct kvm_vcpu *vcpu;
6291 int i, send_ipi = 0;
6294 * We allow guests to temporarily run on slowing clocks,
6295 * provided we notify them after, or to run on accelerating
6296 * clocks, provided we notify them before. Thus time never
6299 * However, we have a problem. We can't atomically update
6300 * the frequency of a given CPU from this function; it is
6301 * merely a notifier, which can be called from any CPU.
6302 * Changing the TSC frequency at arbitrary points in time
6303 * requires a recomputation of local variables related to
6304 * the TSC for each VCPU. We must flag these local variables
6305 * to be updated and be sure the update takes place with the
6306 * new frequency before any guests proceed.
6308 * Unfortunately, the combination of hotplug CPU and frequency
6309 * change creates an intractable locking scenario; the order
6310 * of when these callouts happen is undefined with respect to
6311 * CPU hotplug, and they can race with each other. As such,
6312 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6313 * undefined; you can actually have a CPU frequency change take
6314 * place in between the computation of X and the setting of the
6315 * variable. To protect against this problem, all updates of
6316 * the per_cpu tsc_khz variable are done in an interrupt
6317 * protected IPI, and all callers wishing to update the value
6318 * must wait for a synchronous IPI to complete (which is trivial
6319 * if the caller is on the CPU already). This establishes the
6320 * necessary total order on variable updates.
6322 * Note that because a guest time update may take place
6323 * anytime after the setting of the VCPU's request bit, the
6324 * correct TSC value must be set before the request. However,
6325 * to ensure the update actually makes it to any guest which
6326 * starts running in hardware virtualization between the set
6327 * and the acquisition of the spinlock, we must also ping the
6328 * CPU after setting the request bit.
6332 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6334 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6337 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6339 spin_lock(&kvm_lock);
6340 list_for_each_entry(kvm, &vm_list, vm_list) {
6341 kvm_for_each_vcpu(i, vcpu, kvm) {
6342 if (vcpu->cpu != freq->cpu)
6344 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6345 if (vcpu->cpu != smp_processor_id())
6349 spin_unlock(&kvm_lock);
6351 if (freq->old < freq->new && send_ipi) {
6353 * We upscale the frequency. Must make the guest
6354 * doesn't see old kvmclock values while running with
6355 * the new frequency, otherwise we risk the guest sees
6356 * time go backwards.
6358 * In case we update the frequency for another cpu
6359 * (which might be in guest context) send an interrupt
6360 * to kick the cpu out of guest context. Next time
6361 * guest context is entered kvmclock will be updated,
6362 * so the guest will not see stale values.
6364 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6369 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6370 .notifier_call = kvmclock_cpufreq_notifier
6373 static int kvmclock_cpu_online(unsigned int cpu)
6375 tsc_khz_changed(NULL);
6379 static void kvm_timer_init(void)
6381 max_tsc_khz = tsc_khz;
6383 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6384 #ifdef CONFIG_CPU_FREQ
6385 struct cpufreq_policy policy;
6388 memset(&policy, 0, sizeof(policy));
6390 cpufreq_get_policy(&policy, cpu);
6391 if (policy.cpuinfo.max_freq)
6392 max_tsc_khz = policy.cpuinfo.max_freq;
6395 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6396 CPUFREQ_TRANSITION_NOTIFIER);
6398 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6400 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6401 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6404 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6405 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6407 int kvm_is_in_guest(void)
6409 return __this_cpu_read(current_vcpu) != NULL;
6412 static int kvm_is_user_mode(void)
6416 if (__this_cpu_read(current_vcpu))
6417 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6419 return user_mode != 0;
6422 static unsigned long kvm_get_guest_ip(void)
6424 unsigned long ip = 0;
6426 if (__this_cpu_read(current_vcpu))
6427 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6432 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6433 .is_in_guest = kvm_is_in_guest,
6434 .is_user_mode = kvm_is_user_mode,
6435 .get_guest_ip = kvm_get_guest_ip,
6438 static void kvm_set_mmio_spte_mask(void)
6441 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6444 * Set the reserved bits and the present bit of an paging-structure
6445 * entry to generate page fault with PFER.RSV = 1.
6447 /* Mask the reserved physical address bits. */
6448 mask = rsvd_bits(maxphyaddr, 51);
6450 /* Set the present bit. */
6453 #ifdef CONFIG_X86_64
6455 * If reserved bit is not supported, clear the present bit to disable
6458 if (maxphyaddr == 52)
6462 kvm_mmu_set_mmio_spte_mask(mask, mask);
6465 #ifdef CONFIG_X86_64
6466 static void pvclock_gtod_update_fn(struct work_struct *work)
6470 struct kvm_vcpu *vcpu;
6473 spin_lock(&kvm_lock);
6474 list_for_each_entry(kvm, &vm_list, vm_list)
6475 kvm_for_each_vcpu(i, vcpu, kvm)
6476 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6477 atomic_set(&kvm_guest_has_master_clock, 0);
6478 spin_unlock(&kvm_lock);
6481 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6484 * Notification about pvclock gtod data update.
6486 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6489 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6490 struct timekeeper *tk = priv;
6492 update_pvclock_gtod(tk);
6494 /* disable master clock if host does not trust, or does not
6495 * use, TSC based clocksource.
6497 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6498 atomic_read(&kvm_guest_has_master_clock) != 0)
6499 queue_work(system_long_wq, &pvclock_gtod_work);
6504 static struct notifier_block pvclock_gtod_notifier = {
6505 .notifier_call = pvclock_gtod_notify,
6509 int kvm_arch_init(void *opaque)
6512 struct kvm_x86_ops *ops = opaque;
6515 printk(KERN_ERR "kvm: already loaded the other module\n");
6520 if (!ops->cpu_has_kvm_support()) {
6521 printk(KERN_ERR "kvm: no hardware support\n");
6525 if (ops->disabled_by_bios()) {
6526 printk(KERN_ERR "kvm: disabled by bios\n");
6532 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6534 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6538 r = kvm_mmu_module_init();
6540 goto out_free_percpu;
6542 kvm_set_mmio_spte_mask();
6546 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6547 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6548 PT_PRESENT_MASK, 0, sme_me_mask);
6551 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6553 if (boot_cpu_has(X86_FEATURE_XSAVE))
6554 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6557 #ifdef CONFIG_X86_64
6558 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6560 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6561 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6567 free_percpu(shared_msrs);
6572 void kvm_arch_exit(void)
6574 #ifdef CONFIG_X86_64
6575 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6576 clear_hv_tscchange_cb();
6579 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6581 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6582 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6583 CPUFREQ_TRANSITION_NOTIFIER);
6584 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6585 #ifdef CONFIG_X86_64
6586 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6589 kvm_mmu_module_exit();
6590 free_percpu(shared_msrs);
6593 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6595 ++vcpu->stat.halt_exits;
6596 if (lapic_in_kernel(vcpu)) {
6597 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6600 vcpu->run->exit_reason = KVM_EXIT_HLT;
6604 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6606 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6608 int ret = kvm_skip_emulated_instruction(vcpu);
6610 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6611 * KVM_EXIT_DEBUG here.
6613 return kvm_vcpu_halt(vcpu) && ret;
6615 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6617 #ifdef CONFIG_X86_64
6618 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6619 unsigned long clock_type)
6621 struct kvm_clock_pairing clock_pairing;
6626 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6627 return -KVM_EOPNOTSUPP;
6629 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6630 return -KVM_EOPNOTSUPP;
6632 clock_pairing.sec = ts.tv_sec;
6633 clock_pairing.nsec = ts.tv_nsec;
6634 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6635 clock_pairing.flags = 0;
6638 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6639 sizeof(struct kvm_clock_pairing)))
6647 * kvm_pv_kick_cpu_op: Kick a vcpu.
6649 * @apicid - apicid of vcpu to be kicked.
