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>
71 #define CREATE_TRACE_POINTS
74 #define MAX_IO_MSRS 256
75 #define KVM_MAX_MCE_BANKS 32
76 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
77 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
79 #define emul_to_vcpu(ctxt) \
80 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
83 * - enable syscall per default because its emulated by KVM
84 * - enable LME and LMA per default on 64 bit KVM
88 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
90 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
93 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
94 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
96 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
97 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
99 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
100 static void process_nmi(struct kvm_vcpu *vcpu);
101 static void enter_smm(struct kvm_vcpu *vcpu);
102 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
104 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
105 EXPORT_SYMBOL_GPL(kvm_x86_ops);
107 static bool __read_mostly ignore_msrs = 0;
108 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
110 unsigned int min_timer_period_us = 500;
111 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
113 static bool __read_mostly kvmclock_periodic_sync = true;
114 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
116 bool __read_mostly kvm_has_tsc_control;
117 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
118 u32 __read_mostly kvm_max_guest_tsc_khz;
119 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
120 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
121 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
122 u64 __read_mostly kvm_max_tsc_scaling_ratio;
123 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
124 u64 __read_mostly kvm_default_tsc_scaling_ratio;
125 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
127 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
128 static u32 __read_mostly tsc_tolerance_ppm = 250;
129 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
131 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
132 unsigned int __read_mostly lapic_timer_advance_ns = 0;
133 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
135 static bool __read_mostly vector_hashing = true;
136 module_param(vector_hashing, bool, S_IRUGO);
138 #define KVM_NR_SHARED_MSRS 16
140 struct kvm_shared_msrs_global {
142 u32 msrs[KVM_NR_SHARED_MSRS];
145 struct kvm_shared_msrs {
146 struct user_return_notifier urn;
148 struct kvm_shared_msr_values {
151 } values[KVM_NR_SHARED_MSRS];
154 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
155 static struct kvm_shared_msrs __percpu *shared_msrs;
157 struct kvm_stats_debugfs_item debugfs_entries[] = {
158 { "pf_fixed", VCPU_STAT(pf_fixed) },
159 { "pf_guest", VCPU_STAT(pf_guest) },
160 { "tlb_flush", VCPU_STAT(tlb_flush) },
161 { "invlpg", VCPU_STAT(invlpg) },
162 { "exits", VCPU_STAT(exits) },
163 { "io_exits", VCPU_STAT(io_exits) },
164 { "mmio_exits", VCPU_STAT(mmio_exits) },
165 { "signal_exits", VCPU_STAT(signal_exits) },
166 { "irq_window", VCPU_STAT(irq_window_exits) },
167 { "nmi_window", VCPU_STAT(nmi_window_exits) },
168 { "halt_exits", VCPU_STAT(halt_exits) },
169 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
170 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
171 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
172 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
173 { "hypercalls", VCPU_STAT(hypercalls) },
174 { "request_irq", VCPU_STAT(request_irq_exits) },
175 { "irq_exits", VCPU_STAT(irq_exits) },
176 { "host_state_reload", VCPU_STAT(host_state_reload) },
177 { "efer_reload", VCPU_STAT(efer_reload) },
178 { "fpu_reload", VCPU_STAT(fpu_reload) },
179 { "insn_emulation", VCPU_STAT(insn_emulation) },
180 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
181 { "irq_injections", VCPU_STAT(irq_injections) },
182 { "nmi_injections", VCPU_STAT(nmi_injections) },
183 { "req_event", VCPU_STAT(req_event) },
184 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
185 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
186 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
187 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
188 { "mmu_flooded", VM_STAT(mmu_flooded) },
189 { "mmu_recycled", VM_STAT(mmu_recycled) },
190 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
191 { "mmu_unsync", VM_STAT(mmu_unsync) },
192 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
193 { "largepages", VM_STAT(lpages) },
194 { "max_mmu_page_hash_collisions",
195 VM_STAT(max_mmu_page_hash_collisions) },
199 u64 __read_mostly host_xcr0;
201 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
203 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
206 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
207 vcpu->arch.apf.gfns[i] = ~0;
210 static void kvm_on_user_return(struct user_return_notifier *urn)
213 struct kvm_shared_msrs *locals
214 = container_of(urn, struct kvm_shared_msrs, urn);
215 struct kvm_shared_msr_values *values;
219 * Disabling irqs at this point since the following code could be
220 * interrupted and executed through kvm_arch_hardware_disable()
222 local_irq_save(flags);
223 if (locals->registered) {
224 locals->registered = false;
225 user_return_notifier_unregister(urn);
227 local_irq_restore(flags);
228 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
229 values = &locals->values[slot];
230 if (values->host != values->curr) {
231 wrmsrl(shared_msrs_global.msrs[slot], values->host);
232 values->curr = values->host;
237 static void shared_msr_update(unsigned slot, u32 msr)
240 unsigned int cpu = smp_processor_id();
241 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
243 /* only read, and nobody should modify it at this time,
244 * so don't need lock */
245 if (slot >= shared_msrs_global.nr) {
246 printk(KERN_ERR "kvm: invalid MSR slot!");
249 rdmsrl_safe(msr, &value);
250 smsr->values[slot].host = value;
251 smsr->values[slot].curr = value;
254 void kvm_define_shared_msr(unsigned slot, u32 msr)
256 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
257 shared_msrs_global.msrs[slot] = msr;
258 if (slot >= shared_msrs_global.nr)
259 shared_msrs_global.nr = slot + 1;
261 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
263 static void kvm_shared_msr_cpu_online(void)
267 for (i = 0; i < shared_msrs_global.nr; ++i)
268 shared_msr_update(i, shared_msrs_global.msrs[i]);
271 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
273 unsigned int cpu = smp_processor_id();
274 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
277 if (((value ^ smsr->values[slot].curr) & mask) == 0)
279 smsr->values[slot].curr = value;
280 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
284 if (!smsr->registered) {
285 smsr->urn.on_user_return = kvm_on_user_return;
286 user_return_notifier_register(&smsr->urn);
287 smsr->registered = true;
291 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
293 static void drop_user_return_notifiers(void)
295 unsigned int cpu = smp_processor_id();
296 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
298 if (smsr->registered)
299 kvm_on_user_return(&smsr->urn);
302 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
304 return vcpu->arch.apic_base;
306 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
308 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
310 u64 old_state = vcpu->arch.apic_base &
311 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
312 u64 new_state = msr_info->data &
313 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
314 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
315 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
317 if ((msr_info->data & reserved_bits) || new_state == X2APIC_ENABLE)
319 if (!msr_info->host_initiated &&
320 ((new_state == MSR_IA32_APICBASE_ENABLE &&
321 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
322 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
326 kvm_lapic_set_base(vcpu, msr_info->data);
329 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
331 asmlinkage __visible void kvm_spurious_fault(void)
333 /* Fault while not rebooting. We want the trace. */
336 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
338 #define EXCPT_BENIGN 0
339 #define EXCPT_CONTRIBUTORY 1
342 static int exception_class(int vector)
352 return EXCPT_CONTRIBUTORY;
359 #define EXCPT_FAULT 0
361 #define EXCPT_ABORT 2
362 #define EXCPT_INTERRUPT 3
364 static int exception_type(int vector)
368 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
369 return EXCPT_INTERRUPT;
373 /* #DB is trap, as instruction watchpoints are handled elsewhere */
374 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
377 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
380 /* Reserved exceptions will result in fault */
384 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
385 unsigned nr, bool has_error, u32 error_code,
391 kvm_make_request(KVM_REQ_EVENT, vcpu);
393 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
395 if (has_error && !is_protmode(vcpu))
399 * On vmentry, vcpu->arch.exception.pending is only
400 * true if an event injection was blocked by
401 * nested_run_pending. In that case, however,
402 * vcpu_enter_guest requests an immediate exit,
403 * and the guest shouldn't proceed far enough to
406 WARN_ON_ONCE(vcpu->arch.exception.pending);
407 vcpu->arch.exception.injected = true;
409 vcpu->arch.exception.pending = true;
410 vcpu->arch.exception.injected = false;
412 vcpu->arch.exception.has_error_code = has_error;
413 vcpu->arch.exception.nr = nr;
414 vcpu->arch.exception.error_code = error_code;
418 /* to check exception */
419 prev_nr = vcpu->arch.exception.nr;
420 if (prev_nr == DF_VECTOR) {
421 /* triple fault -> shutdown */
422 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
425 class1 = exception_class(prev_nr);
426 class2 = exception_class(nr);
427 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
428 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
430 * Generate double fault per SDM Table 5-5. Set
431 * exception.pending = true so that the double fault
432 * can trigger a nested vmexit.
434 vcpu->arch.exception.pending = true;
435 vcpu->arch.exception.injected = false;
436 vcpu->arch.exception.has_error_code = true;
437 vcpu->arch.exception.nr = DF_VECTOR;
438 vcpu->arch.exception.error_code = 0;
440 /* replace previous exception with a new one in a hope
441 that instruction re-execution will regenerate lost
446 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
448 kvm_multiple_exception(vcpu, nr, false, 0, false);
450 EXPORT_SYMBOL_GPL(kvm_queue_exception);
452 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
454 kvm_multiple_exception(vcpu, nr, false, 0, true);
456 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
458 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
461 kvm_inject_gp(vcpu, 0);
463 return kvm_skip_emulated_instruction(vcpu);
467 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
469 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
471 ++vcpu->stat.pf_guest;
472 vcpu->arch.exception.nested_apf =
473 is_guest_mode(vcpu) && fault->async_page_fault;
474 if (vcpu->arch.exception.nested_apf)
475 vcpu->arch.apf.nested_apf_token = fault->address;
477 vcpu->arch.cr2 = fault->address;
478 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
480 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
482 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
484 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
485 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
487 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
489 return fault->nested_page_fault;
492 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
494 atomic_inc(&vcpu->arch.nmi_queued);
495 kvm_make_request(KVM_REQ_NMI, vcpu);
497 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
499 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
501 kvm_multiple_exception(vcpu, nr, true, error_code, false);
503 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
505 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
507 kvm_multiple_exception(vcpu, nr, true, error_code, true);
509 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
512 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
513 * a #GP and return false.
515 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
517 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
519 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
522 EXPORT_SYMBOL_GPL(kvm_require_cpl);
524 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
526 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
529 kvm_queue_exception(vcpu, UD_VECTOR);
532 EXPORT_SYMBOL_GPL(kvm_require_dr);
535 * This function will be used to read from the physical memory of the currently
536 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
537 * can read from guest physical or from the guest's guest physical memory.
539 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
540 gfn_t ngfn, void *data, int offset, int len,
543 struct x86_exception exception;
547 ngpa = gfn_to_gpa(ngfn);
548 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
549 if (real_gfn == UNMAPPED_GVA)
552 real_gfn = gpa_to_gfn(real_gfn);
554 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
556 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
558 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
559 void *data, int offset, int len, u32 access)
561 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
562 data, offset, len, access);
566 * Load the pae pdptrs. Return true is they are all valid.
568 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
570 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
571 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
574 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
576 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
577 offset * sizeof(u64), sizeof(pdpte),
578 PFERR_USER_MASK|PFERR_WRITE_MASK);
583 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
584 if ((pdpte[i] & PT_PRESENT_MASK) &&
586 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
593 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
594 __set_bit(VCPU_EXREG_PDPTR,
595 (unsigned long *)&vcpu->arch.regs_avail);
596 __set_bit(VCPU_EXREG_PDPTR,
597 (unsigned long *)&vcpu->arch.regs_dirty);
602 EXPORT_SYMBOL_GPL(load_pdptrs);
604 bool pdptrs_changed(struct kvm_vcpu *vcpu)
606 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
612 if (is_long_mode(vcpu) || !is_pae(vcpu))
615 if (!test_bit(VCPU_EXREG_PDPTR,
616 (unsigned long *)&vcpu->arch.regs_avail))
619 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
620 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
621 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
622 PFERR_USER_MASK | PFERR_WRITE_MASK);
625 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
630 EXPORT_SYMBOL_GPL(pdptrs_changed);
632 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
634 unsigned long old_cr0 = kvm_read_cr0(vcpu);
635 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
640 if (cr0 & 0xffffffff00000000UL)
644 cr0 &= ~CR0_RESERVED_BITS;
646 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
649 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
652 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
654 if ((vcpu->arch.efer & EFER_LME)) {
659 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
664 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
669 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
672 kvm_x86_ops->set_cr0(vcpu, cr0);
674 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
675 kvm_clear_async_pf_completion_queue(vcpu);
676 kvm_async_pf_hash_reset(vcpu);
679 if ((cr0 ^ old_cr0) & update_bits)
680 kvm_mmu_reset_context(vcpu);
682 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
683 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
684 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
685 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
689 EXPORT_SYMBOL_GPL(kvm_set_cr0);
691 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
693 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
695 EXPORT_SYMBOL_GPL(kvm_lmsw);
697 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
699 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
700 !vcpu->guest_xcr0_loaded) {
701 /* kvm_set_xcr() also depends on this */
702 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
703 vcpu->guest_xcr0_loaded = 1;
707 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
709 if (vcpu->guest_xcr0_loaded) {
710 if (vcpu->arch.xcr0 != host_xcr0)
711 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
712 vcpu->guest_xcr0_loaded = 0;
716 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
719 u64 old_xcr0 = vcpu->arch.xcr0;
722 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
723 if (index != XCR_XFEATURE_ENABLED_MASK)
725 if (!(xcr0 & XFEATURE_MASK_FP))
727 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
731 * Do not allow the guest to set bits that we do not support
732 * saving. However, xcr0 bit 0 is always set, even if the
733 * emulated CPU does not support XSAVE (see fx_init).
735 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
736 if (xcr0 & ~valid_bits)
739 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
740 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
743 if (xcr0 & XFEATURE_MASK_AVX512) {
744 if (!(xcr0 & XFEATURE_MASK_YMM))
746 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
749 vcpu->arch.xcr0 = xcr0;
751 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
752 kvm_update_cpuid(vcpu);
756 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
758 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
759 __kvm_set_xcr(vcpu, index, xcr)) {
760 kvm_inject_gp(vcpu, 0);
765 EXPORT_SYMBOL_GPL(kvm_set_xcr);
767 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
769 unsigned long old_cr4 = kvm_read_cr4(vcpu);
770 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
771 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
773 if (cr4 & CR4_RESERVED_BITS)
776 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
779 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
782 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
785 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
788 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
791 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
794 if (is_long_mode(vcpu)) {
795 if (!(cr4 & X86_CR4_PAE))
797 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
798 && ((cr4 ^ old_cr4) & pdptr_bits)
799 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
803 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
804 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
807 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
808 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
812 if (kvm_x86_ops->set_cr4(vcpu, cr4))
815 if (((cr4 ^ old_cr4) & pdptr_bits) ||
816 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
817 kvm_mmu_reset_context(vcpu);
819 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
820 kvm_update_cpuid(vcpu);
824 EXPORT_SYMBOL_GPL(kvm_set_cr4);
826 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
829 cr3 &= ~CR3_PCID_INVD;
832 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
833 kvm_mmu_sync_roots(vcpu);
834 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
838 if (is_long_mode(vcpu) &&
839 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 62)))
841 else if (is_pae(vcpu) && is_paging(vcpu) &&
842 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
845 vcpu->arch.cr3 = cr3;
846 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
847 kvm_mmu_new_cr3(vcpu);
850 EXPORT_SYMBOL_GPL(kvm_set_cr3);
852 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
854 if (cr8 & CR8_RESERVED_BITS)
856 if (lapic_in_kernel(vcpu))
857 kvm_lapic_set_tpr(vcpu, cr8);
859 vcpu->arch.cr8 = cr8;
862 EXPORT_SYMBOL_GPL(kvm_set_cr8);
864 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
866 if (lapic_in_kernel(vcpu))
867 return kvm_lapic_get_cr8(vcpu);
869 return vcpu->arch.cr8;
871 EXPORT_SYMBOL_GPL(kvm_get_cr8);
873 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
877 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
878 for (i = 0; i < KVM_NR_DB_REGS; i++)
879 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
880 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
884 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
886 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
887 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
890 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
894 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
895 dr7 = vcpu->arch.guest_debug_dr7;
897 dr7 = vcpu->arch.dr7;
898 kvm_x86_ops->set_dr7(vcpu, dr7);
899 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
900 if (dr7 & DR7_BP_EN_MASK)
901 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
904 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
906 u64 fixed = DR6_FIXED_1;
908 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
913 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
917 vcpu->arch.db[dr] = val;
918 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
919 vcpu->arch.eff_db[dr] = val;
924 if (val & 0xffffffff00000000ULL)
926 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
927 kvm_update_dr6(vcpu);
932 if (val & 0xffffffff00000000ULL)
934 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
935 kvm_update_dr7(vcpu);
942 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
944 if (__kvm_set_dr(vcpu, dr, val)) {
945 kvm_inject_gp(vcpu, 0);
950 EXPORT_SYMBOL_GPL(kvm_set_dr);
952 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
956 *val = vcpu->arch.db[dr];
961 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
962 *val = vcpu->arch.dr6;
964 *val = kvm_x86_ops->get_dr6(vcpu);
969 *val = vcpu->arch.dr7;
974 EXPORT_SYMBOL_GPL(kvm_get_dr);
976 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
978 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
982 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
985 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
986 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
989 EXPORT_SYMBOL_GPL(kvm_rdpmc);
992 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
993 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
995 * This list is modified at module load time to reflect the
996 * capabilities of the host cpu. This capabilities test skips MSRs that are
997 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
998 * may depend on host virtualization features rather than host cpu features.
1001 static u32 msrs_to_save[] = {
1002 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1004 #ifdef CONFIG_X86_64
1005 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1007 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1008 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1011 static unsigned num_msrs_to_save;
1013 static u32 emulated_msrs[] = {
1014 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1015 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1016 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1017 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1018 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1019 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1020 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1022 HV_X64_MSR_VP_INDEX,
1023 HV_X64_MSR_VP_RUNTIME,
1024 HV_X64_MSR_SCONTROL,
1025 HV_X64_MSR_STIMER0_CONFIG,
1026 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1029 MSR_IA32_TSC_ADJUST,
1030 MSR_IA32_TSCDEADLINE,
1031 MSR_IA32_MISC_ENABLE,
1032 MSR_IA32_MCG_STATUS,
1034 MSR_IA32_MCG_EXT_CTL,
1037 MSR_MISC_FEATURES_ENABLES,
1040 static unsigned num_emulated_msrs;
1042 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1044 if (efer & efer_reserved_bits)
1047 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1050 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1055 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1057 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1059 u64 old_efer = vcpu->arch.efer;
1061 if (!kvm_valid_efer(vcpu, efer))
1065 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1069 efer |= vcpu->arch.efer & EFER_LMA;
1071 kvm_x86_ops->set_efer(vcpu, efer);
1073 /* Update reserved bits */
1074 if ((efer ^ old_efer) & EFER_NX)
1075 kvm_mmu_reset_context(vcpu);
1080 void kvm_enable_efer_bits(u64 mask)
1082 efer_reserved_bits &= ~mask;
1084 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1087 * Writes msr value into into the appropriate "register".
1088 * Returns 0 on success, non-0 otherwise.
1089 * Assumes vcpu_load() was already called.
1091 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1093 switch (msr->index) {
1096 case MSR_KERNEL_GS_BASE:
1099 if (is_noncanonical_address(msr->data, vcpu))
1102 case MSR_IA32_SYSENTER_EIP:
1103 case MSR_IA32_SYSENTER_ESP:
1105 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1106 * non-canonical address is written on Intel but not on
1107 * AMD (which ignores the top 32-bits, because it does
1108 * not implement 64-bit SYSENTER).
1110 * 64-bit code should hence be able to write a non-canonical
1111 * value on AMD. Making the address canonical ensures that
1112 * vmentry does not fail on Intel after writing a non-canonical
1113 * value, and that something deterministic happens if the guest
1114 * invokes 64-bit SYSENTER.
1116 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1118 return kvm_x86_ops->set_msr(vcpu, msr);
1120 EXPORT_SYMBOL_GPL(kvm_set_msr);
1123 * Adapt set_msr() to msr_io()'s calling convention
1125 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1127 struct msr_data msr;
1131 msr.host_initiated = true;
1132 r = kvm_get_msr(vcpu, &msr);
1140 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1142 struct msr_data msr;
1146 msr.host_initiated = true;
1147 return kvm_set_msr(vcpu, &msr);
1150 #ifdef CONFIG_X86_64
1151 struct pvclock_gtod_data {
1154 struct { /* extract of a clocksource struct */
1167 static struct pvclock_gtod_data pvclock_gtod_data;
1169 static void update_pvclock_gtod(struct timekeeper *tk)
1171 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1174 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1176 write_seqcount_begin(&vdata->seq);
1178 /* copy pvclock gtod data */
1179 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1180 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1181 vdata->clock.mask = tk->tkr_mono.mask;
1182 vdata->clock.mult = tk->tkr_mono.mult;
1183 vdata->clock.shift = tk->tkr_mono.shift;
1185 vdata->boot_ns = boot_ns;
1186 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1188 vdata->wall_time_sec = tk->xtime_sec;
1190 write_seqcount_end(&vdata->seq);
1194 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1197 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1198 * vcpu_enter_guest. This function is only called from
1199 * the physical CPU that is running vcpu.
1201 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1204 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1208 struct pvclock_wall_clock wc;
1209 struct timespec64 boot;
1214 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1219 ++version; /* first time write, random junk */
1223 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1227 * The guest calculates current wall clock time by adding
1228 * system time (updated by kvm_guest_time_update below) to the
1229 * wall clock specified here. guest system time equals host
1230 * system time for us, thus we must fill in host boot time here.
