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 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);
580 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
582 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
583 u32 error_code, unsigned long payload)
585 kvm_multiple_exception(vcpu, nr, true, error_code,
586 true, payload, false);
589 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
592 kvm_inject_gp(vcpu, 0);
594 return kvm_skip_emulated_instruction(vcpu);
598 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
600 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
602 ++vcpu->stat.pf_guest;
603 vcpu->arch.exception.nested_apf =
604 is_guest_mode(vcpu) && fault->async_page_fault;
605 if (vcpu->arch.exception.nested_apf) {
606 vcpu->arch.apf.nested_apf_token = fault->address;
607 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
609 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
613 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
615 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
617 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
618 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
620 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
622 return fault->nested_page_fault;
625 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
627 atomic_inc(&vcpu->arch.nmi_queued);
628 kvm_make_request(KVM_REQ_NMI, vcpu);
630 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
632 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
634 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
636 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
638 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
640 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
642 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
645 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
646 * a #GP and return false.
648 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
650 if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
652 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
655 EXPORT_SYMBOL_GPL(kvm_require_cpl);
657 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
659 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
662 kvm_queue_exception(vcpu, UD_VECTOR);
665 EXPORT_SYMBOL_GPL(kvm_require_dr);
668 * This function will be used to read from the physical memory of the currently
669 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
670 * can read from guest physical or from the guest's guest physical memory.
672 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
673 gfn_t ngfn, void *data, int offset, int len,
676 struct x86_exception exception;
680 ngpa = gfn_to_gpa(ngfn);
681 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
682 if (real_gfn == UNMAPPED_GVA)
685 real_gfn = gpa_to_gfn(real_gfn);
687 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
689 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
691 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
692 void *data, int offset, int len, u32 access)
694 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
695 data, offset, len, access);
698 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
700 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
705 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
707 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
709 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
710 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
713 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
715 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
716 offset * sizeof(u64), sizeof(pdpte),
717 PFERR_USER_MASK|PFERR_WRITE_MASK);
722 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
723 if ((pdpte[i] & PT_PRESENT_MASK) &&
724 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
731 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
732 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
738 EXPORT_SYMBOL_GPL(load_pdptrs);
740 bool pdptrs_changed(struct kvm_vcpu *vcpu)
742 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
747 if (!is_pae_paging(vcpu))
750 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
753 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
754 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
755 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
756 PFERR_USER_MASK | PFERR_WRITE_MASK);
760 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
762 EXPORT_SYMBOL_GPL(pdptrs_changed);
764 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
766 unsigned long old_cr0 = kvm_read_cr0(vcpu);
767 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
772 if (cr0 & 0xffffffff00000000UL)
776 cr0 &= ~CR0_RESERVED_BITS;
778 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
781 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
784 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
786 if ((vcpu->arch.efer & EFER_LME)) {
791 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
796 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
801 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
804 kvm_x86_ops.set_cr0(vcpu, cr0);
806 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
807 kvm_clear_async_pf_completion_queue(vcpu);
808 kvm_async_pf_hash_reset(vcpu);
811 if ((cr0 ^ old_cr0) & update_bits)
812 kvm_mmu_reset_context(vcpu);
814 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
815 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
816 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
817 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
821 EXPORT_SYMBOL_GPL(kvm_set_cr0);
823 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
825 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
827 EXPORT_SYMBOL_GPL(kvm_lmsw);
829 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
831 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
833 if (vcpu->arch.xcr0 != host_xcr0)
834 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
836 if (vcpu->arch.xsaves_enabled &&
837 vcpu->arch.ia32_xss != host_xss)
838 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
841 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
843 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
845 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
847 if (vcpu->arch.xcr0 != host_xcr0)
848 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
850 if (vcpu->arch.xsaves_enabled &&
851 vcpu->arch.ia32_xss != host_xss)
852 wrmsrl(MSR_IA32_XSS, host_xss);
856 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
858 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
861 u64 old_xcr0 = vcpu->arch.xcr0;
864 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
865 if (index != XCR_XFEATURE_ENABLED_MASK)
867 if (!(xcr0 & XFEATURE_MASK_FP))
869 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
873 * Do not allow the guest to set bits that we do not support
874 * saving. However, xcr0 bit 0 is always set, even if the
875 * emulated CPU does not support XSAVE (see fx_init).
877 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
878 if (xcr0 & ~valid_bits)
881 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
882 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
885 if (xcr0 & XFEATURE_MASK_AVX512) {
886 if (!(xcr0 & XFEATURE_MASK_YMM))
888 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
891 vcpu->arch.xcr0 = xcr0;
893 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
894 kvm_update_cpuid(vcpu);
898 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
900 if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
901 __kvm_set_xcr(vcpu, index, xcr)) {
902 kvm_inject_gp(vcpu, 0);
907 EXPORT_SYMBOL_GPL(kvm_set_xcr);
909 #define __cr4_reserved_bits(__cpu_has, __c) \
911 u64 __reserved_bits = CR4_RESERVED_BITS; \
913 if (!__cpu_has(__c, X86_FEATURE_XSAVE)) \
914 __reserved_bits |= X86_CR4_OSXSAVE; \
915 if (!__cpu_has(__c, X86_FEATURE_SMEP)) \
916 __reserved_bits |= X86_CR4_SMEP; \
917 if (!__cpu_has(__c, X86_FEATURE_SMAP)) \
918 __reserved_bits |= X86_CR4_SMAP; \
919 if (!__cpu_has(__c, X86_FEATURE_FSGSBASE)) \
920 __reserved_bits |= X86_CR4_FSGSBASE; \
921 if (!__cpu_has(__c, X86_FEATURE_PKU)) \
922 __reserved_bits |= X86_CR4_PKE; \
923 if (!__cpu_has(__c, X86_FEATURE_LA57)) \
924 __reserved_bits |= X86_CR4_LA57; \
925 if (!__cpu_has(__c, X86_FEATURE_UMIP)) \
926 __reserved_bits |= X86_CR4_UMIP; \
930 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
932 if (cr4 & cr4_reserved_bits)
935 if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu))
941 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
943 unsigned long old_cr4 = kvm_read_cr4(vcpu);
944 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
945 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
947 if (kvm_valid_cr4(vcpu, cr4))
950 if (is_long_mode(vcpu)) {
951 if (!(cr4 & X86_CR4_PAE))
953 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
954 && ((cr4 ^ old_cr4) & pdptr_bits)
955 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
959 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
960 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
963 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
964 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
968 if (kvm_x86_ops.set_cr4(vcpu, cr4))
971 if (((cr4 ^ old_cr4) & pdptr_bits) ||
972 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
973 kvm_mmu_reset_context(vcpu);
975 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
976 kvm_update_cpuid(vcpu);
980 EXPORT_SYMBOL_GPL(kvm_set_cr4);
982 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
984 bool skip_tlb_flush = false;
986 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
989 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
990 cr3 &= ~X86_CR3_PCID_NOFLUSH;
994 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
995 if (!skip_tlb_flush) {
996 kvm_mmu_sync_roots(vcpu);
997 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1002 if (is_long_mode(vcpu) &&
1003 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
1005 else if (is_pae_paging(vcpu) &&
1006 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1009 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
1010 vcpu->arch.cr3 = cr3;
1011 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1015 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1017 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1019 if (cr8 & CR8_RESERVED_BITS)
1021 if (lapic_in_kernel(vcpu))
1022 kvm_lapic_set_tpr(vcpu, cr8);
1024 vcpu->arch.cr8 = cr8;
1027 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1029 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1031 if (lapic_in_kernel(vcpu))
1032 return kvm_lapic_get_cr8(vcpu);
1034 return vcpu->arch.cr8;
1036 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1038 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1042 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1043 for (i = 0; i < KVM_NR_DB_REGS; i++)
1044 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1045 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1049 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1053 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1054 dr7 = vcpu->arch.guest_debug_dr7;
1056 dr7 = vcpu->arch.dr7;
1057 kvm_x86_ops.set_dr7(vcpu, dr7);
1058 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1059 if (dr7 & DR7_BP_EN_MASK)
1060 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1063 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1065 u64 fixed = DR6_FIXED_1;
1067 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1072 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1074 size_t size = ARRAY_SIZE(vcpu->arch.db);
1078 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1079 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1080 vcpu->arch.eff_db[dr] = val;
1085 if (val & 0xffffffff00000000ULL)
1086 return -1; /* #GP */
1087 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1092 if (!kvm_dr7_valid(val))
1093 return -1; /* #GP */
1094 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1095 kvm_update_dr7(vcpu);
1102 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1104 if (__kvm_set_dr(vcpu, dr, val)) {
1105 kvm_inject_gp(vcpu, 0);
1110 EXPORT_SYMBOL_GPL(kvm_set_dr);
1112 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1114 size_t size = ARRAY_SIZE(vcpu->arch.db);
1118 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1123 *val = vcpu->arch.dr6;
1128 *val = vcpu->arch.dr7;
1133 EXPORT_SYMBOL_GPL(kvm_get_dr);
1135 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1137 u32 ecx = kvm_rcx_read(vcpu);
1141 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1144 kvm_rax_write(vcpu, (u32)data);
1145 kvm_rdx_write(vcpu, data >> 32);
1148 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1151 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1152 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1154 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1155 * extract the supported MSRs from the related const lists.
1156 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1157 * capabilities of the host cpu. This capabilities test skips MSRs that are
1158 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1159 * may depend on host virtualization features rather than host cpu features.
1162 static const u32 msrs_to_save_all[] = {
1163 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1165 #ifdef CONFIG_X86_64
1166 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1168 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1169 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1171 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1172 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1173 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1174 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1175 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1176 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1177 MSR_IA32_UMWAIT_CONTROL,
1179 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1180 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1181 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1182 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1183 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1184 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1185 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1186 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1187 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1188 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1189 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1190 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1191 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1192 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1193 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1194 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1195 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1196 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1197 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1198 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1199 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1200 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1203 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1204 static unsigned num_msrs_to_save;
1206 static const u32 emulated_msrs_all[] = {
1207 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1208 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1209 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1210 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1211 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1212 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1213 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1215 HV_X64_MSR_VP_INDEX,
1216 HV_X64_MSR_VP_RUNTIME,
1217 HV_X64_MSR_SCONTROL,
1218 HV_X64_MSR_STIMER0_CONFIG,
1219 HV_X64_MSR_VP_ASSIST_PAGE,
1220 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1221 HV_X64_MSR_TSC_EMULATION_STATUS,
1223 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1226 MSR_IA32_TSC_ADJUST,
1227 MSR_IA32_TSCDEADLINE,
1228 MSR_IA32_ARCH_CAPABILITIES,
1229 MSR_IA32_MISC_ENABLE,
1230 MSR_IA32_MCG_STATUS,
1232 MSR_IA32_MCG_EXT_CTL,
1236 MSR_MISC_FEATURES_ENABLES,
1237 MSR_AMD64_VIRT_SPEC_CTRL,
1242 * The following list leaves out MSRs whose values are determined
1243 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1244 * We always support the "true" VMX control MSRs, even if the host
1245 * processor does not, so I am putting these registers here rather
1246 * than in msrs_to_save_all.
1249 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1250 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1251 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1252 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1254 MSR_IA32_VMX_CR0_FIXED0,
1255 MSR_IA32_VMX_CR4_FIXED0,
1256 MSR_IA32_VMX_VMCS_ENUM,
1257 MSR_IA32_VMX_PROCBASED_CTLS2,
1258 MSR_IA32_VMX_EPT_VPID_CAP,
1259 MSR_IA32_VMX_VMFUNC,
1262 MSR_KVM_POLL_CONTROL,
1265 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1266 static unsigned num_emulated_msrs;
1269 * List of msr numbers which are used to expose MSR-based features that
1270 * can be used by a hypervisor to validate requested CPU features.
1272 static const u32 msr_based_features_all[] = {
1274 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1275 MSR_IA32_VMX_PINBASED_CTLS,
1276 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1277 MSR_IA32_VMX_PROCBASED_CTLS,
1278 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1279 MSR_IA32_VMX_EXIT_CTLS,
1280 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1281 MSR_IA32_VMX_ENTRY_CTLS,
1283 MSR_IA32_VMX_CR0_FIXED0,
1284 MSR_IA32_VMX_CR0_FIXED1,
1285 MSR_IA32_VMX_CR4_FIXED0,
1286 MSR_IA32_VMX_CR4_FIXED1,
1287 MSR_IA32_VMX_VMCS_ENUM,
1288 MSR_IA32_VMX_PROCBASED_CTLS2,
1289 MSR_IA32_VMX_EPT_VPID_CAP,
1290 MSR_IA32_VMX_VMFUNC,
1294 MSR_IA32_ARCH_CAPABILITIES,
1297 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1298 static unsigned int num_msr_based_features;
1300 static u64 kvm_get_arch_capabilities(void)
1304 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1305 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1308 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1309 * the nested hypervisor runs with NX huge pages. If it is not,
1310 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1311 * L1 guests, so it need not worry about its own (L2) guests.
1313 data |= ARCH_CAP_PSCHANGE_MC_NO;
1316 * If we're doing cache flushes (either "always" or "cond")
1317 * we will do one whenever the guest does a vmlaunch/vmresume.
1318 * If an outer hypervisor is doing the cache flush for us
1319 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1320 * capability to the guest too, and if EPT is disabled we're not
1321 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1322 * require a nested hypervisor to do a flush of its own.
1324 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1325 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1327 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1328 data |= ARCH_CAP_RDCL_NO;
1329 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1330 data |= ARCH_CAP_SSB_NO;
1331 if (!boot_cpu_has_bug(X86_BUG_MDS))
1332 data |= ARCH_CAP_MDS_NO;
1335 * On TAA affected systems:
1336 * - nothing to do if TSX is disabled on the host.
1337 * - we emulate TSX_CTRL if present on the host.
1338 * This lets the guest use VERW to clear CPU buffers.
1340 if (!boot_cpu_has(X86_FEATURE_RTM))
1341 data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1342 else if (!boot_cpu_has_bug(X86_BUG_TAA))
1343 data |= ARCH_CAP_TAA_NO;
1348 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1350 switch (msr->index) {
1351 case MSR_IA32_ARCH_CAPABILITIES:
1352 msr->data = kvm_get_arch_capabilities();
1354 case MSR_IA32_UCODE_REV:
1355 rdmsrl_safe(msr->index, &msr->data);
1358 if (kvm_x86_ops.get_msr_feature(msr))
1364 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1366 struct kvm_msr_entry msr;
1370 r = kvm_get_msr_feature(&msr);
1379 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1381 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1384 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1387 if (efer & (EFER_LME | EFER_LMA) &&
1388 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1391 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1397 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1399 if (efer & efer_reserved_bits)
1402 return __kvm_valid_efer(vcpu, efer);
1404 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1406 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1408 u64 old_efer = vcpu->arch.efer;
1409 u64 efer = msr_info->data;
1411 if (efer & efer_reserved_bits)
1414 if (!msr_info->host_initiated) {
1415 if (!__kvm_valid_efer(vcpu, efer))
1418 if (is_paging(vcpu) &&
1419 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1424 efer |= vcpu->arch.efer & EFER_LMA;
1426 kvm_x86_ops.set_efer(vcpu, efer);
1428 /* Update reserved bits */
1429 if ((efer ^ old_efer) & EFER_NX)
1430 kvm_mmu_reset_context(vcpu);
1435 void kvm_enable_efer_bits(u64 mask)
1437 efer_reserved_bits &= ~mask;
1439 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1442 * Write @data into the MSR specified by @index. Select MSR specific fault
1443 * checks are bypassed if @host_initiated is %true.
1444 * Returns 0 on success, non-0 otherwise.
1445 * Assumes vcpu_load() was already called.
1447 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1448 bool host_initiated)
1450 struct msr_data msr;
1455 case MSR_KERNEL_GS_BASE:
1458 if (is_noncanonical_address(data, vcpu))
1461 case MSR_IA32_SYSENTER_EIP:
1462 case MSR_IA32_SYSENTER_ESP:
1464 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1465 * non-canonical address is written on Intel but not on
1466 * AMD (which ignores the top 32-bits, because it does
1467 * not implement 64-bit SYSENTER).
1469 * 64-bit code should hence be able to write a non-canonical
1470 * value on AMD. Making the address canonical ensures that
1471 * vmentry does not fail on Intel after writing a non-canonical
1472 * value, and that something deterministic happens if the guest
1473 * invokes 64-bit SYSENTER.
1475 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1480 msr.host_initiated = host_initiated;
1482 return kvm_x86_ops.set_msr(vcpu, &msr);
1486 * Read the MSR specified by @index into @data. Select MSR specific fault
1487 * checks are bypassed if @host_initiated is %true.
1488 * Returns 0 on success, non-0 otherwise.
1489 * Assumes vcpu_load() was already called.
1491 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1492 bool host_initiated)
1494 struct msr_data msr;
1498 msr.host_initiated = host_initiated;
1500 ret = kvm_x86_ops.get_msr(vcpu, &msr);
1506 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1508 return __kvm_get_msr(vcpu, index, data, false);
1510 EXPORT_SYMBOL_GPL(kvm_get_msr);
1512 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1514 return __kvm_set_msr(vcpu, index, data, false);
1516 EXPORT_SYMBOL_GPL(kvm_set_msr);
1518 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1520 u32 ecx = kvm_rcx_read(vcpu);
1523 if (kvm_get_msr(vcpu, ecx, &data)) {
1524 trace_kvm_msr_read_ex(ecx);
1525 kvm_inject_gp(vcpu, 0);
1529 trace_kvm_msr_read(ecx, data);
1531 kvm_rax_write(vcpu, data & -1u);
1532 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1533 return kvm_skip_emulated_instruction(vcpu);
1535 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1537 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1539 u32 ecx = kvm_rcx_read(vcpu);
1540 u64 data = kvm_read_edx_eax(vcpu);
1542 if (kvm_set_msr(vcpu, ecx, data)) {
1543 trace_kvm_msr_write_ex(ecx, data);
1544 kvm_inject_gp(vcpu, 0);
1548 trace_kvm_msr_write(ecx, data);
1549 return kvm_skip_emulated_instruction(vcpu);
1551 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1554 * The fast path for frequent and performance sensitive wrmsr emulation,
1555 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1556 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1557 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1558 * other cases which must be called after interrupts are enabled on the host.
1560 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1562 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1565 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1566 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1567 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1568 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1571 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1572 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1573 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1574 trace_kvm_apic_write(APIC_ICR, (u32)data);
1581 enum exit_fastpath_completion handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1583 u32 msr = kvm_rcx_read(vcpu);
1588 case APIC_BASE_MSR + (APIC_ICR >> 4):
1589 data = kvm_read_edx_eax(vcpu);
1590 ret = handle_fastpath_set_x2apic_icr_irqoff(vcpu, data);
1593 return EXIT_FASTPATH_NONE;
1597 trace_kvm_msr_write(msr, data);
1598 return EXIT_FASTPATH_SKIP_EMUL_INS;
1601 return EXIT_FASTPATH_NONE;
1603 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1606 * Adapt set_msr() to msr_io()'s calling convention
1608 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1610 return __kvm_get_msr(vcpu, index, data, true);
1613 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1615 return __kvm_set_msr(vcpu, index, *data, true);
1618 #ifdef CONFIG_X86_64
1619 struct pvclock_clock {
1629 struct pvclock_gtod_data {
1632 struct pvclock_clock clock; /* extract of a clocksource struct */
1633 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1639 static struct pvclock_gtod_data pvclock_gtod_data;
1641 static void update_pvclock_gtod(struct timekeeper *tk)
1643 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1645 write_seqcount_begin(&vdata->seq);
1647 /* copy pvclock gtod data */
1648 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1649 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1650 vdata->clock.mask = tk->tkr_mono.mask;
1651 vdata->clock.mult = tk->tkr_mono.mult;
1652 vdata->clock.shift = tk->tkr_mono.shift;
1653 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1654 vdata->clock.offset = tk->tkr_mono.base;
1656 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1657 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1658 vdata->raw_clock.mask = tk->tkr_raw.mask;
1659 vdata->raw_clock.mult = tk->tkr_raw.mult;
1660 vdata->raw_clock.shift = tk->tkr_raw.shift;
1661 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1662 vdata->raw_clock.offset = tk->tkr_raw.base;
1664 vdata->wall_time_sec = tk->xtime_sec;
1666 vdata->offs_boot = tk->offs_boot;
1668 write_seqcount_end(&vdata->seq);
1671 static s64 get_kvmclock_base_ns(void)
1673 /* Count up from boot time, but with the frequency of the raw clock. */
1674 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1677 static s64 get_kvmclock_base_ns(void)
1679 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1680 return ktime_get_boottime_ns();
1684 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1686 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1687 kvm_vcpu_kick(vcpu);
1690 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1694 struct pvclock_wall_clock wc;
1700 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1705 ++version; /* first time write, random junk */
1709 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1713 * The guest calculates current wall clock time by adding
1714 * system time (updated by kvm_guest_time_update below) to the
1715 * wall clock specified here. We do the reverse here.
1717 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1719 wc.nsec = do_div(wall_nsec, 1000000000);
1720 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1721 wc.version = version;
1723 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1726 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1729 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1731 do_shl32_div32(dividend, divisor);
1735 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1736 s8 *pshift, u32 *pmultiplier)
1744 scaled64 = scaled_hz;
1745 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1750 tps32 = (uint32_t)tps64;
1751 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1752 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1760 *pmultiplier = div_frac(scaled64, tps32);
1763 #ifdef CONFIG_X86_64
1764 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1767 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1768 static unsigned long max_tsc_khz;
1770 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1772 u64 v = (u64)khz * (1000000 + ppm);
1777 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1781 /* Guest TSC same frequency as host TSC? */
1783 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1787 /* TSC scaling supported? */
1788 if (!kvm_has_tsc_control) {
1789 if (user_tsc_khz > tsc_khz) {
1790 vcpu->arch.tsc_catchup = 1;
1791 vcpu->arch.tsc_always_catchup = 1;
1794 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1799 /* TSC scaling required - calculate ratio */
1800 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1801 user_tsc_khz, tsc_khz);
1803 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1804 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1809 vcpu->arch.tsc_scaling_ratio = ratio;
1813 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1815 u32 thresh_lo, thresh_hi;
1816 int use_scaling = 0;
1818 /* tsc_khz can be zero if TSC calibration fails */
1819 if (user_tsc_khz == 0) {
1820 /* set tsc_scaling_ratio to a safe value */
1821 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1825 /* Compute a scale to convert nanoseconds in TSC cycles */
1826 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1827 &vcpu->arch.virtual_tsc_shift,
1828 &vcpu->arch.virtual_tsc_mult);
1829 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1832 * Compute the variation in TSC rate which is acceptable
1833 * within the range of tolerance and decide if the
1834 * rate being applied is within that bounds of the hardware
1835 * rate. If so, no scaling or compensation need be done.
