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,
1066 static unsigned num_emulated_msrs;
1069 * List of msr numbers which are used to expose MSR-based features that
1070 * can be used by a hypervisor to validate requested CPU features.
1072 static u32 msr_based_features[] = {
1074 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1075 MSR_IA32_VMX_PINBASED_CTLS,
1076 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1077 MSR_IA32_VMX_PROCBASED_CTLS,
1078 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1079 MSR_IA32_VMX_EXIT_CTLS,
1080 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1081 MSR_IA32_VMX_ENTRY_CTLS,
1083 MSR_IA32_VMX_CR0_FIXED0,
1084 MSR_IA32_VMX_CR0_FIXED1,
1085 MSR_IA32_VMX_CR4_FIXED0,
1086 MSR_IA32_VMX_CR4_FIXED1,
1087 MSR_IA32_VMX_VMCS_ENUM,
1088 MSR_IA32_VMX_PROCBASED_CTLS2,
1089 MSR_IA32_VMX_EPT_VPID_CAP,
1090 MSR_IA32_VMX_VMFUNC,
1096 static unsigned int num_msr_based_features;
1098 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1100 switch (msr->index) {
1101 case MSR_IA32_UCODE_REV:
1102 rdmsrl(msr->index, msr->data);
1105 if (kvm_x86_ops->get_msr_feature(msr))
1111 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1113 struct kvm_msr_entry msr;
1117 r = kvm_get_msr_feature(&msr);
1126 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1128 if (efer & efer_reserved_bits)
1131 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1134 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1139 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1141 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1143 u64 old_efer = vcpu->arch.efer;
1145 if (!kvm_valid_efer(vcpu, efer))
1149 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1153 efer |= vcpu->arch.efer & EFER_LMA;
1155 kvm_x86_ops->set_efer(vcpu, efer);
1157 /* Update reserved bits */
1158 if ((efer ^ old_efer) & EFER_NX)
1159 kvm_mmu_reset_context(vcpu);
1164 void kvm_enable_efer_bits(u64 mask)
1166 efer_reserved_bits &= ~mask;
1168 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1171 * Writes msr value into into the appropriate "register".
1172 * Returns 0 on success, non-0 otherwise.
1173 * Assumes vcpu_load() was already called.
1175 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1177 switch (msr->index) {
1180 case MSR_KERNEL_GS_BASE:
1183 if (is_noncanonical_address(msr->data, vcpu))
1186 case MSR_IA32_SYSENTER_EIP:
1187 case MSR_IA32_SYSENTER_ESP:
1189 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1190 * non-canonical address is written on Intel but not on
1191 * AMD (which ignores the top 32-bits, because it does
1192 * not implement 64-bit SYSENTER).
1194 * 64-bit code should hence be able to write a non-canonical
1195 * value on AMD. Making the address canonical ensures that
1196 * vmentry does not fail on Intel after writing a non-canonical
1197 * value, and that something deterministic happens if the guest
1198 * invokes 64-bit SYSENTER.
1200 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1202 return kvm_x86_ops->set_msr(vcpu, msr);
1204 EXPORT_SYMBOL_GPL(kvm_set_msr);
1207 * Adapt set_msr() to msr_io()'s calling convention
1209 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1211 struct msr_data msr;
1215 msr.host_initiated = true;
1216 r = kvm_get_msr(vcpu, &msr);
1224 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1226 struct msr_data msr;
1230 msr.host_initiated = true;
1231 return kvm_set_msr(vcpu, &msr);
1234 #ifdef CONFIG_X86_64
1235 struct pvclock_gtod_data {
1238 struct { /* extract of a clocksource struct */
1251 static struct pvclock_gtod_data pvclock_gtod_data;
1253 static void update_pvclock_gtod(struct timekeeper *tk)
1255 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1258 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1260 write_seqcount_begin(&vdata->seq);
1262 /* copy pvclock gtod data */
1263 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1264 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1265 vdata->clock.mask = tk->tkr_mono.mask;
1266 vdata->clock.mult = tk->tkr_mono.mult;
1267 vdata->clock.shift = tk->tkr_mono.shift;
1269 vdata->boot_ns = boot_ns;
1270 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1272 vdata->wall_time_sec = tk->xtime_sec;
1274 write_seqcount_end(&vdata->seq);
1278 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1281 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1282 * vcpu_enter_guest. This function is only called from
1283 * the physical CPU that is running vcpu.
1285 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1288 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1292 struct pvclock_wall_clock wc;
1293 struct timespec64 boot;
1298 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1303 ++version; /* first time write, random junk */
1307 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1311 * The guest calculates current wall clock time by adding
1312 * system time (updated by kvm_guest_time_update below) to the
1313 * wall clock specified here. guest system time equals host
1314 * system time for us, thus we must fill in host boot time here.
1316 getboottime64(&boot);
1318 if (kvm->arch.kvmclock_offset) {
1319 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1320 boot = timespec64_sub(boot, ts);
1322 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1323 wc.nsec = boot.tv_nsec;
1324 wc.version = version;
1326 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1329 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1332 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1334 do_shl32_div32(dividend, divisor);
1338 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1339 s8 *pshift, u32 *pmultiplier)
1347 scaled64 = scaled_hz;
1348 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1353 tps32 = (uint32_t)tps64;
1354 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1355 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1363 *pmultiplier = div_frac(scaled64, tps32);
1365 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1366 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1369 #ifdef CONFIG_X86_64
1370 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1373 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1374 static unsigned long max_tsc_khz;
1376 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1378 u64 v = (u64)khz * (1000000 + ppm);
1383 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1387 /* Guest TSC same frequency as host TSC? */
1389 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1393 /* TSC scaling supported? */
1394 if (!kvm_has_tsc_control) {
1395 if (user_tsc_khz > tsc_khz) {
1396 vcpu->arch.tsc_catchup = 1;
1397 vcpu->arch.tsc_always_catchup = 1;
1400 WARN(1, "user requested TSC rate below hardware speed\n");
1405 /* TSC scaling required - calculate ratio */
1406 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1407 user_tsc_khz, tsc_khz);
1409 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1410 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1415 vcpu->arch.tsc_scaling_ratio = ratio;
1419 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1421 u32 thresh_lo, thresh_hi;
1422 int use_scaling = 0;
1424 /* tsc_khz can be zero if TSC calibration fails */
1425 if (user_tsc_khz == 0) {
1426 /* set tsc_scaling_ratio to a safe value */
1427 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1431 /* Compute a scale to convert nanoseconds in TSC cycles */
1432 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1433 &vcpu->arch.virtual_tsc_shift,
1434 &vcpu->arch.virtual_tsc_mult);
1435 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1438 * Compute the variation in TSC rate which is acceptable
1439 * within the range of tolerance and decide if the
1440 * rate being applied is within that bounds of the hardware
1441 * rate. If so, no scaling or compensation need be done.
1443 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1444 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1445 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1446 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1449 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1452 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1454 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1455 vcpu->arch.virtual_tsc_mult,
1456 vcpu->arch.virtual_tsc_shift);
1457 tsc += vcpu->arch.this_tsc_write;
1461 static inline int gtod_is_based_on_tsc(int mode)
1463 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1466 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1468 #ifdef CONFIG_X86_64
1470 struct kvm_arch *ka = &vcpu->kvm->arch;
1471 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1473 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1474 atomic_read(&vcpu->kvm->online_vcpus));
1477 * Once the masterclock is enabled, always perform request in
1478 * order to update it.
1480 * In order to enable masterclock, the host clocksource must be TSC
1481 * and the vcpus need to have matched TSCs. When that happens,
1482 * perform request to enable masterclock.
1484 if (ka->use_master_clock ||
1485 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1486 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1488 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1489 atomic_read(&vcpu->kvm->online_vcpus),
1490 ka->use_master_clock, gtod->clock.vclock_mode);
1494 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1496 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1497 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1501 * Multiply tsc by a fixed point number represented by ratio.
1503 * The most significant 64-N bits (mult) of ratio represent the
1504 * integral part of the fixed point number; the remaining N bits
1505 * (frac) represent the fractional part, ie. ratio represents a fixed
1506 * point number (mult + frac * 2^(-N)).
1508 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1510 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1512 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1515 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1518 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1520 if (ratio != kvm_default_tsc_scaling_ratio)
1521 _tsc = __scale_tsc(ratio, tsc);
1525 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1527 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1531 tsc = kvm_scale_tsc(vcpu, rdtsc());
1533 return target_tsc - tsc;
1536 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1538 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1540 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1542 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1544 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1546 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1547 vcpu->arch.tsc_offset = offset;
1550 static inline bool kvm_check_tsc_unstable(void)
1552 #ifdef CONFIG_X86_64
1554 * TSC is marked unstable when we're running on Hyper-V,
1555 * 'TSC page' clocksource is good.
1557 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1560 return check_tsc_unstable();
1563 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1565 struct kvm *kvm = vcpu->kvm;
1566 u64 offset, ns, elapsed;
1567 unsigned long flags;
1569 bool already_matched;
1570 u64 data = msr->data;
1571 bool synchronizing = false;
1573 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1574 offset = kvm_compute_tsc_offset(vcpu, data);
1575 ns = ktime_get_boot_ns();
1576 elapsed = ns - kvm->arch.last_tsc_nsec;
1578 if (vcpu->arch.virtual_tsc_khz) {
1579 if (data == 0 && msr->host_initiated) {
1581 * detection of vcpu initialization -- need to sync
1582 * with other vCPUs. This particularly helps to keep
1583 * kvm_clock stable after CPU hotplug
1585 synchronizing = true;
1587 u64 tsc_exp = kvm->arch.last_tsc_write +
1588 nsec_to_cycles(vcpu, elapsed);
1589 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1591 * Special case: TSC write with a small delta (1 second)
1592 * of virtual cycle time against real time is
1593 * interpreted as an attempt to synchronize the CPU.
1595 synchronizing = data < tsc_exp + tsc_hz &&
1596 data + tsc_hz > tsc_exp;
1601 * For a reliable TSC, we can match TSC offsets, and for an unstable
1602 * TSC, we add elapsed time in this computation. We could let the
1603 * compensation code attempt to catch up if we fall behind, but
1604 * it's better to try to match offsets from the beginning.
1606 if (synchronizing &&
1607 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1608 if (!kvm_check_tsc_unstable()) {
1609 offset = kvm->arch.cur_tsc_offset;
1610 pr_debug("kvm: matched tsc offset for %llu\n", data);
1612 u64 delta = nsec_to_cycles(vcpu, elapsed);
1614 offset = kvm_compute_tsc_offset(vcpu, data);
1615 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1618 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1621 * We split periods of matched TSC writes into generations.
1622 * For each generation, we track the original measured
1623 * nanosecond time, offset, and write, so if TSCs are in
1624 * sync, we can match exact offset, and if not, we can match
1625 * exact software computation in compute_guest_tsc()
1627 * These values are tracked in kvm->arch.cur_xxx variables.
1629 kvm->arch.cur_tsc_generation++;
1630 kvm->arch.cur_tsc_nsec = ns;
1631 kvm->arch.cur_tsc_write = data;
1632 kvm->arch.cur_tsc_offset = offset;
1634 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1635 kvm->arch.cur_tsc_generation, data);
1639 * We also track th most recent recorded KHZ, write and time to
1640 * allow the matching interval to be extended at each write.
1642 kvm->arch.last_tsc_nsec = ns;
1643 kvm->arch.last_tsc_write = data;
1644 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1646 vcpu->arch.last_guest_tsc = data;
1648 /* Keep track of which generation this VCPU has synchronized to */
1649 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1650 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1651 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1653 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1654 update_ia32_tsc_adjust_msr(vcpu, offset);
1656 kvm_vcpu_write_tsc_offset(vcpu, offset);
1657 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1659 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1661 kvm->arch.nr_vcpus_matched_tsc = 0;
1662 } else if (!already_matched) {
1663 kvm->arch.nr_vcpus_matched_tsc++;
1666 kvm_track_tsc_matching(vcpu);
1667 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1670 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1672 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1675 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1678 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1680 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1681 WARN_ON(adjustment < 0);
1682 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1683 adjust_tsc_offset_guest(vcpu, adjustment);
1686 #ifdef CONFIG_X86_64
1688 static u64 read_tsc(void)
1690 u64 ret = (u64)rdtsc_ordered();
1691 u64 last = pvclock_gtod_data.clock.cycle_last;
1693 if (likely(ret >= last))
1697 * GCC likes to generate cmov here, but this branch is extremely
1698 * predictable (it's just a function of time and the likely is
1699 * very likely) and there's a data dependence, so force GCC
1700 * to generate a branch instead. I don't barrier() because
1701 * we don't actually need a barrier, and if this function
1702 * ever gets inlined it will generate worse code.
1708 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1711 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1714 switch (gtod->clock.vclock_mode) {
1715 case VCLOCK_HVCLOCK:
1716 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1718 if (tsc_pg_val != U64_MAX) {
1719 /* TSC page valid */
1720 *mode = VCLOCK_HVCLOCK;
1721 v = (tsc_pg_val - gtod->clock.cycle_last) &
1724 /* TSC page invalid */
1725 *mode = VCLOCK_NONE;
1730 *tsc_timestamp = read_tsc();
1731 v = (*tsc_timestamp - gtod->clock.cycle_last) &
1735 *mode = VCLOCK_NONE;
1738 if (*mode == VCLOCK_NONE)
1739 *tsc_timestamp = v = 0;
1741 return v * gtod->clock.mult;
1744 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1746 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1752 seq = read_seqcount_begin(>od->seq);
1753 ns = gtod->nsec_base;
1754 ns += vgettsc(tsc_timestamp, &mode);
1755 ns >>= gtod->clock.shift;
1756 ns += gtod->boot_ns;
1757 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1763 static int do_realtime(struct timespec *ts, u64 *tsc_timestamp)
1765 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1771 seq = read_seqcount_begin(>od->seq);
1772 ts->tv_sec = gtod->wall_time_sec;
1773 ns = gtod->nsec_base;
1774 ns += vgettsc(tsc_timestamp, &mode);
1775 ns >>= gtod->clock.shift;
1776 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1778 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1784 /* returns true if host is using TSC based clocksource */
1785 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1787 /* checked again under seqlock below */
1788 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1791 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1795 /* returns true if host is using TSC based clocksource */
1796 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1799 /* checked again under seqlock below */
1800 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1803 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1809 * Assuming a stable TSC across physical CPUS, and a stable TSC
1810 * across virtual CPUs, the following condition is possible.
1811 * Each numbered line represents an event visible to both
1812 * CPUs at the next numbered event.
1814 * "timespecX" represents host monotonic time. "tscX" represents
1817 * VCPU0 on CPU0 | VCPU1 on CPU1
1819 * 1. read timespec0,tsc0
1820 * 2. | timespec1 = timespec0 + N
1822 * 3. transition to guest | transition to guest
1823 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1824 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1825 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1827 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1830 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1832 * - 0 < N - M => M < N
1834 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1835 * always the case (the difference between two distinct xtime instances
1836 * might be smaller then the difference between corresponding TSC reads,
1837 * when updating guest vcpus pvclock areas).
1839 * To avoid that problem, do not allow visibility of distinct
1840 * system_timestamp/tsc_timestamp values simultaneously: use a master
1841 * copy of host monotonic time values. Update that master copy
1844 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1848 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1850 #ifdef CONFIG_X86_64
1851 struct kvm_arch *ka = &kvm->arch;
1853 bool host_tsc_clocksource, vcpus_matched;
1855 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1856 atomic_read(&kvm->online_vcpus));
1859 * If the host uses TSC clock, then passthrough TSC as stable
1862 host_tsc_clocksource = kvm_get_time_and_clockread(
1863 &ka->master_kernel_ns,
1864 &ka->master_cycle_now);
1866 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1867 && !ka->backwards_tsc_observed
1868 && !ka->boot_vcpu_runs_old_kvmclock;
1870 if (ka->use_master_clock)
1871 atomic_set(&kvm_guest_has_master_clock, 1);
1873 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1874 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1879 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1881 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1884 static void kvm_gen_update_masterclock(struct kvm *kvm)
1886 #ifdef CONFIG_X86_64
1888 struct kvm_vcpu *vcpu;
1889 struct kvm_arch *ka = &kvm->arch;
1891 spin_lock(&ka->pvclock_gtod_sync_lock);
1892 kvm_make_mclock_inprogress_request(kvm);
1893 /* no guest entries from this point */
1894 pvclock_update_vm_gtod_copy(kvm);
1896 kvm_for_each_vcpu(i, vcpu, kvm)
1897 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1899 /* guest entries allowed */
1900 kvm_for_each_vcpu(i, vcpu, kvm)
1901 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1903 spin_unlock(&ka->pvclock_gtod_sync_lock);
1907 u64 get_kvmclock_ns(struct kvm *kvm)
1909 struct kvm_arch *ka = &kvm->arch;
1910 struct pvclock_vcpu_time_info hv_clock;
1913 spin_lock(&ka->pvclock_gtod_sync_lock);
1914 if (!ka->use_master_clock) {
1915 spin_unlock(&ka->pvclock_gtod_sync_lock);
1916 return ktime_get_boot_ns() + ka->kvmclock_offset;
1919 hv_clock.tsc_timestamp = ka->master_cycle_now;
1920 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1921 spin_unlock(&ka->pvclock_gtod_sync_lock);
1923 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1926 if (__this_cpu_read(cpu_tsc_khz)) {
1927 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1928 &hv_clock.tsc_shift,
1929 &hv_clock.tsc_to_system_mul);
1930 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1932 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
1939 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1941 struct kvm_vcpu_arch *vcpu = &v->arch;
1942 struct pvclock_vcpu_time_info guest_hv_clock;
1944 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1945 &guest_hv_clock, sizeof(guest_hv_clock))))
1948 /* This VCPU is paused, but it's legal for a guest to read another
1949 * VCPU's kvmclock, so we really have to follow the specification where
1950 * it says that version is odd if data is being modified, and even after
1953 * Version field updates must be kept separate. This is because
1954 * kvm_write_guest_cached might use a "rep movs" instruction, and
1955 * writes within a string instruction are weakly ordered. So there
1956 * are three writes overall.
1958 * As a small optimization, only write the version field in the first
1959 * and third write. The vcpu->pv_time cache is still valid, because the
1960 * version field is the first in the struct.
1962 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1964 if (guest_hv_clock.version & 1)
1965 ++guest_hv_clock.version; /* first time write, random junk */
1967 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1968 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1970 sizeof(vcpu->hv_clock.version));
1974 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1975 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1977 if (vcpu->pvclock_set_guest_stopped_request) {
1978 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1979 vcpu->pvclock_set_guest_stopped_request = false;
1982 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1984 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1986 sizeof(vcpu->hv_clock));
1990 vcpu->hv_clock.version++;
1991 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1993 sizeof(vcpu->hv_clock.version));
1996 static int kvm_guest_time_update(struct kvm_vcpu *v)
1998 unsigned long flags, tgt_tsc_khz;
1999 struct kvm_vcpu_arch *vcpu = &v->arch;
2000 struct kvm_arch *ka = &v->kvm->arch;
2002 u64 tsc_timestamp, host_tsc;
2004 bool use_master_clock;
2010 * If the host uses TSC clock, then passthrough TSC as stable
2013 spin_lock(&ka->pvclock_gtod_sync_lock);
2014 use_master_clock = ka->use_master_clock;
2015 if (use_master_clock) {
2016 host_tsc = ka->master_cycle_now;
2017 kernel_ns = ka->master_kernel_ns;
2019 spin_unlock(&ka->pvclock_gtod_sync_lock);
2021 /* Keep irq disabled to prevent changes to the clock */
2022 local_irq_save(flags);
2023 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2024 if (unlikely(tgt_tsc_khz == 0)) {
2025 local_irq_restore(flags);
2026 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2029 if (!use_master_clock) {
2031 kernel_ns = ktime_get_boot_ns();
2034 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2037 * We may have to catch up the TSC to match elapsed wall clock
2038 * time for two reasons, even if kvmclock is used.
2039 * 1) CPU could have been running below the maximum TSC rate
2040 * 2) Broken TSC compensation resets the base at each VCPU
2041 * entry to avoid unknown leaps of TSC even when running
2042 * again on the same CPU. This may cause apparent elapsed
2043 * time to disappear, and the guest to stand still or run
2046 if (vcpu->tsc_catchup) {
2047 u64 tsc = compute_guest_tsc(v, kernel_ns);
2048 if (tsc > tsc_timestamp) {
2049 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2050 tsc_timestamp = tsc;
2054 local_irq_restore(flags);
2056 /* With all the info we got, fill in the values */
2058 if (kvm_has_tsc_control)
2059 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2061 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2062 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2063 &vcpu->hv_clock.tsc_shift,
2064 &vcpu->hv_clock.tsc_to_system_mul);
2065 vcpu->hw_tsc_khz = tgt_tsc_khz;
2068 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2069 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2070 vcpu->last_guest_tsc = tsc_timestamp;
2072 /* If the host uses TSC clocksource, then it is stable */
2074 if (use_master_clock)
2075 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2077 vcpu->hv_clock.flags = pvclock_flags;
2079 if (vcpu->pv_time_enabled)
2080 kvm_setup_pvclock_page(v);
2081 if (v == kvm_get_vcpu(v->kvm, 0))
2082 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2087 * kvmclock updates which are isolated to a given vcpu, such as
2088 * vcpu->cpu migration, should not allow system_timestamp from
2089 * the rest of the vcpus to remain static. Otherwise ntp frequency
2090 * correction applies to one vcpu's system_timestamp but not
2093 * So in those cases, request a kvmclock update for all vcpus.