6651 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6653 struct kvm_lapic_irq lapic_irq;
6655 lapic_irq.shorthand = 0;
6656 lapic_irq.dest_mode = 0;
6657 lapic_irq.level = 0;
6658 lapic_irq.dest_id = apicid;
6659 lapic_irq.msi_redir_hint = false;
6661 lapic_irq.delivery_mode = APIC_DM_REMRD;
6662 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6665 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6667 vcpu->arch.apicv_active = false;
6668 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6671 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6673 unsigned long nr, a0, a1, a2, a3, ret;
6676 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6677 return kvm_hv_hypercall(vcpu);
6679 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6680 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6681 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6682 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6683 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6685 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6687 op_64_bit = is_64_bit_mode(vcpu);
6696 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6702 case KVM_HC_VAPIC_POLL_IRQ:
6705 case KVM_HC_KICK_CPU:
6706 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6709 #ifdef CONFIG_X86_64
6710 case KVM_HC_CLOCK_PAIRING:
6711 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6721 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6723 ++vcpu->stat.hypercalls;
6724 return kvm_skip_emulated_instruction(vcpu);
6726 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6728 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6730 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6731 char instruction[3];
6732 unsigned long rip = kvm_rip_read(vcpu);
6734 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6736 return emulator_write_emulated(ctxt, rip, instruction, 3,
6740 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6742 return vcpu->run->request_interrupt_window &&
6743 likely(!pic_in_kernel(vcpu->kvm));
6746 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6748 struct kvm_run *kvm_run = vcpu->run;
6750 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6751 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6752 kvm_run->cr8 = kvm_get_cr8(vcpu);
6753 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6754 kvm_run->ready_for_interrupt_injection =
6755 pic_in_kernel(vcpu->kvm) ||
6756 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6759 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6763 if (!kvm_x86_ops->update_cr8_intercept)
6766 if (!lapic_in_kernel(vcpu))
6769 if (vcpu->arch.apicv_active)
6772 if (!vcpu->arch.apic->vapic_addr)
6773 max_irr = kvm_lapic_find_highest_irr(vcpu);
6780 tpr = kvm_lapic_get_cr8(vcpu);
6782 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6785 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6789 /* try to reinject previous events if any */
6791 if (vcpu->arch.exception.injected)
6792 kvm_x86_ops->queue_exception(vcpu);
6794 * Do not inject an NMI or interrupt if there is a pending
6795 * exception. Exceptions and interrupts are recognized at
6796 * instruction boundaries, i.e. the start of an instruction.
6797 * Trap-like exceptions, e.g. #DB, have higher priority than
6798 * NMIs and interrupts, i.e. traps are recognized before an
6799 * NMI/interrupt that's pending on the same instruction.
6800 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
6801 * priority, but are only generated (pended) during instruction
6802 * execution, i.e. a pending fault-like exception means the
6803 * fault occurred on the *previous* instruction and must be
6804 * serviced prior to recognizing any new events in order to
6805 * fully complete the previous instruction.
6807 else if (!vcpu->arch.exception.pending) {
6808 if (vcpu->arch.nmi_injected)
6809 kvm_x86_ops->set_nmi(vcpu);
6810 else if (vcpu->arch.interrupt.injected)
6811 kvm_x86_ops->set_irq(vcpu);
6815 * Call check_nested_events() even if we reinjected a previous event
6816 * in order for caller to determine if it should require immediate-exit
6817 * from L2 to L1 due to pending L1 events which require exit
6820 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6821 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6826 /* try to inject new event if pending */
6827 if (vcpu->arch.exception.pending) {
6828 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6829 vcpu->arch.exception.has_error_code,
6830 vcpu->arch.exception.error_code);
6832 WARN_ON_ONCE(vcpu->arch.exception.injected);
6833 vcpu->arch.exception.pending = false;
6834 vcpu->arch.exception.injected = true;
6836 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6837 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6840 if (vcpu->arch.exception.nr == DB_VECTOR &&
6841 (vcpu->arch.dr7 & DR7_GD)) {
6842 vcpu->arch.dr7 &= ~DR7_GD;
6843 kvm_update_dr7(vcpu);
6846 kvm_x86_ops->queue_exception(vcpu);
6849 /* Don't consider new event if we re-injected an event */
6850 if (kvm_event_needs_reinjection(vcpu))
6853 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
6854 kvm_x86_ops->smi_allowed(vcpu)) {
6855 vcpu->arch.smi_pending = false;
6856 ++vcpu->arch.smi_count;
6858 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6859 --vcpu->arch.nmi_pending;
6860 vcpu->arch.nmi_injected = true;
6861 kvm_x86_ops->set_nmi(vcpu);
6862 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6864 * Because interrupts can be injected asynchronously, we are
6865 * calling check_nested_events again here to avoid a race condition.
6866 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6867 * proposal and current concerns. Perhaps we should be setting
6868 * KVM_REQ_EVENT only on certain events and not unconditionally?
6870 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6871 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6875 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6876 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6878 kvm_x86_ops->set_irq(vcpu);
6885 static void process_nmi(struct kvm_vcpu *vcpu)
6890 * x86 is limited to one NMI running, and one NMI pending after it.
6891 * If an NMI is already in progress, limit further NMIs to just one.
6892 * Otherwise, allow two (and we'll inject the first one immediately).
6894 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6897 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6898 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6899 kvm_make_request(KVM_REQ_EVENT, vcpu);
6902 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6905 flags |= seg->g << 23;
6906 flags |= seg->db << 22;
6907 flags |= seg->l << 21;
6908 flags |= seg->avl << 20;
6909 flags |= seg->present << 15;
6910 flags |= seg->dpl << 13;
6911 flags |= seg->s << 12;
6912 flags |= seg->type << 8;
6916 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6918 struct kvm_segment seg;
6921 kvm_get_segment(vcpu, &seg, n);
6922 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6925 offset = 0x7f84 + n * 12;
6927 offset = 0x7f2c + (n - 3) * 12;
6929 put_smstate(u32, buf, offset + 8, seg.base);
6930 put_smstate(u32, buf, offset + 4, seg.limit);
6931 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6934 #ifdef CONFIG_X86_64
6935 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6937 struct kvm_segment seg;
6941 kvm_get_segment(vcpu, &seg, n);
6942 offset = 0x7e00 + n * 16;
6944 flags = enter_smm_get_segment_flags(&seg) >> 8;
6945 put_smstate(u16, buf, offset, seg.selector);
6946 put_smstate(u16, buf, offset + 2, flags);
6947 put_smstate(u32, buf, offset + 4, seg.limit);
6948 put_smstate(u64, buf, offset + 8, seg.base);
6952 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6955 struct kvm_segment seg;
6959 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6960 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6961 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6962 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6964 for (i = 0; i < 8; i++)
6965 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6967 kvm_get_dr(vcpu, 6, &val);
6968 put_smstate(u32, buf, 0x7fcc, (u32)val);
6969 kvm_get_dr(vcpu, 7, &val);
6970 put_smstate(u32, buf, 0x7fc8, (u32)val);
6972 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6973 put_smstate(u32, buf, 0x7fc4, seg.selector);
6974 put_smstate(u32, buf, 0x7f64, seg.base);
6975 put_smstate(u32, buf, 0x7f60, seg.limit);
6976 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6978 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6979 put_smstate(u32, buf, 0x7fc0, seg.selector);
6980 put_smstate(u32, buf, 0x7f80, seg.base);
6981 put_smstate(u32, buf, 0x7f7c, seg.limit);
6982 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6984 kvm_x86_ops->get_gdt(vcpu, &dt);
6985 put_smstate(u32, buf, 0x7f74, dt.address);
6986 put_smstate(u32, buf, 0x7f70, dt.size);
6988 kvm_x86_ops->get_idt(vcpu, &dt);
6989 put_smstate(u32, buf, 0x7f58, dt.address);
6990 put_smstate(u32, buf, 0x7f54, dt.size);
6992 for (i = 0; i < 6; i++)
6993 enter_smm_save_seg_32(vcpu, buf, i);
6995 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6998 put_smstate(u32, buf, 0x7efc, 0x00020000);
6999 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7002 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7004 #ifdef CONFIG_X86_64
7006 struct kvm_segment seg;
7010 for (i = 0; i < 16; i++)
7011 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7013 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7014 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7016 kvm_get_dr(vcpu, 6, &val);
7017 put_smstate(u64, buf, 0x7f68, val);
7018 kvm_get_dr(vcpu, 7, &val);
7019 put_smstate(u64, buf, 0x7f60, val);
7021 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7022 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7023 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7025 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7028 put_smstate(u32, buf, 0x7efc, 0x00020064);
7030 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7032 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7033 put_smstate(u16, buf, 0x7e90, seg.selector);
7034 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7035 put_smstate(u32, buf, 0x7e94, seg.limit);
7036 put_smstate(u64, buf, 0x7e98, seg.base);
7038 kvm_x86_ops->get_idt(vcpu, &dt);
7039 put_smstate(u32, buf, 0x7e84, dt.size);
7040 put_smstate(u64, buf, 0x7e88, dt.address);
7042 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7043 put_smstate(u16, buf, 0x7e70, seg.selector);
7044 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7045 put_smstate(u32, buf, 0x7e74, seg.limit);
7046 put_smstate(u64, buf, 0x7e78, seg.base);
7048 kvm_x86_ops->get_gdt(vcpu, &dt);
7049 put_smstate(u32, buf, 0x7e64, dt.size);
7050 put_smstate(u64, buf, 0x7e68, dt.address);
7052 for (i = 0; i < 6; i++)
7053 enter_smm_save_seg_64(vcpu, buf, i);
7059 static void enter_smm(struct kvm_vcpu *vcpu)
7061 struct kvm_segment cs, ds;
7066 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7067 memset(buf, 0, 512);
7068 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7069 enter_smm_save_state_64(vcpu, buf);
7071 enter_smm_save_state_32(vcpu, buf);
7074 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7075 * vCPU state (e.g. leave guest mode) after we've saved the state into
7076 * the SMM state-save area.