1837 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1838 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1839 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1840 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1843 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1846 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1848 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1849 vcpu->arch.virtual_tsc_mult,
1850 vcpu->arch.virtual_tsc_shift);
1851 tsc += vcpu->arch.this_tsc_write;
1855 static inline int gtod_is_based_on_tsc(int mode)
1857 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
1860 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1862 #ifdef CONFIG_X86_64
1864 struct kvm_arch *ka = &vcpu->kvm->arch;
1865 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1867 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1868 atomic_read(&vcpu->kvm->online_vcpus));
1871 * Once the masterclock is enabled, always perform request in
1872 * order to update it.
1874 * In order to enable masterclock, the host clocksource must be TSC
1875 * and the vcpus need to have matched TSCs. When that happens,
1876 * perform request to enable masterclock.
1878 if (ka->use_master_clock ||
1879 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1880 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1882 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1883 atomic_read(&vcpu->kvm->online_vcpus),
1884 ka->use_master_clock, gtod->clock.vclock_mode);
1888 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1890 u64 curr_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
1891 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1895 * Multiply tsc by a fixed point number represented by ratio.
1897 * The most significant 64-N bits (mult) of ratio represent the
1898 * integral part of the fixed point number; the remaining N bits
1899 * (frac) represent the fractional part, ie. ratio represents a fixed
1900 * point number (mult + frac * 2^(-N)).
1902 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1904 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1906 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1909 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1912 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1914 if (ratio != kvm_default_tsc_scaling_ratio)
1915 _tsc = __scale_tsc(ratio, tsc);
1919 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1921 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1925 tsc = kvm_scale_tsc(vcpu, rdtsc());
1927 return target_tsc - tsc;
1930 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1932 u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
1934 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1936 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1938 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1940 vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
1943 static inline bool kvm_check_tsc_unstable(void)
1945 #ifdef CONFIG_X86_64
1947 * TSC is marked unstable when we're running on Hyper-V,
1948 * 'TSC page' clocksource is good.
1950 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
1953 return check_tsc_unstable();
1956 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1958 struct kvm *kvm = vcpu->kvm;
1959 u64 offset, ns, elapsed;
1960 unsigned long flags;
1962 bool already_matched;
1963 u64 data = msr->data;
1964 bool synchronizing = false;
1966 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1967 offset = kvm_compute_tsc_offset(vcpu, data);
1968 ns = get_kvmclock_base_ns();
1969 elapsed = ns - kvm->arch.last_tsc_nsec;
1971 if (vcpu->arch.virtual_tsc_khz) {
1972 if (data == 0 && msr->host_initiated) {
1974 * detection of vcpu initialization -- need to sync
1975 * with other vCPUs. This particularly helps to keep
1976 * kvm_clock stable after CPU hotplug
1978 synchronizing = true;
1980 u64 tsc_exp = kvm->arch.last_tsc_write +
1981 nsec_to_cycles(vcpu, elapsed);
1982 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1984 * Special case: TSC write with a small delta (1 second)
1985 * of virtual cycle time against real time is
1986 * interpreted as an attempt to synchronize the CPU.
1988 synchronizing = data < tsc_exp + tsc_hz &&
1989 data + tsc_hz > tsc_exp;
1994 * For a reliable TSC, we can match TSC offsets, and for an unstable
1995 * TSC, we add elapsed time in this computation. We could let the
1996 * compensation code attempt to catch up if we fall behind, but
1997 * it's better to try to match offsets from the beginning.
1999 if (synchronizing &&
2000 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2001 if (!kvm_check_tsc_unstable()) {
2002 offset = kvm->arch.cur_tsc_offset;
2004 u64 delta = nsec_to_cycles(vcpu, elapsed);
2006 offset = kvm_compute_tsc_offset(vcpu, data);
2009 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2012 * We split periods of matched TSC writes into generations.
2013 * For each generation, we track the original measured
2014 * nanosecond time, offset, and write, so if TSCs are in
2015 * sync, we can match exact offset, and if not, we can match
2016 * exact software computation in compute_guest_tsc()
2018 * These values are tracked in kvm->arch.cur_xxx variables.
2020 kvm->arch.cur_tsc_generation++;
2021 kvm->arch.cur_tsc_nsec = ns;
2022 kvm->arch.cur_tsc_write = data;
2023 kvm->arch.cur_tsc_offset = offset;
2028 * We also track th most recent recorded KHZ, write and time to
2029 * allow the matching interval to be extended at each write.
2031 kvm->arch.last_tsc_nsec = ns;
2032 kvm->arch.last_tsc_write = data;
2033 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2035 vcpu->arch.last_guest_tsc = data;
2037 /* Keep track of which generation this VCPU has synchronized to */
2038 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2039 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2040 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2042 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
2043 update_ia32_tsc_adjust_msr(vcpu, offset);
2045 kvm_vcpu_write_tsc_offset(vcpu, offset);
2046 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2048 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2050 kvm->arch.nr_vcpus_matched_tsc = 0;
2051 } else if (!already_matched) {
2052 kvm->arch.nr_vcpus_matched_tsc++;
2055 kvm_track_tsc_matching(vcpu);
2056 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2059 EXPORT_SYMBOL_GPL(kvm_write_tsc);
2061 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2064 u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
2065 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2068 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2070 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2071 WARN_ON(adjustment < 0);
2072 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2073 adjust_tsc_offset_guest(vcpu, adjustment);
2076 #ifdef CONFIG_X86_64
2078 static u64 read_tsc(void)
2080 u64 ret = (u64)rdtsc_ordered();
2081 u64 last = pvclock_gtod_data.clock.cycle_last;
2083 if (likely(ret >= last))
2087 * GCC likes to generate cmov here, but this branch is extremely
2088 * predictable (it's just a function of time and the likely is
2089 * very likely) and there's a data dependence, so force GCC
2090 * to generate a branch instead. I don't barrier() because
2091 * we don't actually need a barrier, and if this function
2092 * ever gets inlined it will generate worse code.
2098 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2104 switch (clock->vclock_mode) {
2105 case VDSO_CLOCKMODE_HVCLOCK:
2106 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2108 if (tsc_pg_val != U64_MAX) {
2109 /* TSC page valid */
2110 *mode = VDSO_CLOCKMODE_HVCLOCK;
2111 v = (tsc_pg_val - clock->cycle_last) &
2114 /* TSC page invalid */
2115 *mode = VDSO_CLOCKMODE_NONE;
2118 case VDSO_CLOCKMODE_TSC:
2119 *mode = VDSO_CLOCKMODE_TSC;
2120 *tsc_timestamp = read_tsc();
2121 v = (*tsc_timestamp - clock->cycle_last) &
2125 *mode = VDSO_CLOCKMODE_NONE;
2128 if (*mode == VDSO_CLOCKMODE_NONE)
2129 *tsc_timestamp = v = 0;
2131 return v * clock->mult;
2134 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2136 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2142 seq = read_seqcount_begin(>od->seq);
2143 ns = gtod->raw_clock.base_cycles;
2144 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2145 ns >>= gtod->raw_clock.shift;
2146 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2147 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2153 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2155 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2161 seq = read_seqcount_begin(>od->seq);
2162 ts->tv_sec = gtod->wall_time_sec;
2163 ns = gtod->clock.base_cycles;
2164 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2165 ns >>= gtod->clock.shift;
2166 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2168 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2174 /* returns true if host is using TSC based clocksource */
2175 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2177 /* checked again under seqlock below */
2178 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2181 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2185 /* returns true if host is using TSC based clocksource */
2186 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2189 /* checked again under seqlock below */
2190 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2193 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2199 * Assuming a stable TSC across physical CPUS, and a stable TSC
2200 * across virtual CPUs, the following condition is possible.
2201 * Each numbered line represents an event visible to both
2202 * CPUs at the next numbered event.
2204 * "timespecX" represents host monotonic time. "tscX" represents
2207 * VCPU0 on CPU0 | VCPU1 on CPU1
2209 * 1. read timespec0,tsc0
2210 * 2. | timespec1 = timespec0 + N
2212 * 3. transition to guest | transition to guest
2213 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2214 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2215 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2217 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2220 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2222 * - 0 < N - M => M < N
2224 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2225 * always the case (the difference between two distinct xtime instances
2226 * might be smaller then the difference between corresponding TSC reads,
2227 * when updating guest vcpus pvclock areas).
2229 * To avoid that problem, do not allow visibility of distinct
2230 * system_timestamp/tsc_timestamp values simultaneously: use a master
2231 * copy of host monotonic time values. Update that master copy
2234 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2238 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2240 #ifdef CONFIG_X86_64
2241 struct kvm_arch *ka = &kvm->arch;
2243 bool host_tsc_clocksource, vcpus_matched;
2245 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2246 atomic_read(&kvm->online_vcpus));
2249 * If the host uses TSC clock, then passthrough TSC as stable
2252 host_tsc_clocksource = kvm_get_time_and_clockread(
2253 &ka->master_kernel_ns,
2254 &ka->master_cycle_now);
2256 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2257 && !ka->backwards_tsc_observed
2258 && !ka->boot_vcpu_runs_old_kvmclock;
2260 if (ka->use_master_clock)
2261 atomic_set(&kvm_guest_has_master_clock, 1);
2263 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2264 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2269 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2271 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2274 static void kvm_gen_update_masterclock(struct kvm *kvm)
2276 #ifdef CONFIG_X86_64
2278 struct kvm_vcpu *vcpu;
2279 struct kvm_arch *ka = &kvm->arch;
2281 spin_lock(&ka->pvclock_gtod_sync_lock);
2282 kvm_make_mclock_inprogress_request(kvm);
2283 /* no guest entries from this point */
2284 pvclock_update_vm_gtod_copy(kvm);
2286 kvm_for_each_vcpu(i, vcpu, kvm)
2287 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2289 /* guest entries allowed */
2290 kvm_for_each_vcpu(i, vcpu, kvm)
2291 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2293 spin_unlock(&ka->pvclock_gtod_sync_lock);
2297 u64 get_kvmclock_ns(struct kvm *kvm)
2299 struct kvm_arch *ka = &kvm->arch;
2300 struct pvclock_vcpu_time_info hv_clock;
2303 spin_lock(&ka->pvclock_gtod_sync_lock);
2304 if (!ka->use_master_clock) {
2305 spin_unlock(&ka->pvclock_gtod_sync_lock);
2306 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2309 hv_clock.tsc_timestamp = ka->master_cycle_now;
2310 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2311 spin_unlock(&ka->pvclock_gtod_sync_lock);
2313 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2316 if (__this_cpu_read(cpu_tsc_khz)) {
2317 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2318 &hv_clock.tsc_shift,
2319 &hv_clock.tsc_to_system_mul);
2320 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2322 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2329 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2331 struct kvm_vcpu_arch *vcpu = &v->arch;
2332 struct pvclock_vcpu_time_info guest_hv_clock;
2334 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2335 &guest_hv_clock, sizeof(guest_hv_clock))))
2338 /* This VCPU is paused, but it's legal for a guest to read another
2339 * VCPU's kvmclock, so we really have to follow the specification where
2340 * it says that version is odd if data is being modified, and even after
2343 * Version field updates must be kept separate. This is because
2344 * kvm_write_guest_cached might use a "rep movs" instruction, and
2345 * writes within a string instruction are weakly ordered. So there
2346 * are three writes overall.
2348 * As a small optimization, only write the version field in the first
2349 * and third write. The vcpu->pv_time cache is still valid, because the
2350 * version field is the first in the struct.
2352 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2354 if (guest_hv_clock.version & 1)
2355 ++guest_hv_clock.version; /* first time write, random junk */
2357 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2358 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2360 sizeof(vcpu->hv_clock.version));
2364 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2365 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2367 if (vcpu->pvclock_set_guest_stopped_request) {
2368 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2369 vcpu->pvclock_set_guest_stopped_request = false;
2372 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2374 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2376 sizeof(vcpu->hv_clock));
2380 vcpu->hv_clock.version++;
2381 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2383 sizeof(vcpu->hv_clock.version));
2386 static int kvm_guest_time_update(struct kvm_vcpu *v)
2388 unsigned long flags, tgt_tsc_khz;
2389 struct kvm_vcpu_arch *vcpu = &v->arch;
2390 struct kvm_arch *ka = &v->kvm->arch;
2392 u64 tsc_timestamp, host_tsc;
2394 bool use_master_clock;
2400 * If the host uses TSC clock, then passthrough TSC as stable
2403 spin_lock(&ka->pvclock_gtod_sync_lock);
2404 use_master_clock = ka->use_master_clock;
2405 if (use_master_clock) {
2406 host_tsc = ka->master_cycle_now;
2407 kernel_ns = ka->master_kernel_ns;
2409 spin_unlock(&ka->pvclock_gtod_sync_lock);
2411 /* Keep irq disabled to prevent changes to the clock */
2412 local_irq_save(flags);
2413 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2414 if (unlikely(tgt_tsc_khz == 0)) {
2415 local_irq_restore(flags);
2416 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2419 if (!use_master_clock) {
2421 kernel_ns = get_kvmclock_base_ns();
2424 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2427 * We may have to catch up the TSC to match elapsed wall clock
2428 * time for two reasons, even if kvmclock is used.
2429 * 1) CPU could have been running below the maximum TSC rate
2430 * 2) Broken TSC compensation resets the base at each VCPU
2431 * entry to avoid unknown leaps of TSC even when running
2432 * again on the same CPU. This may cause apparent elapsed
2433 * time to disappear, and the guest to stand still or run
2436 if (vcpu->tsc_catchup) {
2437 u64 tsc = compute_guest_tsc(v, kernel_ns);
2438 if (tsc > tsc_timestamp) {
2439 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2440 tsc_timestamp = tsc;
2444 local_irq_restore(flags);
2446 /* With all the info we got, fill in the values */
2448 if (kvm_has_tsc_control)
2449 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2451 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2452 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2453 &vcpu->hv_clock.tsc_shift,
2454 &vcpu->hv_clock.tsc_to_system_mul);
2455 vcpu->hw_tsc_khz = tgt_tsc_khz;
2458 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2459 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2460 vcpu->last_guest_tsc = tsc_timestamp;
2462 /* If the host uses TSC clocksource, then it is stable */
2464 if (use_master_clock)
2465 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2467 vcpu->hv_clock.flags = pvclock_flags;
2469 if (vcpu->pv_time_enabled)
2470 kvm_setup_pvclock_page(v);
2471 if (v == kvm_get_vcpu(v->kvm, 0))
2472 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2477 * kvmclock updates which are isolated to a given vcpu, such as
2478 * vcpu->cpu migration, should not allow system_timestamp from
2479 * the rest of the vcpus to remain static. Otherwise ntp frequency
2480 * correction applies to one vcpu's system_timestamp but not
2483 * So in those cases, request a kvmclock update for all vcpus.
2484 * We need to rate-limit these requests though, as they can
2485 * considerably slow guests that have a large number of vcpus.
2486 * The time for a remote vcpu to update its kvmclock is bound
2487 * by the delay we use to rate-limit the updates.
2490 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2492 static void kvmclock_update_fn(struct work_struct *work)
2495 struct delayed_work *dwork = to_delayed_work(work);
2496 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2497 kvmclock_update_work);
2498 struct kvm *kvm = container_of(ka, struct kvm, arch);
2499 struct kvm_vcpu *vcpu;
2501 kvm_for_each_vcpu(i, vcpu, kvm) {
2502 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2503 kvm_vcpu_kick(vcpu);
2507 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2509 struct kvm *kvm = v->kvm;
2511 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2512 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2513 KVMCLOCK_UPDATE_DELAY);
2516 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2518 static void kvmclock_sync_fn(struct work_struct *work)
2520 struct delayed_work *dwork = to_delayed_work(work);
2521 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2522 kvmclock_sync_work);
2523 struct kvm *kvm = container_of(ka, struct kvm, arch);
2525 if (!kvmclock_periodic_sync)
2528 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2529 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2530 KVMCLOCK_SYNC_PERIOD);
2534 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2536 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2538 /* McStatusWrEn enabled? */
2539 if (guest_cpuid_is_amd_or_hygon(vcpu))
2540 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2545 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2547 u64 mcg_cap = vcpu->arch.mcg_cap;
2548 unsigned bank_num = mcg_cap & 0xff;
2549 u32 msr = msr_info->index;
2550 u64 data = msr_info->data;
2553 case MSR_IA32_MCG_STATUS:
2554 vcpu->arch.mcg_status = data;
2556 case MSR_IA32_MCG_CTL:
2557 if (!(mcg_cap & MCG_CTL_P) &&
2558 (data || !msr_info->host_initiated))
2560 if (data != 0 && data != ~(u64)0)
2562 vcpu->arch.mcg_ctl = data;
2565 if (msr >= MSR_IA32_MC0_CTL &&
2566 msr < MSR_IA32_MCx_CTL(bank_num)) {
2567 u32 offset = array_index_nospec(
2568 msr - MSR_IA32_MC0_CTL,
2569 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2571 /* only 0 or all 1s can be written to IA32_MCi_CTL
2572 * some Linux kernels though clear bit 10 in bank 4 to
2573 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2574 * this to avoid an uncatched #GP in the guest
2576 if ((offset & 0x3) == 0 &&
2577 data != 0 && (data | (1 << 10)) != ~(u64)0)
2581 if (!msr_info->host_initiated &&
2582 (offset & 0x3) == 1 && data != 0) {
2583 if (!can_set_mci_status(vcpu))
2587 vcpu->arch.mce_banks[offset] = data;
2595 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2597 struct kvm *kvm = vcpu->kvm;
2598 int lm = is_long_mode(vcpu);
2599 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2600 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2601 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2602 : kvm->arch.xen_hvm_config.blob_size_32;
2603 u32 page_num = data & ~PAGE_MASK;
2604 u64 page_addr = data & PAGE_MASK;
2609 if (page_num >= blob_size)
2612 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2617 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2626 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2628 gpa_t gpa = data & ~0x3f;
2630 /* Bits 3:5 are reserved, Should be zero */
2634 vcpu->arch.apf.msr_val = data;
2636 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2637 kvm_clear_async_pf_completion_queue(vcpu);
2638 kvm_async_pf_hash_reset(vcpu);
2642 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2646 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2647 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2648 kvm_async_pf_wakeup_all(vcpu);
2652 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2654 vcpu->arch.pv_time_enabled = false;
2655 vcpu->arch.time = 0;
2658 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2660 ++vcpu->stat.tlb_flush;
2661 kvm_x86_ops.tlb_flush(vcpu, invalidate_gpa);
2664 static void record_steal_time(struct kvm_vcpu *vcpu)
2666 struct kvm_host_map map;
2667 struct kvm_steal_time *st;
2669 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2672 /* -EAGAIN is returned in atomic context so we can just return. */
2673 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2674 &map, &vcpu->arch.st.cache, false))
2678 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2681 * Doing a TLB flush here, on the guest's behalf, can avoid
2684 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2685 st->preempted & KVM_VCPU_FLUSH_TLB);
2686 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2687 kvm_vcpu_flush_tlb(vcpu, false);
2689 vcpu->arch.st.preempted = 0;
2691 if (st->version & 1)
2692 st->version += 1; /* first time write, random junk */
2698 st->steal += current->sched_info.run_delay -
2699 vcpu->arch.st.last_steal;
2700 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2706 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2709 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2712 u32 msr = msr_info->index;
2713 u64 data = msr_info->data;
2716 case MSR_AMD64_NB_CFG:
2717 case MSR_IA32_UCODE_WRITE:
2718 case MSR_VM_HSAVE_PA:
2719 case MSR_AMD64_PATCH_LOADER:
2720 case MSR_AMD64_BU_CFG2:
2721 case MSR_AMD64_DC_CFG:
2722 case MSR_F15H_EX_CFG:
2725 case MSR_IA32_UCODE_REV:
2726 if (msr_info->host_initiated)
2727 vcpu->arch.microcode_version = data;
2729 case MSR_IA32_ARCH_CAPABILITIES:
2730 if (!msr_info->host_initiated)
2732 vcpu->arch.arch_capabilities = data;
2735 return set_efer(vcpu, msr_info);
2737 data &= ~(u64)0x40; /* ignore flush filter disable */
2738 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2739 data &= ~(u64)0x8; /* ignore TLB cache disable */
2741 /* Handle McStatusWrEn */
2742 if (data == BIT_ULL(18)) {
2743 vcpu->arch.msr_hwcr = data;
2744 } else if (data != 0) {
2745 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2750 case MSR_FAM10H_MMIO_CONF_BASE:
2752 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2757 case MSR_IA32_DEBUGCTLMSR:
2759 /* We support the non-activated case already */
2761 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2762 /* Values other than LBR and BTF are vendor-specific,
2763 thus reserved and should throw a #GP */
2766 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2769 case 0x200 ... 0x2ff:
2770 return kvm_mtrr_set_msr(vcpu, msr, data);
2771 case MSR_IA32_APICBASE:
2772 return kvm_set_apic_base(vcpu, msr_info);
2773 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2774 return kvm_x2apic_msr_write(vcpu, msr, data);
2775 case MSR_IA32_TSCDEADLINE:
2776 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2778 case MSR_IA32_TSC_ADJUST:
2779 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2780 if (!msr_info->host_initiated) {
2781 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2782 adjust_tsc_offset_guest(vcpu, adj);
2784 vcpu->arch.ia32_tsc_adjust_msr = data;
2787 case MSR_IA32_MISC_ENABLE:
2788 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2789 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2790 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2792 vcpu->arch.ia32_misc_enable_msr = data;
2793 kvm_update_cpuid(vcpu);
2795 vcpu->arch.ia32_misc_enable_msr = data;
2798 case MSR_IA32_SMBASE:
2799 if (!msr_info->host_initiated)
2801 vcpu->arch.smbase = data;
2803 case MSR_IA32_POWER_CTL:
2804 vcpu->arch.msr_ia32_power_ctl = data;
2807 kvm_write_tsc(vcpu, msr_info);
2810 if (!msr_info->host_initiated &&
2811 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2814 * KVM supports exposing PT to the guest, but does not support
2815 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
2816 * XSAVES/XRSTORS to save/restore PT MSRs.