2094 * We need to rate-limit these requests though, as they can
2095 * considerably slow guests that have a large number of vcpus.
2096 * The time for a remote vcpu to update its kvmclock is bound
2097 * by the delay we use to rate-limit the updates.
2100 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2102 static void kvmclock_update_fn(struct work_struct *work)
2105 struct delayed_work *dwork = to_delayed_work(work);
2106 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2107 kvmclock_update_work);
2108 struct kvm *kvm = container_of(ka, struct kvm, arch);
2109 struct kvm_vcpu *vcpu;
2111 kvm_for_each_vcpu(i, vcpu, kvm) {
2112 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2113 kvm_vcpu_kick(vcpu);
2117 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2119 struct kvm *kvm = v->kvm;
2121 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2122 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2123 KVMCLOCK_UPDATE_DELAY);
2126 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2128 static void kvmclock_sync_fn(struct work_struct *work)
2130 struct delayed_work *dwork = to_delayed_work(work);
2131 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2132 kvmclock_sync_work);
2133 struct kvm *kvm = container_of(ka, struct kvm, arch);
2135 if (!kvmclock_periodic_sync)
2138 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2139 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2140 KVMCLOCK_SYNC_PERIOD);
2143 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2145 u64 mcg_cap = vcpu->arch.mcg_cap;
2146 unsigned bank_num = mcg_cap & 0xff;
2147 u32 msr = msr_info->index;
2148 u64 data = msr_info->data;
2151 case MSR_IA32_MCG_STATUS:
2152 vcpu->arch.mcg_status = data;
2154 case MSR_IA32_MCG_CTL:
2155 if (!(mcg_cap & MCG_CTL_P))
2157 if (data != 0 && data != ~(u64)0)
2159 vcpu->arch.mcg_ctl = data;
2162 if (msr >= MSR_IA32_MC0_CTL &&
2163 msr < MSR_IA32_MCx_CTL(bank_num)) {
2164 u32 offset = msr - MSR_IA32_MC0_CTL;
2165 /* only 0 or all 1s can be written to IA32_MCi_CTL
2166 * some Linux kernels though clear bit 10 in bank 4 to
2167 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2168 * this to avoid an uncatched #GP in the guest
2170 if ((offset & 0x3) == 0 &&
2171 data != 0 && (data | (1 << 10)) != ~(u64)0)
2173 if (!msr_info->host_initiated &&
2174 (offset & 0x3) == 1 && data != 0)
2176 vcpu->arch.mce_banks[offset] = data;
2184 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2186 struct kvm *kvm = vcpu->kvm;
2187 int lm = is_long_mode(vcpu);
2188 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2189 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2190 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2191 : kvm->arch.xen_hvm_config.blob_size_32;
2192 u32 page_num = data & ~PAGE_MASK;
2193 u64 page_addr = data & PAGE_MASK;
2198 if (page_num >= blob_size)
2201 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2206 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2215 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2217 gpa_t gpa = data & ~0x3f;
2219 /* Bits 3:5 are reserved, Should be zero */
2223 vcpu->arch.apf.msr_val = data;
2225 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2226 kvm_clear_async_pf_completion_queue(vcpu);
2227 kvm_async_pf_hash_reset(vcpu);
2231 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2235 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2236 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2237 kvm_async_pf_wakeup_all(vcpu);
2241 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2243 vcpu->arch.pv_time_enabled = false;
2246 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2248 ++vcpu->stat.tlb_flush;
2249 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2252 static void record_steal_time(struct kvm_vcpu *vcpu)
2254 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2257 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2258 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2262 * Doing a TLB flush here, on the guest's behalf, can avoid
2265 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2266 kvm_vcpu_flush_tlb(vcpu, false);
2268 if (vcpu->arch.st.steal.version & 1)
2269 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2271 vcpu->arch.st.steal.version += 1;
2273 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2274 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2278 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2279 vcpu->arch.st.last_steal;
2280 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2282 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2283 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2287 vcpu->arch.st.steal.version += 1;
2289 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2290 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2293 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2296 u32 msr = msr_info->index;
2297 u64 data = msr_info->data;
2300 case MSR_AMD64_NB_CFG:
2301 case MSR_IA32_UCODE_WRITE:
2302 case MSR_VM_HSAVE_PA:
2303 case MSR_AMD64_PATCH_LOADER:
2304 case MSR_AMD64_BU_CFG2:
2305 case MSR_AMD64_DC_CFG:
2308 case MSR_IA32_UCODE_REV:
2309 if (msr_info->host_initiated)
2310 vcpu->arch.microcode_version = data;
2313 return set_efer(vcpu, data);
2315 data &= ~(u64)0x40; /* ignore flush filter disable */
2316 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2317 data &= ~(u64)0x8; /* ignore TLB cache disable */
2318 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2320 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2325 case MSR_FAM10H_MMIO_CONF_BASE:
2327 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2332 case MSR_IA32_DEBUGCTLMSR:
2334 /* We support the non-activated case already */
2336 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2337 /* Values other than LBR and BTF are vendor-specific,
2338 thus reserved and should throw a #GP */
2341 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2344 case 0x200 ... 0x2ff:
2345 return kvm_mtrr_set_msr(vcpu, msr, data);
2346 case MSR_IA32_APICBASE:
2347 return kvm_set_apic_base(vcpu, msr_info);
2348 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2349 return kvm_x2apic_msr_write(vcpu, msr, data);
2350 case MSR_IA32_TSCDEADLINE:
2351 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2353 case MSR_IA32_TSC_ADJUST:
2354 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2355 if (!msr_info->host_initiated) {
2356 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2357 adjust_tsc_offset_guest(vcpu, adj);
2359 vcpu->arch.ia32_tsc_adjust_msr = data;
2362 case MSR_IA32_MISC_ENABLE:
2363 vcpu->arch.ia32_misc_enable_msr = data;
2365 case MSR_IA32_SMBASE:
2366 if (!msr_info->host_initiated)
2368 vcpu->arch.smbase = data;
2371 kvm_write_tsc(vcpu, msr_info);
2374 if (!msr_info->host_initiated)
2376 vcpu->arch.smi_count = data;
2378 case MSR_KVM_WALL_CLOCK_NEW:
2379 case MSR_KVM_WALL_CLOCK:
2380 vcpu->kvm->arch.wall_clock = data;
2381 kvm_write_wall_clock(vcpu->kvm, data);
2383 case MSR_KVM_SYSTEM_TIME_NEW:
2384 case MSR_KVM_SYSTEM_TIME: {
2385 struct kvm_arch *ka = &vcpu->kvm->arch;
2387 kvmclock_reset(vcpu);
2389 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2390 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2392 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2393 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2395 ka->boot_vcpu_runs_old_kvmclock = tmp;
2398 vcpu->arch.time = data;
2399 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2401 /* we verify if the enable bit is set... */
2405 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2406 &vcpu->arch.pv_time, data & ~1ULL,
2407 sizeof(struct pvclock_vcpu_time_info)))
2408 vcpu->arch.pv_time_enabled = false;
2410 vcpu->arch.pv_time_enabled = true;
2414 case MSR_KVM_ASYNC_PF_EN:
2415 if (kvm_pv_enable_async_pf(vcpu, data))
2418 case MSR_KVM_STEAL_TIME:
2420 if (unlikely(!sched_info_on()))
2423 if (data & KVM_STEAL_RESERVED_MASK)
2426 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2427 data & KVM_STEAL_VALID_BITS,
2428 sizeof(struct kvm_steal_time)))
2431 vcpu->arch.st.msr_val = data;
2433 if (!(data & KVM_MSR_ENABLED))
2436 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2439 case MSR_KVM_PV_EOI_EN:
2440 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2444 case MSR_IA32_MCG_CTL:
2445 case MSR_IA32_MCG_STATUS:
2446 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2447 return set_msr_mce(vcpu, msr_info);
2449 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2450 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2451 pr = true; /* fall through */
2452 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2453 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2454 if (kvm_pmu_is_valid_msr(vcpu, msr))
2455 return kvm_pmu_set_msr(vcpu, msr_info);
2457 if (pr || data != 0)
2458 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2459 "0x%x data 0x%llx\n", msr, data);
2461 case MSR_K7_CLK_CTL:
2463 * Ignore all writes to this no longer documented MSR.
2464 * Writes are only relevant for old K7 processors,
2465 * all pre-dating SVM, but a recommended workaround from
2466 * AMD for these chips. It is possible to specify the
2467 * affected processor models on the command line, hence
2468 * the need to ignore the workaround.
2471 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2472 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2473 case HV_X64_MSR_CRASH_CTL:
2474 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2475 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2476 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2477 case HV_X64_MSR_TSC_EMULATION_STATUS:
2478 return kvm_hv_set_msr_common(vcpu, msr, data,
2479 msr_info->host_initiated);
2480 case MSR_IA32_BBL_CR_CTL3:
2481 /* Drop writes to this legacy MSR -- see rdmsr
2482 * counterpart for further detail.
2484 if (report_ignored_msrs)
2485 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2488 case MSR_AMD64_OSVW_ID_LENGTH:
2489 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2491 vcpu->arch.osvw.length = data;
2493 case MSR_AMD64_OSVW_STATUS:
2494 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2496 vcpu->arch.osvw.status = data;
2498 case MSR_PLATFORM_INFO:
2499 if (!msr_info->host_initiated ||
2500 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2501 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2502 cpuid_fault_enabled(vcpu)))
2504 vcpu->arch.msr_platform_info = data;
2506 case MSR_MISC_FEATURES_ENABLES:
2507 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2508 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2509 !supports_cpuid_fault(vcpu)))
2511 vcpu->arch.msr_misc_features_enables = data;
2514 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2515 return xen_hvm_config(vcpu, data);
2516 if (kvm_pmu_is_valid_msr(vcpu, msr))
2517 return kvm_pmu_set_msr(vcpu, msr_info);
2519 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2523 if (report_ignored_msrs)
2525 "ignored wrmsr: 0x%x data 0x%llx\n",
2532 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2536 * Reads an msr value (of 'msr_index') into 'pdata'.
2537 * Returns 0 on success, non-0 otherwise.
2538 * Assumes vcpu_load() was already called.
2540 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2542 return kvm_x86_ops->get_msr(vcpu, msr);
2544 EXPORT_SYMBOL_GPL(kvm_get_msr);
2546 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2549 u64 mcg_cap = vcpu->arch.mcg_cap;
2550 unsigned bank_num = mcg_cap & 0xff;
2553 case MSR_IA32_P5_MC_ADDR:
2554 case MSR_IA32_P5_MC_TYPE:
2557 case MSR_IA32_MCG_CAP:
2558 data = vcpu->arch.mcg_cap;
2560 case MSR_IA32_MCG_CTL:
2561 if (!(mcg_cap & MCG_CTL_P))
2563 data = vcpu->arch.mcg_ctl;
2565 case MSR_IA32_MCG_STATUS:
2566 data = vcpu->arch.mcg_status;
2569 if (msr >= MSR_IA32_MC0_CTL &&
2570 msr < MSR_IA32_MCx_CTL(bank_num)) {
2571 u32 offset = msr - MSR_IA32_MC0_CTL;
2572 data = vcpu->arch.mce_banks[offset];
2581 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2583 switch (msr_info->index) {
2584 case MSR_IA32_PLATFORM_ID:
2585 case MSR_IA32_EBL_CR_POWERON:
2586 case MSR_IA32_DEBUGCTLMSR:
2587 case MSR_IA32_LASTBRANCHFROMIP:
2588 case MSR_IA32_LASTBRANCHTOIP:
2589 case MSR_IA32_LASTINTFROMIP:
2590 case MSR_IA32_LASTINTTOIP:
2592 case MSR_K8_TSEG_ADDR:
2593 case MSR_K8_TSEG_MASK:
2595 case MSR_VM_HSAVE_PA:
2596 case MSR_K8_INT_PENDING_MSG:
2597 case MSR_AMD64_NB_CFG:
2598 case MSR_FAM10H_MMIO_CONF_BASE:
2599 case MSR_AMD64_BU_CFG2:
2600 case MSR_IA32_PERF_CTL:
2601 case MSR_AMD64_DC_CFG:
2604 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2605 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2606 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2607 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2608 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2609 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2610 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2613 case MSR_IA32_UCODE_REV:
2614 msr_info->data = vcpu->arch.microcode_version;
2617 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2620 case 0x200 ... 0x2ff:
2621 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2622 case 0xcd: /* fsb frequency */
2626 * MSR_EBC_FREQUENCY_ID
2627 * Conservative value valid for even the basic CPU models.
2628 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2629 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2630 * and 266MHz for model 3, or 4. Set Core Clock
2631 * Frequency to System Bus Frequency Ratio to 1 (bits
2632 * 31:24) even though these are only valid for CPU
2633 * models > 2, however guests may end up dividing or
2634 * multiplying by zero otherwise.
2636 case MSR_EBC_FREQUENCY_ID:
2637 msr_info->data = 1 << 24;
2639 case MSR_IA32_APICBASE:
2640 msr_info->data = kvm_get_apic_base(vcpu);
2642 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2643 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2645 case MSR_IA32_TSCDEADLINE:
2646 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2648 case MSR_IA32_TSC_ADJUST:
2649 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2651 case MSR_IA32_MISC_ENABLE:
2652 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2654 case MSR_IA32_SMBASE:
2655 if (!msr_info->host_initiated)
2657 msr_info->data = vcpu->arch.smbase;
2660 msr_info->data = vcpu->arch.smi_count;
2662 case MSR_IA32_PERF_STATUS:
2663 /* TSC increment by tick */
2664 msr_info->data = 1000ULL;
2665 /* CPU multiplier */
2666 msr_info->data |= (((uint64_t)4ULL) << 40);
2669 msr_info->data = vcpu->arch.efer;
2671 case MSR_KVM_WALL_CLOCK:
2672 case MSR_KVM_WALL_CLOCK_NEW:
2673 msr_info->data = vcpu->kvm->arch.wall_clock;
2675 case MSR_KVM_SYSTEM_TIME:
2676 case MSR_KVM_SYSTEM_TIME_NEW:
2677 msr_info->data = vcpu->arch.time;
2679 case MSR_KVM_ASYNC_PF_EN:
2680 msr_info->data = vcpu->arch.apf.msr_val;
2682 case MSR_KVM_STEAL_TIME:
2683 msr_info->data = vcpu->arch.st.msr_val;
2685 case MSR_KVM_PV_EOI_EN:
2686 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2688 case MSR_IA32_P5_MC_ADDR:
2689 case MSR_IA32_P5_MC_TYPE:
2690 case MSR_IA32_MCG_CAP:
2691 case MSR_IA32_MCG_CTL:
2692 case MSR_IA32_MCG_STATUS:
2693 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2694 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2695 case MSR_K7_CLK_CTL:
2697 * Provide expected ramp-up count for K7. All other
2698 * are set to zero, indicating minimum divisors for
2701 * This prevents guest kernels on AMD host with CPU
2702 * type 6, model 8 and higher from exploding due to
2703 * the rdmsr failing.
2705 msr_info->data = 0x20000000;
2707 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2708 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2709 case HV_X64_MSR_CRASH_CTL:
2710 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2711 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2712 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2713 case HV_X64_MSR_TSC_EMULATION_STATUS:
2714 return kvm_hv_get_msr_common(vcpu,
2715 msr_info->index, &msr_info->data);
2717 case MSR_IA32_BBL_CR_CTL3:
2718 /* This legacy MSR exists but isn't fully documented in current
2719 * silicon. It is however accessed by winxp in very narrow
2720 * scenarios where it sets bit #19, itself documented as
2721 * a "reserved" bit. Best effort attempt to source coherent
2722 * read data here should the balance of the register be
2723 * interpreted by the guest:
2725 * L2 cache control register 3: 64GB range, 256KB size,
2726 * enabled, latency 0x1, configured
2728 msr_info->data = 0xbe702111;
2730 case MSR_AMD64_OSVW_ID_LENGTH:
2731 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2733 msr_info->data = vcpu->arch.osvw.length;
2735 case MSR_AMD64_OSVW_STATUS:
2736 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2738 msr_info->data = vcpu->arch.osvw.status;
2740 case MSR_PLATFORM_INFO:
2741 msr_info->data = vcpu->arch.msr_platform_info;
2743 case MSR_MISC_FEATURES_ENABLES:
2744 msr_info->data = vcpu->arch.msr_misc_features_enables;
2747 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2748 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2750 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2754 if (report_ignored_msrs)
2755 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2763 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2766 * Read or write a bunch of msrs. All parameters are kernel addresses.
2768 * @return number of msrs set successfully.
2770 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2771 struct kvm_msr_entry *entries,
2772 int (*do_msr)(struct kvm_vcpu *vcpu,
2773 unsigned index, u64 *data))
2777 for (i = 0; i < msrs->nmsrs; ++i)
2778 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2785 * Read or write a bunch of msrs. Parameters are user addresses.
2787 * @return number of msrs set successfully.
2789 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2790 int (*do_msr)(struct kvm_vcpu *vcpu,
2791 unsigned index, u64 *data),
2794 struct kvm_msrs msrs;
2795 struct kvm_msr_entry *entries;
2800 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2804 if (msrs.nmsrs >= MAX_IO_MSRS)
2807 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2808 entries = memdup_user(user_msrs->entries, size);
2809 if (IS_ERR(entries)) {
2810 r = PTR_ERR(entries);
2814 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2819 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2830 static inline bool kvm_can_mwait_in_guest(void)
2832 return boot_cpu_has(X86_FEATURE_MWAIT) &&
2833 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
2834 boot_cpu_has(X86_FEATURE_ARAT);
2837 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2842 case KVM_CAP_IRQCHIP:
2844 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2845 case KVM_CAP_SET_TSS_ADDR:
2846 case KVM_CAP_EXT_CPUID:
2847 case KVM_CAP_EXT_EMUL_CPUID:
2848 case KVM_CAP_CLOCKSOURCE:
2850 case KVM_CAP_NOP_IO_DELAY:
2851 case KVM_CAP_MP_STATE:
2852 case KVM_CAP_SYNC_MMU:
2853 case KVM_CAP_USER_NMI:
2854 case KVM_CAP_REINJECT_CONTROL:
2855 case KVM_CAP_IRQ_INJECT_STATUS:
2856 case KVM_CAP_IOEVENTFD:
2857 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2859 case KVM_CAP_PIT_STATE2:
2860 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2861 case KVM_CAP_XEN_HVM:
2862 case KVM_CAP_VCPU_EVENTS:
2863 case KVM_CAP_HYPERV:
2864 case KVM_CAP_HYPERV_VAPIC:
2865 case KVM_CAP_HYPERV_SPIN:
2866 case KVM_CAP_HYPERV_SYNIC:
2867 case KVM_CAP_HYPERV_SYNIC2:
2868 case KVM_CAP_HYPERV_VP_INDEX:
2869 case KVM_CAP_HYPERV_EVENTFD:
2870 case KVM_CAP_PCI_SEGMENT:
2871 case KVM_CAP_DEBUGREGS:
2872 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2874 case KVM_CAP_ASYNC_PF:
2875 case KVM_CAP_GET_TSC_KHZ:
2876 case KVM_CAP_KVMCLOCK_CTRL:
2877 case KVM_CAP_READONLY_MEM:
2878 case KVM_CAP_HYPERV_TIME:
2879 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2880 case KVM_CAP_TSC_DEADLINE_TIMER:
2881 case KVM_CAP_ENABLE_CAP_VM:
2882 case KVM_CAP_DISABLE_QUIRKS:
2883 case KVM_CAP_SET_BOOT_CPU_ID:
2884 case KVM_CAP_SPLIT_IRQCHIP:
2885 case KVM_CAP_IMMEDIATE_EXIT:
2886 case KVM_CAP_GET_MSR_FEATURES:
2889 case KVM_CAP_SYNC_REGS:
2890 r = KVM_SYNC_X86_VALID_FIELDS;
2892 case KVM_CAP_ADJUST_CLOCK:
2893 r = KVM_CLOCK_TSC_STABLE;
2895 case KVM_CAP_X86_DISABLE_EXITS:
2896 r |= KVM_X86_DISABLE_EXITS_HTL | KVM_X86_DISABLE_EXITS_PAUSE;
2897 if(kvm_can_mwait_in_guest())
2898 r |= KVM_X86_DISABLE_EXITS_MWAIT;
2900 case KVM_CAP_X86_SMM:
2901 /* SMBASE is usually relocated above 1M on modern chipsets,
2902 * and SMM handlers might indeed rely on 4G segment limits,
2903 * so do not report SMM to be available if real mode is
2904 * emulated via vm86 mode. Still, do not go to great lengths
2905 * to avoid userspace's usage of the feature, because it is a
2906 * fringe case that is not enabled except via specific settings
2907 * of the module parameters.