7078 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7080 vcpu->arch.hflags |= HF_SMM_MASK;
7081 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7083 if (kvm_x86_ops->get_nmi_mask(vcpu))
7084 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7086 kvm_x86_ops->set_nmi_mask(vcpu, true);
7088 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7089 kvm_rip_write(vcpu, 0x8000);
7091 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7092 kvm_x86_ops->set_cr0(vcpu, cr0);
7093 vcpu->arch.cr0 = cr0;
7095 kvm_x86_ops->set_cr4(vcpu, 0);
7097 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7098 dt.address = dt.size = 0;
7099 kvm_x86_ops->set_idt(vcpu, &dt);
7101 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7103 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7104 cs.base = vcpu->arch.smbase;
7109 cs.limit = ds.limit = 0xffffffff;
7110 cs.type = ds.type = 0x3;
7111 cs.dpl = ds.dpl = 0;
7116 cs.avl = ds.avl = 0;
7117 cs.present = ds.present = 1;
7118 cs.unusable = ds.unusable = 0;
7119 cs.padding = ds.padding = 0;
7121 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7122 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7123 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7124 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7125 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7126 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7128 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7129 kvm_x86_ops->set_efer(vcpu, 0);
7131 kvm_update_cpuid(vcpu);
7132 kvm_mmu_reset_context(vcpu);
7135 static void process_smi(struct kvm_vcpu *vcpu)
7137 vcpu->arch.smi_pending = true;
7138 kvm_make_request(KVM_REQ_EVENT, vcpu);
7141 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7143 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7146 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7148 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7151 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7153 if (irqchip_split(vcpu->kvm))
7154 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7156 if (vcpu->arch.apicv_active)
7157 kvm_x86_ops->sync_pir_to_irr(vcpu);
7158 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7161 if (is_guest_mode(vcpu))
7162 vcpu->arch.load_eoi_exitmap_pending = true;
7164 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7167 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7169 u64 eoi_exit_bitmap[4];
7171 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7174 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7175 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7176 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7179 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7180 unsigned long start, unsigned long end)
7182 unsigned long apic_address;
7185 * The physical address of apic access page is stored in the VMCS.
7186 * Update it when it becomes invalid.
7188 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7189 if (start <= apic_address && apic_address < end)
7190 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7193 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7195 struct page *page = NULL;
7197 if (!lapic_in_kernel(vcpu))
7200 if (!kvm_x86_ops->set_apic_access_page_addr)
7203 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7204 if (is_error_page(page))
7206 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7209 * Do not pin apic access page in memory, the MMU notifier
7210 * will call us again if it is migrated or swapped out.
7214 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7217 * Returns 1 to let vcpu_run() continue the guest execution loop without
7218 * exiting to the userspace. Otherwise, the value will be returned to the
7221 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7225 dm_request_for_irq_injection(vcpu) &&
7226 kvm_cpu_accept_dm_intr(vcpu);
7228 bool req_immediate_exit = false;
7230 if (kvm_request_pending(vcpu)) {
7231 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7232 kvm_mmu_unload(vcpu);
7233 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7234 __kvm_migrate_timers(vcpu);
7235 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7236 kvm_gen_update_masterclock(vcpu->kvm);
7237 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7238 kvm_gen_kvmclock_update(vcpu);
7239 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7240 r = kvm_guest_time_update(vcpu);
7244 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7245 kvm_mmu_sync_roots(vcpu);
7246 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7247 kvm_vcpu_flush_tlb(vcpu, true);
7248 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7249 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7253 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7254 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7255 vcpu->mmio_needed = 0;
7259 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7260 /* Page is swapped out. Do synthetic halt */
7261 vcpu->arch.apf.halted = true;
7265 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7266 record_steal_time(vcpu);
7267 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7269 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7271 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7272 kvm_pmu_handle_event(vcpu);
7273 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7274 kvm_pmu_deliver_pmi(vcpu);
7275 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7276 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7277 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7278 vcpu->arch.ioapic_handled_vectors)) {
7279 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7280 vcpu->run->eoi.vector =
7281 vcpu->arch.pending_ioapic_eoi;
7286 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7287 vcpu_scan_ioapic(vcpu);
7288 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7289 vcpu_load_eoi_exitmap(vcpu);
7290 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7291 kvm_vcpu_reload_apic_access_page(vcpu);
7292 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7293 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7294 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7298 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7299 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7300 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7304 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7305 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7306 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7312 * KVM_REQ_HV_STIMER has to be processed after
7313 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7314 * depend on the guest clock being up-to-date
7316 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7317 kvm_hv_process_stimers(vcpu);
7320 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7321 ++vcpu->stat.req_event;
7322 kvm_apic_accept_events(vcpu);
7323 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7328 if (inject_pending_event(vcpu, req_int_win) != 0)
7329 req_immediate_exit = true;
7331 /* Enable SMI/NMI/IRQ window open exits if needed.
7333 * SMIs have three cases:
7334 * 1) They can be nested, and then there is nothing to
7335 * do here because RSM will cause a vmexit anyway.
7336 * 2) There is an ISA-specific reason why SMI cannot be
7337 * injected, and the moment when this changes can be
7339 * 3) Or the SMI can be pending because
7340 * inject_pending_event has completed the injection
7341 * of an IRQ or NMI from the previous vmexit, and
7342 * then we request an immediate exit to inject the
7345 if (vcpu->arch.smi_pending && !is_smm(vcpu))
7346 if (!kvm_x86_ops->enable_smi_window(vcpu))
7347 req_immediate_exit = true;
7348 if (vcpu->arch.nmi_pending)
7349 kvm_x86_ops->enable_nmi_window(vcpu);
7350 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7351 kvm_x86_ops->enable_irq_window(vcpu);
7352 WARN_ON(vcpu->arch.exception.pending);
7355 if (kvm_lapic_enabled(vcpu)) {
7356 update_cr8_intercept(vcpu);
7357 kvm_lapic_sync_to_vapic(vcpu);
7361 r = kvm_mmu_reload(vcpu);
7363 goto cancel_injection;
7368 kvm_x86_ops->prepare_guest_switch(vcpu);
7371 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7372 * IPI are then delayed after guest entry, which ensures that they
7373 * result in virtual interrupt delivery.
7375 local_irq_disable();
7376 vcpu->mode = IN_GUEST_MODE;
7378 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7381 * 1) We should set ->mode before checking ->requests. Please see
7382 * the comment in kvm_vcpu_exiting_guest_mode().
7384 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7385 * pairs with the memory barrier implicit in pi_test_and_set_on
7386 * (see vmx_deliver_posted_interrupt).
7388 * 3) This also orders the write to mode from any reads to the page
7389 * tables done while the VCPU is running. Please see the comment
7390 * in kvm_flush_remote_tlbs.
7392 smp_mb__after_srcu_read_unlock();
7395 * This handles the case where a posted interrupt was
7396 * notified with kvm_vcpu_kick.