1232 getboottime64(&boot);
1234 if (kvm->arch.kvmclock_offset) {
1235 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1236 boot = timespec64_sub(boot, ts);
1238 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1239 wc.nsec = boot.tv_nsec;
1240 wc.version = version;
1242 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1245 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1248 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1250 do_shl32_div32(dividend, divisor);
1254 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1255 s8 *pshift, u32 *pmultiplier)
1263 scaled64 = scaled_hz;
1264 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1269 tps32 = (uint32_t)tps64;
1270 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1271 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1279 *pmultiplier = div_frac(scaled64, tps32);
1281 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1282 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1285 #ifdef CONFIG_X86_64
1286 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1289 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1290 static unsigned long max_tsc_khz;
1292 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1294 u64 v = (u64)khz * (1000000 + ppm);
1299 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1303 /* Guest TSC same frequency as host TSC? */
1305 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1309 /* TSC scaling supported? */
1310 if (!kvm_has_tsc_control) {
1311 if (user_tsc_khz > tsc_khz) {
1312 vcpu->arch.tsc_catchup = 1;
1313 vcpu->arch.tsc_always_catchup = 1;
1316 WARN(1, "user requested TSC rate below hardware speed\n");
1321 /* TSC scaling required - calculate ratio */
1322 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1323 user_tsc_khz, tsc_khz);
1325 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1326 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1331 vcpu->arch.tsc_scaling_ratio = ratio;
1335 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1337 u32 thresh_lo, thresh_hi;
1338 int use_scaling = 0;
1340 /* tsc_khz can be zero if TSC calibration fails */
1341 if (user_tsc_khz == 0) {
1342 /* set tsc_scaling_ratio to a safe value */
1343 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1347 /* Compute a scale to convert nanoseconds in TSC cycles */
1348 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1349 &vcpu->arch.virtual_tsc_shift,
1350 &vcpu->arch.virtual_tsc_mult);
1351 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1354 * Compute the variation in TSC rate which is acceptable
1355 * within the range of tolerance and decide if the
1356 * rate being applied is within that bounds of the hardware
1357 * rate. If so, no scaling or compensation need be done.
1359 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1360 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1361 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1362 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1365 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1368 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1370 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1371 vcpu->arch.virtual_tsc_mult,
1372 vcpu->arch.virtual_tsc_shift);
1373 tsc += vcpu->arch.this_tsc_write;
1377 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1379 #ifdef CONFIG_X86_64
1381 struct kvm_arch *ka = &vcpu->kvm->arch;
1382 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1384 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1385 atomic_read(&vcpu->kvm->online_vcpus));
1388 * Once the masterclock is enabled, always perform request in
1389 * order to update it.
1391 * In order to enable masterclock, the host clocksource must be TSC
1392 * and the vcpus need to have matched TSCs. When that happens,
1393 * perform request to enable masterclock.
1395 if (ka->use_master_clock ||
1396 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1397 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1399 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1400 atomic_read(&vcpu->kvm->online_vcpus),
1401 ka->use_master_clock, gtod->clock.vclock_mode);
1405 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1407 u64 curr_offset = vcpu->arch.tsc_offset;
1408 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1412 * Multiply tsc by a fixed point number represented by ratio.
1414 * The most significant 64-N bits (mult) of ratio represent the
1415 * integral part of the fixed point number; the remaining N bits
1416 * (frac) represent the fractional part, ie. ratio represents a fixed
1417 * point number (mult + frac * 2^(-N)).
1419 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1421 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1423 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1426 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1429 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1431 if (ratio != kvm_default_tsc_scaling_ratio)
1432 _tsc = __scale_tsc(ratio, tsc);
1436 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1438 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1442 tsc = kvm_scale_tsc(vcpu, rdtsc());
1444 return target_tsc - tsc;
1447 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1449 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1451 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1453 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1455 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1456 vcpu->arch.tsc_offset = offset;
1459 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1461 struct kvm *kvm = vcpu->kvm;
1462 u64 offset, ns, elapsed;
1463 unsigned long flags;
1465 bool already_matched;
1466 u64 data = msr->data;
1467 bool synchronizing = false;
1469 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1470 offset = kvm_compute_tsc_offset(vcpu, data);
1471 ns = ktime_get_boot_ns();
1472 elapsed = ns - kvm->arch.last_tsc_nsec;
1474 if (vcpu->arch.virtual_tsc_khz) {
1475 if (data == 0 && msr->host_initiated) {
1477 * detection of vcpu initialization -- need to sync
1478 * with other vCPUs. This particularly helps to keep
1479 * kvm_clock stable after CPU hotplug
1481 synchronizing = true;
1483 u64 tsc_exp = kvm->arch.last_tsc_write +
1484 nsec_to_cycles(vcpu, elapsed);
1485 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1487 * Special case: TSC write with a small delta (1 second)
1488 * of virtual cycle time against real time is
1489 * interpreted as an attempt to synchronize the CPU.
1491 synchronizing = data < tsc_exp + tsc_hz &&
1492 data + tsc_hz > tsc_exp;
1497 * For a reliable TSC, we can match TSC offsets, and for an unstable
1498 * TSC, we add elapsed time in this computation. We could let the
1499 * compensation code attempt to catch up if we fall behind, but
1500 * it's better to try to match offsets from the beginning.
1502 if (synchronizing &&
1503 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1504 if (!check_tsc_unstable()) {
1505 offset = kvm->arch.cur_tsc_offset;
1506 pr_debug("kvm: matched tsc offset for %llu\n", data);
1508 u64 delta = nsec_to_cycles(vcpu, elapsed);
1510 offset = kvm_compute_tsc_offset(vcpu, data);
1511 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1514 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1517 * We split periods of matched TSC writes into generations.
1518 * For each generation, we track the original measured
1519 * nanosecond time, offset, and write, so if TSCs are in
1520 * sync, we can match exact offset, and if not, we can match
1521 * exact software computation in compute_guest_tsc()
1523 * These values are tracked in kvm->arch.cur_xxx variables.
1525 kvm->arch.cur_tsc_generation++;
1526 kvm->arch.cur_tsc_nsec = ns;
1527 kvm->arch.cur_tsc_write = data;
1528 kvm->arch.cur_tsc_offset = offset;
1530 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1531 kvm->arch.cur_tsc_generation, data);
1535 * We also track th most recent recorded KHZ, write and time to
1536 * allow the matching interval to be extended at each write.
1538 kvm->arch.last_tsc_nsec = ns;
1539 kvm->arch.last_tsc_write = data;
1540 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1542 vcpu->arch.last_guest_tsc = data;
1544 /* Keep track of which generation this VCPU has synchronized to */
1545 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1546 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1547 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1549 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1550 update_ia32_tsc_adjust_msr(vcpu, offset);
1552 kvm_vcpu_write_tsc_offset(vcpu, offset);
1553 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1555 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1557 kvm->arch.nr_vcpus_matched_tsc = 0;
1558 } else if (!already_matched) {
1559 kvm->arch.nr_vcpus_matched_tsc++;
1562 kvm_track_tsc_matching(vcpu);
1563 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1566 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1568 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1571 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1574 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1576 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1577 WARN_ON(adjustment < 0);
1578 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1579 adjust_tsc_offset_guest(vcpu, adjustment);
1582 #ifdef CONFIG_X86_64
1584 static u64 read_tsc(void)
1586 u64 ret = (u64)rdtsc_ordered();
1587 u64 last = pvclock_gtod_data.clock.cycle_last;
1589 if (likely(ret >= last))
1593 * GCC likes to generate cmov here, but this branch is extremely
1594 * predictable (it's just a function of time and the likely is
1595 * very likely) and there's a data dependence, so force GCC
1596 * to generate a branch instead. I don't barrier() because
1597 * we don't actually need a barrier, and if this function
1598 * ever gets inlined it will generate worse code.
1604 static inline u64 vgettsc(u64 *cycle_now)
1607 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1609 *cycle_now = read_tsc();
1611 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1612 return v * gtod->clock.mult;
1615 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1617 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1623 seq = read_seqcount_begin(>od->seq);
1624 mode = gtod->clock.vclock_mode;
1625 ns = gtod->nsec_base;
1626 ns += vgettsc(cycle_now);
1627 ns >>= gtod->clock.shift;
1628 ns += gtod->boot_ns;
1629 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1635 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1637 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1643 seq = read_seqcount_begin(>od->seq);
1644 mode = gtod->clock.vclock_mode;
1645 ts->tv_sec = gtod->wall_time_sec;
1646 ns = gtod->nsec_base;
1647 ns += vgettsc(cycle_now);
1648 ns >>= gtod->clock.shift;
1649 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1651 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1657 /* returns true if host is using tsc clocksource */
1658 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1660 /* checked again under seqlock below */
1661 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1664 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1667 /* returns true if host is using tsc clocksource */
1668 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1671 /* checked again under seqlock below */
1672 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1675 return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1681 * Assuming a stable TSC across physical CPUS, and a stable TSC
1682 * across virtual CPUs, the following condition is possible.
1683 * Each numbered line represents an event visible to both
1684 * CPUs at the next numbered event.
1686 * "timespecX" represents host monotonic time. "tscX" represents
1689 * VCPU0 on CPU0 | VCPU1 on CPU1
1691 * 1. read timespec0,tsc0
1692 * 2. | timespec1 = timespec0 + N
1694 * 3. transition to guest | transition to guest
1695 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1696 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1697 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1699 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1702 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1704 * - 0 < N - M => M < N
1706 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1707 * always the case (the difference between two distinct xtime instances
1708 * might be smaller then the difference between corresponding TSC reads,
1709 * when updating guest vcpus pvclock areas).
1711 * To avoid that problem, do not allow visibility of distinct
1712 * system_timestamp/tsc_timestamp values simultaneously: use a master
1713 * copy of host monotonic time values. Update that master copy
1716 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1720 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1722 #ifdef CONFIG_X86_64
1723 struct kvm_arch *ka = &kvm->arch;
1725 bool host_tsc_clocksource, vcpus_matched;
1727 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1728 atomic_read(&kvm->online_vcpus));
1731 * If the host uses TSC clock, then passthrough TSC as stable
1734 host_tsc_clocksource = kvm_get_time_and_clockread(
1735 &ka->master_kernel_ns,
1736 &ka->master_cycle_now);
1738 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1739 && !ka->backwards_tsc_observed
1740 && !ka->boot_vcpu_runs_old_kvmclock;
1742 if (ka->use_master_clock)
1743 atomic_set(&kvm_guest_has_master_clock, 1);
1745 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1746 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1751 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1753 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1756 static void kvm_gen_update_masterclock(struct kvm *kvm)
1758 #ifdef CONFIG_X86_64
1760 struct kvm_vcpu *vcpu;
1761 struct kvm_arch *ka = &kvm->arch;
1763 spin_lock(&ka->pvclock_gtod_sync_lock);
1764 kvm_make_mclock_inprogress_request(kvm);
1765 /* no guest entries from this point */
1766 pvclock_update_vm_gtod_copy(kvm);
1768 kvm_for_each_vcpu(i, vcpu, kvm)
1769 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1771 /* guest entries allowed */
1772 kvm_for_each_vcpu(i, vcpu, kvm)
1773 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1775 spin_unlock(&ka->pvclock_gtod_sync_lock);
1779 u64 get_kvmclock_ns(struct kvm *kvm)
1781 struct kvm_arch *ka = &kvm->arch;
1782 struct pvclock_vcpu_time_info hv_clock;
1785 spin_lock(&ka->pvclock_gtod_sync_lock);
1786 if (!ka->use_master_clock) {
1787 spin_unlock(&ka->pvclock_gtod_sync_lock);
1788 return ktime_get_boot_ns() + ka->kvmclock_offset;
1791 hv_clock.tsc_timestamp = ka->master_cycle_now;
1792 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1793 spin_unlock(&ka->pvclock_gtod_sync_lock);
1795 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1798 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1799 &hv_clock.tsc_shift,
1800 &hv_clock.tsc_to_system_mul);
1801 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1808 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1810 struct kvm_vcpu_arch *vcpu = &v->arch;
1811 struct pvclock_vcpu_time_info guest_hv_clock;
1813 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1814 &guest_hv_clock, sizeof(guest_hv_clock))))
1817 /* This VCPU is paused, but it's legal for a guest to read another
1818 * VCPU's kvmclock, so we really have to follow the specification where
1819 * it says that version is odd if data is being modified, and even after
1822 * Version field updates must be kept separate. This is because
1823 * kvm_write_guest_cached might use a "rep movs" instruction, and
1824 * writes within a string instruction are weakly ordered. So there
1825 * are three writes overall.
1827 * As a small optimization, only write the version field in the first
1828 * and third write. The vcpu->pv_time cache is still valid, because the
1829 * version field is the first in the struct.
1831 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1833 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1834 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1836 sizeof(vcpu->hv_clock.version));
1840 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1841 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1843 if (vcpu->pvclock_set_guest_stopped_request) {
1844 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1845 vcpu->pvclock_set_guest_stopped_request = false;
1848 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1850 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1852 sizeof(vcpu->hv_clock));
1856 vcpu->hv_clock.version++;
1857 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1859 sizeof(vcpu->hv_clock.version));
1862 static int kvm_guest_time_update(struct kvm_vcpu *v)
1864 unsigned long flags, tgt_tsc_khz;
1865 struct kvm_vcpu_arch *vcpu = &v->arch;
1866 struct kvm_arch *ka = &v->kvm->arch;
1868 u64 tsc_timestamp, host_tsc;
1870 bool use_master_clock;
1876 * If the host uses TSC clock, then passthrough TSC as stable
1879 spin_lock(&ka->pvclock_gtod_sync_lock);
1880 use_master_clock = ka->use_master_clock;
1881 if (use_master_clock) {
1882 host_tsc = ka->master_cycle_now;
1883 kernel_ns = ka->master_kernel_ns;
1885 spin_unlock(&ka->pvclock_gtod_sync_lock);
1887 /* Keep irq disabled to prevent changes to the clock */
1888 local_irq_save(flags);
1889 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1890 if (unlikely(tgt_tsc_khz == 0)) {
1891 local_irq_restore(flags);
1892 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1895 if (!use_master_clock) {
1897 kernel_ns = ktime_get_boot_ns();
1900 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1903 * We may have to catch up the TSC to match elapsed wall clock
1904 * time for two reasons, even if kvmclock is used.
1905 * 1) CPU could have been running below the maximum TSC rate
1906 * 2) Broken TSC compensation resets the base at each VCPU
1907 * entry to avoid unknown leaps of TSC even when running
1908 * again on the same CPU. This may cause apparent elapsed
1909 * time to disappear, and the guest to stand still or run
1912 if (vcpu->tsc_catchup) {
1913 u64 tsc = compute_guest_tsc(v, kernel_ns);
1914 if (tsc > tsc_timestamp) {
1915 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1916 tsc_timestamp = tsc;
1920 local_irq_restore(flags);
1922 /* With all the info we got, fill in the values */
1924 if (kvm_has_tsc_control)
1925 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1927 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1928 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1929 &vcpu->hv_clock.tsc_shift,
1930 &vcpu->hv_clock.tsc_to_system_mul);
1931 vcpu->hw_tsc_khz = tgt_tsc_khz;
1934 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1935 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1936 vcpu->last_guest_tsc = tsc_timestamp;
1938 /* If the host uses TSC clocksource, then it is stable */
1940 if (use_master_clock)
1941 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1943 vcpu->hv_clock.flags = pvclock_flags;
1945 if (vcpu->pv_time_enabled)
1946 kvm_setup_pvclock_page(v);
1947 if (v == kvm_get_vcpu(v->kvm, 0))
1948 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
1953 * kvmclock updates which are isolated to a given vcpu, such as
1954 * vcpu->cpu migration, should not allow system_timestamp from
1955 * the rest of the vcpus to remain static. Otherwise ntp frequency
1956 * correction applies to one vcpu's system_timestamp but not
1959 * So in those cases, request a kvmclock update for all vcpus.
1960 * We need to rate-limit these requests though, as they can
1961 * considerably slow guests that have a large number of vcpus.
1962 * The time for a remote vcpu to update its kvmclock is bound
1963 * by the delay we use to rate-limit the updates.
1966 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1968 static void kvmclock_update_fn(struct work_struct *work)
1971 struct delayed_work *dwork = to_delayed_work(work);
1972 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1973 kvmclock_update_work);
1974 struct kvm *kvm = container_of(ka, struct kvm, arch);
1975 struct kvm_vcpu *vcpu;
1977 kvm_for_each_vcpu(i, vcpu, kvm) {
1978 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1979 kvm_vcpu_kick(vcpu);
1983 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1985 struct kvm *kvm = v->kvm;
1987 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1988 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1989 KVMCLOCK_UPDATE_DELAY);
1992 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1994 static void kvmclock_sync_fn(struct work_struct *work)
1996 struct delayed_work *dwork = to_delayed_work(work);
1997 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1998 kvmclock_sync_work);
1999 struct kvm *kvm = container_of(ka, struct kvm, arch);
2001 if (!kvmclock_periodic_sync)
2004 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2005 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2006 KVMCLOCK_SYNC_PERIOD);
2009 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2011 u64 mcg_cap = vcpu->arch.mcg_cap;
2012 unsigned bank_num = mcg_cap & 0xff;
2015 case MSR_IA32_MCG_STATUS:
2016 vcpu->arch.mcg_status = data;
2018 case MSR_IA32_MCG_CTL:
2019 if (!(mcg_cap & MCG_CTL_P))
2021 if (data != 0 && data != ~(u64)0)
2023 vcpu->arch.mcg_ctl = data;
2026 if (msr >= MSR_IA32_MC0_CTL &&
2027 msr < MSR_IA32_MCx_CTL(bank_num)) {
2028 u32 offset = msr - MSR_IA32_MC0_CTL;
2029 /* only 0 or all 1s can be written to IA32_MCi_CTL
2030 * some Linux kernels though clear bit 10 in bank 4 to
2031 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2032 * this to avoid an uncatched #GP in the guest
2034 if ((offset & 0x3) == 0 &&
2035 data != 0 && (data | (1 << 10)) != ~(u64)0)
2037 vcpu->arch.mce_banks[offset] = data;
2045 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2047 struct kvm *kvm = vcpu->kvm;
2048 int lm = is_long_mode(vcpu);
2049 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2050 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2051 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2052 : kvm->arch.xen_hvm_config.blob_size_32;
2053 u32 page_num = data & ~PAGE_MASK;
2054 u64 page_addr = data & PAGE_MASK;
2059 if (page_num >= blob_size)
2062 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2067 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2076 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2078 gpa_t gpa = data & ~0x3f;
2080 /* Bits 3:5 are reserved, Should be zero */
2084 vcpu->arch.apf.msr_val = data;
2086 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2087 kvm_clear_async_pf_completion_queue(vcpu);
2088 kvm_async_pf_hash_reset(vcpu);
2092 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2096 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2097 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2098 kvm_async_pf_wakeup_all(vcpu);
2102 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2104 vcpu->arch.pv_time_enabled = false;
2107 static void record_steal_time(struct kvm_vcpu *vcpu)
2109 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2112 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2113 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2116 vcpu->arch.st.steal.preempted = 0;
2118 if (vcpu->arch.st.steal.version & 1)
2119 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2121 vcpu->arch.st.steal.version += 1;
2123 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2124 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2128 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2129 vcpu->arch.st.last_steal;
2130 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2132 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2133 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2137 vcpu->arch.st.steal.version += 1;
2139 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2140 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2143 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2146 u32 msr = msr_info->index;
2147 u64 data = msr_info->data;
2150 case MSR_AMD64_NB_CFG:
2151 case MSR_IA32_UCODE_REV:
2152 case MSR_IA32_UCODE_WRITE:
2153 case MSR_VM_HSAVE_PA:
2154 case MSR_AMD64_PATCH_LOADER:
2155 case MSR_AMD64_BU_CFG2:
2156 case MSR_AMD64_DC_CFG:
2160 return set_efer(vcpu, data);
2162 data &= ~(u64)0x40; /* ignore flush filter disable */
2163 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2164 data &= ~(u64)0x8; /* ignore TLB cache disable */
2165 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2167 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2172 case MSR_FAM10H_MMIO_CONF_BASE:
2174 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2179 case MSR_IA32_DEBUGCTLMSR:
2181 /* We support the non-activated case already */
2183 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2184 /* Values other than LBR and BTF are vendor-specific,
2185 thus reserved and should throw a #GP */
2188 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2191 case 0x200 ... 0x2ff:
2192 return kvm_mtrr_set_msr(vcpu, msr, data);
2193 case MSR_IA32_APICBASE:
2194 return kvm_set_apic_base(vcpu, msr_info);
2195 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2196 return kvm_x2apic_msr_write(vcpu, msr, data);
2197 case MSR_IA32_TSCDEADLINE:
2198 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2200 case MSR_IA32_TSC_ADJUST:
2201 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2202 if (!msr_info->host_initiated) {
2203 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2204 adjust_tsc_offset_guest(vcpu, adj);
2206 vcpu->arch.ia32_tsc_adjust_msr = data;
2209 case MSR_IA32_MISC_ENABLE:
2210 vcpu->arch.ia32_misc_enable_msr = data;
2212 case MSR_IA32_SMBASE:
2213 if (!msr_info->host_initiated)
2215 vcpu->arch.smbase = data;
2217 case MSR_KVM_WALL_CLOCK_NEW:
2218 case MSR_KVM_WALL_CLOCK:
2219 vcpu->kvm->arch.wall_clock = data;
2220 kvm_write_wall_clock(vcpu->kvm, data);
2222 case MSR_KVM_SYSTEM_TIME_NEW:
2223 case MSR_KVM_SYSTEM_TIME: {
2224 struct kvm_arch *ka = &vcpu->kvm->arch;
2226 kvmclock_reset(vcpu);
2228 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2229 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2231 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2232 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2234 ka->boot_vcpu_runs_old_kvmclock = tmp;
2237 vcpu->arch.time = data;
2238 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2240 /* we verify if the enable bit is set... */
2244 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2245 &vcpu->arch.pv_time, data & ~1ULL,
2246 sizeof(struct pvclock_vcpu_time_info)))
2247 vcpu->arch.pv_time_enabled = false;
2249 vcpu->arch.pv_time_enabled = true;
2253 case MSR_KVM_ASYNC_PF_EN:
2254 if (kvm_pv_enable_async_pf(vcpu, data))
2257 case MSR_KVM_STEAL_TIME:
2259 if (unlikely(!sched_info_on()))
2262 if (data & KVM_STEAL_RESERVED_MASK)
2265 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2266 data & KVM_STEAL_VALID_BITS,
2267 sizeof(struct kvm_steal_time)))
2270 vcpu->arch.st.msr_val = data;
2272 if (!(data & KVM_MSR_ENABLED))
2275 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2278 case MSR_KVM_PV_EOI_EN:
2279 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2283 case MSR_IA32_MCG_CTL:
2284 case MSR_IA32_MCG_STATUS:
2285 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2286 return set_msr_mce(vcpu, msr, data);
2288 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2289 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2290 pr = true; /* fall through */
2291 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2292 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2293 if (kvm_pmu_is_valid_msr(vcpu, msr))
2294 return kvm_pmu_set_msr(vcpu, msr_info);
2296 if (pr || data != 0)
2297 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2298 "0x%x data 0x%llx\n", msr, data);
2300 case MSR_K7_CLK_CTL:
2302 * Ignore all writes to this no longer documented MSR.
