1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
24 #include "kvm_cache_regs.h"
25 #include "kvm_emulate.h"
32 #include <linux/clocksource.h>
33 #include <linux/interrupt.h>
34 #include <linux/kvm.h>
36 #include <linux/vmalloc.h>
37 #include <linux/export.h>
38 #include <linux/moduleparam.h>
39 #include <linux/mman.h>
40 #include <linux/highmem.h>
41 #include <linux/iommu.h>
42 #include <linux/intel-iommu.h>
43 #include <linux/cpufreq.h>
44 #include <linux/user-return-notifier.h>
45 #include <linux/srcu.h>
46 #include <linux/slab.h>
47 #include <linux/perf_event.h>
48 #include <linux/uaccess.h>
49 #include <linux/hash.h>
50 #include <linux/pci.h>
51 #include <linux/timekeeper_internal.h>
52 #include <linux/pvclock_gtod.h>
53 #include <linux/kvm_irqfd.h>
54 #include <linux/irqbypass.h>
55 #include <linux/sched/stat.h>
56 #include <linux/sched/isolation.h>
57 #include <linux/mem_encrypt.h>
59 #include <trace/events/kvm.h>
61 #include <asm/debugreg.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
72 #include <asm/intel_pt.h>
73 #include <asm/emulate_prefix.h>
74 #include <clocksource/hyperv_timer.h>
76 #define CREATE_TRACE_POINTS
79 #define MAX_IO_MSRS 256
80 #define KVM_MAX_MCE_BANKS 32
81 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
82 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
84 #define emul_to_vcpu(ctxt) \
85 ((struct kvm_vcpu *)(ctxt)->vcpu)
88 * - enable syscall per default because its emulated by KVM
89 * - enable LME and LMA per default on 64 bit KVM
93 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
95 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
98 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
100 #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__
101 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
103 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
104 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
106 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
107 static void process_nmi(struct kvm_vcpu *vcpu);
108 static void enter_smm(struct kvm_vcpu *vcpu);
109 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
110 static void store_regs(struct kvm_vcpu *vcpu);
111 static int sync_regs(struct kvm_vcpu *vcpu);
113 struct kvm_x86_ops kvm_x86_ops __read_mostly;
114 EXPORT_SYMBOL_GPL(kvm_x86_ops);
116 static bool __read_mostly ignore_msrs = 0;
117 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
119 static bool __read_mostly report_ignored_msrs = true;
120 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
122 unsigned int min_timer_period_us = 200;
123 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
125 static bool __read_mostly kvmclock_periodic_sync = true;
126 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
128 bool __read_mostly kvm_has_tsc_control;
129 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
130 u32 __read_mostly kvm_max_guest_tsc_khz;
131 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
132 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
133 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
134 u64 __read_mostly kvm_max_tsc_scaling_ratio;
135 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
136 u64 __read_mostly kvm_default_tsc_scaling_ratio;
137 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
139 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
140 static u32 __read_mostly tsc_tolerance_ppm = 250;
141 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
144 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
145 * adaptive tuning starting from default advancment of 1000ns. '0' disables
146 * advancement entirely. Any other value is used as-is and disables adaptive
147 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
149 static int __read_mostly lapic_timer_advance_ns = -1;
150 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
152 static bool __read_mostly vector_hashing = true;
153 module_param(vector_hashing, bool, S_IRUGO);
155 bool __read_mostly enable_vmware_backdoor = false;
156 module_param(enable_vmware_backdoor, bool, S_IRUGO);
157 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
159 static bool __read_mostly force_emulation_prefix = false;
160 module_param(force_emulation_prefix, bool, S_IRUGO);
162 int __read_mostly pi_inject_timer = -1;
163 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
165 #define KVM_NR_SHARED_MSRS 16
167 struct kvm_shared_msrs_global {
169 u32 msrs[KVM_NR_SHARED_MSRS];
172 struct kvm_shared_msrs {
173 struct user_return_notifier urn;
175 struct kvm_shared_msr_values {
178 } values[KVM_NR_SHARED_MSRS];
181 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
182 static struct kvm_shared_msrs __percpu *shared_msrs;
184 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
185 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
186 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
187 | XFEATURE_MASK_PKRU)
189 u64 __read_mostly host_efer;
190 EXPORT_SYMBOL_GPL(host_efer);
192 static u64 __read_mostly host_xss;
193 u64 __read_mostly supported_xss;
194 EXPORT_SYMBOL_GPL(supported_xss);
196 struct kvm_stats_debugfs_item debugfs_entries[] = {
197 { "pf_fixed", VCPU_STAT(pf_fixed) },
198 { "pf_guest", VCPU_STAT(pf_guest) },
199 { "tlb_flush", VCPU_STAT(tlb_flush) },
200 { "invlpg", VCPU_STAT(invlpg) },
201 { "exits", VCPU_STAT(exits) },
202 { "io_exits", VCPU_STAT(io_exits) },
203 { "mmio_exits", VCPU_STAT(mmio_exits) },
204 { "signal_exits", VCPU_STAT(signal_exits) },
205 { "irq_window", VCPU_STAT(irq_window_exits) },
206 { "nmi_window", VCPU_STAT(nmi_window_exits) },
207 { "halt_exits", VCPU_STAT(halt_exits) },
208 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
209 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
210 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
211 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
212 { "hypercalls", VCPU_STAT(hypercalls) },
213 { "request_irq", VCPU_STAT(request_irq_exits) },
214 { "irq_exits", VCPU_STAT(irq_exits) },
215 { "host_state_reload", VCPU_STAT(host_state_reload) },
216 { "fpu_reload", VCPU_STAT(fpu_reload) },
217 { "insn_emulation", VCPU_STAT(insn_emulation) },
218 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
219 { "irq_injections", VCPU_STAT(irq_injections) },
220 { "nmi_injections", VCPU_STAT(nmi_injections) },
221 { "req_event", VCPU_STAT(req_event) },
222 { "l1d_flush", VCPU_STAT(l1d_flush) },
223 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
224 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
225 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
226 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
227 { "mmu_flooded", VM_STAT(mmu_flooded) },
228 { "mmu_recycled", VM_STAT(mmu_recycled) },
229 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
230 { "mmu_unsync", VM_STAT(mmu_unsync) },
231 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
232 { "largepages", VM_STAT(lpages, .mode = 0444) },
233 { "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) },
234 { "max_mmu_page_hash_collisions",
235 VM_STAT(max_mmu_page_hash_collisions) },
239 u64 __read_mostly host_xcr0;
240 u64 __read_mostly supported_xcr0;
241 EXPORT_SYMBOL_GPL(supported_xcr0);
243 struct kmem_cache *x86_fpu_cache;
244 EXPORT_SYMBOL_GPL(x86_fpu_cache);
246 static struct kmem_cache *x86_emulator_cache;
248 static struct kmem_cache *kvm_alloc_emulator_cache(void)
250 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
251 unsigned int size = sizeof(struct x86_emulate_ctxt);
253 return kmem_cache_create_usercopy("x86_emulator", size,
254 __alignof__(struct x86_emulate_ctxt),
255 SLAB_ACCOUNT, useroffset,
256 size - useroffset, NULL);
259 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
261 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
264 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
265 vcpu->arch.apf.gfns[i] = ~0;
268 static void kvm_on_user_return(struct user_return_notifier *urn)
271 struct kvm_shared_msrs *locals
272 = container_of(urn, struct kvm_shared_msrs, urn);
273 struct kvm_shared_msr_values *values;
277 * Disabling irqs at this point since the following code could be
278 * interrupted and executed through kvm_arch_hardware_disable()
280 local_irq_save(flags);
281 if (locals->registered) {
282 locals->registered = false;
283 user_return_notifier_unregister(urn);
285 local_irq_restore(flags);
286 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
287 values = &locals->values[slot];
288 if (values->host != values->curr) {
289 wrmsrl(shared_msrs_global.msrs[slot], values->host);
290 values->curr = values->host;
295 void kvm_define_shared_msr(unsigned slot, u32 msr)
297 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
298 shared_msrs_global.msrs[slot] = msr;
299 if (slot >= shared_msrs_global.nr)
300 shared_msrs_global.nr = slot + 1;
302 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
304 static void kvm_shared_msr_cpu_online(void)
306 unsigned int cpu = smp_processor_id();
307 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
311 for (i = 0; i < shared_msrs_global.nr; ++i) {
312 rdmsrl_safe(shared_msrs_global.msrs[i], &value);
313 smsr->values[i].host = value;
314 smsr->values[i].curr = value;
318 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
320 unsigned int cpu = smp_processor_id();
321 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
324 value = (value & mask) | (smsr->values[slot].host & ~mask);
325 if (value == smsr->values[slot].curr)
327 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
331 smsr->values[slot].curr = value;
332 if (!smsr->registered) {
333 smsr->urn.on_user_return = kvm_on_user_return;
334 user_return_notifier_register(&smsr->urn);
335 smsr->registered = true;
339 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
341 static void drop_user_return_notifiers(void)
343 unsigned int cpu = smp_processor_id();
344 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
346 if (smsr->registered)
347 kvm_on_user_return(&smsr->urn);
350 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
352 return vcpu->arch.apic_base;
354 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
356 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
358 return kvm_apic_mode(kvm_get_apic_base(vcpu));
360 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
362 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
364 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
365 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
366 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
367 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
369 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
371 if (!msr_info->host_initiated) {
372 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
374 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
378 kvm_lapic_set_base(vcpu, msr_info->data);
379 kvm_recalculate_apic_map(vcpu->kvm);
382 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
384 asmlinkage __visible void kvm_spurious_fault(void)
386 /* Fault while not rebooting. We want the trace. */
387 BUG_ON(!kvm_rebooting);
389 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
391 #define EXCPT_BENIGN 0
392 #define EXCPT_CONTRIBUTORY 1
395 static int exception_class(int vector)
405 return EXCPT_CONTRIBUTORY;
412 #define EXCPT_FAULT 0
414 #define EXCPT_ABORT 2
415 #define EXCPT_INTERRUPT 3
417 static int exception_type(int vector)
421 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
422 return EXCPT_INTERRUPT;
426 /* #DB is trap, as instruction watchpoints are handled elsewhere */
427 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
430 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
433 /* Reserved exceptions will result in fault */
437 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
439 unsigned nr = vcpu->arch.exception.nr;
440 bool has_payload = vcpu->arch.exception.has_payload;
441 unsigned long payload = vcpu->arch.exception.payload;
449 * "Certain debug exceptions may clear bit 0-3. The
450 * remaining contents of the DR6 register are never
451 * cleared by the processor".
453 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
455 * DR6.RTM is set by all #DB exceptions that don't clear it.
457 vcpu->arch.dr6 |= DR6_RTM;
458 vcpu->arch.dr6 |= payload;
460 * Bit 16 should be set in the payload whenever the #DB
461 * exception should clear DR6.RTM. This makes the payload
462 * compatible with the pending debug exceptions under VMX.
463 * Though not currently documented in the SDM, this also
464 * makes the payload compatible with the exit qualification
465 * for #DB exceptions under VMX.
467 vcpu->arch.dr6 ^= payload & DR6_RTM;
470 * The #DB payload is defined as compatible with the 'pending
471 * debug exceptions' field under VMX, not DR6. While bit 12 is
472 * defined in the 'pending debug exceptions' field (enabled
473 * breakpoint), it is reserved and must be zero in DR6.
475 vcpu->arch.dr6 &= ~BIT(12);
478 vcpu->arch.cr2 = payload;
482 vcpu->arch.exception.has_payload = false;
483 vcpu->arch.exception.payload = 0;
485 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
487 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
488 unsigned nr, bool has_error, u32 error_code,
489 bool has_payload, unsigned long payload, bool reinject)
494 kvm_make_request(KVM_REQ_EVENT, vcpu);
496 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
498 if (has_error && !is_protmode(vcpu))
502 * On vmentry, vcpu->arch.exception.pending is only
503 * true if an event injection was blocked by
504 * nested_run_pending. In that case, however,
505 * vcpu_enter_guest requests an immediate exit,
506 * and the guest shouldn't proceed far enough to
509 WARN_ON_ONCE(vcpu->arch.exception.pending);
510 vcpu->arch.exception.injected = true;
511 if (WARN_ON_ONCE(has_payload)) {
513 * A reinjected event has already
514 * delivered its payload.
520 vcpu->arch.exception.pending = true;
521 vcpu->arch.exception.injected = false;
523 vcpu->arch.exception.has_error_code = has_error;
524 vcpu->arch.exception.nr = nr;
525 vcpu->arch.exception.error_code = error_code;
526 vcpu->arch.exception.has_payload = has_payload;
527 vcpu->arch.exception.payload = payload;
528 if (!is_guest_mode(vcpu))
529 kvm_deliver_exception_payload(vcpu);
533 /* to check exception */
534 prev_nr = vcpu->arch.exception.nr;
535 if (prev_nr == DF_VECTOR) {
536 /* triple fault -> shutdown */
537 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
540 class1 = exception_class(prev_nr);
541 class2 = exception_class(nr);
542 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
543 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
545 * Generate double fault per SDM Table 5-5. Set
546 * exception.pending = true so that the double fault
547 * can trigger a nested vmexit.
549 vcpu->arch.exception.pending = true;
550 vcpu->arch.exception.injected = false;
551 vcpu->arch.exception.has_error_code = true;
552 vcpu->arch.exception.nr = DF_VECTOR;
553 vcpu->arch.exception.error_code = 0;
554 vcpu->arch.exception.has_payload = false;
555 vcpu->arch.exception.payload = 0;
557 /* replace previous exception with a new one in a hope
558 that instruction re-execution will regenerate lost
563 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
565 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
567 EXPORT_SYMBOL_GPL(kvm_queue_exception);
569 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
571 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
573 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
575 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
576 unsigned long payload)
578 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
581 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
582 u32 error_code, unsigned long payload)
584 kvm_multiple_exception(vcpu, nr, true, error_code,
585 true, payload, false);
588 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
591 kvm_inject_gp(vcpu, 0);
593 return kvm_skip_emulated_instruction(vcpu);
597 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
599 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
601 ++vcpu->stat.pf_guest;
602 vcpu->arch.exception.nested_apf =
603 is_guest_mode(vcpu) && fault->async_page_fault;
604 if (vcpu->arch.exception.nested_apf) {
605 vcpu->arch.apf.nested_apf_token = fault->address;
606 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
608 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
612 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
614 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
616 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
617 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
619 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
621 return fault->nested_page_fault;
624 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
626 atomic_inc(&vcpu->arch.nmi_queued);
627 kvm_make_request(KVM_REQ_NMI, vcpu);
629 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
631 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
633 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
635 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
637 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
639 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
641 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
644 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
645 * a #GP and return false.
647 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
649 if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
651 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
654 EXPORT_SYMBOL_GPL(kvm_require_cpl);
656 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
658 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
661 kvm_queue_exception(vcpu, UD_VECTOR);
664 EXPORT_SYMBOL_GPL(kvm_require_dr);
667 * This function will be used to read from the physical memory of the currently
668 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
669 * can read from guest physical or from the guest's guest physical memory.
671 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
672 gfn_t ngfn, void *data, int offset, int len,
675 struct x86_exception exception;
679 ngpa = gfn_to_gpa(ngfn);
680 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
681 if (real_gfn == UNMAPPED_GVA)
684 real_gfn = gpa_to_gfn(real_gfn);
686 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
688 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
690 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
691 void *data, int offset, int len, u32 access)
693 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
694 data, offset, len, access);
697 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
699 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
704 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
706 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
708 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
709 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
712 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
714 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
715 offset * sizeof(u64), sizeof(pdpte),
716 PFERR_USER_MASK|PFERR_WRITE_MASK);
721 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
722 if ((pdpte[i] & PT_PRESENT_MASK) &&
723 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
730 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
731 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
737 EXPORT_SYMBOL_GPL(load_pdptrs);
739 bool pdptrs_changed(struct kvm_vcpu *vcpu)
741 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
746 if (!is_pae_paging(vcpu))
749 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
752 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
753 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
754 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
755 PFERR_USER_MASK | PFERR_WRITE_MASK);
759 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
761 EXPORT_SYMBOL_GPL(pdptrs_changed);
763 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
765 unsigned long old_cr0 = kvm_read_cr0(vcpu);
766 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
771 if (cr0 & 0xffffffff00000000UL)
775 cr0 &= ~CR0_RESERVED_BITS;
777 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
780 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
783 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
785 if ((vcpu->arch.efer & EFER_LME)) {
790 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
795 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
800 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
803 kvm_x86_ops.set_cr0(vcpu, cr0);
805 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
806 kvm_clear_async_pf_completion_queue(vcpu);
807 kvm_async_pf_hash_reset(vcpu);
810 if ((cr0 ^ old_cr0) & update_bits)
811 kvm_mmu_reset_context(vcpu);
813 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
814 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
815 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
816 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
820 EXPORT_SYMBOL_GPL(kvm_set_cr0);
822 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
824 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
826 EXPORT_SYMBOL_GPL(kvm_lmsw);
828 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
830 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
832 if (vcpu->arch.xcr0 != host_xcr0)
833 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
835 if (vcpu->arch.xsaves_enabled &&
836 vcpu->arch.ia32_xss != host_xss)
837 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
840 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
842 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
844 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
846 if (vcpu->arch.xcr0 != host_xcr0)
847 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
849 if (vcpu->arch.xsaves_enabled &&
850 vcpu->arch.ia32_xss != host_xss)
851 wrmsrl(MSR_IA32_XSS, host_xss);
855 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
857 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
860 u64 old_xcr0 = vcpu->arch.xcr0;
863 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
864 if (index != XCR_XFEATURE_ENABLED_MASK)
866 if (!(xcr0 & XFEATURE_MASK_FP))
868 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
872 * Do not allow the guest to set bits that we do not support
873 * saving. However, xcr0 bit 0 is always set, even if the
874 * emulated CPU does not support XSAVE (see fx_init).
876 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
877 if (xcr0 & ~valid_bits)
880 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
881 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
884 if (xcr0 & XFEATURE_MASK_AVX512) {
885 if (!(xcr0 & XFEATURE_MASK_YMM))
887 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
890 vcpu->arch.xcr0 = xcr0;
892 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
893 kvm_update_cpuid(vcpu);
897 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
899 if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
900 __kvm_set_xcr(vcpu, index, xcr)) {
901 kvm_inject_gp(vcpu, 0);
906 EXPORT_SYMBOL_GPL(kvm_set_xcr);
908 #define __cr4_reserved_bits(__cpu_has, __c) \
910 u64 __reserved_bits = CR4_RESERVED_BITS; \
912 if (!__cpu_has(__c, X86_FEATURE_XSAVE)) \
913 __reserved_bits |= X86_CR4_OSXSAVE; \
914 if (!__cpu_has(__c, X86_FEATURE_SMEP)) \
915 __reserved_bits |= X86_CR4_SMEP; \
916 if (!__cpu_has(__c, X86_FEATURE_SMAP)) \
917 __reserved_bits |= X86_CR4_SMAP; \
918 if (!__cpu_has(__c, X86_FEATURE_FSGSBASE)) \
919 __reserved_bits |= X86_CR4_FSGSBASE; \
920 if (!__cpu_has(__c, X86_FEATURE_PKU)) \
921 __reserved_bits |= X86_CR4_PKE; \
922 if (!__cpu_has(__c, X86_FEATURE_LA57)) \
923 __reserved_bits |= X86_CR4_LA57; \
924 if (!__cpu_has(__c, X86_FEATURE_UMIP)) \
925 __reserved_bits |= X86_CR4_UMIP; \
929 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
931 if (cr4 & cr4_reserved_bits)
934 if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu))
940 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
942 unsigned long old_cr4 = kvm_read_cr4(vcpu);
943 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
944 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
946 if (kvm_valid_cr4(vcpu, cr4))
949 if (is_long_mode(vcpu)) {
950 if (!(cr4 & X86_CR4_PAE))
952 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
953 && ((cr4 ^ old_cr4) & pdptr_bits)
954 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
958 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
959 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
962 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
963 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
967 if (kvm_x86_ops.set_cr4(vcpu, cr4))
970 if (((cr4 ^ old_cr4) & pdptr_bits) ||
971 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
972 kvm_mmu_reset_context(vcpu);
974 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
975 kvm_update_cpuid(vcpu);
979 EXPORT_SYMBOL_GPL(kvm_set_cr4);
981 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
983 bool skip_tlb_flush = false;
985 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
988 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
989 cr3 &= ~X86_CR3_PCID_NOFLUSH;
993 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
994 if (!skip_tlb_flush) {
995 kvm_mmu_sync_roots(vcpu);
996 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1001 if (is_long_mode(vcpu) &&
1002 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
1004 else if (is_pae_paging(vcpu) &&
1005 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1008 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
1009 vcpu->arch.cr3 = cr3;
1010 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1014 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1016 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1018 if (cr8 & CR8_RESERVED_BITS)
1020 if (lapic_in_kernel(vcpu))
1021 kvm_lapic_set_tpr(vcpu, cr8);
1023 vcpu->arch.cr8 = cr8;
1026 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1028 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1030 if (lapic_in_kernel(vcpu))
1031 return kvm_lapic_get_cr8(vcpu);
1033 return vcpu->arch.cr8;
1035 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1037 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1041 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1042 for (i = 0; i < KVM_NR_DB_REGS; i++)
1043 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1044 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1048 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1050 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1051 kvm_x86_ops.set_dr6(vcpu, vcpu->arch.dr6);
1054 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1058 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1059 dr7 = vcpu->arch.guest_debug_dr7;
1061 dr7 = vcpu->arch.dr7;
1062 kvm_x86_ops.set_dr7(vcpu, dr7);
1063 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1064 if (dr7 & DR7_BP_EN_MASK)
1065 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1068 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1070 u64 fixed = DR6_FIXED_1;
1072 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1077 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1079 size_t size = ARRAY_SIZE(vcpu->arch.db);
1083 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1084 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1085 vcpu->arch.eff_db[dr] = val;
1090 if (val & 0xffffffff00000000ULL)
1091 return -1; /* #GP */
1092 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1093 kvm_update_dr6(vcpu);
1098 if (!kvm_dr7_valid(val))
1099 return -1; /* #GP */
1100 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1101 kvm_update_dr7(vcpu);
1108 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1110 if (__kvm_set_dr(vcpu, dr, val)) {
1111 kvm_inject_gp(vcpu, 0);
1116 EXPORT_SYMBOL_GPL(kvm_set_dr);
1118 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1120 size_t size = ARRAY_SIZE(vcpu->arch.db);
1124 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1129 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1130 *val = vcpu->arch.dr6;
1132 *val = kvm_x86_ops.get_dr6(vcpu);
1137 *val = vcpu->arch.dr7;
1142 EXPORT_SYMBOL_GPL(kvm_get_dr);
1144 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1146 u32 ecx = kvm_rcx_read(vcpu);
1150 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1153 kvm_rax_write(vcpu, (u32)data);
1154 kvm_rdx_write(vcpu, data >> 32);
1157 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1160 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1161 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1163 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1164 * extract the supported MSRs from the related const lists.
1165 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1166 * capabilities of the host cpu. This capabilities test skips MSRs that are
1167 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1168 * may depend on host virtualization features rather than host cpu features.
1171 static const u32 msrs_to_save_all[] = {
1172 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1174 #ifdef CONFIG_X86_64
1175 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1177 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1178 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1180 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1181 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1182 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1183 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1184 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1185 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1186 MSR_IA32_UMWAIT_CONTROL,
1188 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1189 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1190 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1191 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1192 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1193 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1194 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1195 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1196 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1197 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1198 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1199 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1200 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1201 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1202 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1203 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1204 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1205 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1206 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1207 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1208 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1209 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1212 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1213 static unsigned num_msrs_to_save;
1215 static const u32 emulated_msrs_all[] = {
1216 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1217 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1218 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1219 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1220 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1221 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1222 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1224 HV_X64_MSR_VP_INDEX,
1225 HV_X64_MSR_VP_RUNTIME,
1226 HV_X64_MSR_SCONTROL,
1227 HV_X64_MSR_STIMER0_CONFIG,
1228 HV_X64_MSR_VP_ASSIST_PAGE,
1229 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1230 HV_X64_MSR_TSC_EMULATION_STATUS,
1232 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1235 MSR_IA32_TSC_ADJUST,
1236 MSR_IA32_TSCDEADLINE,
1237 MSR_IA32_ARCH_CAPABILITIES,
1238 MSR_IA32_MISC_ENABLE,
1239 MSR_IA32_MCG_STATUS,
1241 MSR_IA32_MCG_EXT_CTL,
1245 MSR_MISC_FEATURES_ENABLES,
1246 MSR_AMD64_VIRT_SPEC_CTRL,
1251 * The following list leaves out MSRs whose values are determined
1252 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1253 * We always support the "true" VMX control MSRs, even if the host
1254 * processor does not, so I am putting these registers here rather
1255 * than in msrs_to_save_all.
1258 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1259 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1260 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1261 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1263 MSR_IA32_VMX_CR0_FIXED0,
1264 MSR_IA32_VMX_CR4_FIXED0,
1265 MSR_IA32_VMX_VMCS_ENUM,
1266 MSR_IA32_VMX_PROCBASED_CTLS2,
1267 MSR_IA32_VMX_EPT_VPID_CAP,
1268 MSR_IA32_VMX_VMFUNC,
1271 MSR_KVM_POLL_CONTROL,
1274 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1275 static unsigned num_emulated_msrs;
1278 * List of msr numbers which are used to expose MSR-based features that
1279 * can be used by a hypervisor to validate requested CPU features.
1281 static const u32 msr_based_features_all[] = {
1283 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1284 MSR_IA32_VMX_PINBASED_CTLS,
1285 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1286 MSR_IA32_VMX_PROCBASED_CTLS,
1287 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1288 MSR_IA32_VMX_EXIT_CTLS,
1289 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1290 MSR_IA32_VMX_ENTRY_CTLS,
1292 MSR_IA32_VMX_CR0_FIXED0,
1293 MSR_IA32_VMX_CR0_FIXED1,
1294 MSR_IA32_VMX_CR4_FIXED0,
1295 MSR_IA32_VMX_CR4_FIXED1,
1296 MSR_IA32_VMX_VMCS_ENUM,
1297 MSR_IA32_VMX_PROCBASED_CTLS2,
1298 MSR_IA32_VMX_EPT_VPID_CAP,
1299 MSR_IA32_VMX_VMFUNC,
1303 MSR_IA32_ARCH_CAPABILITIES,
1306 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1307 static unsigned int num_msr_based_features;
1309 static u64 kvm_get_arch_capabilities(void)
1313 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1314 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1317 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1318 * the nested hypervisor runs with NX huge pages. If it is not,
1319 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1320 * L1 guests, so it need not worry about its own (L2) guests.
1322 data |= ARCH_CAP_PSCHANGE_MC_NO;
1325 * If we're doing cache flushes (either "always" or "cond")
1326 * we will do one whenever the guest does a vmlaunch/vmresume.
1327 * If an outer hypervisor is doing the cache flush for us
1328 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1329 * capability to the guest too, and if EPT is disabled we're not
1330 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1331 * require a nested hypervisor to do a flush of its own.
1333 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1334 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1336 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1337 data |= ARCH_CAP_RDCL_NO;
1338 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1339 data |= ARCH_CAP_SSB_NO;
1340 if (!boot_cpu_has_bug(X86_BUG_MDS))
1341 data |= ARCH_CAP_MDS_NO;
1344 * On TAA affected systems:
1345 * - nothing to do if TSX is disabled on the host.