2818 if (data & ~supported_xss)
2820 vcpu->arch.ia32_xss = data;
2823 if (!msr_info->host_initiated)
2825 vcpu->arch.smi_count = data;
2827 case MSR_KVM_WALL_CLOCK_NEW:
2828 case MSR_KVM_WALL_CLOCK:
2829 vcpu->kvm->arch.wall_clock = data;
2830 kvm_write_wall_clock(vcpu->kvm, data);
2832 case MSR_KVM_SYSTEM_TIME_NEW:
2833 case MSR_KVM_SYSTEM_TIME: {
2834 struct kvm_arch *ka = &vcpu->kvm->arch;
2836 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2837 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2839 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2840 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2842 ka->boot_vcpu_runs_old_kvmclock = tmp;
2845 vcpu->arch.time = data;
2846 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2848 /* we verify if the enable bit is set... */
2849 vcpu->arch.pv_time_enabled = false;
2853 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2854 &vcpu->arch.pv_time, data & ~1ULL,
2855 sizeof(struct pvclock_vcpu_time_info)))
2856 vcpu->arch.pv_time_enabled = true;
2860 case MSR_KVM_ASYNC_PF_EN:
2861 if (kvm_pv_enable_async_pf(vcpu, data))
2864 case MSR_KVM_STEAL_TIME:
2866 if (unlikely(!sched_info_on()))
2869 if (data & KVM_STEAL_RESERVED_MASK)
2872 vcpu->arch.st.msr_val = data;
2874 if (!(data & KVM_MSR_ENABLED))
2877 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2880 case MSR_KVM_PV_EOI_EN:
2881 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2885 case MSR_KVM_POLL_CONTROL:
2886 /* only enable bit supported */
2887 if (data & (-1ULL << 1))
2890 vcpu->arch.msr_kvm_poll_control = data;
2893 case MSR_IA32_MCG_CTL:
2894 case MSR_IA32_MCG_STATUS:
2895 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2896 return set_msr_mce(vcpu, msr_info);
2898 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2899 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2900 pr = true; /* fall through */
2901 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2902 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2903 if (kvm_pmu_is_valid_msr(vcpu, msr))
2904 return kvm_pmu_set_msr(vcpu, msr_info);
2906 if (pr || data != 0)
2907 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2908 "0x%x data 0x%llx\n", msr, data);
2910 case MSR_K7_CLK_CTL:
2912 * Ignore all writes to this no longer documented MSR.
2913 * Writes are only relevant for old K7 processors,
2914 * all pre-dating SVM, but a recommended workaround from
2915 * AMD for these chips. It is possible to specify the
2916 * affected processor models on the command line, hence
2917 * the need to ignore the workaround.
2920 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2921 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2922 case HV_X64_MSR_CRASH_CTL:
2923 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2924 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2925 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2926 case HV_X64_MSR_TSC_EMULATION_STATUS:
2927 return kvm_hv_set_msr_common(vcpu, msr, data,
2928 msr_info->host_initiated);
2929 case MSR_IA32_BBL_CR_CTL3:
2930 /* Drop writes to this legacy MSR -- see rdmsr
2931 * counterpart for further detail.
2933 if (report_ignored_msrs)
2934 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2937 case MSR_AMD64_OSVW_ID_LENGTH:
2938 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2940 vcpu->arch.osvw.length = data;
2942 case MSR_AMD64_OSVW_STATUS:
2943 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2945 vcpu->arch.osvw.status = data;
2947 case MSR_PLATFORM_INFO:
2948 if (!msr_info->host_initiated ||
2949 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2950 cpuid_fault_enabled(vcpu)))
2952 vcpu->arch.msr_platform_info = data;
2954 case MSR_MISC_FEATURES_ENABLES:
2955 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2956 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2957 !supports_cpuid_fault(vcpu)))
2959 vcpu->arch.msr_misc_features_enables = data;
2962 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2963 return xen_hvm_config(vcpu, data);
2964 if (kvm_pmu_is_valid_msr(vcpu, msr))
2965 return kvm_pmu_set_msr(vcpu, msr_info);
2967 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2971 if (report_ignored_msrs)
2973 "ignored wrmsr: 0x%x data 0x%llx\n",
2980 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2982 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2985 u64 mcg_cap = vcpu->arch.mcg_cap;
2986 unsigned bank_num = mcg_cap & 0xff;
2989 case MSR_IA32_P5_MC_ADDR:
2990 case MSR_IA32_P5_MC_TYPE:
2993 case MSR_IA32_MCG_CAP:
2994 data = vcpu->arch.mcg_cap;
2996 case MSR_IA32_MCG_CTL:
2997 if (!(mcg_cap & MCG_CTL_P) && !host)
2999 data = vcpu->arch.mcg_ctl;
3001 case MSR_IA32_MCG_STATUS:
3002 data = vcpu->arch.mcg_status;
3005 if (msr >= MSR_IA32_MC0_CTL &&
3006 msr < MSR_IA32_MCx_CTL(bank_num)) {
3007 u32 offset = array_index_nospec(
3008 msr - MSR_IA32_MC0_CTL,
3009 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3011 data = vcpu->arch.mce_banks[offset];
3020 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3022 switch (msr_info->index) {
3023 case MSR_IA32_PLATFORM_ID:
3024 case MSR_IA32_EBL_CR_POWERON:
3025 case MSR_IA32_DEBUGCTLMSR:
3026 case MSR_IA32_LASTBRANCHFROMIP:
3027 case MSR_IA32_LASTBRANCHTOIP:
3028 case MSR_IA32_LASTINTFROMIP:
3029 case MSR_IA32_LASTINTTOIP:
3031 case MSR_K8_TSEG_ADDR:
3032 case MSR_K8_TSEG_MASK:
3033 case MSR_VM_HSAVE_PA:
3034 case MSR_K8_INT_PENDING_MSG:
3035 case MSR_AMD64_NB_CFG:
3036 case MSR_FAM10H_MMIO_CONF_BASE:
3037 case MSR_AMD64_BU_CFG2:
3038 case MSR_IA32_PERF_CTL:
3039 case MSR_AMD64_DC_CFG:
3040 case MSR_F15H_EX_CFG:
3042 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3043 * limit) MSRs. Just return 0, as we do not want to expose the host
3044 * data here. Do not conditionalize this on CPUID, as KVM does not do
3045 * so for existing CPU-specific MSRs.
3047 case MSR_RAPL_POWER_UNIT:
3048 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3049 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3050 case MSR_PKG_ENERGY_STATUS: /* Total package */
3051 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3054 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3055 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3056 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3057 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3058 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3059 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3060 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3063 case MSR_IA32_UCODE_REV:
3064 msr_info->data = vcpu->arch.microcode_version;
3066 case MSR_IA32_ARCH_CAPABILITIES:
3067 if (!msr_info->host_initiated &&
3068 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3070 msr_info->data = vcpu->arch.arch_capabilities;
3072 case MSR_IA32_POWER_CTL:
3073 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3076 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
3079 case 0x200 ... 0x2ff:
3080 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3081 case 0xcd: /* fsb frequency */
3085 * MSR_EBC_FREQUENCY_ID
3086 * Conservative value valid for even the basic CPU models.
3087 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3088 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3089 * and 266MHz for model 3, or 4. Set Core Clock
3090 * Frequency to System Bus Frequency Ratio to 1 (bits
3091 * 31:24) even though these are only valid for CPU
3092 * models > 2, however guests may end up dividing or
3093 * multiplying by zero otherwise.
3095 case MSR_EBC_FREQUENCY_ID:
3096 msr_info->data = 1 << 24;
3098 case MSR_IA32_APICBASE:
3099 msr_info->data = kvm_get_apic_base(vcpu);
3101 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
3102 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3103 case MSR_IA32_TSCDEADLINE:
3104 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3106 case MSR_IA32_TSC_ADJUST:
3107 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3109 case MSR_IA32_MISC_ENABLE:
3110 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3112 case MSR_IA32_SMBASE:
3113 if (!msr_info->host_initiated)
3115 msr_info->data = vcpu->arch.smbase;
3118 msr_info->data = vcpu->arch.smi_count;
3120 case MSR_IA32_PERF_STATUS:
3121 /* TSC increment by tick */
3122 msr_info->data = 1000ULL;
3123 /* CPU multiplier */
3124 msr_info->data |= (((uint64_t)4ULL) << 40);
3127 msr_info->data = vcpu->arch.efer;
3129 case MSR_KVM_WALL_CLOCK:
3130 case MSR_KVM_WALL_CLOCK_NEW:
3131 msr_info->data = vcpu->kvm->arch.wall_clock;
3133 case MSR_KVM_SYSTEM_TIME:
3134 case MSR_KVM_SYSTEM_TIME_NEW:
3135 msr_info->data = vcpu->arch.time;
3137 case MSR_KVM_ASYNC_PF_EN:
3138 msr_info->data = vcpu->arch.apf.msr_val;
3140 case MSR_KVM_STEAL_TIME:
3141 msr_info->data = vcpu->arch.st.msr_val;
3143 case MSR_KVM_PV_EOI_EN:
3144 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3146 case MSR_KVM_POLL_CONTROL:
3147 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3149 case MSR_IA32_P5_MC_ADDR:
3150 case MSR_IA32_P5_MC_TYPE:
3151 case MSR_IA32_MCG_CAP:
3152 case MSR_IA32_MCG_CTL:
3153 case MSR_IA32_MCG_STATUS:
3154 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3155 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3156 msr_info->host_initiated);
3158 if (!msr_info->host_initiated &&
3159 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3161 msr_info->data = vcpu->arch.ia32_xss;
3163 case MSR_K7_CLK_CTL:
3165 * Provide expected ramp-up count for K7. All other
3166 * are set to zero, indicating minimum divisors for
3169 * This prevents guest kernels on AMD host with CPU
3170 * type 6, model 8 and higher from exploding due to
3171 * the rdmsr failing.
3173 msr_info->data = 0x20000000;
3175 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3176 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3177 case HV_X64_MSR_CRASH_CTL:
3178 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3179 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3180 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3181 case HV_X64_MSR_TSC_EMULATION_STATUS:
3182 return kvm_hv_get_msr_common(vcpu,
3183 msr_info->index, &msr_info->data,
3184 msr_info->host_initiated);
3185 case MSR_IA32_BBL_CR_CTL3:
3186 /* This legacy MSR exists but isn't fully documented in current
3187 * silicon. It is however accessed by winxp in very narrow
3188 * scenarios where it sets bit #19, itself documented as
3189 * a "reserved" bit. Best effort attempt to source coherent
3190 * read data here should the balance of the register be
3191 * interpreted by the guest:
3193 * L2 cache control register 3: 64GB range, 256KB size,
3194 * enabled, latency 0x1, configured
3196 msr_info->data = 0xbe702111;
3198 case MSR_AMD64_OSVW_ID_LENGTH:
3199 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3201 msr_info->data = vcpu->arch.osvw.length;
3203 case MSR_AMD64_OSVW_STATUS:
3204 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3206 msr_info->data = vcpu->arch.osvw.status;
3208 case MSR_PLATFORM_INFO:
3209 if (!msr_info->host_initiated &&
3210 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3212 msr_info->data = vcpu->arch.msr_platform_info;
3214 case MSR_MISC_FEATURES_ENABLES:
3215 msr_info->data = vcpu->arch.msr_misc_features_enables;
3218 msr_info->data = vcpu->arch.msr_hwcr;
3221 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3222 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3224 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3228 if (report_ignored_msrs)
3229 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3237 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3240 * Read or write a bunch of msrs. All parameters are kernel addresses.
3242 * @return number of msrs set successfully.
3244 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3245 struct kvm_msr_entry *entries,
3246 int (*do_msr)(struct kvm_vcpu *vcpu,
3247 unsigned index, u64 *data))
3251 for (i = 0; i < msrs->nmsrs; ++i)
3252 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3259 * Read or write a bunch of msrs. Parameters are user addresses.
3261 * @return number of msrs set successfully.
3263 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3264 int (*do_msr)(struct kvm_vcpu *vcpu,
3265 unsigned index, u64 *data),
3268 struct kvm_msrs msrs;
3269 struct kvm_msr_entry *entries;
3274 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3278 if (msrs.nmsrs >= MAX_IO_MSRS)
3281 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3282 entries = memdup_user(user_msrs->entries, size);
3283 if (IS_ERR(entries)) {
3284 r = PTR_ERR(entries);
3288 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3293 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3304 static inline bool kvm_can_mwait_in_guest(void)
3306 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3307 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3308 boot_cpu_has(X86_FEATURE_ARAT);
3311 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3316 case KVM_CAP_IRQCHIP:
3318 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3319 case KVM_CAP_SET_TSS_ADDR:
3320 case KVM_CAP_EXT_CPUID:
3321 case KVM_CAP_EXT_EMUL_CPUID:
3322 case KVM_CAP_CLOCKSOURCE:
3324 case KVM_CAP_NOP_IO_DELAY:
3325 case KVM_CAP_MP_STATE:
3326 case KVM_CAP_SYNC_MMU:
3327 case KVM_CAP_USER_NMI:
3328 case KVM_CAP_REINJECT_CONTROL:
3329 case KVM_CAP_IRQ_INJECT_STATUS:
3330 case KVM_CAP_IOEVENTFD:
3331 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3333 case KVM_CAP_PIT_STATE2:
3334 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3335 case KVM_CAP_XEN_HVM:
3336 case KVM_CAP_VCPU_EVENTS:
3337 case KVM_CAP_HYPERV:
3338 case KVM_CAP_HYPERV_VAPIC:
3339 case KVM_CAP_HYPERV_SPIN:
3340 case KVM_CAP_HYPERV_SYNIC:
3341 case KVM_CAP_HYPERV_SYNIC2:
3342 case KVM_CAP_HYPERV_VP_INDEX:
3343 case KVM_CAP_HYPERV_EVENTFD:
3344 case KVM_CAP_HYPERV_TLBFLUSH:
3345 case KVM_CAP_HYPERV_SEND_IPI:
3346 case KVM_CAP_HYPERV_CPUID:
3347 case KVM_CAP_PCI_SEGMENT:
3348 case KVM_CAP_DEBUGREGS:
3349 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3351 case KVM_CAP_ASYNC_PF:
3352 case KVM_CAP_GET_TSC_KHZ:
3353 case KVM_CAP_KVMCLOCK_CTRL:
3354 case KVM_CAP_READONLY_MEM:
3355 case KVM_CAP_HYPERV_TIME:
3356 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3357 case KVM_CAP_TSC_DEADLINE_TIMER:
3358 case KVM_CAP_DISABLE_QUIRKS:
3359 case KVM_CAP_SET_BOOT_CPU_ID:
3360 case KVM_CAP_SPLIT_IRQCHIP:
3361 case KVM_CAP_IMMEDIATE_EXIT:
3362 case KVM_CAP_PMU_EVENT_FILTER:
3363 case KVM_CAP_GET_MSR_FEATURES:
3364 case KVM_CAP_MSR_PLATFORM_INFO:
3365 case KVM_CAP_EXCEPTION_PAYLOAD:
3366 case KVM_CAP_SET_GUEST_DEBUG:
3369 case KVM_CAP_SYNC_REGS:
3370 r = KVM_SYNC_X86_VALID_FIELDS;
3372 case KVM_CAP_ADJUST_CLOCK:
3373 r = KVM_CLOCK_TSC_STABLE;
3375 case KVM_CAP_X86_DISABLE_EXITS:
3376 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3377 KVM_X86_DISABLE_EXITS_CSTATE;
3378 if(kvm_can_mwait_in_guest())
3379 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3381 case KVM_CAP_X86_SMM:
3382 /* SMBASE is usually relocated above 1M on modern chipsets,
3383 * and SMM handlers might indeed rely on 4G segment limits,
3384 * so do not report SMM to be available if real mode is
3385 * emulated via vm86 mode. Still, do not go to great lengths
3386 * to avoid userspace's usage of the feature, because it is a
3387 * fringe case that is not enabled except via specific settings
3388 * of the module parameters.
3390 r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3393 r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3395 case KVM_CAP_NR_VCPUS:
3396 r = KVM_SOFT_MAX_VCPUS;
3398 case KVM_CAP_MAX_VCPUS:
3401 case KVM_CAP_MAX_VCPU_ID:
3402 r = KVM_MAX_VCPU_ID;
3404 case KVM_CAP_PV_MMU: /* obsolete */
3408 r = KVM_MAX_MCE_BANKS;
3411 r = boot_cpu_has(X86_FEATURE_XSAVE);
3413 case KVM_CAP_TSC_CONTROL:
3414 r = kvm_has_tsc_control;
3416 case KVM_CAP_X2APIC_API:
3417 r = KVM_X2APIC_API_VALID_FLAGS;
3419 case KVM_CAP_NESTED_STATE:
3420 r = kvm_x86_ops.get_nested_state ?
3421 kvm_x86_ops.get_nested_state(NULL, NULL, 0) : 0;
3423 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3424 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3426 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3427 r = kvm_x86_ops.nested_enable_evmcs != NULL;
3436 long kvm_arch_dev_ioctl(struct file *filp,
3437 unsigned int ioctl, unsigned long arg)
3439 void __user *argp = (void __user *)arg;
3443 case KVM_GET_MSR_INDEX_LIST: {
3444 struct kvm_msr_list __user *user_msr_list = argp;
3445 struct kvm_msr_list msr_list;
3449 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3452 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3453 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3456 if (n < msr_list.nmsrs)
3459 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3460 num_msrs_to_save * sizeof(u32)))
3462 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3464 num_emulated_msrs * sizeof(u32)))
3469 case KVM_GET_SUPPORTED_CPUID:
3470 case KVM_GET_EMULATED_CPUID: {
3471 struct kvm_cpuid2 __user *cpuid_arg = argp;
3472 struct kvm_cpuid2 cpuid;
3475 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3478 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3484 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3489 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3491 if (copy_to_user(argp, &kvm_mce_cap_supported,
3492 sizeof(kvm_mce_cap_supported)))
3496 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3497 struct kvm_msr_list __user *user_msr_list = argp;
3498 struct kvm_msr_list msr_list;
3502 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3505 msr_list.nmsrs = num_msr_based_features;
3506 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3509 if (n < msr_list.nmsrs)
3512 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3513 num_msr_based_features * sizeof(u32)))
3519 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3529 static void wbinvd_ipi(void *garbage)
3534 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3536 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3539 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3541 /* Address WBINVD may be executed by guest */
3542 if (need_emulate_wbinvd(vcpu)) {
3543 if (kvm_x86_ops.has_wbinvd_exit())
3544 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3545 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3546 smp_call_function_single(vcpu->cpu,
3547 wbinvd_ipi, NULL, 1);
3550 kvm_x86_ops.vcpu_load(vcpu, cpu);
3552 /* Apply any externally detected TSC adjustments (due to suspend) */
3553 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3554 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3555 vcpu->arch.tsc_offset_adjustment = 0;
3556 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3559 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3560 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3561 rdtsc() - vcpu->arch.last_host_tsc;
3563 mark_tsc_unstable("KVM discovered backwards TSC");
3565 if (kvm_check_tsc_unstable()) {
3566 u64 offset = kvm_compute_tsc_offset(vcpu,
3567 vcpu->arch.last_guest_tsc);
3568 kvm_vcpu_write_tsc_offset(vcpu, offset);
3569 vcpu->arch.tsc_catchup = 1;
3572 if (kvm_lapic_hv_timer_in_use(vcpu))
3573 kvm_lapic_restart_hv_timer(vcpu);
3576 * On a host with synchronized TSC, there is no need to update
3577 * kvmclock on vcpu->cpu migration
3579 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3580 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3581 if (vcpu->cpu != cpu)
3582 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3586 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3589 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3591 struct kvm_host_map map;
3592 struct kvm_steal_time *st;
3594 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3597 if (vcpu->arch.st.preempted)
3600 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3601 &vcpu->arch.st.cache, true))
3605 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3607 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3609 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3612 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3616 if (vcpu->preempted)
3617 vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
3620 * Disable page faults because we're in atomic context here.
3621 * kvm_write_guest_offset_cached() would call might_fault()
3622 * that relies on pagefault_disable() to tell if there's a
3623 * bug. NOTE: the write to guest memory may not go through if
3624 * during postcopy live migration or if there's heavy guest
3627 pagefault_disable();
3629 * kvm_memslots() will be called by
3630 * kvm_write_guest_offset_cached() so take the srcu lock.
3632 idx = srcu_read_lock(&vcpu->kvm->srcu);
3633 kvm_steal_time_set_preempted(vcpu);
3634 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3636 kvm_x86_ops.vcpu_put(vcpu);
3637 vcpu->arch.last_host_tsc = rdtsc();
3639 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3640 * on every vmexit, but if not, we might have a stale dr6 from the
3641 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3646 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3647 struct kvm_lapic_state *s)
3649 if (vcpu->arch.apicv_active)
3650 kvm_x86_ops.sync_pir_to_irr(vcpu);
3652 return kvm_apic_get_state(vcpu, s);
3655 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3656 struct kvm_lapic_state *s)
3660 r = kvm_apic_set_state(vcpu, s);
3663 update_cr8_intercept(vcpu);
3668 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3670 return (!lapic_in_kernel(vcpu) ||
3671 kvm_apic_accept_pic_intr(vcpu));
3675 * if userspace requested an interrupt window, check that the
3676 * interrupt window is open.
3678 * No need to exit to userspace if we already have an interrupt queued.
3680 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3682 return kvm_arch_interrupt_allowed(vcpu) &&
3683 !kvm_cpu_has_interrupt(vcpu) &&
3684 !kvm_event_needs_reinjection(vcpu) &&
3685 kvm_cpu_accept_dm_intr(vcpu);
3688 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3689 struct kvm_interrupt *irq)
3691 if (irq->irq >= KVM_NR_INTERRUPTS)
3694 if (!irqchip_in_kernel(vcpu->kvm)) {
3695 kvm_queue_interrupt(vcpu, irq->irq, false);
3696 kvm_make_request(KVM_REQ_EVENT, vcpu);
3701 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3702 * fail for in-kernel 8259.