7398 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7399 kvm_x86_ops->sync_pir_to_irr(vcpu);
7401 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7402 || need_resched() || signal_pending(current)) {
7403 vcpu->mode = OUTSIDE_GUEST_MODE;
7407 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7409 goto cancel_injection;
7412 kvm_load_guest_xcr0(vcpu);
7414 if (req_immediate_exit) {
7415 kvm_make_request(KVM_REQ_EVENT, vcpu);
7416 smp_send_reschedule(vcpu->cpu);
7419 trace_kvm_entry(vcpu->vcpu_id);
7420 if (lapic_timer_advance_ns)
7421 wait_lapic_expire(vcpu);
7422 guest_enter_irqoff();
7424 if (unlikely(vcpu->arch.switch_db_regs)) {
7426 set_debugreg(vcpu->arch.eff_db[0], 0);
7427 set_debugreg(vcpu->arch.eff_db[1], 1);
7428 set_debugreg(vcpu->arch.eff_db[2], 2);
7429 set_debugreg(vcpu->arch.eff_db[3], 3);
7430 set_debugreg(vcpu->arch.dr6, 6);
7431 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7434 kvm_x86_ops->run(vcpu);
7437 * Do this here before restoring debug registers on the host. And
7438 * since we do this before handling the vmexit, a DR access vmexit
7439 * can (a) read the correct value of the debug registers, (b) set
7440 * KVM_DEBUGREG_WONT_EXIT again.
7442 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7443 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7444 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7445 kvm_update_dr0123(vcpu);
7446 kvm_update_dr6(vcpu);
7447 kvm_update_dr7(vcpu);
7448 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7452 * If the guest has used debug registers, at least dr7
7453 * will be disabled while returning to the host.
7454 * If we don't have active breakpoints in the host, we don't
7455 * care about the messed up debug address registers. But if
7456 * we have some of them active, restore the old state.
7458 if (hw_breakpoint_active())
7459 hw_breakpoint_restore();
7461 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7463 vcpu->mode = OUTSIDE_GUEST_MODE;
7466 kvm_put_guest_xcr0(vcpu);
7468 kvm_before_interrupt(vcpu);
7469 kvm_x86_ops->handle_external_intr(vcpu);
7470 kvm_after_interrupt(vcpu);
7474 guest_exit_irqoff();
7479 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7482 * Profile KVM exit RIPs:
7484 if (unlikely(prof_on == KVM_PROFILING)) {
7485 unsigned long rip = kvm_rip_read(vcpu);
7486 profile_hit(KVM_PROFILING, (void *)rip);
7489 if (unlikely(vcpu->arch.tsc_always_catchup))
7490 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7492 if (vcpu->arch.apic_attention)
7493 kvm_lapic_sync_from_vapic(vcpu);
7495 vcpu->arch.gpa_available = false;
7496 r = kvm_x86_ops->handle_exit(vcpu);
7500 kvm_x86_ops->cancel_injection(vcpu);
7501 if (unlikely(vcpu->arch.apic_attention))
7502 kvm_lapic_sync_from_vapic(vcpu);
7507 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7509 if (!kvm_arch_vcpu_runnable(vcpu) &&
7510 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7511 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7512 kvm_vcpu_block(vcpu);
7513 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7515 if (kvm_x86_ops->post_block)
7516 kvm_x86_ops->post_block(vcpu);
7518 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7522 kvm_apic_accept_events(vcpu);
7523 switch(vcpu->arch.mp_state) {
7524 case KVM_MP_STATE_HALTED:
7525 vcpu->arch.pv.pv_unhalted = false;
7526 vcpu->arch.mp_state =
7527 KVM_MP_STATE_RUNNABLE;
7528 case KVM_MP_STATE_RUNNABLE:
7529 vcpu->arch.apf.halted = false;
7531 case KVM_MP_STATE_INIT_RECEIVED:
7540 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7542 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7543 kvm_x86_ops->check_nested_events(vcpu, false);
7545 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7546 !vcpu->arch.apf.halted);
7549 static int vcpu_run(struct kvm_vcpu *vcpu)
7552 struct kvm *kvm = vcpu->kvm;
7554 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7557 if (kvm_vcpu_running(vcpu)) {
7558 r = vcpu_enter_guest(vcpu);
7560 r = vcpu_block(kvm, vcpu);
7566 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7567 if (kvm_cpu_has_pending_timer(vcpu))
7568 kvm_inject_pending_timer_irqs(vcpu);
7570 if (dm_request_for_irq_injection(vcpu) &&
7571 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7573 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7574 ++vcpu->stat.request_irq_exits;
7578 kvm_check_async_pf_completion(vcpu);
7580 if (signal_pending(current)) {
7582 vcpu->run->exit_reason = KVM_EXIT_INTR;
7583 ++vcpu->stat.signal_exits;
7586 if (need_resched()) {
7587 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7589 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7593 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7598 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7601 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7602 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7603 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7604 if (r != EMULATE_DONE)
7609 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7611 BUG_ON(!vcpu->arch.pio.count);
7613 return complete_emulated_io(vcpu);
7617 * Implements the following, as a state machine:
7621 * for each mmio piece in the fragment
7629 * for each mmio piece in the fragment
7634 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7636 struct kvm_run *run = vcpu->run;
7637 struct kvm_mmio_fragment *frag;
7640 BUG_ON(!vcpu->mmio_needed);
7642 /* Complete previous fragment */
7643 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7644 len = min(8u, frag->len);
7645 if (!vcpu->mmio_is_write)
7646 memcpy(frag->data, run->mmio.data, len);
7648 if (frag->len <= 8) {
7649 /* Switch to the next fragment. */
7651 vcpu->mmio_cur_fragment++;
7653 /* Go forward to the next mmio piece. */
7659 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7660 vcpu->mmio_needed = 0;
7662 /* FIXME: return into emulator if single-stepping. */
7663 if (vcpu->mmio_is_write)
7665 vcpu->mmio_read_completed = 1;
7666 return complete_emulated_io(vcpu);
7669 run->exit_reason = KVM_EXIT_MMIO;
7670 run->mmio.phys_addr = frag->gpa;
7671 if (vcpu->mmio_is_write)
7672 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7673 run->mmio.len = min(8u, frag->len);
7674 run->mmio.is_write = vcpu->mmio_is_write;
7675 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7679 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7684 kvm_sigset_activate(vcpu);
7685 kvm_load_guest_fpu(vcpu);
7687 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7688 if (kvm_run->immediate_exit) {
7692 kvm_vcpu_block(vcpu);
7693 kvm_apic_accept_events(vcpu);
7694 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7696 if (signal_pending(current)) {
7698 vcpu->run->exit_reason = KVM_EXIT_INTR;
7699 ++vcpu->stat.signal_exits;
7704 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
7709 if (vcpu->run->kvm_dirty_regs) {
7710 r = sync_regs(vcpu);
7715 /* re-sync apic's tpr */
7716 if (!lapic_in_kernel(vcpu)) {
7717 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7723 if (unlikely(vcpu->arch.complete_userspace_io)) {
7724 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7725 vcpu->arch.complete_userspace_io = NULL;
7730 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7732 if (kvm_run->immediate_exit)
7738 kvm_put_guest_fpu(vcpu);
7739 if (vcpu->run->kvm_valid_regs)
7741 post_kvm_run_save(vcpu);
7742 kvm_sigset_deactivate(vcpu);
7748 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7750 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7752 * We are here if userspace calls get_regs() in the middle of
7753 * instruction emulation. Registers state needs to be copied
7754 * back from emulation context to vcpu. Userspace shouldn't do
7755 * that usually, but some bad designed PV devices (vmware
7756 * backdoor interface) need this to work
7758 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7759 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7761 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7762 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7763 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7764 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7765 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7766 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7767 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7768 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7769 #ifdef CONFIG_X86_64
7770 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7771 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7772 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7773 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7774 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7775 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7776 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7777 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7780 regs->rip = kvm_rip_read(vcpu);
7781 regs->rflags = kvm_get_rflags(vcpu);
7784 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7787 __get_regs(vcpu, regs);
7792 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7794 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7795 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7797 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7798 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7799 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7800 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7801 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7802 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7803 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7804 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7805 #ifdef CONFIG_X86_64
7806 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7807 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7808 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7809 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7810 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7811 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7812 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7813 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7816 kvm_rip_write(vcpu, regs->rip);
7817 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
7819 vcpu->arch.exception.pending = false;
7821 kvm_make_request(KVM_REQ_EVENT, vcpu);
7824 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7827 __set_regs(vcpu, regs);
7832 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7834 struct kvm_segment cs;
7836 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7840 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7842 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7846 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7847 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7848 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7849 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7850 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7851 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7853 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7854 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7856 kvm_x86_ops->get_idt(vcpu, &dt);
7857 sregs->idt.limit = dt.size;
7858 sregs->idt.base = dt.address;
7859 kvm_x86_ops->get_gdt(vcpu, &dt);
7860 sregs->gdt.limit = dt.size;
7861 sregs->gdt.base = dt.address;
7863 sregs->cr0 = kvm_read_cr0(vcpu);
7864 sregs->cr2 = vcpu->arch.cr2;
7865 sregs->cr3 = kvm_read_cr3(vcpu);
7866 sregs->cr4 = kvm_read_cr4(vcpu);
7867 sregs->cr8 = kvm_get_cr8(vcpu);
7868 sregs->efer = vcpu->arch.efer;
7869 sregs->apic_base = kvm_get_apic_base(vcpu);
7871 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7873 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
7874 set_bit(vcpu->arch.interrupt.nr,
7875 (unsigned long *)sregs->interrupt_bitmap);
7878 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7879 struct kvm_sregs *sregs)
7882 __get_sregs(vcpu, sregs);
7887 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7888 struct kvm_mp_state *mp_state)
7892 kvm_apic_accept_events(vcpu);
7893 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7894 vcpu->arch.pv.pv_unhalted)
7895 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7897 mp_state->mp_state = vcpu->arch.mp_state;
7903 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7904 struct kvm_mp_state *mp_state)
7910 if (!lapic_in_kernel(vcpu) &&
7911 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7914 /* INITs are latched while in SMM */
7915 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7916 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7917 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7920 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7921 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7922 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7924 vcpu->arch.mp_state = mp_state->mp_state;
7925 kvm_make_request(KVM_REQ_EVENT, vcpu);
7933 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7934 int reason, bool has_error_code, u32 error_code)
7936 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7939 init_emulate_ctxt(vcpu);
7941 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7942 has_error_code, error_code);
7945 return EMULATE_FAIL;
7947 kvm_rip_write(vcpu, ctxt->eip);
7948 kvm_set_rflags(vcpu, ctxt->eflags);
7949 kvm_make_request(KVM_REQ_EVENT, vcpu);
7950 return EMULATE_DONE;
7952 EXPORT_SYMBOL_GPL(kvm_task_switch);
7954 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7956 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
7958 * When EFER.LME and CR0.PG are set, the processor is in
7959 * 64-bit mode (though maybe in a 32-bit code segment).