2909 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2912 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2914 case KVM_CAP_NR_VCPUS:
2915 r = KVM_SOFT_MAX_VCPUS;
2917 case KVM_CAP_MAX_VCPUS:
2920 case KVM_CAP_NR_MEMSLOTS:
2921 r = KVM_USER_MEM_SLOTS;
2923 case KVM_CAP_PV_MMU: /* obsolete */
2927 r = KVM_MAX_MCE_BANKS;
2930 r = boot_cpu_has(X86_FEATURE_XSAVE);
2932 case KVM_CAP_TSC_CONTROL:
2933 r = kvm_has_tsc_control;
2935 case KVM_CAP_X2APIC_API:
2936 r = KVM_X2APIC_API_VALID_FLAGS;
2945 long kvm_arch_dev_ioctl(struct file *filp,
2946 unsigned int ioctl, unsigned long arg)
2948 void __user *argp = (void __user *)arg;
2952 case KVM_GET_MSR_INDEX_LIST: {
2953 struct kvm_msr_list __user *user_msr_list = argp;
2954 struct kvm_msr_list msr_list;
2958 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2961 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2962 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2965 if (n < msr_list.nmsrs)
2968 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2969 num_msrs_to_save * sizeof(u32)))
2971 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2973 num_emulated_msrs * sizeof(u32)))
2978 case KVM_GET_SUPPORTED_CPUID:
2979 case KVM_GET_EMULATED_CPUID: {
2980 struct kvm_cpuid2 __user *cpuid_arg = argp;
2981 struct kvm_cpuid2 cpuid;
2984 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2987 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2993 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2998 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3000 if (copy_to_user(argp, &kvm_mce_cap_supported,
3001 sizeof(kvm_mce_cap_supported)))
3005 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3006 struct kvm_msr_list __user *user_msr_list = argp;
3007 struct kvm_msr_list msr_list;
3011 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3014 msr_list.nmsrs = num_msr_based_features;
3015 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3018 if (n < msr_list.nmsrs)
3021 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3022 num_msr_based_features * sizeof(u32)))
3028 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3038 static void wbinvd_ipi(void *garbage)
3043 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3045 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3048 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3050 /* Address WBINVD may be executed by guest */
3051 if (need_emulate_wbinvd(vcpu)) {
3052 if (kvm_x86_ops->has_wbinvd_exit())
3053 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3054 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3055 smp_call_function_single(vcpu->cpu,
3056 wbinvd_ipi, NULL, 1);
3059 kvm_x86_ops->vcpu_load(vcpu, cpu);
3061 /* Apply any externally detected TSC adjustments (due to suspend) */
3062 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3063 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3064 vcpu->arch.tsc_offset_adjustment = 0;
3065 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3068 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3069 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3070 rdtsc() - vcpu->arch.last_host_tsc;
3072 mark_tsc_unstable("KVM discovered backwards TSC");
3074 if (kvm_check_tsc_unstable()) {
3075 u64 offset = kvm_compute_tsc_offset(vcpu,
3076 vcpu->arch.last_guest_tsc);
3077 kvm_vcpu_write_tsc_offset(vcpu, offset);
3078 vcpu->arch.tsc_catchup = 1;
3081 if (kvm_lapic_hv_timer_in_use(vcpu))
3082 kvm_lapic_restart_hv_timer(vcpu);
3085 * On a host with synchronized TSC, there is no need to update
3086 * kvmclock on vcpu->cpu migration
3088 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3089 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3090 if (vcpu->cpu != cpu)
3091 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3095 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3098 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3100 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3103 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3105 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3106 &vcpu->arch.st.steal.preempted,
3107 offsetof(struct kvm_steal_time, preempted),
3108 sizeof(vcpu->arch.st.steal.preempted));
3111 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3115 if (vcpu->preempted)
3116 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3119 * Disable page faults because we're in atomic context here.
3120 * kvm_write_guest_offset_cached() would call might_fault()
3121 * that relies on pagefault_disable() to tell if there's a
3122 * bug. NOTE: the write to guest memory may not go through if
3123 * during postcopy live migration or if there's heavy guest
3126 pagefault_disable();
3128 * kvm_memslots() will be called by
3129 * kvm_write_guest_offset_cached() so take the srcu lock.
3131 idx = srcu_read_lock(&vcpu->kvm->srcu);
3132 kvm_steal_time_set_preempted(vcpu);
3133 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3135 kvm_x86_ops->vcpu_put(vcpu);
3136 vcpu->arch.last_host_tsc = rdtsc();
3138 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3139 * on every vmexit, but if not, we might have a stale dr6 from the
3140 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3145 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3146 struct kvm_lapic_state *s)
3148 if (vcpu->arch.apicv_active)
3149 kvm_x86_ops->sync_pir_to_irr(vcpu);
3151 return kvm_apic_get_state(vcpu, s);
3154 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3155 struct kvm_lapic_state *s)
3159 r = kvm_apic_set_state(vcpu, s);
3162 update_cr8_intercept(vcpu);
3167 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3169 return (!lapic_in_kernel(vcpu) ||
3170 kvm_apic_accept_pic_intr(vcpu));
3174 * if userspace requested an interrupt window, check that the
3175 * interrupt window is open.
3177 * No need to exit to userspace if we already have an interrupt queued.
3179 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3181 return kvm_arch_interrupt_allowed(vcpu) &&
3182 !kvm_cpu_has_interrupt(vcpu) &&
3183 !kvm_event_needs_reinjection(vcpu) &&
3184 kvm_cpu_accept_dm_intr(vcpu);
3187 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3188 struct kvm_interrupt *irq)
3190 if (irq->irq >= KVM_NR_INTERRUPTS)
3193 if (!irqchip_in_kernel(vcpu->kvm)) {
3194 kvm_queue_interrupt(vcpu, irq->irq, false);
3195 kvm_make_request(KVM_REQ_EVENT, vcpu);
3200 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3201 * fail for in-kernel 8259.
3203 if (pic_in_kernel(vcpu->kvm))
3206 if (vcpu->arch.pending_external_vector != -1)
3209 vcpu->arch.pending_external_vector = irq->irq;
3210 kvm_make_request(KVM_REQ_EVENT, vcpu);
3214 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3216 kvm_inject_nmi(vcpu);
3221 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3223 kvm_make_request(KVM_REQ_SMI, vcpu);
3228 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3229 struct kvm_tpr_access_ctl *tac)
3233 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3237 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3241 unsigned bank_num = mcg_cap & 0xff, bank;
3244 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3246 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3249 vcpu->arch.mcg_cap = mcg_cap;
3250 /* Init IA32_MCG_CTL to all 1s */
3251 if (mcg_cap & MCG_CTL_P)
3252 vcpu->arch.mcg_ctl = ~(u64)0;
3253 /* Init IA32_MCi_CTL to all 1s */
3254 for (bank = 0; bank < bank_num; bank++)
3255 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3257 if (kvm_x86_ops->setup_mce)
3258 kvm_x86_ops->setup_mce(vcpu);
3263 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3264 struct kvm_x86_mce *mce)
3266 u64 mcg_cap = vcpu->arch.mcg_cap;
3267 unsigned bank_num = mcg_cap & 0xff;
3268 u64 *banks = vcpu->arch.mce_banks;
3270 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3273 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3274 * reporting is disabled
3276 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3277 vcpu->arch.mcg_ctl != ~(u64)0)
3279 banks += 4 * mce->bank;
3281 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3282 * reporting is disabled for the bank
3284 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3286 if (mce->status & MCI_STATUS_UC) {
3287 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3288 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3289 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3292 if (banks[1] & MCI_STATUS_VAL)
3293 mce->status |= MCI_STATUS_OVER;
3294 banks[2] = mce->addr;
3295 banks[3] = mce->misc;
3296 vcpu->arch.mcg_status = mce->mcg_status;
3297 banks[1] = mce->status;
3298 kvm_queue_exception(vcpu, MC_VECTOR);
3299 } else if (!(banks[1] & MCI_STATUS_VAL)
3300 || !(banks[1] & MCI_STATUS_UC)) {
3301 if (banks[1] & MCI_STATUS_VAL)
3302 mce->status |= MCI_STATUS_OVER;
3303 banks[2] = mce->addr;
3304 banks[3] = mce->misc;
3305 banks[1] = mce->status;
3307 banks[1] |= MCI_STATUS_OVER;
3311 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3312 struct kvm_vcpu_events *events)
3316 * FIXME: pass injected and pending separately. This is only
3317 * needed for nested virtualization, whose state cannot be
3318 * migrated yet. For now we can combine them.
3320 events->exception.injected =
3321 (vcpu->arch.exception.pending ||
3322 vcpu->arch.exception.injected) &&
3323 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3324 events->exception.nr = vcpu->arch.exception.nr;
3325 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3326 events->exception.pad = 0;
3327 events->exception.error_code = vcpu->arch.exception.error_code;
3329 events->interrupt.injected =
3330 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3331 events->interrupt.nr = vcpu->arch.interrupt.nr;
3332 events->interrupt.soft = 0;
3333 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3335 events->nmi.injected = vcpu->arch.nmi_injected;
3336 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3337 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3338 events->nmi.pad = 0;
3340 events->sipi_vector = 0; /* never valid when reporting to user space */
3342 events->smi.smm = is_smm(vcpu);
3343 events->smi.pending = vcpu->arch.smi_pending;
3344 events->smi.smm_inside_nmi =
3345 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3346 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3348 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3349 | KVM_VCPUEVENT_VALID_SHADOW
3350 | KVM_VCPUEVENT_VALID_SMM);
3351 memset(&events->reserved, 0, sizeof(events->reserved));
3354 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3356 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3357 struct kvm_vcpu_events *events)
3359 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3360 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3361 | KVM_VCPUEVENT_VALID_SHADOW
3362 | KVM_VCPUEVENT_VALID_SMM))
3365 if (events->exception.injected &&
3366 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3367 is_guest_mode(vcpu)))
3370 /* INITs are latched while in SMM */
3371 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3372 (events->smi.smm || events->smi.pending) &&
3373 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3377 vcpu->arch.exception.injected = false;
3378 vcpu->arch.exception.pending = events->exception.injected;
3379 vcpu->arch.exception.nr = events->exception.nr;
3380 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3381 vcpu->arch.exception.error_code = events->exception.error_code;
3383 vcpu->arch.interrupt.injected = events->interrupt.injected;
3384 vcpu->arch.interrupt.nr = events->interrupt.nr;
3385 vcpu->arch.interrupt.soft = events->interrupt.soft;
3386 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3387 kvm_x86_ops->set_interrupt_shadow(vcpu,
3388 events->interrupt.shadow);
3390 vcpu->arch.nmi_injected = events->nmi.injected;
3391 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3392 vcpu->arch.nmi_pending = events->nmi.pending;
3393 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3395 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3396 lapic_in_kernel(vcpu))
3397 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3399 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3400 u32 hflags = vcpu->arch.hflags;
3401 if (events->smi.smm)
3402 hflags |= HF_SMM_MASK;
3404 hflags &= ~HF_SMM_MASK;
3405 kvm_set_hflags(vcpu, hflags);
3407 vcpu->arch.smi_pending = events->smi.pending;
3409 if (events->smi.smm) {
3410 if (events->smi.smm_inside_nmi)
3411 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3413 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3414 if (lapic_in_kernel(vcpu)) {
3415 if (events->smi.latched_init)
3416 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3418 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3423 kvm_make_request(KVM_REQ_EVENT, vcpu);
3428 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3429 struct kvm_debugregs *dbgregs)
3433 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3434 kvm_get_dr(vcpu, 6, &val);
3436 dbgregs->dr7 = vcpu->arch.dr7;
3438 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3441 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3442 struct kvm_debugregs *dbgregs)
3447 if (dbgregs->dr6 & ~0xffffffffull)
3449 if (dbgregs->dr7 & ~0xffffffffull)
3452 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3453 kvm_update_dr0123(vcpu);
3454 vcpu->arch.dr6 = dbgregs->dr6;
3455 kvm_update_dr6(vcpu);
3456 vcpu->arch.dr7 = dbgregs->dr7;
3457 kvm_update_dr7(vcpu);
3462 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3464 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3466 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3467 u64 xstate_bv = xsave->header.xfeatures;
3471 * Copy legacy XSAVE area, to avoid complications with CPUID
3472 * leaves 0 and 1 in the loop below.
3474 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3477 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3478 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3481 * Copy each region from the possibly compacted offset to the
3482 * non-compacted offset.
3484 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3486 u64 feature = valid & -valid;
3487 int index = fls64(feature) - 1;
3488 void *src = get_xsave_addr(xsave, feature);
3491 u32 size, offset, ecx, edx;
3492 cpuid_count(XSTATE_CPUID, index,
3493 &size, &offset, &ecx, &edx);
3494 if (feature == XFEATURE_MASK_PKRU)
3495 memcpy(dest + offset, &vcpu->arch.pkru,
3496 sizeof(vcpu->arch.pkru));
3498 memcpy(dest + offset, src, size);
3506 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3508 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3509 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3513 * Copy legacy XSAVE area, to avoid complications with CPUID
3514 * leaves 0 and 1 in the loop below.
3516 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3518 /* Set XSTATE_BV and possibly XCOMP_BV. */
3519 xsave->header.xfeatures = xstate_bv;
3520 if (boot_cpu_has(X86_FEATURE_XSAVES))
3521 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3524 * Copy each region from the non-compacted offset to the
3525 * possibly compacted offset.
3527 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3529 u64 feature = valid & -valid;
3530 int index = fls64(feature) - 1;
3531 void *dest = get_xsave_addr(xsave, feature);
3534 u32 size, offset, ecx, edx;
3535 cpuid_count(XSTATE_CPUID, index,
3536 &size, &offset, &ecx, &edx);
3537 if (feature == XFEATURE_MASK_PKRU)
3538 memcpy(&vcpu->arch.pkru, src + offset,
3539 sizeof(vcpu->arch.pkru));
3541 memcpy(dest, src + offset, size);
3548 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3549 struct kvm_xsave *guest_xsave)
3551 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3552 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3553 fill_xsave((u8 *) guest_xsave->region, vcpu);
3555 memcpy(guest_xsave->region,
3556 &vcpu->arch.guest_fpu.state.fxsave,
3557 sizeof(struct fxregs_state));
3558 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3559 XFEATURE_MASK_FPSSE;
3563 #define XSAVE_MXCSR_OFFSET 24
3565 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3566 struct kvm_xsave *guest_xsave)
3569 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3570 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3572 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3574 * Here we allow setting states that are not present in
3575 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3576 * with old userspace.
3578 if (xstate_bv & ~kvm_supported_xcr0() ||
3579 mxcsr & ~mxcsr_feature_mask)
3581 load_xsave(vcpu, (u8 *)guest_xsave->region);
3583 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3584 mxcsr & ~mxcsr_feature_mask)
3586 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3587 guest_xsave->region, sizeof(struct fxregs_state));
3592 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3593 struct kvm_xcrs *guest_xcrs)
3595 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3596 guest_xcrs->nr_xcrs = 0;
3600 guest_xcrs->nr_xcrs = 1;
3601 guest_xcrs->flags = 0;
3602 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3603 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3606 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3607 struct kvm_xcrs *guest_xcrs)
3611 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3614 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3617 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3618 /* Only support XCR0 currently */
3619 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3620 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3621 guest_xcrs->xcrs[i].value);
3630 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3631 * stopped by the hypervisor. This function will be called from the host only.
3632 * EINVAL is returned when the host attempts to set the flag for a guest that
3633 * does not support pv clocks.
3635 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3637 if (!vcpu->arch.pv_time_enabled)
3639 vcpu->arch.pvclock_set_guest_stopped_request = true;
3640 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3644 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3645 struct kvm_enable_cap *cap)
3651 case KVM_CAP_HYPERV_SYNIC2:
3654 case KVM_CAP_HYPERV_SYNIC:
3655 if (!irqchip_in_kernel(vcpu->kvm))
3657 return kvm_hv_activate_synic(vcpu, cap->cap ==
3658 KVM_CAP_HYPERV_SYNIC2);
3664 long kvm_arch_vcpu_ioctl(struct file *filp,
3665 unsigned int ioctl, unsigned long arg)
3667 struct kvm_vcpu *vcpu = filp->private_data;
3668 void __user *argp = (void __user *)arg;
3671 struct kvm_lapic_state *lapic;
3672 struct kvm_xsave *xsave;
3673 struct kvm_xcrs *xcrs;
3681 case KVM_GET_LAPIC: {
3683 if (!lapic_in_kernel(vcpu))
3685 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3690 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3694 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3699 case KVM_SET_LAPIC: {
3701 if (!lapic_in_kernel(vcpu))
3703 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3704 if (IS_ERR(u.lapic)) {
3705 r = PTR_ERR(u.lapic);
3709 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3712 case KVM_INTERRUPT: {
3713 struct kvm_interrupt irq;
3716 if (copy_from_user(&irq, argp, sizeof irq))
3718 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3722 r = kvm_vcpu_ioctl_nmi(vcpu);
3726 r = kvm_vcpu_ioctl_smi(vcpu);
3729 case KVM_SET_CPUID: {
3730 struct kvm_cpuid __user *cpuid_arg = argp;
3731 struct kvm_cpuid cpuid;
3734 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3736 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3739 case KVM_SET_CPUID2: {
3740 struct kvm_cpuid2 __user *cpuid_arg = argp;
3741 struct kvm_cpuid2 cpuid;
3744 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3746 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3747 cpuid_arg->entries);
3750 case KVM_GET_CPUID2: {
3751 struct kvm_cpuid2 __user *cpuid_arg = argp;
3752 struct kvm_cpuid2 cpuid;
3755 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3757 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3758 cpuid_arg->entries);
3762 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3767 case KVM_GET_MSRS: {
3768 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3769 r = msr_io(vcpu, argp, do_get_msr, 1);
3770 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3773 case KVM_SET_MSRS: {
3774 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3775 r = msr_io(vcpu, argp, do_set_msr, 0);
3776 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3779 case KVM_TPR_ACCESS_REPORTING: {
3780 struct kvm_tpr_access_ctl tac;
3783 if (copy_from_user(&tac, argp, sizeof tac))
3785 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3789 if (copy_to_user(argp, &tac, sizeof tac))
3794 case KVM_SET_VAPIC_ADDR: {
3795 struct kvm_vapic_addr va;
3799 if (!lapic_in_kernel(vcpu))
3802 if (copy_from_user(&va, argp, sizeof va))
3804 idx = srcu_read_lock(&vcpu->kvm->srcu);
3805 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3806 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3809 case KVM_X86_SETUP_MCE: {
3813 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3815 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3818 case KVM_X86_SET_MCE: {
3819 struct kvm_x86_mce mce;
3822 if (copy_from_user(&mce, argp, sizeof mce))
3824 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3827 case KVM_GET_VCPU_EVENTS: {
3828 struct kvm_vcpu_events events;
3830 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3833 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3838 case KVM_SET_VCPU_EVENTS: {
3839 struct kvm_vcpu_events events;
3842 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3845 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3848 case KVM_GET_DEBUGREGS: {
3849 struct kvm_debugregs dbgregs;
3851 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3854 if (copy_to_user(argp, &dbgregs,
3855 sizeof(struct kvm_debugregs)))
3860 case KVM_SET_DEBUGREGS: {
3861 struct kvm_debugregs dbgregs;
3864 if (copy_from_user(&dbgregs, argp,
3865 sizeof(struct kvm_debugregs)))
3868 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3871 case KVM_GET_XSAVE: {
3872 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3877 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3880 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3885 case KVM_SET_XSAVE: {
3886 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3887 if (IS_ERR(u.xsave)) {
3888 r = PTR_ERR(u.xsave);
3892 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3895 case KVM_GET_XCRS: {
3896 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3901 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3904 if (copy_to_user(argp, u.xcrs,
3905 sizeof(struct kvm_xcrs)))
3910 case KVM_SET_XCRS: {
3911 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3912 if (IS_ERR(u.xcrs)) {
3913 r = PTR_ERR(u.xcrs);
3917 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3920 case KVM_SET_TSC_KHZ: {
3924 user_tsc_khz = (u32)arg;
3926 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3929 if (user_tsc_khz == 0)
3930 user_tsc_khz = tsc_khz;
3932 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3937 case KVM_GET_TSC_KHZ: {
3938 r = vcpu->arch.virtual_tsc_khz;
3941 case KVM_KVMCLOCK_CTRL: {
3942 r = kvm_set_guest_paused(vcpu);
3945 case KVM_ENABLE_CAP: {
3946 struct kvm_enable_cap cap;
3949 if (copy_from_user(&cap, argp, sizeof(cap)))
3951 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3964 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3966 return VM_FAULT_SIGBUS;
3969 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3973 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3975 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3979 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3982 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
3985 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3986 u32 kvm_nr_mmu_pages)
3988 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3991 mutex_lock(&kvm->slots_lock);
3993 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3994 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3996 mutex_unlock(&kvm->slots_lock);
4000 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4002 return kvm->arch.n_max_mmu_pages;
4005 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4007 struct kvm_pic *pic = kvm->arch.vpic;
4011 switch (chip->chip_id) {
4012 case KVM_IRQCHIP_PIC_MASTER:
4013 memcpy(&chip->chip.pic, &pic->pics[0],
4014 sizeof(struct kvm_pic_state));
4016 case KVM_IRQCHIP_PIC_SLAVE:
4017 memcpy(&chip->chip.pic, &pic->pics[1],
4018 sizeof(struct kvm_pic_state));
4020 case KVM_IRQCHIP_IOAPIC:
4021 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4030 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4032 struct kvm_pic *pic = kvm->arch.vpic;
4036 switch (chip->chip_id) {
4037 case KVM_IRQCHIP_PIC_MASTER:
4038 spin_lock(&pic->lock);
4039 memcpy(&pic->pics[0], &chip->chip.pic,
4040 sizeof(struct kvm_pic_state));
4041 spin_unlock(&pic->lock);
4043 case KVM_IRQCHIP_PIC_SLAVE:
4044 spin_lock(&pic->lock);
4045 memcpy(&pic->pics[1], &chip->chip.pic,
4046 sizeof(struct kvm_pic_state));
4047 spin_unlock(&pic->lock);
4049 case KVM_IRQCHIP_IOAPIC:
4050 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4056 kvm_pic_update_irq(pic);
4060 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4062 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4064 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4066 mutex_lock(&kps->lock);
4067 memcpy(ps, &kps->channels, sizeof(*ps));
4068 mutex_unlock(&kps->lock);
4072 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4075 struct kvm_pit *pit = kvm->arch.vpit;
4077 mutex_lock(&pit->pit_state.lock);
4078 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4079 for (i = 0; i < 3; i++)
4080 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4081 mutex_unlock(&pit->pit_state.lock);
4085 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4087 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4088 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4089 sizeof(ps->channels));
4090 ps->flags = kvm->arch.vpit->pit_state.flags;
4091 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4092 memset(&ps->reserved, 0, sizeof(ps->reserved));
4096 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4100 u32 prev_legacy, cur_legacy;
4101 struct kvm_pit *pit = kvm->arch.vpit;
4103 mutex_lock(&pit->pit_state.lock);
4104 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4105 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4106 if (!prev_legacy && cur_legacy)
4108 memcpy(&pit->pit_state.channels, &ps->channels,
4109 sizeof(pit->pit_state.channels));
4110 pit->pit_state.flags = ps->flags;
4111 for (i = 0; i < 3; i++)
4112 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4114 mutex_unlock(&pit->pit_state.lock);
4118 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4119 struct kvm_reinject_control *control)
4121 struct kvm_pit *pit = kvm->arch.vpit;
4126 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4127 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4128 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4130 mutex_lock(&pit->pit_state.lock);
4131 kvm_pit_set_reinject(pit, control->pit_reinject);
4132 mutex_unlock(&pit->pit_state.lock);
4138 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4139 * @kvm: kvm instance
4140 * @log: slot id and address to which we copy the log
4142 * Steps 1-4 below provide general overview of dirty page logging. See
4143 * kvm_get_dirty_log_protect() function description for additional details.