3704 if (pic_in_kernel(vcpu->kvm))
3707 if (vcpu->arch.pending_external_vector != -1)
3710 vcpu->arch.pending_external_vector = irq->irq;
3711 kvm_make_request(KVM_REQ_EVENT, vcpu);
3715 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3717 kvm_inject_nmi(vcpu);
3722 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3724 kvm_make_request(KVM_REQ_SMI, vcpu);
3729 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3730 struct kvm_tpr_access_ctl *tac)
3734 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3738 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3742 unsigned bank_num = mcg_cap & 0xff, bank;
3745 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3747 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3750 vcpu->arch.mcg_cap = mcg_cap;
3751 /* Init IA32_MCG_CTL to all 1s */
3752 if (mcg_cap & MCG_CTL_P)
3753 vcpu->arch.mcg_ctl = ~(u64)0;
3754 /* Init IA32_MCi_CTL to all 1s */
3755 for (bank = 0; bank < bank_num; bank++)
3756 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3758 kvm_x86_ops.setup_mce(vcpu);
3763 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3764 struct kvm_x86_mce *mce)
3766 u64 mcg_cap = vcpu->arch.mcg_cap;
3767 unsigned bank_num = mcg_cap & 0xff;
3768 u64 *banks = vcpu->arch.mce_banks;
3770 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3773 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3774 * reporting is disabled
3776 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3777 vcpu->arch.mcg_ctl != ~(u64)0)
3779 banks += 4 * mce->bank;
3781 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3782 * reporting is disabled for the bank
3784 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3786 if (mce->status & MCI_STATUS_UC) {
3787 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3788 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3789 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3792 if (banks[1] & MCI_STATUS_VAL)
3793 mce->status |= MCI_STATUS_OVER;
3794 banks[2] = mce->addr;
3795 banks[3] = mce->misc;
3796 vcpu->arch.mcg_status = mce->mcg_status;
3797 banks[1] = mce->status;
3798 kvm_queue_exception(vcpu, MC_VECTOR);
3799 } else if (!(banks[1] & MCI_STATUS_VAL)
3800 || !(banks[1] & MCI_STATUS_UC)) {
3801 if (banks[1] & MCI_STATUS_VAL)
3802 mce->status |= MCI_STATUS_OVER;
3803 banks[2] = mce->addr;
3804 banks[3] = mce->misc;
3805 banks[1] = mce->status;
3807 banks[1] |= MCI_STATUS_OVER;
3811 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3812 struct kvm_vcpu_events *events)
3817 * In guest mode, payload delivery should be deferred,
3818 * so that the L1 hypervisor can intercept #PF before
3819 * CR2 is modified (or intercept #DB before DR6 is
3820 * modified under nVMX). Unless the per-VM capability,
3821 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
3822 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
3823 * opportunistically defer the exception payload, deliver it if the
3824 * capability hasn't been requested before processing a
3825 * KVM_GET_VCPU_EVENTS.
3827 if (!vcpu->kvm->arch.exception_payload_enabled &&
3828 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
3829 kvm_deliver_exception_payload(vcpu);
3832 * The API doesn't provide the instruction length for software
3833 * exceptions, so don't report them. As long as the guest RIP
3834 * isn't advanced, we should expect to encounter the exception
3837 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3838 events->exception.injected = 0;
3839 events->exception.pending = 0;
3841 events->exception.injected = vcpu->arch.exception.injected;
3842 events->exception.pending = vcpu->arch.exception.pending;
3844 * For ABI compatibility, deliberately conflate
3845 * pending and injected exceptions when
3846 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3848 if (!vcpu->kvm->arch.exception_payload_enabled)
3849 events->exception.injected |=
3850 vcpu->arch.exception.pending;
3852 events->exception.nr = vcpu->arch.exception.nr;
3853 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3854 events->exception.error_code = vcpu->arch.exception.error_code;
3855 events->exception_has_payload = vcpu->arch.exception.has_payload;
3856 events->exception_payload = vcpu->arch.exception.payload;
3858 events->interrupt.injected =
3859 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3860 events->interrupt.nr = vcpu->arch.interrupt.nr;
3861 events->interrupt.soft = 0;
3862 events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
3864 events->nmi.injected = vcpu->arch.nmi_injected;
3865 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3866 events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
3867 events->nmi.pad = 0;
3869 events->sipi_vector = 0; /* never valid when reporting to user space */
3871 events->smi.smm = is_smm(vcpu);
3872 events->smi.pending = vcpu->arch.smi_pending;
3873 events->smi.smm_inside_nmi =
3874 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3875 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3877 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3878 | KVM_VCPUEVENT_VALID_SHADOW
3879 | KVM_VCPUEVENT_VALID_SMM);
3880 if (vcpu->kvm->arch.exception_payload_enabled)
3881 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3883 memset(&events->reserved, 0, sizeof(events->reserved));
3886 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3888 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3889 struct kvm_vcpu_events *events)
3891 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3892 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3893 | KVM_VCPUEVENT_VALID_SHADOW
3894 | KVM_VCPUEVENT_VALID_SMM
3895 | KVM_VCPUEVENT_VALID_PAYLOAD))
3898 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3899 if (!vcpu->kvm->arch.exception_payload_enabled)
3901 if (events->exception.pending)
3902 events->exception.injected = 0;
3904 events->exception_has_payload = 0;
3906 events->exception.pending = 0;
3907 events->exception_has_payload = 0;
3910 if ((events->exception.injected || events->exception.pending) &&
3911 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3914 /* INITs are latched while in SMM */
3915 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3916 (events->smi.smm || events->smi.pending) &&
3917 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3921 vcpu->arch.exception.injected = events->exception.injected;
3922 vcpu->arch.exception.pending = events->exception.pending;
3923 vcpu->arch.exception.nr = events->exception.nr;
3924 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3925 vcpu->arch.exception.error_code = events->exception.error_code;
3926 vcpu->arch.exception.has_payload = events->exception_has_payload;
3927 vcpu->arch.exception.payload = events->exception_payload;
3929 vcpu->arch.interrupt.injected = events->interrupt.injected;
3930 vcpu->arch.interrupt.nr = events->interrupt.nr;
3931 vcpu->arch.interrupt.soft = events->interrupt.soft;
3932 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3933 kvm_x86_ops.set_interrupt_shadow(vcpu,
3934 events->interrupt.shadow);
3936 vcpu->arch.nmi_injected = events->nmi.injected;
3937 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3938 vcpu->arch.nmi_pending = events->nmi.pending;
3939 kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
3941 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3942 lapic_in_kernel(vcpu))
3943 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3945 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3946 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3947 if (events->smi.smm)
3948 vcpu->arch.hflags |= HF_SMM_MASK;
3950 vcpu->arch.hflags &= ~HF_SMM_MASK;
3951 kvm_smm_changed(vcpu);
3954 vcpu->arch.smi_pending = events->smi.pending;
3956 if (events->smi.smm) {
3957 if (events->smi.smm_inside_nmi)
3958 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3960 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3963 if (lapic_in_kernel(vcpu)) {
3964 if (events->smi.latched_init)
3965 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3967 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3971 kvm_make_request(KVM_REQ_EVENT, vcpu);
3976 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3977 struct kvm_debugregs *dbgregs)
3981 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3982 kvm_get_dr(vcpu, 6, &val);
3984 dbgregs->dr7 = vcpu->arch.dr7;
3986 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3989 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3990 struct kvm_debugregs *dbgregs)
3995 if (dbgregs->dr6 & ~0xffffffffull)
3997 if (dbgregs->dr7 & ~0xffffffffull)
4000 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4001 kvm_update_dr0123(vcpu);
4002 vcpu->arch.dr6 = dbgregs->dr6;
4003 vcpu->arch.dr7 = dbgregs->dr7;
4004 kvm_update_dr7(vcpu);
4009 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4011 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4013 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4014 u64 xstate_bv = xsave->header.xfeatures;
4018 * Copy legacy XSAVE area, to avoid complications with CPUID
4019 * leaves 0 and 1 in the loop below.
4021 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4024 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4025 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4028 * Copy each region from the possibly compacted offset to the
4029 * non-compacted offset.
4031 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4033 u64 xfeature_mask = valid & -valid;
4034 int xfeature_nr = fls64(xfeature_mask) - 1;
4035 void *src = get_xsave_addr(xsave, xfeature_nr);
4038 u32 size, offset, ecx, edx;
4039 cpuid_count(XSTATE_CPUID, xfeature_nr,
4040 &size, &offset, &ecx, &edx);
4041 if (xfeature_nr == XFEATURE_PKRU)
4042 memcpy(dest + offset, &vcpu->arch.pkru,
4043 sizeof(vcpu->arch.pkru));
4045 memcpy(dest + offset, src, size);
4049 valid -= xfeature_mask;
4053 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4055 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4056 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4060 * Copy legacy XSAVE area, to avoid complications with CPUID
4061 * leaves 0 and 1 in the loop below.
4063 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4065 /* Set XSTATE_BV and possibly XCOMP_BV. */
4066 xsave->header.xfeatures = xstate_bv;
4067 if (boot_cpu_has(X86_FEATURE_XSAVES))
4068 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4071 * Copy each region from the non-compacted offset to the
4072 * possibly compacted offset.
4074 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4076 u64 xfeature_mask = valid & -valid;
4077 int xfeature_nr = fls64(xfeature_mask) - 1;
4078 void *dest = get_xsave_addr(xsave, xfeature_nr);
4081 u32 size, offset, ecx, edx;
4082 cpuid_count(XSTATE_CPUID, xfeature_nr,
4083 &size, &offset, &ecx, &edx);
4084 if (xfeature_nr == XFEATURE_PKRU)
4085 memcpy(&vcpu->arch.pkru, src + offset,
4086 sizeof(vcpu->arch.pkru));
4088 memcpy(dest, src + offset, size);
4091 valid -= xfeature_mask;
4095 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4096 struct kvm_xsave *guest_xsave)
4098 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4099 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4100 fill_xsave((u8 *) guest_xsave->region, vcpu);
4102 memcpy(guest_xsave->region,
4103 &vcpu->arch.guest_fpu->state.fxsave,
4104 sizeof(struct fxregs_state));
4105 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4106 XFEATURE_MASK_FPSSE;
4110 #define XSAVE_MXCSR_OFFSET 24
4112 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4113 struct kvm_xsave *guest_xsave)
4116 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4117 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4119 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4121 * Here we allow setting states that are not present in
4122 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4123 * with old userspace.
4125 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4127 load_xsave(vcpu, (u8 *)guest_xsave->region);
4129 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4130 mxcsr & ~mxcsr_feature_mask)
4132 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4133 guest_xsave->region, sizeof(struct fxregs_state));
4138 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4139 struct kvm_xcrs *guest_xcrs)
4141 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4142 guest_xcrs->nr_xcrs = 0;
4146 guest_xcrs->nr_xcrs = 1;
4147 guest_xcrs->flags = 0;
4148 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4149 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4152 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4153 struct kvm_xcrs *guest_xcrs)
4157 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4160 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4163 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4164 /* Only support XCR0 currently */
4165 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4166 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4167 guest_xcrs->xcrs[i].value);
4176 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4177 * stopped by the hypervisor. This function will be called from the host only.
4178 * EINVAL is returned when the host attempts to set the flag for a guest that
4179 * does not support pv clocks.
4181 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4183 if (!vcpu->arch.pv_time_enabled)
4185 vcpu->arch.pvclock_set_guest_stopped_request = true;
4186 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4190 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4191 struct kvm_enable_cap *cap)
4194 uint16_t vmcs_version;
4195 void __user *user_ptr;
4201 case KVM_CAP_HYPERV_SYNIC2:
4206 case KVM_CAP_HYPERV_SYNIC:
4207 if (!irqchip_in_kernel(vcpu->kvm))
4209 return kvm_hv_activate_synic(vcpu, cap->cap ==
4210 KVM_CAP_HYPERV_SYNIC2);
4211 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4212 if (!kvm_x86_ops.nested_enable_evmcs)
4214 r = kvm_x86_ops.nested_enable_evmcs(vcpu, &vmcs_version);
4216 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4217 if (copy_to_user(user_ptr, &vmcs_version,
4218 sizeof(vmcs_version)))
4222 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4223 if (!kvm_x86_ops.enable_direct_tlbflush)
4226 return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4233 long kvm_arch_vcpu_ioctl(struct file *filp,
4234 unsigned int ioctl, unsigned long arg)
4236 struct kvm_vcpu *vcpu = filp->private_data;
4237 void __user *argp = (void __user *)arg;
4240 struct kvm_lapic_state *lapic;
4241 struct kvm_xsave *xsave;
4242 struct kvm_xcrs *xcrs;
4250 case KVM_GET_LAPIC: {
4252 if (!lapic_in_kernel(vcpu))
4254 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4255 GFP_KERNEL_ACCOUNT);
4260 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4264 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4269 case KVM_SET_LAPIC: {
4271 if (!lapic_in_kernel(vcpu))
4273 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4274 if (IS_ERR(u.lapic)) {
4275 r = PTR_ERR(u.lapic);
4279 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4282 case KVM_INTERRUPT: {
4283 struct kvm_interrupt irq;
4286 if (copy_from_user(&irq, argp, sizeof(irq)))
4288 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4292 r = kvm_vcpu_ioctl_nmi(vcpu);
4296 r = kvm_vcpu_ioctl_smi(vcpu);
4299 case KVM_SET_CPUID: {
4300 struct kvm_cpuid __user *cpuid_arg = argp;
4301 struct kvm_cpuid cpuid;
4304 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4306 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4309 case KVM_SET_CPUID2: {
4310 struct kvm_cpuid2 __user *cpuid_arg = argp;
4311 struct kvm_cpuid2 cpuid;
4314 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4316 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4317 cpuid_arg->entries);
4320 case KVM_GET_CPUID2: {
4321 struct kvm_cpuid2 __user *cpuid_arg = argp;
4322 struct kvm_cpuid2 cpuid;
4325 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4327 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4328 cpuid_arg->entries);
4332 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4337 case KVM_GET_MSRS: {
4338 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4339 r = msr_io(vcpu, argp, do_get_msr, 1);
4340 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4343 case KVM_SET_MSRS: {
4344 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4345 r = msr_io(vcpu, argp, do_set_msr, 0);
4346 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4349 case KVM_TPR_ACCESS_REPORTING: {
4350 struct kvm_tpr_access_ctl tac;
4353 if (copy_from_user(&tac, argp, sizeof(tac)))
4355 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4359 if (copy_to_user(argp, &tac, sizeof(tac)))
4364 case KVM_SET_VAPIC_ADDR: {
4365 struct kvm_vapic_addr va;
4369 if (!lapic_in_kernel(vcpu))
4372 if (copy_from_user(&va, argp, sizeof(va)))
4374 idx = srcu_read_lock(&vcpu->kvm->srcu);
4375 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4376 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4379 case KVM_X86_SETUP_MCE: {
4383 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4385 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4388 case KVM_X86_SET_MCE: {
4389 struct kvm_x86_mce mce;
4392 if (copy_from_user(&mce, argp, sizeof(mce)))
4394 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4397 case KVM_GET_VCPU_EVENTS: {
4398 struct kvm_vcpu_events events;
4400 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4403 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4408 case KVM_SET_VCPU_EVENTS: {
4409 struct kvm_vcpu_events events;
4412 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4415 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4418 case KVM_GET_DEBUGREGS: {
4419 struct kvm_debugregs dbgregs;
4421 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4424 if (copy_to_user(argp, &dbgregs,
4425 sizeof(struct kvm_debugregs)))
4430 case KVM_SET_DEBUGREGS: {
4431 struct kvm_debugregs dbgregs;
4434 if (copy_from_user(&dbgregs, argp,
4435 sizeof(struct kvm_debugregs)))
4438 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4441 case KVM_GET_XSAVE: {
4442 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4447 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4450 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4455 case KVM_SET_XSAVE: {
4456 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4457 if (IS_ERR(u.xsave)) {
4458 r = PTR_ERR(u.xsave);
4462 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4465 case KVM_GET_XCRS: {
4466 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4471 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4474 if (copy_to_user(argp, u.xcrs,
4475 sizeof(struct kvm_xcrs)))
4480 case KVM_SET_XCRS: {
4481 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4482 if (IS_ERR(u.xcrs)) {
4483 r = PTR_ERR(u.xcrs);
4487 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4490 case KVM_SET_TSC_KHZ: {
4494 user_tsc_khz = (u32)arg;
4496 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4499 if (user_tsc_khz == 0)
4500 user_tsc_khz = tsc_khz;
4502 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4507 case KVM_GET_TSC_KHZ: {
4508 r = vcpu->arch.virtual_tsc_khz;
4511 case KVM_KVMCLOCK_CTRL: {
4512 r = kvm_set_guest_paused(vcpu);
4515 case KVM_ENABLE_CAP: {
4516 struct kvm_enable_cap cap;
4519 if (copy_from_user(&cap, argp, sizeof(cap)))
4521 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4524 case KVM_GET_NESTED_STATE: {
4525 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4529 if (!kvm_x86_ops.get_nested_state)
4532 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4534 if (get_user(user_data_size, &user_kvm_nested_state->size))
4537 r = kvm_x86_ops.get_nested_state(vcpu, user_kvm_nested_state,
4542 if (r > user_data_size) {
4543 if (put_user(r, &user_kvm_nested_state->size))
4553 case KVM_SET_NESTED_STATE: {
4554 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4555 struct kvm_nested_state kvm_state;
4559 if (!kvm_x86_ops.set_nested_state)
4563 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4567 if (kvm_state.size < sizeof(kvm_state))
4570 if (kvm_state.flags &
4571 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4572 | KVM_STATE_NESTED_EVMCS))
4575 /* nested_run_pending implies guest_mode. */
4576 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4577 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4580 idx = srcu_read_lock(&vcpu->kvm->srcu);
4581 r = kvm_x86_ops.set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4582 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4585 case KVM_GET_SUPPORTED_HV_CPUID: {
4586 struct kvm_cpuid2 __user *cpuid_arg = argp;
4587 struct kvm_cpuid2 cpuid;
4590 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4593 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4594 cpuid_arg->entries);
4599 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4614 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4616 return VM_FAULT_SIGBUS;
4619 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4623 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4625 ret = kvm_x86_ops.set_tss_addr(kvm, addr);
4629 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4632 return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
4635 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4636 unsigned long kvm_nr_mmu_pages)
4638 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4641 mutex_lock(&kvm->slots_lock);
4643 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4644 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4646 mutex_unlock(&kvm->slots_lock);
4650 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4652 return kvm->arch.n_max_mmu_pages;
4655 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4657 struct kvm_pic *pic = kvm->arch.vpic;
4661 switch (chip->chip_id) {
4662 case KVM_IRQCHIP_PIC_MASTER:
4663 memcpy(&chip->chip.pic, &pic->pics[0],
4664 sizeof(struct kvm_pic_state));
4666 case KVM_IRQCHIP_PIC_SLAVE:
4667 memcpy(&chip->chip.pic, &pic->pics[1],
4668 sizeof(struct kvm_pic_state));
4670 case KVM_IRQCHIP_IOAPIC:
4671 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4680 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4682 struct kvm_pic *pic = kvm->arch.vpic;
4686 switch (chip->chip_id) {
4687 case KVM_IRQCHIP_PIC_MASTER:
4688 spin_lock(&pic->lock);
4689 memcpy(&pic->pics[0], &chip->chip.pic,
4690 sizeof(struct kvm_pic_state));
4691 spin_unlock(&pic->lock);
4693 case KVM_IRQCHIP_PIC_SLAVE:
4694 spin_lock(&pic->lock);
4695 memcpy(&pic->pics[1], &chip->chip.pic,
4696 sizeof(struct kvm_pic_state));
4697 spin_unlock(&pic->lock);
4699 case KVM_IRQCHIP_IOAPIC:
4700 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4706 kvm_pic_update_irq(pic);
4710 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4712 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4714 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4716 mutex_lock(&kps->lock);
4717 memcpy(ps, &kps->channels, sizeof(*ps));
4718 mutex_unlock(&kps->lock);
4722 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4725 struct kvm_pit *pit = kvm->arch.vpit;
4727 mutex_lock(&pit->pit_state.lock);
4728 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4729 for (i = 0; i < 3; i++)
4730 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4731 mutex_unlock(&pit->pit_state.lock);
4735 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4737 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4738 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4739 sizeof(ps->channels));
4740 ps->flags = kvm->arch.vpit->pit_state.flags;
4741 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4742 memset(&ps->reserved, 0, sizeof(ps->reserved));
4746 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4750 u32 prev_legacy, cur_legacy;
4751 struct kvm_pit *pit = kvm->arch.vpit;
4753 mutex_lock(&pit->pit_state.lock);
4754 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4755 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4756 if (!prev_legacy && cur_legacy)
4758 memcpy(&pit->pit_state.channels, &ps->channels,
4759 sizeof(pit->pit_state.channels));
4760 pit->pit_state.flags = ps->flags;
4761 for (i = 0; i < 3; i++)
4762 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4764 mutex_unlock(&pit->pit_state.lock);
4768 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4769 struct kvm_reinject_control *control)
4771 struct kvm_pit *pit = kvm->arch.vpit;
4773 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4774 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4775 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4777 mutex_lock(&pit->pit_state.lock);
4778 kvm_pit_set_reinject(pit, control->pit_reinject);
4779 mutex_unlock(&pit->pit_state.lock);
4784 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
4787 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4789 if (kvm_x86_ops.flush_log_dirty)
4790 kvm_x86_ops.flush_log_dirty(kvm);
4793 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4796 if (!irqchip_in_kernel(kvm))
4799 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4800 irq_event->irq, irq_event->level,
4805 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4806 struct kvm_enable_cap *cap)
4814 case KVM_CAP_DISABLE_QUIRKS:
4815 kvm->arch.disabled_quirks = cap->args[0];
4818 case KVM_CAP_SPLIT_IRQCHIP: {
4819 mutex_lock(&kvm->lock);
4821 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4822 goto split_irqchip_unlock;
4824 if (irqchip_in_kernel(kvm))
4825 goto split_irqchip_unlock;
4826 if (kvm->created_vcpus)
4827 goto split_irqchip_unlock;
4828 r = kvm_setup_empty_irq_routing(kvm);
4830 goto split_irqchip_unlock;
4831 /* Pairs with irqchip_in_kernel. */
4833 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4834 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4836 split_irqchip_unlock:
4837 mutex_unlock(&kvm->lock);
4840 case KVM_CAP_X2APIC_API:
4842 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4845 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4846 kvm->arch.x2apic_format = true;
4847 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4848 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4852 case KVM_CAP_X86_DISABLE_EXITS:
4854 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4857 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4858 kvm_can_mwait_in_guest())
4859 kvm->arch.mwait_in_guest = true;
4860 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4861 kvm->arch.hlt_in_guest = true;
4862 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4863 kvm->arch.pause_in_guest = true;
4864 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4865 kvm->arch.cstate_in_guest = true;
4868 case KVM_CAP_MSR_PLATFORM_INFO:
4869 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4872 case KVM_CAP_EXCEPTION_PAYLOAD:
4873 kvm->arch.exception_payload_enabled = cap->args[0];
4883 long kvm_arch_vm_ioctl(struct file *filp,
4884 unsigned int ioctl, unsigned long arg)
4886 struct kvm *kvm = filp->private_data;
4887 void __user *argp = (void __user *)arg;
4890 * This union makes it completely explicit to gcc-3.x
4891 * that these two variables' stack usage should be
4892 * combined, not added together.