1346 * - we emulate TSX_CTRL if present on the host.
1347 * This lets the guest use VERW to clear CPU buffers.
1349 if (!boot_cpu_has(X86_FEATURE_RTM))
1350 data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1351 else if (!boot_cpu_has_bug(X86_BUG_TAA))
1352 data |= ARCH_CAP_TAA_NO;
1357 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1359 switch (msr->index) {
1360 case MSR_IA32_ARCH_CAPABILITIES:
1361 msr->data = kvm_get_arch_capabilities();
1363 case MSR_IA32_UCODE_REV:
1364 rdmsrl_safe(msr->index, &msr->data);
1367 if (kvm_x86_ops.get_msr_feature(msr))
1373 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1375 struct kvm_msr_entry msr;
1379 r = kvm_get_msr_feature(&msr);
1388 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1390 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1393 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1396 if (efer & (EFER_LME | EFER_LMA) &&
1397 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1400 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1406 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1408 if (efer & efer_reserved_bits)
1411 return __kvm_valid_efer(vcpu, efer);
1413 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1415 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1417 u64 old_efer = vcpu->arch.efer;
1418 u64 efer = msr_info->data;
1420 if (efer & efer_reserved_bits)
1423 if (!msr_info->host_initiated) {
1424 if (!__kvm_valid_efer(vcpu, efer))
1427 if (is_paging(vcpu) &&
1428 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1433 efer |= vcpu->arch.efer & EFER_LMA;
1435 kvm_x86_ops.set_efer(vcpu, efer);
1437 /* Update reserved bits */
1438 if ((efer ^ old_efer) & EFER_NX)
1439 kvm_mmu_reset_context(vcpu);
1444 void kvm_enable_efer_bits(u64 mask)
1446 efer_reserved_bits &= ~mask;
1448 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1451 * Write @data into the MSR specified by @index. Select MSR specific fault
1452 * checks are bypassed if @host_initiated is %true.
1453 * Returns 0 on success, non-0 otherwise.
1454 * Assumes vcpu_load() was already called.
1456 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1457 bool host_initiated)
1459 struct msr_data msr;
1464 case MSR_KERNEL_GS_BASE:
1467 if (is_noncanonical_address(data, vcpu))
1470 case MSR_IA32_SYSENTER_EIP:
1471 case MSR_IA32_SYSENTER_ESP:
1473 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1474 * non-canonical address is written on Intel but not on
1475 * AMD (which ignores the top 32-bits, because it does
1476 * not implement 64-bit SYSENTER).
1478 * 64-bit code should hence be able to write a non-canonical
1479 * value on AMD. Making the address canonical ensures that
1480 * vmentry does not fail on Intel after writing a non-canonical
1481 * value, and that something deterministic happens if the guest
1482 * invokes 64-bit SYSENTER.
1484 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1489 msr.host_initiated = host_initiated;
1491 return kvm_x86_ops.set_msr(vcpu, &msr);
1495 * Read the MSR specified by @index into @data. Select MSR specific fault
1496 * checks are bypassed if @host_initiated is %true.
1497 * Returns 0 on success, non-0 otherwise.
1498 * Assumes vcpu_load() was already called.
1500 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1501 bool host_initiated)
1503 struct msr_data msr;
1507 msr.host_initiated = host_initiated;
1509 ret = kvm_x86_ops.get_msr(vcpu, &msr);
1515 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1517 return __kvm_get_msr(vcpu, index, data, false);
1519 EXPORT_SYMBOL_GPL(kvm_get_msr);
1521 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1523 return __kvm_set_msr(vcpu, index, data, false);
1525 EXPORT_SYMBOL_GPL(kvm_set_msr);
1527 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1529 u32 ecx = kvm_rcx_read(vcpu);
1532 if (kvm_get_msr(vcpu, ecx, &data)) {
1533 trace_kvm_msr_read_ex(ecx);
1534 kvm_inject_gp(vcpu, 0);
1538 trace_kvm_msr_read(ecx, data);
1540 kvm_rax_write(vcpu, data & -1u);
1541 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1542 return kvm_skip_emulated_instruction(vcpu);
1544 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1546 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1548 u32 ecx = kvm_rcx_read(vcpu);
1549 u64 data = kvm_read_edx_eax(vcpu);
1551 if (kvm_set_msr(vcpu, ecx, data)) {
1552 trace_kvm_msr_write_ex(ecx, data);
1553 kvm_inject_gp(vcpu, 0);
1557 trace_kvm_msr_write(ecx, data);
1558 return kvm_skip_emulated_instruction(vcpu);
1560 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1563 * The fast path for frequent and performance sensitive wrmsr emulation,
1564 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1565 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1566 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1567 * other cases which must be called after interrupts are enabled on the host.
1569 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1571 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1574 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1575 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1576 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1577 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1580 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1581 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1582 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1583 trace_kvm_apic_write(APIC_ICR, (u32)data);
1590 enum exit_fastpath_completion handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1592 u32 msr = kvm_rcx_read(vcpu);
1597 case APIC_BASE_MSR + (APIC_ICR >> 4):
1598 data = kvm_read_edx_eax(vcpu);
1599 ret = handle_fastpath_set_x2apic_icr_irqoff(vcpu, data);
1602 return EXIT_FASTPATH_NONE;
1606 trace_kvm_msr_write(msr, data);
1607 return EXIT_FASTPATH_SKIP_EMUL_INS;
1610 return EXIT_FASTPATH_NONE;
1612 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1615 * Adapt set_msr() to msr_io()'s calling convention
1617 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1619 return __kvm_get_msr(vcpu, index, data, true);
1622 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1624 return __kvm_set_msr(vcpu, index, *data, true);
1627 #ifdef CONFIG_X86_64
1628 struct pvclock_clock {
1638 struct pvclock_gtod_data {
1641 struct pvclock_clock clock; /* extract of a clocksource struct */
1642 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1648 static struct pvclock_gtod_data pvclock_gtod_data;
1650 static void update_pvclock_gtod(struct timekeeper *tk)
1652 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1654 write_seqcount_begin(&vdata->seq);
1656 /* copy pvclock gtod data */
1657 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1658 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1659 vdata->clock.mask = tk->tkr_mono.mask;
1660 vdata->clock.mult = tk->tkr_mono.mult;
1661 vdata->clock.shift = tk->tkr_mono.shift;
1662 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1663 vdata->clock.offset = tk->tkr_mono.base;
1665 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1666 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1667 vdata->raw_clock.mask = tk->tkr_raw.mask;
1668 vdata->raw_clock.mult = tk->tkr_raw.mult;
1669 vdata->raw_clock.shift = tk->tkr_raw.shift;
1670 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1671 vdata->raw_clock.offset = tk->tkr_raw.base;
1673 vdata->wall_time_sec = tk->xtime_sec;
1675 vdata->offs_boot = tk->offs_boot;
1677 write_seqcount_end(&vdata->seq);
1680 static s64 get_kvmclock_base_ns(void)
1682 /* Count up from boot time, but with the frequency of the raw clock. */
1683 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1686 static s64 get_kvmclock_base_ns(void)
1688 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1689 return ktime_get_boottime_ns();
1693 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1695 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1696 kvm_vcpu_kick(vcpu);
1699 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1703 struct pvclock_wall_clock wc;
1709 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1714 ++version; /* first time write, random junk */
1718 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1722 * The guest calculates current wall clock time by adding
1723 * system time (updated by kvm_guest_time_update below) to the
1724 * wall clock specified here. We do the reverse here.
1726 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1728 wc.nsec = do_div(wall_nsec, 1000000000);
1729 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1730 wc.version = version;
1732 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1735 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1738 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1740 do_shl32_div32(dividend, divisor);
1744 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1745 s8 *pshift, u32 *pmultiplier)
1753 scaled64 = scaled_hz;
1754 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1759 tps32 = (uint32_t)tps64;
1760 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1761 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1769 *pmultiplier = div_frac(scaled64, tps32);
1772 #ifdef CONFIG_X86_64
1773 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1776 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1777 static unsigned long max_tsc_khz;
1779 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1781 u64 v = (u64)khz * (1000000 + ppm);
1786 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1790 /* Guest TSC same frequency as host TSC? */
1792 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1796 /* TSC scaling supported? */
1797 if (!kvm_has_tsc_control) {
1798 if (user_tsc_khz > tsc_khz) {
1799 vcpu->arch.tsc_catchup = 1;
1800 vcpu->arch.tsc_always_catchup = 1;
1803 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1808 /* TSC scaling required - calculate ratio */
1809 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1810 user_tsc_khz, tsc_khz);
1812 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1813 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1818 vcpu->arch.tsc_scaling_ratio = ratio;
1822 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1824 u32 thresh_lo, thresh_hi;
1825 int use_scaling = 0;
1827 /* tsc_khz can be zero if TSC calibration fails */
1828 if (user_tsc_khz == 0) {
1829 /* set tsc_scaling_ratio to a safe value */
1830 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1834 /* Compute a scale to convert nanoseconds in TSC cycles */
1835 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1836 &vcpu->arch.virtual_tsc_shift,
1837 &vcpu->arch.virtual_tsc_mult);
1838 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1841 * Compute the variation in TSC rate which is acceptable
1842 * within the range of tolerance and decide if the
1843 * rate being applied is within that bounds of the hardware
1844 * rate. If so, no scaling or compensation need be done.
1846 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1847 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1848 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1849 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1852 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1855 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1857 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1858 vcpu->arch.virtual_tsc_mult,
1859 vcpu->arch.virtual_tsc_shift);
1860 tsc += vcpu->arch.this_tsc_write;
1864 static inline int gtod_is_based_on_tsc(int mode)
1866 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
1869 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1871 #ifdef CONFIG_X86_64
1873 struct kvm_arch *ka = &vcpu->kvm->arch;
1874 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1876 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1877 atomic_read(&vcpu->kvm->online_vcpus));
1880 * Once the masterclock is enabled, always perform request in
1881 * order to update it.
1883 * In order to enable masterclock, the host clocksource must be TSC
1884 * and the vcpus need to have matched TSCs. When that happens,
1885 * perform request to enable masterclock.
1887 if (ka->use_master_clock ||
1888 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1889 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1891 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1892 atomic_read(&vcpu->kvm->online_vcpus),
1893 ka->use_master_clock, gtod->clock.vclock_mode);
1897 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1899 u64 curr_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
1900 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1904 * Multiply tsc by a fixed point number represented by ratio.
1906 * The most significant 64-N bits (mult) of ratio represent the
1907 * integral part of the fixed point number; the remaining N bits
1908 * (frac) represent the fractional part, ie. ratio represents a fixed
1909 * point number (mult + frac * 2^(-N)).
1911 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1913 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1915 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1918 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1921 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1923 if (ratio != kvm_default_tsc_scaling_ratio)
1924 _tsc = __scale_tsc(ratio, tsc);
1928 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1930 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1934 tsc = kvm_scale_tsc(vcpu, rdtsc());
1936 return target_tsc - tsc;
1939 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1941 u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
1943 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1945 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1947 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1949 vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
1952 static inline bool kvm_check_tsc_unstable(void)
1954 #ifdef CONFIG_X86_64
1956 * TSC is marked unstable when we're running on Hyper-V,
1957 * 'TSC page' clocksource is good.
1959 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
1962 return check_tsc_unstable();
1965 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1967 struct kvm *kvm = vcpu->kvm;
1968 u64 offset, ns, elapsed;
1969 unsigned long flags;
1971 bool already_matched;
1972 u64 data = msr->data;
1973 bool synchronizing = false;
1975 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1976 offset = kvm_compute_tsc_offset(vcpu, data);
1977 ns = get_kvmclock_base_ns();
1978 elapsed = ns - kvm->arch.last_tsc_nsec;
1980 if (vcpu->arch.virtual_tsc_khz) {
1981 if (data == 0 && msr->host_initiated) {
1983 * detection of vcpu initialization -- need to sync
1984 * with other vCPUs. This particularly helps to keep
1985 * kvm_clock stable after CPU hotplug
1987 synchronizing = true;
1989 u64 tsc_exp = kvm->arch.last_tsc_write +
1990 nsec_to_cycles(vcpu, elapsed);
1991 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1993 * Special case: TSC write with a small delta (1 second)
1994 * of virtual cycle time against real time is
1995 * interpreted as an attempt to synchronize the CPU.
1997 synchronizing = data < tsc_exp + tsc_hz &&
1998 data + tsc_hz > tsc_exp;
2003 * For a reliable TSC, we can match TSC offsets, and for an unstable
2004 * TSC, we add elapsed time in this computation. We could let the
2005 * compensation code attempt to catch up if we fall behind, but
2006 * it's better to try to match offsets from the beginning.
2008 if (synchronizing &&
2009 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2010 if (!kvm_check_tsc_unstable()) {
2011 offset = kvm->arch.cur_tsc_offset;
2013 u64 delta = nsec_to_cycles(vcpu, elapsed);
2015 offset = kvm_compute_tsc_offset(vcpu, data);
2018 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2021 * We split periods of matched TSC writes into generations.
2022 * For each generation, we track the original measured
2023 * nanosecond time, offset, and write, so if TSCs are in
2024 * sync, we can match exact offset, and if not, we can match
2025 * exact software computation in compute_guest_tsc()
2027 * These values are tracked in kvm->arch.cur_xxx variables.
2029 kvm->arch.cur_tsc_generation++;
2030 kvm->arch.cur_tsc_nsec = ns;
2031 kvm->arch.cur_tsc_write = data;
2032 kvm->arch.cur_tsc_offset = offset;
2037 * We also track th most recent recorded KHZ, write and time to
2038 * allow the matching interval to be extended at each write.
2040 kvm->arch.last_tsc_nsec = ns;
2041 kvm->arch.last_tsc_write = data;
2042 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2044 vcpu->arch.last_guest_tsc = data;
2046 /* Keep track of which generation this VCPU has synchronized to */
2047 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2048 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2049 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2051 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
2052 update_ia32_tsc_adjust_msr(vcpu, offset);
2054 kvm_vcpu_write_tsc_offset(vcpu, offset);
2055 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2057 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2059 kvm->arch.nr_vcpus_matched_tsc = 0;
2060 } else if (!already_matched) {
2061 kvm->arch.nr_vcpus_matched_tsc++;
2064 kvm_track_tsc_matching(vcpu);
2065 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2068 EXPORT_SYMBOL_GPL(kvm_write_tsc);
2070 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2073 u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
2074 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2077 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2079 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2080 WARN_ON(adjustment < 0);
2081 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2082 adjust_tsc_offset_guest(vcpu, adjustment);
2085 #ifdef CONFIG_X86_64
2087 static u64 read_tsc(void)
2089 u64 ret = (u64)rdtsc_ordered();
2090 u64 last = pvclock_gtod_data.clock.cycle_last;
2092 if (likely(ret >= last))
2096 * GCC likes to generate cmov here, but this branch is extremely
2097 * predictable (it's just a function of time and the likely is
2098 * very likely) and there's a data dependence, so force GCC
2099 * to generate a branch instead. I don't barrier() because
2100 * we don't actually need a barrier, and if this function
2101 * ever gets inlined it will generate worse code.
2107 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2113 switch (clock->vclock_mode) {
2114 case VDSO_CLOCKMODE_HVCLOCK:
2115 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2117 if (tsc_pg_val != U64_MAX) {
2118 /* TSC page valid */
2119 *mode = VDSO_CLOCKMODE_HVCLOCK;
2120 v = (tsc_pg_val - clock->cycle_last) &
2123 /* TSC page invalid */
2124 *mode = VDSO_CLOCKMODE_NONE;
2127 case VDSO_CLOCKMODE_TSC:
2128 *mode = VDSO_CLOCKMODE_TSC;
2129 *tsc_timestamp = read_tsc();
2130 v = (*tsc_timestamp - clock->cycle_last) &
2134 *mode = VDSO_CLOCKMODE_NONE;
2137 if (*mode == VDSO_CLOCKMODE_NONE)
2138 *tsc_timestamp = v = 0;
2140 return v * clock->mult;
2143 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2145 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2151 seq = read_seqcount_begin(>od->seq);
2152 ns = gtod->raw_clock.base_cycles;
2153 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2154 ns >>= gtod->raw_clock.shift;
2155 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2156 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2162 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2164 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2170 seq = read_seqcount_begin(>od->seq);
2171 ts->tv_sec = gtod->wall_time_sec;
2172 ns = gtod->clock.base_cycles;
2173 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2174 ns >>= gtod->clock.shift;
2175 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2177 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2183 /* returns true if host is using TSC based clocksource */
2184 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2186 /* checked again under seqlock below */
2187 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2190 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2194 /* returns true if host is using TSC based clocksource */
2195 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2198 /* checked again under seqlock below */
2199 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2202 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2208 * Assuming a stable TSC across physical CPUS, and a stable TSC
2209 * across virtual CPUs, the following condition is possible.
2210 * Each numbered line represents an event visible to both
2211 * CPUs at the next numbered event.
2213 * "timespecX" represents host monotonic time. "tscX" represents
2216 * VCPU0 on CPU0 | VCPU1 on CPU1
2218 * 1. read timespec0,tsc0
2219 * 2. | timespec1 = timespec0 + N
2221 * 3. transition to guest | transition to guest
2222 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2223 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2224 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2226 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2229 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2231 * - 0 < N - M => M < N
2233 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2234 * always the case (the difference between two distinct xtime instances
2235 * might be smaller then the difference between corresponding TSC reads,
2236 * when updating guest vcpus pvclock areas).
2238 * To avoid that problem, do not allow visibility of distinct
2239 * system_timestamp/tsc_timestamp values simultaneously: use a master
2240 * copy of host monotonic time values. Update that master copy
2243 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2247 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2249 #ifdef CONFIG_X86_64
2250 struct kvm_arch *ka = &kvm->arch;
2252 bool host_tsc_clocksource, vcpus_matched;
2254 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2255 atomic_read(&kvm->online_vcpus));
2258 * If the host uses TSC clock, then passthrough TSC as stable
2261 host_tsc_clocksource = kvm_get_time_and_clockread(
2262 &ka->master_kernel_ns,
2263 &ka->master_cycle_now);
2265 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2266 && !ka->backwards_tsc_observed
2267 && !ka->boot_vcpu_runs_old_kvmclock;
2269 if (ka->use_master_clock)
2270 atomic_set(&kvm_guest_has_master_clock, 1);
2272 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2273 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2278 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2280 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2283 static void kvm_gen_update_masterclock(struct kvm *kvm)
2285 #ifdef CONFIG_X86_64
2287 struct kvm_vcpu *vcpu;
2288 struct kvm_arch *ka = &kvm->arch;
2290 spin_lock(&ka->pvclock_gtod_sync_lock);
2291 kvm_make_mclock_inprogress_request(kvm);
2292 /* no guest entries from this point */
2293 pvclock_update_vm_gtod_copy(kvm);
2295 kvm_for_each_vcpu(i, vcpu, kvm)
2296 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2298 /* guest entries allowed */
2299 kvm_for_each_vcpu(i, vcpu, kvm)
2300 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2302 spin_unlock(&ka->pvclock_gtod_sync_lock);
2306 u64 get_kvmclock_ns(struct kvm *kvm)
2308 struct kvm_arch *ka = &kvm->arch;
2309 struct pvclock_vcpu_time_info hv_clock;
2312 spin_lock(&ka->pvclock_gtod_sync_lock);
2313 if (!ka->use_master_clock) {
2314 spin_unlock(&ka->pvclock_gtod_sync_lock);
2315 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2318 hv_clock.tsc_timestamp = ka->master_cycle_now;
2319 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2320 spin_unlock(&ka->pvclock_gtod_sync_lock);
2322 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2325 if (__this_cpu_read(cpu_tsc_khz)) {
2326 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2327 &hv_clock.tsc_shift,
2328 &hv_clock.tsc_to_system_mul);
2329 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2331 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2338 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2340 struct kvm_vcpu_arch *vcpu = &v->arch;
2341 struct pvclock_vcpu_time_info guest_hv_clock;
2343 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2344 &guest_hv_clock, sizeof(guest_hv_clock))))
2347 /* This VCPU is paused, but it's legal for a guest to read another
2348 * VCPU's kvmclock, so we really have to follow the specification where
2349 * it says that version is odd if data is being modified, and even after
2352 * Version field updates must be kept separate. This is because
2353 * kvm_write_guest_cached might use a "rep movs" instruction, and
2354 * writes within a string instruction are weakly ordered. So there
2355 * are three writes overall.
2357 * As a small optimization, only write the version field in the first
2358 * and third write. The vcpu->pv_time cache is still valid, because the
2359 * version field is the first in the struct.
2361 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2363 if (guest_hv_clock.version & 1)
2364 ++guest_hv_clock.version; /* first time write, random junk */
2366 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2367 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2369 sizeof(vcpu->hv_clock.version));
2373 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2374 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2376 if (vcpu->pvclock_set_guest_stopped_request) {
2377 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2378 vcpu->pvclock_set_guest_stopped_request = false;
2381 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2383 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2385 sizeof(vcpu->hv_clock));
2389 vcpu->hv_clock.version++;
2390 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2392 sizeof(vcpu->hv_clock.version));
2395 static int kvm_guest_time_update(struct kvm_vcpu *v)
2397 unsigned long flags, tgt_tsc_khz;
2398 struct kvm_vcpu_arch *vcpu = &v->arch;
2399 struct kvm_arch *ka = &v->kvm->arch;
2401 u64 tsc_timestamp, host_tsc;
2403 bool use_master_clock;
2409 * If the host uses TSC clock, then passthrough TSC as stable
2412 spin_lock(&ka->pvclock_gtod_sync_lock);
2413 use_master_clock = ka->use_master_clock;
2414 if (use_master_clock) {
2415 host_tsc = ka->master_cycle_now;
2416 kernel_ns = ka->master_kernel_ns;
2418 spin_unlock(&ka->pvclock_gtod_sync_lock);
2420 /* Keep irq disabled to prevent changes to the clock */
2421 local_irq_save(flags);
2422 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2423 if (unlikely(tgt_tsc_khz == 0)) {
2424 local_irq_restore(flags);
2425 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2428 if (!use_master_clock) {
2430 kernel_ns = get_kvmclock_base_ns();
2433 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2436 * We may have to catch up the TSC to match elapsed wall clock
2437 * time for two reasons, even if kvmclock is used.
2438 * 1) CPU could have been running below the maximum TSC rate
2439 * 2) Broken TSC compensation resets the base at each VCPU
2440 * entry to avoid unknown leaps of TSC even when running
2441 * again on the same CPU. This may cause apparent elapsed
2442 * time to disappear, and the guest to stand still or run
2445 if (vcpu->tsc_catchup) {
2446 u64 tsc = compute_guest_tsc(v, kernel_ns);
2447 if (tsc > tsc_timestamp) {
2448 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2449 tsc_timestamp = tsc;
2453 local_irq_restore(flags);
2455 /* With all the info we got, fill in the values */
2457 if (kvm_has_tsc_control)
2458 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2460 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2461 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2462 &vcpu->hv_clock.tsc_shift,
2463 &vcpu->hv_clock.tsc_to_system_mul);
2464 vcpu->hw_tsc_khz = tgt_tsc_khz;
2467 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2468 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2469 vcpu->last_guest_tsc = tsc_timestamp;
2471 /* If the host uses TSC clocksource, then it is stable */
2473 if (use_master_clock)
2474 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2476 vcpu->hv_clock.flags = pvclock_flags;
2478 if (vcpu->pv_time_enabled)
2479 kvm_setup_pvclock_page(v);
2480 if (v == kvm_get_vcpu(v->kvm, 0))
2481 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2486 * kvmclock updates which are isolated to a given vcpu, such as
2487 * vcpu->cpu migration, should not allow system_timestamp from
2488 * the rest of the vcpus to remain static. Otherwise ntp frequency
2489 * correction applies to one vcpu's system_timestamp but not
2492 * So in those cases, request a kvmclock update for all vcpus.
2493 * We need to rate-limit these requests though, as they can
2494 * considerably slow guests that have a large number of vcpus.
2495 * The time for a remote vcpu to update its kvmclock is bound
2496 * by the delay we use to rate-limit the updates.