2303 * Writes are only relevant for old K7 processors,
2304 * all pre-dating SVM, but a recommended workaround from
2305 * AMD for these chips. It is possible to specify the
2306 * affected processor models on the command line, hence
2307 * the need to ignore the workaround.
2310 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2311 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2312 case HV_X64_MSR_CRASH_CTL:
2313 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2314 return kvm_hv_set_msr_common(vcpu, msr, data,
2315 msr_info->host_initiated);
2316 case MSR_IA32_BBL_CR_CTL3:
2317 /* Drop writes to this legacy MSR -- see rdmsr
2318 * counterpart for further detail.
2320 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", msr, data);
2322 case MSR_AMD64_OSVW_ID_LENGTH:
2323 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2325 vcpu->arch.osvw.length = data;
2327 case MSR_AMD64_OSVW_STATUS:
2328 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2330 vcpu->arch.osvw.status = data;
2332 case MSR_PLATFORM_INFO:
2333 if (!msr_info->host_initiated ||
2334 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2335 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2336 cpuid_fault_enabled(vcpu)))
2338 vcpu->arch.msr_platform_info = data;
2340 case MSR_MISC_FEATURES_ENABLES:
2341 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2342 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2343 !supports_cpuid_fault(vcpu)))
2345 vcpu->arch.msr_misc_features_enables = data;
2348 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2349 return xen_hvm_config(vcpu, data);
2350 if (kvm_pmu_is_valid_msr(vcpu, msr))
2351 return kvm_pmu_set_msr(vcpu, msr_info);
2353 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2357 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2364 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2368 * Reads an msr value (of 'msr_index') into 'pdata'.
2369 * Returns 0 on success, non-0 otherwise.
2370 * Assumes vcpu_load() was already called.
2372 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2374 return kvm_x86_ops->get_msr(vcpu, msr);
2376 EXPORT_SYMBOL_GPL(kvm_get_msr);
2378 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2381 u64 mcg_cap = vcpu->arch.mcg_cap;
2382 unsigned bank_num = mcg_cap & 0xff;
2385 case MSR_IA32_P5_MC_ADDR:
2386 case MSR_IA32_P5_MC_TYPE:
2389 case MSR_IA32_MCG_CAP:
2390 data = vcpu->arch.mcg_cap;
2392 case MSR_IA32_MCG_CTL:
2393 if (!(mcg_cap & MCG_CTL_P))
2395 data = vcpu->arch.mcg_ctl;
2397 case MSR_IA32_MCG_STATUS:
2398 data = vcpu->arch.mcg_status;
2401 if (msr >= MSR_IA32_MC0_CTL &&
2402 msr < MSR_IA32_MCx_CTL(bank_num)) {
2403 u32 offset = msr - MSR_IA32_MC0_CTL;
2404 data = vcpu->arch.mce_banks[offset];
2413 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2415 switch (msr_info->index) {
2416 case MSR_IA32_PLATFORM_ID:
2417 case MSR_IA32_EBL_CR_POWERON:
2418 case MSR_IA32_DEBUGCTLMSR:
2419 case MSR_IA32_LASTBRANCHFROMIP:
2420 case MSR_IA32_LASTBRANCHTOIP:
2421 case MSR_IA32_LASTINTFROMIP:
2422 case MSR_IA32_LASTINTTOIP:
2424 case MSR_K8_TSEG_ADDR:
2425 case MSR_K8_TSEG_MASK:
2427 case MSR_VM_HSAVE_PA:
2428 case MSR_K8_INT_PENDING_MSG:
2429 case MSR_AMD64_NB_CFG:
2430 case MSR_FAM10H_MMIO_CONF_BASE:
2431 case MSR_AMD64_BU_CFG2:
2432 case MSR_IA32_PERF_CTL:
2433 case MSR_AMD64_DC_CFG:
2436 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2437 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2438 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2439 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2440 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2441 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2444 case MSR_IA32_UCODE_REV:
2445 msr_info->data = 0x100000000ULL;
2448 case 0x200 ... 0x2ff:
2449 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2450 case 0xcd: /* fsb frequency */
2454 * MSR_EBC_FREQUENCY_ID
2455 * Conservative value valid for even the basic CPU models.
2456 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2457 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2458 * and 266MHz for model 3, or 4. Set Core Clock
2459 * Frequency to System Bus Frequency Ratio to 1 (bits
2460 * 31:24) even though these are only valid for CPU
2461 * models > 2, however guests may end up dividing or
2462 * multiplying by zero otherwise.
2464 case MSR_EBC_FREQUENCY_ID:
2465 msr_info->data = 1 << 24;
2467 case MSR_IA32_APICBASE:
2468 msr_info->data = kvm_get_apic_base(vcpu);
2470 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2471 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2473 case MSR_IA32_TSCDEADLINE:
2474 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2476 case MSR_IA32_TSC_ADJUST:
2477 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2479 case MSR_IA32_MISC_ENABLE:
2480 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2482 case MSR_IA32_SMBASE:
2483 if (!msr_info->host_initiated)
2485 msr_info->data = vcpu->arch.smbase;
2487 case MSR_IA32_PERF_STATUS:
2488 /* TSC increment by tick */
2489 msr_info->data = 1000ULL;
2490 /* CPU multiplier */
2491 msr_info->data |= (((uint64_t)4ULL) << 40);
2494 msr_info->data = vcpu->arch.efer;
2496 case MSR_KVM_WALL_CLOCK:
2497 case MSR_KVM_WALL_CLOCK_NEW:
2498 msr_info->data = vcpu->kvm->arch.wall_clock;
2500 case MSR_KVM_SYSTEM_TIME:
2501 case MSR_KVM_SYSTEM_TIME_NEW:
2502 msr_info->data = vcpu->arch.time;
2504 case MSR_KVM_ASYNC_PF_EN:
2505 msr_info->data = vcpu->arch.apf.msr_val;
2507 case MSR_KVM_STEAL_TIME:
2508 msr_info->data = vcpu->arch.st.msr_val;
2510 case MSR_KVM_PV_EOI_EN:
2511 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2513 case MSR_IA32_P5_MC_ADDR:
2514 case MSR_IA32_P5_MC_TYPE:
2515 case MSR_IA32_MCG_CAP:
2516 case MSR_IA32_MCG_CTL:
2517 case MSR_IA32_MCG_STATUS:
2518 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2519 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2520 case MSR_K7_CLK_CTL:
2522 * Provide expected ramp-up count for K7. All other
2523 * are set to zero, indicating minimum divisors for
2526 * This prevents guest kernels on AMD host with CPU
2527 * type 6, model 8 and higher from exploding due to
2528 * the rdmsr failing.
2530 msr_info->data = 0x20000000;
2532 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2533 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2534 case HV_X64_MSR_CRASH_CTL:
2535 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2536 return kvm_hv_get_msr_common(vcpu,
2537 msr_info->index, &msr_info->data);
2539 case MSR_IA32_BBL_CR_CTL3:
2540 /* This legacy MSR exists but isn't fully documented in current
2541 * silicon. It is however accessed by winxp in very narrow
2542 * scenarios where it sets bit #19, itself documented as
2543 * a "reserved" bit. Best effort attempt to source coherent
2544 * read data here should the balance of the register be
2545 * interpreted by the guest:
2547 * L2 cache control register 3: 64GB range, 256KB size,
2548 * enabled, latency 0x1, configured
2550 msr_info->data = 0xbe702111;
2552 case MSR_AMD64_OSVW_ID_LENGTH:
2553 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2555 msr_info->data = vcpu->arch.osvw.length;
2557 case MSR_AMD64_OSVW_STATUS:
2558 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2560 msr_info->data = vcpu->arch.osvw.status;
2562 case MSR_PLATFORM_INFO:
2563 msr_info->data = vcpu->arch.msr_platform_info;
2565 case MSR_MISC_FEATURES_ENABLES:
2566 msr_info->data = vcpu->arch.msr_misc_features_enables;
2569 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2570 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2572 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2576 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2583 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2586 * Read or write a bunch of msrs. All parameters are kernel addresses.
2588 * @return number of msrs set successfully.
2590 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2591 struct kvm_msr_entry *entries,
2592 int (*do_msr)(struct kvm_vcpu *vcpu,
2593 unsigned index, u64 *data))
2597 idx = srcu_read_lock(&vcpu->kvm->srcu);
2598 for (i = 0; i < msrs->nmsrs; ++i)
2599 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2601 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2607 * Read or write a bunch of msrs. Parameters are user addresses.
2609 * @return number of msrs set successfully.
2611 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2612 int (*do_msr)(struct kvm_vcpu *vcpu,
2613 unsigned index, u64 *data),
2616 struct kvm_msrs msrs;
2617 struct kvm_msr_entry *entries;
2622 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2626 if (msrs.nmsrs >= MAX_IO_MSRS)
2629 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2630 entries = memdup_user(user_msrs->entries, size);
2631 if (IS_ERR(entries)) {
2632 r = PTR_ERR(entries);
2636 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2641 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2652 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2657 case KVM_CAP_IRQCHIP:
2659 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2660 case KVM_CAP_SET_TSS_ADDR:
2661 case KVM_CAP_EXT_CPUID:
2662 case KVM_CAP_EXT_EMUL_CPUID:
2663 case KVM_CAP_CLOCKSOURCE:
2665 case KVM_CAP_NOP_IO_DELAY:
2666 case KVM_CAP_MP_STATE:
2667 case KVM_CAP_SYNC_MMU:
2668 case KVM_CAP_USER_NMI:
2669 case KVM_CAP_REINJECT_CONTROL:
2670 case KVM_CAP_IRQ_INJECT_STATUS:
2671 case KVM_CAP_IOEVENTFD:
2672 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2674 case KVM_CAP_PIT_STATE2:
2675 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2676 case KVM_CAP_XEN_HVM:
2677 case KVM_CAP_VCPU_EVENTS:
2678 case KVM_CAP_HYPERV:
2679 case KVM_CAP_HYPERV_VAPIC:
2680 case KVM_CAP_HYPERV_SPIN:
2681 case KVM_CAP_HYPERV_SYNIC:
2682 case KVM_CAP_HYPERV_SYNIC2:
2683 case KVM_CAP_HYPERV_VP_INDEX:
2684 case KVM_CAP_PCI_SEGMENT:
2685 case KVM_CAP_DEBUGREGS:
2686 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2688 case KVM_CAP_ASYNC_PF:
2689 case KVM_CAP_GET_TSC_KHZ:
2690 case KVM_CAP_KVMCLOCK_CTRL:
2691 case KVM_CAP_READONLY_MEM:
2692 case KVM_CAP_HYPERV_TIME:
2693 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2694 case KVM_CAP_TSC_DEADLINE_TIMER:
2695 case KVM_CAP_ENABLE_CAP_VM:
2696 case KVM_CAP_DISABLE_QUIRKS:
2697 case KVM_CAP_SET_BOOT_CPU_ID:
2698 case KVM_CAP_SPLIT_IRQCHIP:
2699 case KVM_CAP_IMMEDIATE_EXIT:
2702 case KVM_CAP_ADJUST_CLOCK:
2703 r = KVM_CLOCK_TSC_STABLE;
2705 case KVM_CAP_X86_GUEST_MWAIT:
2706 r = kvm_mwait_in_guest();
2708 case KVM_CAP_X86_SMM:
2709 /* SMBASE is usually relocated above 1M on modern chipsets,
2710 * and SMM handlers might indeed rely on 4G segment limits,
2711 * so do not report SMM to be available if real mode is
2712 * emulated via vm86 mode. Still, do not go to great lengths
2713 * to avoid userspace's usage of the feature, because it is a
2714 * fringe case that is not enabled except via specific settings
2715 * of the module parameters.
2717 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2720 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2722 case KVM_CAP_NR_VCPUS:
2723 r = KVM_SOFT_MAX_VCPUS;
2725 case KVM_CAP_MAX_VCPUS:
2728 case KVM_CAP_NR_MEMSLOTS:
2729 r = KVM_USER_MEM_SLOTS;
2731 case KVM_CAP_PV_MMU: /* obsolete */
2735 r = KVM_MAX_MCE_BANKS;
2738 r = boot_cpu_has(X86_FEATURE_XSAVE);
2740 case KVM_CAP_TSC_CONTROL:
2741 r = kvm_has_tsc_control;
2743 case KVM_CAP_X2APIC_API:
2744 r = KVM_X2APIC_API_VALID_FLAGS;
2754 long kvm_arch_dev_ioctl(struct file *filp,
2755 unsigned int ioctl, unsigned long arg)
2757 void __user *argp = (void __user *)arg;
2761 case KVM_GET_MSR_INDEX_LIST: {
2762 struct kvm_msr_list __user *user_msr_list = argp;
2763 struct kvm_msr_list msr_list;
2767 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2770 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2771 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2774 if (n < msr_list.nmsrs)
2777 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2778 num_msrs_to_save * sizeof(u32)))
2780 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2782 num_emulated_msrs * sizeof(u32)))
2787 case KVM_GET_SUPPORTED_CPUID:
2788 case KVM_GET_EMULATED_CPUID: {
2789 struct kvm_cpuid2 __user *cpuid_arg = argp;
2790 struct kvm_cpuid2 cpuid;
2793 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2796 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2802 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2807 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2809 if (copy_to_user(argp, &kvm_mce_cap_supported,
2810 sizeof(kvm_mce_cap_supported)))
2822 static void wbinvd_ipi(void *garbage)
2827 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2829 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2832 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2834 /* Address WBINVD may be executed by guest */
2835 if (need_emulate_wbinvd(vcpu)) {
2836 if (kvm_x86_ops->has_wbinvd_exit())
2837 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2838 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2839 smp_call_function_single(vcpu->cpu,
2840 wbinvd_ipi, NULL, 1);
2843 kvm_x86_ops->vcpu_load(vcpu, cpu);
2845 /* Apply any externally detected TSC adjustments (due to suspend) */
2846 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2847 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2848 vcpu->arch.tsc_offset_adjustment = 0;
2849 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2852 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2853 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2854 rdtsc() - vcpu->arch.last_host_tsc;
2856 mark_tsc_unstable("KVM discovered backwards TSC");
2858 if (check_tsc_unstable()) {
2859 u64 offset = kvm_compute_tsc_offset(vcpu,
2860 vcpu->arch.last_guest_tsc);
2861 kvm_vcpu_write_tsc_offset(vcpu, offset);
2862 vcpu->arch.tsc_catchup = 1;
2865 if (kvm_lapic_hv_timer_in_use(vcpu))
2866 kvm_lapic_restart_hv_timer(vcpu);
2869 * On a host with synchronized TSC, there is no need to update
2870 * kvmclock on vcpu->cpu migration
2872 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2873 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2874 if (vcpu->cpu != cpu)
2875 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
2879 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2882 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2884 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2887 vcpu->arch.st.steal.preempted = 1;
2889 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2890 &vcpu->arch.st.steal.preempted,
2891 offsetof(struct kvm_steal_time, preempted),
2892 sizeof(vcpu->arch.st.steal.preempted));
2895 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2899 if (vcpu->preempted)
2900 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
2903 * Disable page faults because we're in atomic context here.
2904 * kvm_write_guest_offset_cached() would call might_fault()
2905 * that relies on pagefault_disable() to tell if there's a
2906 * bug. NOTE: the write to guest memory may not go through if
2907 * during postcopy live migration or if there's heavy guest
2910 pagefault_disable();
2912 * kvm_memslots() will be called by
2913 * kvm_write_guest_offset_cached() so take the srcu lock.
2915 idx = srcu_read_lock(&vcpu->kvm->srcu);
2916 kvm_steal_time_set_preempted(vcpu);
2917 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2919 kvm_x86_ops->vcpu_put(vcpu);
2920 kvm_put_guest_fpu(vcpu);
2921 vcpu->arch.last_host_tsc = rdtsc();
2924 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2925 struct kvm_lapic_state *s)
2927 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
2928 kvm_x86_ops->sync_pir_to_irr(vcpu);
2930 return kvm_apic_get_state(vcpu, s);
2933 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2934 struct kvm_lapic_state *s)
2938 r = kvm_apic_set_state(vcpu, s);
2941 update_cr8_intercept(vcpu);
2946 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2948 return (!lapic_in_kernel(vcpu) ||
2949 kvm_apic_accept_pic_intr(vcpu));
2953 * if userspace requested an interrupt window, check that the
2954 * interrupt window is open.
2956 * No need to exit to userspace if we already have an interrupt queued.
2958 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2960 return kvm_arch_interrupt_allowed(vcpu) &&
2961 !kvm_cpu_has_interrupt(vcpu) &&
2962 !kvm_event_needs_reinjection(vcpu) &&
2963 kvm_cpu_accept_dm_intr(vcpu);
2966 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2967 struct kvm_interrupt *irq)
2969 if (irq->irq >= KVM_NR_INTERRUPTS)
2972 if (!irqchip_in_kernel(vcpu->kvm)) {
2973 kvm_queue_interrupt(vcpu, irq->irq, false);
2974 kvm_make_request(KVM_REQ_EVENT, vcpu);
2979 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2980 * fail for in-kernel 8259.
2982 if (pic_in_kernel(vcpu->kvm))
2985 if (vcpu->arch.pending_external_vector != -1)
2988 vcpu->arch.pending_external_vector = irq->irq;
2989 kvm_make_request(KVM_REQ_EVENT, vcpu);
2993 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2995 kvm_inject_nmi(vcpu);
3000 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3002 kvm_make_request(KVM_REQ_SMI, vcpu);
3007 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3008 struct kvm_tpr_access_ctl *tac)
3012 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3016 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3020 unsigned bank_num = mcg_cap & 0xff, bank;
3023 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3025 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3028 vcpu->arch.mcg_cap = mcg_cap;
3029 /* Init IA32_MCG_CTL to all 1s */
3030 if (mcg_cap & MCG_CTL_P)
3031 vcpu->arch.mcg_ctl = ~(u64)0;
3032 /* Init IA32_MCi_CTL to all 1s */
3033 for (bank = 0; bank < bank_num; bank++)
3034 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3036 if (kvm_x86_ops->setup_mce)
3037 kvm_x86_ops->setup_mce(vcpu);
3042 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3043 struct kvm_x86_mce *mce)
3045 u64 mcg_cap = vcpu->arch.mcg_cap;
3046 unsigned bank_num = mcg_cap & 0xff;
3047 u64 *banks = vcpu->arch.mce_banks;
3049 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3052 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3053 * reporting is disabled
3055 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3056 vcpu->arch.mcg_ctl != ~(u64)0)
3058 banks += 4 * mce->bank;
3060 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3061 * reporting is disabled for the bank
3063 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3065 if (mce->status & MCI_STATUS_UC) {
3066 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3067 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3068 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3071 if (banks[1] & MCI_STATUS_VAL)
3072 mce->status |= MCI_STATUS_OVER;
3073 banks[2] = mce->addr;
3074 banks[3] = mce->misc;
3075 vcpu->arch.mcg_status = mce->mcg_status;
3076 banks[1] = mce->status;
3077 kvm_queue_exception(vcpu, MC_VECTOR);
3078 } else if (!(banks[1] & MCI_STATUS_VAL)
3079 || !(banks[1] & MCI_STATUS_UC)) {
3080 if (banks[1] & MCI_STATUS_VAL)
3081 mce->status |= MCI_STATUS_OVER;
3082 banks[2] = mce->addr;
3083 banks[3] = mce->misc;
3084 banks[1] = mce->status;
3086 banks[1] |= MCI_STATUS_OVER;
3090 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3091 struct kvm_vcpu_events *events)
3095 * FIXME: pass injected and pending separately. This is only
3096 * needed for nested virtualization, whose state cannot be
3097 * migrated yet. For now we can combine them.