4895 struct kvm_pit_state ps;
4896 struct kvm_pit_state2 ps2;
4897 struct kvm_pit_config pit_config;
4901 case KVM_SET_TSS_ADDR:
4902 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4904 case KVM_SET_IDENTITY_MAP_ADDR: {
4907 mutex_lock(&kvm->lock);
4909 if (kvm->created_vcpus)
4910 goto set_identity_unlock;
4912 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4913 goto set_identity_unlock;
4914 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4915 set_identity_unlock:
4916 mutex_unlock(&kvm->lock);
4919 case KVM_SET_NR_MMU_PAGES:
4920 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4922 case KVM_GET_NR_MMU_PAGES:
4923 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4925 case KVM_CREATE_IRQCHIP: {
4926 mutex_lock(&kvm->lock);
4929 if (irqchip_in_kernel(kvm))
4930 goto create_irqchip_unlock;
4933 if (kvm->created_vcpus)
4934 goto create_irqchip_unlock;
4936 r = kvm_pic_init(kvm);
4938 goto create_irqchip_unlock;
4940 r = kvm_ioapic_init(kvm);
4942 kvm_pic_destroy(kvm);
4943 goto create_irqchip_unlock;
4946 r = kvm_setup_default_irq_routing(kvm);
4948 kvm_ioapic_destroy(kvm);
4949 kvm_pic_destroy(kvm);
4950 goto create_irqchip_unlock;
4952 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4954 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4955 create_irqchip_unlock:
4956 mutex_unlock(&kvm->lock);
4959 case KVM_CREATE_PIT:
4960 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4962 case KVM_CREATE_PIT2:
4964 if (copy_from_user(&u.pit_config, argp,
4965 sizeof(struct kvm_pit_config)))
4968 mutex_lock(&kvm->lock);
4971 goto create_pit_unlock;
4973 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4977 mutex_unlock(&kvm->lock);
4979 case KVM_GET_IRQCHIP: {
4980 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4981 struct kvm_irqchip *chip;
4983 chip = memdup_user(argp, sizeof(*chip));
4990 if (!irqchip_kernel(kvm))
4991 goto get_irqchip_out;
4992 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4994 goto get_irqchip_out;
4996 if (copy_to_user(argp, chip, sizeof(*chip)))
4997 goto get_irqchip_out;
5003 case KVM_SET_IRQCHIP: {
5004 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5005 struct kvm_irqchip *chip;
5007 chip = memdup_user(argp, sizeof(*chip));
5014 if (!irqchip_kernel(kvm))
5015 goto set_irqchip_out;
5016 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5023 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5026 if (!kvm->arch.vpit)
5028 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5032 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5039 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5041 mutex_lock(&kvm->lock);
5043 if (!kvm->arch.vpit)
5045 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5047 mutex_unlock(&kvm->lock);
5050 case KVM_GET_PIT2: {
5052 if (!kvm->arch.vpit)
5054 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5058 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5063 case KVM_SET_PIT2: {
5065 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5067 mutex_lock(&kvm->lock);
5069 if (!kvm->arch.vpit)
5071 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5073 mutex_unlock(&kvm->lock);
5076 case KVM_REINJECT_CONTROL: {
5077 struct kvm_reinject_control control;
5079 if (copy_from_user(&control, argp, sizeof(control)))
5082 if (!kvm->arch.vpit)
5084 r = kvm_vm_ioctl_reinject(kvm, &control);
5087 case KVM_SET_BOOT_CPU_ID:
5089 mutex_lock(&kvm->lock);
5090 if (kvm->created_vcpus)
5093 kvm->arch.bsp_vcpu_id = arg;
5094 mutex_unlock(&kvm->lock);
5096 case KVM_XEN_HVM_CONFIG: {
5097 struct kvm_xen_hvm_config xhc;
5099 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5104 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5108 case KVM_SET_CLOCK: {
5109 struct kvm_clock_data user_ns;
5113 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5122 * TODO: userspace has to take care of races with VCPU_RUN, so
5123 * kvm_gen_update_masterclock() can be cut down to locked
5124 * pvclock_update_vm_gtod_copy().
5126 kvm_gen_update_masterclock(kvm);
5127 now_ns = get_kvmclock_ns(kvm);
5128 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5129 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5132 case KVM_GET_CLOCK: {
5133 struct kvm_clock_data user_ns;
5136 now_ns = get_kvmclock_ns(kvm);
5137 user_ns.clock = now_ns;
5138 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5139 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5142 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5147 case KVM_MEMORY_ENCRYPT_OP: {
5149 if (kvm_x86_ops.mem_enc_op)
5150 r = kvm_x86_ops.mem_enc_op(kvm, argp);
5153 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5154 struct kvm_enc_region region;
5157 if (copy_from_user(®ion, argp, sizeof(region)))
5161 if (kvm_x86_ops.mem_enc_reg_region)
5162 r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion);
5165 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5166 struct kvm_enc_region region;
5169 if (copy_from_user(®ion, argp, sizeof(region)))
5173 if (kvm_x86_ops.mem_enc_unreg_region)
5174 r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion);
5177 case KVM_HYPERV_EVENTFD: {
5178 struct kvm_hyperv_eventfd hvevfd;
5181 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5183 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5186 case KVM_SET_PMU_EVENT_FILTER:
5187 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5196 static void kvm_init_msr_list(void)
5198 struct x86_pmu_capability x86_pmu;
5202 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5203 "Please update the fixed PMCs in msrs_to_saved_all[]");
5205 perf_get_x86_pmu_capability(&x86_pmu);
5207 num_msrs_to_save = 0;
5208 num_emulated_msrs = 0;
5209 num_msr_based_features = 0;
5211 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5212 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5216 * Even MSRs that are valid in the host may not be exposed
5217 * to the guests in some cases.
5219 switch (msrs_to_save_all[i]) {
5220 case MSR_IA32_BNDCFGS:
5221 if (!kvm_mpx_supported())
5225 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5228 case MSR_IA32_RTIT_CTL:
5229 case MSR_IA32_RTIT_STATUS:
5230 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5233 case MSR_IA32_RTIT_CR3_MATCH:
5234 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5235 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5238 case MSR_IA32_RTIT_OUTPUT_BASE:
5239 case MSR_IA32_RTIT_OUTPUT_MASK:
5240 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5241 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5242 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5245 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5246 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5247 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5248 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5251 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5252 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5253 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5256 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5257 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5258 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5265 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5268 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5269 if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5272 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5275 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5276 struct kvm_msr_entry msr;
5278 msr.index = msr_based_features_all[i];
5279 if (kvm_get_msr_feature(&msr))
5282 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5286 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5294 if (!(lapic_in_kernel(vcpu) &&
5295 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5296 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5307 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5314 if (!(lapic_in_kernel(vcpu) &&
5315 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5317 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5319 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5329 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5330 struct kvm_segment *var, int seg)
5332 kvm_x86_ops.set_segment(vcpu, var, seg);
5335 void kvm_get_segment(struct kvm_vcpu *vcpu,
5336 struct kvm_segment *var, int seg)
5338 kvm_x86_ops.get_segment(vcpu, var, seg);
5341 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5342 struct x86_exception *exception)
5346 BUG_ON(!mmu_is_nested(vcpu));
5348 /* NPT walks are always user-walks */
5349 access |= PFERR_USER_MASK;
5350 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5355 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5356 struct x86_exception *exception)
5358 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5359 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5362 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5363 struct x86_exception *exception)
5365 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5366 access |= PFERR_FETCH_MASK;
5367 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5370 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5371 struct x86_exception *exception)
5373 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5374 access |= PFERR_WRITE_MASK;
5375 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5378 /* uses this to access any guest's mapped memory without checking CPL */
5379 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5380 struct x86_exception *exception)
5382 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5385 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5386 struct kvm_vcpu *vcpu, u32 access,
5387 struct x86_exception *exception)
5390 int r = X86EMUL_CONTINUE;
5393 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5395 unsigned offset = addr & (PAGE_SIZE-1);
5396 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5399 if (gpa == UNMAPPED_GVA)
5400 return X86EMUL_PROPAGATE_FAULT;
5401 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5404 r = X86EMUL_IO_NEEDED;
5416 /* used for instruction fetching */
5417 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5418 gva_t addr, void *val, unsigned int bytes,
5419 struct x86_exception *exception)
5421 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5422 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5426 /* Inline kvm_read_guest_virt_helper for speed. */
5427 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5429 if (unlikely(gpa == UNMAPPED_GVA))
5430 return X86EMUL_PROPAGATE_FAULT;
5432 offset = addr & (PAGE_SIZE-1);
5433 if (WARN_ON(offset + bytes > PAGE_SIZE))
5434 bytes = (unsigned)PAGE_SIZE - offset;
5435 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5437 if (unlikely(ret < 0))
5438 return X86EMUL_IO_NEEDED;
5440 return X86EMUL_CONTINUE;
5443 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5444 gva_t addr, void *val, unsigned int bytes,
5445 struct x86_exception *exception)
5447 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5450 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5451 * is returned, but our callers are not ready for that and they blindly
5452 * call kvm_inject_page_fault. Ensure that they at least do not leak
5453 * uninitialized kernel stack memory into cr2 and error code.
5455 memset(exception, 0, sizeof(*exception));
5456 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5459 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5461 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5462 gva_t addr, void *val, unsigned int bytes,
5463 struct x86_exception *exception, bool system)
5465 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5468 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5469 access |= PFERR_USER_MASK;
5471 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5474 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5475 unsigned long addr, void *val, unsigned int bytes)
5477 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5478 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5480 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5483 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5484 struct kvm_vcpu *vcpu, u32 access,
5485 struct x86_exception *exception)
5488 int r = X86EMUL_CONTINUE;
5491 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5494 unsigned offset = addr & (PAGE_SIZE-1);
5495 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5498 if (gpa == UNMAPPED_GVA)
5499 return X86EMUL_PROPAGATE_FAULT;
5500 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5502 r = X86EMUL_IO_NEEDED;
5514 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5515 unsigned int bytes, struct x86_exception *exception,
5518 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5519 u32 access = PFERR_WRITE_MASK;
5521 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5522 access |= PFERR_USER_MASK;
5524 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5528 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5529 unsigned int bytes, struct x86_exception *exception)
5531 /* kvm_write_guest_virt_system can pull in tons of pages. */
5532 vcpu->arch.l1tf_flush_l1d = true;
5535 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5536 * is returned, but our callers are not ready for that and they blindly
5537 * call kvm_inject_page_fault. Ensure that they at least do not leak
5538 * uninitialized kernel stack memory into cr2 and error code.
5540 memset(exception, 0, sizeof(*exception));
5541 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5542 PFERR_WRITE_MASK, exception);
5544 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5546 int handle_ud(struct kvm_vcpu *vcpu)
5548 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5549 int emul_type = EMULTYPE_TRAP_UD;
5550 char sig[5]; /* ud2; .ascii "kvm" */
5551 struct x86_exception e;
5553 if (force_emulation_prefix &&
5554 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5555 sig, sizeof(sig), &e) == 0 &&
5556 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5557 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5558 emul_type = EMULTYPE_TRAP_UD_FORCED;
5561 return kvm_emulate_instruction(vcpu, emul_type);
5563 EXPORT_SYMBOL_GPL(handle_ud);
5565 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5566 gpa_t gpa, bool write)
5568 /* For APIC access vmexit */
5569 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5572 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5573 trace_vcpu_match_mmio(gva, gpa, write, true);
5580 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5581 gpa_t *gpa, struct x86_exception *exception,
5584 u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5585 | (write ? PFERR_WRITE_MASK : 0);
5588 * currently PKRU is only applied to ept enabled guest so
5589 * there is no pkey in EPT page table for L1 guest or EPT
5590 * shadow page table for L2 guest.
5592 if (vcpu_match_mmio_gva(vcpu, gva)
5593 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5594 vcpu->arch.mmio_access, 0, access)) {
5595 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5596 (gva & (PAGE_SIZE - 1));
5597 trace_vcpu_match_mmio(gva, *gpa, write, false);
5601 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5603 if (*gpa == UNMAPPED_GVA)
5606 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5609 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5610 const void *val, int bytes)
5614 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5617 kvm_page_track_write(vcpu, gpa, val, bytes);
5621 struct read_write_emulator_ops {
5622 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5624 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5625 void *val, int bytes);
5626 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5627 int bytes, void *val);
5628 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5629 void *val, int bytes);
5633 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5635 if (vcpu->mmio_read_completed) {
5636 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5637 vcpu->mmio_fragments[0].gpa, val);
5638 vcpu->mmio_read_completed = 0;
5645 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5646 void *val, int bytes)
5648 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5651 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5652 void *val, int bytes)
5654 return emulator_write_phys(vcpu, gpa, val, bytes);
5657 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5659 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5660 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5663 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5664 void *val, int bytes)
5666 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5667 return X86EMUL_IO_NEEDED;
5670 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5671 void *val, int bytes)
5673 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5675 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5676 return X86EMUL_CONTINUE;
5679 static const struct read_write_emulator_ops read_emultor = {
5680 .read_write_prepare = read_prepare,
5681 .read_write_emulate = read_emulate,
5682 .read_write_mmio = vcpu_mmio_read,
5683 .read_write_exit_mmio = read_exit_mmio,
5686 static const struct read_write_emulator_ops write_emultor = {
5687 .read_write_emulate = write_emulate,
5688 .read_write_mmio = write_mmio,
5689 .read_write_exit_mmio = write_exit_mmio,
5693 static int emulator_read_write_onepage(unsigned long addr, void *val,
5695 struct x86_exception *exception,
5696 struct kvm_vcpu *vcpu,
5697 const struct read_write_emulator_ops *ops)
5701 bool write = ops->write;
5702 struct kvm_mmio_fragment *frag;
5703 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
5706 * If the exit was due to a NPF we may already have a GPA.
5707 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5708 * Note, this cannot be used on string operations since string
5709 * operation using rep will only have the initial GPA from the NPF
5712 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
5713 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
5714 gpa = ctxt->gpa_val;
5715 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5717 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5719 return X86EMUL_PROPAGATE_FAULT;
5722 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5723 return X86EMUL_CONTINUE;
5726 * Is this MMIO handled locally?
5728 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5729 if (handled == bytes)
5730 return X86EMUL_CONTINUE;
5736 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5737 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5741 return X86EMUL_CONTINUE;
5744 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5746 void *val, unsigned int bytes,
5747 struct x86_exception *exception,
5748 const struct read_write_emulator_ops *ops)
5750 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5754 if (ops->read_write_prepare &&
5755 ops->read_write_prepare(vcpu, val, bytes))
5756 return X86EMUL_CONTINUE;
5758 vcpu->mmio_nr_fragments = 0;
5760 /* Crossing a page boundary? */
5761 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5764 now = -addr & ~PAGE_MASK;
5765 rc = emulator_read_write_onepage(addr, val, now, exception,
5768 if (rc != X86EMUL_CONTINUE)
5771 if (ctxt->mode != X86EMUL_MODE_PROT64)
5777 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5779 if (rc != X86EMUL_CONTINUE)
5782 if (!vcpu->mmio_nr_fragments)
5785 gpa = vcpu->mmio_fragments[0].gpa;
5787 vcpu->mmio_needed = 1;
5788 vcpu->mmio_cur_fragment = 0;
5790 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5791 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5792 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5793 vcpu->run->mmio.phys_addr = gpa;
5795 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5798 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5802 struct x86_exception *exception)
5804 return emulator_read_write(ctxt, addr, val, bytes,
5805 exception, &read_emultor);
5808 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5812 struct x86_exception *exception)
5814 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5815 exception, &write_emultor);
5818 #define CMPXCHG_TYPE(t, ptr, old, new) \
5819 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5821 #ifdef CONFIG_X86_64
5822 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5824 # define CMPXCHG64(ptr, old, new) \
5825 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5828 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5833 struct x86_exception *exception)
5835 struct kvm_host_map map;
5836 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5842 /* guests cmpxchg8b have to be emulated atomically */
5843 if (bytes > 8 || (bytes & (bytes - 1)))
5846 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5848 if (gpa == UNMAPPED_GVA ||
5849 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5853 * Emulate the atomic as a straight write to avoid #AC if SLD is
5854 * enabled in the host and the access splits a cache line.
5856 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
5857 page_line_mask = ~(cache_line_size() - 1);
5859 page_line_mask = PAGE_MASK;
5861 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
5864 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5867 kaddr = map.hva + offset_in_page(gpa);
5871 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5874 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5877 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5880 exchanged = CMPXCHG64(kaddr, old, new);
5886 kvm_vcpu_unmap(vcpu, &map, true);
5889 return X86EMUL_CMPXCHG_FAILED;
5891 kvm_page_track_write(vcpu, gpa, new, bytes);
5893 return X86EMUL_CONTINUE;
5896 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5898 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5901 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5905 for (i = 0; i < vcpu->arch.pio.count; i++) {
5906 if (vcpu->arch.pio.in)
5907 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5908 vcpu->arch.pio.size, pd);
5910 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5911 vcpu->arch.pio.port, vcpu->arch.pio.size,
5915 pd += vcpu->arch.pio.size;
5920 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5921 unsigned short port, void *val,
5922 unsigned int count, bool in)
5924 vcpu->arch.pio.port = port;
5925 vcpu->arch.pio.in = in;
5926 vcpu->arch.pio.count = count;
5927 vcpu->arch.pio.size = size;
5929 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5930 vcpu->arch.pio.count = 0;
5934 vcpu->run->exit_reason = KVM_EXIT_IO;
5935 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5936 vcpu->run->io.size = size;
5937 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5938 vcpu->run->io.count = count;
5939 vcpu->run->io.port = port;
5944 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
5945 unsigned short port, void *val, unsigned int count)
5949 if (vcpu->arch.pio.count)
5952 memset(vcpu->arch.pio_data, 0, size * count);
5954 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5957 memcpy(val, vcpu->arch.pio_data, size * count);
5958 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5959 vcpu->arch.pio.count = 0;
5966 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5967 int size, unsigned short port, void *val,
5970 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
5974 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
5975 unsigned short port, const void *val,
5978 memcpy(vcpu->arch.pio_data, val, size * count);
5979 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5980 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5983 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5984 int size, unsigned short port,
5985 const void *val, unsigned int count)
5987 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
5990 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5992 return kvm_x86_ops.get_segment_base(vcpu, seg);
5995 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5997 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6000 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6002 if (!need_emulate_wbinvd(vcpu))
6003 return X86EMUL_CONTINUE;
6005 if (kvm_x86_ops.has_wbinvd_exit()) {
6006 int cpu = get_cpu();
6008 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6009 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6010 wbinvd_ipi, NULL, 1);
6012 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6015 return X86EMUL_CONTINUE;
6018 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6020 kvm_emulate_wbinvd_noskip(vcpu);
6021 return kvm_skip_emulated_instruction(vcpu);
6023 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6027 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6029 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6032 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6033 unsigned long *dest)
6035 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6038 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6039 unsigned long value)
6042 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6045 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6047 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6050 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6052 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6053 unsigned long value;
6057 value = kvm_read_cr0(vcpu);
6060 value = vcpu->arch.cr2;
6063 value = kvm_read_cr3(vcpu);
6066 value = kvm_read_cr4(vcpu);
6069 value = kvm_get_cr8(vcpu);
6072 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6079 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6081 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6086 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6089 vcpu->arch.cr2 = val;
6092 res = kvm_set_cr3(vcpu, val);
6095 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6098 res = kvm_set_cr8(vcpu, val);
6101 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6108 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6110 return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6113 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6115 kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6118 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6120 kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6123 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6125 kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6128 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6130 kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6133 static unsigned long emulator_get_cached_segment_base(
6134 struct x86_emulate_ctxt *ctxt, int seg)
6136 return get_segment_base(emul_to_vcpu(ctxt), seg);
6139 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6140 struct desc_struct *desc, u32 *base3,
6143 struct kvm_segment var;
6145 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6146 *selector = var.selector;
6149 memset(desc, 0, sizeof(*desc));
6157 set_desc_limit(desc, var.limit);
6158 set_desc_base(desc, (unsigned long)var.base);
6159 #ifdef CONFIG_X86_64
6161 *base3 = var.base >> 32;
6163 desc->type = var.type;
6165 desc->dpl = var.dpl;
6166 desc->p = var.present;
6167 desc->avl = var.avl;
6175 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6176 struct desc_struct *desc, u32 base3,
6179 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6180 struct kvm_segment var;
6182 var.selector = selector;
6183 var.base = get_desc_base(desc);
6184 #ifdef CONFIG_X86_64
6185 var.base |= ((u64)base3) << 32;
6187 var.limit = get_desc_limit(desc);
6189 var.limit = (var.limit << 12) | 0xfff;
6190 var.type = desc->type;
6191 var.dpl = desc->dpl;
6196 var.avl = desc->avl;
6197 var.present = desc->p;
6198 var.unusable = !var.present;
6201 kvm_set_segment(vcpu, &var, seg);
6205 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6206 u32 msr_index, u64 *pdata)
6208 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6211 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6212 u32 msr_index, u64 data)
6214 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6217 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6219 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6221 return vcpu->arch.smbase;
6224 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6226 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6228 vcpu->arch.smbase = smbase;
6231 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6234 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6237 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6238 u32 pmc, u64 *pdata)
6240 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6243 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6245 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6248 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6249 struct x86_instruction_info *info,
6250 enum x86_intercept_stage stage)
6252 return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6256 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6257 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6260 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6263 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6265 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6268 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6270 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6273 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6275 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6278 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6280 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6283 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6285 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6288 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6290 kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6293 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6295 return emul_to_vcpu(ctxt)->arch.hflags;
6298 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6300 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6303 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6304 const char *smstate)
6306 return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6309 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6311 kvm_smm_changed(emul_to_vcpu(ctxt));
6314 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6316 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6319 static const struct x86_emulate_ops emulate_ops = {
6320 .read_gpr = emulator_read_gpr,
6321 .write_gpr = emulator_write_gpr,
6322 .read_std = emulator_read_std,
6323 .write_std = emulator_write_std,
6324 .read_phys = kvm_read_guest_phys_system,
6325 .fetch = kvm_fetch_guest_virt,
6326 .read_emulated = emulator_read_emulated,
6327 .write_emulated = emulator_write_emulated,
6328 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6329 .invlpg = emulator_invlpg,
6330 .pio_in_emulated = emulator_pio_in_emulated,
6331 .pio_out_emulated = emulator_pio_out_emulated,
6332 .get_segment = emulator_get_segment,
6333 .set_segment = emulator_set_segment,
6334 .get_cached_segment_base = emulator_get_cached_segment_base,
6335 .get_gdt = emulator_get_gdt,
6336 .get_idt = emulator_get_idt,
6337 .set_gdt = emulator_set_gdt,
6338 .set_idt = emulator_set_idt,
6339 .get_cr = emulator_get_cr,
6340 .set_cr = emulator_set_cr,
6341 .cpl = emulator_get_cpl,
6342 .get_dr = emulator_get_dr,
6343 .set_dr = emulator_set_dr,
6344 .get_smbase = emulator_get_smbase,
6345 .set_smbase = emulator_set_smbase,
6346 .set_msr = emulator_set_msr,
6347 .get_msr = emulator_get_msr,
6348 .check_pmc = emulator_check_pmc,
6349 .read_pmc = emulator_read_pmc,
6350 .halt = emulator_halt,
6351 .wbinvd = emulator_wbinvd,
6352 .fix_hypercall = emulator_fix_hypercall,
6353 .intercept = emulator_intercept,
6354 .get_cpuid = emulator_get_cpuid,
6355 .guest_has_long_mode = emulator_guest_has_long_mode,
6356 .guest_has_movbe = emulator_guest_has_movbe,
6357 .guest_has_fxsr = emulator_guest_has_fxsr,
6358 .set_nmi_mask = emulator_set_nmi_mask,
6359 .get_hflags = emulator_get_hflags,
6360 .set_hflags = emulator_set_hflags,
6361 .pre_leave_smm = emulator_pre_leave_smm,
6362 .post_leave_smm = emulator_post_leave_smm,
6363 .set_xcr = emulator_set_xcr,
6366 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6368 u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6370 * an sti; sti; sequence only disable interrupts for the first
6371 * instruction. So, if the last instruction, be it emulated or
6372 * not, left the system with the INT_STI flag enabled, it
6373 * means that the last instruction is an sti. We should not
6374 * leave the flag on in this case. The same goes for mov ss
6376 if (int_shadow & mask)
6378 if (unlikely(int_shadow || mask)) {
6379 kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6381 kvm_make_request(KVM_REQ_EVENT, vcpu);
6385 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6387 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6388 if (ctxt->exception.vector == PF_VECTOR)
6389 return kvm_propagate_fault(vcpu, &ctxt->exception);
6391 if (ctxt->exception.error_code_valid)
6392 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6393 ctxt->exception.error_code);
6395 kvm_queue_exception(vcpu, ctxt->exception.vector);
6399 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6401 struct x86_emulate_ctxt *ctxt;
6403 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6405 pr_err("kvm: failed to allocate vcpu's emulator\n");
6410 ctxt->ops = &emulate_ops;
6411 vcpu->arch.emulate_ctxt = ctxt;
6416 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6418 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6421 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6423 ctxt->gpa_available = false;
6424 ctxt->eflags = kvm_get_rflags(vcpu);
6425 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6427 ctxt->eip = kvm_rip_read(vcpu);
6428 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6429 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6430 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6431 cs_db ? X86EMUL_MODE_PROT32 :
6432 X86EMUL_MODE_PROT16;
6433 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6434 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6435 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6437 init_decode_cache(ctxt);
6438 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6441 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6443 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6446 init_emulate_ctxt(vcpu);
6450 ctxt->_eip = ctxt->eip + inc_eip;
6451 ret = emulate_int_real(ctxt, irq);
6453 if (ret != X86EMUL_CONTINUE) {
6454 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6456 ctxt->eip = ctxt->_eip;
6457 kvm_rip_write(vcpu, ctxt->eip);
6458 kvm_set_rflags(vcpu, ctxt->eflags);
6461 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6463 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6465 ++vcpu->stat.insn_emulation_fail;
6466 trace_kvm_emulate_insn_failed(vcpu);
6468 if (emulation_type & EMULTYPE_VMWARE_GP) {
6469 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6473 if (emulation_type & EMULTYPE_SKIP) {
6474 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6475 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6476 vcpu->run->internal.ndata = 0;
6480 kvm_queue_exception(vcpu, UD_VECTOR);
6482 if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
6483 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6484 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6485 vcpu->run->internal.ndata = 0;
6492 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6493 bool write_fault_to_shadow_pgtable,
6496 gpa_t gpa = cr2_or_gpa;
6499 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6502 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6503 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6506 if (!vcpu->arch.mmu->direct_map) {
6508 * Write permission should be allowed since only
6509 * write access need to be emulated.