2499 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2501 static void kvmclock_update_fn(struct work_struct *work)
2504 struct delayed_work *dwork = to_delayed_work(work);
2505 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2506 kvmclock_update_work);
2507 struct kvm *kvm = container_of(ka, struct kvm, arch);
2508 struct kvm_vcpu *vcpu;
2510 kvm_for_each_vcpu(i, vcpu, kvm) {
2511 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2512 kvm_vcpu_kick(vcpu);
2516 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2518 struct kvm *kvm = v->kvm;
2520 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2521 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2522 KVMCLOCK_UPDATE_DELAY);
2525 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2527 static void kvmclock_sync_fn(struct work_struct *work)
2529 struct delayed_work *dwork = to_delayed_work(work);
2530 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2531 kvmclock_sync_work);
2532 struct kvm *kvm = container_of(ka, struct kvm, arch);
2534 if (!kvmclock_periodic_sync)
2537 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2538 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2539 KVMCLOCK_SYNC_PERIOD);
2543 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2545 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2547 /* McStatusWrEn enabled? */
2548 if (guest_cpuid_is_amd_or_hygon(vcpu))
2549 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2554 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2556 u64 mcg_cap = vcpu->arch.mcg_cap;
2557 unsigned bank_num = mcg_cap & 0xff;
2558 u32 msr = msr_info->index;
2559 u64 data = msr_info->data;
2562 case MSR_IA32_MCG_STATUS:
2563 vcpu->arch.mcg_status = data;
2565 case MSR_IA32_MCG_CTL:
2566 if (!(mcg_cap & MCG_CTL_P) &&
2567 (data || !msr_info->host_initiated))
2569 if (data != 0 && data != ~(u64)0)
2571 vcpu->arch.mcg_ctl = data;
2574 if (msr >= MSR_IA32_MC0_CTL &&
2575 msr < MSR_IA32_MCx_CTL(bank_num)) {
2576 u32 offset = array_index_nospec(
2577 msr - MSR_IA32_MC0_CTL,
2578 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2580 /* only 0 or all 1s can be written to IA32_MCi_CTL
2581 * some Linux kernels though clear bit 10 in bank 4 to
2582 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2583 * this to avoid an uncatched #GP in the guest
2585 if ((offset & 0x3) == 0 &&
2586 data != 0 && (data | (1 << 10)) != ~(u64)0)
2590 if (!msr_info->host_initiated &&
2591 (offset & 0x3) == 1 && data != 0) {
2592 if (!can_set_mci_status(vcpu))
2596 vcpu->arch.mce_banks[offset] = data;
2604 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2606 struct kvm *kvm = vcpu->kvm;
2607 int lm = is_long_mode(vcpu);
2608 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2609 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2610 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2611 : kvm->arch.xen_hvm_config.blob_size_32;
2612 u32 page_num = data & ~PAGE_MASK;
2613 u64 page_addr = data & PAGE_MASK;
2618 if (page_num >= blob_size)
2621 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2626 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2635 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2637 gpa_t gpa = data & ~0x3f;
2639 /* Bits 3:5 are reserved, Should be zero */
2643 vcpu->arch.apf.msr_val = data;
2645 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2646 kvm_clear_async_pf_completion_queue(vcpu);
2647 kvm_async_pf_hash_reset(vcpu);
2651 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2655 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2656 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2657 kvm_async_pf_wakeup_all(vcpu);
2661 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2663 vcpu->arch.pv_time_enabled = false;
2664 vcpu->arch.time = 0;
2667 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2669 ++vcpu->stat.tlb_flush;
2670 kvm_x86_ops.tlb_flush(vcpu, invalidate_gpa);
2673 static void record_steal_time(struct kvm_vcpu *vcpu)
2675 struct kvm_host_map map;
2676 struct kvm_steal_time *st;
2678 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2681 /* -EAGAIN is returned in atomic context so we can just return. */
2682 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2683 &map, &vcpu->arch.st.cache, false))
2687 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2690 * Doing a TLB flush here, on the guest's behalf, can avoid
2693 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2694 st->preempted & KVM_VCPU_FLUSH_TLB);
2695 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2696 kvm_vcpu_flush_tlb(vcpu, false);
2698 vcpu->arch.st.preempted = 0;
2700 if (st->version & 1)
2701 st->version += 1; /* first time write, random junk */
2707 st->steal += current->sched_info.run_delay -
2708 vcpu->arch.st.last_steal;
2709 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2715 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2718 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2721 u32 msr = msr_info->index;
2722 u64 data = msr_info->data;
2725 case MSR_AMD64_NB_CFG:
2726 case MSR_IA32_UCODE_WRITE:
2727 case MSR_VM_HSAVE_PA:
2728 case MSR_AMD64_PATCH_LOADER:
2729 case MSR_AMD64_BU_CFG2:
2730 case MSR_AMD64_DC_CFG:
2731 case MSR_F15H_EX_CFG:
2734 case MSR_IA32_UCODE_REV:
2735 if (msr_info->host_initiated)
2736 vcpu->arch.microcode_version = data;
2738 case MSR_IA32_ARCH_CAPABILITIES:
2739 if (!msr_info->host_initiated)
2741 vcpu->arch.arch_capabilities = data;
2744 return set_efer(vcpu, msr_info);
2746 data &= ~(u64)0x40; /* ignore flush filter disable */
2747 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2748 data &= ~(u64)0x8; /* ignore TLB cache disable */
2750 /* Handle McStatusWrEn */
2751 if (data == BIT_ULL(18)) {
2752 vcpu->arch.msr_hwcr = data;
2753 } else if (data != 0) {
2754 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2759 case MSR_FAM10H_MMIO_CONF_BASE:
2761 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2766 case MSR_IA32_DEBUGCTLMSR:
2768 /* We support the non-activated case already */
2770 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2771 /* Values other than LBR and BTF are vendor-specific,
2772 thus reserved and should throw a #GP */
2775 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2778 case 0x200 ... 0x2ff:
2779 return kvm_mtrr_set_msr(vcpu, msr, data);
2780 case MSR_IA32_APICBASE:
2781 return kvm_set_apic_base(vcpu, msr_info);
2782 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2783 return kvm_x2apic_msr_write(vcpu, msr, data);
2784 case MSR_IA32_TSCDEADLINE:
2785 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2787 case MSR_IA32_TSC_ADJUST:
2788 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2789 if (!msr_info->host_initiated) {
2790 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2791 adjust_tsc_offset_guest(vcpu, adj);
2793 vcpu->arch.ia32_tsc_adjust_msr = data;
2796 case MSR_IA32_MISC_ENABLE:
2797 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2798 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2799 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2801 vcpu->arch.ia32_misc_enable_msr = data;
2802 kvm_update_cpuid(vcpu);
2804 vcpu->arch.ia32_misc_enable_msr = data;
2807 case MSR_IA32_SMBASE:
2808 if (!msr_info->host_initiated)
2810 vcpu->arch.smbase = data;
2812 case MSR_IA32_POWER_CTL:
2813 vcpu->arch.msr_ia32_power_ctl = data;
2816 kvm_write_tsc(vcpu, msr_info);
2819 if (!msr_info->host_initiated &&
2820 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2823 * KVM supports exposing PT to the guest, but does not support
2824 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
2825 * XSAVES/XRSTORS to save/restore PT MSRs.
2827 if (data & ~supported_xss)
2829 vcpu->arch.ia32_xss = data;
2832 if (!msr_info->host_initiated)
2834 vcpu->arch.smi_count = data;
2836 case MSR_KVM_WALL_CLOCK_NEW:
2837 case MSR_KVM_WALL_CLOCK:
2838 vcpu->kvm->arch.wall_clock = data;
2839 kvm_write_wall_clock(vcpu->kvm, data);
2841 case MSR_KVM_SYSTEM_TIME_NEW:
2842 case MSR_KVM_SYSTEM_TIME: {
2843 struct kvm_arch *ka = &vcpu->kvm->arch;
2845 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2846 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2848 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2849 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2851 ka->boot_vcpu_runs_old_kvmclock = tmp;
2854 vcpu->arch.time = data;
2855 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2857 /* we verify if the enable bit is set... */
2858 vcpu->arch.pv_time_enabled = false;
2862 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2863 &vcpu->arch.pv_time, data & ~1ULL,
2864 sizeof(struct pvclock_vcpu_time_info)))
2865 vcpu->arch.pv_time_enabled = true;
2869 case MSR_KVM_ASYNC_PF_EN:
2870 if (kvm_pv_enable_async_pf(vcpu, data))
2873 case MSR_KVM_STEAL_TIME:
2875 if (unlikely(!sched_info_on()))
2878 if (data & KVM_STEAL_RESERVED_MASK)
2881 vcpu->arch.st.msr_val = data;
2883 if (!(data & KVM_MSR_ENABLED))
2886 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2889 case MSR_KVM_PV_EOI_EN:
2890 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2894 case MSR_KVM_POLL_CONTROL:
2895 /* only enable bit supported */
2896 if (data & (-1ULL << 1))
2899 vcpu->arch.msr_kvm_poll_control = data;
2902 case MSR_IA32_MCG_CTL:
2903 case MSR_IA32_MCG_STATUS:
2904 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2905 return set_msr_mce(vcpu, msr_info);
2907 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2908 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2909 pr = true; /* fall through */
2910 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2911 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2912 if (kvm_pmu_is_valid_msr(vcpu, msr))
2913 return kvm_pmu_set_msr(vcpu, msr_info);
2915 if (pr || data != 0)
2916 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2917 "0x%x data 0x%llx\n", msr, data);
2919 case MSR_K7_CLK_CTL:
2921 * Ignore all writes to this no longer documented MSR.
2922 * Writes are only relevant for old K7 processors,
2923 * all pre-dating SVM, but a recommended workaround from
2924 * AMD for these chips. It is possible to specify the
2925 * affected processor models on the command line, hence
2926 * the need to ignore the workaround.
2929 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2930 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2931 case HV_X64_MSR_CRASH_CTL:
2932 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2933 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2934 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2935 case HV_X64_MSR_TSC_EMULATION_STATUS:
2936 return kvm_hv_set_msr_common(vcpu, msr, data,
2937 msr_info->host_initiated);
2938 case MSR_IA32_BBL_CR_CTL3:
2939 /* Drop writes to this legacy MSR -- see rdmsr
2940 * counterpart for further detail.
2942 if (report_ignored_msrs)
2943 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2946 case MSR_AMD64_OSVW_ID_LENGTH:
2947 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2949 vcpu->arch.osvw.length = data;
2951 case MSR_AMD64_OSVW_STATUS:
2952 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2954 vcpu->arch.osvw.status = data;
2956 case MSR_PLATFORM_INFO:
2957 if (!msr_info->host_initiated ||
2958 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2959 cpuid_fault_enabled(vcpu)))
2961 vcpu->arch.msr_platform_info = data;
2963 case MSR_MISC_FEATURES_ENABLES:
2964 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2965 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2966 !supports_cpuid_fault(vcpu)))
2968 vcpu->arch.msr_misc_features_enables = data;
2971 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2972 return xen_hvm_config(vcpu, data);
2973 if (kvm_pmu_is_valid_msr(vcpu, msr))
2974 return kvm_pmu_set_msr(vcpu, msr_info);
2976 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2980 if (report_ignored_msrs)
2982 "ignored wrmsr: 0x%x data 0x%llx\n",
2989 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2991 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2994 u64 mcg_cap = vcpu->arch.mcg_cap;
2995 unsigned bank_num = mcg_cap & 0xff;
2998 case MSR_IA32_P5_MC_ADDR:
2999 case MSR_IA32_P5_MC_TYPE:
3002 case MSR_IA32_MCG_CAP:
3003 data = vcpu->arch.mcg_cap;
3005 case MSR_IA32_MCG_CTL:
3006 if (!(mcg_cap & MCG_CTL_P) && !host)
3008 data = vcpu->arch.mcg_ctl;
3010 case MSR_IA32_MCG_STATUS:
3011 data = vcpu->arch.mcg_status;
3014 if (msr >= MSR_IA32_MC0_CTL &&
3015 msr < MSR_IA32_MCx_CTL(bank_num)) {
3016 u32 offset = array_index_nospec(
3017 msr - MSR_IA32_MC0_CTL,
3018 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3020 data = vcpu->arch.mce_banks[offset];
3029 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3031 switch (msr_info->index) {
3032 case MSR_IA32_PLATFORM_ID:
3033 case MSR_IA32_EBL_CR_POWERON:
3034 case MSR_IA32_DEBUGCTLMSR:
3035 case MSR_IA32_LASTBRANCHFROMIP:
3036 case MSR_IA32_LASTBRANCHTOIP:
3037 case MSR_IA32_LASTINTFROMIP:
3038 case MSR_IA32_LASTINTTOIP:
3040 case MSR_K8_TSEG_ADDR:
3041 case MSR_K8_TSEG_MASK:
3042 case MSR_VM_HSAVE_PA:
3043 case MSR_K8_INT_PENDING_MSG:
3044 case MSR_AMD64_NB_CFG:
3045 case MSR_FAM10H_MMIO_CONF_BASE:
3046 case MSR_AMD64_BU_CFG2:
3047 case MSR_IA32_PERF_CTL:
3048 case MSR_AMD64_DC_CFG:
3049 case MSR_F15H_EX_CFG:
3051 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3052 * limit) MSRs. Just return 0, as we do not want to expose the host
3053 * data here. Do not conditionalize this on CPUID, as KVM does not do
3054 * so for existing CPU-specific MSRs.
3056 case MSR_RAPL_POWER_UNIT:
3057 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3058 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3059 case MSR_PKG_ENERGY_STATUS: /* Total package */
3060 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3063 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3064 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3065 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3066 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3067 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3068 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3069 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3072 case MSR_IA32_UCODE_REV:
3073 msr_info->data = vcpu->arch.microcode_version;
3075 case MSR_IA32_ARCH_CAPABILITIES:
3076 if (!msr_info->host_initiated &&
3077 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3079 msr_info->data = vcpu->arch.arch_capabilities;
3081 case MSR_IA32_POWER_CTL:
3082 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3085 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
3088 case 0x200 ... 0x2ff:
3089 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3090 case 0xcd: /* fsb frequency */
3094 * MSR_EBC_FREQUENCY_ID
3095 * Conservative value valid for even the basic CPU models.
3096 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3097 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3098 * and 266MHz for model 3, or 4. Set Core Clock
3099 * Frequency to System Bus Frequency Ratio to 1 (bits
3100 * 31:24) even though these are only valid for CPU
3101 * models > 2, however guests may end up dividing or
3102 * multiplying by zero otherwise.
3104 case MSR_EBC_FREQUENCY_ID:
3105 msr_info->data = 1 << 24;
3107 case MSR_IA32_APICBASE:
3108 msr_info->data = kvm_get_apic_base(vcpu);
3110 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
3111 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3112 case MSR_IA32_TSCDEADLINE:
3113 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3115 case MSR_IA32_TSC_ADJUST:
3116 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3118 case MSR_IA32_MISC_ENABLE:
3119 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3121 case MSR_IA32_SMBASE:
3122 if (!msr_info->host_initiated)
3124 msr_info->data = vcpu->arch.smbase;
3127 msr_info->data = vcpu->arch.smi_count;
3129 case MSR_IA32_PERF_STATUS:
3130 /* TSC increment by tick */
3131 msr_info->data = 1000ULL;
3132 /* CPU multiplier */
3133 msr_info->data |= (((uint64_t)4ULL) << 40);
3136 msr_info->data = vcpu->arch.efer;
3138 case MSR_KVM_WALL_CLOCK:
3139 case MSR_KVM_WALL_CLOCK_NEW:
3140 msr_info->data = vcpu->kvm->arch.wall_clock;
3142 case MSR_KVM_SYSTEM_TIME:
3143 case MSR_KVM_SYSTEM_TIME_NEW:
3144 msr_info->data = vcpu->arch.time;
3146 case MSR_KVM_ASYNC_PF_EN:
3147 msr_info->data = vcpu->arch.apf.msr_val;
3149 case MSR_KVM_STEAL_TIME:
3150 msr_info->data = vcpu->arch.st.msr_val;
3152 case MSR_KVM_PV_EOI_EN:
3153 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3155 case MSR_KVM_POLL_CONTROL:
3156 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3158 case MSR_IA32_P5_MC_ADDR:
3159 case MSR_IA32_P5_MC_TYPE:
3160 case MSR_IA32_MCG_CAP:
3161 case MSR_IA32_MCG_CTL:
3162 case MSR_IA32_MCG_STATUS:
3163 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3164 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3165 msr_info->host_initiated);
3167 if (!msr_info->host_initiated &&
3168 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3170 msr_info->data = vcpu->arch.ia32_xss;
3172 case MSR_K7_CLK_CTL:
3174 * Provide expected ramp-up count for K7. All other
3175 * are set to zero, indicating minimum divisors for
3178 * This prevents guest kernels on AMD host with CPU
3179 * type 6, model 8 and higher from exploding due to
3180 * the rdmsr failing.
3182 msr_info->data = 0x20000000;
3184 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3185 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3186 case HV_X64_MSR_CRASH_CTL:
3187 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3188 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3189 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3190 case HV_X64_MSR_TSC_EMULATION_STATUS:
3191 return kvm_hv_get_msr_common(vcpu,
3192 msr_info->index, &msr_info->data,
3193 msr_info->host_initiated);
3194 case MSR_IA32_BBL_CR_CTL3:
3195 /* This legacy MSR exists but isn't fully documented in current
3196 * silicon. It is however accessed by winxp in very narrow
3197 * scenarios where it sets bit #19, itself documented as
3198 * a "reserved" bit. Best effort attempt to source coherent
3199 * read data here should the balance of the register be
3200 * interpreted by the guest:
3202 * L2 cache control register 3: 64GB range, 256KB size,
3203 * enabled, latency 0x1, configured
3205 msr_info->data = 0xbe702111;
3207 case MSR_AMD64_OSVW_ID_LENGTH:
3208 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3210 msr_info->data = vcpu->arch.osvw.length;
3212 case MSR_AMD64_OSVW_STATUS:
3213 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3215 msr_info->data = vcpu->arch.osvw.status;
3217 case MSR_PLATFORM_INFO:
3218 if (!msr_info->host_initiated &&
3219 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3221 msr_info->data = vcpu->arch.msr_platform_info;
3223 case MSR_MISC_FEATURES_ENABLES:
3224 msr_info->data = vcpu->arch.msr_misc_features_enables;
3227 msr_info->data = vcpu->arch.msr_hwcr;
3230 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3231 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3233 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3237 if (report_ignored_msrs)
3238 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3246 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3249 * Read or write a bunch of msrs. All parameters are kernel addresses.
3251 * @return number of msrs set successfully.
3253 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3254 struct kvm_msr_entry *entries,
3255 int (*do_msr)(struct kvm_vcpu *vcpu,
3256 unsigned index, u64 *data))
3260 for (i = 0; i < msrs->nmsrs; ++i)
3261 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3268 * Read or write a bunch of msrs. Parameters are user addresses.
3270 * @return number of msrs set successfully.
3272 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3273 int (*do_msr)(struct kvm_vcpu *vcpu,
3274 unsigned index, u64 *data),
3277 struct kvm_msrs msrs;
3278 struct kvm_msr_entry *entries;
3283 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3287 if (msrs.nmsrs >= MAX_IO_MSRS)
3290 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3291 entries = memdup_user(user_msrs->entries, size);
3292 if (IS_ERR(entries)) {
3293 r = PTR_ERR(entries);
3297 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3302 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3313 static inline bool kvm_can_mwait_in_guest(void)
3315 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3316 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3317 boot_cpu_has(X86_FEATURE_ARAT);
3320 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3325 case KVM_CAP_IRQCHIP:
3327 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3328 case KVM_CAP_SET_TSS_ADDR:
3329 case KVM_CAP_EXT_CPUID:
3330 case KVM_CAP_EXT_EMUL_CPUID:
3331 case KVM_CAP_CLOCKSOURCE:
3333 case KVM_CAP_NOP_IO_DELAY:
3334 case KVM_CAP_MP_STATE:
3335 case KVM_CAP_SYNC_MMU:
3336 case KVM_CAP_USER_NMI:
3337 case KVM_CAP_REINJECT_CONTROL:
3338 case KVM_CAP_IRQ_INJECT_STATUS:
3339 case KVM_CAP_IOEVENTFD:
3340 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3342 case KVM_CAP_PIT_STATE2:
3343 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3344 case KVM_CAP_XEN_HVM:
3345 case KVM_CAP_VCPU_EVENTS:
3346 case KVM_CAP_HYPERV:
3347 case KVM_CAP_HYPERV_VAPIC:
3348 case KVM_CAP_HYPERV_SPIN:
3349 case KVM_CAP_HYPERV_SYNIC:
3350 case KVM_CAP_HYPERV_SYNIC2:
3351 case KVM_CAP_HYPERV_VP_INDEX:
3352 case KVM_CAP_HYPERV_EVENTFD:
3353 case KVM_CAP_HYPERV_TLBFLUSH:
3354 case KVM_CAP_HYPERV_SEND_IPI:
3355 case KVM_CAP_HYPERV_CPUID:
3356 case KVM_CAP_PCI_SEGMENT:
3357 case KVM_CAP_DEBUGREGS:
3358 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3360 case KVM_CAP_ASYNC_PF:
3361 case KVM_CAP_GET_TSC_KHZ:
3362 case KVM_CAP_KVMCLOCK_CTRL:
3363 case KVM_CAP_READONLY_MEM:
3364 case KVM_CAP_HYPERV_TIME:
3365 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3366 case KVM_CAP_TSC_DEADLINE_TIMER:
3367 case KVM_CAP_DISABLE_QUIRKS:
3368 case KVM_CAP_SET_BOOT_CPU_ID:
3369 case KVM_CAP_SPLIT_IRQCHIP:
3370 case KVM_CAP_IMMEDIATE_EXIT:
3371 case KVM_CAP_PMU_EVENT_FILTER:
3372 case KVM_CAP_GET_MSR_FEATURES:
3373 case KVM_CAP_MSR_PLATFORM_INFO:
3374 case KVM_CAP_EXCEPTION_PAYLOAD:
3375 case KVM_CAP_SET_GUEST_DEBUG:
3378 case KVM_CAP_SYNC_REGS:
3379 r = KVM_SYNC_X86_VALID_FIELDS;
3381 case KVM_CAP_ADJUST_CLOCK:
3382 r = KVM_CLOCK_TSC_STABLE;
3384 case KVM_CAP_X86_DISABLE_EXITS:
3385 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3386 KVM_X86_DISABLE_EXITS_CSTATE;
3387 if(kvm_can_mwait_in_guest())
3388 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3390 case KVM_CAP_X86_SMM:
3391 /* SMBASE is usually relocated above 1M on modern chipsets,
3392 * and SMM handlers might indeed rely on 4G segment limits,
3393 * so do not report SMM to be available if real mode is
3394 * emulated via vm86 mode. Still, do not go to great lengths
3395 * to avoid userspace's usage of the feature, because it is a
3396 * fringe case that is not enabled except via specific settings
3397 * of the module parameters.
3399 r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3402 r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3404 case KVM_CAP_NR_VCPUS:
3405 r = KVM_SOFT_MAX_VCPUS;
3407 case KVM_CAP_MAX_VCPUS:
3410 case KVM_CAP_MAX_VCPU_ID:
3411 r = KVM_MAX_VCPU_ID;
3413 case KVM_CAP_PV_MMU: /* obsolete */
3417 r = KVM_MAX_MCE_BANKS;
3420 r = boot_cpu_has(X86_FEATURE_XSAVE);
3422 case KVM_CAP_TSC_CONTROL:
3423 r = kvm_has_tsc_control;
3425 case KVM_CAP_X2APIC_API:
3426 r = KVM_X2APIC_API_VALID_FLAGS;
3428 case KVM_CAP_NESTED_STATE:
3429 r = kvm_x86_ops.get_nested_state ?
3430 kvm_x86_ops.get_nested_state(NULL, NULL, 0) : 0;
3432 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3433 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3435 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3436 r = kvm_x86_ops.nested_enable_evmcs != NULL;
3445 long kvm_arch_dev_ioctl(struct file *filp,
3446 unsigned int ioctl, unsigned long arg)
3448 void __user *argp = (void __user *)arg;
3452 case KVM_GET_MSR_INDEX_LIST: {
3453 struct kvm_msr_list __user *user_msr_list = argp;
3454 struct kvm_msr_list msr_list;
3458 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3461 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3462 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3465 if (n < msr_list.nmsrs)
3468 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3469 num_msrs_to_save * sizeof(u32)))
3471 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3473 num_emulated_msrs * sizeof(u32)))
3478 case KVM_GET_SUPPORTED_CPUID:
3479 case KVM_GET_EMULATED_CPUID: {
3480 struct kvm_cpuid2 __user *cpuid_arg = argp;
3481 struct kvm_cpuid2 cpuid;
3484 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3487 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3493 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3498 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3500 if (copy_to_user(argp, &kvm_mce_cap_supported,
3501 sizeof(kvm_mce_cap_supported)))
3505 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3506 struct kvm_msr_list __user *user_msr_list = argp;
3507 struct kvm_msr_list msr_list;
3511 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3514 msr_list.nmsrs = num_msr_based_features;
3515 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3518 if (n < msr_list.nmsrs)
3521 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3522 num_msr_based_features * sizeof(u32)))
3528 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3538 static void wbinvd_ipi(void *garbage)
3543 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3545 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3548 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3550 /* Address WBINVD may be executed by guest */
3551 if (need_emulate_wbinvd(vcpu)) {
3552 if (kvm_x86_ops.has_wbinvd_exit())
3553 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3554 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3555 smp_call_function_single(vcpu->cpu,
3556 wbinvd_ipi, NULL, 1);
3559 kvm_x86_ops.vcpu_load(vcpu, cpu);
3561 /* Apply any externally detected TSC adjustments (due to suspend) */
3562 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3563 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3564 vcpu->arch.tsc_offset_adjustment = 0;
3565 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3568 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3569 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3570 rdtsc() - vcpu->arch.last_host_tsc;
3572 mark_tsc_unstable("KVM discovered backwards TSC");
3574 if (kvm_check_tsc_unstable()) {
3575 u64 offset = kvm_compute_tsc_offset(vcpu,
3576 vcpu->arch.last_guest_tsc);
3577 kvm_vcpu_write_tsc_offset(vcpu, offset);
3578 vcpu->arch.tsc_catchup = 1;
3581 if (kvm_lapic_hv_timer_in_use(vcpu))
3582 kvm_lapic_restart_hv_timer(vcpu);
3585 * On a host with synchronized TSC, there is no need to update
3586 * kvmclock on vcpu->cpu migration
3588 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3589 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3590 if (vcpu->cpu != cpu)
3591 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3595 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3598 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3600 struct kvm_host_map map;
3601 struct kvm_steal_time *st;
3603 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3606 if (vcpu->arch.st.preempted)
3609 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3610 &vcpu->arch.st.cache, true))
3614 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3616 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3618 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3621 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3625 if (vcpu->preempted)
3626 vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
3629 * Disable page faults because we're in atomic context here.
3630 * kvm_write_guest_offset_cached() would call might_fault()
3631 * that relies on pagefault_disable() to tell if there's a
3632 * bug. NOTE: the write to guest memory may not go through if
3633 * during postcopy live migration or if there's heavy guest
3636 pagefault_disable();
3638 * kvm_memslots() will be called by
3639 * kvm_write_guest_offset_cached() so take the srcu lock.
3641 idx = srcu_read_lock(&vcpu->kvm->srcu);
3642 kvm_steal_time_set_preempted(vcpu);
3643 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3645 kvm_x86_ops.vcpu_put(vcpu);
3646 vcpu->arch.last_host_tsc = rdtsc();
3648 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3649 * on every vmexit, but if not, we might have a stale dr6 from the
3650 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3655 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3656 struct kvm_lapic_state *s)
3658 if (vcpu->arch.apicv_active)
3659 kvm_x86_ops.sync_pir_to_irr(vcpu);
3661 return kvm_apic_get_state(vcpu, s);
3664 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3665 struct kvm_lapic_state *s)
3669 r = kvm_apic_set_state(vcpu, s);
3672 update_cr8_intercept(vcpu);
3677 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3679 return (!lapic_in_kernel(vcpu) ||
3680 kvm_apic_accept_pic_intr(vcpu));
3684 * if userspace requested an interrupt window, check that the
3685 * interrupt window is open.
3687 * No need to exit to userspace if we already have an interrupt queued.
3689 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3691 return kvm_arch_interrupt_allowed(vcpu) &&
3692 !kvm_cpu_has_interrupt(vcpu) &&
3693 !kvm_event_needs_reinjection(vcpu) &&
3694 kvm_cpu_accept_dm_intr(vcpu);
3697 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3698 struct kvm_interrupt *irq)
3700 if (irq->irq >= KVM_NR_INTERRUPTS)
3703 if (!irqchip_in_kernel(vcpu->kvm)) {
3704 kvm_queue_interrupt(vcpu, irq->irq, false);
3705 kvm_make_request(KVM_REQ_EVENT, vcpu);
3710 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3711 * fail for in-kernel 8259.
3713 if (pic_in_kernel(vcpu->kvm))
3716 if (vcpu->arch.pending_external_vector != -1)
3719 vcpu->arch.pending_external_vector = irq->irq;
3720 kvm_make_request(KVM_REQ_EVENT, vcpu);
3724 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3726 kvm_inject_nmi(vcpu);
3731 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3733 kvm_make_request(KVM_REQ_SMI, vcpu);
3738 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3739 struct kvm_tpr_access_ctl *tac)
3743 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3747 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3751 unsigned bank_num = mcg_cap & 0xff, bank;
3754 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3756 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3759 vcpu->arch.mcg_cap = mcg_cap;
3760 /* Init IA32_MCG_CTL to all 1s */
3761 if (mcg_cap & MCG_CTL_P)
3762 vcpu->arch.mcg_ctl = ~(u64)0;
3763 /* Init IA32_MCi_CTL to all 1s */
3764 for (bank = 0; bank < bank_num; bank++)
3765 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3767 kvm_x86_ops.setup_mce(vcpu);
3772 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3773 struct kvm_x86_mce *mce)
3775 u64 mcg_cap = vcpu->arch.mcg_cap;
3776 unsigned bank_num = mcg_cap & 0xff;
3777 u64 *banks = vcpu->arch.mce_banks;
3779 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3782 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3783 * reporting is disabled
3785 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3786 vcpu->arch.mcg_ctl != ~(u64)0)
3788 banks += 4 * mce->bank;
3790 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3791 * reporting is disabled for the bank
3793 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3795 if (mce->status & MCI_STATUS_UC) {
3796 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3797 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3798 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3801 if (banks[1] & MCI_STATUS_VAL)
3802 mce->status |= MCI_STATUS_OVER;
3803 banks[2] = mce->addr;
3804 banks[3] = mce->misc;
3805 vcpu->arch.mcg_status = mce->mcg_status;
3806 banks[1] = mce->status;
3807 kvm_queue_exception(vcpu, MC_VECTOR);
3808 } else if (!(banks[1] & MCI_STATUS_VAL)
3809 || !(banks[1] & MCI_STATUS_UC)) {
3810 if (banks[1] & MCI_STATUS_VAL)
3811 mce->status |= MCI_STATUS_OVER;
3812 banks[2] = mce->addr;
3813 banks[3] = mce->misc;
3814 banks[1] = mce->status;
3816 banks[1] |= MCI_STATUS_OVER;
3820 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3821 struct kvm_vcpu_events *events)
3826 * In guest mode, payload delivery should be deferred,
3827 * so that the L1 hypervisor can intercept #PF before
3828 * CR2 is modified (or intercept #DB before DR6 is
3829 * modified under nVMX). Unless the per-VM capability,
3830 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
3831 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
3832 * opportunistically defer the exception payload, deliver it if the
3833 * capability hasn't been requested before processing a
3834 * KVM_GET_VCPU_EVENTS.