3099 events->exception.injected =
3100 (vcpu->arch.exception.pending ||
3101 vcpu->arch.exception.injected) &&
3102 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3103 events->exception.nr = vcpu->arch.exception.nr;
3104 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3105 events->exception.pad = 0;
3106 events->exception.error_code = vcpu->arch.exception.error_code;
3108 events->interrupt.injected =
3109 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3110 events->interrupt.nr = vcpu->arch.interrupt.nr;
3111 events->interrupt.soft = 0;
3112 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3114 events->nmi.injected = vcpu->arch.nmi_injected;
3115 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3116 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3117 events->nmi.pad = 0;
3119 events->sipi_vector = 0; /* never valid when reporting to user space */
3121 events->smi.smm = is_smm(vcpu);
3122 events->smi.pending = vcpu->arch.smi_pending;
3123 events->smi.smm_inside_nmi =
3124 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3125 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3127 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3128 | KVM_VCPUEVENT_VALID_SHADOW
3129 | KVM_VCPUEVENT_VALID_SMM);
3130 memset(&events->reserved, 0, sizeof(events->reserved));
3133 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3135 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3136 struct kvm_vcpu_events *events)
3138 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3139 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3140 | KVM_VCPUEVENT_VALID_SHADOW
3141 | KVM_VCPUEVENT_VALID_SMM))
3144 if (events->exception.injected &&
3145 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3146 is_guest_mode(vcpu)))
3149 /* INITs are latched while in SMM */
3150 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3151 (events->smi.smm || events->smi.pending) &&
3152 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3156 vcpu->arch.exception.injected = false;
3157 vcpu->arch.exception.pending = events->exception.injected;
3158 vcpu->arch.exception.nr = events->exception.nr;
3159 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3160 vcpu->arch.exception.error_code = events->exception.error_code;
3162 vcpu->arch.interrupt.pending = events->interrupt.injected;
3163 vcpu->arch.interrupt.nr = events->interrupt.nr;
3164 vcpu->arch.interrupt.soft = events->interrupt.soft;
3165 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3166 kvm_x86_ops->set_interrupt_shadow(vcpu,
3167 events->interrupt.shadow);
3169 vcpu->arch.nmi_injected = events->nmi.injected;
3170 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3171 vcpu->arch.nmi_pending = events->nmi.pending;
3172 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3174 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3175 lapic_in_kernel(vcpu))
3176 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3178 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3179 u32 hflags = vcpu->arch.hflags;
3180 if (events->smi.smm)
3181 hflags |= HF_SMM_MASK;
3183 hflags &= ~HF_SMM_MASK;
3184 kvm_set_hflags(vcpu, hflags);
3186 vcpu->arch.smi_pending = events->smi.pending;
3188 if (events->smi.smm) {
3189 if (events->smi.smm_inside_nmi)
3190 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3192 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3193 if (lapic_in_kernel(vcpu)) {
3194 if (events->smi.latched_init)
3195 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3197 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3202 kvm_make_request(KVM_REQ_EVENT, vcpu);
3207 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3208 struct kvm_debugregs *dbgregs)
3212 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3213 kvm_get_dr(vcpu, 6, &val);
3215 dbgregs->dr7 = vcpu->arch.dr7;
3217 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3220 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3221 struct kvm_debugregs *dbgregs)
3226 if (dbgregs->dr6 & ~0xffffffffull)
3228 if (dbgregs->dr7 & ~0xffffffffull)
3231 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3232 kvm_update_dr0123(vcpu);
3233 vcpu->arch.dr6 = dbgregs->dr6;
3234 kvm_update_dr6(vcpu);
3235 vcpu->arch.dr7 = dbgregs->dr7;
3236 kvm_update_dr7(vcpu);
3241 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3243 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3245 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3246 u64 xstate_bv = xsave->header.xfeatures;
3250 * Copy legacy XSAVE area, to avoid complications with CPUID
3251 * leaves 0 and 1 in the loop below.
3253 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3256 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3257 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3260 * Copy each region from the possibly compacted offset to the
3261 * non-compacted offset.
3263 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3265 u64 feature = valid & -valid;
3266 int index = fls64(feature) - 1;
3267 void *src = get_xsave_addr(xsave, feature);
3270 u32 size, offset, ecx, edx;
3271 cpuid_count(XSTATE_CPUID, index,
3272 &size, &offset, &ecx, &edx);
3273 if (feature == XFEATURE_MASK_PKRU)
3274 memcpy(dest + offset, &vcpu->arch.pkru,
3275 sizeof(vcpu->arch.pkru));
3277 memcpy(dest + offset, src, size);
3285 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3287 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3288 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3292 * Copy legacy XSAVE area, to avoid complications with CPUID
3293 * leaves 0 and 1 in the loop below.
3295 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3297 /* Set XSTATE_BV and possibly XCOMP_BV. */
3298 xsave->header.xfeatures = xstate_bv;
3299 if (boot_cpu_has(X86_FEATURE_XSAVES))
3300 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3303 * Copy each region from the non-compacted offset to the
3304 * possibly compacted offset.
3306 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3308 u64 feature = valid & -valid;
3309 int index = fls64(feature) - 1;
3310 void *dest = get_xsave_addr(xsave, feature);
3313 u32 size, offset, ecx, edx;
3314 cpuid_count(XSTATE_CPUID, index,
3315 &size, &offset, &ecx, &edx);
3316 if (feature == XFEATURE_MASK_PKRU)
3317 memcpy(&vcpu->arch.pkru, src + offset,
3318 sizeof(vcpu->arch.pkru));
3320 memcpy(dest, src + offset, size);
3327 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3328 struct kvm_xsave *guest_xsave)
3330 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3331 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3332 fill_xsave((u8 *) guest_xsave->region, vcpu);
3334 memcpy(guest_xsave->region,
3335 &vcpu->arch.guest_fpu.state.fxsave,
3336 sizeof(struct fxregs_state));
3337 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3338 XFEATURE_MASK_FPSSE;
3342 #define XSAVE_MXCSR_OFFSET 24
3344 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3345 struct kvm_xsave *guest_xsave)
3348 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3349 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3351 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3353 * Here we allow setting states that are not present in
3354 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3355 * with old userspace.
3357 if (xstate_bv & ~kvm_supported_xcr0() ||
3358 mxcsr & ~mxcsr_feature_mask)
3360 load_xsave(vcpu, (u8 *)guest_xsave->region);
3362 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3363 mxcsr & ~mxcsr_feature_mask)
3365 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3366 guest_xsave->region, sizeof(struct fxregs_state));
3371 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3372 struct kvm_xcrs *guest_xcrs)
3374 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3375 guest_xcrs->nr_xcrs = 0;
3379 guest_xcrs->nr_xcrs = 1;
3380 guest_xcrs->flags = 0;
3381 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3382 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3385 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3386 struct kvm_xcrs *guest_xcrs)
3390 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3393 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3396 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3397 /* Only support XCR0 currently */
3398 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3399 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3400 guest_xcrs->xcrs[i].value);
3409 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3410 * stopped by the hypervisor. This function will be called from the host only.
3411 * EINVAL is returned when the host attempts to set the flag for a guest that
3412 * does not support pv clocks.
3414 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3416 if (!vcpu->arch.pv_time_enabled)
3418 vcpu->arch.pvclock_set_guest_stopped_request = true;
3419 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3423 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3424 struct kvm_enable_cap *cap)
3430 case KVM_CAP_HYPERV_SYNIC2:
3433 case KVM_CAP_HYPERV_SYNIC:
3434 if (!irqchip_in_kernel(vcpu->kvm))
3436 return kvm_hv_activate_synic(vcpu, cap->cap ==
3437 KVM_CAP_HYPERV_SYNIC2);
3443 long kvm_arch_vcpu_ioctl(struct file *filp,
3444 unsigned int ioctl, unsigned long arg)
3446 struct kvm_vcpu *vcpu = filp->private_data;
3447 void __user *argp = (void __user *)arg;
3450 struct kvm_lapic_state *lapic;
3451 struct kvm_xsave *xsave;
3452 struct kvm_xcrs *xcrs;
3458 case KVM_GET_LAPIC: {
3460 if (!lapic_in_kernel(vcpu))
3462 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3467 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3471 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3476 case KVM_SET_LAPIC: {
3478 if (!lapic_in_kernel(vcpu))
3480 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3481 if (IS_ERR(u.lapic))
3482 return PTR_ERR(u.lapic);
3484 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3487 case KVM_INTERRUPT: {
3488 struct kvm_interrupt irq;
3491 if (copy_from_user(&irq, argp, sizeof irq))
3493 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3497 r = kvm_vcpu_ioctl_nmi(vcpu);
3501 r = kvm_vcpu_ioctl_smi(vcpu);
3504 case KVM_SET_CPUID: {
3505 struct kvm_cpuid __user *cpuid_arg = argp;
3506 struct kvm_cpuid cpuid;
3509 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3511 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3514 case KVM_SET_CPUID2: {
3515 struct kvm_cpuid2 __user *cpuid_arg = argp;
3516 struct kvm_cpuid2 cpuid;
3519 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3521 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3522 cpuid_arg->entries);
3525 case KVM_GET_CPUID2: {
3526 struct kvm_cpuid2 __user *cpuid_arg = argp;
3527 struct kvm_cpuid2 cpuid;
3530 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3532 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3533 cpuid_arg->entries);
3537 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3543 r = msr_io(vcpu, argp, do_get_msr, 1);
3546 r = msr_io(vcpu, argp, do_set_msr, 0);
3548 case KVM_TPR_ACCESS_REPORTING: {
3549 struct kvm_tpr_access_ctl tac;
3552 if (copy_from_user(&tac, argp, sizeof tac))
3554 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3558 if (copy_to_user(argp, &tac, sizeof tac))
3563 case KVM_SET_VAPIC_ADDR: {
3564 struct kvm_vapic_addr va;
3568 if (!lapic_in_kernel(vcpu))
3571 if (copy_from_user(&va, argp, sizeof va))
3573 idx = srcu_read_lock(&vcpu->kvm->srcu);
3574 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3575 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3578 case KVM_X86_SETUP_MCE: {
3582 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3584 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3587 case KVM_X86_SET_MCE: {
3588 struct kvm_x86_mce mce;
3591 if (copy_from_user(&mce, argp, sizeof mce))
3593 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3596 case KVM_GET_VCPU_EVENTS: {
3597 struct kvm_vcpu_events events;
3599 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3602 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3607 case KVM_SET_VCPU_EVENTS: {
3608 struct kvm_vcpu_events events;
3611 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3614 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3617 case KVM_GET_DEBUGREGS: {
3618 struct kvm_debugregs dbgregs;
3620 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3623 if (copy_to_user(argp, &dbgregs,
3624 sizeof(struct kvm_debugregs)))
3629 case KVM_SET_DEBUGREGS: {
3630 struct kvm_debugregs dbgregs;
3633 if (copy_from_user(&dbgregs, argp,
3634 sizeof(struct kvm_debugregs)))
3637 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3640 case KVM_GET_XSAVE: {
3641 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3646 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3649 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3654 case KVM_SET_XSAVE: {
3655 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3656 if (IS_ERR(u.xsave))
3657 return PTR_ERR(u.xsave);
3659 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3662 case KVM_GET_XCRS: {
3663 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3668 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3671 if (copy_to_user(argp, u.xcrs,
3672 sizeof(struct kvm_xcrs)))
3677 case KVM_SET_XCRS: {
3678 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3680 return PTR_ERR(u.xcrs);
3682 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3685 case KVM_SET_TSC_KHZ: {
3689 user_tsc_khz = (u32)arg;
3691 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3694 if (user_tsc_khz == 0)
3695 user_tsc_khz = tsc_khz;
3697 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3702 case KVM_GET_TSC_KHZ: {
3703 r = vcpu->arch.virtual_tsc_khz;
3706 case KVM_KVMCLOCK_CTRL: {
3707 r = kvm_set_guest_paused(vcpu);
3710 case KVM_ENABLE_CAP: {
3711 struct kvm_enable_cap cap;
3714 if (copy_from_user(&cap, argp, sizeof(cap)))
3716 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3727 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3729 return VM_FAULT_SIGBUS;
3732 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3736 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3738 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3742 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3745 kvm->arch.ept_identity_map_addr = ident_addr;
3749 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3750 u32 kvm_nr_mmu_pages)
3752 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3755 mutex_lock(&kvm->slots_lock);
3757 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3758 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3760 mutex_unlock(&kvm->slots_lock);
3764 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3766 return kvm->arch.n_max_mmu_pages;
3769 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3771 struct kvm_pic *pic = kvm->arch.vpic;
3775 switch (chip->chip_id) {
3776 case KVM_IRQCHIP_PIC_MASTER:
3777 memcpy(&chip->chip.pic, &pic->pics[0],
3778 sizeof(struct kvm_pic_state));
3780 case KVM_IRQCHIP_PIC_SLAVE:
3781 memcpy(&chip->chip.pic, &pic->pics[1],
3782 sizeof(struct kvm_pic_state));
3784 case KVM_IRQCHIP_IOAPIC:
3785 kvm_get_ioapic(kvm, &chip->chip.ioapic);
3794 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3796 struct kvm_pic *pic = kvm->arch.vpic;
3800 switch (chip->chip_id) {
3801 case KVM_IRQCHIP_PIC_MASTER:
3802 spin_lock(&pic->lock);
3803 memcpy(&pic->pics[0], &chip->chip.pic,
3804 sizeof(struct kvm_pic_state));
3805 spin_unlock(&pic->lock);
3807 case KVM_IRQCHIP_PIC_SLAVE:
3808 spin_lock(&pic->lock);
3809 memcpy(&pic->pics[1], &chip->chip.pic,
3810 sizeof(struct kvm_pic_state));
3811 spin_unlock(&pic->lock);
3813 case KVM_IRQCHIP_IOAPIC:
3814 kvm_set_ioapic(kvm, &chip->chip.ioapic);
3820 kvm_pic_update_irq(pic);
3824 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3826 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3828 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3830 mutex_lock(&kps->lock);
3831 memcpy(ps, &kps->channels, sizeof(*ps));
3832 mutex_unlock(&kps->lock);
3836 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3839 struct kvm_pit *pit = kvm->arch.vpit;
3841 mutex_lock(&pit->pit_state.lock);
3842 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3843 for (i = 0; i < 3; i++)
3844 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3845 mutex_unlock(&pit->pit_state.lock);
3849 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3851 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3852 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3853 sizeof(ps->channels));
3854 ps->flags = kvm->arch.vpit->pit_state.flags;
3855 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3856 memset(&ps->reserved, 0, sizeof(ps->reserved));
3860 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3864 u32 prev_legacy, cur_legacy;
3865 struct kvm_pit *pit = kvm->arch.vpit;
3867 mutex_lock(&pit->pit_state.lock);
3868 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3869 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3870 if (!prev_legacy && cur_legacy)
3872 memcpy(&pit->pit_state.channels, &ps->channels,
3873 sizeof(pit->pit_state.channels));
3874 pit->pit_state.flags = ps->flags;
3875 for (i = 0; i < 3; i++)
3876 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3878 mutex_unlock(&pit->pit_state.lock);
3882 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3883 struct kvm_reinject_control *control)
3885 struct kvm_pit *pit = kvm->arch.vpit;
3890 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3891 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3892 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3894 mutex_lock(&pit->pit_state.lock);
3895 kvm_pit_set_reinject(pit, control->pit_reinject);
3896 mutex_unlock(&pit->pit_state.lock);
3902 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3903 * @kvm: kvm instance
3904 * @log: slot id and address to which we copy the log
3906 * Steps 1-4 below provide general overview of dirty page logging. See
3907 * kvm_get_dirty_log_protect() function description for additional details.
3909 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3910 * always flush the TLB (step 4) even if previous step failed and the dirty
3911 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3912 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3913 * writes will be marked dirty for next log read.
3915 * 1. Take a snapshot of the bit and clear it if needed.
3916 * 2. Write protect the corresponding page.
3917 * 3. Copy the snapshot to the userspace.
3918 * 4. Flush TLB's if needed.
3920 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3922 bool is_dirty = false;
3925 mutex_lock(&kvm->slots_lock);
3928 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3930 if (kvm_x86_ops->flush_log_dirty)
3931 kvm_x86_ops->flush_log_dirty(kvm);
3933 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3936 * All the TLBs can be flushed out of mmu lock, see the comments in
3937 * kvm_mmu_slot_remove_write_access().
3939 lockdep_assert_held(&kvm->slots_lock);
3941 kvm_flush_remote_tlbs(kvm);
3943 mutex_unlock(&kvm->slots_lock);
3947 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3950 if (!irqchip_in_kernel(kvm))
3953 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3954 irq_event->irq, irq_event->level,
3959 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3960 struct kvm_enable_cap *cap)
3968 case KVM_CAP_DISABLE_QUIRKS:
3969 kvm->arch.disabled_quirks = cap->args[0];
3972 case KVM_CAP_SPLIT_IRQCHIP: {
3973 mutex_lock(&kvm->lock);
3975 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3976 goto split_irqchip_unlock;
3978 if (irqchip_in_kernel(kvm))
3979 goto split_irqchip_unlock;
3980 if (kvm->created_vcpus)
3981 goto split_irqchip_unlock;
3982 r = kvm_setup_empty_irq_routing(kvm);
3984 goto split_irqchip_unlock;
3985 /* Pairs with irqchip_in_kernel. */
3987 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
3988 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3990 split_irqchip_unlock:
3991 mutex_unlock(&kvm->lock);
3994 case KVM_CAP_X2APIC_API:
3996 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3999 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4000 kvm->arch.x2apic_format = true;
4001 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4002 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4013 long kvm_arch_vm_ioctl(struct file *filp,
4014 unsigned int ioctl, unsigned long arg)
4016 struct kvm *kvm = filp->private_data;
4017 void __user *argp = (void __user *)arg;
4020 * This union makes it completely explicit to gcc-3.x
4021 * that these two variables' stack usage should be
4022 * combined, not added together.
4025 struct kvm_pit_state ps;
4026 struct kvm_pit_state2 ps2;
4027 struct kvm_pit_config pit_config;
4031 case KVM_SET_TSS_ADDR:
4032 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4034 case KVM_SET_IDENTITY_MAP_ADDR: {
4038 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4040 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4043 case KVM_SET_NR_MMU_PAGES:
4044 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4046 case KVM_GET_NR_MMU_PAGES:
4047 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4049 case KVM_CREATE_IRQCHIP: {
4050 mutex_lock(&kvm->lock);
4053 if (irqchip_in_kernel(kvm))
4054 goto create_irqchip_unlock;
4057 if (kvm->created_vcpus)
4058 goto create_irqchip_unlock;
4060 r = kvm_pic_init(kvm);
4062 goto create_irqchip_unlock;
4064 r = kvm_ioapic_init(kvm);
4066 kvm_pic_destroy(kvm);
4067 goto create_irqchip_unlock;
4070 r = kvm_setup_default_irq_routing(kvm);
4072 kvm_ioapic_destroy(kvm);
4073 kvm_pic_destroy(kvm);
4074 goto create_irqchip_unlock;
4076 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4078 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4079 create_irqchip_unlock:
4080 mutex_unlock(&kvm->lock);
4083 case KVM_CREATE_PIT:
4084 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4086 case KVM_CREATE_PIT2:
4088 if (copy_from_user(&u.pit_config, argp,
4089 sizeof(struct kvm_pit_config)))
4092 mutex_lock(&kvm->lock);
4095 goto create_pit_unlock;
4097 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4101 mutex_unlock(&kvm->lock);
4103 case KVM_GET_IRQCHIP: {
4104 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4105 struct kvm_irqchip *chip;
4107 chip = memdup_user(argp, sizeof(*chip));
4114 if (!irqchip_kernel(kvm))
4115 goto get_irqchip_out;
4116 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4118 goto get_irqchip_out;
4120 if (copy_to_user(argp, chip, sizeof *chip))
4121 goto get_irqchip_out;
4127 case KVM_SET_IRQCHIP: {
4128 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4129 struct kvm_irqchip *chip;
4131 chip = memdup_user(argp, sizeof(*chip));
4138 if (!irqchip_kernel(kvm))
4139 goto set_irqchip_out;
4140 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4142 goto set_irqchip_out;
4150 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4153 if (!kvm->arch.vpit)
4155 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4159 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4166 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4169 if (!kvm->arch.vpit)
4171 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4174 case KVM_GET_PIT2: {
4176 if (!kvm->arch.vpit)
4178 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4182 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4187 case KVM_SET_PIT2: {
4189 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4192 if (!kvm->arch.vpit)
4194 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4197 case KVM_REINJECT_CONTROL: {
4198 struct kvm_reinject_control control;
4200 if (copy_from_user(&control, argp, sizeof(control)))
4202 r = kvm_vm_ioctl_reinject(kvm, &control);
4205 case KVM_SET_BOOT_CPU_ID:
4207 mutex_lock(&kvm->lock);
4208 if (kvm->created_vcpus)
4211 kvm->arch.bsp_vcpu_id = arg;
4212 mutex_unlock(&kvm->lock);
4214 case KVM_XEN_HVM_CONFIG: {
4216 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4217 sizeof(struct kvm_xen_hvm_config)))
4220 if (kvm->arch.xen_hvm_config.flags)
4225 case KVM_SET_CLOCK: {
4226 struct kvm_clock_data user_ns;
4230 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4239 * TODO: userspace has to take care of races with VCPU_RUN, so
4240 * kvm_gen_update_masterclock() can be cut down to locked
4241 * pvclock_update_vm_gtod_copy().
4243 kvm_gen_update_masterclock(kvm);
4244 now_ns = get_kvmclock_ns(kvm);
4245 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4246 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4249 case KVM_GET_CLOCK: {
4250 struct kvm_clock_data user_ns;
4253 now_ns = get_kvmclock_ns(kvm);
4254 user_ns.clock = now_ns;
4255 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4256 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4259 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4264 case KVM_ENABLE_CAP: {
4265 struct kvm_enable_cap cap;
4268 if (copy_from_user(&cap, argp, sizeof(cap)))
4270 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4280 static void kvm_init_msr_list(void)
4285 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4286 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4290 * Even MSRs that are valid in the host may not be exposed
4291 * to the guests in some cases.