6511 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6514 * If the mapping is invalid in guest, let cpu retry
6515 * it to generate fault.
6517 if (gpa == UNMAPPED_GVA)
6522 * Do not retry the unhandleable instruction if it faults on the
6523 * readonly host memory, otherwise it will goto a infinite loop:
6524 * retry instruction -> write #PF -> emulation fail -> retry
6525 * instruction -> ...
6527 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6530 * If the instruction failed on the error pfn, it can not be fixed,
6531 * report the error to userspace.
6533 if (is_error_noslot_pfn(pfn))
6536 kvm_release_pfn_clean(pfn);
6538 /* The instructions are well-emulated on direct mmu. */
6539 if (vcpu->arch.mmu->direct_map) {
6540 unsigned int indirect_shadow_pages;
6542 spin_lock(&vcpu->kvm->mmu_lock);
6543 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6544 spin_unlock(&vcpu->kvm->mmu_lock);
6546 if (indirect_shadow_pages)
6547 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6553 * if emulation was due to access to shadowed page table
6554 * and it failed try to unshadow page and re-enter the
6555 * guest to let CPU execute the instruction.
6557 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6560 * If the access faults on its page table, it can not
6561 * be fixed by unprotecting shadow page and it should
6562 * be reported to userspace.
6564 return !write_fault_to_shadow_pgtable;
6567 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6568 gpa_t cr2_or_gpa, int emulation_type)
6570 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6571 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
6573 last_retry_eip = vcpu->arch.last_retry_eip;
6574 last_retry_addr = vcpu->arch.last_retry_addr;
6577 * If the emulation is caused by #PF and it is non-page_table
6578 * writing instruction, it means the VM-EXIT is caused by shadow
6579 * page protected, we can zap the shadow page and retry this
6580 * instruction directly.
6582 * Note: if the guest uses a non-page-table modifying instruction
6583 * on the PDE that points to the instruction, then we will unmap
6584 * the instruction and go to an infinite loop. So, we cache the
6585 * last retried eip and the last fault address, if we meet the eip
6586 * and the address again, we can break out of the potential infinite
6589 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6591 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6594 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6595 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6598 if (x86_page_table_writing_insn(ctxt))
6601 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
6604 vcpu->arch.last_retry_eip = ctxt->eip;
6605 vcpu->arch.last_retry_addr = cr2_or_gpa;
6607 if (!vcpu->arch.mmu->direct_map)
6608 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6610 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6615 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6616 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6618 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6620 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6621 /* This is a good place to trace that we are exiting SMM. */
6622 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6624 /* Process a latched INIT or SMI, if any. */
6625 kvm_make_request(KVM_REQ_EVENT, vcpu);
6628 kvm_mmu_reset_context(vcpu);
6631 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6640 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6641 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6646 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6648 struct kvm_run *kvm_run = vcpu->run;
6650 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6651 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6652 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6653 kvm_run->debug.arch.exception = DB_VECTOR;
6654 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6657 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6661 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6663 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6666 r = kvm_x86_ops.skip_emulated_instruction(vcpu);
6671 * rflags is the old, "raw" value of the flags. The new value has
6672 * not been saved yet.
6674 * This is correct even for TF set by the guest, because "the
6675 * processor will not generate this exception after the instruction
6676 * that sets the TF flag".
6678 if (unlikely(rflags & X86_EFLAGS_TF))
6679 r = kvm_vcpu_do_singlestep(vcpu);
6682 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6684 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6686 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6687 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6688 struct kvm_run *kvm_run = vcpu->run;
6689 unsigned long eip = kvm_get_linear_rip(vcpu);
6690 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6691 vcpu->arch.guest_debug_dr7,
6695 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6696 kvm_run->debug.arch.pc = eip;
6697 kvm_run->debug.arch.exception = DB_VECTOR;
6698 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6704 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6705 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6706 unsigned long eip = kvm_get_linear_rip(vcpu);
6707 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6712 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
6721 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6723 switch (ctxt->opcode_len) {
6730 case 0xe6: /* OUT */
6734 case 0x6c: /* INS */
6736 case 0x6e: /* OUTS */
6743 case 0x33: /* RDPMC */
6752 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6753 int emulation_type, void *insn, int insn_len)
6756 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6757 bool writeback = true;
6758 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6760 vcpu->arch.l1tf_flush_l1d = true;
6763 * Clear write_fault_to_shadow_pgtable here to ensure it is
6766 vcpu->arch.write_fault_to_shadow_pgtable = false;
6767 kvm_clear_exception_queue(vcpu);
6769 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6770 init_emulate_ctxt(vcpu);
6773 * We will reenter on the same instruction since
6774 * we do not set complete_userspace_io. This does not
6775 * handle watchpoints yet, those would be handled in
6778 if (!(emulation_type & EMULTYPE_SKIP) &&
6779 kvm_vcpu_check_breakpoint(vcpu, &r))
6782 ctxt->interruptibility = 0;
6783 ctxt->have_exception = false;
6784 ctxt->exception.vector = -1;
6785 ctxt->perm_ok = false;
6787 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6789 r = x86_decode_insn(ctxt, insn, insn_len);
6791 trace_kvm_emulate_insn_start(vcpu);
6792 ++vcpu->stat.insn_emulation;
6793 if (r != EMULATION_OK) {
6794 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6795 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6796 kvm_queue_exception(vcpu, UD_VECTOR);
6799 if (reexecute_instruction(vcpu, cr2_or_gpa,
6803 if (ctxt->have_exception) {
6805 * #UD should result in just EMULATION_FAILED, and trap-like
6806 * exception should not be encountered during decode.
6808 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6809 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6810 inject_emulated_exception(vcpu);
6813 return handle_emulation_failure(vcpu, emulation_type);
6817 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6818 !is_vmware_backdoor_opcode(ctxt)) {
6819 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6824 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6825 * for kvm_skip_emulated_instruction(). The caller is responsible for
6826 * updating interruptibility state and injecting single-step #DBs.
6828 if (emulation_type & EMULTYPE_SKIP) {
6829 kvm_rip_write(vcpu, ctxt->_eip);
6830 if (ctxt->eflags & X86_EFLAGS_RF)
6831 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6835 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
6838 /* this is needed for vmware backdoor interface to work since it
6839 changes registers values during IO operation */
6840 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6841 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6842 emulator_invalidate_register_cache(ctxt);
6846 if (emulation_type & EMULTYPE_PF) {
6847 /* Save the faulting GPA (cr2) in the address field */
6848 ctxt->exception.address = cr2_or_gpa;
6850 /* With shadow page tables, cr2 contains a GVA or nGPA. */
6851 if (vcpu->arch.mmu->direct_map) {
6852 ctxt->gpa_available = true;
6853 ctxt->gpa_val = cr2_or_gpa;
6856 /* Sanitize the address out of an abundance of paranoia. */
6857 ctxt->exception.address = 0;
6860 r = x86_emulate_insn(ctxt);
6862 if (r == EMULATION_INTERCEPTED)
6865 if (r == EMULATION_FAILED) {
6866 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
6870 return handle_emulation_failure(vcpu, emulation_type);
6873 if (ctxt->have_exception) {
6875 if (inject_emulated_exception(vcpu))
6877 } else if (vcpu->arch.pio.count) {
6878 if (!vcpu->arch.pio.in) {
6879 /* FIXME: return into emulator if single-stepping. */
6880 vcpu->arch.pio.count = 0;
6883 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6886 } else if (vcpu->mmio_needed) {
6887 ++vcpu->stat.mmio_exits;
6889 if (!vcpu->mmio_is_write)
6892 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6893 } else if (r == EMULATION_RESTART)
6899 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6900 toggle_interruptibility(vcpu, ctxt->interruptibility);
6901 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6902 if (!ctxt->have_exception ||
6903 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6904 kvm_rip_write(vcpu, ctxt->eip);
6906 r = kvm_vcpu_do_singlestep(vcpu);
6907 if (kvm_x86_ops.update_emulated_instruction)
6908 kvm_x86_ops.update_emulated_instruction(vcpu);
6909 __kvm_set_rflags(vcpu, ctxt->eflags);
6913 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6914 * do nothing, and it will be requested again as soon as
6915 * the shadow expires. But we still need to check here,
6916 * because POPF has no interrupt shadow.
6918 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6919 kvm_make_request(KVM_REQ_EVENT, vcpu);
6921 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6926 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6928 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6930 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6932 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6933 void *insn, int insn_len)
6935 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6937 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6939 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6941 vcpu->arch.pio.count = 0;
6945 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6947 vcpu->arch.pio.count = 0;
6949 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6952 return kvm_skip_emulated_instruction(vcpu);
6955 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6956 unsigned short port)
6958 unsigned long val = kvm_rax_read(vcpu);
6959 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
6965 * Workaround userspace that relies on old KVM behavior of %rip being
6966 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6969 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6970 vcpu->arch.complete_userspace_io =
6971 complete_fast_pio_out_port_0x7e;
6972 kvm_skip_emulated_instruction(vcpu);
6974 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6975 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6980 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6984 /* We should only ever be called with arch.pio.count equal to 1 */
6985 BUG_ON(vcpu->arch.pio.count != 1);
6987 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6988 vcpu->arch.pio.count = 0;
6992 /* For size less than 4 we merge, else we zero extend */
6993 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6996 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
6997 * the copy and tracing
6999 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7000 kvm_rax_write(vcpu, val);
7002 return kvm_skip_emulated_instruction(vcpu);
7005 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7006 unsigned short port)
7011 /* For size less than 4 we merge, else we zero extend */
7012 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7014 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7016 kvm_rax_write(vcpu, val);
7020 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7021 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7026 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7031 ret = kvm_fast_pio_in(vcpu, size, port);
7033 ret = kvm_fast_pio_out(vcpu, size, port);
7034 return ret && kvm_skip_emulated_instruction(vcpu);
7036 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7038 static int kvmclock_cpu_down_prep(unsigned int cpu)
7040 __this_cpu_write(cpu_tsc_khz, 0);
7044 static void tsc_khz_changed(void *data)
7046 struct cpufreq_freqs *freq = data;
7047 unsigned long khz = 0;
7051 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7052 khz = cpufreq_quick_get(raw_smp_processor_id());
7055 __this_cpu_write(cpu_tsc_khz, khz);
7058 #ifdef CONFIG_X86_64
7059 static void kvm_hyperv_tsc_notifier(void)
7062 struct kvm_vcpu *vcpu;
7065 mutex_lock(&kvm_lock);
7066 list_for_each_entry(kvm, &vm_list, vm_list)
7067 kvm_make_mclock_inprogress_request(kvm);
7069 hyperv_stop_tsc_emulation();
7071 /* TSC frequency always matches when on Hyper-V */
7072 for_each_present_cpu(cpu)
7073 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7074 kvm_max_guest_tsc_khz = tsc_khz;
7076 list_for_each_entry(kvm, &vm_list, vm_list) {
7077 struct kvm_arch *ka = &kvm->arch;
7079 spin_lock(&ka->pvclock_gtod_sync_lock);
7081 pvclock_update_vm_gtod_copy(kvm);
7083 kvm_for_each_vcpu(cpu, vcpu, kvm)
7084 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7086 kvm_for_each_vcpu(cpu, vcpu, kvm)
7087 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7089 spin_unlock(&ka->pvclock_gtod_sync_lock);
7091 mutex_unlock(&kvm_lock);
7095 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7098 struct kvm_vcpu *vcpu;
7099 int i, send_ipi = 0;
7102 * We allow guests to temporarily run on slowing clocks,
7103 * provided we notify them after, or to run on accelerating
7104 * clocks, provided we notify them before. Thus time never
7107 * However, we have a problem. We can't atomically update
7108 * the frequency of a given CPU from this function; it is
7109 * merely a notifier, which can be called from any CPU.
7110 * Changing the TSC frequency at arbitrary points in time
7111 * requires a recomputation of local variables related to
7112 * the TSC for each VCPU. We must flag these local variables
7113 * to be updated and be sure the update takes place with the
7114 * new frequency before any guests proceed.
7116 * Unfortunately, the combination of hotplug CPU and frequency
7117 * change creates an intractable locking scenario; the order
7118 * of when these callouts happen is undefined with respect to
7119 * CPU hotplug, and they can race with each other. As such,
7120 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7121 * undefined; you can actually have a CPU frequency change take
7122 * place in between the computation of X and the setting of the
7123 * variable. To protect against this problem, all updates of
7124 * the per_cpu tsc_khz variable are done in an interrupt
7125 * protected IPI, and all callers wishing to update the value
7126 * must wait for a synchronous IPI to complete (which is trivial
7127 * if the caller is on the CPU already). This establishes the
7128 * necessary total order on variable updates.
7130 * Note that because a guest time update may take place
7131 * anytime after the setting of the VCPU's request bit, the
7132 * correct TSC value must be set before the request. However,
7133 * to ensure the update actually makes it to any guest which
7134 * starts running in hardware virtualization between the set
7135 * and the acquisition of the spinlock, we must also ping the
7136 * CPU after setting the request bit.
7140 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7142 mutex_lock(&kvm_lock);
7143 list_for_each_entry(kvm, &vm_list, vm_list) {
7144 kvm_for_each_vcpu(i, vcpu, kvm) {
7145 if (vcpu->cpu != cpu)
7147 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7148 if (vcpu->cpu != raw_smp_processor_id())
7152 mutex_unlock(&kvm_lock);
7154 if (freq->old < freq->new && send_ipi) {
7156 * We upscale the frequency. Must make the guest
7157 * doesn't see old kvmclock values while running with
7158 * the new frequency, otherwise we risk the guest sees
7159 * time go backwards.
7161 * In case we update the frequency for another cpu
7162 * (which might be in guest context) send an interrupt
7163 * to kick the cpu out of guest context. Next time
7164 * guest context is entered kvmclock will be updated,
7165 * so the guest will not see stale values.
7167 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7171 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7174 struct cpufreq_freqs *freq = data;
7177 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7179 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7182 for_each_cpu(cpu, freq->policy->cpus)
7183 __kvmclock_cpufreq_notifier(freq, cpu);
7188 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7189 .notifier_call = kvmclock_cpufreq_notifier
7192 static int kvmclock_cpu_online(unsigned int cpu)
7194 tsc_khz_changed(NULL);
7198 static void kvm_timer_init(void)
7200 max_tsc_khz = tsc_khz;
7202 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7203 #ifdef CONFIG_CPU_FREQ
7204 struct cpufreq_policy *policy;
7208 policy = cpufreq_cpu_get(cpu);
7210 if (policy->cpuinfo.max_freq)
7211 max_tsc_khz = policy->cpuinfo.max_freq;
7212 cpufreq_cpu_put(policy);
7216 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7217 CPUFREQ_TRANSITION_NOTIFIER);
7220 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7221 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7224 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7225 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7227 int kvm_is_in_guest(void)
7229 return __this_cpu_read(current_vcpu) != NULL;
7232 static int kvm_is_user_mode(void)
7236 if (__this_cpu_read(current_vcpu))
7237 user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7239 return user_mode != 0;
7242 static unsigned long kvm_get_guest_ip(void)
7244 unsigned long ip = 0;
7246 if (__this_cpu_read(current_vcpu))
7247 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7252 static void kvm_handle_intel_pt_intr(void)
7254 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7256 kvm_make_request(KVM_REQ_PMI, vcpu);
7257 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7258 (unsigned long *)&vcpu->arch.pmu.global_status);
7261 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7262 .is_in_guest = kvm_is_in_guest,
7263 .is_user_mode = kvm_is_user_mode,
7264 .get_guest_ip = kvm_get_guest_ip,
7265 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7268 #ifdef CONFIG_X86_64
7269 static void pvclock_gtod_update_fn(struct work_struct *work)
7273 struct kvm_vcpu *vcpu;
7276 mutex_lock(&kvm_lock);
7277 list_for_each_entry(kvm, &vm_list, vm_list)
7278 kvm_for_each_vcpu(i, vcpu, kvm)
7279 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7280 atomic_set(&kvm_guest_has_master_clock, 0);
7281 mutex_unlock(&kvm_lock);
7284 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7287 * Notification about pvclock gtod data update.
7289 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7292 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7293 struct timekeeper *tk = priv;
7295 update_pvclock_gtod(tk);
7297 /* disable master clock if host does not trust, or does not
7298 * use, TSC based clocksource.
7300 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7301 atomic_read(&kvm_guest_has_master_clock) != 0)
7302 queue_work(system_long_wq, &pvclock_gtod_work);
7307 static struct notifier_block pvclock_gtod_notifier = {
7308 .notifier_call = pvclock_gtod_notify,
7312 int kvm_arch_init(void *opaque)
7314 struct kvm_x86_init_ops *ops = opaque;
7317 if (kvm_x86_ops.hardware_enable) {
7318 printk(KERN_ERR "kvm: already loaded the other module\n");
7323 if (!ops->cpu_has_kvm_support()) {
7324 pr_err_ratelimited("kvm: no hardware support\n");
7328 if (ops->disabled_by_bios()) {
7329 pr_err_ratelimited("kvm: disabled by bios\n");
7335 * KVM explicitly assumes that the guest has an FPU and
7336 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7337 * vCPU's FPU state as a fxregs_state struct.