3836 if (!vcpu->kvm->arch.exception_payload_enabled &&
3837 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
3838 kvm_deliver_exception_payload(vcpu);
3841 * The API doesn't provide the instruction length for software
3842 * exceptions, so don't report them. As long as the guest RIP
3843 * isn't advanced, we should expect to encounter the exception
3846 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3847 events->exception.injected = 0;
3848 events->exception.pending = 0;
3850 events->exception.injected = vcpu->arch.exception.injected;
3851 events->exception.pending = vcpu->arch.exception.pending;
3853 * For ABI compatibility, deliberately conflate
3854 * pending and injected exceptions when
3855 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3857 if (!vcpu->kvm->arch.exception_payload_enabled)
3858 events->exception.injected |=
3859 vcpu->arch.exception.pending;
3861 events->exception.nr = vcpu->arch.exception.nr;
3862 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3863 events->exception.error_code = vcpu->arch.exception.error_code;
3864 events->exception_has_payload = vcpu->arch.exception.has_payload;
3865 events->exception_payload = vcpu->arch.exception.payload;
3867 events->interrupt.injected =
3868 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3869 events->interrupt.nr = vcpu->arch.interrupt.nr;
3870 events->interrupt.soft = 0;
3871 events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
3873 events->nmi.injected = vcpu->arch.nmi_injected;
3874 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3875 events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
3876 events->nmi.pad = 0;
3878 events->sipi_vector = 0; /* never valid when reporting to user space */
3880 events->smi.smm = is_smm(vcpu);
3881 events->smi.pending = vcpu->arch.smi_pending;
3882 events->smi.smm_inside_nmi =
3883 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3884 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3886 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3887 | KVM_VCPUEVENT_VALID_SHADOW
3888 | KVM_VCPUEVENT_VALID_SMM);
3889 if (vcpu->kvm->arch.exception_payload_enabled)
3890 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3892 memset(&events->reserved, 0, sizeof(events->reserved));
3895 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3897 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3898 struct kvm_vcpu_events *events)
3900 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3901 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3902 | KVM_VCPUEVENT_VALID_SHADOW
3903 | KVM_VCPUEVENT_VALID_SMM
3904 | KVM_VCPUEVENT_VALID_PAYLOAD))
3907 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3908 if (!vcpu->kvm->arch.exception_payload_enabled)
3910 if (events->exception.pending)
3911 events->exception.injected = 0;
3913 events->exception_has_payload = 0;
3915 events->exception.pending = 0;
3916 events->exception_has_payload = 0;
3919 if ((events->exception.injected || events->exception.pending) &&
3920 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3923 /* INITs are latched while in SMM */
3924 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3925 (events->smi.smm || events->smi.pending) &&
3926 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3930 vcpu->arch.exception.injected = events->exception.injected;
3931 vcpu->arch.exception.pending = events->exception.pending;
3932 vcpu->arch.exception.nr = events->exception.nr;
3933 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3934 vcpu->arch.exception.error_code = events->exception.error_code;
3935 vcpu->arch.exception.has_payload = events->exception_has_payload;
3936 vcpu->arch.exception.payload = events->exception_payload;
3938 vcpu->arch.interrupt.injected = events->interrupt.injected;
3939 vcpu->arch.interrupt.nr = events->interrupt.nr;
3940 vcpu->arch.interrupt.soft = events->interrupt.soft;
3941 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3942 kvm_x86_ops.set_interrupt_shadow(vcpu,
3943 events->interrupt.shadow);
3945 vcpu->arch.nmi_injected = events->nmi.injected;
3946 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3947 vcpu->arch.nmi_pending = events->nmi.pending;
3948 kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
3950 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3951 lapic_in_kernel(vcpu))
3952 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3954 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3955 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3956 if (events->smi.smm)
3957 vcpu->arch.hflags |= HF_SMM_MASK;
3959 vcpu->arch.hflags &= ~HF_SMM_MASK;
3960 kvm_smm_changed(vcpu);
3963 vcpu->arch.smi_pending = events->smi.pending;
3965 if (events->smi.smm) {
3966 if (events->smi.smm_inside_nmi)
3967 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3969 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3972 if (lapic_in_kernel(vcpu)) {
3973 if (events->smi.latched_init)
3974 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3976 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3980 kvm_make_request(KVM_REQ_EVENT, vcpu);
3985 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3986 struct kvm_debugregs *dbgregs)
3990 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3991 kvm_get_dr(vcpu, 6, &val);
3993 dbgregs->dr7 = vcpu->arch.dr7;
3995 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3998 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3999 struct kvm_debugregs *dbgregs)
4004 if (dbgregs->dr6 & ~0xffffffffull)
4006 if (dbgregs->dr7 & ~0xffffffffull)
4009 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4010 kvm_update_dr0123(vcpu);
4011 vcpu->arch.dr6 = dbgregs->dr6;
4012 kvm_update_dr6(vcpu);
4013 vcpu->arch.dr7 = dbgregs->dr7;
4014 kvm_update_dr7(vcpu);
4019 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4021 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4023 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4024 u64 xstate_bv = xsave->header.xfeatures;
4028 * Copy legacy XSAVE area, to avoid complications with CPUID
4029 * leaves 0 and 1 in the loop below.
4031 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4034 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4035 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4038 * Copy each region from the possibly compacted offset to the
4039 * non-compacted offset.
4041 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4043 u64 xfeature_mask = valid & -valid;
4044 int xfeature_nr = fls64(xfeature_mask) - 1;
4045 void *src = get_xsave_addr(xsave, xfeature_nr);
4048 u32 size, offset, ecx, edx;
4049 cpuid_count(XSTATE_CPUID, xfeature_nr,
4050 &size, &offset, &ecx, &edx);
4051 if (xfeature_nr == XFEATURE_PKRU)
4052 memcpy(dest + offset, &vcpu->arch.pkru,
4053 sizeof(vcpu->arch.pkru));
4055 memcpy(dest + offset, src, size);
4059 valid -= xfeature_mask;
4063 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4065 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4066 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4070 * Copy legacy XSAVE area, to avoid complications with CPUID
4071 * leaves 0 and 1 in the loop below.
4073 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4075 /* Set XSTATE_BV and possibly XCOMP_BV. */
4076 xsave->header.xfeatures = xstate_bv;
4077 if (boot_cpu_has(X86_FEATURE_XSAVES))
4078 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4081 * Copy each region from the non-compacted offset to the
4082 * possibly compacted offset.
4084 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4086 u64 xfeature_mask = valid & -valid;
4087 int xfeature_nr = fls64(xfeature_mask) - 1;
4088 void *dest = get_xsave_addr(xsave, xfeature_nr);
4091 u32 size, offset, ecx, edx;
4092 cpuid_count(XSTATE_CPUID, xfeature_nr,
4093 &size, &offset, &ecx, &edx);
4094 if (xfeature_nr == XFEATURE_PKRU)
4095 memcpy(&vcpu->arch.pkru, src + offset,
4096 sizeof(vcpu->arch.pkru));
4098 memcpy(dest, src + offset, size);
4101 valid -= xfeature_mask;
4105 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4106 struct kvm_xsave *guest_xsave)
4108 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4109 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4110 fill_xsave((u8 *) guest_xsave->region, vcpu);
4112 memcpy(guest_xsave->region,
4113 &vcpu->arch.guest_fpu->state.fxsave,
4114 sizeof(struct fxregs_state));
4115 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4116 XFEATURE_MASK_FPSSE;
4120 #define XSAVE_MXCSR_OFFSET 24
4122 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4123 struct kvm_xsave *guest_xsave)
4126 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4127 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4129 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4131 * Here we allow setting states that are not present in
4132 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4133 * with old userspace.
4135 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4137 load_xsave(vcpu, (u8 *)guest_xsave->region);
4139 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4140 mxcsr & ~mxcsr_feature_mask)
4142 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4143 guest_xsave->region, sizeof(struct fxregs_state));
4148 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4149 struct kvm_xcrs *guest_xcrs)
4151 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4152 guest_xcrs->nr_xcrs = 0;
4156 guest_xcrs->nr_xcrs = 1;
4157 guest_xcrs->flags = 0;
4158 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4159 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4162 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4163 struct kvm_xcrs *guest_xcrs)
4167 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4170 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4173 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4174 /* Only support XCR0 currently */
4175 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4176 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4177 guest_xcrs->xcrs[i].value);
4186 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4187 * stopped by the hypervisor. This function will be called from the host only.
4188 * EINVAL is returned when the host attempts to set the flag for a guest that
4189 * does not support pv clocks.
4191 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4193 if (!vcpu->arch.pv_time_enabled)
4195 vcpu->arch.pvclock_set_guest_stopped_request = true;
4196 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4200 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4201 struct kvm_enable_cap *cap)
4204 uint16_t vmcs_version;
4205 void __user *user_ptr;
4211 case KVM_CAP_HYPERV_SYNIC2:
4216 case KVM_CAP_HYPERV_SYNIC:
4217 if (!irqchip_in_kernel(vcpu->kvm))
4219 return kvm_hv_activate_synic(vcpu, cap->cap ==
4220 KVM_CAP_HYPERV_SYNIC2);
4221 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4222 if (!kvm_x86_ops.nested_enable_evmcs)
4224 r = kvm_x86_ops.nested_enable_evmcs(vcpu, &vmcs_version);
4226 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4227 if (copy_to_user(user_ptr, &vmcs_version,
4228 sizeof(vmcs_version)))
4232 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4233 if (!kvm_x86_ops.enable_direct_tlbflush)
4236 return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4243 long kvm_arch_vcpu_ioctl(struct file *filp,
4244 unsigned int ioctl, unsigned long arg)
4246 struct kvm_vcpu *vcpu = filp->private_data;
4247 void __user *argp = (void __user *)arg;
4250 struct kvm_lapic_state *lapic;
4251 struct kvm_xsave *xsave;
4252 struct kvm_xcrs *xcrs;
4260 case KVM_GET_LAPIC: {
4262 if (!lapic_in_kernel(vcpu))
4264 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4265 GFP_KERNEL_ACCOUNT);
4270 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4274 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4279 case KVM_SET_LAPIC: {
4281 if (!lapic_in_kernel(vcpu))
4283 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4284 if (IS_ERR(u.lapic)) {
4285 r = PTR_ERR(u.lapic);
4289 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4292 case KVM_INTERRUPT: {
4293 struct kvm_interrupt irq;
4296 if (copy_from_user(&irq, argp, sizeof(irq)))
4298 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4302 r = kvm_vcpu_ioctl_nmi(vcpu);
4306 r = kvm_vcpu_ioctl_smi(vcpu);
4309 case KVM_SET_CPUID: {
4310 struct kvm_cpuid __user *cpuid_arg = argp;
4311 struct kvm_cpuid cpuid;
4314 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4316 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4319 case KVM_SET_CPUID2: {
4320 struct kvm_cpuid2 __user *cpuid_arg = argp;
4321 struct kvm_cpuid2 cpuid;
4324 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4326 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4327 cpuid_arg->entries);
4330 case KVM_GET_CPUID2: {
4331 struct kvm_cpuid2 __user *cpuid_arg = argp;
4332 struct kvm_cpuid2 cpuid;
4335 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4337 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4338 cpuid_arg->entries);
4342 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4347 case KVM_GET_MSRS: {
4348 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4349 r = msr_io(vcpu, argp, do_get_msr, 1);
4350 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4353 case KVM_SET_MSRS: {
4354 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4355 r = msr_io(vcpu, argp, do_set_msr, 0);
4356 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4359 case KVM_TPR_ACCESS_REPORTING: {
4360 struct kvm_tpr_access_ctl tac;
4363 if (copy_from_user(&tac, argp, sizeof(tac)))
4365 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4369 if (copy_to_user(argp, &tac, sizeof(tac)))
4374 case KVM_SET_VAPIC_ADDR: {
4375 struct kvm_vapic_addr va;
4379 if (!lapic_in_kernel(vcpu))
4382 if (copy_from_user(&va, argp, sizeof(va)))
4384 idx = srcu_read_lock(&vcpu->kvm->srcu);
4385 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4386 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4389 case KVM_X86_SETUP_MCE: {
4393 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4395 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4398 case KVM_X86_SET_MCE: {
4399 struct kvm_x86_mce mce;
4402 if (copy_from_user(&mce, argp, sizeof(mce)))
4404 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4407 case KVM_GET_VCPU_EVENTS: {
4408 struct kvm_vcpu_events events;
4410 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4413 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4418 case KVM_SET_VCPU_EVENTS: {
4419 struct kvm_vcpu_events events;
4422 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4425 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4428 case KVM_GET_DEBUGREGS: {
4429 struct kvm_debugregs dbgregs;
4431 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4434 if (copy_to_user(argp, &dbgregs,
4435 sizeof(struct kvm_debugregs)))
4440 case KVM_SET_DEBUGREGS: {
4441 struct kvm_debugregs dbgregs;
4444 if (copy_from_user(&dbgregs, argp,
4445 sizeof(struct kvm_debugregs)))
4448 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4451 case KVM_GET_XSAVE: {
4452 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4457 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4460 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4465 case KVM_SET_XSAVE: {
4466 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4467 if (IS_ERR(u.xsave)) {
4468 r = PTR_ERR(u.xsave);
4472 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4475 case KVM_GET_XCRS: {
4476 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4481 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4484 if (copy_to_user(argp, u.xcrs,
4485 sizeof(struct kvm_xcrs)))
4490 case KVM_SET_XCRS: {
4491 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4492 if (IS_ERR(u.xcrs)) {
4493 r = PTR_ERR(u.xcrs);
4497 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4500 case KVM_SET_TSC_KHZ: {
4504 user_tsc_khz = (u32)arg;
4506 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4509 if (user_tsc_khz == 0)
4510 user_tsc_khz = tsc_khz;
4512 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4517 case KVM_GET_TSC_KHZ: {
4518 r = vcpu->arch.virtual_tsc_khz;
4521 case KVM_KVMCLOCK_CTRL: {
4522 r = kvm_set_guest_paused(vcpu);
4525 case KVM_ENABLE_CAP: {
4526 struct kvm_enable_cap cap;
4529 if (copy_from_user(&cap, argp, sizeof(cap)))
4531 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4534 case KVM_GET_NESTED_STATE: {
4535 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4539 if (!kvm_x86_ops.get_nested_state)
4542 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4544 if (get_user(user_data_size, &user_kvm_nested_state->size))
4547 r = kvm_x86_ops.get_nested_state(vcpu, user_kvm_nested_state,
4552 if (r > user_data_size) {
4553 if (put_user(r, &user_kvm_nested_state->size))
4563 case KVM_SET_NESTED_STATE: {
4564 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4565 struct kvm_nested_state kvm_state;
4569 if (!kvm_x86_ops.set_nested_state)
4573 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4577 if (kvm_state.size < sizeof(kvm_state))
4580 if (kvm_state.flags &
4581 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4582 | KVM_STATE_NESTED_EVMCS))
4585 /* nested_run_pending implies guest_mode. */
4586 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4587 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4590 idx = srcu_read_lock(&vcpu->kvm->srcu);
4591 r = kvm_x86_ops.set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4592 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4595 case KVM_GET_SUPPORTED_HV_CPUID: {
4596 struct kvm_cpuid2 __user *cpuid_arg = argp;
4597 struct kvm_cpuid2 cpuid;
4600 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4603 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4604 cpuid_arg->entries);
4609 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4624 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4626 return VM_FAULT_SIGBUS;
4629 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4633 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4635 ret = kvm_x86_ops.set_tss_addr(kvm, addr);
4639 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4642 return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
4645 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4646 unsigned long kvm_nr_mmu_pages)
4648 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4651 mutex_lock(&kvm->slots_lock);
4653 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4654 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4656 mutex_unlock(&kvm->slots_lock);
4660 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4662 return kvm->arch.n_max_mmu_pages;
4665 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4667 struct kvm_pic *pic = kvm->arch.vpic;
4671 switch (chip->chip_id) {
4672 case KVM_IRQCHIP_PIC_MASTER:
4673 memcpy(&chip->chip.pic, &pic->pics[0],
4674 sizeof(struct kvm_pic_state));
4676 case KVM_IRQCHIP_PIC_SLAVE:
4677 memcpy(&chip->chip.pic, &pic->pics[1],
4678 sizeof(struct kvm_pic_state));
4680 case KVM_IRQCHIP_IOAPIC:
4681 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4690 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4692 struct kvm_pic *pic = kvm->arch.vpic;
4696 switch (chip->chip_id) {
4697 case KVM_IRQCHIP_PIC_MASTER:
4698 spin_lock(&pic->lock);
4699 memcpy(&pic->pics[0], &chip->chip.pic,
4700 sizeof(struct kvm_pic_state));
4701 spin_unlock(&pic->lock);
4703 case KVM_IRQCHIP_PIC_SLAVE:
4704 spin_lock(&pic->lock);
4705 memcpy(&pic->pics[1], &chip->chip.pic,
4706 sizeof(struct kvm_pic_state));
4707 spin_unlock(&pic->lock);
4709 case KVM_IRQCHIP_IOAPIC:
4710 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4716 kvm_pic_update_irq(pic);
4720 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4722 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4724 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4726 mutex_lock(&kps->lock);
4727 memcpy(ps, &kps->channels, sizeof(*ps));
4728 mutex_unlock(&kps->lock);
4732 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4735 struct kvm_pit *pit = kvm->arch.vpit;
4737 mutex_lock(&pit->pit_state.lock);
4738 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4739 for (i = 0; i < 3; i++)
4740 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4741 mutex_unlock(&pit->pit_state.lock);
4745 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4747 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4748 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4749 sizeof(ps->channels));
4750 ps->flags = kvm->arch.vpit->pit_state.flags;
4751 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4752 memset(&ps->reserved, 0, sizeof(ps->reserved));
4756 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4760 u32 prev_legacy, cur_legacy;
4761 struct kvm_pit *pit = kvm->arch.vpit;
4763 mutex_lock(&pit->pit_state.lock);
4764 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4765 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4766 if (!prev_legacy && cur_legacy)
4768 memcpy(&pit->pit_state.channels, &ps->channels,
4769 sizeof(pit->pit_state.channels));
4770 pit->pit_state.flags = ps->flags;
4771 for (i = 0; i < 3; i++)
4772 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4774 mutex_unlock(&pit->pit_state.lock);
4778 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4779 struct kvm_reinject_control *control)
4781 struct kvm_pit *pit = kvm->arch.vpit;
4783 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4784 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4785 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4787 mutex_lock(&pit->pit_state.lock);
4788 kvm_pit_set_reinject(pit, control->pit_reinject);
4789 mutex_unlock(&pit->pit_state.lock);
4794 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
4797 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4799 if (kvm_x86_ops.flush_log_dirty)
4800 kvm_x86_ops.flush_log_dirty(kvm);
4803 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4806 if (!irqchip_in_kernel(kvm))
4809 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4810 irq_event->irq, irq_event->level,
4815 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4816 struct kvm_enable_cap *cap)
4824 case KVM_CAP_DISABLE_QUIRKS:
4825 kvm->arch.disabled_quirks = cap->args[0];
4828 case KVM_CAP_SPLIT_IRQCHIP: {
4829 mutex_lock(&kvm->lock);
4831 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4832 goto split_irqchip_unlock;
4834 if (irqchip_in_kernel(kvm))
4835 goto split_irqchip_unlock;
4836 if (kvm->created_vcpus)
4837 goto split_irqchip_unlock;
4838 r = kvm_setup_empty_irq_routing(kvm);
4840 goto split_irqchip_unlock;
4841 /* Pairs with irqchip_in_kernel. */
4843 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4844 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4846 split_irqchip_unlock:
4847 mutex_unlock(&kvm->lock);
4850 case KVM_CAP_X2APIC_API:
4852 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4855 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4856 kvm->arch.x2apic_format = true;
4857 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4858 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4862 case KVM_CAP_X86_DISABLE_EXITS:
4864 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4867 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4868 kvm_can_mwait_in_guest())
4869 kvm->arch.mwait_in_guest = true;
4870 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4871 kvm->arch.hlt_in_guest = true;
4872 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4873 kvm->arch.pause_in_guest = true;
4874 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4875 kvm->arch.cstate_in_guest = true;
4878 case KVM_CAP_MSR_PLATFORM_INFO:
4879 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4882 case KVM_CAP_EXCEPTION_PAYLOAD:
4883 kvm->arch.exception_payload_enabled = cap->args[0];
4893 long kvm_arch_vm_ioctl(struct file *filp,
4894 unsigned int ioctl, unsigned long arg)
4896 struct kvm *kvm = filp->private_data;
4897 void __user *argp = (void __user *)arg;
4900 * This union makes it completely explicit to gcc-3.x
4901 * that these two variables' stack usage should be
4902 * combined, not added together.
4905 struct kvm_pit_state ps;
4906 struct kvm_pit_state2 ps2;
4907 struct kvm_pit_config pit_config;
4911 case KVM_SET_TSS_ADDR:
4912 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4914 case KVM_SET_IDENTITY_MAP_ADDR: {
4917 mutex_lock(&kvm->lock);
4919 if (kvm->created_vcpus)
4920 goto set_identity_unlock;
4922 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4923 goto set_identity_unlock;
4924 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4925 set_identity_unlock:
4926 mutex_unlock(&kvm->lock);
4929 case KVM_SET_NR_MMU_PAGES:
4930 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4932 case KVM_GET_NR_MMU_PAGES:
4933 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4935 case KVM_CREATE_IRQCHIP: {
4936 mutex_lock(&kvm->lock);
4939 if (irqchip_in_kernel(kvm))
4940 goto create_irqchip_unlock;
4943 if (kvm->created_vcpus)
4944 goto create_irqchip_unlock;
4946 r = kvm_pic_init(kvm);
4948 goto create_irqchip_unlock;
4950 r = kvm_ioapic_init(kvm);
4952 kvm_pic_destroy(kvm);
4953 goto create_irqchip_unlock;
4956 r = kvm_setup_default_irq_routing(kvm);
4958 kvm_ioapic_destroy(kvm);
4959 kvm_pic_destroy(kvm);
4960 goto create_irqchip_unlock;
4962 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4964 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4965 create_irqchip_unlock:
4966 mutex_unlock(&kvm->lock);
4969 case KVM_CREATE_PIT:
4970 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4972 case KVM_CREATE_PIT2:
4974 if (copy_from_user(&u.pit_config, argp,
4975 sizeof(struct kvm_pit_config)))
4978 mutex_lock(&kvm->lock);
4981 goto create_pit_unlock;
4983 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4987 mutex_unlock(&kvm->lock);
4989 case KVM_GET_IRQCHIP: {
4990 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4991 struct kvm_irqchip *chip;
4993 chip = memdup_user(argp, sizeof(*chip));
5000 if (!irqchip_kernel(kvm))
5001 goto get_irqchip_out;
5002 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5004 goto get_irqchip_out;
5006 if (copy_to_user(argp, chip, sizeof(*chip)))
5007 goto get_irqchip_out;
5013 case KVM_SET_IRQCHIP: {
5014 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5015 struct kvm_irqchip *chip;
5017 chip = memdup_user(argp, sizeof(*chip));
5024 if (!irqchip_kernel(kvm))
5025 goto set_irqchip_out;
5026 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5033 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5036 if (!kvm->arch.vpit)
5038 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5042 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5049 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5051 mutex_lock(&kvm->lock);
5053 if (!kvm->arch.vpit)
5055 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5057 mutex_unlock(&kvm->lock);
5060 case KVM_GET_PIT2: {
5062 if (!kvm->arch.vpit)
5064 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5068 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5073 case KVM_SET_PIT2: {
5075 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5077 mutex_lock(&kvm->lock);
5079 if (!kvm->arch.vpit)
5081 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5083 mutex_unlock(&kvm->lock);
5086 case KVM_REINJECT_CONTROL: {
5087 struct kvm_reinject_control control;
5089 if (copy_from_user(&control, argp, sizeof(control)))
5092 if (!kvm->arch.vpit)
5094 r = kvm_vm_ioctl_reinject(kvm, &control);
5097 case KVM_SET_BOOT_CPU_ID:
5099 mutex_lock(&kvm->lock);
5100 if (kvm->created_vcpus)
5103 kvm->arch.bsp_vcpu_id = arg;
5104 mutex_unlock(&kvm->lock);
5106 case KVM_XEN_HVM_CONFIG: {
5107 struct kvm_xen_hvm_config xhc;
5109 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5114 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5118 case KVM_SET_CLOCK: {
5119 struct kvm_clock_data user_ns;
5123 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5132 * TODO: userspace has to take care of races with VCPU_RUN, so
5133 * kvm_gen_update_masterclock() can be cut down to locked
5134 * pvclock_update_vm_gtod_copy().
5136 kvm_gen_update_masterclock(kvm);
5137 now_ns = get_kvmclock_ns(kvm);
5138 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5139 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5142 case KVM_GET_CLOCK: {
5143 struct kvm_clock_data user_ns;
5146 now_ns = get_kvmclock_ns(kvm);
5147 user_ns.clock = now_ns;
5148 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5149 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5152 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5157 case KVM_MEMORY_ENCRYPT_OP: {
5159 if (kvm_x86_ops.mem_enc_op)
5160 r = kvm_x86_ops.mem_enc_op(kvm, argp);
5163 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5164 struct kvm_enc_region region;
5167 if (copy_from_user(®ion, argp, sizeof(region)))
5171 if (kvm_x86_ops.mem_enc_reg_region)
5172 r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion);
5175 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5176 struct kvm_enc_region region;
5179 if (copy_from_user(®ion, argp, sizeof(region)))
5183 if (kvm_x86_ops.mem_enc_unreg_region)
5184 r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion);
5187 case KVM_HYPERV_EVENTFD: {
5188 struct kvm_hyperv_eventfd hvevfd;
5191 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5193 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5196 case KVM_SET_PMU_EVENT_FILTER:
5197 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5206 static void kvm_init_msr_list(void)
5208 struct x86_pmu_capability x86_pmu;
5212 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5213 "Please update the fixed PMCs in msrs_to_saved_all[]");
5215 perf_get_x86_pmu_capability(&x86_pmu);
5217 num_msrs_to_save = 0;
5218 num_emulated_msrs = 0;
5219 num_msr_based_features = 0;
5221 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5222 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5226 * Even MSRs that are valid in the host may not be exposed
5227 * to the guests in some cases.