4145 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4146 * always flush the TLB (step 4) even if previous step failed and the dirty
4147 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4148 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4149 * writes will be marked dirty for next log read.
4151 * 1. Take a snapshot of the bit and clear it if needed.
4152 * 2. Write protect the corresponding page.
4153 * 3. Copy the snapshot to the userspace.
4154 * 4. Flush TLB's if needed.
4156 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4158 bool is_dirty = false;
4161 mutex_lock(&kvm->slots_lock);
4164 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4166 if (kvm_x86_ops->flush_log_dirty)
4167 kvm_x86_ops->flush_log_dirty(kvm);
4169 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
4172 * All the TLBs can be flushed out of mmu lock, see the comments in
4173 * kvm_mmu_slot_remove_write_access().
4175 lockdep_assert_held(&kvm->slots_lock);
4177 kvm_flush_remote_tlbs(kvm);
4179 mutex_unlock(&kvm->slots_lock);
4183 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4186 if (!irqchip_in_kernel(kvm))
4189 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4190 irq_event->irq, irq_event->level,
4195 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4196 struct kvm_enable_cap *cap)
4204 case KVM_CAP_DISABLE_QUIRKS:
4205 kvm->arch.disabled_quirks = cap->args[0];
4208 case KVM_CAP_SPLIT_IRQCHIP: {
4209 mutex_lock(&kvm->lock);
4211 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4212 goto split_irqchip_unlock;
4214 if (irqchip_in_kernel(kvm))
4215 goto split_irqchip_unlock;
4216 if (kvm->created_vcpus)
4217 goto split_irqchip_unlock;
4218 r = kvm_setup_empty_irq_routing(kvm);
4220 goto split_irqchip_unlock;
4221 /* Pairs with irqchip_in_kernel. */
4223 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4224 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4226 split_irqchip_unlock:
4227 mutex_unlock(&kvm->lock);
4230 case KVM_CAP_X2APIC_API:
4232 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4235 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4236 kvm->arch.x2apic_format = true;
4237 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4238 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4242 case KVM_CAP_X86_DISABLE_EXITS:
4244 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4247 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4248 kvm_can_mwait_in_guest())
4249 kvm->arch.mwait_in_guest = true;
4250 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HTL)
4251 kvm->arch.hlt_in_guest = true;
4252 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4253 kvm->arch.pause_in_guest = true;
4263 long kvm_arch_vm_ioctl(struct file *filp,
4264 unsigned int ioctl, unsigned long arg)
4266 struct kvm *kvm = filp->private_data;
4267 void __user *argp = (void __user *)arg;
4270 * This union makes it completely explicit to gcc-3.x
4271 * that these two variables' stack usage should be
4272 * combined, not added together.
4275 struct kvm_pit_state ps;
4276 struct kvm_pit_state2 ps2;
4277 struct kvm_pit_config pit_config;
4281 case KVM_SET_TSS_ADDR:
4282 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4284 case KVM_SET_IDENTITY_MAP_ADDR: {
4287 mutex_lock(&kvm->lock);
4289 if (kvm->created_vcpus)
4290 goto set_identity_unlock;
4292 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4293 goto set_identity_unlock;
4294 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4295 set_identity_unlock:
4296 mutex_unlock(&kvm->lock);
4299 case KVM_SET_NR_MMU_PAGES:
4300 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4302 case KVM_GET_NR_MMU_PAGES:
4303 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4305 case KVM_CREATE_IRQCHIP: {
4306 mutex_lock(&kvm->lock);
4309 if (irqchip_in_kernel(kvm))
4310 goto create_irqchip_unlock;
4313 if (kvm->created_vcpus)
4314 goto create_irqchip_unlock;
4316 r = kvm_pic_init(kvm);
4318 goto create_irqchip_unlock;
4320 r = kvm_ioapic_init(kvm);
4322 kvm_pic_destroy(kvm);
4323 goto create_irqchip_unlock;
4326 r = kvm_setup_default_irq_routing(kvm);
4328 kvm_ioapic_destroy(kvm);
4329 kvm_pic_destroy(kvm);
4330 goto create_irqchip_unlock;
4332 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4334 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4335 create_irqchip_unlock:
4336 mutex_unlock(&kvm->lock);
4339 case KVM_CREATE_PIT:
4340 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4342 case KVM_CREATE_PIT2:
4344 if (copy_from_user(&u.pit_config, argp,
4345 sizeof(struct kvm_pit_config)))
4348 mutex_lock(&kvm->lock);
4351 goto create_pit_unlock;
4353 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4357 mutex_unlock(&kvm->lock);
4359 case KVM_GET_IRQCHIP: {
4360 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4361 struct kvm_irqchip *chip;
4363 chip = memdup_user(argp, sizeof(*chip));
4370 if (!irqchip_kernel(kvm))
4371 goto get_irqchip_out;
4372 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4374 goto get_irqchip_out;
4376 if (copy_to_user(argp, chip, sizeof *chip))
4377 goto get_irqchip_out;
4383 case KVM_SET_IRQCHIP: {
4384 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4385 struct kvm_irqchip *chip;
4387 chip = memdup_user(argp, sizeof(*chip));
4394 if (!irqchip_kernel(kvm))
4395 goto set_irqchip_out;
4396 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4398 goto set_irqchip_out;
4406 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4409 if (!kvm->arch.vpit)
4411 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4415 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4422 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4425 if (!kvm->arch.vpit)
4427 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4430 case KVM_GET_PIT2: {
4432 if (!kvm->arch.vpit)
4434 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4438 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4443 case KVM_SET_PIT2: {
4445 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4448 if (!kvm->arch.vpit)
4450 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4453 case KVM_REINJECT_CONTROL: {
4454 struct kvm_reinject_control control;
4456 if (copy_from_user(&control, argp, sizeof(control)))
4458 r = kvm_vm_ioctl_reinject(kvm, &control);
4461 case KVM_SET_BOOT_CPU_ID:
4463 mutex_lock(&kvm->lock);
4464 if (kvm->created_vcpus)
4467 kvm->arch.bsp_vcpu_id = arg;
4468 mutex_unlock(&kvm->lock);
4470 case KVM_XEN_HVM_CONFIG: {
4471 struct kvm_xen_hvm_config xhc;
4473 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4478 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4482 case KVM_SET_CLOCK: {
4483 struct kvm_clock_data user_ns;
4487 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4496 * TODO: userspace has to take care of races with VCPU_RUN, so
4497 * kvm_gen_update_masterclock() can be cut down to locked
4498 * pvclock_update_vm_gtod_copy().
4500 kvm_gen_update_masterclock(kvm);
4501 now_ns = get_kvmclock_ns(kvm);
4502 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4503 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4506 case KVM_GET_CLOCK: {
4507 struct kvm_clock_data user_ns;
4510 now_ns = get_kvmclock_ns(kvm);
4511 user_ns.clock = now_ns;
4512 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4513 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4516 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4521 case KVM_ENABLE_CAP: {
4522 struct kvm_enable_cap cap;
4525 if (copy_from_user(&cap, argp, sizeof(cap)))
4527 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4530 case KVM_MEMORY_ENCRYPT_OP: {
4532 if (kvm_x86_ops->mem_enc_op)
4533 r = kvm_x86_ops->mem_enc_op(kvm, argp);
4536 case KVM_MEMORY_ENCRYPT_REG_REGION: {
4537 struct kvm_enc_region region;
4540 if (copy_from_user(®ion, argp, sizeof(region)))
4544 if (kvm_x86_ops->mem_enc_reg_region)
4545 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
4548 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4549 struct kvm_enc_region region;
4552 if (copy_from_user(®ion, argp, sizeof(region)))
4556 if (kvm_x86_ops->mem_enc_unreg_region)
4557 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
4560 case KVM_HYPERV_EVENTFD: {
4561 struct kvm_hyperv_eventfd hvevfd;
4564 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4566 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4576 static void kvm_init_msr_list(void)
4581 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4582 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4586 * Even MSRs that are valid in the host may not be exposed
4587 * to the guests in some cases.
4589 switch (msrs_to_save[i]) {
4590 case MSR_IA32_BNDCFGS:
4591 if (!kvm_x86_ops->mpx_supported())
4595 if (!kvm_x86_ops->rdtscp_supported())
4603 msrs_to_save[j] = msrs_to_save[i];
4606 num_msrs_to_save = j;
4608 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4609 switch (emulated_msrs[i]) {
4610 case MSR_IA32_SMBASE:
4611 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4619 emulated_msrs[j] = emulated_msrs[i];
4622 num_emulated_msrs = j;
4624 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4625 struct kvm_msr_entry msr;
4627 msr.index = msr_based_features[i];
4628 if (kvm_get_msr_feature(&msr))
4632 msr_based_features[j] = msr_based_features[i];
4635 num_msr_based_features = j;
4638 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4646 if (!(lapic_in_kernel(vcpu) &&
4647 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4648 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4659 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4666 if (!(lapic_in_kernel(vcpu) &&
4667 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4669 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4671 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4681 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4682 struct kvm_segment *var, int seg)
4684 kvm_x86_ops->set_segment(vcpu, var, seg);
4687 void kvm_get_segment(struct kvm_vcpu *vcpu,
4688 struct kvm_segment *var, int seg)
4690 kvm_x86_ops->get_segment(vcpu, var, seg);
4693 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4694 struct x86_exception *exception)
4698 BUG_ON(!mmu_is_nested(vcpu));
4700 /* NPT walks are always user-walks */
4701 access |= PFERR_USER_MASK;
4702 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4707 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4708 struct x86_exception *exception)
4710 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4711 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4714 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4715 struct x86_exception *exception)
4717 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4718 access |= PFERR_FETCH_MASK;
4719 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4722 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4723 struct x86_exception *exception)
4725 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4726 access |= PFERR_WRITE_MASK;
4727 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4730 /* uses this to access any guest's mapped memory without checking CPL */
4731 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4732 struct x86_exception *exception)
4734 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4737 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4738 struct kvm_vcpu *vcpu, u32 access,
4739 struct x86_exception *exception)
4742 int r = X86EMUL_CONTINUE;
4745 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4747 unsigned offset = addr & (PAGE_SIZE-1);
4748 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4751 if (gpa == UNMAPPED_GVA)
4752 return X86EMUL_PROPAGATE_FAULT;
4753 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4756 r = X86EMUL_IO_NEEDED;
4768 /* used for instruction fetching */
4769 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4770 gva_t addr, void *val, unsigned int bytes,
4771 struct x86_exception *exception)
4773 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4774 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4778 /* Inline kvm_read_guest_virt_helper for speed. */
4779 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4781 if (unlikely(gpa == UNMAPPED_GVA))
4782 return X86EMUL_PROPAGATE_FAULT;
4784 offset = addr & (PAGE_SIZE-1);
4785 if (WARN_ON(offset + bytes > PAGE_SIZE))
4786 bytes = (unsigned)PAGE_SIZE - offset;
4787 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4789 if (unlikely(ret < 0))
4790 return X86EMUL_IO_NEEDED;
4792 return X86EMUL_CONTINUE;
4795 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4796 gva_t addr, void *val, unsigned int bytes,
4797 struct x86_exception *exception)
4799 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4800 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4802 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4805 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4807 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4808 gva_t addr, void *val, unsigned int bytes,
4809 struct x86_exception *exception)
4811 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4812 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4815 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4816 unsigned long addr, void *val, unsigned int bytes)
4818 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4819 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4821 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4824 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4825 gva_t addr, void *val,
4827 struct x86_exception *exception)
4829 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4831 int r = X86EMUL_CONTINUE;
4834 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4837 unsigned offset = addr & (PAGE_SIZE-1);
4838 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4841 if (gpa == UNMAPPED_GVA)
4842 return X86EMUL_PROPAGATE_FAULT;
4843 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4845 r = X86EMUL_IO_NEEDED;
4856 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4858 int handle_ud(struct kvm_vcpu *vcpu)
4860 int emul_type = EMULTYPE_TRAP_UD;
4861 enum emulation_result er;
4862 char sig[5]; /* ud2; .ascii "kvm" */
4863 struct x86_exception e;
4865 if (force_emulation_prefix &&
4866 kvm_read_guest_virt(&vcpu->arch.emulate_ctxt,
4867 kvm_get_linear_rip(vcpu), sig, sizeof(sig), &e) == 0 &&
4868 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
4869 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
4873 er = emulate_instruction(vcpu, emul_type);
4874 if (er == EMULATE_USER_EXIT)
4876 if (er != EMULATE_DONE)
4877 kvm_queue_exception(vcpu, UD_VECTOR);
4880 EXPORT_SYMBOL_GPL(handle_ud);
4882 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4883 gpa_t gpa, bool write)
4885 /* For APIC access vmexit */
4886 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4889 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4890 trace_vcpu_match_mmio(gva, gpa, write, true);
4897 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4898 gpa_t *gpa, struct x86_exception *exception,
4901 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4902 | (write ? PFERR_WRITE_MASK : 0);
4905 * currently PKRU is only applied to ept enabled guest so
4906 * there is no pkey in EPT page table for L1 guest or EPT
4907 * shadow page table for L2 guest.
4909 if (vcpu_match_mmio_gva(vcpu, gva)
4910 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4911 vcpu->arch.access, 0, access)) {
4912 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4913 (gva & (PAGE_SIZE - 1));
4914 trace_vcpu_match_mmio(gva, *gpa, write, false);
4918 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4920 if (*gpa == UNMAPPED_GVA)
4923 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4926 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4927 const void *val, int bytes)
4931 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4934 kvm_page_track_write(vcpu, gpa, val, bytes);
4938 struct read_write_emulator_ops {
4939 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4941 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4942 void *val, int bytes);
4943 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4944 int bytes, void *val);
4945 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4946 void *val, int bytes);
4950 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4952 if (vcpu->mmio_read_completed) {
4953 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4954 vcpu->mmio_fragments[0].gpa, val);
4955 vcpu->mmio_read_completed = 0;
4962 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4963 void *val, int bytes)
4965 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4968 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4969 void *val, int bytes)
4971 return emulator_write_phys(vcpu, gpa, val, bytes);
4974 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4976 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
4977 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4980 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4981 void *val, int bytes)
4983 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
4984 return X86EMUL_IO_NEEDED;
4987 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4988 void *val, int bytes)
4990 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4992 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4993 return X86EMUL_CONTINUE;
4996 static const struct read_write_emulator_ops read_emultor = {
4997 .read_write_prepare = read_prepare,
4998 .read_write_emulate = read_emulate,
4999 .read_write_mmio = vcpu_mmio_read,
5000 .read_write_exit_mmio = read_exit_mmio,
5003 static const struct read_write_emulator_ops write_emultor = {
5004 .read_write_emulate = write_emulate,
5005 .read_write_mmio = write_mmio,
5006 .read_write_exit_mmio = write_exit_mmio,
5010 static int emulator_read_write_onepage(unsigned long addr, void *val,
5012 struct x86_exception *exception,
5013 struct kvm_vcpu *vcpu,
5014 const struct read_write_emulator_ops *ops)
5018 bool write = ops->write;
5019 struct kvm_mmio_fragment *frag;
5020 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5023 * If the exit was due to a NPF we may already have a GPA.
5024 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5025 * Note, this cannot be used on string operations since string
5026 * operation using rep will only have the initial GPA from the NPF
5029 if (vcpu->arch.gpa_available &&
5030 emulator_can_use_gpa(ctxt) &&
5031 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5032 gpa = vcpu->arch.gpa_val;
5033 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5035 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5037 return X86EMUL_PROPAGATE_FAULT;
5040 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5041 return X86EMUL_CONTINUE;
5044 * Is this MMIO handled locally?