4293 switch (msrs_to_save[i]) {
4294 case MSR_IA32_BNDCFGS:
4295 if (!kvm_x86_ops->mpx_supported())
4299 if (!kvm_x86_ops->rdtscp_supported())
4307 msrs_to_save[j] = msrs_to_save[i];
4310 num_msrs_to_save = j;
4312 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4313 switch (emulated_msrs[i]) {
4314 case MSR_IA32_SMBASE:
4315 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4323 emulated_msrs[j] = emulated_msrs[i];
4326 num_emulated_msrs = j;
4329 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4337 if (!(lapic_in_kernel(vcpu) &&
4338 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4339 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4350 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4357 if (!(lapic_in_kernel(vcpu) &&
4358 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4360 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4362 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4372 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4373 struct kvm_segment *var, int seg)
4375 kvm_x86_ops->set_segment(vcpu, var, seg);
4378 void kvm_get_segment(struct kvm_vcpu *vcpu,
4379 struct kvm_segment *var, int seg)
4381 kvm_x86_ops->get_segment(vcpu, var, seg);
4384 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4385 struct x86_exception *exception)
4389 BUG_ON(!mmu_is_nested(vcpu));
4391 /* NPT walks are always user-walks */
4392 access |= PFERR_USER_MASK;
4393 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4398 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4399 struct x86_exception *exception)
4401 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4402 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4405 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4406 struct x86_exception *exception)
4408 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4409 access |= PFERR_FETCH_MASK;
4410 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4413 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4414 struct x86_exception *exception)
4416 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4417 access |= PFERR_WRITE_MASK;
4418 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4421 /* uses this to access any guest's mapped memory without checking CPL */
4422 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4423 struct x86_exception *exception)
4425 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4428 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4429 struct kvm_vcpu *vcpu, u32 access,
4430 struct x86_exception *exception)
4433 int r = X86EMUL_CONTINUE;
4436 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4438 unsigned offset = addr & (PAGE_SIZE-1);
4439 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4442 if (gpa == UNMAPPED_GVA)
4443 return X86EMUL_PROPAGATE_FAULT;
4444 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4447 r = X86EMUL_IO_NEEDED;
4459 /* used for instruction fetching */
4460 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4461 gva_t addr, void *val, unsigned int bytes,
4462 struct x86_exception *exception)
4464 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4465 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4469 /* Inline kvm_read_guest_virt_helper for speed. */
4470 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4472 if (unlikely(gpa == UNMAPPED_GVA))
4473 return X86EMUL_PROPAGATE_FAULT;
4475 offset = addr & (PAGE_SIZE-1);
4476 if (WARN_ON(offset + bytes > PAGE_SIZE))
4477 bytes = (unsigned)PAGE_SIZE - offset;
4478 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4480 if (unlikely(ret < 0))
4481 return X86EMUL_IO_NEEDED;
4483 return X86EMUL_CONTINUE;
4486 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4487 gva_t addr, void *val, unsigned int bytes,
4488 struct x86_exception *exception)
4490 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4491 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4493 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4496 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4498 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4499 gva_t addr, void *val, unsigned int bytes,
4500 struct x86_exception *exception)
4502 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4503 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4506 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4507 unsigned long addr, void *val, unsigned int bytes)
4509 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4510 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4512 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4515 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4516 gva_t addr, void *val,
4518 struct x86_exception *exception)
4520 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4522 int r = X86EMUL_CONTINUE;
4525 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4528 unsigned offset = addr & (PAGE_SIZE-1);
4529 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4532 if (gpa == UNMAPPED_GVA)
4533 return X86EMUL_PROPAGATE_FAULT;
4534 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4536 r = X86EMUL_IO_NEEDED;
4547 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4549 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4550 gpa_t gpa, bool write)
4552 /* For APIC access vmexit */
4553 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4556 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4557 trace_vcpu_match_mmio(gva, gpa, write, true);
4564 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4565 gpa_t *gpa, struct x86_exception *exception,
4568 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4569 | (write ? PFERR_WRITE_MASK : 0);
4572 * currently PKRU is only applied to ept enabled guest so
4573 * there is no pkey in EPT page table for L1 guest or EPT
4574 * shadow page table for L2 guest.
4576 if (vcpu_match_mmio_gva(vcpu, gva)
4577 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4578 vcpu->arch.access, 0, access)) {
4579 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4580 (gva & (PAGE_SIZE - 1));
4581 trace_vcpu_match_mmio(gva, *gpa, write, false);
4585 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4587 if (*gpa == UNMAPPED_GVA)
4590 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4593 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4594 const void *val, int bytes)
4598 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4601 kvm_page_track_write(vcpu, gpa, val, bytes);
4605 struct read_write_emulator_ops {
4606 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4608 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4609 void *val, int bytes);
4610 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4611 int bytes, void *val);
4612 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4613 void *val, int bytes);
4617 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4619 if (vcpu->mmio_read_completed) {
4620 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4621 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4622 vcpu->mmio_read_completed = 0;
4629 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4630 void *val, int bytes)
4632 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4635 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4636 void *val, int bytes)
4638 return emulator_write_phys(vcpu, gpa, val, bytes);
4641 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4643 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4644 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4647 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4648 void *val, int bytes)
4650 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4651 return X86EMUL_IO_NEEDED;
4654 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4655 void *val, int bytes)
4657 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4659 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4660 return X86EMUL_CONTINUE;
4663 static const struct read_write_emulator_ops read_emultor = {
4664 .read_write_prepare = read_prepare,
4665 .read_write_emulate = read_emulate,
4666 .read_write_mmio = vcpu_mmio_read,
4667 .read_write_exit_mmio = read_exit_mmio,
4670 static const struct read_write_emulator_ops write_emultor = {
4671 .read_write_emulate = write_emulate,
4672 .read_write_mmio = write_mmio,
4673 .read_write_exit_mmio = write_exit_mmio,
4677 static int emulator_read_write_onepage(unsigned long addr, void *val,
4679 struct x86_exception *exception,
4680 struct kvm_vcpu *vcpu,
4681 const struct read_write_emulator_ops *ops)
4685 bool write = ops->write;
4686 struct kvm_mmio_fragment *frag;
4687 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4690 * If the exit was due to a NPF we may already have a GPA.
4691 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4692 * Note, this cannot be used on string operations since string
4693 * operation using rep will only have the initial GPA from the NPF
4696 if (vcpu->arch.gpa_available &&
4697 emulator_can_use_gpa(ctxt) &&
4698 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
4699 gpa = vcpu->arch.gpa_val;
4700 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
4702 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4704 return X86EMUL_PROPAGATE_FAULT;
4707 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
4708 return X86EMUL_CONTINUE;
4711 * Is this MMIO handled locally?
4713 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4714 if (handled == bytes)
4715 return X86EMUL_CONTINUE;
4721 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4722 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4726 return X86EMUL_CONTINUE;
4729 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4731 void *val, unsigned int bytes,
4732 struct x86_exception *exception,
4733 const struct read_write_emulator_ops *ops)
4735 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4739 if (ops->read_write_prepare &&
4740 ops->read_write_prepare(vcpu, val, bytes))
4741 return X86EMUL_CONTINUE;
4743 vcpu->mmio_nr_fragments = 0;
4745 /* Crossing a page boundary? */
4746 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4749 now = -addr & ~PAGE_MASK;
4750 rc = emulator_read_write_onepage(addr, val, now, exception,
4753 if (rc != X86EMUL_CONTINUE)
4756 if (ctxt->mode != X86EMUL_MODE_PROT64)
4762 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4764 if (rc != X86EMUL_CONTINUE)
4767 if (!vcpu->mmio_nr_fragments)
4770 gpa = vcpu->mmio_fragments[0].gpa;
4772 vcpu->mmio_needed = 1;
4773 vcpu->mmio_cur_fragment = 0;
4775 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4776 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4777 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4778 vcpu->run->mmio.phys_addr = gpa;
4780 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4783 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4787 struct x86_exception *exception)
4789 return emulator_read_write(ctxt, addr, val, bytes,
4790 exception, &read_emultor);
4793 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4797 struct x86_exception *exception)
4799 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4800 exception, &write_emultor);
4803 #define CMPXCHG_TYPE(t, ptr, old, new) \
4804 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4806 #ifdef CONFIG_X86_64
4807 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4809 # define CMPXCHG64(ptr, old, new) \
4810 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4813 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4818 struct x86_exception *exception)
4820 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4826 /* guests cmpxchg8b have to be emulated atomically */
4827 if (bytes > 8 || (bytes & (bytes - 1)))
4830 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4832 if (gpa == UNMAPPED_GVA ||
4833 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4836 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4839 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4840 if (is_error_page(page))
4843 kaddr = kmap_atomic(page);
4844 kaddr += offset_in_page(gpa);
4847 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4850 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4853 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4856 exchanged = CMPXCHG64(kaddr, old, new);
4861 kunmap_atomic(kaddr);
4862 kvm_release_page_dirty(page);
4865 return X86EMUL_CMPXCHG_FAILED;
4867 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4868 kvm_page_track_write(vcpu, gpa, new, bytes);
4870 return X86EMUL_CONTINUE;
4873 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4875 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4878 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4882 for (i = 0; i < vcpu->arch.pio.count; i++) {
4883 if (vcpu->arch.pio.in)
4884 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4885 vcpu->arch.pio.size, pd);
4887 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4888 vcpu->arch.pio.port, vcpu->arch.pio.size,
4892 pd += vcpu->arch.pio.size;
4897 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4898 unsigned short port, void *val,
4899 unsigned int count, bool in)
4901 vcpu->arch.pio.port = port;
4902 vcpu->arch.pio.in = in;
4903 vcpu->arch.pio.count = count;
4904 vcpu->arch.pio.size = size;
4906 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4907 vcpu->arch.pio.count = 0;
4911 vcpu->run->exit_reason = KVM_EXIT_IO;
4912 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4913 vcpu->run->io.size = size;
4914 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4915 vcpu->run->io.count = count;
4916 vcpu->run->io.port = port;
4921 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4922 int size, unsigned short port, void *val,
4925 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4928 if (vcpu->arch.pio.count)
4931 memset(vcpu->arch.pio_data, 0, size * count);
4933 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4936 memcpy(val, vcpu->arch.pio_data, size * count);
4937 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4938 vcpu->arch.pio.count = 0;
4945 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4946 int size, unsigned short port,
4947 const void *val, unsigned int count)
4949 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4951 memcpy(vcpu->arch.pio_data, val, size * count);
4952 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4953 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4956 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4958 return kvm_x86_ops->get_segment_base(vcpu, seg);
4961 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4963 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4966 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4968 if (!need_emulate_wbinvd(vcpu))
4969 return X86EMUL_CONTINUE;
4971 if (kvm_x86_ops->has_wbinvd_exit()) {
4972 int cpu = get_cpu();
4974 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4975 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4976 wbinvd_ipi, NULL, 1);
4978 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4981 return X86EMUL_CONTINUE;
4984 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4986 kvm_emulate_wbinvd_noskip(vcpu);
4987 return kvm_skip_emulated_instruction(vcpu);
4989 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4993 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4995 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4998 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4999 unsigned long *dest)
5001 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5004 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5005 unsigned long value)
5008 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5011 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5013 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5016 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5018 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5019 unsigned long value;
5023 value = kvm_read_cr0(vcpu);
5026 value = vcpu->arch.cr2;
5029 value = kvm_read_cr3(vcpu);
5032 value = kvm_read_cr4(vcpu);
5035 value = kvm_get_cr8(vcpu);
5038 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5045 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5047 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5052 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5055 vcpu->arch.cr2 = val;
5058 res = kvm_set_cr3(vcpu, val);
5061 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5064 res = kvm_set_cr8(vcpu, val);
5067 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5074 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5076 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5079 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5081 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5084 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5086 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5089 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5091 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5094 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5096 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5099 static unsigned long emulator_get_cached_segment_base(
5100 struct x86_emulate_ctxt *ctxt, int seg)
5102 return get_segment_base(emul_to_vcpu(ctxt), seg);
5105 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5106 struct desc_struct *desc, u32 *base3,
5109 struct kvm_segment var;
5111 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5112 *selector = var.selector;
5115 memset(desc, 0, sizeof(*desc));
5123 set_desc_limit(desc, var.limit);
5124 set_desc_base(desc, (unsigned long)var.base);
5125 #ifdef CONFIG_X86_64
5127 *base3 = var.base >> 32;
5129 desc->type = var.type;
5131 desc->dpl = var.dpl;
5132 desc->p = var.present;
5133 desc->avl = var.avl;
5141 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5142 struct desc_struct *desc, u32 base3,
5145 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5146 struct kvm_segment var;
5148 var.selector = selector;
5149 var.base = get_desc_base(desc);
5150 #ifdef CONFIG_X86_64
5151 var.base |= ((u64)base3) << 32;
5153 var.limit = get_desc_limit(desc);
5155 var.limit = (var.limit << 12) | 0xfff;
5156 var.type = desc->type;
5157 var.dpl = desc->dpl;
5162 var.avl = desc->avl;
5163 var.present = desc->p;
5164 var.unusable = !var.present;
5167 kvm_set_segment(vcpu, &var, seg);
5171 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5172 u32 msr_index, u64 *pdata)
5174 struct msr_data msr;
5177 msr.index = msr_index;
5178 msr.host_initiated = false;
5179 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5187 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5188 u32 msr_index, u64 data)
5190 struct msr_data msr;
5193 msr.index = msr_index;
5194 msr.host_initiated = false;
5195 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5198 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5200 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5202 return vcpu->arch.smbase;
5205 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5207 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5209 vcpu->arch.smbase = smbase;
5212 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5215 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5218 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5219 u32 pmc, u64 *pdata)
5221 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5224 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5226 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5229 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5232 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5235 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5240 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5241 struct x86_instruction_info *info,
5242 enum x86_intercept_stage stage)
5244 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5247 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5248 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5250 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5253 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5255 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5258 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5260 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5263 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5265 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5268 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5270 return emul_to_vcpu(ctxt)->arch.hflags;
5273 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5275 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5278 static const struct x86_emulate_ops emulate_ops = {
5279 .read_gpr = emulator_read_gpr,
5280 .write_gpr = emulator_write_gpr,
5281 .read_std = kvm_read_guest_virt_system,
5282 .write_std = kvm_write_guest_virt_system,
5283 .read_phys = kvm_read_guest_phys_system,
5284 .fetch = kvm_fetch_guest_virt,
5285 .read_emulated = emulator_read_emulated,
5286 .write_emulated = emulator_write_emulated,
5287 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5288 .invlpg = emulator_invlpg,
5289 .pio_in_emulated = emulator_pio_in_emulated,
5290 .pio_out_emulated = emulator_pio_out_emulated,
5291 .get_segment = emulator_get_segment,
5292 .set_segment = emulator_set_segment,
5293 .get_cached_segment_base = emulator_get_cached_segment_base,
5294 .get_gdt = emulator_get_gdt,
5295 .get_idt = emulator_get_idt,
5296 .set_gdt = emulator_set_gdt,
5297 .set_idt = emulator_set_idt,
5298 .get_cr = emulator_get_cr,
5299 .set_cr = emulator_set_cr,
5300 .cpl = emulator_get_cpl,
5301 .get_dr = emulator_get_dr,
5302 .set_dr = emulator_set_dr,
5303 .get_smbase = emulator_get_smbase,
5304 .set_smbase = emulator_set_smbase,
5305 .set_msr = emulator_set_msr,
5306 .get_msr = emulator_get_msr,
5307 .check_pmc = emulator_check_pmc,
5308 .read_pmc = emulator_read_pmc,
5309 .halt = emulator_halt,
5310 .wbinvd = emulator_wbinvd,
5311 .fix_hypercall = emulator_fix_hypercall,
5312 .get_fpu = emulator_get_fpu,
5313 .put_fpu = emulator_put_fpu,
5314 .intercept = emulator_intercept,
5315 .get_cpuid = emulator_get_cpuid,
5316 .set_nmi_mask = emulator_set_nmi_mask,
5317 .get_hflags = emulator_get_hflags,
5318 .set_hflags = emulator_set_hflags,
5321 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5323 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5325 * an sti; sti; sequence only disable interrupts for the first
5326 * instruction. So, if the last instruction, be it emulated or
5327 * not, left the system with the INT_STI flag enabled, it
5328 * means that the last instruction is an sti. We should not
5329 * leave the flag on in this case. The same goes for mov ss
5331 if (int_shadow & mask)
5333 if (unlikely(int_shadow || mask)) {
5334 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5336 kvm_make_request(KVM_REQ_EVENT, vcpu);
5340 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5342 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5343 if (ctxt->exception.vector == PF_VECTOR)
5344 return kvm_propagate_fault(vcpu, &ctxt->exception);
5346 if (ctxt->exception.error_code_valid)
5347 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5348 ctxt->exception.error_code);
5350 kvm_queue_exception(vcpu, ctxt->exception.vector);
5354 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5356 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5359 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5361 ctxt->eflags = kvm_get_rflags(vcpu);
5362 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5364 ctxt->eip = kvm_rip_read(vcpu);
5365 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5366 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5367 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5368 cs_db ? X86EMUL_MODE_PROT32 :
5369 X86EMUL_MODE_PROT16;
5370 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5371 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5372 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5374 init_decode_cache(ctxt);
5375 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5378 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5380 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5383 init_emulate_ctxt(vcpu);
5387 ctxt->_eip = ctxt->eip + inc_eip;
5388 ret = emulate_int_real(ctxt, irq);
5390 if (ret != X86EMUL_CONTINUE)
5391 return EMULATE_FAIL;
5393 ctxt->eip = ctxt->_eip;
5394 kvm_rip_write(vcpu, ctxt->eip);
5395 kvm_set_rflags(vcpu, ctxt->eflags);
5397 if (irq == NMI_VECTOR)
5398 vcpu->arch.nmi_pending = 0;
5400 vcpu->arch.interrupt.pending = false;
5402 return EMULATE_DONE;
5404 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5406 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5408 int r = EMULATE_DONE;
5410 ++vcpu->stat.insn_emulation_fail;
5411 trace_kvm_emulate_insn_failed(vcpu);
5412 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5413 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5414 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5415 vcpu->run->internal.ndata = 0;
5418 kvm_queue_exception(vcpu, UD_VECTOR);
5423 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5424 bool write_fault_to_shadow_pgtable,
5430 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5433 if (!vcpu->arch.mmu.direct_map) {
5435 * Write permission should be allowed since only
5436 * write access need to be emulated.
5438 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5441 * If the mapping is invalid in guest, let cpu retry
5442 * it to generate fault.
5444 if (gpa == UNMAPPED_GVA)
5449 * Do not retry the unhandleable instruction if it faults on the
5450 * readonly host memory, otherwise it will goto a infinite loop:
5451 * retry instruction -> write #PF -> emulation fail -> retry
5452 * instruction -> ...
5454 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5457 * If the instruction failed on the error pfn, it can not be fixed,
5458 * report the error to userspace.
5460 if (is_error_noslot_pfn(pfn))
5463 kvm_release_pfn_clean(pfn);
5465 /* The instructions are well-emulated on direct mmu. */
5466 if (vcpu->arch.mmu.direct_map) {
5467 unsigned int indirect_shadow_pages;
5469 spin_lock(&vcpu->kvm->mmu_lock);
5470 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5471 spin_unlock(&vcpu->kvm->mmu_lock);
5473 if (indirect_shadow_pages)
5474 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5480 * if emulation was due to access to shadowed page table
5481 * and it failed try to unshadow page and re-enter the
5482 * guest to let CPU execute the instruction.
5484 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5487 * If the access faults on its page table, it can not
5488 * be fixed by unprotecting shadow page and it should
5489 * be reported to userspace.
5491 return !write_fault_to_shadow_pgtable;
5494 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5495 unsigned long cr2, int emulation_type)
5497 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5498 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5500 last_retry_eip = vcpu->arch.last_retry_eip;
5501 last_retry_addr = vcpu->arch.last_retry_addr;
5504 * If the emulation is caused by #PF and it is non-page_table
5505 * writing instruction, it means the VM-EXIT is caused by shadow
5506 * page protected, we can zap the shadow page and retry this
5507 * instruction directly.
5509 * Note: if the guest uses a non-page-table modifying instruction
5510 * on the PDE that points to the instruction, then we will unmap
5511 * the instruction and go to an infinite loop. So, we cache the
5512 * last retried eip and the last fault address, if we meet the eip
5513 * and the address again, we can break out of the potential infinite
5516 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5518 if (!(emulation_type & EMULTYPE_RETRY))
5521 if (x86_page_table_writing_insn(ctxt))
5524 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5527 vcpu->arch.last_retry_eip = ctxt->eip;
5528 vcpu->arch.last_retry_addr = cr2;
5530 if (!vcpu->arch.mmu.direct_map)
5531 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5533 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5538 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5539 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5541 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5543 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5544 /* This is a good place to trace that we are exiting SMM. */
5545 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5547 /* Process a latched INIT or SMI, if any. */
5548 kvm_make_request(KVM_REQ_EVENT, vcpu);
5551 kvm_mmu_reset_context(vcpu);
5554 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5556 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5558 vcpu->arch.hflags = emul_flags;
5560 if (changed & HF_SMM_MASK)
5561 kvm_smm_changed(vcpu);
5564 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5573 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5574 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5579 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5581 struct kvm_run *kvm_run = vcpu->run;
5583 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5584 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5585 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5586 kvm_run->debug.arch.exception = DB_VECTOR;
5587 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5588 *r = EMULATE_USER_EXIT;
5591 * "Certain debug exceptions may clear bit 0-3. The
5592 * remaining contents of the DR6 register are never
5593 * cleared by the processor".
5595 vcpu->arch.dr6 &= ~15;
5596 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5597 kvm_queue_exception(vcpu, DB_VECTOR);
5601 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5603 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5604 int r = EMULATE_DONE;
5606 kvm_x86_ops->skip_emulated_instruction(vcpu);
5609 * rflags is the old, "raw" value of the flags. The new value has
5610 * not been saved yet.
5612 * This is correct even for TF set by the guest, because "the
5613 * processor will not generate this exception after the instruction
5614 * that sets the TF flag".