5229 switch (msrs_to_save_all[i]) {
5230 case MSR_IA32_BNDCFGS:
5231 if (!kvm_mpx_supported())
5235 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5238 case MSR_IA32_RTIT_CTL:
5239 case MSR_IA32_RTIT_STATUS:
5240 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5243 case MSR_IA32_RTIT_CR3_MATCH:
5244 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5245 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5248 case MSR_IA32_RTIT_OUTPUT_BASE:
5249 case MSR_IA32_RTIT_OUTPUT_MASK:
5250 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5251 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5252 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5255 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5256 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5257 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5258 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5261 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5262 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5263 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5266 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5267 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5268 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5275 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5278 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5279 if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5282 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5285 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5286 struct kvm_msr_entry msr;
5288 msr.index = msr_based_features_all[i];
5289 if (kvm_get_msr_feature(&msr))
5292 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5296 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5304 if (!(lapic_in_kernel(vcpu) &&
5305 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5306 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5317 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5324 if (!(lapic_in_kernel(vcpu) &&
5325 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5327 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5329 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5339 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5340 struct kvm_segment *var, int seg)
5342 kvm_x86_ops.set_segment(vcpu, var, seg);
5345 void kvm_get_segment(struct kvm_vcpu *vcpu,
5346 struct kvm_segment *var, int seg)
5348 kvm_x86_ops.get_segment(vcpu, var, seg);
5351 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5352 struct x86_exception *exception)
5356 BUG_ON(!mmu_is_nested(vcpu));
5358 /* NPT walks are always user-walks */
5359 access |= PFERR_USER_MASK;
5360 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5365 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5366 struct x86_exception *exception)
5368 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5369 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5372 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5373 struct x86_exception *exception)
5375 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5376 access |= PFERR_FETCH_MASK;
5377 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5380 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5381 struct x86_exception *exception)
5383 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5384 access |= PFERR_WRITE_MASK;
5385 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5388 /* uses this to access any guest's mapped memory without checking CPL */
5389 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5390 struct x86_exception *exception)
5392 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5395 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5396 struct kvm_vcpu *vcpu, u32 access,
5397 struct x86_exception *exception)
5400 int r = X86EMUL_CONTINUE;
5403 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5405 unsigned offset = addr & (PAGE_SIZE-1);
5406 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5409 if (gpa == UNMAPPED_GVA)
5410 return X86EMUL_PROPAGATE_FAULT;
5411 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5414 r = X86EMUL_IO_NEEDED;
5426 /* used for instruction fetching */
5427 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5428 gva_t addr, void *val, unsigned int bytes,
5429 struct x86_exception *exception)
5431 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5432 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5436 /* Inline kvm_read_guest_virt_helper for speed. */
5437 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5439 if (unlikely(gpa == UNMAPPED_GVA))
5440 return X86EMUL_PROPAGATE_FAULT;
5442 offset = addr & (PAGE_SIZE-1);
5443 if (WARN_ON(offset + bytes > PAGE_SIZE))
5444 bytes = (unsigned)PAGE_SIZE - offset;
5445 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5447 if (unlikely(ret < 0))
5448 return X86EMUL_IO_NEEDED;
5450 return X86EMUL_CONTINUE;
5453 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5454 gva_t addr, void *val, unsigned int bytes,
5455 struct x86_exception *exception)
5457 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5460 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5461 * is returned, but our callers are not ready for that and they blindly
5462 * call kvm_inject_page_fault. Ensure that they at least do not leak
5463 * uninitialized kernel stack memory into cr2 and error code.
5465 memset(exception, 0, sizeof(*exception));
5466 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5469 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5471 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5472 gva_t addr, void *val, unsigned int bytes,
5473 struct x86_exception *exception, bool system)
5475 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5478 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5479 access |= PFERR_USER_MASK;
5481 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5484 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5485 unsigned long addr, void *val, unsigned int bytes)
5487 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5488 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5490 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5493 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5494 struct kvm_vcpu *vcpu, u32 access,
5495 struct x86_exception *exception)
5498 int r = X86EMUL_CONTINUE;
5501 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5504 unsigned offset = addr & (PAGE_SIZE-1);
5505 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5508 if (gpa == UNMAPPED_GVA)
5509 return X86EMUL_PROPAGATE_FAULT;
5510 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5512 r = X86EMUL_IO_NEEDED;
5524 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5525 unsigned int bytes, struct x86_exception *exception,
5528 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5529 u32 access = PFERR_WRITE_MASK;
5531 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5532 access |= PFERR_USER_MASK;
5534 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5538 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5539 unsigned int bytes, struct x86_exception *exception)
5541 /* kvm_write_guest_virt_system can pull in tons of pages. */
5542 vcpu->arch.l1tf_flush_l1d = true;
5545 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5546 * is returned, but our callers are not ready for that and they blindly
5547 * call kvm_inject_page_fault. Ensure that they at least do not leak
5548 * uninitialized kernel stack memory into cr2 and error code.
5550 memset(exception, 0, sizeof(*exception));
5551 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5552 PFERR_WRITE_MASK, exception);
5554 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5556 int handle_ud(struct kvm_vcpu *vcpu)
5558 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5559 int emul_type = EMULTYPE_TRAP_UD;
5560 char sig[5]; /* ud2; .ascii "kvm" */
5561 struct x86_exception e;
5563 if (force_emulation_prefix &&
5564 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5565 sig, sizeof(sig), &e) == 0 &&
5566 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5567 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5568 emul_type = EMULTYPE_TRAP_UD_FORCED;
5571 return kvm_emulate_instruction(vcpu, emul_type);
5573 EXPORT_SYMBOL_GPL(handle_ud);
5575 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5576 gpa_t gpa, bool write)
5578 /* For APIC access vmexit */
5579 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5582 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5583 trace_vcpu_match_mmio(gva, gpa, write, true);
5590 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5591 gpa_t *gpa, struct x86_exception *exception,
5594 u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5595 | (write ? PFERR_WRITE_MASK : 0);
5598 * currently PKRU is only applied to ept enabled guest so
5599 * there is no pkey in EPT page table for L1 guest or EPT
5600 * shadow page table for L2 guest.
5602 if (vcpu_match_mmio_gva(vcpu, gva)
5603 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5604 vcpu->arch.mmio_access, 0, access)) {
5605 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5606 (gva & (PAGE_SIZE - 1));
5607 trace_vcpu_match_mmio(gva, *gpa, write, false);
5611 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5613 if (*gpa == UNMAPPED_GVA)
5616 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5619 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5620 const void *val, int bytes)
5624 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5627 kvm_page_track_write(vcpu, gpa, val, bytes);
5631 struct read_write_emulator_ops {
5632 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5634 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5635 void *val, int bytes);
5636 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5637 int bytes, void *val);
5638 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5639 void *val, int bytes);
5643 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5645 if (vcpu->mmio_read_completed) {
5646 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5647 vcpu->mmio_fragments[0].gpa, val);
5648 vcpu->mmio_read_completed = 0;
5655 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5656 void *val, int bytes)
5658 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5661 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5662 void *val, int bytes)
5664 return emulator_write_phys(vcpu, gpa, val, bytes);
5667 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5669 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5670 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5673 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5674 void *val, int bytes)
5676 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5677 return X86EMUL_IO_NEEDED;
5680 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5681 void *val, int bytes)
5683 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5685 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5686 return X86EMUL_CONTINUE;
5689 static const struct read_write_emulator_ops read_emultor = {
5690 .read_write_prepare = read_prepare,
5691 .read_write_emulate = read_emulate,
5692 .read_write_mmio = vcpu_mmio_read,
5693 .read_write_exit_mmio = read_exit_mmio,
5696 static const struct read_write_emulator_ops write_emultor = {
5697 .read_write_emulate = write_emulate,
5698 .read_write_mmio = write_mmio,
5699 .read_write_exit_mmio = write_exit_mmio,
5703 static int emulator_read_write_onepage(unsigned long addr, void *val,
5705 struct x86_exception *exception,
5706 struct kvm_vcpu *vcpu,
5707 const struct read_write_emulator_ops *ops)
5711 bool write = ops->write;
5712 struct kvm_mmio_fragment *frag;
5713 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
5716 * If the exit was due to a NPF we may already have a GPA.
5717 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5718 * Note, this cannot be used on string operations since string
5719 * operation using rep will only have the initial GPA from the NPF
5722 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
5723 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
5724 gpa = ctxt->gpa_val;
5725 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5727 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5729 return X86EMUL_PROPAGATE_FAULT;
5732 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5733 return X86EMUL_CONTINUE;
5736 * Is this MMIO handled locally?
5738 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5739 if (handled == bytes)
5740 return X86EMUL_CONTINUE;
5746 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5747 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5751 return X86EMUL_CONTINUE;
5754 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5756 void *val, unsigned int bytes,
5757 struct x86_exception *exception,
5758 const struct read_write_emulator_ops *ops)
5760 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5764 if (ops->read_write_prepare &&
5765 ops->read_write_prepare(vcpu, val, bytes))
5766 return X86EMUL_CONTINUE;
5768 vcpu->mmio_nr_fragments = 0;
5770 /* Crossing a page boundary? */
5771 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5774 now = -addr & ~PAGE_MASK;
5775 rc = emulator_read_write_onepage(addr, val, now, exception,
5778 if (rc != X86EMUL_CONTINUE)
5781 if (ctxt->mode != X86EMUL_MODE_PROT64)
5787 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5789 if (rc != X86EMUL_CONTINUE)
5792 if (!vcpu->mmio_nr_fragments)
5795 gpa = vcpu->mmio_fragments[0].gpa;
5797 vcpu->mmio_needed = 1;
5798 vcpu->mmio_cur_fragment = 0;
5800 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5801 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5802 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5803 vcpu->run->mmio.phys_addr = gpa;
5805 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5808 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5812 struct x86_exception *exception)
5814 return emulator_read_write(ctxt, addr, val, bytes,
5815 exception, &read_emultor);
5818 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5822 struct x86_exception *exception)
5824 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5825 exception, &write_emultor);
5828 #define CMPXCHG_TYPE(t, ptr, old, new) \
5829 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5831 #ifdef CONFIG_X86_64
5832 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5834 # define CMPXCHG64(ptr, old, new) \
5835 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5838 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5843 struct x86_exception *exception)
5845 struct kvm_host_map map;
5846 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5852 /* guests cmpxchg8b have to be emulated atomically */
5853 if (bytes > 8 || (bytes & (bytes - 1)))
5856 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5858 if (gpa == UNMAPPED_GVA ||
5859 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5863 * Emulate the atomic as a straight write to avoid #AC if SLD is
5864 * enabled in the host and the access splits a cache line.
5866 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
5867 page_line_mask = ~(cache_line_size() - 1);
5869 page_line_mask = PAGE_MASK;
5871 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
5874 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5877 kaddr = map.hva + offset_in_page(gpa);
5881 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5884 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5887 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5890 exchanged = CMPXCHG64(kaddr, old, new);
5896 kvm_vcpu_unmap(vcpu, &map, true);
5899 return X86EMUL_CMPXCHG_FAILED;
5901 kvm_page_track_write(vcpu, gpa, new, bytes);
5903 return X86EMUL_CONTINUE;
5906 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5908 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5911 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5915 for (i = 0; i < vcpu->arch.pio.count; i++) {
5916 if (vcpu->arch.pio.in)
5917 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5918 vcpu->arch.pio.size, pd);
5920 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5921 vcpu->arch.pio.port, vcpu->arch.pio.size,
5925 pd += vcpu->arch.pio.size;
5930 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5931 unsigned short port, void *val,
5932 unsigned int count, bool in)
5934 vcpu->arch.pio.port = port;
5935 vcpu->arch.pio.in = in;
5936 vcpu->arch.pio.count = count;
5937 vcpu->arch.pio.size = size;
5939 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5940 vcpu->arch.pio.count = 0;
5944 vcpu->run->exit_reason = KVM_EXIT_IO;
5945 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5946 vcpu->run->io.size = size;
5947 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5948 vcpu->run->io.count = count;
5949 vcpu->run->io.port = port;
5954 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
5955 unsigned short port, void *val, unsigned int count)
5959 if (vcpu->arch.pio.count)
5962 memset(vcpu->arch.pio_data, 0, size * count);
5964 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5967 memcpy(val, vcpu->arch.pio_data, size * count);
5968 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5969 vcpu->arch.pio.count = 0;
5976 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5977 int size, unsigned short port, void *val,
5980 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
5984 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
5985 unsigned short port, const void *val,
5988 memcpy(vcpu->arch.pio_data, val, size * count);
5989 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5990 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5993 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5994 int size, unsigned short port,
5995 const void *val, unsigned int count)
5997 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6000 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6002 return kvm_x86_ops.get_segment_base(vcpu, seg);
6005 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6007 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6010 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6012 if (!need_emulate_wbinvd(vcpu))
6013 return X86EMUL_CONTINUE;
6015 if (kvm_x86_ops.has_wbinvd_exit()) {
6016 int cpu = get_cpu();
6018 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6019 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6020 wbinvd_ipi, NULL, 1);
6022 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6025 return X86EMUL_CONTINUE;
6028 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6030 kvm_emulate_wbinvd_noskip(vcpu);
6031 return kvm_skip_emulated_instruction(vcpu);
6033 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6037 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6039 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6042 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6043 unsigned long *dest)
6045 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6048 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6049 unsigned long value)
6052 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6055 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6057 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6060 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6062 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6063 unsigned long value;
6067 value = kvm_read_cr0(vcpu);
6070 value = vcpu->arch.cr2;
6073 value = kvm_read_cr3(vcpu);
6076 value = kvm_read_cr4(vcpu);
6079 value = kvm_get_cr8(vcpu);
6082 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6089 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6091 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6096 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6099 vcpu->arch.cr2 = val;
6102 res = kvm_set_cr3(vcpu, val);
6105 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6108 res = kvm_set_cr8(vcpu, val);
6111 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6118 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6120 return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6123 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6125 kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6128 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6130 kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6133 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6135 kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6138 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6140 kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6143 static unsigned long emulator_get_cached_segment_base(
6144 struct x86_emulate_ctxt *ctxt, int seg)
6146 return get_segment_base(emul_to_vcpu(ctxt), seg);
6149 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6150 struct desc_struct *desc, u32 *base3,
6153 struct kvm_segment var;
6155 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6156 *selector = var.selector;
6159 memset(desc, 0, sizeof(*desc));
6167 set_desc_limit(desc, var.limit);
6168 set_desc_base(desc, (unsigned long)var.base);
6169 #ifdef CONFIG_X86_64
6171 *base3 = var.base >> 32;
6173 desc->type = var.type;
6175 desc->dpl = var.dpl;
6176 desc->p = var.present;
6177 desc->avl = var.avl;
6185 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6186 struct desc_struct *desc, u32 base3,
6189 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6190 struct kvm_segment var;
6192 var.selector = selector;
6193 var.base = get_desc_base(desc);
6194 #ifdef CONFIG_X86_64
6195 var.base |= ((u64)base3) << 32;
6197 var.limit = get_desc_limit(desc);
6199 var.limit = (var.limit << 12) | 0xfff;
6200 var.type = desc->type;
6201 var.dpl = desc->dpl;
6206 var.avl = desc->avl;
6207 var.present = desc->p;
6208 var.unusable = !var.present;
6211 kvm_set_segment(vcpu, &var, seg);
6215 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6216 u32 msr_index, u64 *pdata)
6218 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6221 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6222 u32 msr_index, u64 data)
6224 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6227 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6229 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6231 return vcpu->arch.smbase;
6234 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6236 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6238 vcpu->arch.smbase = smbase;
6241 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6244 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6247 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6248 u32 pmc, u64 *pdata)
6250 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6253 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6255 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6258 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6259 struct x86_instruction_info *info,
6260 enum x86_intercept_stage stage)
6262 return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6266 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6267 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6270 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6273 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6275 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6278 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6280 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6283 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6285 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6288 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6290 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6293 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6295 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6298 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6300 kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6303 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6305 return emul_to_vcpu(ctxt)->arch.hflags;
6308 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6310 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6313 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6314 const char *smstate)
6316 return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6319 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6321 kvm_smm_changed(emul_to_vcpu(ctxt));
6324 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6326 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6329 static const struct x86_emulate_ops emulate_ops = {
6330 .read_gpr = emulator_read_gpr,
6331 .write_gpr = emulator_write_gpr,
6332 .read_std = emulator_read_std,
6333 .write_std = emulator_write_std,
6334 .read_phys = kvm_read_guest_phys_system,
6335 .fetch = kvm_fetch_guest_virt,
6336 .read_emulated = emulator_read_emulated,
6337 .write_emulated = emulator_write_emulated,
6338 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6339 .invlpg = emulator_invlpg,
6340 .pio_in_emulated = emulator_pio_in_emulated,
6341 .pio_out_emulated = emulator_pio_out_emulated,
6342 .get_segment = emulator_get_segment,
6343 .set_segment = emulator_set_segment,
6344 .get_cached_segment_base = emulator_get_cached_segment_base,
6345 .get_gdt = emulator_get_gdt,
6346 .get_idt = emulator_get_idt,
6347 .set_gdt = emulator_set_gdt,
6348 .set_idt = emulator_set_idt,
6349 .get_cr = emulator_get_cr,
6350 .set_cr = emulator_set_cr,
6351 .cpl = emulator_get_cpl,
6352 .get_dr = emulator_get_dr,
6353 .set_dr = emulator_set_dr,
6354 .get_smbase = emulator_get_smbase,
6355 .set_smbase = emulator_set_smbase,
6356 .set_msr = emulator_set_msr,
6357 .get_msr = emulator_get_msr,
6358 .check_pmc = emulator_check_pmc,
6359 .read_pmc = emulator_read_pmc,
6360 .halt = emulator_halt,
6361 .wbinvd = emulator_wbinvd,
6362 .fix_hypercall = emulator_fix_hypercall,
6363 .intercept = emulator_intercept,
6364 .get_cpuid = emulator_get_cpuid,
6365 .guest_has_long_mode = emulator_guest_has_long_mode,
6366 .guest_has_movbe = emulator_guest_has_movbe,
6367 .guest_has_fxsr = emulator_guest_has_fxsr,
6368 .set_nmi_mask = emulator_set_nmi_mask,
6369 .get_hflags = emulator_get_hflags,
6370 .set_hflags = emulator_set_hflags,
6371 .pre_leave_smm = emulator_pre_leave_smm,
6372 .post_leave_smm = emulator_post_leave_smm,
6373 .set_xcr = emulator_set_xcr,
6376 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6378 u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6380 * an sti; sti; sequence only disable interrupts for the first
6381 * instruction. So, if the last instruction, be it emulated or
6382 * not, left the system with the INT_STI flag enabled, it
6383 * means that the last instruction is an sti. We should not
6384 * leave the flag on in this case. The same goes for mov ss
6386 if (int_shadow & mask)
6388 if (unlikely(int_shadow || mask)) {
6389 kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6391 kvm_make_request(KVM_REQ_EVENT, vcpu);
6395 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6397 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6398 if (ctxt->exception.vector == PF_VECTOR)
6399 return kvm_propagate_fault(vcpu, &ctxt->exception);
6401 if (ctxt->exception.error_code_valid)
6402 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6403 ctxt->exception.error_code);
6405 kvm_queue_exception(vcpu, ctxt->exception.vector);
6409 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6411 struct x86_emulate_ctxt *ctxt;
6413 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6415 pr_err("kvm: failed to allocate vcpu's emulator\n");
6420 ctxt->ops = &emulate_ops;
6421 vcpu->arch.emulate_ctxt = ctxt;
6426 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6428 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6431 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6433 ctxt->gpa_available = false;
6434 ctxt->eflags = kvm_get_rflags(vcpu);
6435 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6437 ctxt->eip = kvm_rip_read(vcpu);
6438 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6439 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6440 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6441 cs_db ? X86EMUL_MODE_PROT32 :
6442 X86EMUL_MODE_PROT16;
6443 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6444 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6445 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6447 init_decode_cache(ctxt);
6448 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6451 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6453 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6456 init_emulate_ctxt(vcpu);
6460 ctxt->_eip = ctxt->eip + inc_eip;
6461 ret = emulate_int_real(ctxt, irq);
6463 if (ret != X86EMUL_CONTINUE) {
6464 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6466 ctxt->eip = ctxt->_eip;
6467 kvm_rip_write(vcpu, ctxt->eip);
6468 kvm_set_rflags(vcpu, ctxt->eflags);
6471 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6473 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6475 ++vcpu->stat.insn_emulation_fail;
6476 trace_kvm_emulate_insn_failed(vcpu);
6478 if (emulation_type & EMULTYPE_VMWARE_GP) {
6479 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6483 if (emulation_type & EMULTYPE_SKIP) {
6484 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6485 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6486 vcpu->run->internal.ndata = 0;
6490 kvm_queue_exception(vcpu, UD_VECTOR);
6492 if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
6493 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6494 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6495 vcpu->run->internal.ndata = 0;
6502 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6503 bool write_fault_to_shadow_pgtable,
6506 gpa_t gpa = cr2_or_gpa;
6509 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6512 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6513 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6516 if (!vcpu->arch.mmu->direct_map) {
6518 * Write permission should be allowed since only
6519 * write access need to be emulated.
6521 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6524 * If the mapping is invalid in guest, let cpu retry
6525 * it to generate fault.
6527 if (gpa == UNMAPPED_GVA)
6532 * Do not retry the unhandleable instruction if it faults on the
6533 * readonly host memory, otherwise it will goto a infinite loop:
6534 * retry instruction -> write #PF -> emulation fail -> retry
6535 * instruction -> ...
6537 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6540 * If the instruction failed on the error pfn, it can not be fixed,
6541 * report the error to userspace.
6543 if (is_error_noslot_pfn(pfn))
6546 kvm_release_pfn_clean(pfn);
6548 /* The instructions are well-emulated on direct mmu. */
6549 if (vcpu->arch.mmu->direct_map) {
6550 unsigned int indirect_shadow_pages;
6552 spin_lock(&vcpu->kvm->mmu_lock);
6553 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6554 spin_unlock(&vcpu->kvm->mmu_lock);
6556 if (indirect_shadow_pages)
6557 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6563 * if emulation was due to access to shadowed page table
6564 * and it failed try to unshadow page and re-enter the
6565 * guest to let CPU execute the instruction.
6567 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6570 * If the access faults on its page table, it can not
6571 * be fixed by unprotecting shadow page and it should
6572 * be reported to userspace.
6574 return !write_fault_to_shadow_pgtable;
6577 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6578 gpa_t cr2_or_gpa, int emulation_type)
6580 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6581 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
6583 last_retry_eip = vcpu->arch.last_retry_eip;
6584 last_retry_addr = vcpu->arch.last_retry_addr;
6587 * If the emulation is caused by #PF and it is non-page_table
6588 * writing instruction, it means the VM-EXIT is caused by shadow
6589 * page protected, we can zap the shadow page and retry this
6590 * instruction directly.
6592 * Note: if the guest uses a non-page-table modifying instruction
6593 * on the PDE that points to the instruction, then we will unmap
6594 * the instruction and go to an infinite loop. So, we cache the
6595 * last retried eip and the last fault address, if we meet the eip
6596 * and the address again, we can break out of the potential infinite
6599 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6601 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6604 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6605 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6608 if (x86_page_table_writing_insn(ctxt))
6611 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
6614 vcpu->arch.last_retry_eip = ctxt->eip;
6615 vcpu->arch.last_retry_addr = cr2_or_gpa;
6617 if (!vcpu->arch.mmu->direct_map)
6618 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6620 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6625 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6626 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6628 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6630 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6631 /* This is a good place to trace that we are exiting SMM. */
6632 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6634 /* Process a latched INIT or SMI, if any. */
6635 kvm_make_request(KVM_REQ_EVENT, vcpu);
6638 kvm_mmu_reset_context(vcpu);
6641 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6650 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6651 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6656 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6658 struct kvm_run *kvm_run = vcpu->run;
6660 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6661 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6662 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6663 kvm_run->debug.arch.exception = DB_VECTOR;
6664 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6667 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6671 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6673 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6676 r = kvm_x86_ops.skip_emulated_instruction(vcpu);
6681 * rflags is the old, "raw" value of the flags. The new value has
6682 * not been saved yet.
6684 * This is correct even for TF set by the guest, because "the
6685 * processor will not generate this exception after the instruction
6686 * that sets the TF flag".
6688 if (unlikely(rflags & X86_EFLAGS_TF))
6689 r = kvm_vcpu_do_singlestep(vcpu);
6692 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6694 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6696 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6697 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6698 struct kvm_run *kvm_run = vcpu->run;
6699 unsigned long eip = kvm_get_linear_rip(vcpu);
6700 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6701 vcpu->arch.guest_debug_dr7,
6705 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6706 kvm_run->debug.arch.pc = eip;
6707 kvm_run->debug.arch.exception = DB_VECTOR;
6708 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6714 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6715 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6716 unsigned long eip = kvm_get_linear_rip(vcpu);
6717 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6722 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6723 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6724 kvm_queue_exception(vcpu, DB_VECTOR);
6733 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6735 switch (ctxt->opcode_len) {
6742 case 0xe6: /* OUT */
6746 case 0x6c: /* INS */
6748 case 0x6e: /* OUTS */
6755 case 0x33: /* RDPMC */
6764 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6765 int emulation_type, void *insn, int insn_len)
6768 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6769 bool writeback = true;
6770 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6772 vcpu->arch.l1tf_flush_l1d = true;
6775 * Clear write_fault_to_shadow_pgtable here to ensure it is
6778 vcpu->arch.write_fault_to_shadow_pgtable = false;
6779 kvm_clear_exception_queue(vcpu);
6781 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6782 init_emulate_ctxt(vcpu);
6785 * We will reenter on the same instruction since
6786 * we do not set complete_userspace_io. This does not
6787 * handle watchpoints yet, those would be handled in
6790 if (!(emulation_type & EMULTYPE_SKIP) &&
6791 kvm_vcpu_check_breakpoint(vcpu, &r))
6794 ctxt->interruptibility = 0;
6795 ctxt->have_exception = false;
6796 ctxt->exception.vector = -1;
6797 ctxt->perm_ok = false;
6799 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6801 r = x86_decode_insn(ctxt, insn, insn_len);
6803 trace_kvm_emulate_insn_start(vcpu);
6804 ++vcpu->stat.insn_emulation;
6805 if (r != EMULATION_OK) {
6806 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6807 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6808 kvm_queue_exception(vcpu, UD_VECTOR);
6811 if (reexecute_instruction(vcpu, cr2_or_gpa,
6815 if (ctxt->have_exception) {
6817 * #UD should result in just EMULATION_FAILED, and trap-like
6818 * exception should not be encountered during decode.
6820 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6821 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6822 inject_emulated_exception(vcpu);
6825 return handle_emulation_failure(vcpu, emulation_type);
6829 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6830 !is_vmware_backdoor_opcode(ctxt)) {
6831 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6836 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6837 * for kvm_skip_emulated_instruction(). The caller is responsible for
6838 * updating interruptibility state and injecting single-step #DBs.