7960 * CR4.PAE and EFER.LMA must be set.
7962 if (!(sregs->cr4 & X86_CR4_PAE)
7963 || !(sregs->efer & EFER_LMA))
7967 * Not in 64-bit mode: EFER.LMA is clear and the code
7968 * segment cannot be 64-bit.
7970 if (sregs->efer & EFER_LMA || sregs->cs.l)
7977 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7979 struct msr_data apic_base_msr;
7980 int mmu_reset_needed = 0;
7981 int cpuid_update_needed = 0;
7982 int pending_vec, max_bits, idx;
7986 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7987 (sregs->cr4 & X86_CR4_OSXSAVE))
7990 if (kvm_valid_sregs(vcpu, sregs))
7993 apic_base_msr.data = sregs->apic_base;
7994 apic_base_msr.host_initiated = true;
7995 if (kvm_set_apic_base(vcpu, &apic_base_msr))
7998 dt.size = sregs->idt.limit;
7999 dt.address = sregs->idt.base;
8000 kvm_x86_ops->set_idt(vcpu, &dt);
8001 dt.size = sregs->gdt.limit;
8002 dt.address = sregs->gdt.base;
8003 kvm_x86_ops->set_gdt(vcpu, &dt);
8005 vcpu->arch.cr2 = sregs->cr2;
8006 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8007 vcpu->arch.cr3 = sregs->cr3;
8008 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8010 kvm_set_cr8(vcpu, sregs->cr8);
8012 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8013 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8015 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8016 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8017 vcpu->arch.cr0 = sregs->cr0;
8019 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8020 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8021 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8022 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8023 if (cpuid_update_needed)
8024 kvm_update_cpuid(vcpu);
8026 idx = srcu_read_lock(&vcpu->kvm->srcu);
8027 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
8028 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8029 mmu_reset_needed = 1;
8031 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8033 if (mmu_reset_needed)
8034 kvm_mmu_reset_context(vcpu);
8036 max_bits = KVM_NR_INTERRUPTS;
8037 pending_vec = find_first_bit(
8038 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8039 if (pending_vec < max_bits) {
8040 kvm_queue_interrupt(vcpu, pending_vec, false);
8041 pr_debug("Set back pending irq %d\n", pending_vec);
8044 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8045 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8046 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8047 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8048 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8049 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8051 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8052 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8054 update_cr8_intercept(vcpu);
8056 /* Older userspace won't unhalt the vcpu on reset. */
8057 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8058 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8060 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8062 kvm_make_request(KVM_REQ_EVENT, vcpu);
8069 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8070 struct kvm_sregs *sregs)
8075 ret = __set_sregs(vcpu, sregs);
8080 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8081 struct kvm_guest_debug *dbg)
8083 unsigned long rflags;
8088 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8090 if (vcpu->arch.exception.pending)
8092 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8093 kvm_queue_exception(vcpu, DB_VECTOR);
8095 kvm_queue_exception(vcpu, BP_VECTOR);
8099 * Read rflags as long as potentially injected trace flags are still
8102 rflags = kvm_get_rflags(vcpu);
8104 vcpu->guest_debug = dbg->control;
8105 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8106 vcpu->guest_debug = 0;
8108 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8109 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8110 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8111 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8113 for (i = 0; i < KVM_NR_DB_REGS; i++)
8114 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8116 kvm_update_dr7(vcpu);
8118 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8119 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8120 get_segment_base(vcpu, VCPU_SREG_CS);
8123 * Trigger an rflags update that will inject or remove the trace
8126 kvm_set_rflags(vcpu, rflags);
8128 kvm_x86_ops->update_bp_intercept(vcpu);
8138 * Translate a guest virtual address to a guest physical address.