5046 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5047 if (handled == bytes)
5048 return X86EMUL_CONTINUE;
5054 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5055 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5059 return X86EMUL_CONTINUE;
5062 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5064 void *val, unsigned int bytes,
5065 struct x86_exception *exception,
5066 const struct read_write_emulator_ops *ops)
5068 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5072 if (ops->read_write_prepare &&
5073 ops->read_write_prepare(vcpu, val, bytes))
5074 return X86EMUL_CONTINUE;
5076 vcpu->mmio_nr_fragments = 0;
5078 /* Crossing a page boundary? */
5079 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5082 now = -addr & ~PAGE_MASK;
5083 rc = emulator_read_write_onepage(addr, val, now, exception,
5086 if (rc != X86EMUL_CONTINUE)
5089 if (ctxt->mode != X86EMUL_MODE_PROT64)
5095 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5097 if (rc != X86EMUL_CONTINUE)
5100 if (!vcpu->mmio_nr_fragments)
5103 gpa = vcpu->mmio_fragments[0].gpa;
5105 vcpu->mmio_needed = 1;
5106 vcpu->mmio_cur_fragment = 0;
5108 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5109 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5110 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5111 vcpu->run->mmio.phys_addr = gpa;
5113 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5116 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5120 struct x86_exception *exception)
5122 return emulator_read_write(ctxt, addr, val, bytes,
5123 exception, &read_emultor);
5126 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5130 struct x86_exception *exception)
5132 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5133 exception, &write_emultor);
5136 #define CMPXCHG_TYPE(t, ptr, old, new) \
5137 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5139 #ifdef CONFIG_X86_64
5140 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5142 # define CMPXCHG64(ptr, old, new) \
5143 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5146 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5151 struct x86_exception *exception)
5153 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5159 /* guests cmpxchg8b have to be emulated atomically */
5160 if (bytes > 8 || (bytes & (bytes - 1)))
5163 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5165 if (gpa == UNMAPPED_GVA ||
5166 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5169 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5172 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5173 if (is_error_page(page))
5176 kaddr = kmap_atomic(page);
5177 kaddr += offset_in_page(gpa);
5180 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5183 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5186 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5189 exchanged = CMPXCHG64(kaddr, old, new);
5194 kunmap_atomic(kaddr);
5195 kvm_release_page_dirty(page);
5198 return X86EMUL_CMPXCHG_FAILED;
5200 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5201 kvm_page_track_write(vcpu, gpa, new, bytes);
5203 return X86EMUL_CONTINUE;
5206 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5208 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5211 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5215 for (i = 0; i < vcpu->arch.pio.count; i++) {
5216 if (vcpu->arch.pio.in)
5217 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5218 vcpu->arch.pio.size, pd);
5220 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5221 vcpu->arch.pio.port, vcpu->arch.pio.size,
5225 pd += vcpu->arch.pio.size;
5230 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5231 unsigned short port, void *val,
5232 unsigned int count, bool in)
5234 vcpu->arch.pio.port = port;
5235 vcpu->arch.pio.in = in;
5236 vcpu->arch.pio.count = count;
5237 vcpu->arch.pio.size = size;
5239 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5240 vcpu->arch.pio.count = 0;
5244 vcpu->run->exit_reason = KVM_EXIT_IO;
5245 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5246 vcpu->run->io.size = size;
5247 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5248 vcpu->run->io.count = count;
5249 vcpu->run->io.port = port;
5254 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5255 int size, unsigned short port, void *val,
5258 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5261 if (vcpu->arch.pio.count)
5264 memset(vcpu->arch.pio_data, 0, size * count);
5266 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5269 memcpy(val, vcpu->arch.pio_data, size * count);
5270 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5271 vcpu->arch.pio.count = 0;
5278 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5279 int size, unsigned short port,
5280 const void *val, unsigned int count)
5282 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5284 memcpy(vcpu->arch.pio_data, val, size * count);
5285 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5286 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5289 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5291 return kvm_x86_ops->get_segment_base(vcpu, seg);
5294 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5296 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5299 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5301 if (!need_emulate_wbinvd(vcpu))
5302 return X86EMUL_CONTINUE;
5304 if (kvm_x86_ops->has_wbinvd_exit()) {
5305 int cpu = get_cpu();
5307 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5308 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5309 wbinvd_ipi, NULL, 1);
5311 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5314 return X86EMUL_CONTINUE;
5317 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5319 kvm_emulate_wbinvd_noskip(vcpu);
5320 return kvm_skip_emulated_instruction(vcpu);
5322 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5326 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5328 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5331 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5332 unsigned long *dest)
5334 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5337 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5338 unsigned long value)
5341 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5344 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5346 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5349 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5351 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5352 unsigned long value;
5356 value = kvm_read_cr0(vcpu);
5359 value = vcpu->arch.cr2;
5362 value = kvm_read_cr3(vcpu);
5365 value = kvm_read_cr4(vcpu);
5368 value = kvm_get_cr8(vcpu);
5371 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5378 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5380 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5385 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5388 vcpu->arch.cr2 = val;
5391 res = kvm_set_cr3(vcpu, val);
5394 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5397 res = kvm_set_cr8(vcpu, val);
5400 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5407 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5409 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5412 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5414 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5417 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5419 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5422 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5424 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5427 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5429 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5432 static unsigned long emulator_get_cached_segment_base(
5433 struct x86_emulate_ctxt *ctxt, int seg)
5435 return get_segment_base(emul_to_vcpu(ctxt), seg);
5438 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5439 struct desc_struct *desc, u32 *base3,
5442 struct kvm_segment var;
5444 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5445 *selector = var.selector;
5448 memset(desc, 0, sizeof(*desc));
5456 set_desc_limit(desc, var.limit);
5457 set_desc_base(desc, (unsigned long)var.base);
5458 #ifdef CONFIG_X86_64
5460 *base3 = var.base >> 32;
5462 desc->type = var.type;
5464 desc->dpl = var.dpl;
5465 desc->p = var.present;
5466 desc->avl = var.avl;
5474 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5475 struct desc_struct *desc, u32 base3,
5478 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5479 struct kvm_segment var;
5481 var.selector = selector;
5482 var.base = get_desc_base(desc);
5483 #ifdef CONFIG_X86_64
5484 var.base |= ((u64)base3) << 32;
5486 var.limit = get_desc_limit(desc);
5488 var.limit = (var.limit << 12) | 0xfff;
5489 var.type = desc->type;
5490 var.dpl = desc->dpl;
5495 var.avl = desc->avl;
5496 var.present = desc->p;
5497 var.unusable = !var.present;
5500 kvm_set_segment(vcpu, &var, seg);
5504 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5505 u32 msr_index, u64 *pdata)
5507 struct msr_data msr;
5510 msr.index = msr_index;
5511 msr.host_initiated = false;
5512 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5520 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5521 u32 msr_index, u64 data)
5523 struct msr_data msr;
5526 msr.index = msr_index;
5527 msr.host_initiated = false;
5528 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5531 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5533 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5535 return vcpu->arch.smbase;
5538 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5540 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5542 vcpu->arch.smbase = smbase;
5545 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5548 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5551 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5552 u32 pmc, u64 *pdata)
5554 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5557 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5559 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5562 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5563 struct x86_instruction_info *info,
5564 enum x86_intercept_stage stage)
5566 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5569 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5570 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5572 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5575 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5577 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5580 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5582 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5585 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5587 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5590 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5592 return emul_to_vcpu(ctxt)->arch.hflags;
5595 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5597 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5600 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5602 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5605 static const struct x86_emulate_ops emulate_ops = {
5606 .read_gpr = emulator_read_gpr,
5607 .write_gpr = emulator_write_gpr,
5608 .read_std = kvm_read_guest_virt_system,
5609 .write_std = kvm_write_guest_virt_system,
5610 .read_phys = kvm_read_guest_phys_system,
5611 .fetch = kvm_fetch_guest_virt,
5612 .read_emulated = emulator_read_emulated,
5613 .write_emulated = emulator_write_emulated,
5614 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5615 .invlpg = emulator_invlpg,
5616 .pio_in_emulated = emulator_pio_in_emulated,
5617 .pio_out_emulated = emulator_pio_out_emulated,
5618 .get_segment = emulator_get_segment,
5619 .set_segment = emulator_set_segment,
5620 .get_cached_segment_base = emulator_get_cached_segment_base,
5621 .get_gdt = emulator_get_gdt,
5622 .get_idt = emulator_get_idt,
5623 .set_gdt = emulator_set_gdt,
5624 .set_idt = emulator_set_idt,
5625 .get_cr = emulator_get_cr,
5626 .set_cr = emulator_set_cr,
5627 .cpl = emulator_get_cpl,
5628 .get_dr = emulator_get_dr,
5629 .set_dr = emulator_set_dr,
5630 .get_smbase = emulator_get_smbase,
5631 .set_smbase = emulator_set_smbase,
5632 .set_msr = emulator_set_msr,
5633 .get_msr = emulator_get_msr,
5634 .check_pmc = emulator_check_pmc,
5635 .read_pmc = emulator_read_pmc,
5636 .halt = emulator_halt,
5637 .wbinvd = emulator_wbinvd,
5638 .fix_hypercall = emulator_fix_hypercall,
5639 .intercept = emulator_intercept,
5640 .get_cpuid = emulator_get_cpuid,
5641 .set_nmi_mask = emulator_set_nmi_mask,
5642 .get_hflags = emulator_get_hflags,
5643 .set_hflags = emulator_set_hflags,
5644 .pre_leave_smm = emulator_pre_leave_smm,
5647 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5649 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5651 * an sti; sti; sequence only disable interrupts for the first
5652 * instruction. So, if the last instruction, be it emulated or
5653 * not, left the system with the INT_STI flag enabled, it
5654 * means that the last instruction is an sti. We should not
5655 * leave the flag on in this case. The same goes for mov ss
5657 if (int_shadow & mask)
5659 if (unlikely(int_shadow || mask)) {
5660 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5662 kvm_make_request(KVM_REQ_EVENT, vcpu);
5666 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5668 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5669 if (ctxt->exception.vector == PF_VECTOR)
5670 return kvm_propagate_fault(vcpu, &ctxt->exception);
5672 if (ctxt->exception.error_code_valid)
5673 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5674 ctxt->exception.error_code);
5676 kvm_queue_exception(vcpu, ctxt->exception.vector);
5680 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5682 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5685 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5687 ctxt->eflags = kvm_get_rflags(vcpu);
5688 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5690 ctxt->eip = kvm_rip_read(vcpu);
5691 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5692 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5693 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5694 cs_db ? X86EMUL_MODE_PROT32 :
5695 X86EMUL_MODE_PROT16;
5696 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5697 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5698 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5700 init_decode_cache(ctxt);
5701 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5704 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5706 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5709 init_emulate_ctxt(vcpu);
5713 ctxt->_eip = ctxt->eip + inc_eip;
5714 ret = emulate_int_real(ctxt, irq);
5716 if (ret != X86EMUL_CONTINUE)
5717 return EMULATE_FAIL;
5719 ctxt->eip = ctxt->_eip;
5720 kvm_rip_write(vcpu, ctxt->eip);
5721 kvm_set_rflags(vcpu, ctxt->eflags);
5723 return EMULATE_DONE;
5725 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5727 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
5729 int r = EMULATE_DONE;
5731 ++vcpu->stat.insn_emulation_fail;
5732 trace_kvm_emulate_insn_failed(vcpu);
5734 if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
5735 return EMULATE_FAIL;
5737 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5738 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5739 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5740 vcpu->run->internal.ndata = 0;
5741 r = EMULATE_USER_EXIT;
5744 kvm_queue_exception(vcpu, UD_VECTOR);
5749 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5750 bool write_fault_to_shadow_pgtable,
5756 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5759 if (!vcpu->arch.mmu.direct_map) {
5761 * Write permission should be allowed since only
5762 * write access need to be emulated.
5764 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5767 * If the mapping is invalid in guest, let cpu retry
5768 * it to generate fault.
5770 if (gpa == UNMAPPED_GVA)
5775 * Do not retry the unhandleable instruction if it faults on the
5776 * readonly host memory, otherwise it will goto a infinite loop:
5777 * retry instruction -> write #PF -> emulation fail -> retry
5778 * instruction -> ...
5780 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5783 * If the instruction failed on the error pfn, it can not be fixed,
5784 * report the error to userspace.
5786 if (is_error_noslot_pfn(pfn))
5789 kvm_release_pfn_clean(pfn);
5791 /* The instructions are well-emulated on direct mmu. */
5792 if (vcpu->arch.mmu.direct_map) {
5793 unsigned int indirect_shadow_pages;
5795 spin_lock(&vcpu->kvm->mmu_lock);
5796 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5797 spin_unlock(&vcpu->kvm->mmu_lock);
5799 if (indirect_shadow_pages)
5800 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5806 * if emulation was due to access to shadowed page table
5807 * and it failed try to unshadow page and re-enter the
5808 * guest to let CPU execute the instruction.
5810 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5813 * If the access faults on its page table, it can not
5814 * be fixed by unprotecting shadow page and it should
5815 * be reported to userspace.
5817 return !write_fault_to_shadow_pgtable;
5820 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5821 unsigned long cr2, int emulation_type)
5823 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5824 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5826 last_retry_eip = vcpu->arch.last_retry_eip;
5827 last_retry_addr = vcpu->arch.last_retry_addr;
5830 * If the emulation is caused by #PF and it is non-page_table
5831 * writing instruction, it means the VM-EXIT is caused by shadow
5832 * page protected, we can zap the shadow page and retry this
5833 * instruction directly.
5835 * Note: if the guest uses a non-page-table modifying instruction
5836 * on the PDE that points to the instruction, then we will unmap
5837 * the instruction and go to an infinite loop. So, we cache the
5838 * last retried eip and the last fault address, if we meet the eip
5839 * and the address again, we can break out of the potential infinite
5842 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5844 if (!(emulation_type & EMULTYPE_RETRY))
5847 if (x86_page_table_writing_insn(ctxt))
5850 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5853 vcpu->arch.last_retry_eip = ctxt->eip;
5854 vcpu->arch.last_retry_addr = cr2;
5856 if (!vcpu->arch.mmu.direct_map)
5857 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5859 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5864 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5865 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5867 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5869 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5870 /* This is a good place to trace that we are exiting SMM. */
5871 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5873 /* Process a latched INIT or SMI, if any. */
5874 kvm_make_request(KVM_REQ_EVENT, vcpu);
5877 kvm_mmu_reset_context(vcpu);
5880 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5882 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5884 vcpu->arch.hflags = emul_flags;
5886 if (changed & HF_SMM_MASK)
5887 kvm_smm_changed(vcpu);
5890 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5899 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5900 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5905 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5907 struct kvm_run *kvm_run = vcpu->run;
5909 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5910 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5911 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5912 kvm_run->debug.arch.exception = DB_VECTOR;
5913 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5914 *r = EMULATE_USER_EXIT;
5917 * "Certain debug exceptions may clear bit 0-3. The
5918 * remaining contents of the DR6 register are never
5919 * cleared by the processor".
5921 vcpu->arch.dr6 &= ~15;
5922 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5923 kvm_queue_exception(vcpu, DB_VECTOR);
5927 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5929 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5930 int r = EMULATE_DONE;
5932 kvm_x86_ops->skip_emulated_instruction(vcpu);
5935 * rflags is the old, "raw" value of the flags. The new value has
5936 * not been saved yet.
5938 * This is correct even for TF set by the guest, because "the
5939 * processor will not generate this exception after the instruction
5940 * that sets the TF flag".
5942 if (unlikely(rflags & X86_EFLAGS_TF))
5943 kvm_vcpu_do_singlestep(vcpu, &r);
5944 return r == EMULATE_DONE;
5946 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5948 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5950 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5951 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5952 struct kvm_run *kvm_run = vcpu->run;
5953 unsigned long eip = kvm_get_linear_rip(vcpu);
5954 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5955 vcpu->arch.guest_debug_dr7,
5959 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5960 kvm_run->debug.arch.pc = eip;
5961 kvm_run->debug.arch.exception = DB_VECTOR;
5962 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5963 *r = EMULATE_USER_EXIT;
5968 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5969 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5970 unsigned long eip = kvm_get_linear_rip(vcpu);
5971 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5976 vcpu->arch.dr6 &= ~15;
5977 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5978 kvm_queue_exception(vcpu, DB_VECTOR);
5987 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
5989 switch (ctxt->opcode_len) {
5996 case 0xe6: /* OUT */
6000 case 0x6c: /* INS */
6002 case 0x6e: /* OUTS */
6009 case 0x33: /* RDPMC */
6018 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6025 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6026 bool writeback = true;
6027 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6030 * Clear write_fault_to_shadow_pgtable here to ensure it is
6033 vcpu->arch.write_fault_to_shadow_pgtable = false;
6034 kvm_clear_exception_queue(vcpu);
6036 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6037 init_emulate_ctxt(vcpu);
6040 * We will reenter on the same instruction since
6041 * we do not set complete_userspace_io. This does not
6042 * handle watchpoints yet, those would be handled in
6045 if (!(emulation_type & EMULTYPE_SKIP) &&
6046 kvm_vcpu_check_breakpoint(vcpu, &r))
6049 ctxt->interruptibility = 0;
6050 ctxt->have_exception = false;
6051 ctxt->exception.vector = -1;
6052 ctxt->perm_ok = false;
6054 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6056 r = x86_decode_insn(ctxt, insn, insn_len);
6058 trace_kvm_emulate_insn_start(vcpu);
6059 ++vcpu->stat.insn_emulation;
6060 if (r != EMULATION_OK) {
6061 if (emulation_type & EMULTYPE_TRAP_UD)
6062 return EMULATE_FAIL;
6063 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6065 return EMULATE_DONE;
6066 if (ctxt->have_exception && inject_emulated_exception(vcpu))
6067 return EMULATE_DONE;
6068 if (emulation_type & EMULTYPE_SKIP)
6069 return EMULATE_FAIL;
6070 return handle_emulation_failure(vcpu, emulation_type);
6074 if ((emulation_type & EMULTYPE_VMWARE) &&
6075 !is_vmware_backdoor_opcode(ctxt))
6076 return EMULATE_FAIL;
6078 if (emulation_type & EMULTYPE_SKIP) {
6079 kvm_rip_write(vcpu, ctxt->_eip);
6080 if (ctxt->eflags & X86_EFLAGS_RF)
6081 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6082 return EMULATE_DONE;
6085 if (retry_instruction(ctxt, cr2, emulation_type))
6086 return EMULATE_DONE;
6088 /* this is needed for vmware backdoor interface to work since it
6089 changes registers values during IO operation */
6090 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6091 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6092 emulator_invalidate_register_cache(ctxt);
6096 /* Save the faulting GPA (cr2) in the address field */
6097 ctxt->exception.address = cr2;
6099 r = x86_emulate_insn(ctxt);
6101 if (r == EMULATION_INTERCEPTED)
6102 return EMULATE_DONE;
6104 if (r == EMULATION_FAILED) {
6105 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6107 return EMULATE_DONE;
6109 return handle_emulation_failure(vcpu, emulation_type);
6112 if (ctxt->have_exception) {
6114 if (inject_emulated_exception(vcpu))
6116 } else if (vcpu->arch.pio.count) {
6117 if (!vcpu->arch.pio.in) {
6118 /* FIXME: return into emulator if single-stepping. */
6119 vcpu->arch.pio.count = 0;
6122 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6124 r = EMULATE_USER_EXIT;
6125 } else if (vcpu->mmio_needed) {
6126 if (!vcpu->mmio_is_write)
6128 r = EMULATE_USER_EXIT;
6129 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6130 } else if (r == EMULATION_RESTART)
6136 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6137 toggle_interruptibility(vcpu, ctxt->interruptibility);
6138 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6139 kvm_rip_write(vcpu, ctxt->eip);
6140 if (r == EMULATE_DONE &&
6141 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
6142 kvm_vcpu_do_singlestep(vcpu, &r);
6143 if (!ctxt->have_exception ||
6144 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6145 __kvm_set_rflags(vcpu, ctxt->eflags);
6148 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6149 * do nothing, and it will be requested again as soon as
6150 * the shadow expires. But we still need to check here,
6151 * because POPF has no interrupt shadow.
6153 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6154 kvm_make_request(KVM_REQ_EVENT, vcpu);
6156 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6160 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
6162 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6163 unsigned short port)
6165 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6166 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6167 size, port, &val, 1);
6168 /* do not return to emulator after return from userspace */
6169 vcpu->arch.pio.count = 0;
6173 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6177 /* We should only ever be called with arch.pio.count equal to 1 */
6178 BUG_ON(vcpu->arch.pio.count != 1);
6180 /* For size less than 4 we merge, else we zero extend */
6181 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6185 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6186 * the copy and tracing
6188 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6189 vcpu->arch.pio.port, &val, 1);
6190 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6195 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6196 unsigned short port)
6201 /* For size less than 4 we merge, else we zero extend */
6202 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6204 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6207 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6211 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6216 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6218 int ret = kvm_skip_emulated_instruction(vcpu);
6221 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6222 * KVM_EXIT_DEBUG here.
6225 return kvm_fast_pio_in(vcpu, size, port) && ret;
6227 return kvm_fast_pio_out(vcpu, size, port) && ret;
6229 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6231 static int kvmclock_cpu_down_prep(unsigned int cpu)
6233 __this_cpu_write(cpu_tsc_khz, 0);
6237 static void tsc_khz_changed(void *data)
6239 struct cpufreq_freqs *freq = data;
6240 unsigned long khz = 0;
6244 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6245 khz = cpufreq_quick_get(raw_smp_processor_id());
6248 __this_cpu_write(cpu_tsc_khz, khz);
6251 #ifdef CONFIG_X86_64
6252 static void kvm_hyperv_tsc_notifier(void)
6255 struct kvm_vcpu *vcpu;
6258 spin_lock(&kvm_lock);
6259 list_for_each_entry(kvm, &vm_list, vm_list)
6260 kvm_make_mclock_inprogress_request(kvm);
6262 hyperv_stop_tsc_emulation();
6264 /* TSC frequency always matches when on Hyper-V */
6265 for_each_present_cpu(cpu)
6266 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6267 kvm_max_guest_tsc_khz = tsc_khz;
6269 list_for_each_entry(kvm, &vm_list, vm_list) {
6270 struct kvm_arch *ka = &kvm->arch;
6272 spin_lock(&ka->pvclock_gtod_sync_lock);
6274 pvclock_update_vm_gtod_copy(kvm);
6276 kvm_for_each_vcpu(cpu, vcpu, kvm)
6277 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6279 kvm_for_each_vcpu(cpu, vcpu, kvm)
6280 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6282 spin_unlock(&ka->pvclock_gtod_sync_lock);
6284 spin_unlock(&kvm_lock);
6288 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6291 struct cpufreq_freqs *freq = data;
6293 struct kvm_vcpu *vcpu;
6294 int i, send_ipi = 0;
6297 * We allow guests to temporarily run on slowing clocks,
6298 * provided we notify them after, or to run on accelerating
6299 * clocks, provided we notify them before. Thus time never
6302 * However, we have a problem. We can't atomically update
6303 * the frequency of a given CPU from this function; it is
6304 * merely a notifier, which can be called from any CPU.
6305 * Changing the TSC frequency at arbitrary points in time
6306 * requires a recomputation of local variables related to
6307 * the TSC for each VCPU. We must flag these local variables
6308 * to be updated and be sure the update takes place with the
6309 * new frequency before any guests proceed.
6311 * Unfortunately, the combination of hotplug CPU and frequency
6312 * change creates an intractable locking scenario; the order
6313 * of when these callouts happen is undefined with respect to
6314 * CPU hotplug, and they can race with each other. As such,
6315 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6316 * undefined; you can actually have a CPU frequency change take
6317 * place in between the computation of X and the setting of the
6318 * variable. To protect against this problem, all updates of
6319 * the per_cpu tsc_khz variable are done in an interrupt
6320 * protected IPI, and all callers wishing to update the value
6321 * must wait for a synchronous IPI to complete (which is trivial
6322 * if the caller is on the CPU already). This establishes the
6323 * necessary total order on variable updates.
6325 * Note that because a guest time update may take place
6326 * anytime after the setting of the VCPU's request bit, the
6327 * correct TSC value must be set before the request. However,
6328 * to ensure the update actually makes it to any guest which
6329 * starts running in hardware virtualization between the set
6330 * and the acquisition of the spinlock, we must also ping the
6331 * CPU after setting the request bit.