5616 if (unlikely(rflags & X86_EFLAGS_TF))
5617 kvm_vcpu_do_singlestep(vcpu, &r);
5618 return r == EMULATE_DONE;
5620 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5622 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5624 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5625 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5626 struct kvm_run *kvm_run = vcpu->run;
5627 unsigned long eip = kvm_get_linear_rip(vcpu);
5628 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5629 vcpu->arch.guest_debug_dr7,
5633 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5634 kvm_run->debug.arch.pc = eip;
5635 kvm_run->debug.arch.exception = DB_VECTOR;
5636 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5637 *r = EMULATE_USER_EXIT;
5642 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5643 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5644 unsigned long eip = kvm_get_linear_rip(vcpu);
5645 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5650 vcpu->arch.dr6 &= ~15;
5651 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5652 kvm_queue_exception(vcpu, DB_VECTOR);
5661 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5668 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5669 bool writeback = true;
5670 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5673 * Clear write_fault_to_shadow_pgtable here to ensure it is
5676 vcpu->arch.write_fault_to_shadow_pgtable = false;
5677 kvm_clear_exception_queue(vcpu);
5679 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5680 init_emulate_ctxt(vcpu);
5683 * We will reenter on the same instruction since
5684 * we do not set complete_userspace_io. This does not
5685 * handle watchpoints yet, those would be handled in
5688 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5691 ctxt->interruptibility = 0;
5692 ctxt->have_exception = false;
5693 ctxt->exception.vector = -1;
5694 ctxt->perm_ok = false;
5696 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5698 r = x86_decode_insn(ctxt, insn, insn_len);
5700 trace_kvm_emulate_insn_start(vcpu);
5701 ++vcpu->stat.insn_emulation;
5702 if (r != EMULATION_OK) {
5703 if (emulation_type & EMULTYPE_TRAP_UD)
5704 return EMULATE_FAIL;
5705 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5707 return EMULATE_DONE;
5708 if (emulation_type & EMULTYPE_SKIP)
5709 return EMULATE_FAIL;
5710 return handle_emulation_failure(vcpu);
5714 if (emulation_type & EMULTYPE_SKIP) {
5715 kvm_rip_write(vcpu, ctxt->_eip);
5716 if (ctxt->eflags & X86_EFLAGS_RF)
5717 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5718 return EMULATE_DONE;
5721 if (retry_instruction(ctxt, cr2, emulation_type))
5722 return EMULATE_DONE;
5724 /* this is needed for vmware backdoor interface to work since it
5725 changes registers values during IO operation */
5726 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5727 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5728 emulator_invalidate_register_cache(ctxt);
5732 /* Save the faulting GPA (cr2) in the address field */
5733 ctxt->exception.address = cr2;
5735 r = x86_emulate_insn(ctxt);
5737 if (r == EMULATION_INTERCEPTED)
5738 return EMULATE_DONE;
5740 if (r == EMULATION_FAILED) {
5741 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5743 return EMULATE_DONE;
5745 return handle_emulation_failure(vcpu);
5748 if (ctxt->have_exception) {
5750 if (inject_emulated_exception(vcpu))
5752 } else if (vcpu->arch.pio.count) {
5753 if (!vcpu->arch.pio.in) {
5754 /* FIXME: return into emulator if single-stepping. */
5755 vcpu->arch.pio.count = 0;
5758 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5760 r = EMULATE_USER_EXIT;
5761 } else if (vcpu->mmio_needed) {
5762 if (!vcpu->mmio_is_write)
5764 r = EMULATE_USER_EXIT;
5765 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5766 } else if (r == EMULATION_RESTART)
5772 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5773 toggle_interruptibility(vcpu, ctxt->interruptibility);
5774 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5775 kvm_rip_write(vcpu, ctxt->eip);
5776 if (r == EMULATE_DONE &&
5777 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
5778 kvm_vcpu_do_singlestep(vcpu, &r);
5779 if (!ctxt->have_exception ||
5780 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5781 __kvm_set_rflags(vcpu, ctxt->eflags);
5784 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5785 * do nothing, and it will be requested again as soon as
5786 * the shadow expires. But we still need to check here,
5787 * because POPF has no interrupt shadow.
5789 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5790 kvm_make_request(KVM_REQ_EVENT, vcpu);
5792 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5796 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5798 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5800 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5801 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5802 size, port, &val, 1);
5803 /* do not return to emulator after return from userspace */
5804 vcpu->arch.pio.count = 0;
5807 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5809 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5813 /* We should only ever be called with arch.pio.count equal to 1 */
5814 BUG_ON(vcpu->arch.pio.count != 1);
5816 /* For size less than 4 we merge, else we zero extend */
5817 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5821 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5822 * the copy and tracing
5824 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5825 vcpu->arch.pio.port, &val, 1);
5826 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5831 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5836 /* For size less than 4 we merge, else we zero extend */
5837 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5839 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5842 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5846 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5850 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5852 static int kvmclock_cpu_down_prep(unsigned int cpu)
5854 __this_cpu_write(cpu_tsc_khz, 0);
5858 static void tsc_khz_changed(void *data)
5860 struct cpufreq_freqs *freq = data;
5861 unsigned long khz = 0;
5865 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5866 khz = cpufreq_quick_get(raw_smp_processor_id());
5869 __this_cpu_write(cpu_tsc_khz, khz);
5872 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5875 struct cpufreq_freqs *freq = data;
5877 struct kvm_vcpu *vcpu;
5878 int i, send_ipi = 0;
5881 * We allow guests to temporarily run on slowing clocks,
5882 * provided we notify them after, or to run on accelerating
5883 * clocks, provided we notify them before. Thus time never
5886 * However, we have a problem. We can't atomically update
5887 * the frequency of a given CPU from this function; it is
5888 * merely a notifier, which can be called from any CPU.
5889 * Changing the TSC frequency at arbitrary points in time
5890 * requires a recomputation of local variables related to
5891 * the TSC for each VCPU. We must flag these local variables
5892 * to be updated and be sure the update takes place with the
5893 * new frequency before any guests proceed.
5895 * Unfortunately, the combination of hotplug CPU and frequency
5896 * change creates an intractable locking scenario; the order
5897 * of when these callouts happen is undefined with respect to
5898 * CPU hotplug, and they can race with each other. As such,
5899 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5900 * undefined; you can actually have a CPU frequency change take
5901 * place in between the computation of X and the setting of the
5902 * variable. To protect against this problem, all updates of
5903 * the per_cpu tsc_khz variable are done in an interrupt
5904 * protected IPI, and all callers wishing to update the value
5905 * must wait for a synchronous IPI to complete (which is trivial
5906 * if the caller is on the CPU already). This establishes the
5907 * necessary total order on variable updates.
5909 * Note that because a guest time update may take place
5910 * anytime after the setting of the VCPU's request bit, the
5911 * correct TSC value must be set before the request. However,
5912 * to ensure the update actually makes it to any guest which
5913 * starts running in hardware virtualization between the set
5914 * and the acquisition of the spinlock, we must also ping the
5915 * CPU after setting the request bit.
5919 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5921 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5924 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5926 spin_lock(&kvm_lock);
5927 list_for_each_entry(kvm, &vm_list, vm_list) {
5928 kvm_for_each_vcpu(i, vcpu, kvm) {
5929 if (vcpu->cpu != freq->cpu)
5931 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5932 if (vcpu->cpu != smp_processor_id())
5936 spin_unlock(&kvm_lock);
5938 if (freq->old < freq->new && send_ipi) {
5940 * We upscale the frequency. Must make the guest
5941 * doesn't see old kvmclock values while running with
5942 * the new frequency, otherwise we risk the guest sees
5943 * time go backwards.
5945 * In case we update the frequency for another cpu
5946 * (which might be in guest context) send an interrupt
5947 * to kick the cpu out of guest context. Next time
5948 * guest context is entered kvmclock will be updated,
5949 * so the guest will not see stale values.
5951 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5956 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5957 .notifier_call = kvmclock_cpufreq_notifier
5960 static int kvmclock_cpu_online(unsigned int cpu)
5962 tsc_khz_changed(NULL);
5966 static void kvm_timer_init(void)
5968 max_tsc_khz = tsc_khz;
5970 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5971 #ifdef CONFIG_CPU_FREQ
5972 struct cpufreq_policy policy;
5975 memset(&policy, 0, sizeof(policy));
5977 cpufreq_get_policy(&policy, cpu);
5978 if (policy.cpuinfo.max_freq)
5979 max_tsc_khz = policy.cpuinfo.max_freq;
5982 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5983 CPUFREQ_TRANSITION_NOTIFIER);
5985 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5987 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
5988 kvmclock_cpu_online, kvmclock_cpu_down_prep);
5991 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5993 int kvm_is_in_guest(void)
5995 return __this_cpu_read(current_vcpu) != NULL;
5998 static int kvm_is_user_mode(void)
6002 if (__this_cpu_read(current_vcpu))
6003 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6005 return user_mode != 0;
6008 static unsigned long kvm_get_guest_ip(void)
6010 unsigned long ip = 0;
6012 if (__this_cpu_read(current_vcpu))
6013 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6018 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6019 .is_in_guest = kvm_is_in_guest,
6020 .is_user_mode = kvm_is_user_mode,
6021 .get_guest_ip = kvm_get_guest_ip,
6024 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
6026 __this_cpu_write(current_vcpu, vcpu);
6028 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
6030 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
6032 __this_cpu_write(current_vcpu, NULL);
6034 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
6036 static void kvm_set_mmio_spte_mask(void)
6039 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6042 * Set the reserved bits and the present bit of an paging-structure
6043 * entry to generate page fault with PFER.RSV = 1.
6045 /* Mask the reserved physical address bits. */
6046 mask = rsvd_bits(maxphyaddr, 51);
6048 /* Set the present bit. */
6051 #ifdef CONFIG_X86_64
6053 * If reserved bit is not supported, clear the present bit to disable
6056 if (maxphyaddr == 52)
6060 kvm_mmu_set_mmio_spte_mask(mask, mask);
6063 #ifdef CONFIG_X86_64
6064 static void pvclock_gtod_update_fn(struct work_struct *work)
6068 struct kvm_vcpu *vcpu;
6071 spin_lock(&kvm_lock);
6072 list_for_each_entry(kvm, &vm_list, vm_list)
6073 kvm_for_each_vcpu(i, vcpu, kvm)
6074 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6075 atomic_set(&kvm_guest_has_master_clock, 0);
6076 spin_unlock(&kvm_lock);
6079 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6082 * Notification about pvclock gtod data update.
6084 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6087 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6088 struct timekeeper *tk = priv;
6090 update_pvclock_gtod(tk);
6092 /* disable master clock if host does not trust, or does not
6093 * use, TSC clocksource
6095 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6096 atomic_read(&kvm_guest_has_master_clock) != 0)
6097 queue_work(system_long_wq, &pvclock_gtod_work);
6102 static struct notifier_block pvclock_gtod_notifier = {
6103 .notifier_call = pvclock_gtod_notify,
6107 int kvm_arch_init(void *opaque)
6110 struct kvm_x86_ops *ops = opaque;
6113 printk(KERN_ERR "kvm: already loaded the other module\n");
6118 if (!ops->cpu_has_kvm_support()) {
6119 printk(KERN_ERR "kvm: no hardware support\n");
6123 if (ops->disabled_by_bios()) {
6124 printk(KERN_ERR "kvm: disabled by bios\n");
6130 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6132 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6136 r = kvm_mmu_module_init();
6138 goto out_free_percpu;
6140 kvm_set_mmio_spte_mask();
6144 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6145 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6146 PT_PRESENT_MASK, 0, sme_me_mask);
6149 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6151 if (boot_cpu_has(X86_FEATURE_XSAVE))
6152 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6155 #ifdef CONFIG_X86_64
6156 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6162 free_percpu(shared_msrs);
6167 void kvm_arch_exit(void)
6170 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6172 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6173 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6174 CPUFREQ_TRANSITION_NOTIFIER);
6175 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6176 #ifdef CONFIG_X86_64
6177 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6180 kvm_mmu_module_exit();
6181 free_percpu(shared_msrs);
6184 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6186 ++vcpu->stat.halt_exits;
6187 if (lapic_in_kernel(vcpu)) {
6188 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6191 vcpu->run->exit_reason = KVM_EXIT_HLT;
6195 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6197 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6199 int ret = kvm_skip_emulated_instruction(vcpu);
6201 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6202 * KVM_EXIT_DEBUG here.
6204 return kvm_vcpu_halt(vcpu) && ret;
6206 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6208 #ifdef CONFIG_X86_64
6209 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6210 unsigned long clock_type)
6212 struct kvm_clock_pairing clock_pairing;
6217 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6218 return -KVM_EOPNOTSUPP;
6220 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6221 return -KVM_EOPNOTSUPP;
6223 clock_pairing.sec = ts.tv_sec;
6224 clock_pairing.nsec = ts.tv_nsec;
6225 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6226 clock_pairing.flags = 0;
6229 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6230 sizeof(struct kvm_clock_pairing)))
6238 * kvm_pv_kick_cpu_op: Kick a vcpu.
6240 * @apicid - apicid of vcpu to be kicked.
6242 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6244 struct kvm_lapic_irq lapic_irq;
6246 lapic_irq.shorthand = 0;
6247 lapic_irq.dest_mode = 0;
6248 lapic_irq.level = 0;
6249 lapic_irq.dest_id = apicid;
6250 lapic_irq.msi_redir_hint = false;
6252 lapic_irq.delivery_mode = APIC_DM_REMRD;
6253 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6256 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6258 vcpu->arch.apicv_active = false;
6259 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6262 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6264 unsigned long nr, a0, a1, a2, a3, ret;
6267 r = kvm_skip_emulated_instruction(vcpu);
6269 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6270 return kvm_hv_hypercall(vcpu);
6272 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6273 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6274 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6275 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6276 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6278 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6280 op_64_bit = is_64_bit_mode(vcpu);
6289 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6295 case KVM_HC_VAPIC_POLL_IRQ:
6298 case KVM_HC_KICK_CPU:
6299 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6302 #ifdef CONFIG_X86_64
6303 case KVM_HC_CLOCK_PAIRING:
6304 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6314 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6315 ++vcpu->stat.hypercalls;
6318 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6320 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6322 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6323 char instruction[3];
6324 unsigned long rip = kvm_rip_read(vcpu);
6326 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6328 return emulator_write_emulated(ctxt, rip, instruction, 3,
6332 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6334 return vcpu->run->request_interrupt_window &&
6335 likely(!pic_in_kernel(vcpu->kvm));
6338 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6340 struct kvm_run *kvm_run = vcpu->run;
6342 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6343 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6344 kvm_run->cr8 = kvm_get_cr8(vcpu);
6345 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6346 kvm_run->ready_for_interrupt_injection =
6347 pic_in_kernel(vcpu->kvm) ||
6348 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6351 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6355 if (!kvm_x86_ops->update_cr8_intercept)
6358 if (!lapic_in_kernel(vcpu))
6361 if (vcpu->arch.apicv_active)
6364 if (!vcpu->arch.apic->vapic_addr)
6365 max_irr = kvm_lapic_find_highest_irr(vcpu);
6372 tpr = kvm_lapic_get_cr8(vcpu);
6374 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6377 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6381 /* try to reinject previous events if any */
6382 if (vcpu->arch.exception.injected) {
6383 kvm_x86_ops->queue_exception(vcpu);
6388 * Exceptions must be injected immediately, or the exception
6389 * frame will have the address of the NMI or interrupt handler.
6391 if (!vcpu->arch.exception.pending) {
6392 if (vcpu->arch.nmi_injected) {
6393 kvm_x86_ops->set_nmi(vcpu);
6397 if (vcpu->arch.interrupt.pending) {
6398 kvm_x86_ops->set_irq(vcpu);
6403 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6404 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6409 /* try to inject new event if pending */
6410 if (vcpu->arch.exception.pending) {
6411 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6412 vcpu->arch.exception.has_error_code,
6413 vcpu->arch.exception.error_code);
6415 vcpu->arch.exception.pending = false;
6416 vcpu->arch.exception.injected = true;
6418 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6419 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6422 if (vcpu->arch.exception.nr == DB_VECTOR &&
6423 (vcpu->arch.dr7 & DR7_GD)) {
6424 vcpu->arch.dr7 &= ~DR7_GD;
6425 kvm_update_dr7(vcpu);
6428 kvm_x86_ops->queue_exception(vcpu);
6429 } else if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6430 vcpu->arch.smi_pending = false;
6432 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6433 --vcpu->arch.nmi_pending;
6434 vcpu->arch.nmi_injected = true;
6435 kvm_x86_ops->set_nmi(vcpu);
6436 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6438 * Because interrupts can be injected asynchronously, we are
6439 * calling check_nested_events again here to avoid a race condition.
6440 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6441 * proposal and current concerns. Perhaps we should be setting
6442 * KVM_REQ_EVENT only on certain events and not unconditionally?
6444 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6445 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6449 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6450 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6452 kvm_x86_ops->set_irq(vcpu);
6459 static void process_nmi(struct kvm_vcpu *vcpu)
6464 * x86 is limited to one NMI running, and one NMI pending after it.
6465 * If an NMI is already in progress, limit further NMIs to just one.
6466 * Otherwise, allow two (and we'll inject the first one immediately).
6468 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6471 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6472 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6473 kvm_make_request(KVM_REQ_EVENT, vcpu);
6476 #define put_smstate(type, buf, offset, val) \
6477 *(type *)((buf) + (offset) - 0x7e00) = val
6479 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6482 flags |= seg->g << 23;
6483 flags |= seg->db << 22;
6484 flags |= seg->l << 21;
6485 flags |= seg->avl << 20;
6486 flags |= seg->present << 15;
6487 flags |= seg->dpl << 13;
6488 flags |= seg->s << 12;
6489 flags |= seg->type << 8;
6493 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6495 struct kvm_segment seg;
6498 kvm_get_segment(vcpu, &seg, n);
6499 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6502 offset = 0x7f84 + n * 12;
6504 offset = 0x7f2c + (n - 3) * 12;
6506 put_smstate(u32, buf, offset + 8, seg.base);
6507 put_smstate(u32, buf, offset + 4, seg.limit);
6508 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6511 #ifdef CONFIG_X86_64
6512 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6514 struct kvm_segment seg;
6518 kvm_get_segment(vcpu, &seg, n);
6519 offset = 0x7e00 + n * 16;
6521 flags = enter_smm_get_segment_flags(&seg) >> 8;
6522 put_smstate(u16, buf, offset, seg.selector);
6523 put_smstate(u16, buf, offset + 2, flags);
6524 put_smstate(u32, buf, offset + 4, seg.limit);
6525 put_smstate(u64, buf, offset + 8, seg.base);
6529 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6532 struct kvm_segment seg;
6536 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6537 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6538 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6539 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6541 for (i = 0; i < 8; i++)
6542 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6544 kvm_get_dr(vcpu, 6, &val);
6545 put_smstate(u32, buf, 0x7fcc, (u32)val);
6546 kvm_get_dr(vcpu, 7, &val);
6547 put_smstate(u32, buf, 0x7fc8, (u32)val);
6549 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6550 put_smstate(u32, buf, 0x7fc4, seg.selector);
6551 put_smstate(u32, buf, 0x7f64, seg.base);
6552 put_smstate(u32, buf, 0x7f60, seg.limit);
6553 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6555 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6556 put_smstate(u32, buf, 0x7fc0, seg.selector);
6557 put_smstate(u32, buf, 0x7f80, seg.base);
6558 put_smstate(u32, buf, 0x7f7c, seg.limit);
6559 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6561 kvm_x86_ops->get_gdt(vcpu, &dt);
6562 put_smstate(u32, buf, 0x7f74, dt.address);
6563 put_smstate(u32, buf, 0x7f70, dt.size);
6565 kvm_x86_ops->get_idt(vcpu, &dt);
6566 put_smstate(u32, buf, 0x7f58, dt.address);
6567 put_smstate(u32, buf, 0x7f54, dt.size);
6569 for (i = 0; i < 6; i++)
6570 enter_smm_save_seg_32(vcpu, buf, i);
6572 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6575 put_smstate(u32, buf, 0x7efc, 0x00020000);
6576 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6579 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6581 #ifdef CONFIG_X86_64
6583 struct kvm_segment seg;
6587 for (i = 0; i < 16; i++)
6588 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6590 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6591 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6593 kvm_get_dr(vcpu, 6, &val);
6594 put_smstate(u64, buf, 0x7f68, val);
6595 kvm_get_dr(vcpu, 7, &val);
6596 put_smstate(u64, buf, 0x7f60, val);
6598 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6599 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6600 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6602 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6605 put_smstate(u32, buf, 0x7efc, 0x00020064);
6607 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6609 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6610 put_smstate(u16, buf, 0x7e90, seg.selector);
6611 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6612 put_smstate(u32, buf, 0x7e94, seg.limit);
6613 put_smstate(u64, buf, 0x7e98, seg.base);
6615 kvm_x86_ops->get_idt(vcpu, &dt);
6616 put_smstate(u32, buf, 0x7e84, dt.size);
6617 put_smstate(u64, buf, 0x7e88, dt.address);
6619 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6620 put_smstate(u16, buf, 0x7e70, seg.selector);
6621 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6622 put_smstate(u32, buf, 0x7e74, seg.limit);
6623 put_smstate(u64, buf, 0x7e78, seg.base);
6625 kvm_x86_ops->get_gdt(vcpu, &dt);
6626 put_smstate(u32, buf, 0x7e64, dt.size);
6627 put_smstate(u64, buf, 0x7e68, dt.address);
6629 for (i = 0; i < 6; i++)
6630 enter_smm_save_seg_64(vcpu, buf, i);
6636 static void enter_smm(struct kvm_vcpu *vcpu)
6638 struct kvm_segment cs, ds;
6643 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6644 vcpu->arch.hflags |= HF_SMM_MASK;
6645 memset(buf, 0, 512);
6646 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6647 enter_smm_save_state_64(vcpu, buf);
6649 enter_smm_save_state_32(vcpu, buf);
6651 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6653 if (kvm_x86_ops->get_nmi_mask(vcpu))
6654 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6656 kvm_x86_ops->set_nmi_mask(vcpu, true);
6658 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6659 kvm_rip_write(vcpu, 0x8000);
6661 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6662 kvm_x86_ops->set_cr0(vcpu, cr0);
6663 vcpu->arch.cr0 = cr0;
6665 kvm_x86_ops->set_cr4(vcpu, 0);
6667 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6668 dt.address = dt.size = 0;
6669 kvm_x86_ops->set_idt(vcpu, &dt);
6671 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6673 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6674 cs.base = vcpu->arch.smbase;
6679 cs.limit = ds.limit = 0xffffffff;
6680 cs.type = ds.type = 0x3;
6681 cs.dpl = ds.dpl = 0;
6686 cs.avl = ds.avl = 0;
6687 cs.present = ds.present = 1;
6688 cs.unusable = ds.unusable = 0;
6689 cs.padding = ds.padding = 0;
6691 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6692 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6693 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6694 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6695 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6696 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6698 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6699 kvm_x86_ops->set_efer(vcpu, 0);
6701 kvm_update_cpuid(vcpu);
6702 kvm_mmu_reset_context(vcpu);
6705 static void process_smi(struct kvm_vcpu *vcpu)
6707 vcpu->arch.smi_pending = true;
6708 kvm_make_request(KVM_REQ_EVENT, vcpu);
6711 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6713 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6716 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6718 u64 eoi_exit_bitmap[4];
6720 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6723 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6725 if (irqchip_split(vcpu->kvm))
6726 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6728 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6729 kvm_x86_ops->sync_pir_to_irr(vcpu);
6730 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6732 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6733 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6734 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6737 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6739 ++vcpu->stat.tlb_flush;
6740 kvm_x86_ops->tlb_flush(vcpu);
6743 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6745 struct page *page = NULL;
6747 if (!lapic_in_kernel(vcpu))
6750 if (!kvm_x86_ops->set_apic_access_page_addr)
6753 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6754 if (is_error_page(page))
6756 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6759 * Do not pin apic access page in memory, the MMU notifier
6760 * will call us again if it is migrated or swapped out.