7339 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7340 printk(KERN_ERR "kvm: inadequate fpu\n");
7346 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7347 __alignof__(struct fpu), SLAB_ACCOUNT,
7349 if (!x86_fpu_cache) {
7350 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7354 x86_emulator_cache = kvm_alloc_emulator_cache();
7355 if (!x86_emulator_cache) {
7356 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7357 goto out_free_x86_fpu_cache;
7360 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7362 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7363 goto out_free_x86_emulator_cache;
7366 r = kvm_mmu_module_init();
7368 goto out_free_percpu;
7370 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7371 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7372 PT_PRESENT_MASK, 0, sme_me_mask);
7375 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7377 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7378 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7379 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7383 if (pi_inject_timer == -1)
7384 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7385 #ifdef CONFIG_X86_64
7386 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7388 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7389 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7395 free_percpu(shared_msrs);
7396 out_free_x86_emulator_cache:
7397 kmem_cache_destroy(x86_emulator_cache);
7398 out_free_x86_fpu_cache:
7399 kmem_cache_destroy(x86_fpu_cache);
7404 void kvm_arch_exit(void)
7406 #ifdef CONFIG_X86_64
7407 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7408 clear_hv_tscchange_cb();
7411 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7413 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7414 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7415 CPUFREQ_TRANSITION_NOTIFIER);
7416 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7417 #ifdef CONFIG_X86_64
7418 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7420 kvm_x86_ops.hardware_enable = NULL;
7421 kvm_mmu_module_exit();
7422 free_percpu(shared_msrs);
7423 kmem_cache_destroy(x86_fpu_cache);
7426 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7428 ++vcpu->stat.halt_exits;
7429 if (lapic_in_kernel(vcpu)) {
7430 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7433 vcpu->run->exit_reason = KVM_EXIT_HLT;
7437 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7439 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7441 int ret = kvm_skip_emulated_instruction(vcpu);
7443 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7444 * KVM_EXIT_DEBUG here.
7446 return kvm_vcpu_halt(vcpu) && ret;
7448 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7450 #ifdef CONFIG_X86_64
7451 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7452 unsigned long clock_type)
7454 struct kvm_clock_pairing clock_pairing;
7455 struct timespec64 ts;
7459 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7460 return -KVM_EOPNOTSUPP;
7462 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7463 return -KVM_EOPNOTSUPP;
7465 clock_pairing.sec = ts.tv_sec;
7466 clock_pairing.nsec = ts.tv_nsec;
7467 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7468 clock_pairing.flags = 0;
7469 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7472 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7473 sizeof(struct kvm_clock_pairing)))
7481 * kvm_pv_kick_cpu_op: Kick a vcpu.
7483 * @apicid - apicid of vcpu to be kicked.
7485 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7487 struct kvm_lapic_irq lapic_irq;
7489 lapic_irq.shorthand = APIC_DEST_NOSHORT;
7490 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7491 lapic_irq.level = 0;
7492 lapic_irq.dest_id = apicid;
7493 lapic_irq.msi_redir_hint = false;
7495 lapic_irq.delivery_mode = APIC_DM_REMRD;
7496 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7499 bool kvm_apicv_activated(struct kvm *kvm)
7501 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
7503 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
7505 void kvm_apicv_init(struct kvm *kvm, bool enable)
7508 clear_bit(APICV_INHIBIT_REASON_DISABLE,
7509 &kvm->arch.apicv_inhibit_reasons);
7511 set_bit(APICV_INHIBIT_REASON_DISABLE,
7512 &kvm->arch.apicv_inhibit_reasons);
7514 EXPORT_SYMBOL_GPL(kvm_apicv_init);
7516 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7518 struct kvm_vcpu *target = NULL;
7519 struct kvm_apic_map *map;
7522 map = rcu_dereference(kvm->arch.apic_map);
7524 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7525 target = map->phys_map[dest_id]->vcpu;
7529 if (target && READ_ONCE(target->ready))
7530 kvm_vcpu_yield_to(target);
7533 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7535 unsigned long nr, a0, a1, a2, a3, ret;
7538 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7539 return kvm_hv_hypercall(vcpu);
7541 nr = kvm_rax_read(vcpu);
7542 a0 = kvm_rbx_read(vcpu);
7543 a1 = kvm_rcx_read(vcpu);
7544 a2 = kvm_rdx_read(vcpu);
7545 a3 = kvm_rsi_read(vcpu);
7547 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7549 op_64_bit = is_64_bit_mode(vcpu);
7558 if (kvm_x86_ops.get_cpl(vcpu) != 0) {
7564 case KVM_HC_VAPIC_POLL_IRQ:
7567 case KVM_HC_KICK_CPU:
7568 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7569 kvm_sched_yield(vcpu->kvm, a1);
7572 #ifdef CONFIG_X86_64
7573 case KVM_HC_CLOCK_PAIRING:
7574 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7577 case KVM_HC_SEND_IPI:
7578 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7580 case KVM_HC_SCHED_YIELD:
7581 kvm_sched_yield(vcpu->kvm, a0);
7591 kvm_rax_write(vcpu, ret);
7593 ++vcpu->stat.hypercalls;
7594 return kvm_skip_emulated_instruction(vcpu);
7596 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7598 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7600 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7601 char instruction[3];
7602 unsigned long rip = kvm_rip_read(vcpu);
7604 kvm_x86_ops.patch_hypercall(vcpu, instruction);
7606 return emulator_write_emulated(ctxt, rip, instruction, 3,
7610 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7612 return vcpu->run->request_interrupt_window &&
7613 likely(!pic_in_kernel(vcpu->kvm));
7616 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7618 struct kvm_run *kvm_run = vcpu->run;
7620 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7621 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7622 kvm_run->cr8 = kvm_get_cr8(vcpu);
7623 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7624 kvm_run->ready_for_interrupt_injection =
7625 pic_in_kernel(vcpu->kvm) ||
7626 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7629 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7633 if (!kvm_x86_ops.update_cr8_intercept)
7636 if (!lapic_in_kernel(vcpu))
7639 if (vcpu->arch.apicv_active)
7642 if (!vcpu->arch.apic->vapic_addr)
7643 max_irr = kvm_lapic_find_highest_irr(vcpu);
7650 tpr = kvm_lapic_get_cr8(vcpu);
7652 kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
7655 static int inject_pending_event(struct kvm_vcpu *vcpu)
7659 /* try to reinject previous events if any */
7661 if (vcpu->arch.exception.injected)
7662 kvm_x86_ops.queue_exception(vcpu);
7664 * Do not inject an NMI or interrupt if there is a pending
7665 * exception. Exceptions and interrupts are recognized at
7666 * instruction boundaries, i.e. the start of an instruction.
7667 * Trap-like exceptions, e.g. #DB, have higher priority than
7668 * NMIs and interrupts, i.e. traps are recognized before an
7669 * NMI/interrupt that's pending on the same instruction.
7670 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7671 * priority, but are only generated (pended) during instruction
7672 * execution, i.e. a pending fault-like exception means the
7673 * fault occurred on the *previous* instruction and must be
7674 * serviced prior to recognizing any new events in order to
7675 * fully complete the previous instruction.
7677 else if (!vcpu->arch.exception.pending) {
7678 if (vcpu->arch.nmi_injected)
7679 kvm_x86_ops.set_nmi(vcpu);
7680 else if (vcpu->arch.interrupt.injected)
7681 kvm_x86_ops.set_irq(vcpu);
7685 * Call check_nested_events() even if we reinjected a previous event
7686 * in order for caller to determine if it should require immediate-exit
7687 * from L2 to L1 due to pending L1 events which require exit
7690 if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) {
7691 r = kvm_x86_ops.check_nested_events(vcpu);
7696 /* try to inject new event if pending */
7697 if (vcpu->arch.exception.pending) {
7698 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7699 vcpu->arch.exception.has_error_code,
7700 vcpu->arch.exception.error_code);
7702 WARN_ON_ONCE(vcpu->arch.exception.injected);
7703 vcpu->arch.exception.pending = false;
7704 vcpu->arch.exception.injected = true;
7706 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7707 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7710 if (vcpu->arch.exception.nr == DB_VECTOR) {
7712 * This code assumes that nSVM doesn't use
7713 * check_nested_events(). If it does, the
7714 * DR6/DR7 changes should happen before L1
7715 * gets a #VMEXIT for an intercepted #DB in
7716 * L2. (Under VMX, on the other hand, the
7717 * DR6/DR7 changes should not happen in the
7718 * event of a VM-exit to L1 for an intercepted
7721 kvm_deliver_exception_payload(vcpu);
7722 if (vcpu->arch.dr7 & DR7_GD) {
7723 vcpu->arch.dr7 &= ~DR7_GD;
7724 kvm_update_dr7(vcpu);
7728 kvm_x86_ops.queue_exception(vcpu);
7731 /* Don't consider new event if we re-injected an event */
7732 if (kvm_event_needs_reinjection(vcpu))
7735 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7736 kvm_x86_ops.smi_allowed(vcpu)) {
7737 vcpu->arch.smi_pending = false;
7738 ++vcpu->arch.smi_count;
7740 } else if (vcpu->arch.nmi_pending && kvm_x86_ops.nmi_allowed(vcpu)) {
7741 --vcpu->arch.nmi_pending;
7742 vcpu->arch.nmi_injected = true;
7743 kvm_x86_ops.set_nmi(vcpu);
7744 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7746 * Because interrupts can be injected asynchronously, we are
7747 * calling check_nested_events again here to avoid a race condition.
7748 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7749 * proposal and current concerns. Perhaps we should be setting
7750 * KVM_REQ_EVENT only on certain events and not unconditionally?
7752 if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) {
7753 r = kvm_x86_ops.check_nested_events(vcpu);
7757 if (kvm_x86_ops.interrupt_allowed(vcpu)) {
7758 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7760 kvm_x86_ops.set_irq(vcpu);
7767 static void process_nmi(struct kvm_vcpu *vcpu)
7772 * x86 is limited to one NMI running, and one NMI pending after it.
7773 * If an NMI is already in progress, limit further NMIs to just one.
7774 * Otherwise, allow two (and we'll inject the first one immediately).
7776 if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7779 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7780 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7781 kvm_make_request(KVM_REQ_EVENT, vcpu);
7784 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7787 flags |= seg->g << 23;
7788 flags |= seg->db << 22;
7789 flags |= seg->l << 21;
7790 flags |= seg->avl << 20;
7791 flags |= seg->present << 15;
7792 flags |= seg->dpl << 13;
7793 flags |= seg->s << 12;
7794 flags |= seg->type << 8;
7798 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7800 struct kvm_segment seg;
7803 kvm_get_segment(vcpu, &seg, n);
7804 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7807 offset = 0x7f84 + n * 12;
7809 offset = 0x7f2c + (n - 3) * 12;
7811 put_smstate(u32, buf, offset + 8, seg.base);
7812 put_smstate(u32, buf, offset + 4, seg.limit);
7813 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7816 #ifdef CONFIG_X86_64
7817 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7819 struct kvm_segment seg;
7823 kvm_get_segment(vcpu, &seg, n);
7824 offset = 0x7e00 + n * 16;
7826 flags = enter_smm_get_segment_flags(&seg) >> 8;
7827 put_smstate(u16, buf, offset, seg.selector);
7828 put_smstate(u16, buf, offset + 2, flags);
7829 put_smstate(u32, buf, offset + 4, seg.limit);
7830 put_smstate(u64, buf, offset + 8, seg.base);
7834 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7837 struct kvm_segment seg;
7841 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7842 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7843 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7844 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7846 for (i = 0; i < 8; i++)
7847 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7849 kvm_get_dr(vcpu, 6, &val);
7850 put_smstate(u32, buf, 0x7fcc, (u32)val);
7851 kvm_get_dr(vcpu, 7, &val);
7852 put_smstate(u32, buf, 0x7fc8, (u32)val);
7854 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7855 put_smstate(u32, buf, 0x7fc4, seg.selector);
7856 put_smstate(u32, buf, 0x7f64, seg.base);
7857 put_smstate(u32, buf, 0x7f60, seg.limit);
7858 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7860 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7861 put_smstate(u32, buf, 0x7fc0, seg.selector);
7862 put_smstate(u32, buf, 0x7f80, seg.base);
7863 put_smstate(u32, buf, 0x7f7c, seg.limit);
7864 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7866 kvm_x86_ops.get_gdt(vcpu, &dt);
7867 put_smstate(u32, buf, 0x7f74, dt.address);
7868 put_smstate(u32, buf, 0x7f70, dt.size);
7870 kvm_x86_ops.get_idt(vcpu, &dt);
7871 put_smstate(u32, buf, 0x7f58, dt.address);
7872 put_smstate(u32, buf, 0x7f54, dt.size);
7874 for (i = 0; i < 6; i++)
7875 enter_smm_save_seg_32(vcpu, buf, i);
7877 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7880 put_smstate(u32, buf, 0x7efc, 0x00020000);
7881 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7884 #ifdef CONFIG_X86_64
7885 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7888 struct kvm_segment seg;
7892 for (i = 0; i < 16; i++)
7893 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7895 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7896 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7898 kvm_get_dr(vcpu, 6, &val);
7899 put_smstate(u64, buf, 0x7f68, val);
7900 kvm_get_dr(vcpu, 7, &val);
7901 put_smstate(u64, buf, 0x7f60, val);
7903 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7904 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7905 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7907 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7910 put_smstate(u32, buf, 0x7efc, 0x00020064);
7912 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7914 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7915 put_smstate(u16, buf, 0x7e90, seg.selector);
7916 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7917 put_smstate(u32, buf, 0x7e94, seg.limit);
7918 put_smstate(u64, buf, 0x7e98, seg.base);
7920 kvm_x86_ops.get_idt(vcpu, &dt);
7921 put_smstate(u32, buf, 0x7e84, dt.size);
7922 put_smstate(u64, buf, 0x7e88, dt.address);
7924 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7925 put_smstate(u16, buf, 0x7e70, seg.selector);
7926 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7927 put_smstate(u32, buf, 0x7e74, seg.limit);
7928 put_smstate(u64, buf, 0x7e78, seg.base);
7930 kvm_x86_ops.get_gdt(vcpu, &dt);
7931 put_smstate(u32, buf, 0x7e64, dt.size);
7932 put_smstate(u64, buf, 0x7e68, dt.address);
7934 for (i = 0; i < 6; i++)
7935 enter_smm_save_seg_64(vcpu, buf, i);
7939 static void enter_smm(struct kvm_vcpu *vcpu)
7941 struct kvm_segment cs, ds;
7946 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7947 memset(buf, 0, 512);
7948 #ifdef CONFIG_X86_64
7949 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7950 enter_smm_save_state_64(vcpu, buf);
7953 enter_smm_save_state_32(vcpu, buf);
7956 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7957 * vCPU state (e.g. leave guest mode) after we've saved the state into
7958 * the SMM state-save area.
7960 kvm_x86_ops.pre_enter_smm(vcpu, buf);
7962 vcpu->arch.hflags |= HF_SMM_MASK;
7963 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7965 if (kvm_x86_ops.get_nmi_mask(vcpu))
7966 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7968 kvm_x86_ops.set_nmi_mask(vcpu, true);
7970 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7971 kvm_rip_write(vcpu, 0x8000);
7973 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7974 kvm_x86_ops.set_cr0(vcpu, cr0);
7975 vcpu->arch.cr0 = cr0;
7977 kvm_x86_ops.set_cr4(vcpu, 0);
7979 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7980 dt.address = dt.size = 0;
7981 kvm_x86_ops.set_idt(vcpu, &dt);
7983 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7985 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7986 cs.base = vcpu->arch.smbase;
7991 cs.limit = ds.limit = 0xffffffff;
7992 cs.type = ds.type = 0x3;
7993 cs.dpl = ds.dpl = 0;
7998 cs.avl = ds.avl = 0;
7999 cs.present = ds.present = 1;
8000 cs.unusable = ds.unusable = 0;
8001 cs.padding = ds.padding = 0;
8003 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8004 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8005 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8006 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8007 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8008 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8010 #ifdef CONFIG_X86_64
8011 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8012 kvm_x86_ops.set_efer(vcpu, 0);
8015 kvm_update_cpuid(vcpu);
8016 kvm_mmu_reset_context(vcpu);
8019 static void process_smi(struct kvm_vcpu *vcpu)
8021 vcpu->arch.smi_pending = true;
8022 kvm_make_request(KVM_REQ_EVENT, vcpu);
8025 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8026 unsigned long *vcpu_bitmap)
8030 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8032 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8035 free_cpumask_var(cpus);
8038 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8040 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8043 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8045 if (!lapic_in_kernel(vcpu))
8048 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8049 kvm_apic_update_apicv(vcpu);
8050 kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8052 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8055 * NOTE: Do not hold any lock prior to calling this.
8057 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8058 * locked, because it calls __x86_set_memory_region() which does
8059 * synchronize_srcu(&kvm->srcu).
8061 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8063 unsigned long old, new, expected;
8065 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8066 !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8069 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8071 expected = new = old;
8073 __clear_bit(bit, &new);
8075 __set_bit(bit, &new);
8078 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8079 } while (old != expected);
8084 trace_kvm_apicv_update_request(activate, bit);
8085 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8086 kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8087 kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
8089 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8091 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8093 if (!kvm_apic_present(vcpu))
8096 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8098 if (irqchip_split(vcpu->kvm))
8099 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8101 if (vcpu->arch.apicv_active)
8102 kvm_x86_ops.sync_pir_to_irr(vcpu);
8103 if (ioapic_in_kernel(vcpu->kvm))
8104 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8107 if (is_guest_mode(vcpu))
8108 vcpu->arch.load_eoi_exitmap_pending = true;
8110 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8113 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8115 u64 eoi_exit_bitmap[4];
8117 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8120 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8121 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8122 kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8125 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8126 unsigned long start, unsigned long end,
8129 unsigned long apic_address;
8132 * The physical address of apic access page is stored in the VMCS.
8133 * Update it when it becomes invalid.
8135 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8136 if (start <= apic_address && apic_address < end)
8137 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8142 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8144 struct page *page = NULL;
8146 if (!lapic_in_kernel(vcpu))
8149 if (!kvm_x86_ops.set_apic_access_page_addr)
8152 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8153 if (is_error_page(page))
8155 kvm_x86_ops.set_apic_access_page_addr(vcpu, page_to_phys(page));
8158 * Do not pin apic access page in memory, the MMU notifier
8159 * will call us again if it is migrated or swapped out.
8164 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8166 smp_send_reschedule(vcpu->cpu);
8168 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8171 * Returns 1 to let vcpu_run() continue the guest execution loop without
8172 * exiting to the userspace. Otherwise, the value will be returned to the
8175 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8179 dm_request_for_irq_injection(vcpu) &&
8180 kvm_cpu_accept_dm_intr(vcpu);
8181 enum exit_fastpath_completion exit_fastpath = EXIT_FASTPATH_NONE;
8183 bool req_immediate_exit = false;
8185 if (kvm_request_pending(vcpu)) {
8186 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8187 if (unlikely(!kvm_x86_ops.get_vmcs12_pages(vcpu))) {
8192 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8193 kvm_mmu_unload(vcpu);
8194 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8195 __kvm_migrate_timers(vcpu);
8196 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8197 kvm_gen_update_masterclock(vcpu->kvm);
8198 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8199 kvm_gen_kvmclock_update(vcpu);
8200 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8201 r = kvm_guest_time_update(vcpu);
8205 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8206 kvm_mmu_sync_roots(vcpu);
8207 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8208 kvm_mmu_load_pgd(vcpu);
8209 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
8210 kvm_vcpu_flush_tlb(vcpu, true);
8211 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8212 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8216 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8217 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8218 vcpu->mmio_needed = 0;
8222 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8223 /* Page is swapped out. Do synthetic halt */
8224 vcpu->arch.apf.halted = true;
8228 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8229 record_steal_time(vcpu);
8230 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8232 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8234 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8235 kvm_pmu_handle_event(vcpu);
8236 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8237 kvm_pmu_deliver_pmi(vcpu);
8238 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8239 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8240 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8241 vcpu->arch.ioapic_handled_vectors)) {
8242 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8243 vcpu->run->eoi.vector =
8244 vcpu->arch.pending_ioapic_eoi;
8249 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8250 vcpu_scan_ioapic(vcpu);
8251 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8252 vcpu_load_eoi_exitmap(vcpu);
8253 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8254 kvm_vcpu_reload_apic_access_page(vcpu);
8255 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8256 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8257 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8261 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8262 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8263 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8267 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8268 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8269 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8275 * KVM_REQ_HV_STIMER has to be processed after
8276 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8277 * depend on the guest clock being up-to-date
8279 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8280 kvm_hv_process_stimers(vcpu);
8281 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8282 kvm_vcpu_update_apicv(vcpu);
8285 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8286 ++vcpu->stat.req_event;
8287 kvm_apic_accept_events(vcpu);
8288 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8293 if (inject_pending_event(vcpu) != 0)
8294 req_immediate_exit = true;
8296 /* Enable SMI/NMI/IRQ window open exits if needed.
8298 * SMIs have three cases:
8299 * 1) They can be nested, and then there is nothing to
8300 * do here because RSM will cause a vmexit anyway.
8301 * 2) There is an ISA-specific reason why SMI cannot be
8302 * injected, and the moment when this changes can be
8304 * 3) Or the SMI can be pending because
8305 * inject_pending_event has completed the injection
8306 * of an IRQ or NMI from the previous vmexit, and
8307 * then we request an immediate exit to inject the
8310 if (vcpu->arch.smi_pending && !is_smm(vcpu))
8311 if (!kvm_x86_ops.enable_smi_window(vcpu))
8312 req_immediate_exit = true;
8313 if (vcpu->arch.nmi_pending)
8314 kvm_x86_ops.enable_nmi_window(vcpu);
8315 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8316 kvm_x86_ops.enable_irq_window(vcpu);
8317 WARN_ON(vcpu->arch.exception.pending);
8320 if (kvm_lapic_enabled(vcpu)) {
8321 update_cr8_intercept(vcpu);
8322 kvm_lapic_sync_to_vapic(vcpu);
8326 r = kvm_mmu_reload(vcpu);
8328 goto cancel_injection;
8333 kvm_x86_ops.prepare_guest_switch(vcpu);
8336 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8337 * IPI are then delayed after guest entry, which ensures that they
8338 * result in virtual interrupt delivery.
8340 local_irq_disable();
8341 vcpu->mode = IN_GUEST_MODE;
8343 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8346 * 1) We should set ->mode before checking ->requests. Please see
8347 * the comment in kvm_vcpu_exiting_guest_mode().