8140 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8141 struct kvm_translation *tr)
8143 unsigned long vaddr = tr->linear_address;
8149 idx = srcu_read_lock(&vcpu->kvm->srcu);
8150 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8151 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8152 tr->physical_address = gpa;
8153 tr->valid = gpa != UNMAPPED_GVA;
8161 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8163 struct fxregs_state *fxsave;
8167 fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8168 memcpy(fpu->fpr, fxsave->st_space, 128);
8169 fpu->fcw = fxsave->cwd;
8170 fpu->fsw = fxsave->swd;
8171 fpu->ftwx = fxsave->twd;
8172 fpu->last_opcode = fxsave->fop;
8173 fpu->last_ip = fxsave->rip;
8174 fpu->last_dp = fxsave->rdp;
8175 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
8181 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8183 struct fxregs_state *fxsave;
8187 fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8189 memcpy(fxsave->st_space, fpu->fpr, 128);
8190 fxsave->cwd = fpu->fcw;
8191 fxsave->swd = fpu->fsw;
8192 fxsave->twd = fpu->ftwx;
8193 fxsave->fop = fpu->last_opcode;
8194 fxsave->rip = fpu->last_ip;
8195 fxsave->rdp = fpu->last_dp;
8196 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
8202 static void store_regs(struct kvm_vcpu *vcpu)
8204 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8206 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8207 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8209 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8210 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8212 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8213 kvm_vcpu_ioctl_x86_get_vcpu_events(
8214 vcpu, &vcpu->run->s.regs.events);
8217 static int sync_regs(struct kvm_vcpu *vcpu)
8219 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8222 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8223 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8224 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8226 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8227 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8229 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8231 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8232 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8233 vcpu, &vcpu->run->s.regs.events))
8235 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8241 static void fx_init(struct kvm_vcpu *vcpu)
8243 fpstate_init(&vcpu->arch.guest_fpu.state);
8244 if (boot_cpu_has(X86_FEATURE_XSAVES))
8245 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
8246 host_xcr0 | XSTATE_COMPACTION_ENABLED;
8249 * Ensure guest xcr0 is valid for loading
8251 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8253 vcpu->arch.cr0 |= X86_CR0_ET;
8256 /* Swap (qemu) user FPU context for the guest FPU context. */
8257 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8260 copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
8261 /* PKRU is separately restored in kvm_x86_ops->run. */
8262 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
8263 ~XFEATURE_MASK_PKRU);
8268 /* When vcpu_run ends, restore user space FPU context. */
8269 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8272 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
8273 copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
8275 ++vcpu->stat.fpu_reload;
8279 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8281 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8283 kvmclock_reset(vcpu);
8285 kvm_x86_ops->vcpu_free(vcpu);
8286 free_cpumask_var(wbinvd_dirty_mask);
8289 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8292 struct kvm_vcpu *vcpu;
8294 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8295 printk_once(KERN_WARNING
8296 "kvm: SMP vm created on host with unstable TSC; "
8297 "guest TSC will not be reliable\n");
8299 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8304 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8306 kvm_vcpu_mtrr_init(vcpu);
8308 kvm_vcpu_reset(vcpu, false);
8309 kvm_mmu_setup(vcpu);
8314 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8316 struct msr_data msr;
8317 struct kvm *kvm = vcpu->kvm;
8319 kvm_hv_vcpu_postcreate(vcpu);
8321 if (mutex_lock_killable(&vcpu->mutex))
8325 msr.index = MSR_IA32_TSC;
8326 msr.host_initiated = true;
8327 kvm_write_tsc(vcpu, &msr);
8329 mutex_unlock(&vcpu->mutex);
8331 if (!kvmclock_periodic_sync)
8334 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8335 KVMCLOCK_SYNC_PERIOD);
8338 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8340 vcpu->arch.apf.msr_val = 0;
8343 kvm_mmu_unload(vcpu);
8346 kvm_x86_ops->vcpu_free(vcpu);
8349 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8351 kvm_lapic_reset(vcpu, init_event);
8353 vcpu->arch.hflags = 0;
8355 vcpu->arch.smi_pending = 0;
8356 vcpu->arch.smi_count = 0;
8357 atomic_set(&vcpu->arch.nmi_queued, 0);
8358 vcpu->arch.nmi_pending = 0;
8359 vcpu->arch.nmi_injected = false;
8360 kvm_clear_interrupt_queue(vcpu);
8361 kvm_clear_exception_queue(vcpu);
8362 vcpu->arch.exception.pending = false;
8364 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8365 kvm_update_dr0123(vcpu);
8366 vcpu->arch.dr6 = DR6_INIT;
8367 kvm_update_dr6(vcpu);
8368 vcpu->arch.dr7 = DR7_FIXED_1;
8369 kvm_update_dr7(vcpu);
8373 kvm_make_request(KVM_REQ_EVENT, vcpu);
8374 vcpu->arch.apf.msr_val = 0;
8375 vcpu->arch.st.msr_val = 0;
8377 kvmclock_reset(vcpu);
8379 kvm_clear_async_pf_completion_queue(vcpu);
8380 kvm_async_pf_hash_reset(vcpu);
8381 vcpu->arch.apf.halted = false;
8383 if (kvm_mpx_supported()) {
8384 void *mpx_state_buffer;
8387 * To avoid have the INIT path from kvm_apic_has_events() that be
8388 * called with loaded FPU and does not let userspace fix the state.
8391 kvm_put_guest_fpu(vcpu);
8392 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8393 XFEATURE_MASK_BNDREGS);
8394 if (mpx_state_buffer)
8395 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8396 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8397 XFEATURE_MASK_BNDCSR);
8398 if (mpx_state_buffer)
8399 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8401 kvm_load_guest_fpu(vcpu);
8405 kvm_pmu_reset(vcpu);
8406 vcpu->arch.smbase = 0x30000;
8408 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8409 vcpu->arch.msr_misc_features_enables = 0;
8411 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8414 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8415 vcpu->arch.regs_avail = ~0;
8416 vcpu->arch.regs_dirty = ~0;
8418 vcpu->arch.ia32_xss = 0;
8420 kvm_x86_ops->vcpu_reset(vcpu, init_event);
8423 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8425 struct kvm_segment cs;
8427 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8428 cs.selector = vector << 8;
8429 cs.base = vector << 12;
8430 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8431 kvm_rip_write(vcpu, 0);
8434 int kvm_arch_hardware_enable(void)
8437 struct kvm_vcpu *vcpu;
8442 bool stable, backwards_tsc = false;
8444 kvm_shared_msr_cpu_online();
8445 ret = kvm_x86_ops->hardware_enable();
8449 local_tsc = rdtsc();
8450 stable = !kvm_check_tsc_unstable();
8451 list_for_each_entry(kvm, &vm_list, vm_list) {
8452 kvm_for_each_vcpu(i, vcpu, kvm) {
8453 if (!stable && vcpu->cpu == smp_processor_id())
8454 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8455 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8456 backwards_tsc = true;
8457 if (vcpu->arch.last_host_tsc > max_tsc)
8458 max_tsc = vcpu->arch.last_host_tsc;
8464 * Sometimes, even reliable TSCs go backwards. This happens on
8465 * platforms that reset TSC during suspend or hibernate actions, but
8466 * maintain synchronization. We must compensate. Fortunately, we can
8467 * detect that condition here, which happens early in CPU bringup,
8468 * before any KVM threads can be running. Unfortunately, we can't
8469 * bring the TSCs fully up to date with real time, as we aren't yet far
8470 * enough into CPU bringup that we know how much real time has actually
8471 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8472 * variables that haven't been updated yet.
8474 * So we simply find the maximum observed TSC above, then record the
8475 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8476 * the adjustment will be applied. Note that we accumulate
8477 * adjustments, in case multiple suspend cycles happen before some VCPU
8478 * gets a chance to run again. In the event that no KVM threads get a
8479 * chance to run, we will miss the entire elapsed period, as we'll have
8480 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8481 * loose cycle time. This isn't too big a deal, since the loss will be
8482 * uniform across all VCPUs (not to mention the scenario is extremely
8483 * unlikely). It is possible that a second hibernate recovery happens
8484 * much faster than a first, causing the observed TSC here to be
8485 * smaller; this would require additional padding adjustment, which is
8486 * why we set last_host_tsc to the local tsc observed here.
8488 * N.B. - this code below runs only on platforms with reliable TSC,
8489 * as that is the only way backwards_tsc is set above. Also note
8490 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8491 * have the same delta_cyc adjustment applied if backwards_tsc
8492 * is detected. Note further, this adjustment is only done once,
8493 * as we reset last_host_tsc on all VCPUs to stop this from being
8494 * called multiple times (one for each physical CPU bringup).
8496 * Platforms with unreliable TSCs don't have to deal with this, they
8497 * will be compensated by the logic in vcpu_load, which sets the TSC to
8498 * catchup mode. This will catchup all VCPUs to real time, but cannot
8499 * guarantee that they stay in perfect synchronization.
8501 if (backwards_tsc) {
8502 u64 delta_cyc = max_tsc - local_tsc;
8503 list_for_each_entry(kvm, &vm_list, vm_list) {
8504 kvm->arch.backwards_tsc_observed = true;
8505 kvm_for_each_vcpu(i, vcpu, kvm) {
8506 vcpu->arch.tsc_offset_adjustment += delta_cyc;
8507 vcpu->arch.last_host_tsc = local_tsc;
8508 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8512 * We have to disable TSC offset matching.. if you were
8513 * booting a VM while issuing an S4 host suspend....
8514 * you may have some problem. Solving this issue is
8515 * left as an exercise to the reader.
8517 kvm->arch.last_tsc_nsec = 0;
8518 kvm->arch.last_tsc_write = 0;
8525 void kvm_arch_hardware_disable(void)
8527 kvm_x86_ops->hardware_disable();
8528 drop_user_return_notifiers();
8531 int kvm_arch_hardware_setup(void)
8535 r = kvm_x86_ops->hardware_setup();
8539 if (kvm_has_tsc_control) {
8541 * Make sure the user can only configure tsc_khz values that
8542 * fit into a signed integer.
8543 * A min value is not calculated needed because it will always
8544 * be 1 on all machines.