6335 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6337 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6340 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6342 spin_lock(&kvm_lock);
6343 list_for_each_entry(kvm, &vm_list, vm_list) {
6344 kvm_for_each_vcpu(i, vcpu, kvm) {
6345 if (vcpu->cpu != freq->cpu)
6347 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6348 if (vcpu->cpu != smp_processor_id())
6352 spin_unlock(&kvm_lock);
6354 if (freq->old < freq->new && send_ipi) {
6356 * We upscale the frequency. Must make the guest
6357 * doesn't see old kvmclock values while running with
6358 * the new frequency, otherwise we risk the guest sees
6359 * time go backwards.
6361 * In case we update the frequency for another cpu
6362 * (which might be in guest context) send an interrupt
6363 * to kick the cpu out of guest context. Next time
6364 * guest context is entered kvmclock will be updated,
6365 * so the guest will not see stale values.
6367 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6372 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6373 .notifier_call = kvmclock_cpufreq_notifier
6376 static int kvmclock_cpu_online(unsigned int cpu)
6378 tsc_khz_changed(NULL);
6382 static void kvm_timer_init(void)
6384 max_tsc_khz = tsc_khz;
6386 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6387 #ifdef CONFIG_CPU_FREQ
6388 struct cpufreq_policy policy;
6391 memset(&policy, 0, sizeof(policy));
6393 cpufreq_get_policy(&policy, cpu);
6394 if (policy.cpuinfo.max_freq)
6395 max_tsc_khz = policy.cpuinfo.max_freq;
6398 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6399 CPUFREQ_TRANSITION_NOTIFIER);
6401 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6403 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6404 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6407 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6408 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6410 int kvm_is_in_guest(void)
6412 return __this_cpu_read(current_vcpu) != NULL;
6415 static int kvm_is_user_mode(void)
6419 if (__this_cpu_read(current_vcpu))
6420 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6422 return user_mode != 0;
6425 static unsigned long kvm_get_guest_ip(void)
6427 unsigned long ip = 0;
6429 if (__this_cpu_read(current_vcpu))
6430 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6435 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6436 .is_in_guest = kvm_is_in_guest,
6437 .is_user_mode = kvm_is_user_mode,
6438 .get_guest_ip = kvm_get_guest_ip,
6441 static void kvm_set_mmio_spte_mask(void)
6444 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6447 * Set the reserved bits and the present bit of an paging-structure
6448 * entry to generate page fault with PFER.RSV = 1.
6450 /* Mask the reserved physical address bits. */
6451 mask = rsvd_bits(maxphyaddr, 51);
6453 /* Set the present bit. */
6456 #ifdef CONFIG_X86_64
6458 * If reserved bit is not supported, clear the present bit to disable
6461 if (maxphyaddr == 52)
6465 kvm_mmu_set_mmio_spte_mask(mask, mask);
6468 #ifdef CONFIG_X86_64
6469 static void pvclock_gtod_update_fn(struct work_struct *work)
6473 struct kvm_vcpu *vcpu;
6476 spin_lock(&kvm_lock);
6477 list_for_each_entry(kvm, &vm_list, vm_list)
6478 kvm_for_each_vcpu(i, vcpu, kvm)
6479 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6480 atomic_set(&kvm_guest_has_master_clock, 0);
6481 spin_unlock(&kvm_lock);
6484 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6487 * Notification about pvclock gtod data update.
6489 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6492 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6493 struct timekeeper *tk = priv;
6495 update_pvclock_gtod(tk);
6497 /* disable master clock if host does not trust, or does not
6498 * use, TSC based clocksource.
6500 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6501 atomic_read(&kvm_guest_has_master_clock) != 0)
6502 queue_work(system_long_wq, &pvclock_gtod_work);
6507 static struct notifier_block pvclock_gtod_notifier = {
6508 .notifier_call = pvclock_gtod_notify,
6512 int kvm_arch_init(void *opaque)
6515 struct kvm_x86_ops *ops = opaque;
6518 printk(KERN_ERR "kvm: already loaded the other module\n");
6523 if (!ops->cpu_has_kvm_support()) {
6524 printk(KERN_ERR "kvm: no hardware support\n");
6528 if (ops->disabled_by_bios()) {
6529 printk(KERN_ERR "kvm: disabled by bios\n");
6535 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6537 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6541 r = kvm_mmu_module_init();
6543 goto out_free_percpu;
6545 kvm_set_mmio_spte_mask();
6549 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6550 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6551 PT_PRESENT_MASK, 0, sme_me_mask);
6554 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6556 if (boot_cpu_has(X86_FEATURE_XSAVE))
6557 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6560 #ifdef CONFIG_X86_64
6561 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6563 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6564 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6570 free_percpu(shared_msrs);
6575 void kvm_arch_exit(void)
6577 #ifdef CONFIG_X86_64
6578 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6579 clear_hv_tscchange_cb();
6582 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6584 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6585 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6586 CPUFREQ_TRANSITION_NOTIFIER);
6587 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6588 #ifdef CONFIG_X86_64
6589 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6592 kvm_mmu_module_exit();
6593 free_percpu(shared_msrs);
6596 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6598 ++vcpu->stat.halt_exits;
6599 if (lapic_in_kernel(vcpu)) {
6600 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6603 vcpu->run->exit_reason = KVM_EXIT_HLT;
6607 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6609 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6611 int ret = kvm_skip_emulated_instruction(vcpu);
6613 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6614 * KVM_EXIT_DEBUG here.
6616 return kvm_vcpu_halt(vcpu) && ret;
6618 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6620 #ifdef CONFIG_X86_64
6621 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6622 unsigned long clock_type)
6624 struct kvm_clock_pairing clock_pairing;
6629 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6630 return -KVM_EOPNOTSUPP;
6632 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6633 return -KVM_EOPNOTSUPP;
6635 clock_pairing.sec = ts.tv_sec;
6636 clock_pairing.nsec = ts.tv_nsec;
6637 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6638 clock_pairing.flags = 0;
6641 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6642 sizeof(struct kvm_clock_pairing)))
6650 * kvm_pv_kick_cpu_op: Kick a vcpu.
6652 * @apicid - apicid of vcpu to be kicked.
6654 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6656 struct kvm_lapic_irq lapic_irq;
6658 lapic_irq.shorthand = 0;
6659 lapic_irq.dest_mode = 0;
6660 lapic_irq.level = 0;
6661 lapic_irq.dest_id = apicid;
6662 lapic_irq.msi_redir_hint = false;
6664 lapic_irq.delivery_mode = APIC_DM_REMRD;
6665 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6668 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6670 vcpu->arch.apicv_active = false;
6671 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6674 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6676 unsigned long nr, a0, a1, a2, a3, ret;
6679 if (kvm_hv_hypercall_enabled(vcpu->kvm)) {
6680 if (!kvm_hv_hypercall(vcpu))
6685 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6686 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6687 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6688 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6689 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6691 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6693 op_64_bit = is_64_bit_mode(vcpu);
6702 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6708 case KVM_HC_VAPIC_POLL_IRQ:
6711 case KVM_HC_KICK_CPU:
6712 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6715 #ifdef CONFIG_X86_64
6716 case KVM_HC_CLOCK_PAIRING:
6717 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6727 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6730 ++vcpu->stat.hypercalls;
6731 return kvm_skip_emulated_instruction(vcpu);
6733 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6735 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6737 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6738 char instruction[3];
6739 unsigned long rip = kvm_rip_read(vcpu);
6741 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6743 return emulator_write_emulated(ctxt, rip, instruction, 3,
6747 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6749 return vcpu->run->request_interrupt_window &&
6750 likely(!pic_in_kernel(vcpu->kvm));
6753 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6755 struct kvm_run *kvm_run = vcpu->run;
6757 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6758 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6759 kvm_run->cr8 = kvm_get_cr8(vcpu);
6760 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6761 kvm_run->ready_for_interrupt_injection =
6762 pic_in_kernel(vcpu->kvm) ||
6763 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6766 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6770 if (!kvm_x86_ops->update_cr8_intercept)
6773 if (!lapic_in_kernel(vcpu))
6776 if (vcpu->arch.apicv_active)
6779 if (!vcpu->arch.apic->vapic_addr)
6780 max_irr = kvm_lapic_find_highest_irr(vcpu);
6787 tpr = kvm_lapic_get_cr8(vcpu);
6789 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6792 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6796 /* try to reinject previous events if any */
6798 if (vcpu->arch.exception.injected)
6799 kvm_x86_ops->queue_exception(vcpu);
6801 * Do not inject an NMI or interrupt if there is a pending
6802 * exception. Exceptions and interrupts are recognized at
6803 * instruction boundaries, i.e. the start of an instruction.
6804 * Trap-like exceptions, e.g. #DB, have higher priority than
6805 * NMIs and interrupts, i.e. traps are recognized before an
6806 * NMI/interrupt that's pending on the same instruction.
6807 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
6808 * priority, but are only generated (pended) during instruction
6809 * execution, i.e. a pending fault-like exception means the
6810 * fault occurred on the *previous* instruction and must be
6811 * serviced prior to recognizing any new events in order to
6812 * fully complete the previous instruction.
6814 else if (!vcpu->arch.exception.pending) {
6815 if (vcpu->arch.nmi_injected)
6816 kvm_x86_ops->set_nmi(vcpu);
6817 else if (vcpu->arch.interrupt.injected)
6818 kvm_x86_ops->set_irq(vcpu);
6822 * Call check_nested_events() even if we reinjected a previous event
6823 * in order for caller to determine if it should require immediate-exit
6824 * from L2 to L1 due to pending L1 events which require exit
6827 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6828 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6833 /* try to inject new event if pending */
6834 if (vcpu->arch.exception.pending) {
6835 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6836 vcpu->arch.exception.has_error_code,
6837 vcpu->arch.exception.error_code);
6839 WARN_ON_ONCE(vcpu->arch.exception.injected);
6840 vcpu->arch.exception.pending = false;
6841 vcpu->arch.exception.injected = true;
6843 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6844 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6847 if (vcpu->arch.exception.nr == DB_VECTOR &&
6848 (vcpu->arch.dr7 & DR7_GD)) {
6849 vcpu->arch.dr7 &= ~DR7_GD;
6850 kvm_update_dr7(vcpu);
6853 kvm_x86_ops->queue_exception(vcpu);
6856 /* Don't consider new event if we re-injected an event */
6857 if (kvm_event_needs_reinjection(vcpu))
6860 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
6861 kvm_x86_ops->smi_allowed(vcpu)) {
6862 vcpu->arch.smi_pending = false;
6863 ++vcpu->arch.smi_count;
6865 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6866 --vcpu->arch.nmi_pending;
6867 vcpu->arch.nmi_injected = true;
6868 kvm_x86_ops->set_nmi(vcpu);
6869 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6871 * Because interrupts can be injected asynchronously, we are
6872 * calling check_nested_events again here to avoid a race condition.
6873 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6874 * proposal and current concerns. Perhaps we should be setting
6875 * KVM_REQ_EVENT only on certain events and not unconditionally?
6877 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6878 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6882 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6883 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6885 kvm_x86_ops->set_irq(vcpu);
6892 static void process_nmi(struct kvm_vcpu *vcpu)
6897 * x86 is limited to one NMI running, and one NMI pending after it.
6898 * If an NMI is already in progress, limit further NMIs to just one.
6899 * Otherwise, allow two (and we'll inject the first one immediately).
6901 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6904 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6905 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6906 kvm_make_request(KVM_REQ_EVENT, vcpu);
6909 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6912 flags |= seg->g << 23;
6913 flags |= seg->db << 22;
6914 flags |= seg->l << 21;
6915 flags |= seg->avl << 20;
6916 flags |= seg->present << 15;
6917 flags |= seg->dpl << 13;
6918 flags |= seg->s << 12;
6919 flags |= seg->type << 8;
6923 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6925 struct kvm_segment seg;
6928 kvm_get_segment(vcpu, &seg, n);
6929 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6932 offset = 0x7f84 + n * 12;
6934 offset = 0x7f2c + (n - 3) * 12;
6936 put_smstate(u32, buf, offset + 8, seg.base);
6937 put_smstate(u32, buf, offset + 4, seg.limit);
6938 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6941 #ifdef CONFIG_X86_64
6942 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6944 struct kvm_segment seg;
6948 kvm_get_segment(vcpu, &seg, n);
6949 offset = 0x7e00 + n * 16;
6951 flags = enter_smm_get_segment_flags(&seg) >> 8;
6952 put_smstate(u16, buf, offset, seg.selector);
6953 put_smstate(u16, buf, offset + 2, flags);
6954 put_smstate(u32, buf, offset + 4, seg.limit);
6955 put_smstate(u64, buf, offset + 8, seg.base);
6959 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6962 struct kvm_segment seg;
6966 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6967 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6968 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6969 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6971 for (i = 0; i < 8; i++)
6972 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6974 kvm_get_dr(vcpu, 6, &val);
6975 put_smstate(u32, buf, 0x7fcc, (u32)val);
6976 kvm_get_dr(vcpu, 7, &val);
6977 put_smstate(u32, buf, 0x7fc8, (u32)val);
6979 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6980 put_smstate(u32, buf, 0x7fc4, seg.selector);
6981 put_smstate(u32, buf, 0x7f64, seg.base);
6982 put_smstate(u32, buf, 0x7f60, seg.limit);
6983 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6985 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6986 put_smstate(u32, buf, 0x7fc0, seg.selector);
6987 put_smstate(u32, buf, 0x7f80, seg.base);
6988 put_smstate(u32, buf, 0x7f7c, seg.limit);
6989 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6991 kvm_x86_ops->get_gdt(vcpu, &dt);
6992 put_smstate(u32, buf, 0x7f74, dt.address);
6993 put_smstate(u32, buf, 0x7f70, dt.size);
6995 kvm_x86_ops->get_idt(vcpu, &dt);
6996 put_smstate(u32, buf, 0x7f58, dt.address);
6997 put_smstate(u32, buf, 0x7f54, dt.size);
6999 for (i = 0; i < 6; i++)
7000 enter_smm_save_seg_32(vcpu, buf, i);
7002 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7005 put_smstate(u32, buf, 0x7efc, 0x00020000);
7006 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7009 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7011 #ifdef CONFIG_X86_64
7013 struct kvm_segment seg;
7017 for (i = 0; i < 16; i++)
7018 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7020 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7021 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7023 kvm_get_dr(vcpu, 6, &val);
7024 put_smstate(u64, buf, 0x7f68, val);
7025 kvm_get_dr(vcpu, 7, &val);
7026 put_smstate(u64, buf, 0x7f60, val);
7028 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7029 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7030 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7032 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7035 put_smstate(u32, buf, 0x7efc, 0x00020064);
7037 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7039 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7040 put_smstate(u16, buf, 0x7e90, seg.selector);
7041 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7042 put_smstate(u32, buf, 0x7e94, seg.limit);
7043 put_smstate(u64, buf, 0x7e98, seg.base);
7045 kvm_x86_ops->get_idt(vcpu, &dt);
7046 put_smstate(u32, buf, 0x7e84, dt.size);
7047 put_smstate(u64, buf, 0x7e88, dt.address);
7049 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7050 put_smstate(u16, buf, 0x7e70, seg.selector);
7051 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7052 put_smstate(u32, buf, 0x7e74, seg.limit);
7053 put_smstate(u64, buf, 0x7e78, seg.base);
7055 kvm_x86_ops->get_gdt(vcpu, &dt);
7056 put_smstate(u32, buf, 0x7e64, dt.size);
7057 put_smstate(u64, buf, 0x7e68, dt.address);
7059 for (i = 0; i < 6; i++)
7060 enter_smm_save_seg_64(vcpu, buf, i);
7066 static void enter_smm(struct kvm_vcpu *vcpu)
7068 struct kvm_segment cs, ds;
7073 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7074 memset(buf, 0, 512);
7075 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7076 enter_smm_save_state_64(vcpu, buf);
7078 enter_smm_save_state_32(vcpu, buf);
7081 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7082 * vCPU state (e.g. leave guest mode) after we've saved the state into
7083 * the SMM state-save area.
7085 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7087 vcpu->arch.hflags |= HF_SMM_MASK;
7088 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7090 if (kvm_x86_ops->get_nmi_mask(vcpu))
7091 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7093 kvm_x86_ops->set_nmi_mask(vcpu, true);
7095 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7096 kvm_rip_write(vcpu, 0x8000);
7098 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7099 kvm_x86_ops->set_cr0(vcpu, cr0);
7100 vcpu->arch.cr0 = cr0;
7102 kvm_x86_ops->set_cr4(vcpu, 0);
7104 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7105 dt.address = dt.size = 0;
7106 kvm_x86_ops->set_idt(vcpu, &dt);
7108 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7110 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7111 cs.base = vcpu->arch.smbase;
7116 cs.limit = ds.limit = 0xffffffff;
7117 cs.type = ds.type = 0x3;
7118 cs.dpl = ds.dpl = 0;
7123 cs.avl = ds.avl = 0;
7124 cs.present = ds.present = 1;
7125 cs.unusable = ds.unusable = 0;
7126 cs.padding = ds.padding = 0;
7128 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7129 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7130 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7131 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7132 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7133 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7135 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7136 kvm_x86_ops->set_efer(vcpu, 0);
7138 kvm_update_cpuid(vcpu);
7139 kvm_mmu_reset_context(vcpu);
7142 static void process_smi(struct kvm_vcpu *vcpu)
7144 vcpu->arch.smi_pending = true;
7145 kvm_make_request(KVM_REQ_EVENT, vcpu);
7148 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7150 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7153 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7155 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7158 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7160 if (irqchip_split(vcpu->kvm))
7161 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7163 if (vcpu->arch.apicv_active)
7164 kvm_x86_ops->sync_pir_to_irr(vcpu);
7165 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7168 if (is_guest_mode(vcpu))
7169 vcpu->arch.load_eoi_exitmap_pending = true;
7171 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7174 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7176 u64 eoi_exit_bitmap[4];
7178 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7181 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7182 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7183 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7186 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7187 unsigned long start, unsigned long end)
7189 unsigned long apic_address;
7192 * The physical address of apic access page is stored in the VMCS.
7193 * Update it when it becomes invalid.
7195 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7196 if (start <= apic_address && apic_address < end)
7197 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7200 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7202 struct page *page = NULL;
7204 if (!lapic_in_kernel(vcpu))
7207 if (!kvm_x86_ops->set_apic_access_page_addr)
7210 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7211 if (is_error_page(page))
7213 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7216 * Do not pin apic access page in memory, the MMU notifier
7217 * will call us again if it is migrated or swapped out.
7221 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7224 * Returns 1 to let vcpu_run() continue the guest execution loop without
7225 * exiting to the userspace. Otherwise, the value will be returned to the
7228 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7232 dm_request_for_irq_injection(vcpu) &&
7233 kvm_cpu_accept_dm_intr(vcpu);
7235 bool req_immediate_exit = false;
7237 if (kvm_request_pending(vcpu)) {
7238 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7239 kvm_mmu_unload(vcpu);
7240 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7241 __kvm_migrate_timers(vcpu);
7242 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7243 kvm_gen_update_masterclock(vcpu->kvm);
7244 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7245 kvm_gen_kvmclock_update(vcpu);
7246 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7247 r = kvm_guest_time_update(vcpu);
7251 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7252 kvm_mmu_sync_roots(vcpu);
7253 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7254 kvm_vcpu_flush_tlb(vcpu, true);
7255 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7256 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7260 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7261 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7262 vcpu->mmio_needed = 0;
7266 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7267 /* Page is swapped out. Do synthetic halt */
7268 vcpu->arch.apf.halted = true;
7272 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7273 record_steal_time(vcpu);
7274 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7276 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7278 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7279 kvm_pmu_handle_event(vcpu);
7280 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7281 kvm_pmu_deliver_pmi(vcpu);
7282 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7283 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7284 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7285 vcpu->arch.ioapic_handled_vectors)) {
7286 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7287 vcpu->run->eoi.vector =
7288 vcpu->arch.pending_ioapic_eoi;
7293 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7294 vcpu_scan_ioapic(vcpu);
7295 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7296 vcpu_load_eoi_exitmap(vcpu);
7297 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7298 kvm_vcpu_reload_apic_access_page(vcpu);
7299 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7300 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7301 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7305 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7306 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7307 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7311 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7312 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7313 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7319 * KVM_REQ_HV_STIMER has to be processed after
7320 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7321 * depend on the guest clock being up-to-date
7323 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7324 kvm_hv_process_stimers(vcpu);
7327 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7328 ++vcpu->stat.req_event;
7329 kvm_apic_accept_events(vcpu);
7330 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7335 if (inject_pending_event(vcpu, req_int_win) != 0)
7336 req_immediate_exit = true;
7338 /* Enable SMI/NMI/IRQ window open exits if needed.
7340 * SMIs have three cases:
7341 * 1) They can be nested, and then there is nothing to
7342 * do here because RSM will cause a vmexit anyway.