6840 if (emulation_type & EMULTYPE_SKIP) {
6841 kvm_rip_write(vcpu, ctxt->_eip);
6842 if (ctxt->eflags & X86_EFLAGS_RF)
6843 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6847 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
6850 /* this is needed for vmware backdoor interface to work since it
6851 changes registers values during IO operation */
6852 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6853 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6854 emulator_invalidate_register_cache(ctxt);
6858 if (emulation_type & EMULTYPE_PF) {
6859 /* Save the faulting GPA (cr2) in the address field */
6860 ctxt->exception.address = cr2_or_gpa;
6862 /* With shadow page tables, cr2 contains a GVA or nGPA. */
6863 if (vcpu->arch.mmu->direct_map) {
6864 ctxt->gpa_available = true;
6865 ctxt->gpa_val = cr2_or_gpa;
6868 /* Sanitize the address out of an abundance of paranoia. */
6869 ctxt->exception.address = 0;
6872 r = x86_emulate_insn(ctxt);
6874 if (r == EMULATION_INTERCEPTED)
6877 if (r == EMULATION_FAILED) {
6878 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
6882 return handle_emulation_failure(vcpu, emulation_type);
6885 if (ctxt->have_exception) {
6887 if (inject_emulated_exception(vcpu))
6889 } else if (vcpu->arch.pio.count) {
6890 if (!vcpu->arch.pio.in) {
6891 /* FIXME: return into emulator if single-stepping. */
6892 vcpu->arch.pio.count = 0;
6895 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6898 } else if (vcpu->mmio_needed) {
6899 ++vcpu->stat.mmio_exits;
6901 if (!vcpu->mmio_is_write)
6904 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6905 } else if (r == EMULATION_RESTART)
6911 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6912 toggle_interruptibility(vcpu, ctxt->interruptibility);
6913 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6914 if (!ctxt->have_exception ||
6915 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6916 kvm_rip_write(vcpu, ctxt->eip);
6918 r = kvm_vcpu_do_singlestep(vcpu);
6919 if (kvm_x86_ops.update_emulated_instruction)
6920 kvm_x86_ops.update_emulated_instruction(vcpu);
6921 __kvm_set_rflags(vcpu, ctxt->eflags);
6925 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6926 * do nothing, and it will be requested again as soon as
6927 * the shadow expires. But we still need to check here,
6928 * because POPF has no interrupt shadow.
6930 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6931 kvm_make_request(KVM_REQ_EVENT, vcpu);
6933 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6938 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6940 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6942 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6944 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6945 void *insn, int insn_len)
6947 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6949 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6951 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6953 vcpu->arch.pio.count = 0;
6957 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6959 vcpu->arch.pio.count = 0;
6961 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6964 return kvm_skip_emulated_instruction(vcpu);
6967 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6968 unsigned short port)
6970 unsigned long val = kvm_rax_read(vcpu);
6971 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
6977 * Workaround userspace that relies on old KVM behavior of %rip being
6978 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6981 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6982 vcpu->arch.complete_userspace_io =
6983 complete_fast_pio_out_port_0x7e;
6984 kvm_skip_emulated_instruction(vcpu);
6986 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6987 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6992 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6996 /* We should only ever be called with arch.pio.count equal to 1 */
6997 BUG_ON(vcpu->arch.pio.count != 1);
6999 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7000 vcpu->arch.pio.count = 0;
7004 /* For size less than 4 we merge, else we zero extend */
7005 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7008 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7009 * the copy and tracing
7011 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7012 kvm_rax_write(vcpu, val);
7014 return kvm_skip_emulated_instruction(vcpu);
7017 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7018 unsigned short port)
7023 /* For size less than 4 we merge, else we zero extend */
7024 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7026 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7028 kvm_rax_write(vcpu, val);
7032 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7033 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7038 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7043 ret = kvm_fast_pio_in(vcpu, size, port);
7045 ret = kvm_fast_pio_out(vcpu, size, port);
7046 return ret && kvm_skip_emulated_instruction(vcpu);
7048 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7050 static int kvmclock_cpu_down_prep(unsigned int cpu)
7052 __this_cpu_write(cpu_tsc_khz, 0);
7056 static void tsc_khz_changed(void *data)
7058 struct cpufreq_freqs *freq = data;
7059 unsigned long khz = 0;
7063 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7064 khz = cpufreq_quick_get(raw_smp_processor_id());
7067 __this_cpu_write(cpu_tsc_khz, khz);
7070 #ifdef CONFIG_X86_64
7071 static void kvm_hyperv_tsc_notifier(void)
7074 struct kvm_vcpu *vcpu;
7077 mutex_lock(&kvm_lock);
7078 list_for_each_entry(kvm, &vm_list, vm_list)
7079 kvm_make_mclock_inprogress_request(kvm);
7081 hyperv_stop_tsc_emulation();
7083 /* TSC frequency always matches when on Hyper-V */
7084 for_each_present_cpu(cpu)
7085 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7086 kvm_max_guest_tsc_khz = tsc_khz;
7088 list_for_each_entry(kvm, &vm_list, vm_list) {
7089 struct kvm_arch *ka = &kvm->arch;
7091 spin_lock(&ka->pvclock_gtod_sync_lock);
7093 pvclock_update_vm_gtod_copy(kvm);
7095 kvm_for_each_vcpu(cpu, vcpu, kvm)
7096 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7098 kvm_for_each_vcpu(cpu, vcpu, kvm)
7099 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7101 spin_unlock(&ka->pvclock_gtod_sync_lock);
7103 mutex_unlock(&kvm_lock);
7107 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7110 struct kvm_vcpu *vcpu;
7111 int i, send_ipi = 0;
7114 * We allow guests to temporarily run on slowing clocks,
7115 * provided we notify them after, or to run on accelerating
7116 * clocks, provided we notify them before. Thus time never
7119 * However, we have a problem. We can't atomically update
7120 * the frequency of a given CPU from this function; it is
7121 * merely a notifier, which can be called from any CPU.
7122 * Changing the TSC frequency at arbitrary points in time
7123 * requires a recomputation of local variables related to
7124 * the TSC for each VCPU. We must flag these local variables
7125 * to be updated and be sure the update takes place with the
7126 * new frequency before any guests proceed.
7128 * Unfortunately, the combination of hotplug CPU and frequency
7129 * change creates an intractable locking scenario; the order
7130 * of when these callouts happen is undefined with respect to
7131 * CPU hotplug, and they can race with each other. As such,
7132 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7133 * undefined; you can actually have a CPU frequency change take
7134 * place in between the computation of X and the setting of the
7135 * variable. To protect against this problem, all updates of
7136 * the per_cpu tsc_khz variable are done in an interrupt
7137 * protected IPI, and all callers wishing to update the value
7138 * must wait for a synchronous IPI to complete (which is trivial
7139 * if the caller is on the CPU already). This establishes the
7140 * necessary total order on variable updates.
7142 * Note that because a guest time update may take place
7143 * anytime after the setting of the VCPU's request bit, the
7144 * correct TSC value must be set before the request. However,
7145 * to ensure the update actually makes it to any guest which
7146 * starts running in hardware virtualization between the set
7147 * and the acquisition of the spinlock, we must also ping the
7148 * CPU after setting the request bit.
7152 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7154 mutex_lock(&kvm_lock);
7155 list_for_each_entry(kvm, &vm_list, vm_list) {
7156 kvm_for_each_vcpu(i, vcpu, kvm) {
7157 if (vcpu->cpu != cpu)
7159 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7160 if (vcpu->cpu != raw_smp_processor_id())
7164 mutex_unlock(&kvm_lock);
7166 if (freq->old < freq->new && send_ipi) {
7168 * We upscale the frequency. Must make the guest
7169 * doesn't see old kvmclock values while running with
7170 * the new frequency, otherwise we risk the guest sees
7171 * time go backwards.
7173 * In case we update the frequency for another cpu
7174 * (which might be in guest context) send an interrupt
7175 * to kick the cpu out of guest context. Next time
7176 * guest context is entered kvmclock will be updated,
7177 * so the guest will not see stale values.
7179 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7183 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7186 struct cpufreq_freqs *freq = data;
7189 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7191 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7194 for_each_cpu(cpu, freq->policy->cpus)
7195 __kvmclock_cpufreq_notifier(freq, cpu);
7200 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7201 .notifier_call = kvmclock_cpufreq_notifier
7204 static int kvmclock_cpu_online(unsigned int cpu)
7206 tsc_khz_changed(NULL);
7210 static void kvm_timer_init(void)
7212 max_tsc_khz = tsc_khz;
7214 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7215 #ifdef CONFIG_CPU_FREQ
7216 struct cpufreq_policy *policy;
7220 policy = cpufreq_cpu_get(cpu);
7222 if (policy->cpuinfo.max_freq)
7223 max_tsc_khz = policy->cpuinfo.max_freq;
7224 cpufreq_cpu_put(policy);
7228 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7229 CPUFREQ_TRANSITION_NOTIFIER);
7232 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7233 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7236 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7237 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7239 int kvm_is_in_guest(void)
7241 return __this_cpu_read(current_vcpu) != NULL;
7244 static int kvm_is_user_mode(void)
7248 if (__this_cpu_read(current_vcpu))
7249 user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7251 return user_mode != 0;
7254 static unsigned long kvm_get_guest_ip(void)
7256 unsigned long ip = 0;
7258 if (__this_cpu_read(current_vcpu))
7259 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7264 static void kvm_handle_intel_pt_intr(void)
7266 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7268 kvm_make_request(KVM_REQ_PMI, vcpu);
7269 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7270 (unsigned long *)&vcpu->arch.pmu.global_status);
7273 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7274 .is_in_guest = kvm_is_in_guest,
7275 .is_user_mode = kvm_is_user_mode,
7276 .get_guest_ip = kvm_get_guest_ip,
7277 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7280 #ifdef CONFIG_X86_64
7281 static void pvclock_gtod_update_fn(struct work_struct *work)
7285 struct kvm_vcpu *vcpu;
7288 mutex_lock(&kvm_lock);
7289 list_for_each_entry(kvm, &vm_list, vm_list)
7290 kvm_for_each_vcpu(i, vcpu, kvm)
7291 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7292 atomic_set(&kvm_guest_has_master_clock, 0);
7293 mutex_unlock(&kvm_lock);
7296 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7299 * Notification about pvclock gtod data update.
7301 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7304 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7305 struct timekeeper *tk = priv;
7307 update_pvclock_gtod(tk);
7309 /* disable master clock if host does not trust, or does not
7310 * use, TSC based clocksource.
7312 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7313 atomic_read(&kvm_guest_has_master_clock) != 0)
7314 queue_work(system_long_wq, &pvclock_gtod_work);
7319 static struct notifier_block pvclock_gtod_notifier = {
7320 .notifier_call = pvclock_gtod_notify,
7324 int kvm_arch_init(void *opaque)
7326 struct kvm_x86_init_ops *ops = opaque;
7329 if (kvm_x86_ops.hardware_enable) {
7330 printk(KERN_ERR "kvm: already loaded the other module\n");
7335 if (!ops->cpu_has_kvm_support()) {
7336 pr_err_ratelimited("kvm: no hardware support\n");
7340 if (ops->disabled_by_bios()) {
7341 pr_err_ratelimited("kvm: disabled by bios\n");
7347 * KVM explicitly assumes that the guest has an FPU and
7348 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7349 * vCPU's FPU state as a fxregs_state struct.
7351 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7352 printk(KERN_ERR "kvm: inadequate fpu\n");
7358 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7359 __alignof__(struct fpu), SLAB_ACCOUNT,
7361 if (!x86_fpu_cache) {
7362 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7366 x86_emulator_cache = kvm_alloc_emulator_cache();
7367 if (!x86_emulator_cache) {
7368 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7369 goto out_free_x86_fpu_cache;
7372 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7374 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7375 goto out_free_x86_emulator_cache;
7378 r = kvm_mmu_module_init();
7380 goto out_free_percpu;
7382 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7383 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7384 PT_PRESENT_MASK, 0, sme_me_mask);
7387 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7389 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7390 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7391 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7395 if (pi_inject_timer == -1)
7396 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7397 #ifdef CONFIG_X86_64
7398 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7400 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7401 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7407 free_percpu(shared_msrs);
7408 out_free_x86_emulator_cache:
7409 kmem_cache_destroy(x86_emulator_cache);
7410 out_free_x86_fpu_cache:
7411 kmem_cache_destroy(x86_fpu_cache);
7416 void kvm_arch_exit(void)
7418 #ifdef CONFIG_X86_64
7419 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7420 clear_hv_tscchange_cb();
7423 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7425 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7426 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7427 CPUFREQ_TRANSITION_NOTIFIER);
7428 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7429 #ifdef CONFIG_X86_64
7430 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7432 kvm_x86_ops.hardware_enable = NULL;
7433 kvm_mmu_module_exit();
7434 free_percpu(shared_msrs);
7435 kmem_cache_destroy(x86_fpu_cache);
7438 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7440 ++vcpu->stat.halt_exits;
7441 if (lapic_in_kernel(vcpu)) {
7442 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7445 vcpu->run->exit_reason = KVM_EXIT_HLT;
7449 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7451 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7453 int ret = kvm_skip_emulated_instruction(vcpu);
7455 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7456 * KVM_EXIT_DEBUG here.
7458 return kvm_vcpu_halt(vcpu) && ret;
7460 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7462 #ifdef CONFIG_X86_64
7463 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7464 unsigned long clock_type)
7466 struct kvm_clock_pairing clock_pairing;
7467 struct timespec64 ts;
7471 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7472 return -KVM_EOPNOTSUPP;
7474 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7475 return -KVM_EOPNOTSUPP;
7477 clock_pairing.sec = ts.tv_sec;
7478 clock_pairing.nsec = ts.tv_nsec;
7479 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7480 clock_pairing.flags = 0;
7481 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7484 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7485 sizeof(struct kvm_clock_pairing)))
7493 * kvm_pv_kick_cpu_op: Kick a vcpu.
7495 * @apicid - apicid of vcpu to be kicked.
7497 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7499 struct kvm_lapic_irq lapic_irq;
7501 lapic_irq.shorthand = APIC_DEST_NOSHORT;
7502 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7503 lapic_irq.level = 0;
7504 lapic_irq.dest_id = apicid;
7505 lapic_irq.msi_redir_hint = false;
7507 lapic_irq.delivery_mode = APIC_DM_REMRD;
7508 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7511 bool kvm_apicv_activated(struct kvm *kvm)
7513 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
7515 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
7517 void kvm_apicv_init(struct kvm *kvm, bool enable)
7520 clear_bit(APICV_INHIBIT_REASON_DISABLE,
7521 &kvm->arch.apicv_inhibit_reasons);
7523 set_bit(APICV_INHIBIT_REASON_DISABLE,
7524 &kvm->arch.apicv_inhibit_reasons);
7526 EXPORT_SYMBOL_GPL(kvm_apicv_init);
7528 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7530 struct kvm_vcpu *target = NULL;
7531 struct kvm_apic_map *map;
7534 map = rcu_dereference(kvm->arch.apic_map);
7536 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7537 target = map->phys_map[dest_id]->vcpu;
7541 if (target && READ_ONCE(target->ready))
7542 kvm_vcpu_yield_to(target);
7545 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7547 unsigned long nr, a0, a1, a2, a3, ret;
7550 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7551 return kvm_hv_hypercall(vcpu);
7553 nr = kvm_rax_read(vcpu);
7554 a0 = kvm_rbx_read(vcpu);
7555 a1 = kvm_rcx_read(vcpu);
7556 a2 = kvm_rdx_read(vcpu);
7557 a3 = kvm_rsi_read(vcpu);
7559 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7561 op_64_bit = is_64_bit_mode(vcpu);
7570 if (kvm_x86_ops.get_cpl(vcpu) != 0) {
7576 case KVM_HC_VAPIC_POLL_IRQ:
7579 case KVM_HC_KICK_CPU:
7580 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7581 kvm_sched_yield(vcpu->kvm, a1);
7584 #ifdef CONFIG_X86_64
7585 case KVM_HC_CLOCK_PAIRING:
7586 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7589 case KVM_HC_SEND_IPI:
7590 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7592 case KVM_HC_SCHED_YIELD:
7593 kvm_sched_yield(vcpu->kvm, a0);
7603 kvm_rax_write(vcpu, ret);
7605 ++vcpu->stat.hypercalls;
7606 return kvm_skip_emulated_instruction(vcpu);
7608 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7610 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7612 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7613 char instruction[3];
7614 unsigned long rip = kvm_rip_read(vcpu);
7616 kvm_x86_ops.patch_hypercall(vcpu, instruction);
7618 return emulator_write_emulated(ctxt, rip, instruction, 3,
7622 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7624 return vcpu->run->request_interrupt_window &&
7625 likely(!pic_in_kernel(vcpu->kvm));
7628 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7630 struct kvm_run *kvm_run = vcpu->run;
7632 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7633 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7634 kvm_run->cr8 = kvm_get_cr8(vcpu);
7635 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7636 kvm_run->ready_for_interrupt_injection =
7637 pic_in_kernel(vcpu->kvm) ||
7638 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7641 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7645 if (!kvm_x86_ops.update_cr8_intercept)
7648 if (!lapic_in_kernel(vcpu))
7651 if (vcpu->arch.apicv_active)
7654 if (!vcpu->arch.apic->vapic_addr)
7655 max_irr = kvm_lapic_find_highest_irr(vcpu);
7662 tpr = kvm_lapic_get_cr8(vcpu);
7664 kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
7667 static int inject_pending_event(struct kvm_vcpu *vcpu)
7671 /* try to reinject previous events if any */
7673 if (vcpu->arch.exception.injected)
7674 kvm_x86_ops.queue_exception(vcpu);
7676 * Do not inject an NMI or interrupt if there is a pending
7677 * exception. Exceptions and interrupts are recognized at
7678 * instruction boundaries, i.e. the start of an instruction.
7679 * Trap-like exceptions, e.g. #DB, have higher priority than
7680 * NMIs and interrupts, i.e. traps are recognized before an
7681 * NMI/interrupt that's pending on the same instruction.
7682 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7683 * priority, but are only generated (pended) during instruction
7684 * execution, i.e. a pending fault-like exception means the
7685 * fault occurred on the *previous* instruction and must be
7686 * serviced prior to recognizing any new events in order to
7687 * fully complete the previous instruction.
7689 else if (!vcpu->arch.exception.pending) {
7690 if (vcpu->arch.nmi_injected)
7691 kvm_x86_ops.set_nmi(vcpu);
7692 else if (vcpu->arch.interrupt.injected)
7693 kvm_x86_ops.set_irq(vcpu);
7697 * Call check_nested_events() even if we reinjected a previous event
7698 * in order for caller to determine if it should require immediate-exit
7699 * from L2 to L1 due to pending L1 events which require exit
7702 if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) {
7703 r = kvm_x86_ops.check_nested_events(vcpu);
7708 /* try to inject new event if pending */
7709 if (vcpu->arch.exception.pending) {
7710 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7711 vcpu->arch.exception.has_error_code,
7712 vcpu->arch.exception.error_code);
7714 WARN_ON_ONCE(vcpu->arch.exception.injected);
7715 vcpu->arch.exception.pending = false;
7716 vcpu->arch.exception.injected = true;
7718 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7719 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7722 if (vcpu->arch.exception.nr == DB_VECTOR) {
7724 * This code assumes that nSVM doesn't use
7725 * check_nested_events(). If it does, the
7726 * DR6/DR7 changes should happen before L1
7727 * gets a #VMEXIT for an intercepted #DB in
7728 * L2. (Under VMX, on the other hand, the
7729 * DR6/DR7 changes should not happen in the
7730 * event of a VM-exit to L1 for an intercepted
7733 kvm_deliver_exception_payload(vcpu);
7734 if (vcpu->arch.dr7 & DR7_GD) {
7735 vcpu->arch.dr7 &= ~DR7_GD;
7736 kvm_update_dr7(vcpu);
7740 kvm_x86_ops.queue_exception(vcpu);
7743 /* Don't consider new event if we re-injected an event */
7744 if (kvm_event_needs_reinjection(vcpu))
7747 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7748 kvm_x86_ops.smi_allowed(vcpu)) {
7749 vcpu->arch.smi_pending = false;
7750 ++vcpu->arch.smi_count;
7752 } else if (vcpu->arch.nmi_pending && kvm_x86_ops.nmi_allowed(vcpu)) {
7753 --vcpu->arch.nmi_pending;
7754 vcpu->arch.nmi_injected = true;
7755 kvm_x86_ops.set_nmi(vcpu);
7756 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7758 * Because interrupts can be injected asynchronously, we are
7759 * calling check_nested_events again here to avoid a race condition.
7760 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7761 * proposal and current concerns. Perhaps we should be setting
7762 * KVM_REQ_EVENT only on certain events and not unconditionally?
7764 if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) {
7765 r = kvm_x86_ops.check_nested_events(vcpu);
7769 if (kvm_x86_ops.interrupt_allowed(vcpu)) {
7770 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7772 kvm_x86_ops.set_irq(vcpu);
7779 static void process_nmi(struct kvm_vcpu *vcpu)
7784 * x86 is limited to one NMI running, and one NMI pending after it.
7785 * If an NMI is already in progress, limit further NMIs to just one.
7786 * Otherwise, allow two (and we'll inject the first one immediately).
7788 if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7791 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7792 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7793 kvm_make_request(KVM_REQ_EVENT, vcpu);
7796 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7799 flags |= seg->g << 23;
7800 flags |= seg->db << 22;
7801 flags |= seg->l << 21;
7802 flags |= seg->avl << 20;
7803 flags |= seg->present << 15;
7804 flags |= seg->dpl << 13;
7805 flags |= seg->s << 12;
7806 flags |= seg->type << 8;
7810 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7812 struct kvm_segment seg;
7815 kvm_get_segment(vcpu, &seg, n);
7816 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7819 offset = 0x7f84 + n * 12;
7821 offset = 0x7f2c + (n - 3) * 12;
7823 put_smstate(u32, buf, offset + 8, seg.base);
7824 put_smstate(u32, buf, offset + 4, seg.limit);
7825 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7828 #ifdef CONFIG_X86_64
7829 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7831 struct kvm_segment seg;
7835 kvm_get_segment(vcpu, &seg, n);
7836 offset = 0x7e00 + n * 16;
7838 flags = enter_smm_get_segment_flags(&seg) >> 8;
7839 put_smstate(u16, buf, offset, seg.selector);
7840 put_smstate(u16, buf, offset + 2, flags);
7841 put_smstate(u32, buf, offset + 4, seg.limit);
7842 put_smstate(u64, buf, offset + 8, seg.base);
7846 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7849 struct kvm_segment seg;
7853 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7854 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7855 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7856 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7858 for (i = 0; i < 8; i++)
7859 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7861 kvm_get_dr(vcpu, 6, &val);
7862 put_smstate(u32, buf, 0x7fcc, (u32)val);
7863 kvm_get_dr(vcpu, 7, &val);
7864 put_smstate(u32, buf, 0x7fc8, (u32)val);
7866 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7867 put_smstate(u32, buf, 0x7fc4, seg.selector);
7868 put_smstate(u32, buf, 0x7f64, seg.base);
7869 put_smstate(u32, buf, 0x7f60, seg.limit);
7870 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7872 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7873 put_smstate(u32, buf, 0x7fc0, seg.selector);
7874 put_smstate(u32, buf, 0x7f80, seg.base);
7875 put_smstate(u32, buf, 0x7f7c, seg.limit);
7876 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7878 kvm_x86_ops.get_gdt(vcpu, &dt);
7879 put_smstate(u32, buf, 0x7f74, dt.address);
7880 put_smstate(u32, buf, 0x7f70, dt.size);
7882 kvm_x86_ops.get_idt(vcpu, &dt);
7883 put_smstate(u32, buf, 0x7f58, dt.address);
7884 put_smstate(u32, buf, 0x7f54, dt.size);
7886 for (i = 0; i < 6; i++)
7887 enter_smm_save_seg_32(vcpu, buf, i);
7889 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7892 put_smstate(u32, buf, 0x7efc, 0x00020000);
7893 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7896 #ifdef CONFIG_X86_64
7897 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7900 struct kvm_segment seg;
7904 for (i = 0; i < 16; i++)
7905 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7907 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7908 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7910 kvm_get_dr(vcpu, 6, &val);
7911 put_smstate(u64, buf, 0x7f68, val);
7912 kvm_get_dr(vcpu, 7, &val);
7913 put_smstate(u64, buf, 0x7f60, val);
7915 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7916 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7917 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7919 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7922 put_smstate(u32, buf, 0x7efc, 0x00020064);
7924 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7926 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7927 put_smstate(u16, buf, 0x7e90, seg.selector);
7928 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7929 put_smstate(u32, buf, 0x7e94, seg.limit);
7930 put_smstate(u64, buf, 0x7e98, seg.base);
7932 kvm_x86_ops.get_idt(vcpu, &dt);
7933 put_smstate(u32, buf, 0x7e84, dt.size);
7934 put_smstate(u64, buf, 0x7e88, dt.address);
7936 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7937 put_smstate(u16, buf, 0x7e70, seg.selector);
7938 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7939 put_smstate(u32, buf, 0x7e74, seg.limit);
7940 put_smstate(u64, buf, 0x7e78, seg.base);
7942 kvm_x86_ops.get_gdt(vcpu, &dt);
7943 put_smstate(u32, buf, 0x7e64, dt.size);
7944 put_smstate(u64, buf, 0x7e68, dt.address);
7946 for (i = 0; i < 6; i++)
7947 enter_smm_save_seg_64(vcpu, buf, i);
7951 static void enter_smm(struct kvm_vcpu *vcpu)
7953 struct kvm_segment cs, ds;
7958 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7959 memset(buf, 0, 512);
7960 #ifdef CONFIG_X86_64
7961 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7962 enter_smm_save_state_64(vcpu, buf);
7965 enter_smm_save_state_32(vcpu, buf);
7968 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7969 * vCPU state (e.g. leave guest mode) after we've saved the state into
7970 * the SMM state-save area.
7972 kvm_x86_ops.pre_enter_smm(vcpu, buf);
7974 vcpu->arch.hflags |= HF_SMM_MASK;
7975 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7977 if (kvm_x86_ops.get_nmi_mask(vcpu))
7978 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7980 kvm_x86_ops.set_nmi_mask(vcpu, true);
7982 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7983 kvm_rip_write(vcpu, 0x8000);
7985 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7986 kvm_x86_ops.set_cr0(vcpu, cr0);
7987 vcpu->arch.cr0 = cr0;
7989 kvm_x86_ops.set_cr4(vcpu, 0);
7991 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7992 dt.address = dt.size = 0;
7993 kvm_x86_ops.set_idt(vcpu, &dt);
7995 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7997 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7998 cs.base = vcpu->arch.smbase;
8003 cs.limit = ds.limit = 0xffffffff;
8004 cs.type = ds.type = 0x3;
8005 cs.dpl = ds.dpl = 0;
8010 cs.avl = ds.avl = 0;
8011 cs.present = ds.present = 1;
8012 cs.unusable = ds.unusable = 0;
8013 cs.padding = ds.padding = 0;
8015 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8016 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8017 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8018 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8019 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8020 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8022 #ifdef CONFIG_X86_64
8023 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8024 kvm_x86_ops.set_efer(vcpu, 0);
8027 kvm_update_cpuid(vcpu);
8028 kvm_mmu_reset_context(vcpu);
8031 static void process_smi(struct kvm_vcpu *vcpu)
8033 vcpu->arch.smi_pending = true;
8034 kvm_make_request(KVM_REQ_EVENT, vcpu);
8037 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8038 unsigned long *vcpu_bitmap)
8042 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8044 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8047 free_cpumask_var(cpus);
8050 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8052 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8055 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8057 if (!lapic_in_kernel(vcpu))
8060 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8061 kvm_apic_update_apicv(vcpu);
8062 kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8064 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8067 * NOTE: Do not hold any lock prior to calling this.