8546 u64 max = min(0x7fffffffULL,
8547 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8548 kvm_max_guest_tsc_khz = max;
8550 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8553 kvm_init_msr_list();
8557 void kvm_arch_hardware_unsetup(void)
8559 kvm_x86_ops->hardware_unsetup();
8562 void kvm_arch_check_processor_compat(void *rtn)
8564 kvm_x86_ops->check_processor_compatibility(rtn);
8567 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8569 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8571 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8573 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8575 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8578 struct static_key kvm_no_apic_vcpu __read_mostly;
8579 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8581 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8586 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8587 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8588 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8589 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8591 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8593 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8598 vcpu->arch.pio_data = page_address(page);
8600 kvm_set_tsc_khz(vcpu, max_tsc_khz);
8602 r = kvm_mmu_create(vcpu);
8604 goto fail_free_pio_data;
8606 if (irqchip_in_kernel(vcpu->kvm)) {
8607 r = kvm_create_lapic(vcpu);
8609 goto fail_mmu_destroy;
8611 static_key_slow_inc(&kvm_no_apic_vcpu);
8613 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8615 if (!vcpu->arch.mce_banks) {
8617 goto fail_free_lapic;
8619 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8621 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8623 goto fail_free_mce_banks;
8628 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8630 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8632 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8634 kvm_async_pf_hash_reset(vcpu);
8637 vcpu->arch.pending_external_vector = -1;
8638 vcpu->arch.preempted_in_kernel = false;
8640 kvm_hv_vcpu_init(vcpu);
8644 fail_free_mce_banks:
8645 kfree(vcpu->arch.mce_banks);
8647 kvm_free_lapic(vcpu);
8649 kvm_mmu_destroy(vcpu);
8651 free_page((unsigned long)vcpu->arch.pio_data);
8656 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8660 kvm_hv_vcpu_uninit(vcpu);
8661 kvm_pmu_destroy(vcpu);
8662 kfree(vcpu->arch.mce_banks);
8663 kvm_free_lapic(vcpu);
8664 idx = srcu_read_lock(&vcpu->kvm->srcu);
8665 kvm_mmu_destroy(vcpu);
8666 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8667 free_page((unsigned long)vcpu->arch.pio_data);
8668 if (!lapic_in_kernel(vcpu))
8669 static_key_slow_dec(&kvm_no_apic_vcpu);
8672 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8674 kvm_x86_ops->sched_in(vcpu, cpu);
8677 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8682 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8683 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8684 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8685 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8686 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8688 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8689 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8690 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8691 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8692 &kvm->arch.irq_sources_bitmap);
8694 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8695 mutex_init(&kvm->arch.apic_map_lock);
8696 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8698 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8699 pvclock_update_vm_gtod_copy(kvm);
8701 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8702 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8704 kvm_hv_init_vm(kvm);
8705 kvm_page_track_init(kvm);
8706 kvm_mmu_init_vm(kvm);
8708 if (kvm_x86_ops->vm_init)
8709 return kvm_x86_ops->vm_init(kvm);
8714 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8717 kvm_mmu_unload(vcpu);
8721 static void kvm_free_vcpus(struct kvm *kvm)
8724 struct kvm_vcpu *vcpu;
8727 * Unpin any mmu pages first.
8729 kvm_for_each_vcpu(i, vcpu, kvm) {
8730 kvm_clear_async_pf_completion_queue(vcpu);
8731 kvm_unload_vcpu_mmu(vcpu);
8733 kvm_for_each_vcpu(i, vcpu, kvm)
8734 kvm_arch_vcpu_free(vcpu);
8736 mutex_lock(&kvm->lock);
8737 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8738 kvm->vcpus[i] = NULL;
8740 atomic_set(&kvm->online_vcpus, 0);
8741 mutex_unlock(&kvm->lock);
8744 void kvm_arch_sync_events(struct kvm *kvm)
8746 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8747 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8751 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8755 struct kvm_memslots *slots = kvm_memslots(kvm);
8756 struct kvm_memory_slot *slot, old;
8758 /* Called with kvm->slots_lock held. */
8759 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8762 slot = id_to_memslot(slots, id);
8768 * MAP_SHARED to prevent internal slot pages from being moved
8771 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8772 MAP_SHARED | MAP_ANONYMOUS, 0);
8773 if (IS_ERR((void *)hva))
8774 return PTR_ERR((void *)hva);
8783 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8784 struct kvm_userspace_memory_region m;
8786 m.slot = id | (i << 16);
8788 m.guest_phys_addr = gpa;
8789 m.userspace_addr = hva;
8790 m.memory_size = size;
8791 r = __kvm_set_memory_region(kvm, &m);
8797 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8801 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8803 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8807 mutex_lock(&kvm->slots_lock);
8808 r = __x86_set_memory_region(kvm, id, gpa, size);
8809 mutex_unlock(&kvm->slots_lock);
8813 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8815 void kvm_arch_destroy_vm(struct kvm *kvm)
8817 if (current->mm == kvm->mm) {
8819 * Free memory regions allocated on behalf of userspace,
8820 * unless the the memory map has changed due to process exit
8823 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8824 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8825 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8827 if (kvm_x86_ops->vm_destroy)
8828 kvm_x86_ops->vm_destroy(kvm);
8829 kvm_pic_destroy(kvm);
8830 kvm_ioapic_destroy(kvm);
8831 kvm_free_vcpus(kvm);
8832 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8833 kvm_mmu_uninit_vm(kvm);
8834 kvm_page_track_cleanup(kvm);
8835 kvm_hv_destroy_vm(kvm);
8838 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8839 struct kvm_memory_slot *dont)
8843 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8844 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8845 kvfree(free->arch.rmap[i]);
8846 free->arch.rmap[i] = NULL;
8851 if (!dont || free->arch.lpage_info[i - 1] !=
8852 dont->arch.lpage_info[i - 1]) {
8853 kvfree(free->arch.lpage_info[i - 1]);
8854 free->arch.lpage_info[i - 1] = NULL;
8858 kvm_page_track_free_memslot(free, dont);
8861 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8862 unsigned long npages)
8866 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8867 struct kvm_lpage_info *linfo;
8872 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8873 slot->base_gfn, level) + 1;
8875 slot->arch.rmap[i] =
8876 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8877 if (!slot->arch.rmap[i])
8882 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8886 slot->arch.lpage_info[i - 1] = linfo;
8888 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8889 linfo[0].disallow_lpage = 1;
8890 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8891 linfo[lpages - 1].disallow_lpage = 1;
8892 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8894 * If the gfn and userspace address are not aligned wrt each
8895 * other, or if explicitly asked to, disable large page
8896 * support for this slot
8898 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8899 !kvm_largepages_enabled()) {
8902 for (j = 0; j < lpages; ++j)
8903 linfo[j].disallow_lpage = 1;
8907 if (kvm_page_track_create_memslot(slot, npages))
8913 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8914 kvfree(slot->arch.rmap[i]);
8915 slot->arch.rmap[i] = NULL;
8919 kvfree(slot->arch.lpage_info[i - 1]);
8920 slot->arch.lpage_info[i - 1] = NULL;
8925 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8928 * memslots->generation has been incremented.
8929 * mmio generation may have reached its maximum value.
8931 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8934 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8935 struct kvm_memory_slot *memslot,
8936 const struct kvm_userspace_memory_region *mem,
8937 enum kvm_mr_change change)
8942 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8943 struct kvm_memory_slot *new)
8945 /* Still write protect RO slot */
8946 if (new->flags & KVM_MEM_READONLY) {
8947 kvm_mmu_slot_remove_write_access(kvm, new);
8952 * Call kvm_x86_ops dirty logging hooks when they are valid.
8954 * kvm_x86_ops->slot_disable_log_dirty is called when:
8956 * - KVM_MR_CREATE with dirty logging is disabled
8957 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8959 * The reason is, in case of PML, we need to set D-bit for any slots
8960 * with dirty logging disabled in order to eliminate unnecessary GPA
8961 * logging in PML buffer (and potential PML buffer full VMEXT). This
8962 * guarantees leaving PML enabled during guest's lifetime won't have
8963 * any additonal overhead from PML when guest is running with dirty
8964 * logging disabled for memory slots.
8966 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8967 * to dirty logging mode.
8969 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8971 * In case of write protect:
8973 * Write protect all pages for dirty logging.
8975 * All the sptes including the large sptes which point to this
8976 * slot are set to readonly. We can not create any new large
8977 * spte on this slot until the end of the logging.
8979 * See the comments in fast_page_fault().