7343 * 2) There is an ISA-specific reason why SMI cannot be
7344 * injected, and the moment when this changes can be
7346 * 3) Or the SMI can be pending because
7347 * inject_pending_event has completed the injection
7348 * of an IRQ or NMI from the previous vmexit, and
7349 * then we request an immediate exit to inject the
7352 if (vcpu->arch.smi_pending && !is_smm(vcpu))
7353 if (!kvm_x86_ops->enable_smi_window(vcpu))
7354 req_immediate_exit = true;
7355 if (vcpu->arch.nmi_pending)
7356 kvm_x86_ops->enable_nmi_window(vcpu);
7357 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7358 kvm_x86_ops->enable_irq_window(vcpu);
7359 WARN_ON(vcpu->arch.exception.pending);
7362 if (kvm_lapic_enabled(vcpu)) {
7363 update_cr8_intercept(vcpu);
7364 kvm_lapic_sync_to_vapic(vcpu);
7368 r = kvm_mmu_reload(vcpu);
7370 goto cancel_injection;
7375 kvm_x86_ops->prepare_guest_switch(vcpu);
7378 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7379 * IPI are then delayed after guest entry, which ensures that they
7380 * result in virtual interrupt delivery.
7382 local_irq_disable();
7383 vcpu->mode = IN_GUEST_MODE;
7385 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7388 * 1) We should set ->mode before checking ->requests. Please see
7389 * the comment in kvm_vcpu_exiting_guest_mode().
7391 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7392 * pairs with the memory barrier implicit in pi_test_and_set_on
7393 * (see vmx_deliver_posted_interrupt).
7395 * 3) This also orders the write to mode from any reads to the page
7396 * tables done while the VCPU is running. Please see the comment
7397 * in kvm_flush_remote_tlbs.
7399 smp_mb__after_srcu_read_unlock();
7402 * This handles the case where a posted interrupt was
7403 * notified with kvm_vcpu_kick.
7405 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7406 kvm_x86_ops->sync_pir_to_irr(vcpu);
7408 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7409 || need_resched() || signal_pending(current)) {
7410 vcpu->mode = OUTSIDE_GUEST_MODE;
7414 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7416 goto cancel_injection;
7419 kvm_load_guest_xcr0(vcpu);
7421 if (req_immediate_exit) {
7422 kvm_make_request(KVM_REQ_EVENT, vcpu);
7423 smp_send_reschedule(vcpu->cpu);
7426 trace_kvm_entry(vcpu->vcpu_id);
7427 if (lapic_timer_advance_ns)
7428 wait_lapic_expire(vcpu);
7429 guest_enter_irqoff();
7431 if (unlikely(vcpu->arch.switch_db_regs)) {
7433 set_debugreg(vcpu->arch.eff_db[0], 0);
7434 set_debugreg(vcpu->arch.eff_db[1], 1);
7435 set_debugreg(vcpu->arch.eff_db[2], 2);
7436 set_debugreg(vcpu->arch.eff_db[3], 3);
7437 set_debugreg(vcpu->arch.dr6, 6);
7438 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7441 kvm_x86_ops->run(vcpu);
7444 * Do this here before restoring debug registers on the host. And
7445 * since we do this before handling the vmexit, a DR access vmexit
7446 * can (a) read the correct value of the debug registers, (b) set
7447 * KVM_DEBUGREG_WONT_EXIT again.
7449 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7450 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7451 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7452 kvm_update_dr0123(vcpu);
7453 kvm_update_dr6(vcpu);
7454 kvm_update_dr7(vcpu);
7455 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7459 * If the guest has used debug registers, at least dr7
7460 * will be disabled while returning to the host.
7461 * If we don't have active breakpoints in the host, we don't
7462 * care about the messed up debug address registers. But if
7463 * we have some of them active, restore the old state.
7465 if (hw_breakpoint_active())
7466 hw_breakpoint_restore();
7468 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7470 vcpu->mode = OUTSIDE_GUEST_MODE;
7473 kvm_put_guest_xcr0(vcpu);
7475 kvm_before_interrupt(vcpu);
7476 kvm_x86_ops->handle_external_intr(vcpu);
7477 kvm_after_interrupt(vcpu);
7481 guest_exit_irqoff();
7486 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7489 * Profile KVM exit RIPs:
7491 if (unlikely(prof_on == KVM_PROFILING)) {
7492 unsigned long rip = kvm_rip_read(vcpu);
7493 profile_hit(KVM_PROFILING, (void *)rip);
7496 if (unlikely(vcpu->arch.tsc_always_catchup))
7497 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7499 if (vcpu->arch.apic_attention)
7500 kvm_lapic_sync_from_vapic(vcpu);
7502 vcpu->arch.gpa_available = false;
7503 r = kvm_x86_ops->handle_exit(vcpu);
7507 kvm_x86_ops->cancel_injection(vcpu);
7508 if (unlikely(vcpu->arch.apic_attention))
7509 kvm_lapic_sync_from_vapic(vcpu);
7514 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7516 if (!kvm_arch_vcpu_runnable(vcpu) &&
7517 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7518 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7519 kvm_vcpu_block(vcpu);
7520 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7522 if (kvm_x86_ops->post_block)
7523 kvm_x86_ops->post_block(vcpu);
7525 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7529 kvm_apic_accept_events(vcpu);
7530 switch(vcpu->arch.mp_state) {
7531 case KVM_MP_STATE_HALTED:
7532 vcpu->arch.pv.pv_unhalted = false;
7533 vcpu->arch.mp_state =
7534 KVM_MP_STATE_RUNNABLE;
7535 case KVM_MP_STATE_RUNNABLE:
7536 vcpu->arch.apf.halted = false;
7538 case KVM_MP_STATE_INIT_RECEIVED:
7547 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7549 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7550 kvm_x86_ops->check_nested_events(vcpu, false);
7552 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7553 !vcpu->arch.apf.halted);
7556 static int vcpu_run(struct kvm_vcpu *vcpu)
7559 struct kvm *kvm = vcpu->kvm;
7561 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7564 if (kvm_vcpu_running(vcpu)) {
7565 r = vcpu_enter_guest(vcpu);
7567 r = vcpu_block(kvm, vcpu);
7573 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7574 if (kvm_cpu_has_pending_timer(vcpu))
7575 kvm_inject_pending_timer_irqs(vcpu);
7577 if (dm_request_for_irq_injection(vcpu) &&
7578 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7580 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7581 ++vcpu->stat.request_irq_exits;
7585 kvm_check_async_pf_completion(vcpu);
7587 if (signal_pending(current)) {
7589 vcpu->run->exit_reason = KVM_EXIT_INTR;
7590 ++vcpu->stat.signal_exits;
7593 if (need_resched()) {
7594 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7596 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7600 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7605 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7608 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7609 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7610 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7611 if (r != EMULATE_DONE)
7616 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7618 BUG_ON(!vcpu->arch.pio.count);
7620 return complete_emulated_io(vcpu);
7624 * Implements the following, as a state machine:
7628 * for each mmio piece in the fragment
7636 * for each mmio piece in the fragment
7641 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7643 struct kvm_run *run = vcpu->run;
7644 struct kvm_mmio_fragment *frag;
7647 BUG_ON(!vcpu->mmio_needed);
7649 /* Complete previous fragment */
7650 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7651 len = min(8u, frag->len);
7652 if (!vcpu->mmio_is_write)
7653 memcpy(frag->data, run->mmio.data, len);
7655 if (frag->len <= 8) {
7656 /* Switch to the next fragment. */
7658 vcpu->mmio_cur_fragment++;
7660 /* Go forward to the next mmio piece. */
7666 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7667 vcpu->mmio_needed = 0;
7669 /* FIXME: return into emulator if single-stepping. */
7670 if (vcpu->mmio_is_write)
7672 vcpu->mmio_read_completed = 1;
7673 return complete_emulated_io(vcpu);
7676 run->exit_reason = KVM_EXIT_MMIO;
7677 run->mmio.phys_addr = frag->gpa;
7678 if (vcpu->mmio_is_write)
7679 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7680 run->mmio.len = min(8u, frag->len);
7681 run->mmio.is_write = vcpu->mmio_is_write;
7682 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7686 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7691 kvm_sigset_activate(vcpu);
7692 kvm_load_guest_fpu(vcpu);
7694 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7695 if (kvm_run->immediate_exit) {
7699 kvm_vcpu_block(vcpu);
7700 kvm_apic_accept_events(vcpu);
7701 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7703 if (signal_pending(current)) {
7705 vcpu->run->exit_reason = KVM_EXIT_INTR;
7706 ++vcpu->stat.signal_exits;
7711 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
7716 if (vcpu->run->kvm_dirty_regs) {
7717 r = sync_regs(vcpu);
7722 /* re-sync apic's tpr */
7723 if (!lapic_in_kernel(vcpu)) {
7724 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7730 if (unlikely(vcpu->arch.complete_userspace_io)) {
7731 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7732 vcpu->arch.complete_userspace_io = NULL;
7737 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7739 if (kvm_run->immediate_exit)
7745 kvm_put_guest_fpu(vcpu);
7746 if (vcpu->run->kvm_valid_regs)
7748 post_kvm_run_save(vcpu);
7749 kvm_sigset_deactivate(vcpu);
7755 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7757 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7759 * We are here if userspace calls get_regs() in the middle of
7760 * instruction emulation. Registers state needs to be copied
7761 * back from emulation context to vcpu. Userspace shouldn't do
7762 * that usually, but some bad designed PV devices (vmware
7763 * backdoor interface) need this to work
7765 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7766 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7768 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7769 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7770 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7771 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7772 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7773 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7774 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7775 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7776 #ifdef CONFIG_X86_64
7777 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7778 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7779 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7780 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7781 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7782 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7783 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7784 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7787 regs->rip = kvm_rip_read(vcpu);
7788 regs->rflags = kvm_get_rflags(vcpu);
7791 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7794 __get_regs(vcpu, regs);
7799 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7801 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7802 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7804 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7805 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7806 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7807 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7808 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7809 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7810 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7811 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7812 #ifdef CONFIG_X86_64
7813 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7814 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7815 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7816 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7817 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7818 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7819 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7820 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7823 kvm_rip_write(vcpu, regs->rip);
7824 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
7826 vcpu->arch.exception.pending = false;
7828 kvm_make_request(KVM_REQ_EVENT, vcpu);
7831 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7834 __set_regs(vcpu, regs);
7839 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7841 struct kvm_segment cs;
7843 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7847 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7849 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7853 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7854 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7855 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7856 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7857 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7858 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7860 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7861 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7863 kvm_x86_ops->get_idt(vcpu, &dt);
7864 sregs->idt.limit = dt.size;
7865 sregs->idt.base = dt.address;
7866 kvm_x86_ops->get_gdt(vcpu, &dt);
7867 sregs->gdt.limit = dt.size;
7868 sregs->gdt.base = dt.address;
7870 sregs->cr0 = kvm_read_cr0(vcpu);
7871 sregs->cr2 = vcpu->arch.cr2;
7872 sregs->cr3 = kvm_read_cr3(vcpu);
7873 sregs->cr4 = kvm_read_cr4(vcpu);
7874 sregs->cr8 = kvm_get_cr8(vcpu);
7875 sregs->efer = vcpu->arch.efer;
7876 sregs->apic_base = kvm_get_apic_base(vcpu);
7878 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7880 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
7881 set_bit(vcpu->arch.interrupt.nr,
7882 (unsigned long *)sregs->interrupt_bitmap);
7885 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7886 struct kvm_sregs *sregs)
7889 __get_sregs(vcpu, sregs);
7894 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7895 struct kvm_mp_state *mp_state)
7899 kvm_apic_accept_events(vcpu);
7900 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7901 vcpu->arch.pv.pv_unhalted)
7902 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7904 mp_state->mp_state = vcpu->arch.mp_state;
7910 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7911 struct kvm_mp_state *mp_state)
7917 if (!lapic_in_kernel(vcpu) &&
7918 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7921 /* INITs are latched while in SMM */
7922 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7923 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7924 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7927 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7928 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7929 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7931 vcpu->arch.mp_state = mp_state->mp_state;
7932 kvm_make_request(KVM_REQ_EVENT, vcpu);
7940 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7941 int reason, bool has_error_code, u32 error_code)
7943 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7946 init_emulate_ctxt(vcpu);
7948 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7949 has_error_code, error_code);
7952 return EMULATE_FAIL;
7954 kvm_rip_write(vcpu, ctxt->eip);
7955 kvm_set_rflags(vcpu, ctxt->eflags);
7956 kvm_make_request(KVM_REQ_EVENT, vcpu);
7957 return EMULATE_DONE;
7959 EXPORT_SYMBOL_GPL(kvm_task_switch);
7961 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7963 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
7965 * When EFER.LME and CR0.PG are set, the processor is in
7966 * 64-bit mode (though maybe in a 32-bit code segment).
7967 * CR4.PAE and EFER.LMA must be set.
7969 if (!(sregs->cr4 & X86_CR4_PAE)
7970 || !(sregs->efer & EFER_LMA))
7974 * Not in 64-bit mode: EFER.LMA is clear and the code
7975 * segment cannot be 64-bit.
7977 if (sregs->efer & EFER_LMA || sregs->cs.l)
7984 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7986 struct msr_data apic_base_msr;
7987 int mmu_reset_needed = 0;
7988 int pending_vec, max_bits, idx;
7992 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7993 (sregs->cr4 & X86_CR4_OSXSAVE))
7996 if (kvm_valid_sregs(vcpu, sregs))
7999 apic_base_msr.data = sregs->apic_base;
8000 apic_base_msr.host_initiated = true;
8001 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8004 dt.size = sregs->idt.limit;
8005 dt.address = sregs->idt.base;
8006 kvm_x86_ops->set_idt(vcpu, &dt);
8007 dt.size = sregs->gdt.limit;
8008 dt.address = sregs->gdt.base;
8009 kvm_x86_ops->set_gdt(vcpu, &dt);
8011 vcpu->arch.cr2 = sregs->cr2;
8012 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8013 vcpu->arch.cr3 = sregs->cr3;
8014 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8016 kvm_set_cr8(vcpu, sregs->cr8);
8018 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8019 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8021 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8022 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8023 vcpu->arch.cr0 = sregs->cr0;
8025 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8026 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8027 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
8028 kvm_update_cpuid(vcpu);
8030 idx = srcu_read_lock(&vcpu->kvm->srcu);
8031 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
8032 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8033 mmu_reset_needed = 1;
8035 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8037 if (mmu_reset_needed)
8038 kvm_mmu_reset_context(vcpu);
8040 max_bits = KVM_NR_INTERRUPTS;
8041 pending_vec = find_first_bit(
8042 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8043 if (pending_vec < max_bits) {
8044 kvm_queue_interrupt(vcpu, pending_vec, false);
8045 pr_debug("Set back pending irq %d\n", pending_vec);
8048 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8049 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8050 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8051 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8052 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8053 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8055 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8056 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8058 update_cr8_intercept(vcpu);
8060 /* Older userspace won't unhalt the vcpu on reset. */
8061 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8062 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8064 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8066 kvm_make_request(KVM_REQ_EVENT, vcpu);
8073 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8074 struct kvm_sregs *sregs)
8079 ret = __set_sregs(vcpu, sregs);
8084 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8085 struct kvm_guest_debug *dbg)
8087 unsigned long rflags;
8092 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8094 if (vcpu->arch.exception.pending)
8096 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8097 kvm_queue_exception(vcpu, DB_VECTOR);
8099 kvm_queue_exception(vcpu, BP_VECTOR);
8103 * Read rflags as long as potentially injected trace flags are still
8106 rflags = kvm_get_rflags(vcpu);
8108 vcpu->guest_debug = dbg->control;
8109 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8110 vcpu->guest_debug = 0;
8112 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8113 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8114 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8115 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8117 for (i = 0; i < KVM_NR_DB_REGS; i++)
8118 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8120 kvm_update_dr7(vcpu);
8122 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8123 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8124 get_segment_base(vcpu, VCPU_SREG_CS);
8127 * Trigger an rflags update that will inject or remove the trace
8130 kvm_set_rflags(vcpu, rflags);
8132 kvm_x86_ops->update_bp_intercept(vcpu);
8142 * Translate a guest virtual address to a guest physical address.
8144 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8145 struct kvm_translation *tr)
8147 unsigned long vaddr = tr->linear_address;
8153 idx = srcu_read_lock(&vcpu->kvm->srcu);
8154 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8155 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8156 tr->physical_address = gpa;
8157 tr->valid = gpa != UNMAPPED_GVA;
8165 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8167 struct fxregs_state *fxsave;
8171 fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8172 memcpy(fpu->fpr, fxsave->st_space, 128);
8173 fpu->fcw = fxsave->cwd;
8174 fpu->fsw = fxsave->swd;
8175 fpu->ftwx = fxsave->twd;
8176 fpu->last_opcode = fxsave->fop;
8177 fpu->last_ip = fxsave->rip;
8178 fpu->last_dp = fxsave->rdp;
8179 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
8185 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8187 struct fxregs_state *fxsave;
8191 fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8193 memcpy(fxsave->st_space, fpu->fpr, 128);
8194 fxsave->cwd = fpu->fcw;
8195 fxsave->swd = fpu->fsw;
8196 fxsave->twd = fpu->ftwx;
8197 fxsave->fop = fpu->last_opcode;
8198 fxsave->rip = fpu->last_ip;
8199 fxsave->rdp = fpu->last_dp;
8200 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
8206 static void store_regs(struct kvm_vcpu *vcpu)
8208 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8210 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8211 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8213 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8214 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8216 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8217 kvm_vcpu_ioctl_x86_get_vcpu_events(
8218 vcpu, &vcpu->run->s.regs.events);
8221 static int sync_regs(struct kvm_vcpu *vcpu)
8223 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8226 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8227 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8228 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8230 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8231 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8233 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8235 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8236 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8237 vcpu, &vcpu->run->s.regs.events))
8239 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8245 static void fx_init(struct kvm_vcpu *vcpu)
8247 fpstate_init(&vcpu->arch.guest_fpu.state);
8248 if (boot_cpu_has(X86_FEATURE_XSAVES))
8249 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
8250 host_xcr0 | XSTATE_COMPACTION_ENABLED;
8253 * Ensure guest xcr0 is valid for loading
8255 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8257 vcpu->arch.cr0 |= X86_CR0_ET;
8260 /* Swap (qemu) user FPU context for the guest FPU context. */
8261 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8264 copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
8265 /* PKRU is separately restored in kvm_x86_ops->run. */
8266 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
8267 ~XFEATURE_MASK_PKRU);
8272 /* When vcpu_run ends, restore user space FPU context. */
8273 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8276 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
8277 copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
8279 ++vcpu->stat.fpu_reload;
8283 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8285 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8287 kvmclock_reset(vcpu);
8289 kvm_x86_ops->vcpu_free(vcpu);
8290 free_cpumask_var(wbinvd_dirty_mask);
8293 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8296 struct kvm_vcpu *vcpu;
8298 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8299 printk_once(KERN_WARNING
8300 "kvm: SMP vm created on host with unstable TSC; "
8301 "guest TSC will not be reliable\n");
8303 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8308 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8310 kvm_vcpu_mtrr_init(vcpu);
8312 kvm_vcpu_reset(vcpu, false);
8313 kvm_mmu_setup(vcpu);
8318 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8320 struct msr_data msr;
8321 struct kvm *kvm = vcpu->kvm;
8323 kvm_hv_vcpu_postcreate(vcpu);
8325 if (mutex_lock_killable(&vcpu->mutex))
8329 msr.index = MSR_IA32_TSC;
8330 msr.host_initiated = true;
8331 kvm_write_tsc(vcpu, &msr);
8333 mutex_unlock(&vcpu->mutex);
8335 if (!kvmclock_periodic_sync)
8338 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8339 KVMCLOCK_SYNC_PERIOD);
8342 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8344 vcpu->arch.apf.msr_val = 0;
8347 kvm_mmu_unload(vcpu);
8350 kvm_x86_ops->vcpu_free(vcpu);
8353 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8355 kvm_lapic_reset(vcpu, init_event);
8357 vcpu->arch.hflags = 0;
8359 vcpu->arch.smi_pending = 0;
8360 vcpu->arch.smi_count = 0;
8361 atomic_set(&vcpu->arch.nmi_queued, 0);
8362 vcpu->arch.nmi_pending = 0;
8363 vcpu->arch.nmi_injected = false;
8364 kvm_clear_interrupt_queue(vcpu);
8365 kvm_clear_exception_queue(vcpu);
8366 vcpu->arch.exception.pending = false;
8368 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8369 kvm_update_dr0123(vcpu);
8370 vcpu->arch.dr6 = DR6_INIT;
8371 kvm_update_dr6(vcpu);
8372 vcpu->arch.dr7 = DR7_FIXED_1;
8373 kvm_update_dr7(vcpu);
8377 kvm_make_request(KVM_REQ_EVENT, vcpu);
8378 vcpu->arch.apf.msr_val = 0;
8379 vcpu->arch.st.msr_val = 0;
8381 kvmclock_reset(vcpu);
8383 kvm_clear_async_pf_completion_queue(vcpu);
8384 kvm_async_pf_hash_reset(vcpu);
8385 vcpu->arch.apf.halted = false;
8387 if (kvm_mpx_supported()) {
8388 void *mpx_state_buffer;
8391 * To avoid have the INIT path from kvm_apic_has_events() that be
8392 * called with loaded FPU and does not let userspace fix the state.