6764 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6767 * Returns 1 to let vcpu_run() continue the guest execution loop without
6768 * exiting to the userspace. Otherwise, the value will be returned to the
6771 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6775 dm_request_for_irq_injection(vcpu) &&
6776 kvm_cpu_accept_dm_intr(vcpu);
6778 bool req_immediate_exit = false;
6780 if (kvm_request_pending(vcpu)) {
6781 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6782 kvm_mmu_unload(vcpu);
6783 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6784 __kvm_migrate_timers(vcpu);
6785 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6786 kvm_gen_update_masterclock(vcpu->kvm);
6787 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6788 kvm_gen_kvmclock_update(vcpu);
6789 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6790 r = kvm_guest_time_update(vcpu);
6794 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6795 kvm_mmu_sync_roots(vcpu);
6796 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6797 kvm_vcpu_flush_tlb(vcpu);
6798 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6799 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6803 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6804 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6805 vcpu->mmio_needed = 0;
6809 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6810 /* Page is swapped out. Do synthetic halt */
6811 vcpu->arch.apf.halted = true;
6815 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6816 record_steal_time(vcpu);
6817 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6819 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6821 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6822 kvm_pmu_handle_event(vcpu);
6823 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6824 kvm_pmu_deliver_pmi(vcpu);
6825 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6826 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6827 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6828 vcpu->arch.ioapic_handled_vectors)) {
6829 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6830 vcpu->run->eoi.vector =
6831 vcpu->arch.pending_ioapic_eoi;
6836 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6837 vcpu_scan_ioapic(vcpu);
6838 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6839 kvm_vcpu_reload_apic_access_page(vcpu);
6840 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6841 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6842 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6846 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6847 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6848 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6852 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6853 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6854 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6860 * KVM_REQ_HV_STIMER has to be processed after
6861 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6862 * depend on the guest clock being up-to-date
6864 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6865 kvm_hv_process_stimers(vcpu);
6868 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6869 ++vcpu->stat.req_event;
6870 kvm_apic_accept_events(vcpu);
6871 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6876 if (inject_pending_event(vcpu, req_int_win) != 0)
6877 req_immediate_exit = true;
6879 /* Enable NMI/IRQ window open exits if needed.
6881 * SMIs have two cases: 1) they can be nested, and
6882 * then there is nothing to do here because RSM will
6883 * cause a vmexit anyway; 2) or the SMI can be pending
6884 * because inject_pending_event has completed the
6885 * injection of an IRQ or NMI from the previous vmexit,
6886 * and then we request an immediate exit to inject the SMI.
6888 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6889 req_immediate_exit = true;
6890 if (vcpu->arch.nmi_pending)
6891 kvm_x86_ops->enable_nmi_window(vcpu);
6892 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6893 kvm_x86_ops->enable_irq_window(vcpu);
6894 WARN_ON(vcpu->arch.exception.pending);
6897 if (kvm_lapic_enabled(vcpu)) {
6898 update_cr8_intercept(vcpu);
6899 kvm_lapic_sync_to_vapic(vcpu);
6903 r = kvm_mmu_reload(vcpu);
6905 goto cancel_injection;
6910 kvm_x86_ops->prepare_guest_switch(vcpu);
6911 kvm_load_guest_fpu(vcpu);
6914 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
6915 * IPI are then delayed after guest entry, which ensures that they
6916 * result in virtual interrupt delivery.
6918 local_irq_disable();
6919 vcpu->mode = IN_GUEST_MODE;
6921 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6924 * 1) We should set ->mode before checking ->requests. Please see
6925 * the comment in kvm_vcpu_exiting_guest_mode().
6927 * 2) For APICv, we should set ->mode before checking PIR.ON. This
6928 * pairs with the memory barrier implicit in pi_test_and_set_on
6929 * (see vmx_deliver_posted_interrupt).
6931 * 3) This also orders the write to mode from any reads to the page
6932 * tables done while the VCPU is running. Please see the comment
6933 * in kvm_flush_remote_tlbs.
6935 smp_mb__after_srcu_read_unlock();
6938 * This handles the case where a posted interrupt was
6939 * notified with kvm_vcpu_kick.
6941 if (kvm_lapic_enabled(vcpu)) {
6942 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6943 kvm_x86_ops->sync_pir_to_irr(vcpu);
6946 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
6947 || need_resched() || signal_pending(current)) {
6948 vcpu->mode = OUTSIDE_GUEST_MODE;
6952 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6954 goto cancel_injection;
6957 kvm_load_guest_xcr0(vcpu);
6959 if (req_immediate_exit) {
6960 kvm_make_request(KVM_REQ_EVENT, vcpu);
6961 smp_send_reschedule(vcpu->cpu);
6964 trace_kvm_entry(vcpu->vcpu_id);
6965 wait_lapic_expire(vcpu);
6966 guest_enter_irqoff();
6968 if (unlikely(vcpu->arch.switch_db_regs)) {
6970 set_debugreg(vcpu->arch.eff_db[0], 0);
6971 set_debugreg(vcpu->arch.eff_db[1], 1);
6972 set_debugreg(vcpu->arch.eff_db[2], 2);
6973 set_debugreg(vcpu->arch.eff_db[3], 3);
6974 set_debugreg(vcpu->arch.dr6, 6);
6975 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6978 kvm_x86_ops->run(vcpu);
6981 * Do this here before restoring debug registers on the host. And
6982 * since we do this before handling the vmexit, a DR access vmexit
6983 * can (a) read the correct value of the debug registers, (b) set
6984 * KVM_DEBUGREG_WONT_EXIT again.
6986 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6987 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6988 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6989 kvm_update_dr0123(vcpu);
6990 kvm_update_dr6(vcpu);
6991 kvm_update_dr7(vcpu);
6992 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6996 * If the guest has used debug registers, at least dr7
6997 * will be disabled while returning to the host.
6998 * If we don't have active breakpoints in the host, we don't
6999 * care about the messed up debug address registers. But if
7000 * we have some of them active, restore the old state.
7002 if (hw_breakpoint_active())
7003 hw_breakpoint_restore();
7005 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7007 vcpu->mode = OUTSIDE_GUEST_MODE;
7010 kvm_put_guest_xcr0(vcpu);
7012 kvm_x86_ops->handle_external_intr(vcpu);
7016 guest_exit_irqoff();
7021 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7024 * Profile KVM exit RIPs:
7026 if (unlikely(prof_on == KVM_PROFILING)) {
7027 unsigned long rip = kvm_rip_read(vcpu);
7028 profile_hit(KVM_PROFILING, (void *)rip);
7031 if (unlikely(vcpu->arch.tsc_always_catchup))
7032 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7034 if (vcpu->arch.apic_attention)
7035 kvm_lapic_sync_from_vapic(vcpu);
7037 vcpu->arch.gpa_available = false;
7038 r = kvm_x86_ops->handle_exit(vcpu);
7042 kvm_x86_ops->cancel_injection(vcpu);
7043 if (unlikely(vcpu->arch.apic_attention))
7044 kvm_lapic_sync_from_vapic(vcpu);
7049 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7051 if (!kvm_arch_vcpu_runnable(vcpu) &&
7052 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7053 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7054 kvm_vcpu_block(vcpu);
7055 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7057 if (kvm_x86_ops->post_block)
7058 kvm_x86_ops->post_block(vcpu);
7060 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7064 kvm_apic_accept_events(vcpu);
7065 switch(vcpu->arch.mp_state) {
7066 case KVM_MP_STATE_HALTED:
7067 vcpu->arch.pv.pv_unhalted = false;
7068 vcpu->arch.mp_state =
7069 KVM_MP_STATE_RUNNABLE;
7070 case KVM_MP_STATE_RUNNABLE:
7071 vcpu->arch.apf.halted = false;
7073 case KVM_MP_STATE_INIT_RECEIVED:
7082 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7084 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7085 kvm_x86_ops->check_nested_events(vcpu, false);
7087 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7088 !vcpu->arch.apf.halted);
7091 static int vcpu_run(struct kvm_vcpu *vcpu)
7094 struct kvm *kvm = vcpu->kvm;
7096 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7099 if (kvm_vcpu_running(vcpu)) {
7100 r = vcpu_enter_guest(vcpu);
7102 r = vcpu_block(kvm, vcpu);
7108 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7109 if (kvm_cpu_has_pending_timer(vcpu))
7110 kvm_inject_pending_timer_irqs(vcpu);
7112 if (dm_request_for_irq_injection(vcpu) &&
7113 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7115 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7116 ++vcpu->stat.request_irq_exits;
7120 kvm_check_async_pf_completion(vcpu);
7122 if (signal_pending(current)) {
7124 vcpu->run->exit_reason = KVM_EXIT_INTR;
7125 ++vcpu->stat.signal_exits;
7128 if (need_resched()) {
7129 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7131 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7135 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7140 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7143 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7144 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7145 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7146 if (r != EMULATE_DONE)
7151 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7153 BUG_ON(!vcpu->arch.pio.count);
7155 return complete_emulated_io(vcpu);
7159 * Implements the following, as a state machine:
7163 * for each mmio piece in the fragment
7171 * for each mmio piece in the fragment
7176 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7178 struct kvm_run *run = vcpu->run;
7179 struct kvm_mmio_fragment *frag;
7182 BUG_ON(!vcpu->mmio_needed);
7184 /* Complete previous fragment */
7185 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7186 len = min(8u, frag->len);
7187 if (!vcpu->mmio_is_write)
7188 memcpy(frag->data, run->mmio.data, len);
7190 if (frag->len <= 8) {
7191 /* Switch to the next fragment. */
7193 vcpu->mmio_cur_fragment++;
7195 /* Go forward to the next mmio piece. */
7201 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7202 vcpu->mmio_needed = 0;
7204 /* FIXME: return into emulator if single-stepping. */
7205 if (vcpu->mmio_is_write)
7207 vcpu->mmio_read_completed = 1;
7208 return complete_emulated_io(vcpu);
7211 run->exit_reason = KVM_EXIT_MMIO;
7212 run->mmio.phys_addr = frag->gpa;
7213 if (vcpu->mmio_is_write)
7214 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7215 run->mmio.len = min(8u, frag->len);
7216 run->mmio.is_write = vcpu->mmio_is_write;
7217 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7222 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7224 struct fpu *fpu = ¤t->thread.fpu;
7228 fpu__initialize(fpu);
7230 if (vcpu->sigset_active)
7231 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7233 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7234 if (kvm_run->immediate_exit) {
7238 kvm_vcpu_block(vcpu);
7239 kvm_apic_accept_events(vcpu);
7240 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7242 if (signal_pending(current)) {
7244 vcpu->run->exit_reason = KVM_EXIT_INTR;
7245 ++vcpu->stat.signal_exits;
7250 /* re-sync apic's tpr */
7251 if (!lapic_in_kernel(vcpu)) {
7252 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7258 if (unlikely(vcpu->arch.complete_userspace_io)) {
7259 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7260 vcpu->arch.complete_userspace_io = NULL;
7265 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7267 if (kvm_run->immediate_exit)
7273 post_kvm_run_save(vcpu);
7274 if (vcpu->sigset_active)
7275 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7280 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7282 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7284 * We are here if userspace calls get_regs() in the middle of
7285 * instruction emulation. Registers state needs to be copied
7286 * back from emulation context to vcpu. Userspace shouldn't do
7287 * that usually, but some bad designed PV devices (vmware
7288 * backdoor interface) need this to work
7290 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7291 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7293 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7294 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7295 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7296 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7297 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7298 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7299 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7300 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7301 #ifdef CONFIG_X86_64
7302 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7303 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7304 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7305 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7306 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7307 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7308 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7309 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7312 regs->rip = kvm_rip_read(vcpu);
7313 regs->rflags = kvm_get_rflags(vcpu);
7318 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7320 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7321 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7323 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7324 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7325 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7326 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7327 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7328 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7329 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7330 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7331 #ifdef CONFIG_X86_64
7332 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7333 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7334 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7335 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7336 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7337 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7338 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7339 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7342 kvm_rip_write(vcpu, regs->rip);
7343 kvm_set_rflags(vcpu, regs->rflags);
7345 vcpu->arch.exception.pending = false;
7347 kvm_make_request(KVM_REQ_EVENT, vcpu);
7352 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7354 struct kvm_segment cs;
7356 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7360 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7362 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7363 struct kvm_sregs *sregs)
7367 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7368 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7369 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7370 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7371 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7372 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7374 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7375 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7377 kvm_x86_ops->get_idt(vcpu, &dt);
7378 sregs->idt.limit = dt.size;
7379 sregs->idt.base = dt.address;
7380 kvm_x86_ops->get_gdt(vcpu, &dt);
7381 sregs->gdt.limit = dt.size;
7382 sregs->gdt.base = dt.address;
7384 sregs->cr0 = kvm_read_cr0(vcpu);
7385 sregs->cr2 = vcpu->arch.cr2;
7386 sregs->cr3 = kvm_read_cr3(vcpu);
7387 sregs->cr4 = kvm_read_cr4(vcpu);
7388 sregs->cr8 = kvm_get_cr8(vcpu);
7389 sregs->efer = vcpu->arch.efer;
7390 sregs->apic_base = kvm_get_apic_base(vcpu);
7392 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7394 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7395 set_bit(vcpu->arch.interrupt.nr,
7396 (unsigned long *)sregs->interrupt_bitmap);
7401 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7402 struct kvm_mp_state *mp_state)
7404 kvm_apic_accept_events(vcpu);
7405 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7406 vcpu->arch.pv.pv_unhalted)
7407 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7409 mp_state->mp_state = vcpu->arch.mp_state;
7414 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7415 struct kvm_mp_state *mp_state)
7417 if (!lapic_in_kernel(vcpu) &&
7418 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7421 /* INITs are latched while in SMM */
7422 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7423 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7424 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7427 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7428 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7429 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7431 vcpu->arch.mp_state = mp_state->mp_state;
7432 kvm_make_request(KVM_REQ_EVENT, vcpu);
7436 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7437 int reason, bool has_error_code, u32 error_code)
7439 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7442 init_emulate_ctxt(vcpu);
7444 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7445 has_error_code, error_code);
7448 return EMULATE_FAIL;
7450 kvm_rip_write(vcpu, ctxt->eip);
7451 kvm_set_rflags(vcpu, ctxt->eflags);
7452 kvm_make_request(KVM_REQ_EVENT, vcpu);
7453 return EMULATE_DONE;
7455 EXPORT_SYMBOL_GPL(kvm_task_switch);
7457 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7458 struct kvm_sregs *sregs)
7460 struct msr_data apic_base_msr;
7461 int mmu_reset_needed = 0;
7462 int pending_vec, max_bits, idx;
7465 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7466 (sregs->cr4 & X86_CR4_OSXSAVE))
7469 apic_base_msr.data = sregs->apic_base;
7470 apic_base_msr.host_initiated = true;
7471 if (kvm_set_apic_base(vcpu, &apic_base_msr))
7474 dt.size = sregs->idt.limit;
7475 dt.address = sregs->idt.base;
7476 kvm_x86_ops->set_idt(vcpu, &dt);
7477 dt.size = sregs->gdt.limit;
7478 dt.address = sregs->gdt.base;
7479 kvm_x86_ops->set_gdt(vcpu, &dt);
7481 vcpu->arch.cr2 = sregs->cr2;
7482 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7483 vcpu->arch.cr3 = sregs->cr3;
7484 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7486 kvm_set_cr8(vcpu, sregs->cr8);
7488 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7489 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7491 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7492 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7493 vcpu->arch.cr0 = sregs->cr0;
7495 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7496 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7497 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7498 kvm_update_cpuid(vcpu);
7500 idx = srcu_read_lock(&vcpu->kvm->srcu);
7501 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7502 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7503 mmu_reset_needed = 1;
7505 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7507 if (mmu_reset_needed)
7508 kvm_mmu_reset_context(vcpu);
7510 max_bits = KVM_NR_INTERRUPTS;
7511 pending_vec = find_first_bit(
7512 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7513 if (pending_vec < max_bits) {
7514 kvm_queue_interrupt(vcpu, pending_vec, false);
7515 pr_debug("Set back pending irq %d\n", pending_vec);
7518 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7519 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7520 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7521 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7522 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7523 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7525 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7526 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7528 update_cr8_intercept(vcpu);
7530 /* Older userspace won't unhalt the vcpu on reset. */
7531 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7532 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7534 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7536 kvm_make_request(KVM_REQ_EVENT, vcpu);
7541 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7542 struct kvm_guest_debug *dbg)
7544 unsigned long rflags;
7547 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7549 if (vcpu->arch.exception.pending)
7551 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7552 kvm_queue_exception(vcpu, DB_VECTOR);
7554 kvm_queue_exception(vcpu, BP_VECTOR);
7558 * Read rflags as long as potentially injected trace flags are still
7561 rflags = kvm_get_rflags(vcpu);
7563 vcpu->guest_debug = dbg->control;
7564 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7565 vcpu->guest_debug = 0;
7567 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7568 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7569 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7570 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7572 for (i = 0; i < KVM_NR_DB_REGS; i++)
7573 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7575 kvm_update_dr7(vcpu);
7577 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7578 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7579 get_segment_base(vcpu, VCPU_SREG_CS);
7582 * Trigger an rflags update that will inject or remove the trace
7585 kvm_set_rflags(vcpu, rflags);
7587 kvm_x86_ops->update_bp_intercept(vcpu);
7597 * Translate a guest virtual address to a guest physical address.
7599 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7600 struct kvm_translation *tr)
7602 unsigned long vaddr = tr->linear_address;
7606 idx = srcu_read_lock(&vcpu->kvm->srcu);
7607 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7608 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7609 tr->physical_address = gpa;
7610 tr->valid = gpa != UNMAPPED_GVA;
7617 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7619 struct fxregs_state *fxsave =
7620 &vcpu->arch.guest_fpu.state.fxsave;
7622 memcpy(fpu->fpr, fxsave->st_space, 128);
7623 fpu->fcw = fxsave->cwd;
7624 fpu->fsw = fxsave->swd;
7625 fpu->ftwx = fxsave->twd;
7626 fpu->last_opcode = fxsave->fop;
7627 fpu->last_ip = fxsave->rip;
7628 fpu->last_dp = fxsave->rdp;
7629 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7634 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7636 struct fxregs_state *fxsave =
7637 &vcpu->arch.guest_fpu.state.fxsave;
7639 memcpy(fxsave->st_space, fpu->fpr, 128);
7640 fxsave->cwd = fpu->fcw;
7641 fxsave->swd = fpu->fsw;
7642 fxsave->twd = fpu->ftwx;
7643 fxsave->fop = fpu->last_opcode;
7644 fxsave->rip = fpu->last_ip;
7645 fxsave->rdp = fpu->last_dp;
7646 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7651 static void fx_init(struct kvm_vcpu *vcpu)
7653 fpstate_init(&vcpu->arch.guest_fpu.state);
7654 if (boot_cpu_has(X86_FEATURE_XSAVES))
7655 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7656 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7659 * Ensure guest xcr0 is valid for loading
7661 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7663 vcpu->arch.cr0 |= X86_CR0_ET;
7666 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7668 if (vcpu->guest_fpu_loaded)
7672 * Restore all possible states in the guest,
7673 * and assume host would use all available bits.
7674 * Guest xcr0 would be loaded later.