8069 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8070 * locked, because it calls __x86_set_memory_region() which does
8071 * synchronize_srcu(&kvm->srcu).
8073 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8075 unsigned long old, new, expected;
8077 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8078 !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8081 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8083 expected = new = old;
8085 __clear_bit(bit, &new);
8087 __set_bit(bit, &new);
8090 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8091 } while (old != expected);
8096 trace_kvm_apicv_update_request(activate, bit);
8097 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8098 kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8099 kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
8101 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8103 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8105 if (!kvm_apic_present(vcpu))
8108 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8110 if (irqchip_split(vcpu->kvm))
8111 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8113 if (vcpu->arch.apicv_active)
8114 kvm_x86_ops.sync_pir_to_irr(vcpu);
8115 if (ioapic_in_kernel(vcpu->kvm))
8116 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8119 if (is_guest_mode(vcpu))
8120 vcpu->arch.load_eoi_exitmap_pending = true;
8122 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8125 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8127 u64 eoi_exit_bitmap[4];
8129 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8132 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8133 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8134 kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8137 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8138 unsigned long start, unsigned long end,
8141 unsigned long apic_address;
8144 * The physical address of apic access page is stored in the VMCS.
8145 * Update it when it becomes invalid.
8147 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8148 if (start <= apic_address && apic_address < end)
8149 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8154 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8156 struct page *page = NULL;
8158 if (!lapic_in_kernel(vcpu))
8161 if (!kvm_x86_ops.set_apic_access_page_addr)
8164 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8165 if (is_error_page(page))
8167 kvm_x86_ops.set_apic_access_page_addr(vcpu, page_to_phys(page));
8170 * Do not pin apic access page in memory, the MMU notifier
8171 * will call us again if it is migrated or swapped out.
8176 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8178 smp_send_reschedule(vcpu->cpu);
8180 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8183 * Returns 1 to let vcpu_run() continue the guest execution loop without
8184 * exiting to the userspace. Otherwise, the value will be returned to the
8187 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8191 dm_request_for_irq_injection(vcpu) &&
8192 kvm_cpu_accept_dm_intr(vcpu);
8193 enum exit_fastpath_completion exit_fastpath = EXIT_FASTPATH_NONE;
8195 bool req_immediate_exit = false;
8197 if (kvm_request_pending(vcpu)) {
8198 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8199 if (unlikely(!kvm_x86_ops.get_vmcs12_pages(vcpu))) {
8204 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8205 kvm_mmu_unload(vcpu);
8206 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8207 __kvm_migrate_timers(vcpu);
8208 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8209 kvm_gen_update_masterclock(vcpu->kvm);
8210 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8211 kvm_gen_kvmclock_update(vcpu);
8212 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8213 r = kvm_guest_time_update(vcpu);
8217 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8218 kvm_mmu_sync_roots(vcpu);
8219 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8220 kvm_mmu_load_pgd(vcpu);
8221 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
8222 kvm_vcpu_flush_tlb(vcpu, true);
8223 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8224 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8228 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8229 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8230 vcpu->mmio_needed = 0;
8234 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8235 /* Page is swapped out. Do synthetic halt */
8236 vcpu->arch.apf.halted = true;
8240 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8241 record_steal_time(vcpu);
8242 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8244 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8246 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8247 kvm_pmu_handle_event(vcpu);
8248 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8249 kvm_pmu_deliver_pmi(vcpu);
8250 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8251 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8252 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8253 vcpu->arch.ioapic_handled_vectors)) {
8254 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8255 vcpu->run->eoi.vector =
8256 vcpu->arch.pending_ioapic_eoi;
8261 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8262 vcpu_scan_ioapic(vcpu);
8263 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8264 vcpu_load_eoi_exitmap(vcpu);
8265 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8266 kvm_vcpu_reload_apic_access_page(vcpu);
8267 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8268 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8269 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8273 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8274 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8275 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8279 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8280 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8281 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8287 * KVM_REQ_HV_STIMER has to be processed after
8288 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8289 * depend on the guest clock being up-to-date
8291 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8292 kvm_hv_process_stimers(vcpu);
8293 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8294 kvm_vcpu_update_apicv(vcpu);
8297 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8298 ++vcpu->stat.req_event;
8299 kvm_apic_accept_events(vcpu);
8300 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8305 if (inject_pending_event(vcpu) != 0)
8306 req_immediate_exit = true;
8308 /* Enable SMI/NMI/IRQ window open exits if needed.
8310 * SMIs have three cases:
8311 * 1) They can be nested, and then there is nothing to
8312 * do here because RSM will cause a vmexit anyway.
8313 * 2) There is an ISA-specific reason why SMI cannot be
8314 * injected, and the moment when this changes can be
8316 * 3) Or the SMI can be pending because
8317 * inject_pending_event has completed the injection
8318 * of an IRQ or NMI from the previous vmexit, and
8319 * then we request an immediate exit to inject the
8322 if (vcpu->arch.smi_pending && !is_smm(vcpu))
8323 if (!kvm_x86_ops.enable_smi_window(vcpu))
8324 req_immediate_exit = true;
8325 if (vcpu->arch.nmi_pending)
8326 kvm_x86_ops.enable_nmi_window(vcpu);
8327 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8328 kvm_x86_ops.enable_irq_window(vcpu);
8329 WARN_ON(vcpu->arch.exception.pending);
8332 if (kvm_lapic_enabled(vcpu)) {
8333 update_cr8_intercept(vcpu);
8334 kvm_lapic_sync_to_vapic(vcpu);
8338 r = kvm_mmu_reload(vcpu);
8340 goto cancel_injection;
8345 kvm_x86_ops.prepare_guest_switch(vcpu);
8348 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8349 * IPI are then delayed after guest entry, which ensures that they
8350 * result in virtual interrupt delivery.
8352 local_irq_disable();
8353 vcpu->mode = IN_GUEST_MODE;
8355 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8358 * 1) We should set ->mode before checking ->requests. Please see
8359 * the comment in kvm_vcpu_exiting_guest_mode().
8361 * 2) For APICv, we should set ->mode before checking PID.ON. This
8362 * pairs with the memory barrier implicit in pi_test_and_set_on
8363 * (see vmx_deliver_posted_interrupt).
8365 * 3) This also orders the write to mode from any reads to the page
8366 * tables done while the VCPU is running. Please see the comment
8367 * in kvm_flush_remote_tlbs.
8369 smp_mb__after_srcu_read_unlock();
8372 * This handles the case where a posted interrupt was
8373 * notified with kvm_vcpu_kick.
8375 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8376 kvm_x86_ops.sync_pir_to_irr(vcpu);
8378 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8379 || need_resched() || signal_pending(current)) {
8380 vcpu->mode = OUTSIDE_GUEST_MODE;
8384 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8386 goto cancel_injection;
8389 if (req_immediate_exit) {
8390 kvm_make_request(KVM_REQ_EVENT, vcpu);
8391 kvm_x86_ops.request_immediate_exit(vcpu);
8394 trace_kvm_entry(vcpu->vcpu_id);
8395 guest_enter_irqoff();
8397 fpregs_assert_state_consistent();
8398 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8399 switch_fpu_return();
8401 if (unlikely(vcpu->arch.switch_db_regs)) {
8403 set_debugreg(vcpu->arch.eff_db[0], 0);
8404 set_debugreg(vcpu->arch.eff_db[1], 1);
8405 set_debugreg(vcpu->arch.eff_db[2], 2);
8406 set_debugreg(vcpu->arch.eff_db[3], 3);
8407 set_debugreg(vcpu->arch.dr6, 6);
8408 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8411 kvm_x86_ops.run(vcpu);
8414 * Do this here before restoring debug registers on the host. And
8415 * since we do this before handling the vmexit, a DR access vmexit
8416 * can (a) read the correct value of the debug registers, (b) set
8417 * KVM_DEBUGREG_WONT_EXIT again.
8419 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8420 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8421 kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8422 kvm_update_dr0123(vcpu);
8423 kvm_update_dr6(vcpu);
8424 kvm_update_dr7(vcpu);
8425 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8429 * If the guest has used debug registers, at least dr7
8430 * will be disabled while returning to the host.
8431 * If we don't have active breakpoints in the host, we don't
8432 * care about the messed up debug address registers. But if
8433 * we have some of them active, restore the old state.
8435 if (hw_breakpoint_active())
8436 hw_breakpoint_restore();
8438 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8440 vcpu->mode = OUTSIDE_GUEST_MODE;
8443 kvm_x86_ops.handle_exit_irqoff(vcpu, &exit_fastpath);
8446 * Consume any pending interrupts, including the possible source of
8447 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8448 * An instruction is required after local_irq_enable() to fully unblock
8449 * interrupts on processors that implement an interrupt shadow, the
8450 * stat.exits increment will do nicely.
8452 kvm_before_interrupt(vcpu);
8455 local_irq_disable();
8456 kvm_after_interrupt(vcpu);
8458 guest_exit_irqoff();
8459 if (lapic_in_kernel(vcpu)) {
8460 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8461 if (delta != S64_MIN) {
8462 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8463 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8470 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8473 * Profile KVM exit RIPs:
8475 if (unlikely(prof_on == KVM_PROFILING)) {
8476 unsigned long rip = kvm_rip_read(vcpu);
8477 profile_hit(KVM_PROFILING, (void *)rip);
8480 if (unlikely(vcpu->arch.tsc_always_catchup))
8481 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8483 if (vcpu->arch.apic_attention)
8484 kvm_lapic_sync_from_vapic(vcpu);
8486 r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
8490 kvm_x86_ops.cancel_injection(vcpu);
8491 if (unlikely(vcpu->arch.apic_attention))
8492 kvm_lapic_sync_from_vapic(vcpu);
8497 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8499 if (!kvm_arch_vcpu_runnable(vcpu) &&
8500 (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
8501 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8502 kvm_vcpu_block(vcpu);
8503 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8505 if (kvm_x86_ops.post_block)
8506 kvm_x86_ops.post_block(vcpu);
8508 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8512 kvm_apic_accept_events(vcpu);
8513 switch(vcpu->arch.mp_state) {
8514 case KVM_MP_STATE_HALTED:
8515 vcpu->arch.pv.pv_unhalted = false;
8516 vcpu->arch.mp_state =
8517 KVM_MP_STATE_RUNNABLE;
8519 case KVM_MP_STATE_RUNNABLE:
8520 vcpu->arch.apf.halted = false;
8522 case KVM_MP_STATE_INIT_RECEIVED:
8530 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8532 if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events)
8533 kvm_x86_ops.check_nested_events(vcpu);
8535 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8536 !vcpu->arch.apf.halted);
8539 static int vcpu_run(struct kvm_vcpu *vcpu)
8542 struct kvm *kvm = vcpu->kvm;
8544 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8545 vcpu->arch.l1tf_flush_l1d = true;
8548 if (kvm_vcpu_running(vcpu)) {
8549 r = vcpu_enter_guest(vcpu);
8551 r = vcpu_block(kvm, vcpu);
8557 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8558 if (kvm_cpu_has_pending_timer(vcpu))
8559 kvm_inject_pending_timer_irqs(vcpu);
8561 if (dm_request_for_irq_injection(vcpu) &&
8562 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8564 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8565 ++vcpu->stat.request_irq_exits;
8569 kvm_check_async_pf_completion(vcpu);
8571 if (signal_pending(current)) {
8573 vcpu->run->exit_reason = KVM_EXIT_INTR;
8574 ++vcpu->stat.signal_exits;
8577 if (need_resched()) {
8578 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8580 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8584 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8589 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8593 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8594 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8595 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8599 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8601 BUG_ON(!vcpu->arch.pio.count);
8603 return complete_emulated_io(vcpu);
8607 * Implements the following, as a state machine:
8611 * for each mmio piece in the fragment
8619 * for each mmio piece in the fragment
8624 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8626 struct kvm_run *run = vcpu->run;
8627 struct kvm_mmio_fragment *frag;
8630 BUG_ON(!vcpu->mmio_needed);
8632 /* Complete previous fragment */
8633 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8634 len = min(8u, frag->len);
8635 if (!vcpu->mmio_is_write)
8636 memcpy(frag->data, run->mmio.data, len);
8638 if (frag->len <= 8) {
8639 /* Switch to the next fragment. */
8641 vcpu->mmio_cur_fragment++;
8643 /* Go forward to the next mmio piece. */
8649 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8650 vcpu->mmio_needed = 0;
8652 /* FIXME: return into emulator if single-stepping. */
8653 if (vcpu->mmio_is_write)
8655 vcpu->mmio_read_completed = 1;
8656 return complete_emulated_io(vcpu);
8659 run->exit_reason = KVM_EXIT_MMIO;
8660 run->mmio.phys_addr = frag->gpa;
8661 if (vcpu->mmio_is_write)
8662 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8663 run->mmio.len = min(8u, frag->len);
8664 run->mmio.is_write = vcpu->mmio_is_write;
8665 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8669 static void kvm_save_current_fpu(struct fpu *fpu)
8672 * If the target FPU state is not resident in the CPU registers, just
8673 * memcpy() from current, else save CPU state directly to the target.
8675 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8676 memcpy(&fpu->state, ¤t->thread.fpu.state,
8677 fpu_kernel_xstate_size);
8679 copy_fpregs_to_fpstate(fpu);
8682 /* Swap (qemu) user FPU context for the guest FPU context. */
8683 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8687 kvm_save_current_fpu(vcpu->arch.user_fpu);
8689 /* PKRU is separately restored in kvm_x86_ops.run. */
8690 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8691 ~XFEATURE_MASK_PKRU);
8693 fpregs_mark_activate();
8699 /* When vcpu_run ends, restore user space FPU context. */
8700 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8704 kvm_save_current_fpu(vcpu->arch.guest_fpu);
8706 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8708 fpregs_mark_activate();
8711 ++vcpu->stat.fpu_reload;
8715 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8720 kvm_sigset_activate(vcpu);
8721 kvm_load_guest_fpu(vcpu);
8723 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8724 if (kvm_run->immediate_exit) {
8728 kvm_vcpu_block(vcpu);
8729 kvm_apic_accept_events(vcpu);
8730 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8732 if (signal_pending(current)) {
8734 vcpu->run->exit_reason = KVM_EXIT_INTR;
8735 ++vcpu->stat.signal_exits;
8740 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8745 if (vcpu->run->kvm_dirty_regs) {
8746 r = sync_regs(vcpu);
8751 /* re-sync apic's tpr */
8752 if (!lapic_in_kernel(vcpu)) {
8753 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8759 if (unlikely(vcpu->arch.complete_userspace_io)) {
8760 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8761 vcpu->arch.complete_userspace_io = NULL;
8766 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8768 if (kvm_run->immediate_exit)
8774 kvm_put_guest_fpu(vcpu);
8775 if (vcpu->run->kvm_valid_regs)
8777 post_kvm_run_save(vcpu);
8778 kvm_sigset_deactivate(vcpu);
8784 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8786 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8788 * We are here if userspace calls get_regs() in the middle of
8789 * instruction emulation. Registers state needs to be copied
8790 * back from emulation context to vcpu. Userspace shouldn't do
8791 * that usually, but some bad designed PV devices (vmware
8792 * backdoor interface) need this to work
8794 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
8795 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8797 regs->rax = kvm_rax_read(vcpu);
8798 regs->rbx = kvm_rbx_read(vcpu);
8799 regs->rcx = kvm_rcx_read(vcpu);
8800 regs->rdx = kvm_rdx_read(vcpu);
8801 regs->rsi = kvm_rsi_read(vcpu);
8802 regs->rdi = kvm_rdi_read(vcpu);
8803 regs->rsp = kvm_rsp_read(vcpu);
8804 regs->rbp = kvm_rbp_read(vcpu);
8805 #ifdef CONFIG_X86_64
8806 regs->r8 = kvm_r8_read(vcpu);
8807 regs->r9 = kvm_r9_read(vcpu);
8808 regs->r10 = kvm_r10_read(vcpu);
8809 regs->r11 = kvm_r11_read(vcpu);
8810 regs->r12 = kvm_r12_read(vcpu);
8811 regs->r13 = kvm_r13_read(vcpu);
8812 regs->r14 = kvm_r14_read(vcpu);
8813 regs->r15 = kvm_r15_read(vcpu);
8816 regs->rip = kvm_rip_read(vcpu);
8817 regs->rflags = kvm_get_rflags(vcpu);
8820 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8823 __get_regs(vcpu, regs);
8828 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8830 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8831 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8833 kvm_rax_write(vcpu, regs->rax);
8834 kvm_rbx_write(vcpu, regs->rbx);
8835 kvm_rcx_write(vcpu, regs->rcx);
8836 kvm_rdx_write(vcpu, regs->rdx);
8837 kvm_rsi_write(vcpu, regs->rsi);
8838 kvm_rdi_write(vcpu, regs->rdi);
8839 kvm_rsp_write(vcpu, regs->rsp);
8840 kvm_rbp_write(vcpu, regs->rbp);
8841 #ifdef CONFIG_X86_64
8842 kvm_r8_write(vcpu, regs->r8);
8843 kvm_r9_write(vcpu, regs->r9);
8844 kvm_r10_write(vcpu, regs->r10);
8845 kvm_r11_write(vcpu, regs->r11);
8846 kvm_r12_write(vcpu, regs->r12);
8847 kvm_r13_write(vcpu, regs->r13);
8848 kvm_r14_write(vcpu, regs->r14);
8849 kvm_r15_write(vcpu, regs->r15);
8852 kvm_rip_write(vcpu, regs->rip);
8853 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8855 vcpu->arch.exception.pending = false;
8857 kvm_make_request(KVM_REQ_EVENT, vcpu);
8860 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8863 __set_regs(vcpu, regs);
8868 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8870 struct kvm_segment cs;
8872 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8876 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8878 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8882 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8883 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8884 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8885 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8886 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8887 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8889 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8890 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8892 kvm_x86_ops.get_idt(vcpu, &dt);
8893 sregs->idt.limit = dt.size;
8894 sregs->idt.base = dt.address;
8895 kvm_x86_ops.get_gdt(vcpu, &dt);
8896 sregs->gdt.limit = dt.size;
8897 sregs->gdt.base = dt.address;
8899 sregs->cr0 = kvm_read_cr0(vcpu);
8900 sregs->cr2 = vcpu->arch.cr2;
8901 sregs->cr3 = kvm_read_cr3(vcpu);
8902 sregs->cr4 = kvm_read_cr4(vcpu);
8903 sregs->cr8 = kvm_get_cr8(vcpu);
8904 sregs->efer = vcpu->arch.efer;
8905 sregs->apic_base = kvm_get_apic_base(vcpu);
8907 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8909 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8910 set_bit(vcpu->arch.interrupt.nr,
8911 (unsigned long *)sregs->interrupt_bitmap);
8914 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8915 struct kvm_sregs *sregs)
8918 __get_sregs(vcpu, sregs);
8923 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8924 struct kvm_mp_state *mp_state)
8927 if (kvm_mpx_supported())
8928 kvm_load_guest_fpu(vcpu);
8930 kvm_apic_accept_events(vcpu);
8931 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8932 vcpu->arch.pv.pv_unhalted)
8933 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8935 mp_state->mp_state = vcpu->arch.mp_state;
8937 if (kvm_mpx_supported())
8938 kvm_put_guest_fpu(vcpu);
8943 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8944 struct kvm_mp_state *mp_state)
8950 if (!lapic_in_kernel(vcpu) &&
8951 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8955 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
8956 * INIT state; latched init should be reported using
8957 * KVM_SET_VCPU_EVENTS, so reject it here.
8959 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
8960 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8961 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8964 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8965 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8966 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8968 vcpu->arch.mp_state = mp_state->mp_state;
8969 kvm_make_request(KVM_REQ_EVENT, vcpu);
8977 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8978 int reason, bool has_error_code, u32 error_code)
8980 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8983 init_emulate_ctxt(vcpu);
8985 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8986 has_error_code, error_code);
8988 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8989 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8990 vcpu->run->internal.ndata = 0;
8994 kvm_rip_write(vcpu, ctxt->eip);
8995 kvm_set_rflags(vcpu, ctxt->eflags);
8998 EXPORT_SYMBOL_GPL(kvm_task_switch);
9000 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9002 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9004 * When EFER.LME and CR0.PG are set, the processor is in
9005 * 64-bit mode (though maybe in a 32-bit code segment).
9006 * CR4.PAE and EFER.LMA must be set.
9008 if (!(sregs->cr4 & X86_CR4_PAE)
9009 || !(sregs->efer & EFER_LMA))
9013 * Not in 64-bit mode: EFER.LMA is clear and the code
9014 * segment cannot be 64-bit.
9016 if (sregs->efer & EFER_LMA || sregs->cs.l)
9020 return kvm_valid_cr4(vcpu, sregs->cr4);
9023 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9025 struct msr_data apic_base_msr;
9026 int mmu_reset_needed = 0;
9027 int cpuid_update_needed = 0;
9028 int pending_vec, max_bits, idx;
9032 if (kvm_valid_sregs(vcpu, sregs))
9035 apic_base_msr.data = sregs->apic_base;
9036 apic_base_msr.host_initiated = true;
9037 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9040 dt.size = sregs->idt.limit;
9041 dt.address = sregs->idt.base;
9042 kvm_x86_ops.set_idt(vcpu, &dt);
9043 dt.size = sregs->gdt.limit;
9044 dt.address = sregs->gdt.base;
9045 kvm_x86_ops.set_gdt(vcpu, &dt);
9047 vcpu->arch.cr2 = sregs->cr2;
9048 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9049 vcpu->arch.cr3 = sregs->cr3;
9050 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9052 kvm_set_cr8(vcpu, sregs->cr8);
9054 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9055 kvm_x86_ops.set_efer(vcpu, sregs->efer);
9057 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9058 kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9059 vcpu->arch.cr0 = sregs->cr0;
9061 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9062 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9063 (X86_CR4_OSXSAVE | X86_CR4_PKE));
9064 kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9065 if (cpuid_update_needed)
9066 kvm_update_cpuid(vcpu);
9068 idx = srcu_read_lock(&vcpu->kvm->srcu);
9069 if (is_pae_paging(vcpu)) {
9070 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9071 mmu_reset_needed = 1;
9073 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9075 if (mmu_reset_needed)
9076 kvm_mmu_reset_context(vcpu);
9078 max_bits = KVM_NR_INTERRUPTS;
9079 pending_vec = find_first_bit(
9080 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9081 if (pending_vec < max_bits) {
9082 kvm_queue_interrupt(vcpu, pending_vec, false);
9083 pr_debug("Set back pending irq %d\n", pending_vec);
9086 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9087 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9088 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9089 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9090 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9091 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9093 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9094 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9096 update_cr8_intercept(vcpu);
9098 /* Older userspace won't unhalt the vcpu on reset. */
9099 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9100 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9102 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9104 kvm_make_request(KVM_REQ_EVENT, vcpu);
9111 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9112 struct kvm_sregs *sregs)
9117 ret = __set_sregs(vcpu, sregs);
9122 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9123 struct kvm_guest_debug *dbg)
9125 unsigned long rflags;
9130 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9132 if (vcpu->arch.exception.pending)
9134 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9135 kvm_queue_exception(vcpu, DB_VECTOR);
9137 kvm_queue_exception(vcpu, BP_VECTOR);
9141 * Read rflags as long as potentially injected trace flags are still
9144 rflags = kvm_get_rflags(vcpu);
9146 vcpu->guest_debug = dbg->control;
9147 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9148 vcpu->guest_debug = 0;
9150 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9151 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9152 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9153 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9155 for (i = 0; i < KVM_NR_DB_REGS; i++)
9156 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9158 kvm_update_dr7(vcpu);
9160 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9161 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9162 get_segment_base(vcpu, VCPU_SREG_CS);
9165 * Trigger an rflags update that will inject or remove the trace
9168 kvm_set_rflags(vcpu, rflags);
9170 kvm_x86_ops.update_bp_intercept(vcpu);
9180 * Translate a guest virtual address to a guest physical address.
9182 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9183 struct kvm_translation *tr)
9185 unsigned long vaddr = tr->linear_address;
9191 idx = srcu_read_lock(&vcpu->kvm->srcu);
9192 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9193 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9194 tr->physical_address = gpa;
9195 tr->valid = gpa != UNMAPPED_GVA;
9203 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9205 struct fxregs_state *fxsave;
9209 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9210 memcpy(fpu->fpr, fxsave->st_space, 128);
9211 fpu->fcw = fxsave->cwd;
9212 fpu->fsw = fxsave->swd;
9213 fpu->ftwx = fxsave->twd;
9214 fpu->last_opcode = fxsave->fop;
9215 fpu->last_ip = fxsave->rip;
9216 fpu->last_dp = fxsave->rdp;
9217 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9223 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9225 struct fxregs_state *fxsave;
9229 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9231 memcpy(fxsave->st_space, fpu->fpr, 128);
9232 fxsave->cwd = fpu->fcw;
9233 fxsave->swd = fpu->fsw;
9234 fxsave->twd = fpu->ftwx;
9235 fxsave->fop = fpu->last_opcode;
9236 fxsave->rip = fpu->last_ip;
9237 fxsave->rdp = fpu->last_dp;
9238 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9244 static void store_regs(struct kvm_vcpu *vcpu)
9246 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9248 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9249 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9251 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9252 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9254 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9255 kvm_vcpu_ioctl_x86_get_vcpu_events(
9256 vcpu, &vcpu->run->s.regs.events);
9259 static int sync_regs(struct kvm_vcpu *vcpu)
9261 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9264 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9265 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9266 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9268 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9269 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9271 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9273 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9274 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9275 vcpu, &vcpu->run->s.regs.events))
9277 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9283 static void fx_init(struct kvm_vcpu *vcpu)
9285 fpstate_init(&vcpu->arch.guest_fpu->state);
9286 if (boot_cpu_has(X86_FEATURE_XSAVES))
9287 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9288 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9291 * Ensure guest xcr0 is valid for loading
9293 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9295 vcpu->arch.cr0 |= X86_CR0_ET;
9298 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9300 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9301 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9302 "guest TSC will not be reliable\n");
9307 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9312 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9313 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9315 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9317 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9319 r = kvm_mmu_create(vcpu);
9323 if (irqchip_in_kernel(vcpu->kvm)) {
9324 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9326 goto fail_mmu_destroy;
9327 if (kvm_apicv_activated(vcpu->kvm))
9328 vcpu->arch.apicv_active = true;
9330 static_key_slow_inc(&kvm_no_apic_vcpu);
9334 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9336 goto fail_free_lapic;
9337 vcpu->arch.pio_data = page_address(page);
9339 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9340 GFP_KERNEL_ACCOUNT);
9341 if (!vcpu->arch.mce_banks)
9342 goto fail_free_pio_data;
9343 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9345 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9346 GFP_KERNEL_ACCOUNT))
9347 goto fail_free_mce_banks;
9349 if (!alloc_emulate_ctxt(vcpu))
9350 goto free_wbinvd_dirty_mask;
9352 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9353 GFP_KERNEL_ACCOUNT);
9354 if (!vcpu->arch.user_fpu) {
9355 pr_err("kvm: failed to allocate userspace's fpu\n");
9356 goto free_emulate_ctxt;
9359 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9360 GFP_KERNEL_ACCOUNT);
9361 if (!vcpu->arch.guest_fpu) {
9362 pr_err("kvm: failed to allocate vcpu's fpu\n");
9367 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9369 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9371 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9373 kvm_async_pf_hash_reset(vcpu);
9376 vcpu->arch.pending_external_vector = -1;
9377 vcpu->arch.preempted_in_kernel = false;
9379 kvm_hv_vcpu_init(vcpu);
9381 r = kvm_x86_ops.vcpu_create(vcpu);
9383 goto free_guest_fpu;
9385 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9386 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9387 kvm_vcpu_mtrr_init(vcpu);
9389 kvm_vcpu_reset(vcpu, false);
9390 kvm_init_mmu(vcpu, false);
9395 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9397 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9399 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9400 free_wbinvd_dirty_mask:
9401 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9402 fail_free_mce_banks:
9403 kfree(vcpu->arch.mce_banks);
9405 free_page((unsigned long)vcpu->arch.pio_data);
9407 kvm_free_lapic(vcpu);
9409 kvm_mmu_destroy(vcpu);
9413 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9415 struct msr_data msr;
9416 struct kvm *kvm = vcpu->kvm;
9418 kvm_hv_vcpu_postcreate(vcpu);
9420 if (mutex_lock_killable(&vcpu->mutex))
9424 msr.index = MSR_IA32_TSC;
9425 msr.host_initiated = true;
9426 kvm_write_tsc(vcpu, &msr);
9429 /* poll control enabled by default */
9430 vcpu->arch.msr_kvm_poll_control = 1;
9432 mutex_unlock(&vcpu->mutex);
9434 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9435 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9436 KVMCLOCK_SYNC_PERIOD);
9439 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9441 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9444 kvm_release_pfn(cache->pfn, cache->dirty, cache);
9446 kvmclock_reset(vcpu);
9448 kvm_x86_ops.vcpu_free(vcpu);
9450 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9451 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9452 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9453 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9455 kvm_hv_vcpu_uninit(vcpu);
9456 kvm_pmu_destroy(vcpu);
9457 kfree(vcpu->arch.mce_banks);
9458 kvm_free_lapic(vcpu);
9459 idx = srcu_read_lock(&vcpu->kvm->srcu);
9460 kvm_mmu_destroy(vcpu);
9461 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9462 free_page((unsigned long)vcpu->arch.pio_data);
9463 if (!lapic_in_kernel(vcpu))
9464 static_key_slow_dec(&kvm_no_apic_vcpu);
9467 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9469 kvm_lapic_reset(vcpu, init_event);
9471 vcpu->arch.hflags = 0;
9473 vcpu->arch.smi_pending = 0;
9474 vcpu->arch.smi_count = 0;
9475 atomic_set(&vcpu->arch.nmi_queued, 0);
9476 vcpu->arch.nmi_pending = 0;
9477 vcpu->arch.nmi_injected = false;
9478 kvm_clear_interrupt_queue(vcpu);
9479 kvm_clear_exception_queue(vcpu);
9481 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9482 kvm_update_dr0123(vcpu);
9483 vcpu->arch.dr6 = DR6_INIT;
9484 kvm_update_dr6(vcpu);
9485 vcpu->arch.dr7 = DR7_FIXED_1;
9486 kvm_update_dr7(vcpu);
9490 kvm_make_request(KVM_REQ_EVENT, vcpu);
9491 vcpu->arch.apf.msr_val = 0;
9492 vcpu->arch.st.msr_val = 0;
9494 kvmclock_reset(vcpu);
9496 kvm_clear_async_pf_completion_queue(vcpu);
9497 kvm_async_pf_hash_reset(vcpu);
9498 vcpu->arch.apf.halted = false;
9500 if (kvm_mpx_supported()) {
9501 void *mpx_state_buffer;
9504 * To avoid have the INIT path from kvm_apic_has_events() that be
9505 * called with loaded FPU and does not let userspace fix the state.