8981 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8982 if (kvm_x86_ops->slot_enable_log_dirty)
8983 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8985 kvm_mmu_slot_remove_write_access(kvm, new);
8987 if (kvm_x86_ops->slot_disable_log_dirty)
8988 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8992 void kvm_arch_commit_memory_region(struct kvm *kvm,
8993 const struct kvm_userspace_memory_region *mem,
8994 const struct kvm_memory_slot *old,
8995 const struct kvm_memory_slot *new,
8996 enum kvm_mr_change change)
8998 int nr_mmu_pages = 0;
9000 if (!kvm->arch.n_requested_mmu_pages)
9001 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9004 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9007 * Dirty logging tracks sptes in 4k granularity, meaning that large
9008 * sptes have to be split. If live migration is successful, the guest
9009 * in the source machine will be destroyed and large sptes will be
9010 * created in the destination. However, if the guest continues to run
9011 * in the source machine (for example if live migration fails), small
9012 * sptes will remain around and cause bad performance.
9014 * Scan sptes if dirty logging has been stopped, dropping those
9015 * which can be collapsed into a single large-page spte. Later
9016 * page faults will create the large-page sptes.
9018 if ((change != KVM_MR_DELETE) &&
9019 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9020 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9021 kvm_mmu_zap_collapsible_sptes(kvm, new);
9024 * Set up write protection and/or dirty logging for the new slot.
9026 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9027 * been zapped so no dirty logging staff is needed for old slot. For
9028 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9029 * new and it's also covered when dealing with the new slot.
9031 * FIXME: const-ify all uses of struct kvm_memory_slot.
9033 if (change != KVM_MR_DELETE)
9034 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9037 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9039 kvm_mmu_invalidate_zap_all_pages(kvm);
9042 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9043 struct kvm_memory_slot *slot)
9045 kvm_page_track_flush_slot(kvm, slot);
9048 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9050 if (!list_empty_careful(&vcpu->async_pf.done))
9053 if (kvm_apic_has_events(vcpu))
9056 if (vcpu->arch.pv.pv_unhalted)
9059 if (vcpu->arch.exception.pending)
9062 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9063 (vcpu->arch.nmi_pending &&
9064 kvm_x86_ops->nmi_allowed(vcpu)))
9067 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9068 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9071 if (kvm_arch_interrupt_allowed(vcpu) &&
9072 kvm_cpu_has_interrupt(vcpu))
9075 if (kvm_hv_has_stimer_pending(vcpu))
9081 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9083 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9086 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9088 return vcpu->arch.preempted_in_kernel;
9091 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9093 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9096 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9098 return kvm_x86_ops->interrupt_allowed(vcpu);
9101 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9103 if (is_64_bit_mode(vcpu))
9104 return kvm_rip_read(vcpu);
9105 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9106 kvm_rip_read(vcpu));
9108 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9110 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9112 return kvm_get_linear_rip(vcpu) == linear_rip;
9114 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9116 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9118 unsigned long rflags;
9120 rflags = kvm_x86_ops->get_rflags(vcpu);
9121 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9122 rflags &= ~X86_EFLAGS_TF;
9125 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9127 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9129 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9130 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9131 rflags |= X86_EFLAGS_TF;
9132 kvm_x86_ops->set_rflags(vcpu, rflags);
9135 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9137 __kvm_set_rflags(vcpu, rflags);
9138 kvm_make_request(KVM_REQ_EVENT, vcpu);
9140 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9142 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9146 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
9150 r = kvm_mmu_reload(vcpu);
9154 if (!vcpu->arch.mmu.direct_map &&
9155 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
9158 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
9161 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9163 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9166 static inline u32 kvm_async_pf_next_probe(u32 key)
9168 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9171 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9173 u32 key = kvm_async_pf_hash_fn(gfn);
9175 while (vcpu->arch.apf.gfns[key] != ~0)
9176 key = kvm_async_pf_next_probe(key);
9178 vcpu->arch.apf.gfns[key] = gfn;
9181 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9184 u32 key = kvm_async_pf_hash_fn(gfn);
9186 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9187 (vcpu->arch.apf.gfns[key] != gfn &&
9188 vcpu->arch.apf.gfns[key] != ~0); i++)
9189 key = kvm_async_pf_next_probe(key);
9194 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9196 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9199 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9203 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9205 vcpu->arch.apf.gfns[i] = ~0;
9207 j = kvm_async_pf_next_probe(j);
9208 if (vcpu->arch.apf.gfns[j] == ~0)
9210 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9212 * k lies cyclically in ]i,j]
9214 * |....j i.k.| or |.k..j i...|
9216 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9217 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9222 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9225 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9229 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9232 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9236 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9237 struct kvm_async_pf *work)
9239 struct x86_exception fault;
9241 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9242 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9244 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9245 (vcpu->arch.apf.send_user_only &&
9246 kvm_x86_ops->get_cpl(vcpu) == 0))
9247 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9248 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9249 fault.vector = PF_VECTOR;
9250 fault.error_code_valid = true;
9251 fault.error_code = 0;
9252 fault.nested_page_fault = false;
9253 fault.address = work->arch.token;
9254 fault.async_page_fault = true;
9255 kvm_inject_page_fault(vcpu, &fault);
9259 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9260 struct kvm_async_pf *work)
9262 struct x86_exception fault;
9265 if (work->wakeup_all)
9266 work->arch.token = ~0; /* broadcast wakeup */
9268 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9269 trace_kvm_async_pf_ready(work->arch.token, work->gva);
9271 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9272 !apf_get_user(vcpu, &val)) {
9273 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9274 vcpu->arch.exception.pending &&
9275 vcpu->arch.exception.nr == PF_VECTOR &&
9276 !apf_put_user(vcpu, 0)) {
9277 vcpu->arch.exception.injected = false;
9278 vcpu->arch.exception.pending = false;
9279 vcpu->arch.exception.nr = 0;
9280 vcpu->arch.exception.has_error_code = false;
9281 vcpu->arch.exception.error_code = 0;
9282 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9283 fault.vector = PF_VECTOR;
9284 fault.error_code_valid = true;
9285 fault.error_code = 0;
9286 fault.nested_page_fault = false;
9287 fault.address = work->arch.token;
9288 fault.async_page_fault = true;
9289 kvm_inject_page_fault(vcpu, &fault);
9292 vcpu->arch.apf.halted = false;
9293 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9296 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9298 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9301 return kvm_can_do_async_pf(vcpu);
9304 void kvm_arch_start_assignment(struct kvm *kvm)
9306 atomic_inc(&kvm->arch.assigned_device_count);
9308 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9310 void kvm_arch_end_assignment(struct kvm *kvm)
9312 atomic_dec(&kvm->arch.assigned_device_count);
9314 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9316 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9318 return atomic_read(&kvm->arch.assigned_device_count);
9320 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9322 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9324 atomic_inc(&kvm->arch.noncoherent_dma_count);
9326 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9328 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9330 atomic_dec(&kvm->arch.noncoherent_dma_count);
9332 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9334 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9336 return atomic_read(&kvm->arch.noncoherent_dma_count);
9338 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9340 bool kvm_arch_has_irq_bypass(void)
9342 return kvm_x86_ops->update_pi_irte != NULL;
9345 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9346 struct irq_bypass_producer *prod)
9348 struct kvm_kernel_irqfd *irqfd =
9349 container_of(cons, struct kvm_kernel_irqfd, consumer);
9351 irqfd->producer = prod;
9353 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9354 prod->irq, irqfd->gsi, 1);
9357 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9358 struct irq_bypass_producer *prod)
9361 struct kvm_kernel_irqfd *irqfd =
9362 container_of(cons, struct kvm_kernel_irqfd, consumer);
9364 WARN_ON(irqfd->producer != prod);
9365 irqfd->producer = NULL;
9368 * When producer of consumer is unregistered, we change back to
9369 * remapped mode, so we can re-use the current implementation
9370 * when the irq is masked/disabled or the consumer side (KVM
9371 * int this case doesn't want to receive the interrupts.
9373 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9375 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9376 " fails: %d\n", irqfd->consumer.token, ret);
9379 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9380 uint32_t guest_irq, bool set)
9382 if (!kvm_x86_ops->update_pi_irte)
9385 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9388 bool kvm_vector_hashing_enabled(void)
9390 return vector_hashing;
9392 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9394 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9395 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9396 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9397 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9398 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9399 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9400 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9401 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9402 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9403 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9404 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9405 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9406 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9407 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9408 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9409 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9410 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9411 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9412 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);