8395 kvm_put_guest_fpu(vcpu);
8396 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8397 XFEATURE_MASK_BNDREGS);
8398 if (mpx_state_buffer)
8399 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8400 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8401 XFEATURE_MASK_BNDCSR);
8402 if (mpx_state_buffer)
8403 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8405 kvm_load_guest_fpu(vcpu);
8409 kvm_pmu_reset(vcpu);
8410 vcpu->arch.smbase = 0x30000;
8412 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8413 vcpu->arch.msr_misc_features_enables = 0;
8415 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8418 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8419 vcpu->arch.regs_avail = ~0;
8420 vcpu->arch.regs_dirty = ~0;
8422 vcpu->arch.ia32_xss = 0;
8424 kvm_x86_ops->vcpu_reset(vcpu, init_event);
8427 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8429 struct kvm_segment cs;
8431 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8432 cs.selector = vector << 8;
8433 cs.base = vector << 12;
8434 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8435 kvm_rip_write(vcpu, 0);
8438 int kvm_arch_hardware_enable(void)
8441 struct kvm_vcpu *vcpu;
8446 bool stable, backwards_tsc = false;
8448 kvm_shared_msr_cpu_online();
8449 ret = kvm_x86_ops->hardware_enable();
8453 local_tsc = rdtsc();
8454 stable = !kvm_check_tsc_unstable();
8455 list_for_each_entry(kvm, &vm_list, vm_list) {
8456 kvm_for_each_vcpu(i, vcpu, kvm) {
8457 if (!stable && vcpu->cpu == smp_processor_id())
8458 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8459 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8460 backwards_tsc = true;
8461 if (vcpu->arch.last_host_tsc > max_tsc)
8462 max_tsc = vcpu->arch.last_host_tsc;
8468 * Sometimes, even reliable TSCs go backwards. This happens on
8469 * platforms that reset TSC during suspend or hibernate actions, but
8470 * maintain synchronization. We must compensate. Fortunately, we can
8471 * detect that condition here, which happens early in CPU bringup,
8472 * before any KVM threads can be running. Unfortunately, we can't
8473 * bring the TSCs fully up to date with real time, as we aren't yet far
8474 * enough into CPU bringup that we know how much real time has actually
8475 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8476 * variables that haven't been updated yet.
8478 * So we simply find the maximum observed TSC above, then record the
8479 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8480 * the adjustment will be applied. Note that we accumulate
8481 * adjustments, in case multiple suspend cycles happen before some VCPU
8482 * gets a chance to run again. In the event that no KVM threads get a
8483 * chance to run, we will miss the entire elapsed period, as we'll have
8484 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8485 * loose cycle time. This isn't too big a deal, since the loss will be
8486 * uniform across all VCPUs (not to mention the scenario is extremely
8487 * unlikely). It is possible that a second hibernate recovery happens
8488 * much faster than a first, causing the observed TSC here to be
8489 * smaller; this would require additional padding adjustment, which is
8490 * why we set last_host_tsc to the local tsc observed here.
8492 * N.B. - this code below runs only on platforms with reliable TSC,
8493 * as that is the only way backwards_tsc is set above. Also note
8494 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8495 * have the same delta_cyc adjustment applied if backwards_tsc
8496 * is detected. Note further, this adjustment is only done once,
8497 * as we reset last_host_tsc on all VCPUs to stop this from being
8498 * called multiple times (one for each physical CPU bringup).
8500 * Platforms with unreliable TSCs don't have to deal with this, they
8501 * will be compensated by the logic in vcpu_load, which sets the TSC to
8502 * catchup mode. This will catchup all VCPUs to real time, but cannot
8503 * guarantee that they stay in perfect synchronization.
8505 if (backwards_tsc) {
8506 u64 delta_cyc = max_tsc - local_tsc;
8507 list_for_each_entry(kvm, &vm_list, vm_list) {
8508 kvm->arch.backwards_tsc_observed = true;
8509 kvm_for_each_vcpu(i, vcpu, kvm) {
8510 vcpu->arch.tsc_offset_adjustment += delta_cyc;
8511 vcpu->arch.last_host_tsc = local_tsc;
8512 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8516 * We have to disable TSC offset matching.. if you were
8517 * booting a VM while issuing an S4 host suspend....
8518 * you may have some problem. Solving this issue is
8519 * left as an exercise to the reader.
8521 kvm->arch.last_tsc_nsec = 0;
8522 kvm->arch.last_tsc_write = 0;
8529 void kvm_arch_hardware_disable(void)
8531 kvm_x86_ops->hardware_disable();
8532 drop_user_return_notifiers();
8535 int kvm_arch_hardware_setup(void)
8539 r = kvm_x86_ops->hardware_setup();
8543 if (kvm_has_tsc_control) {
8545 * Make sure the user can only configure tsc_khz values that
8546 * fit into a signed integer.
8547 * A min value is not calculated needed because it will always
8548 * be 1 on all machines.
8550 u64 max = min(0x7fffffffULL,
8551 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8552 kvm_max_guest_tsc_khz = max;
8554 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8557 kvm_init_msr_list();
8561 void kvm_arch_hardware_unsetup(void)
8563 kvm_x86_ops->hardware_unsetup();
8566 void kvm_arch_check_processor_compat(void *rtn)
8568 kvm_x86_ops->check_processor_compatibility(rtn);
8571 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8573 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8575 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8577 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8579 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8582 struct static_key kvm_no_apic_vcpu __read_mostly;
8583 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8585 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8590 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8591 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8592 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8593 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8595 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8597 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8602 vcpu->arch.pio_data = page_address(page);
8604 kvm_set_tsc_khz(vcpu, max_tsc_khz);
8606 r = kvm_mmu_create(vcpu);
8608 goto fail_free_pio_data;
8610 if (irqchip_in_kernel(vcpu->kvm)) {
8611 r = kvm_create_lapic(vcpu);
8613 goto fail_mmu_destroy;
8615 static_key_slow_inc(&kvm_no_apic_vcpu);
8617 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8619 if (!vcpu->arch.mce_banks) {
8621 goto fail_free_lapic;
8623 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8625 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8627 goto fail_free_mce_banks;
8632 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8634 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8636 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8638 kvm_async_pf_hash_reset(vcpu);
8641 vcpu->arch.pending_external_vector = -1;
8642 vcpu->arch.preempted_in_kernel = false;
8644 kvm_hv_vcpu_init(vcpu);
8648 fail_free_mce_banks:
8649 kfree(vcpu->arch.mce_banks);
8651 kvm_free_lapic(vcpu);
8653 kvm_mmu_destroy(vcpu);
8655 free_page((unsigned long)vcpu->arch.pio_data);
8660 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8664 kvm_hv_vcpu_uninit(vcpu);
8665 kvm_pmu_destroy(vcpu);
8666 kfree(vcpu->arch.mce_banks);
8667 kvm_free_lapic(vcpu);
8668 idx = srcu_read_lock(&vcpu->kvm->srcu);
8669 kvm_mmu_destroy(vcpu);
8670 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8671 free_page((unsigned long)vcpu->arch.pio_data);
8672 if (!lapic_in_kernel(vcpu))
8673 static_key_slow_dec(&kvm_no_apic_vcpu);
8676 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8678 kvm_x86_ops->sched_in(vcpu, cpu);
8681 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8686 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8687 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8688 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8689 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8690 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8692 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8693 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8694 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8695 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8696 &kvm->arch.irq_sources_bitmap);
8698 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8699 mutex_init(&kvm->arch.apic_map_lock);
8700 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8702 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8703 pvclock_update_vm_gtod_copy(kvm);
8705 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8706 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8708 kvm_hv_init_vm(kvm);
8709 kvm_page_track_init(kvm);
8710 kvm_mmu_init_vm(kvm);
8712 if (kvm_x86_ops->vm_init)
8713 return kvm_x86_ops->vm_init(kvm);
8718 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8721 kvm_mmu_unload(vcpu);
8725 static void kvm_free_vcpus(struct kvm *kvm)
8728 struct kvm_vcpu *vcpu;
8731 * Unpin any mmu pages first.
8733 kvm_for_each_vcpu(i, vcpu, kvm) {
8734 kvm_clear_async_pf_completion_queue(vcpu);
8735 kvm_unload_vcpu_mmu(vcpu);
8737 kvm_for_each_vcpu(i, vcpu, kvm)
8738 kvm_arch_vcpu_free(vcpu);
8740 mutex_lock(&kvm->lock);
8741 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8742 kvm->vcpus[i] = NULL;
8744 atomic_set(&kvm->online_vcpus, 0);
8745 mutex_unlock(&kvm->lock);
8748 void kvm_arch_sync_events(struct kvm *kvm)
8750 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8751 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8755 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8759 struct kvm_memslots *slots = kvm_memslots(kvm);
8760 struct kvm_memory_slot *slot, old;
8762 /* Called with kvm->slots_lock held. */
8763 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8766 slot = id_to_memslot(slots, id);
8772 * MAP_SHARED to prevent internal slot pages from being moved
8775 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8776 MAP_SHARED | MAP_ANONYMOUS, 0);
8777 if (IS_ERR((void *)hva))
8778 return PTR_ERR((void *)hva);
8787 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8788 struct kvm_userspace_memory_region m;
8790 m.slot = id | (i << 16);
8792 m.guest_phys_addr = gpa;
8793 m.userspace_addr = hva;
8794 m.memory_size = size;
8795 r = __kvm_set_memory_region(kvm, &m);
8801 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8805 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8807 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8811 mutex_lock(&kvm->slots_lock);
8812 r = __x86_set_memory_region(kvm, id, gpa, size);
8813 mutex_unlock(&kvm->slots_lock);
8817 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8819 void kvm_arch_destroy_vm(struct kvm *kvm)
8821 if (current->mm == kvm->mm) {
8823 * Free memory regions allocated on behalf of userspace,
8824 * unless the the memory map has changed due to process exit
8827 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8828 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8829 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8831 if (kvm_x86_ops->vm_destroy)
8832 kvm_x86_ops->vm_destroy(kvm);
8833 kvm_pic_destroy(kvm);
8834 kvm_ioapic_destroy(kvm);
8835 kvm_free_vcpus(kvm);
8836 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8837 kvm_mmu_uninit_vm(kvm);
8838 kvm_page_track_cleanup(kvm);
8839 kvm_hv_destroy_vm(kvm);
8842 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8843 struct kvm_memory_slot *dont)
8847 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8848 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8849 kvfree(free->arch.rmap[i]);
8850 free->arch.rmap[i] = NULL;
8855 if (!dont || free->arch.lpage_info[i - 1] !=
8856 dont->arch.lpage_info[i - 1]) {
8857 kvfree(free->arch.lpage_info[i - 1]);
8858 free->arch.lpage_info[i - 1] = NULL;
8862 kvm_page_track_free_memslot(free, dont);
8865 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8866 unsigned long npages)
8870 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8871 struct kvm_lpage_info *linfo;
8876 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8877 slot->base_gfn, level) + 1;
8879 slot->arch.rmap[i] =
8880 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8881 if (!slot->arch.rmap[i])
8886 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8890 slot->arch.lpage_info[i - 1] = linfo;
8892 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8893 linfo[0].disallow_lpage = 1;
8894 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8895 linfo[lpages - 1].disallow_lpage = 1;
8896 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8898 * If the gfn and userspace address are not aligned wrt each
8899 * other, or if explicitly asked to, disable large page
8900 * support for this slot
8902 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8903 !kvm_largepages_enabled()) {
8906 for (j = 0; j < lpages; ++j)
8907 linfo[j].disallow_lpage = 1;
8911 if (kvm_page_track_create_memslot(slot, npages))
8917 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8918 kvfree(slot->arch.rmap[i]);
8919 slot->arch.rmap[i] = NULL;
8923 kvfree(slot->arch.lpage_info[i - 1]);
8924 slot->arch.lpage_info[i - 1] = NULL;
8929 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8932 * memslots->generation has been incremented.
8933 * mmio generation may have reached its maximum value.
8935 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8938 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8939 struct kvm_memory_slot *memslot,
8940 const struct kvm_userspace_memory_region *mem,
8941 enum kvm_mr_change change)
8946 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8947 struct kvm_memory_slot *new)
8949 /* Still write protect RO slot */
8950 if (new->flags & KVM_MEM_READONLY) {
8951 kvm_mmu_slot_remove_write_access(kvm, new);
8956 * Call kvm_x86_ops dirty logging hooks when they are valid.
8958 * kvm_x86_ops->slot_disable_log_dirty is called when:
8960 * - KVM_MR_CREATE with dirty logging is disabled
8961 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8963 * The reason is, in case of PML, we need to set D-bit for any slots
8964 * with dirty logging disabled in order to eliminate unnecessary GPA
8965 * logging in PML buffer (and potential PML buffer full VMEXT). This
8966 * guarantees leaving PML enabled during guest's lifetime won't have
8967 * any additonal overhead from PML when guest is running with dirty
8968 * logging disabled for memory slots.
8970 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8971 * to dirty logging mode.
8973 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8975 * In case of write protect:
8977 * Write protect all pages for dirty logging.
8979 * All the sptes including the large sptes which point to this
8980 * slot are set to readonly. We can not create any new large
8981 * spte on this slot until the end of the logging.
8983 * See the comments in fast_page_fault().
8985 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8986 if (kvm_x86_ops->slot_enable_log_dirty)
8987 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8989 kvm_mmu_slot_remove_write_access(kvm, new);
8991 if (kvm_x86_ops->slot_disable_log_dirty)
8992 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8996 void kvm_arch_commit_memory_region(struct kvm *kvm,
8997 const struct kvm_userspace_memory_region *mem,
8998 const struct kvm_memory_slot *old,
8999 const struct kvm_memory_slot *new,
9000 enum kvm_mr_change change)
9002 int nr_mmu_pages = 0;
9004 if (!kvm->arch.n_requested_mmu_pages)
9005 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9008 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9011 * Dirty logging tracks sptes in 4k granularity, meaning that large
9012 * sptes have to be split. If live migration is successful, the guest
9013 * in the source machine will be destroyed and large sptes will be
9014 * created in the destination. However, if the guest continues to run
9015 * in the source machine (for example if live migration fails), small
9016 * sptes will remain around and cause bad performance.
9018 * Scan sptes if dirty logging has been stopped, dropping those
9019 * which can be collapsed into a single large-page spte. Later
9020 * page faults will create the large-page sptes.
9022 if ((change != KVM_MR_DELETE) &&
9023 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9024 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9025 kvm_mmu_zap_collapsible_sptes(kvm, new);
9028 * Set up write protection and/or dirty logging for the new slot.
9030 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9031 * been zapped so no dirty logging staff is needed for old slot. For
9032 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9033 * new and it's also covered when dealing with the new slot.
9035 * FIXME: const-ify all uses of struct kvm_memory_slot.
9037 if (change != KVM_MR_DELETE)
9038 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9041 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9043 kvm_mmu_invalidate_zap_all_pages(kvm);
9046 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9047 struct kvm_memory_slot *slot)
9049 kvm_page_track_flush_slot(kvm, slot);
9052 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9054 if (!list_empty_careful(&vcpu->async_pf.done))
9057 if (kvm_apic_has_events(vcpu))
9060 if (vcpu->arch.pv.pv_unhalted)
9063 if (vcpu->arch.exception.pending)
9066 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9067 (vcpu->arch.nmi_pending &&
9068 kvm_x86_ops->nmi_allowed(vcpu)))
9071 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9072 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9075 if (kvm_arch_interrupt_allowed(vcpu) &&
9076 kvm_cpu_has_interrupt(vcpu))
9079 if (kvm_hv_has_stimer_pending(vcpu))
9085 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9087 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9090 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9092 return vcpu->arch.preempted_in_kernel;
9095 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9097 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9100 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9102 return kvm_x86_ops->interrupt_allowed(vcpu);
9105 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9107 if (is_64_bit_mode(vcpu))
9108 return kvm_rip_read(vcpu);
9109 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9110 kvm_rip_read(vcpu));
9112 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9114 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9116 return kvm_get_linear_rip(vcpu) == linear_rip;
9118 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9120 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9122 unsigned long rflags;
9124 rflags = kvm_x86_ops->get_rflags(vcpu);
9125 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9126 rflags &= ~X86_EFLAGS_TF;
9129 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9131 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9133 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9134 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9135 rflags |= X86_EFLAGS_TF;
9136 kvm_x86_ops->set_rflags(vcpu, rflags);
9139 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9141 __kvm_set_rflags(vcpu, rflags);
9142 kvm_make_request(KVM_REQ_EVENT, vcpu);
9144 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9146 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9150 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
9154 r = kvm_mmu_reload(vcpu);
9158 if (!vcpu->arch.mmu.direct_map &&
9159 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
9162 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
9165 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9167 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9170 static inline u32 kvm_async_pf_next_probe(u32 key)
9172 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9175 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9177 u32 key = kvm_async_pf_hash_fn(gfn);
9179 while (vcpu->arch.apf.gfns[key] != ~0)
9180 key = kvm_async_pf_next_probe(key);
9182 vcpu->arch.apf.gfns[key] = gfn;
9185 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9188 u32 key = kvm_async_pf_hash_fn(gfn);
9190 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9191 (vcpu->arch.apf.gfns[key] != gfn &&
9192 vcpu->arch.apf.gfns[key] != ~0); i++)
9193 key = kvm_async_pf_next_probe(key);
9198 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9200 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9203 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9207 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9209 vcpu->arch.apf.gfns[i] = ~0;
9211 j = kvm_async_pf_next_probe(j);
9212 if (vcpu->arch.apf.gfns[j] == ~0)
9214 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9216 * k lies cyclically in ]i,j]
9218 * |....j i.k.| or |.k..j i...|
9220 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9221 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9226 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9229 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9233 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9236 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9240 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9241 struct kvm_async_pf *work)
9243 struct x86_exception fault;
9245 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9246 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9248 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9249 (vcpu->arch.apf.send_user_only &&
9250 kvm_x86_ops->get_cpl(vcpu) == 0))
9251 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9252 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9253 fault.vector = PF_VECTOR;
9254 fault.error_code_valid = true;
9255 fault.error_code = 0;
9256 fault.nested_page_fault = false;
9257 fault.address = work->arch.token;
9258 fault.async_page_fault = true;
9259 kvm_inject_page_fault(vcpu, &fault);
9263 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9264 struct kvm_async_pf *work)
9266 struct x86_exception fault;
9269 if (work->wakeup_all)
9270 work->arch.token = ~0; /* broadcast wakeup */
9272 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9273 trace_kvm_async_pf_ready(work->arch.token, work->gva);
9275 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9276 !apf_get_user(vcpu, &val)) {
9277 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9278 vcpu->arch.exception.pending &&
9279 vcpu->arch.exception.nr == PF_VECTOR &&
9280 !apf_put_user(vcpu, 0)) {
9281 vcpu->arch.exception.injected = false;
9282 vcpu->arch.exception.pending = false;
9283 vcpu->arch.exception.nr = 0;
9284 vcpu->arch.exception.has_error_code = false;
9285 vcpu->arch.exception.error_code = 0;
9286 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9287 fault.vector = PF_VECTOR;
9288 fault.error_code_valid = true;
9289 fault.error_code = 0;
9290 fault.nested_page_fault = false;
9291 fault.address = work->arch.token;
9292 fault.async_page_fault = true;
9293 kvm_inject_page_fault(vcpu, &fault);
9296 vcpu->arch.apf.halted = false;
9297 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9300 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9302 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9305 return kvm_can_do_async_pf(vcpu);
9308 void kvm_arch_start_assignment(struct kvm *kvm)
9310 atomic_inc(&kvm->arch.assigned_device_count);
9312 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9314 void kvm_arch_end_assignment(struct kvm *kvm)
9316 atomic_dec(&kvm->arch.assigned_device_count);
9318 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9320 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9322 return atomic_read(&kvm->arch.assigned_device_count);
9324 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9326 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9328 atomic_inc(&kvm->arch.noncoherent_dma_count);
9330 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9332 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9334 atomic_dec(&kvm->arch.noncoherent_dma_count);
9336 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9338 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9340 return atomic_read(&kvm->arch.noncoherent_dma_count);
9342 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9344 bool kvm_arch_has_irq_bypass(void)
9346 return kvm_x86_ops->update_pi_irte != NULL;
9349 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9350 struct irq_bypass_producer *prod)
9352 struct kvm_kernel_irqfd *irqfd =
9353 container_of(cons, struct kvm_kernel_irqfd, consumer);
9355 irqfd->producer = prod;
9357 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9358 prod->irq, irqfd->gsi, 1);
9361 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9362 struct irq_bypass_producer *prod)
9365 struct kvm_kernel_irqfd *irqfd =
9366 container_of(cons, struct kvm_kernel_irqfd, consumer);
9368 WARN_ON(irqfd->producer != prod);
9369 irqfd->producer = NULL;
9372 * When producer of consumer is unregistered, we change back to
9373 * remapped mode, so we can re-use the current implementation
9374 * when the irq is masked/disabled or the consumer side (KVM
9375 * int this case doesn't want to receive the interrupts.
9377 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9379 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9380 " fails: %d\n", irqfd->consumer.token, ret);
9383 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9384 uint32_t guest_irq, bool set)
9386 if (!kvm_x86_ops->update_pi_irte)
9389 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9392 bool kvm_vector_hashing_enabled(void)
9394 return vector_hashing;
9396 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9398 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9399 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9400 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9401 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9402 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9403 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9404 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9405 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9406 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9407 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9408 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9409 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9410 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9411 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9412 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9413 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9414 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9415 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9416 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);