9508 kvm_put_guest_fpu(vcpu);
9509 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9511 if (mpx_state_buffer)
9512 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9513 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9515 if (mpx_state_buffer)
9516 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9518 kvm_load_guest_fpu(vcpu);
9522 kvm_pmu_reset(vcpu);
9523 vcpu->arch.smbase = 0x30000;
9525 vcpu->arch.msr_misc_features_enables = 0;
9527 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9530 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9531 vcpu->arch.regs_avail = ~0;
9532 vcpu->arch.regs_dirty = ~0;
9534 vcpu->arch.ia32_xss = 0;
9536 kvm_x86_ops.vcpu_reset(vcpu, init_event);
9539 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9541 struct kvm_segment cs;
9543 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9544 cs.selector = vector << 8;
9545 cs.base = vector << 12;
9546 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9547 kvm_rip_write(vcpu, 0);
9550 int kvm_arch_hardware_enable(void)
9553 struct kvm_vcpu *vcpu;
9558 bool stable, backwards_tsc = false;
9560 kvm_shared_msr_cpu_online();
9561 ret = kvm_x86_ops.hardware_enable();
9565 local_tsc = rdtsc();
9566 stable = !kvm_check_tsc_unstable();
9567 list_for_each_entry(kvm, &vm_list, vm_list) {
9568 kvm_for_each_vcpu(i, vcpu, kvm) {
9569 if (!stable && vcpu->cpu == smp_processor_id())
9570 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9571 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9572 backwards_tsc = true;
9573 if (vcpu->arch.last_host_tsc > max_tsc)
9574 max_tsc = vcpu->arch.last_host_tsc;
9580 * Sometimes, even reliable TSCs go backwards. This happens on
9581 * platforms that reset TSC during suspend or hibernate actions, but
9582 * maintain synchronization. We must compensate. Fortunately, we can
9583 * detect that condition here, which happens early in CPU bringup,
9584 * before any KVM threads can be running. Unfortunately, we can't
9585 * bring the TSCs fully up to date with real time, as we aren't yet far
9586 * enough into CPU bringup that we know how much real time has actually
9587 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9588 * variables that haven't been updated yet.
9590 * So we simply find the maximum observed TSC above, then record the
9591 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9592 * the adjustment will be applied. Note that we accumulate
9593 * adjustments, in case multiple suspend cycles happen before some VCPU
9594 * gets a chance to run again. In the event that no KVM threads get a
9595 * chance to run, we will miss the entire elapsed period, as we'll have
9596 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9597 * loose cycle time. This isn't too big a deal, since the loss will be
9598 * uniform across all VCPUs (not to mention the scenario is extremely
9599 * unlikely). It is possible that a second hibernate recovery happens
9600 * much faster than a first, causing the observed TSC here to be
9601 * smaller; this would require additional padding adjustment, which is
9602 * why we set last_host_tsc to the local tsc observed here.
9604 * N.B. - this code below runs only on platforms with reliable TSC,
9605 * as that is the only way backwards_tsc is set above. Also note
9606 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9607 * have the same delta_cyc adjustment applied if backwards_tsc
9608 * is detected. Note further, this adjustment is only done once,
9609 * as we reset last_host_tsc on all VCPUs to stop this from being
9610 * called multiple times (one for each physical CPU bringup).
9612 * Platforms with unreliable TSCs don't have to deal with this, they
9613 * will be compensated by the logic in vcpu_load, which sets the TSC to
9614 * catchup mode. This will catchup all VCPUs to real time, but cannot
9615 * guarantee that they stay in perfect synchronization.
9617 if (backwards_tsc) {
9618 u64 delta_cyc = max_tsc - local_tsc;
9619 list_for_each_entry(kvm, &vm_list, vm_list) {
9620 kvm->arch.backwards_tsc_observed = true;
9621 kvm_for_each_vcpu(i, vcpu, kvm) {
9622 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9623 vcpu->arch.last_host_tsc = local_tsc;
9624 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9628 * We have to disable TSC offset matching.. if you were
9629 * booting a VM while issuing an S4 host suspend....
9630 * you may have some problem. Solving this issue is
9631 * left as an exercise to the reader.
9633 kvm->arch.last_tsc_nsec = 0;
9634 kvm->arch.last_tsc_write = 0;
9641 void kvm_arch_hardware_disable(void)
9643 kvm_x86_ops.hardware_disable();
9644 drop_user_return_notifiers();
9647 int kvm_arch_hardware_setup(void *opaque)
9649 struct kvm_x86_init_ops *ops = opaque;
9652 rdmsrl_safe(MSR_EFER, &host_efer);
9654 if (boot_cpu_has(X86_FEATURE_XSAVES))
9655 rdmsrl(MSR_IA32_XSS, host_xss);
9657 r = ops->hardware_setup();
9661 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9663 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9666 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
9667 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
9668 #undef __kvm_cpu_cap_has
9670 if (kvm_has_tsc_control) {
9672 * Make sure the user can only configure tsc_khz values that
9673 * fit into a signed integer.
9674 * A min value is not calculated because it will always
9675 * be 1 on all machines.
9677 u64 max = min(0x7fffffffULL,
9678 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9679 kvm_max_guest_tsc_khz = max;
9681 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9684 kvm_init_msr_list();
9688 void kvm_arch_hardware_unsetup(void)
9690 kvm_x86_ops.hardware_unsetup();
9693 int kvm_arch_check_processor_compat(void *opaque)
9695 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
9696 struct kvm_x86_init_ops *ops = opaque;
9698 WARN_ON(!irqs_disabled());
9700 if (__cr4_reserved_bits(cpu_has, c) !=
9701 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
9704 return ops->check_processor_compatibility();
9707 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9709 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9711 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9713 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9715 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9718 struct static_key kvm_no_apic_vcpu __read_mostly;
9719 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9721 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9723 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9725 vcpu->arch.l1tf_flush_l1d = true;
9726 if (pmu->version && unlikely(pmu->event_count)) {
9727 pmu->need_cleanup = true;
9728 kvm_make_request(KVM_REQ_PMU, vcpu);
9730 kvm_x86_ops.sched_in(vcpu, cpu);
9733 void kvm_arch_free_vm(struct kvm *kvm)
9735 kfree(kvm->arch.hyperv.hv_pa_pg);
9740 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9745 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9746 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9747 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9748 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9749 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9750 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9752 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9753 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9754 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9755 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9756 &kvm->arch.irq_sources_bitmap);
9758 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9759 mutex_init(&kvm->arch.apic_map_lock);
9760 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9762 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
9763 pvclock_update_vm_gtod_copy(kvm);
9765 kvm->arch.guest_can_read_msr_platform_info = true;
9767 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9768 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9770 kvm_hv_init_vm(kvm);
9771 kvm_page_track_init(kvm);
9772 kvm_mmu_init_vm(kvm);
9774 return kvm_x86_ops.vm_init(kvm);
9777 int kvm_arch_post_init_vm(struct kvm *kvm)
9779 return kvm_mmu_post_init_vm(kvm);
9782 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9785 kvm_mmu_unload(vcpu);
9789 static void kvm_free_vcpus(struct kvm *kvm)
9792 struct kvm_vcpu *vcpu;
9795 * Unpin any mmu pages first.
9797 kvm_for_each_vcpu(i, vcpu, kvm) {
9798 kvm_clear_async_pf_completion_queue(vcpu);
9799 kvm_unload_vcpu_mmu(vcpu);
9801 kvm_for_each_vcpu(i, vcpu, kvm)
9802 kvm_vcpu_destroy(vcpu);
9804 mutex_lock(&kvm->lock);
9805 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9806 kvm->vcpus[i] = NULL;
9808 atomic_set(&kvm->online_vcpus, 0);
9809 mutex_unlock(&kvm->lock);
9812 void kvm_arch_sync_events(struct kvm *kvm)
9814 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9815 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9819 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9822 unsigned long hva, uninitialized_var(old_npages);
9823 struct kvm_memslots *slots = kvm_memslots(kvm);
9824 struct kvm_memory_slot *slot;
9826 /* Called with kvm->slots_lock held. */
9827 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9830 slot = id_to_memslot(slots, id);
9832 if (slot && slot->npages)
9836 * MAP_SHARED to prevent internal slot pages from being moved
9839 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9840 MAP_SHARED | MAP_ANONYMOUS, 0);
9841 if (IS_ERR((void *)hva))
9842 return PTR_ERR((void *)hva);
9844 if (!slot || !slot->npages)
9848 * Stuff a non-canonical value to catch use-after-delete. This
9849 * ends up being 0 on 32-bit KVM, but there's no better
9852 hva = (unsigned long)(0xdeadull << 48);
9853 old_npages = slot->npages;
9856 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9857 struct kvm_userspace_memory_region m;
9859 m.slot = id | (i << 16);
9861 m.guest_phys_addr = gpa;
9862 m.userspace_addr = hva;
9863 m.memory_size = size;
9864 r = __kvm_set_memory_region(kvm, &m);
9870 vm_munmap(hva, old_npages * PAGE_SIZE);
9874 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9876 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9878 kvm_mmu_pre_destroy_vm(kvm);
9881 void kvm_arch_destroy_vm(struct kvm *kvm)
9883 if (current->mm == kvm->mm) {
9885 * Free memory regions allocated on behalf of userspace,
9886 * unless the the memory map has changed due to process exit
9889 mutex_lock(&kvm->slots_lock);
9890 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
9892 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
9894 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9895 mutex_unlock(&kvm->slots_lock);
9897 if (kvm_x86_ops.vm_destroy)
9898 kvm_x86_ops.vm_destroy(kvm);
9899 kvm_pic_destroy(kvm);
9900 kvm_ioapic_destroy(kvm);
9901 kvm_free_vcpus(kvm);
9902 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9903 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9904 kvm_mmu_uninit_vm(kvm);
9905 kvm_page_track_cleanup(kvm);
9906 kvm_hv_destroy_vm(kvm);
9909 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
9913 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9914 kvfree(slot->arch.rmap[i]);
9915 slot->arch.rmap[i] = NULL;
9920 kvfree(slot->arch.lpage_info[i - 1]);
9921 slot->arch.lpage_info[i - 1] = NULL;
9924 kvm_page_track_free_memslot(slot);
9927 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
9928 unsigned long npages)
9933 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
9934 * old arrays will be freed by __kvm_set_memory_region() if installing
9935 * the new memslot is successful.
9937 memset(&slot->arch, 0, sizeof(slot->arch));
9939 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9940 struct kvm_lpage_info *linfo;
9945 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9946 slot->base_gfn, level) + 1;
9948 slot->arch.rmap[i] =
9949 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9950 GFP_KERNEL_ACCOUNT);
9951 if (!slot->arch.rmap[i])
9956 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9960 slot->arch.lpage_info[i - 1] = linfo;
9962 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9963 linfo[0].disallow_lpage = 1;
9964 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9965 linfo[lpages - 1].disallow_lpage = 1;
9966 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9968 * If the gfn and userspace address are not aligned wrt each
9969 * other, disable large page support for this slot.
9971 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
9974 for (j = 0; j < lpages; ++j)
9975 linfo[j].disallow_lpage = 1;
9979 if (kvm_page_track_create_memslot(slot, npages))
9985 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9986 kvfree(slot->arch.rmap[i]);
9987 slot->arch.rmap[i] = NULL;
9991 kvfree(slot->arch.lpage_info[i - 1]);
9992 slot->arch.lpage_info[i - 1] = NULL;
9997 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9999 struct kvm_vcpu *vcpu;
10003 * memslots->generation has been incremented.
10004 * mmio generation may have reached its maximum value.
10006 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10008 /* Force re-initialization of steal_time cache */
10009 kvm_for_each_vcpu(i, vcpu, kvm)
10010 kvm_vcpu_kick(vcpu);
10013 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10014 struct kvm_memory_slot *memslot,
10015 const struct kvm_userspace_memory_region *mem,
10016 enum kvm_mr_change change)
10018 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10019 return kvm_alloc_memslot_metadata(memslot,
10020 mem->memory_size >> PAGE_SHIFT);
10024 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10025 struct kvm_memory_slot *new)
10027 /* Still write protect RO slot */
10028 if (new->flags & KVM_MEM_READONLY) {
10029 kvm_mmu_slot_remove_write_access(kvm, new, PT_PAGE_TABLE_LEVEL);
10034 * Call kvm_x86_ops dirty logging hooks when they are valid.
10036 * kvm_x86_ops.slot_disable_log_dirty is called when:
10038 * - KVM_MR_CREATE with dirty logging is disabled
10039 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
10041 * The reason is, in case of PML, we need to set D-bit for any slots
10042 * with dirty logging disabled in order to eliminate unnecessary GPA
10043 * logging in PML buffer (and potential PML buffer full VMEXIT). This
10044 * guarantees leaving PML enabled during guest's lifetime won't have
10045 * any additional overhead from PML when guest is running with dirty
10046 * logging disabled for memory slots.
10048 * kvm_x86_ops.slot_enable_log_dirty is called when switching new slot
10049 * to dirty logging mode.
10051 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
10053 * In case of write protect:
10055 * Write protect all pages for dirty logging.
10057 * All the sptes including the large sptes which point to this
10058 * slot are set to readonly. We can not create any new large
10059 * spte on this slot until the end of the logging.
10061 * See the comments in fast_page_fault().
10063 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10064 if (kvm_x86_ops.slot_enable_log_dirty) {
10065 kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10068 kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10069 PT_DIRECTORY_LEVEL : PT_PAGE_TABLE_LEVEL;
10072 * If we're with initial-all-set, we don't need
10073 * to write protect any small page because
10074 * they're reported as dirty already. However
10075 * we still need to write-protect huge pages
10076 * so that the page split can happen lazily on
10077 * the first write to the huge page.
10079 kvm_mmu_slot_remove_write_access(kvm, new, level);
10082 if (kvm_x86_ops.slot_disable_log_dirty)
10083 kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10087 void kvm_arch_commit_memory_region(struct kvm *kvm,
10088 const struct kvm_userspace_memory_region *mem,
10089 struct kvm_memory_slot *old,
10090 const struct kvm_memory_slot *new,
10091 enum kvm_mr_change change)
10093 if (!kvm->arch.n_requested_mmu_pages)
10094 kvm_mmu_change_mmu_pages(kvm,
10095 kvm_mmu_calculate_default_mmu_pages(kvm));
10098 * Dirty logging tracks sptes in 4k granularity, meaning that large
10099 * sptes have to be split. If live migration is successful, the guest
10100 * in the source machine will be destroyed and large sptes will be
10101 * created in the destination. However, if the guest continues to run
10102 * in the source machine (for example if live migration fails), small
10103 * sptes will remain around and cause bad performance.
10105 * Scan sptes if dirty logging has been stopped, dropping those
10106 * which can be collapsed into a single large-page spte. Later
10107 * page faults will create the large-page sptes.
10109 * There is no need to do this in any of the following cases:
10110 * CREATE: No dirty mappings will already exist.
10111 * MOVE/DELETE: The old mappings will already have been cleaned up by
10112 * kvm_arch_flush_shadow_memslot()
10114 if (change == KVM_MR_FLAGS_ONLY &&
10115 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10116 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10117 kvm_mmu_zap_collapsible_sptes(kvm, new);
10120 * Set up write protection and/or dirty logging for the new slot.
10122 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
10123 * been zapped so no dirty logging staff is needed for old slot. For
10124 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
10125 * new and it's also covered when dealing with the new slot.
10127 * FIXME: const-ify all uses of struct kvm_memory_slot.
10129 if (change != KVM_MR_DELETE)
10130 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
10132 /* Free the arrays associated with the old memslot. */
10133 if (change == KVM_MR_MOVE)
10134 kvm_arch_free_memslot(kvm, old);
10137 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10139 kvm_mmu_zap_all(kvm);
10142 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10143 struct kvm_memory_slot *slot)
10145 kvm_page_track_flush_slot(kvm, slot);
10148 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10150 return (is_guest_mode(vcpu) &&
10151 kvm_x86_ops.guest_apic_has_interrupt &&
10152 kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10155 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10157 if (!list_empty_careful(&vcpu->async_pf.done))
10160 if (kvm_apic_has_events(vcpu))
10163 if (vcpu->arch.pv.pv_unhalted)
10166 if (vcpu->arch.exception.pending)
10169 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10170 (vcpu->arch.nmi_pending &&
10171 kvm_x86_ops.nmi_allowed(vcpu)))
10174 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10175 (vcpu->arch.smi_pending && !is_smm(vcpu)))
10178 if (kvm_arch_interrupt_allowed(vcpu) &&
10179 (kvm_cpu_has_interrupt(vcpu) ||
10180 kvm_guest_apic_has_interrupt(vcpu)))
10183 if (kvm_hv_has_stimer_pending(vcpu))
10189 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10191 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10194 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10196 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10199 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10200 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10201 kvm_test_request(KVM_REQ_EVENT, vcpu))
10204 if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10210 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10212 return vcpu->arch.preempted_in_kernel;
10215 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10217 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10220 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10222 return kvm_x86_ops.interrupt_allowed(vcpu);
10225 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10227 if (is_64_bit_mode(vcpu))
10228 return kvm_rip_read(vcpu);
10229 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10230 kvm_rip_read(vcpu));
10232 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10234 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10236 return kvm_get_linear_rip(vcpu) == linear_rip;
10238 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10240 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10242 unsigned long rflags;
10244 rflags = kvm_x86_ops.get_rflags(vcpu);
10245 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10246 rflags &= ~X86_EFLAGS_TF;
10249 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10251 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10253 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10254 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10255 rflags |= X86_EFLAGS_TF;
10256 kvm_x86_ops.set_rflags(vcpu, rflags);
10259 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10261 __kvm_set_rflags(vcpu, rflags);
10262 kvm_make_request(KVM_REQ_EVENT, vcpu);
10264 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10266 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10270 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10274 r = kvm_mmu_reload(vcpu);
10278 if (!vcpu->arch.mmu->direct_map &&
10279 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10282 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10285 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10287 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10290 static inline u32 kvm_async_pf_next_probe(u32 key)
10292 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
10295 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10297 u32 key = kvm_async_pf_hash_fn(gfn);
10299 while (vcpu->arch.apf.gfns[key] != ~0)
10300 key = kvm_async_pf_next_probe(key);
10302 vcpu->arch.apf.gfns[key] = gfn;
10305 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10308 u32 key = kvm_async_pf_hash_fn(gfn);
10310 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
10311 (vcpu->arch.apf.gfns[key] != gfn &&
10312 vcpu->arch.apf.gfns[key] != ~0); i++)
10313 key = kvm_async_pf_next_probe(key);
10318 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10320 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10323 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10327 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10329 vcpu->arch.apf.gfns[i] = ~0;
10331 j = kvm_async_pf_next_probe(j);
10332 if (vcpu->arch.apf.gfns[j] == ~0)
10334 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10336 * k lies cyclically in ]i,j]
10338 * |....j i.k.| or |.k..j i...|
10340 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10341 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10346 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10349 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10353 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10356 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10360 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10362 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10365 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10366 (vcpu->arch.apf.send_user_only &&
10367 kvm_x86_ops.get_cpl(vcpu) == 0))
10373 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10375 if (unlikely(!lapic_in_kernel(vcpu) ||
10376 kvm_event_needs_reinjection(vcpu) ||
10377 vcpu->arch.exception.pending))
10380 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10384 * If interrupts are off we cannot even use an artificial
10387 return kvm_x86_ops.interrupt_allowed(vcpu);
10390 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10391 struct kvm_async_pf *work)
10393 struct x86_exception fault;
10395 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10396 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10398 if (kvm_can_deliver_async_pf(vcpu) &&
10399 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10400 fault.vector = PF_VECTOR;
10401 fault.error_code_valid = true;
10402 fault.error_code = 0;
10403 fault.nested_page_fault = false;
10404 fault.address = work->arch.token;
10405 fault.async_page_fault = true;
10406 kvm_inject_page_fault(vcpu, &fault);
10409 * It is not possible to deliver a paravirtualized asynchronous
10410 * page fault, but putting the guest in an artificial halt state
10411 * can be beneficial nevertheless: if an interrupt arrives, we
10412 * can deliver it timely and perhaps the guest will schedule
10413 * another process. When the instruction that triggered a page
10414 * fault is retried, hopefully the page will be ready in the host.
10416 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10420 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10421 struct kvm_async_pf *work)
10423 struct x86_exception fault;
10426 if (work->wakeup_all)
10427 work->arch.token = ~0; /* broadcast wakeup */
10429 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10430 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10432 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10433 !apf_get_user(vcpu, &val)) {
10434 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10435 vcpu->arch.exception.pending &&
10436 vcpu->arch.exception.nr == PF_VECTOR &&
10437 !apf_put_user(vcpu, 0)) {
10438 vcpu->arch.exception.injected = false;
10439 vcpu->arch.exception.pending = false;
10440 vcpu->arch.exception.nr = 0;
10441 vcpu->arch.exception.has_error_code = false;
10442 vcpu->arch.exception.error_code = 0;
10443 vcpu->arch.exception.has_payload = false;
10444 vcpu->arch.exception.payload = 0;
10445 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10446 fault.vector = PF_VECTOR;
10447 fault.error_code_valid = true;
10448 fault.error_code = 0;
10449 fault.nested_page_fault = false;
10450 fault.address = work->arch.token;
10451 fault.async_page_fault = true;
10452 kvm_inject_page_fault(vcpu, &fault);
10455 vcpu->arch.apf.halted = false;
10456 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10459 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10461 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10464 return kvm_can_do_async_pf(vcpu);
10467 void kvm_arch_start_assignment(struct kvm *kvm)
10469 atomic_inc(&kvm->arch.assigned_device_count);
10471 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10473 void kvm_arch_end_assignment(struct kvm *kvm)
10475 atomic_dec(&kvm->arch.assigned_device_count);
10477 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10479 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10481 return atomic_read(&kvm->arch.assigned_device_count);
10483 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10485 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10487 atomic_inc(&kvm->arch.noncoherent_dma_count);
10489 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10491 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10493 atomic_dec(&kvm->arch.noncoherent_dma_count);
10495 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10497 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10499 return atomic_read(&kvm->arch.noncoherent_dma_count);
10501 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10503 bool kvm_arch_has_irq_bypass(void)
10508 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10509 struct irq_bypass_producer *prod)
10511 struct kvm_kernel_irqfd *irqfd =
10512 container_of(cons, struct kvm_kernel_irqfd, consumer);
10514 irqfd->producer = prod;
10516 return kvm_x86_ops.update_pi_irte(irqfd->kvm,
10517 prod->irq, irqfd->gsi, 1);
10520 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10521 struct irq_bypass_producer *prod)
10524 struct kvm_kernel_irqfd *irqfd =
10525 container_of(cons, struct kvm_kernel_irqfd, consumer);
10527 WARN_ON(irqfd->producer != prod);
10528 irqfd->producer = NULL;
10531 * When producer of consumer is unregistered, we change back to
10532 * remapped mode, so we can re-use the current implementation
10533 * when the irq is masked/disabled or the consumer side (KVM
10534 * int this case doesn't want to receive the interrupts.
10536 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10538 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10539 " fails: %d\n", irqfd->consumer.token, ret);
10542 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10543 uint32_t guest_irq, bool set)
10545 return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
10548 bool kvm_vector_hashing_enabled(void)
10550 return vector_hashing;
10553 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10555 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10557 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10559 u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu)
10561 uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD;
10563 /* The STIBP bit doesn't fault even if it's not advertised */
10564 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
10565 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
10566 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10567 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
10568 !boot_cpu_has(X86_FEATURE_AMD_IBRS))
10569 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10571 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) &&
10572 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
10573 bits &= ~SPEC_CTRL_SSBD;
10574 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
10575 !boot_cpu_has(X86_FEATURE_AMD_SSBD))
10576 bits &= ~SPEC_CTRL_SSBD;
10580 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits);
10582 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10583 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10584 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10585 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10586 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10587 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10588 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10589 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10590 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10591 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10592 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10593 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10594 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10595 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10596 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10597 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10598 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10599 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10600 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10601 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10602 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
10603 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);