7676 vcpu->guest_fpu_loaded = 1;
7677 __kernel_fpu_begin();
7678 /* PKRU is separately restored in kvm_x86_ops->run. */
7679 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
7680 ~XFEATURE_MASK_PKRU);
7684 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7686 if (!vcpu->guest_fpu_loaded)
7689 vcpu->guest_fpu_loaded = 0;
7690 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7692 ++vcpu->stat.fpu_reload;
7696 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7698 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7700 kvmclock_reset(vcpu);
7702 kvm_x86_ops->vcpu_free(vcpu);
7703 free_cpumask_var(wbinvd_dirty_mask);
7706 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7709 struct kvm_vcpu *vcpu;
7711 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7712 printk_once(KERN_WARNING
7713 "kvm: SMP vm created on host with unstable TSC; "
7714 "guest TSC will not be reliable\n");
7716 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7721 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7725 kvm_vcpu_mtrr_init(vcpu);
7726 r = vcpu_load(vcpu);
7729 kvm_vcpu_reset(vcpu, false);
7730 kvm_mmu_setup(vcpu);
7735 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7737 struct msr_data msr;
7738 struct kvm *kvm = vcpu->kvm;
7740 kvm_hv_vcpu_postcreate(vcpu);
7742 if (vcpu_load(vcpu))
7745 msr.index = MSR_IA32_TSC;
7746 msr.host_initiated = true;
7747 kvm_write_tsc(vcpu, &msr);
7750 if (!kvmclock_periodic_sync)
7753 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7754 KVMCLOCK_SYNC_PERIOD);
7757 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7760 vcpu->arch.apf.msr_val = 0;
7762 r = vcpu_load(vcpu);
7764 kvm_mmu_unload(vcpu);
7767 kvm_x86_ops->vcpu_free(vcpu);
7770 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7772 vcpu->arch.hflags = 0;
7774 vcpu->arch.smi_pending = 0;
7775 atomic_set(&vcpu->arch.nmi_queued, 0);
7776 vcpu->arch.nmi_pending = 0;
7777 vcpu->arch.nmi_injected = false;
7778 kvm_clear_interrupt_queue(vcpu);
7779 kvm_clear_exception_queue(vcpu);
7780 vcpu->arch.exception.pending = false;
7782 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7783 kvm_update_dr0123(vcpu);
7784 vcpu->arch.dr6 = DR6_INIT;
7785 kvm_update_dr6(vcpu);
7786 vcpu->arch.dr7 = DR7_FIXED_1;
7787 kvm_update_dr7(vcpu);
7791 kvm_make_request(KVM_REQ_EVENT, vcpu);
7792 vcpu->arch.apf.msr_val = 0;
7793 vcpu->arch.st.msr_val = 0;
7795 kvmclock_reset(vcpu);
7797 kvm_clear_async_pf_completion_queue(vcpu);
7798 kvm_async_pf_hash_reset(vcpu);
7799 vcpu->arch.apf.halted = false;
7802 kvm_pmu_reset(vcpu);
7803 vcpu->arch.smbase = 0x30000;
7805 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
7806 vcpu->arch.msr_misc_features_enables = 0;
7809 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7810 vcpu->arch.regs_avail = ~0;
7811 vcpu->arch.regs_dirty = ~0;
7813 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7816 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7818 struct kvm_segment cs;
7820 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7821 cs.selector = vector << 8;
7822 cs.base = vector << 12;
7823 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7824 kvm_rip_write(vcpu, 0);
7827 int kvm_arch_hardware_enable(void)
7830 struct kvm_vcpu *vcpu;
7835 bool stable, backwards_tsc = false;
7837 kvm_shared_msr_cpu_online();
7838 ret = kvm_x86_ops->hardware_enable();
7842 local_tsc = rdtsc();
7843 stable = !check_tsc_unstable();
7844 list_for_each_entry(kvm, &vm_list, vm_list) {
7845 kvm_for_each_vcpu(i, vcpu, kvm) {
7846 if (!stable && vcpu->cpu == smp_processor_id())
7847 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7848 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7849 backwards_tsc = true;
7850 if (vcpu->arch.last_host_tsc > max_tsc)
7851 max_tsc = vcpu->arch.last_host_tsc;
7857 * Sometimes, even reliable TSCs go backwards. This happens on
7858 * platforms that reset TSC during suspend or hibernate actions, but
7859 * maintain synchronization. We must compensate. Fortunately, we can
7860 * detect that condition here, which happens early in CPU bringup,
7861 * before any KVM threads can be running. Unfortunately, we can't
7862 * bring the TSCs fully up to date with real time, as we aren't yet far
7863 * enough into CPU bringup that we know how much real time has actually
7864 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
7865 * variables that haven't been updated yet.
7867 * So we simply find the maximum observed TSC above, then record the
7868 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7869 * the adjustment will be applied. Note that we accumulate
7870 * adjustments, in case multiple suspend cycles happen before some VCPU
7871 * gets a chance to run again. In the event that no KVM threads get a
7872 * chance to run, we will miss the entire elapsed period, as we'll have
7873 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7874 * loose cycle time. This isn't too big a deal, since the loss will be
7875 * uniform across all VCPUs (not to mention the scenario is extremely
7876 * unlikely). It is possible that a second hibernate recovery happens
7877 * much faster than a first, causing the observed TSC here to be
7878 * smaller; this would require additional padding adjustment, which is
7879 * why we set last_host_tsc to the local tsc observed here.
7881 * N.B. - this code below runs only on platforms with reliable TSC,
7882 * as that is the only way backwards_tsc is set above. Also note
7883 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7884 * have the same delta_cyc adjustment applied if backwards_tsc
7885 * is detected. Note further, this adjustment is only done once,
7886 * as we reset last_host_tsc on all VCPUs to stop this from being
7887 * called multiple times (one for each physical CPU bringup).
7889 * Platforms with unreliable TSCs don't have to deal with this, they
7890 * will be compensated by the logic in vcpu_load, which sets the TSC to
7891 * catchup mode. This will catchup all VCPUs to real time, but cannot
7892 * guarantee that they stay in perfect synchronization.
7894 if (backwards_tsc) {
7895 u64 delta_cyc = max_tsc - local_tsc;
7896 list_for_each_entry(kvm, &vm_list, vm_list) {
7897 kvm->arch.backwards_tsc_observed = true;
7898 kvm_for_each_vcpu(i, vcpu, kvm) {
7899 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7900 vcpu->arch.last_host_tsc = local_tsc;
7901 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7905 * We have to disable TSC offset matching.. if you were
7906 * booting a VM while issuing an S4 host suspend....
7907 * you may have some problem. Solving this issue is
7908 * left as an exercise to the reader.
7910 kvm->arch.last_tsc_nsec = 0;
7911 kvm->arch.last_tsc_write = 0;
7918 void kvm_arch_hardware_disable(void)
7920 kvm_x86_ops->hardware_disable();
7921 drop_user_return_notifiers();
7924 int kvm_arch_hardware_setup(void)
7928 r = kvm_x86_ops->hardware_setup();
7932 if (kvm_has_tsc_control) {
7934 * Make sure the user can only configure tsc_khz values that
7935 * fit into a signed integer.
7936 * A min value is not calculated needed because it will always
7937 * be 1 on all machines.
7939 u64 max = min(0x7fffffffULL,
7940 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7941 kvm_max_guest_tsc_khz = max;
7943 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7946 kvm_init_msr_list();
7950 void kvm_arch_hardware_unsetup(void)
7952 kvm_x86_ops->hardware_unsetup();
7955 void kvm_arch_check_processor_compat(void *rtn)
7957 kvm_x86_ops->check_processor_compatibility(rtn);
7960 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7962 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7964 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7966 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7968 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7971 struct static_key kvm_no_apic_vcpu __read_mostly;
7972 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7974 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7980 BUG_ON(vcpu->kvm == NULL);
7983 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
7984 vcpu->arch.pv.pv_unhalted = false;
7985 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7986 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7987 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7989 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7991 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7996 vcpu->arch.pio_data = page_address(page);
7998 kvm_set_tsc_khz(vcpu, max_tsc_khz);
8000 r = kvm_mmu_create(vcpu);
8002 goto fail_free_pio_data;
8004 if (irqchip_in_kernel(kvm)) {
8005 r = kvm_create_lapic(vcpu);
8007 goto fail_mmu_destroy;
8009 static_key_slow_inc(&kvm_no_apic_vcpu);
8011 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8013 if (!vcpu->arch.mce_banks) {
8015 goto fail_free_lapic;
8017 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8019 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8021 goto fail_free_mce_banks;
8026 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
8027 vcpu->arch.pv_time_enabled = false;
8029 vcpu->arch.guest_supported_xcr0 = 0;
8030 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8032 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8034 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8036 kvm_async_pf_hash_reset(vcpu);
8039 vcpu->arch.pending_external_vector = -1;
8040 vcpu->arch.preempted_in_kernel = false;
8042 kvm_hv_vcpu_init(vcpu);
8046 fail_free_mce_banks:
8047 kfree(vcpu->arch.mce_banks);
8049 kvm_free_lapic(vcpu);
8051 kvm_mmu_destroy(vcpu);
8053 free_page((unsigned long)vcpu->arch.pio_data);
8058 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8062 kvm_hv_vcpu_uninit(vcpu);
8063 kvm_pmu_destroy(vcpu);
8064 kfree(vcpu->arch.mce_banks);
8065 kvm_free_lapic(vcpu);
8066 idx = srcu_read_lock(&vcpu->kvm->srcu);
8067 kvm_mmu_destroy(vcpu);
8068 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8069 free_page((unsigned long)vcpu->arch.pio_data);
8070 if (!lapic_in_kernel(vcpu))
8071 static_key_slow_dec(&kvm_no_apic_vcpu);
8074 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8076 kvm_x86_ops->sched_in(vcpu, cpu);
8079 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8084 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8085 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8086 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8087 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8088 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8090 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8091 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8092 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8093 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8094 &kvm->arch.irq_sources_bitmap);
8096 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8097 mutex_init(&kvm->arch.apic_map_lock);
8098 mutex_init(&kvm->arch.hyperv.hv_lock);
8099 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8101 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8102 pvclock_update_vm_gtod_copy(kvm);
8104 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8105 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8107 kvm_page_track_init(kvm);
8108 kvm_mmu_init_vm(kvm);
8110 if (kvm_x86_ops->vm_init)
8111 return kvm_x86_ops->vm_init(kvm);
8116 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8119 r = vcpu_load(vcpu);
8121 kvm_mmu_unload(vcpu);
8125 static void kvm_free_vcpus(struct kvm *kvm)
8128 struct kvm_vcpu *vcpu;
8131 * Unpin any mmu pages first.
8133 kvm_for_each_vcpu(i, vcpu, kvm) {
8134 kvm_clear_async_pf_completion_queue(vcpu);
8135 kvm_unload_vcpu_mmu(vcpu);
8137 kvm_for_each_vcpu(i, vcpu, kvm)
8138 kvm_arch_vcpu_free(vcpu);
8140 mutex_lock(&kvm->lock);
8141 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8142 kvm->vcpus[i] = NULL;
8144 atomic_set(&kvm->online_vcpus, 0);
8145 mutex_unlock(&kvm->lock);
8148 void kvm_arch_sync_events(struct kvm *kvm)
8150 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8151 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8155 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8159 struct kvm_memslots *slots = kvm_memslots(kvm);
8160 struct kvm_memory_slot *slot, old;
8162 /* Called with kvm->slots_lock held. */
8163 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8166 slot = id_to_memslot(slots, id);
8172 * MAP_SHARED to prevent internal slot pages from being moved
8175 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8176 MAP_SHARED | MAP_ANONYMOUS, 0);
8177 if (IS_ERR((void *)hva))
8178 return PTR_ERR((void *)hva);
8187 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8188 struct kvm_userspace_memory_region m;
8190 m.slot = id | (i << 16);
8192 m.guest_phys_addr = gpa;
8193 m.userspace_addr = hva;
8194 m.memory_size = size;
8195 r = __kvm_set_memory_region(kvm, &m);
8201 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8207 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8209 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8213 mutex_lock(&kvm->slots_lock);
8214 r = __x86_set_memory_region(kvm, id, gpa, size);
8215 mutex_unlock(&kvm->slots_lock);
8219 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8221 void kvm_arch_destroy_vm(struct kvm *kvm)
8223 if (current->mm == kvm->mm) {
8225 * Free memory regions allocated on behalf of userspace,
8226 * unless the the memory map has changed due to process exit
8229 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8230 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8231 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8233 if (kvm_x86_ops->vm_destroy)
8234 kvm_x86_ops->vm_destroy(kvm);
8235 kvm_pic_destroy(kvm);
8236 kvm_ioapic_destroy(kvm);
8237 kvm_free_vcpus(kvm);
8238 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8239 kvm_mmu_uninit_vm(kvm);
8240 kvm_page_track_cleanup(kvm);
8243 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8244 struct kvm_memory_slot *dont)
8248 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8249 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8250 kvfree(free->arch.rmap[i]);
8251 free->arch.rmap[i] = NULL;
8256 if (!dont || free->arch.lpage_info[i - 1] !=
8257 dont->arch.lpage_info[i - 1]) {
8258 kvfree(free->arch.lpage_info[i - 1]);
8259 free->arch.lpage_info[i - 1] = NULL;
8263 kvm_page_track_free_memslot(free, dont);
8266 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8267 unsigned long npages)
8271 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8272 struct kvm_lpage_info *linfo;
8277 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8278 slot->base_gfn, level) + 1;
8280 slot->arch.rmap[i] =
8281 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8282 if (!slot->arch.rmap[i])
8287 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8291 slot->arch.lpage_info[i - 1] = linfo;
8293 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8294 linfo[0].disallow_lpage = 1;
8295 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8296 linfo[lpages - 1].disallow_lpage = 1;
8297 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8299 * If the gfn and userspace address are not aligned wrt each
8300 * other, or if explicitly asked to, disable large page
8301 * support for this slot
8303 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8304 !kvm_largepages_enabled()) {
8307 for (j = 0; j < lpages; ++j)
8308 linfo[j].disallow_lpage = 1;
8312 if (kvm_page_track_create_memslot(slot, npages))
8318 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8319 kvfree(slot->arch.rmap[i]);
8320 slot->arch.rmap[i] = NULL;
8324 kvfree(slot->arch.lpage_info[i - 1]);
8325 slot->arch.lpage_info[i - 1] = NULL;
8330 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8333 * memslots->generation has been incremented.
8334 * mmio generation may have reached its maximum value.
8336 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8339 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8340 struct kvm_memory_slot *memslot,
8341 const struct kvm_userspace_memory_region *mem,
8342 enum kvm_mr_change change)
8347 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8348 struct kvm_memory_slot *new)
8350 /* Still write protect RO slot */
8351 if (new->flags & KVM_MEM_READONLY) {
8352 kvm_mmu_slot_remove_write_access(kvm, new);
8357 * Call kvm_x86_ops dirty logging hooks when they are valid.
8359 * kvm_x86_ops->slot_disable_log_dirty is called when:
8361 * - KVM_MR_CREATE with dirty logging is disabled
8362 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8364 * The reason is, in case of PML, we need to set D-bit for any slots
8365 * with dirty logging disabled in order to eliminate unnecessary GPA
8366 * logging in PML buffer (and potential PML buffer full VMEXT). This
8367 * guarantees leaving PML enabled during guest's lifetime won't have
8368 * any additonal overhead from PML when guest is running with dirty
8369 * logging disabled for memory slots.
8371 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8372 * to dirty logging mode.
8374 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8376 * In case of write protect:
8378 * Write protect all pages for dirty logging.
8380 * All the sptes including the large sptes which point to this
8381 * slot are set to readonly. We can not create any new large
8382 * spte on this slot until the end of the logging.
8384 * See the comments in fast_page_fault().
8386 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8387 if (kvm_x86_ops->slot_enable_log_dirty)
8388 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8390 kvm_mmu_slot_remove_write_access(kvm, new);
8392 if (kvm_x86_ops->slot_disable_log_dirty)
8393 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8397 void kvm_arch_commit_memory_region(struct kvm *kvm,
8398 const struct kvm_userspace_memory_region *mem,
8399 const struct kvm_memory_slot *old,
8400 const struct kvm_memory_slot *new,
8401 enum kvm_mr_change change)
8403 int nr_mmu_pages = 0;
8405 if (!kvm->arch.n_requested_mmu_pages)
8406 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8409 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8412 * Dirty logging tracks sptes in 4k granularity, meaning that large
8413 * sptes have to be split. If live migration is successful, the guest
8414 * in the source machine will be destroyed and large sptes will be
8415 * created in the destination. However, if the guest continues to run
8416 * in the source machine (for example if live migration fails), small
8417 * sptes will remain around and cause bad performance.
8419 * Scan sptes if dirty logging has been stopped, dropping those
8420 * which can be collapsed into a single large-page spte. Later
8421 * page faults will create the large-page sptes.
8423 if ((change != KVM_MR_DELETE) &&
8424 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8425 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8426 kvm_mmu_zap_collapsible_sptes(kvm, new);
8429 * Set up write protection and/or dirty logging for the new slot.
8431 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8432 * been zapped so no dirty logging staff is needed for old slot. For
8433 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8434 * new and it's also covered when dealing with the new slot.
8436 * FIXME: const-ify all uses of struct kvm_memory_slot.
8438 if (change != KVM_MR_DELETE)
8439 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8442 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8444 kvm_mmu_invalidate_zap_all_pages(kvm);
8447 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8448 struct kvm_memory_slot *slot)
8450 kvm_page_track_flush_slot(kvm, slot);
8453 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8455 if (!list_empty_careful(&vcpu->async_pf.done))
8458 if (kvm_apic_has_events(vcpu))
8461 if (vcpu->arch.pv.pv_unhalted)
8464 if (vcpu->arch.exception.pending)
8467 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
8468 (vcpu->arch.nmi_pending &&
8469 kvm_x86_ops->nmi_allowed(vcpu)))
8472 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
8473 (vcpu->arch.smi_pending && !is_smm(vcpu)))
8476 if (kvm_arch_interrupt_allowed(vcpu) &&
8477 kvm_cpu_has_interrupt(vcpu))
8480 if (kvm_hv_has_stimer_pending(vcpu))
8486 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8488 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8491 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
8493 return vcpu->arch.preempted_in_kernel;
8496 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8498 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8501 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8503 return kvm_x86_ops->interrupt_allowed(vcpu);
8506 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8508 if (is_64_bit_mode(vcpu))
8509 return kvm_rip_read(vcpu);
8510 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8511 kvm_rip_read(vcpu));
8513 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8515 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8517 return kvm_get_linear_rip(vcpu) == linear_rip;
8519 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8521 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8523 unsigned long rflags;
8525 rflags = kvm_x86_ops->get_rflags(vcpu);
8526 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8527 rflags &= ~X86_EFLAGS_TF;
8530 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8532 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8534 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8535 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8536 rflags |= X86_EFLAGS_TF;
8537 kvm_x86_ops->set_rflags(vcpu, rflags);
8540 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8542 __kvm_set_rflags(vcpu, rflags);
8543 kvm_make_request(KVM_REQ_EVENT, vcpu);
8545 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8547 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8551 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8555 r = kvm_mmu_reload(vcpu);
8559 if (!vcpu->arch.mmu.direct_map &&
8560 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8563 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8566 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8568 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8571 static inline u32 kvm_async_pf_next_probe(u32 key)
8573 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8576 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8578 u32 key = kvm_async_pf_hash_fn(gfn);
8580 while (vcpu->arch.apf.gfns[key] != ~0)
8581 key = kvm_async_pf_next_probe(key);
8583 vcpu->arch.apf.gfns[key] = gfn;
8586 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8589 u32 key = kvm_async_pf_hash_fn(gfn);
8591 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8592 (vcpu->arch.apf.gfns[key] != gfn &&
8593 vcpu->arch.apf.gfns[key] != ~0); i++)
8594 key = kvm_async_pf_next_probe(key);
8599 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8601 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8604 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8608 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8610 vcpu->arch.apf.gfns[i] = ~0;
8612 j = kvm_async_pf_next_probe(j);
8613 if (vcpu->arch.apf.gfns[j] == ~0)
8615 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8617 * k lies cyclically in ]i,j]
8619 * |....j i.k.| or |.k..j i...|
8621 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8622 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8627 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8630 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8634 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
8637 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
8641 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8642 struct kvm_async_pf *work)
8644 struct x86_exception fault;
8646 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8647 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8649 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8650 (vcpu->arch.apf.send_user_only &&
8651 kvm_x86_ops->get_cpl(vcpu) == 0))
8652 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8653 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8654 fault.vector = PF_VECTOR;
8655 fault.error_code_valid = true;
8656 fault.error_code = 0;
8657 fault.nested_page_fault = false;
8658 fault.address = work->arch.token;
8659 fault.async_page_fault = true;
8660 kvm_inject_page_fault(vcpu, &fault);
8664 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8665 struct kvm_async_pf *work)
8667 struct x86_exception fault;
8670 if (work->wakeup_all)
8671 work->arch.token = ~0; /* broadcast wakeup */
8673 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8674 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8676 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
8677 !apf_get_user(vcpu, &val)) {
8678 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
8679 vcpu->arch.exception.pending &&
8680 vcpu->arch.exception.nr == PF_VECTOR &&
8681 !apf_put_user(vcpu, 0)) {
8682 vcpu->arch.exception.injected = false;
8683 vcpu->arch.exception.pending = false;
8684 vcpu->arch.exception.nr = 0;
8685 vcpu->arch.exception.has_error_code = false;
8686 vcpu->arch.exception.error_code = 0;
8687 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8688 fault.vector = PF_VECTOR;
8689 fault.error_code_valid = true;
8690 fault.error_code = 0;
8691 fault.nested_page_fault = false;
8692 fault.address = work->arch.token;
8693 fault.async_page_fault = true;
8694 kvm_inject_page_fault(vcpu, &fault);
8697 vcpu->arch.apf.halted = false;
8698 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8701 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8703 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8706 return kvm_can_do_async_pf(vcpu);
8709 void kvm_arch_start_assignment(struct kvm *kvm)
8711 atomic_inc(&kvm->arch.assigned_device_count);
8713 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8715 void kvm_arch_end_assignment(struct kvm *kvm)
8717 atomic_dec(&kvm->arch.assigned_device_count);
8719 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8721 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8723 return atomic_read(&kvm->arch.assigned_device_count);
8725 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8727 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8729 atomic_inc(&kvm->arch.noncoherent_dma_count);
8731 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8733 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8735 atomic_dec(&kvm->arch.noncoherent_dma_count);
8737 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8739 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8741 return atomic_read(&kvm->arch.noncoherent_dma_count);
8743 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8745 bool kvm_arch_has_irq_bypass(void)
8747 return kvm_x86_ops->update_pi_irte != NULL;
8750 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8751 struct irq_bypass_producer *prod)
8753 struct kvm_kernel_irqfd *irqfd =
8754 container_of(cons, struct kvm_kernel_irqfd, consumer);
8756 irqfd->producer = prod;
8758 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8759 prod->irq, irqfd->gsi, 1);
8762 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8763 struct irq_bypass_producer *prod)
8766 struct kvm_kernel_irqfd *irqfd =
8767 container_of(cons, struct kvm_kernel_irqfd, consumer);
8769 WARN_ON(irqfd->producer != prod);
8770 irqfd->producer = NULL;
8773 * When producer of consumer is unregistered, we change back to
8774 * remapped mode, so we can re-use the current implementation
8775 * when the irq is masked/disabled or the consumer side (KVM
8776 * int this case doesn't want to receive the interrupts.
8778 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8780 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8781 " fails: %d\n", irqfd->consumer.token, ret);
8784 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8785 uint32_t guest_irq, bool set)
8787 if (!kvm_x86_ops->update_pi_irte)
8790 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8793 bool kvm_vector_hashing_enabled(void)
8795 return vector_hashing;
8797 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8799 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8800 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8801 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8802 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8803 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8804 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8805 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8806 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8807 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8808 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8809 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8810 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8811 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8812 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8813 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8814 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8815 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8816 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8817 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);