8349 * 2) For APICv, we should set ->mode before checking PID.ON. This
8350 * pairs with the memory barrier implicit in pi_test_and_set_on
8351 * (see vmx_deliver_posted_interrupt).
8353 * 3) This also orders the write to mode from any reads to the page
8354 * tables done while the VCPU is running. Please see the comment
8355 * in kvm_flush_remote_tlbs.
8357 smp_mb__after_srcu_read_unlock();
8360 * This handles the case where a posted interrupt was
8361 * notified with kvm_vcpu_kick.
8363 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8364 kvm_x86_ops.sync_pir_to_irr(vcpu);
8366 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8367 || need_resched() || signal_pending(current)) {
8368 vcpu->mode = OUTSIDE_GUEST_MODE;
8372 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8374 goto cancel_injection;
8377 if (req_immediate_exit) {
8378 kvm_make_request(KVM_REQ_EVENT, vcpu);
8379 kvm_x86_ops.request_immediate_exit(vcpu);
8382 trace_kvm_entry(vcpu->vcpu_id);
8383 guest_enter_irqoff();
8385 fpregs_assert_state_consistent();
8386 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8387 switch_fpu_return();
8389 if (unlikely(vcpu->arch.switch_db_regs)) {
8391 set_debugreg(vcpu->arch.eff_db[0], 0);
8392 set_debugreg(vcpu->arch.eff_db[1], 1);
8393 set_debugreg(vcpu->arch.eff_db[2], 2);
8394 set_debugreg(vcpu->arch.eff_db[3], 3);
8395 set_debugreg(vcpu->arch.dr6, 6);
8396 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8399 kvm_x86_ops.run(vcpu);
8402 * Do this here before restoring debug registers on the host. And
8403 * since we do this before handling the vmexit, a DR access vmexit
8404 * can (a) read the correct value of the debug registers, (b) set
8405 * KVM_DEBUGREG_WONT_EXIT again.
8407 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8408 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8409 kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8410 kvm_update_dr0123(vcpu);
8411 kvm_update_dr7(vcpu);
8412 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8416 * If the guest has used debug registers, at least dr7
8417 * will be disabled while returning to the host.
8418 * If we don't have active breakpoints in the host, we don't
8419 * care about the messed up debug address registers. But if
8420 * we have some of them active, restore the old state.
8422 if (hw_breakpoint_active())
8423 hw_breakpoint_restore();
8425 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8427 vcpu->mode = OUTSIDE_GUEST_MODE;
8430 kvm_x86_ops.handle_exit_irqoff(vcpu, &exit_fastpath);
8433 * Consume any pending interrupts, including the possible source of
8434 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8435 * An instruction is required after local_irq_enable() to fully unblock
8436 * interrupts on processors that implement an interrupt shadow, the
8437 * stat.exits increment will do nicely.
8439 kvm_before_interrupt(vcpu);
8442 local_irq_disable();
8443 kvm_after_interrupt(vcpu);
8445 guest_exit_irqoff();
8446 if (lapic_in_kernel(vcpu)) {
8447 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8448 if (delta != S64_MIN) {
8449 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8450 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8457 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8460 * Profile KVM exit RIPs:
8462 if (unlikely(prof_on == KVM_PROFILING)) {
8463 unsigned long rip = kvm_rip_read(vcpu);
8464 profile_hit(KVM_PROFILING, (void *)rip);
8467 if (unlikely(vcpu->arch.tsc_always_catchup))
8468 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8470 if (vcpu->arch.apic_attention)
8471 kvm_lapic_sync_from_vapic(vcpu);
8473 r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
8477 kvm_x86_ops.cancel_injection(vcpu);
8478 if (unlikely(vcpu->arch.apic_attention))
8479 kvm_lapic_sync_from_vapic(vcpu);
8484 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8486 if (!kvm_arch_vcpu_runnable(vcpu) &&
8487 (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
8488 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8489 kvm_vcpu_block(vcpu);
8490 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8492 if (kvm_x86_ops.post_block)
8493 kvm_x86_ops.post_block(vcpu);
8495 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8499 kvm_apic_accept_events(vcpu);
8500 switch(vcpu->arch.mp_state) {
8501 case KVM_MP_STATE_HALTED:
8502 vcpu->arch.pv.pv_unhalted = false;
8503 vcpu->arch.mp_state =
8504 KVM_MP_STATE_RUNNABLE;
8506 case KVM_MP_STATE_RUNNABLE:
8507 vcpu->arch.apf.halted = false;
8509 case KVM_MP_STATE_INIT_RECEIVED:
8517 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8519 if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events)
8520 kvm_x86_ops.check_nested_events(vcpu);
8522 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8523 !vcpu->arch.apf.halted);
8526 static int vcpu_run(struct kvm_vcpu *vcpu)
8529 struct kvm *kvm = vcpu->kvm;
8531 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8532 vcpu->arch.l1tf_flush_l1d = true;
8535 if (kvm_vcpu_running(vcpu)) {
8536 r = vcpu_enter_guest(vcpu);
8538 r = vcpu_block(kvm, vcpu);
8544 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8545 if (kvm_cpu_has_pending_timer(vcpu))
8546 kvm_inject_pending_timer_irqs(vcpu);
8548 if (dm_request_for_irq_injection(vcpu) &&
8549 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8551 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8552 ++vcpu->stat.request_irq_exits;
8556 kvm_check_async_pf_completion(vcpu);
8558 if (signal_pending(current)) {
8560 vcpu->run->exit_reason = KVM_EXIT_INTR;
8561 ++vcpu->stat.signal_exits;
8564 if (need_resched()) {
8565 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8567 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8571 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8576 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8580 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8581 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8582 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8586 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8588 BUG_ON(!vcpu->arch.pio.count);
8590 return complete_emulated_io(vcpu);
8594 * Implements the following, as a state machine:
8598 * for each mmio piece in the fragment
8606 * for each mmio piece in the fragment
8611 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8613 struct kvm_run *run = vcpu->run;
8614 struct kvm_mmio_fragment *frag;
8617 BUG_ON(!vcpu->mmio_needed);
8619 /* Complete previous fragment */
8620 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8621 len = min(8u, frag->len);
8622 if (!vcpu->mmio_is_write)
8623 memcpy(frag->data, run->mmio.data, len);
8625 if (frag->len <= 8) {
8626 /* Switch to the next fragment. */
8628 vcpu->mmio_cur_fragment++;
8630 /* Go forward to the next mmio piece. */
8636 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8637 vcpu->mmio_needed = 0;
8639 /* FIXME: return into emulator if single-stepping. */
8640 if (vcpu->mmio_is_write)
8642 vcpu->mmio_read_completed = 1;
8643 return complete_emulated_io(vcpu);
8646 run->exit_reason = KVM_EXIT_MMIO;
8647 run->mmio.phys_addr = frag->gpa;
8648 if (vcpu->mmio_is_write)
8649 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8650 run->mmio.len = min(8u, frag->len);
8651 run->mmio.is_write = vcpu->mmio_is_write;
8652 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8656 static void kvm_save_current_fpu(struct fpu *fpu)
8659 * If the target FPU state is not resident in the CPU registers, just
8660 * memcpy() from current, else save CPU state directly to the target.
8662 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8663 memcpy(&fpu->state, ¤t->thread.fpu.state,
8664 fpu_kernel_xstate_size);
8666 copy_fpregs_to_fpstate(fpu);
8669 /* Swap (qemu) user FPU context for the guest FPU context. */
8670 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8674 kvm_save_current_fpu(vcpu->arch.user_fpu);
8676 /* PKRU is separately restored in kvm_x86_ops.run. */
8677 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8678 ~XFEATURE_MASK_PKRU);
8680 fpregs_mark_activate();
8686 /* When vcpu_run ends, restore user space FPU context. */
8687 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8691 kvm_save_current_fpu(vcpu->arch.guest_fpu);
8693 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8695 fpregs_mark_activate();
8698 ++vcpu->stat.fpu_reload;
8702 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8707 kvm_sigset_activate(vcpu);
8708 kvm_load_guest_fpu(vcpu);
8710 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8711 if (kvm_run->immediate_exit) {
8715 kvm_vcpu_block(vcpu);
8716 kvm_apic_accept_events(vcpu);
8717 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8719 if (signal_pending(current)) {
8721 vcpu->run->exit_reason = KVM_EXIT_INTR;
8722 ++vcpu->stat.signal_exits;
8727 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8732 if (vcpu->run->kvm_dirty_regs) {
8733 r = sync_regs(vcpu);
8738 /* re-sync apic's tpr */
8739 if (!lapic_in_kernel(vcpu)) {
8740 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8746 if (unlikely(vcpu->arch.complete_userspace_io)) {
8747 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8748 vcpu->arch.complete_userspace_io = NULL;
8753 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8755 if (kvm_run->immediate_exit)
8761 kvm_put_guest_fpu(vcpu);
8762 if (vcpu->run->kvm_valid_regs)
8764 post_kvm_run_save(vcpu);
8765 kvm_sigset_deactivate(vcpu);
8771 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8773 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8775 * We are here if userspace calls get_regs() in the middle of
8776 * instruction emulation. Registers state needs to be copied
8777 * back from emulation context to vcpu. Userspace shouldn't do
8778 * that usually, but some bad designed PV devices (vmware
8779 * backdoor interface) need this to work
8781 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
8782 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8784 regs->rax = kvm_rax_read(vcpu);
8785 regs->rbx = kvm_rbx_read(vcpu);
8786 regs->rcx = kvm_rcx_read(vcpu);
8787 regs->rdx = kvm_rdx_read(vcpu);
8788 regs->rsi = kvm_rsi_read(vcpu);
8789 regs->rdi = kvm_rdi_read(vcpu);
8790 regs->rsp = kvm_rsp_read(vcpu);
8791 regs->rbp = kvm_rbp_read(vcpu);
8792 #ifdef CONFIG_X86_64
8793 regs->r8 = kvm_r8_read(vcpu);
8794 regs->r9 = kvm_r9_read(vcpu);
8795 regs->r10 = kvm_r10_read(vcpu);
8796 regs->r11 = kvm_r11_read(vcpu);
8797 regs->r12 = kvm_r12_read(vcpu);
8798 regs->r13 = kvm_r13_read(vcpu);
8799 regs->r14 = kvm_r14_read(vcpu);
8800 regs->r15 = kvm_r15_read(vcpu);
8803 regs->rip = kvm_rip_read(vcpu);
8804 regs->rflags = kvm_get_rflags(vcpu);
8807 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8810 __get_regs(vcpu, regs);
8815 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8817 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8818 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8820 kvm_rax_write(vcpu, regs->rax);
8821 kvm_rbx_write(vcpu, regs->rbx);
8822 kvm_rcx_write(vcpu, regs->rcx);
8823 kvm_rdx_write(vcpu, regs->rdx);
8824 kvm_rsi_write(vcpu, regs->rsi);
8825 kvm_rdi_write(vcpu, regs->rdi);
8826 kvm_rsp_write(vcpu, regs->rsp);
8827 kvm_rbp_write(vcpu, regs->rbp);
8828 #ifdef CONFIG_X86_64
8829 kvm_r8_write(vcpu, regs->r8);
8830 kvm_r9_write(vcpu, regs->r9);
8831 kvm_r10_write(vcpu, regs->r10);
8832 kvm_r11_write(vcpu, regs->r11);
8833 kvm_r12_write(vcpu, regs->r12);
8834 kvm_r13_write(vcpu, regs->r13);
8835 kvm_r14_write(vcpu, regs->r14);
8836 kvm_r15_write(vcpu, regs->r15);
8839 kvm_rip_write(vcpu, regs->rip);
8840 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8842 vcpu->arch.exception.pending = false;
8844 kvm_make_request(KVM_REQ_EVENT, vcpu);
8847 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8850 __set_regs(vcpu, regs);
8855 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8857 struct kvm_segment cs;
8859 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8863 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8865 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8869 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8870 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8871 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8872 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8873 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8874 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8876 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8877 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8879 kvm_x86_ops.get_idt(vcpu, &dt);
8880 sregs->idt.limit = dt.size;
8881 sregs->idt.base = dt.address;
8882 kvm_x86_ops.get_gdt(vcpu, &dt);
8883 sregs->gdt.limit = dt.size;
8884 sregs->gdt.base = dt.address;
8886 sregs->cr0 = kvm_read_cr0(vcpu);
8887 sregs->cr2 = vcpu->arch.cr2;
8888 sregs->cr3 = kvm_read_cr3(vcpu);
8889 sregs->cr4 = kvm_read_cr4(vcpu);
8890 sregs->cr8 = kvm_get_cr8(vcpu);
8891 sregs->efer = vcpu->arch.efer;
8892 sregs->apic_base = kvm_get_apic_base(vcpu);
8894 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8896 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8897 set_bit(vcpu->arch.interrupt.nr,
8898 (unsigned long *)sregs->interrupt_bitmap);
8901 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8902 struct kvm_sregs *sregs)
8905 __get_sregs(vcpu, sregs);
8910 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8911 struct kvm_mp_state *mp_state)
8914 if (kvm_mpx_supported())
8915 kvm_load_guest_fpu(vcpu);
8917 kvm_apic_accept_events(vcpu);
8918 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8919 vcpu->arch.pv.pv_unhalted)
8920 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8922 mp_state->mp_state = vcpu->arch.mp_state;
8924 if (kvm_mpx_supported())
8925 kvm_put_guest_fpu(vcpu);
8930 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8931 struct kvm_mp_state *mp_state)
8937 if (!lapic_in_kernel(vcpu) &&
8938 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8942 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
8943 * INIT state; latched init should be reported using
8944 * KVM_SET_VCPU_EVENTS, so reject it here.
8946 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
8947 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8948 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8951 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8952 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8953 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8955 vcpu->arch.mp_state = mp_state->mp_state;
8956 kvm_make_request(KVM_REQ_EVENT, vcpu);
8964 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8965 int reason, bool has_error_code, u32 error_code)
8967 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8970 init_emulate_ctxt(vcpu);
8972 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8973 has_error_code, error_code);
8975 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8976 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8977 vcpu->run->internal.ndata = 0;
8981 kvm_rip_write(vcpu, ctxt->eip);
8982 kvm_set_rflags(vcpu, ctxt->eflags);
8985 EXPORT_SYMBOL_GPL(kvm_task_switch);
8987 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8989 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8991 * When EFER.LME and CR0.PG are set, the processor is in
8992 * 64-bit mode (though maybe in a 32-bit code segment).
8993 * CR4.PAE and EFER.LMA must be set.
8995 if (!(sregs->cr4 & X86_CR4_PAE)
8996 || !(sregs->efer & EFER_LMA))
9000 * Not in 64-bit mode: EFER.LMA is clear and the code
9001 * segment cannot be 64-bit.
9003 if (sregs->efer & EFER_LMA || sregs->cs.l)
9007 return kvm_valid_cr4(vcpu, sregs->cr4);
9010 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9012 struct msr_data apic_base_msr;
9013 int mmu_reset_needed = 0;
9014 int cpuid_update_needed = 0;
9015 int pending_vec, max_bits, idx;
9019 if (kvm_valid_sregs(vcpu, sregs))
9022 apic_base_msr.data = sregs->apic_base;
9023 apic_base_msr.host_initiated = true;
9024 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9027 dt.size = sregs->idt.limit;
9028 dt.address = sregs->idt.base;
9029 kvm_x86_ops.set_idt(vcpu, &dt);
9030 dt.size = sregs->gdt.limit;
9031 dt.address = sregs->gdt.base;
9032 kvm_x86_ops.set_gdt(vcpu, &dt);
9034 vcpu->arch.cr2 = sregs->cr2;
9035 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9036 vcpu->arch.cr3 = sregs->cr3;
9037 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9039 kvm_set_cr8(vcpu, sregs->cr8);
9041 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9042 kvm_x86_ops.set_efer(vcpu, sregs->efer);
9044 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9045 kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9046 vcpu->arch.cr0 = sregs->cr0;
9048 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9049 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9050 (X86_CR4_OSXSAVE | X86_CR4_PKE));
9051 kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9052 if (cpuid_update_needed)
9053 kvm_update_cpuid(vcpu);
9055 idx = srcu_read_lock(&vcpu->kvm->srcu);
9056 if (is_pae_paging(vcpu)) {
9057 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9058 mmu_reset_needed = 1;
9060 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9062 if (mmu_reset_needed)
9063 kvm_mmu_reset_context(vcpu);
9065 max_bits = KVM_NR_INTERRUPTS;
9066 pending_vec = find_first_bit(
9067 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9068 if (pending_vec < max_bits) {
9069 kvm_queue_interrupt(vcpu, pending_vec, false);
9070 pr_debug("Set back pending irq %d\n", pending_vec);
9073 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9074 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9075 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9076 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9077 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9078 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9080 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9081 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9083 update_cr8_intercept(vcpu);
9085 /* Older userspace won't unhalt the vcpu on reset. */
9086 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9087 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9089 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9091 kvm_make_request(KVM_REQ_EVENT, vcpu);
9098 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9099 struct kvm_sregs *sregs)
9104 ret = __set_sregs(vcpu, sregs);
9109 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9110 struct kvm_guest_debug *dbg)
9112 unsigned long rflags;
9117 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9119 if (vcpu->arch.exception.pending)
9121 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9122 kvm_queue_exception(vcpu, DB_VECTOR);
9124 kvm_queue_exception(vcpu, BP_VECTOR);
9128 * Read rflags as long as potentially injected trace flags are still
9131 rflags = kvm_get_rflags(vcpu);
9133 vcpu->guest_debug = dbg->control;
9134 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9135 vcpu->guest_debug = 0;
9137 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9138 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9139 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9140 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9142 for (i = 0; i < KVM_NR_DB_REGS; i++)
9143 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9145 kvm_update_dr7(vcpu);
9147 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9148 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9149 get_segment_base(vcpu, VCPU_SREG_CS);
9152 * Trigger an rflags update that will inject or remove the trace
9155 kvm_set_rflags(vcpu, rflags);
9157 kvm_x86_ops.update_bp_intercept(vcpu);
9167 * Translate a guest virtual address to a guest physical address.
9169 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9170 struct kvm_translation *tr)
9172 unsigned long vaddr = tr->linear_address;
9178 idx = srcu_read_lock(&vcpu->kvm->srcu);
9179 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9180 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9181 tr->physical_address = gpa;
9182 tr->valid = gpa != UNMAPPED_GVA;
9190 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9192 struct fxregs_state *fxsave;
9196 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9197 memcpy(fpu->fpr, fxsave->st_space, 128);
9198 fpu->fcw = fxsave->cwd;
9199 fpu->fsw = fxsave->swd;
9200 fpu->ftwx = fxsave->twd;
9201 fpu->last_opcode = fxsave->fop;
9202 fpu->last_ip = fxsave->rip;
9203 fpu->last_dp = fxsave->rdp;
9204 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9210 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9212 struct fxregs_state *fxsave;
9216 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9218 memcpy(fxsave->st_space, fpu->fpr, 128);
9219 fxsave->cwd = fpu->fcw;
9220 fxsave->swd = fpu->fsw;
9221 fxsave->twd = fpu->ftwx;
9222 fxsave->fop = fpu->last_opcode;
9223 fxsave->rip = fpu->last_ip;
9224 fxsave->rdp = fpu->last_dp;
9225 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9231 static void store_regs(struct kvm_vcpu *vcpu)
9233 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9235 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9236 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9238 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9239 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9241 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9242 kvm_vcpu_ioctl_x86_get_vcpu_events(
9243 vcpu, &vcpu->run->s.regs.events);
9246 static int sync_regs(struct kvm_vcpu *vcpu)
9248 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9251 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9252 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9253 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9255 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9256 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9258 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9260 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9261 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9262 vcpu, &vcpu->run->s.regs.events))
9264 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9270 static void fx_init(struct kvm_vcpu *vcpu)
9272 fpstate_init(&vcpu->arch.guest_fpu->state);
9273 if (boot_cpu_has(X86_FEATURE_XSAVES))
9274 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9275 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9278 * Ensure guest xcr0 is valid for loading
9280 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9282 vcpu->arch.cr0 |= X86_CR0_ET;
9285 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9287 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9288 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9289 "guest TSC will not be reliable\n");
9294 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9299 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9300 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9302 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9304 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9306 r = kvm_mmu_create(vcpu);
9310 if (irqchip_in_kernel(vcpu->kvm)) {
9311 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9313 goto fail_mmu_destroy;
9314 if (kvm_apicv_activated(vcpu->kvm))
9315 vcpu->arch.apicv_active = true;
9317 static_key_slow_inc(&kvm_no_apic_vcpu);
9321 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9323 goto fail_free_lapic;
9324 vcpu->arch.pio_data = page_address(page);
9326 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9327 GFP_KERNEL_ACCOUNT);
9328 if (!vcpu->arch.mce_banks)
9329 goto fail_free_pio_data;
9330 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9332 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9333 GFP_KERNEL_ACCOUNT))
9334 goto fail_free_mce_banks;
9336 if (!alloc_emulate_ctxt(vcpu))
9337 goto free_wbinvd_dirty_mask;
9339 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9340 GFP_KERNEL_ACCOUNT);
9341 if (!vcpu->arch.user_fpu) {
9342 pr_err("kvm: failed to allocate userspace's fpu\n");
9343 goto free_emulate_ctxt;
9346 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9347 GFP_KERNEL_ACCOUNT);
9348 if (!vcpu->arch.guest_fpu) {
9349 pr_err("kvm: failed to allocate vcpu's fpu\n");
9354 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9356 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9358 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9360 kvm_async_pf_hash_reset(vcpu);
9363 vcpu->arch.pending_external_vector = -1;
9364 vcpu->arch.preempted_in_kernel = false;
9366 kvm_hv_vcpu_init(vcpu);
9368 r = kvm_x86_ops.vcpu_create(vcpu);
9370 goto free_guest_fpu;
9372 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9373 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9374 kvm_vcpu_mtrr_init(vcpu);
9376 kvm_vcpu_reset(vcpu, false);
9377 kvm_init_mmu(vcpu, false);
9382 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9384 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9386 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9387 free_wbinvd_dirty_mask:
9388 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9389 fail_free_mce_banks:
9390 kfree(vcpu->arch.mce_banks);
9392 free_page((unsigned long)vcpu->arch.pio_data);
9394 kvm_free_lapic(vcpu);
9396 kvm_mmu_destroy(vcpu);
9400 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9402 struct msr_data msr;
9403 struct kvm *kvm = vcpu->kvm;
9405 kvm_hv_vcpu_postcreate(vcpu);
9407 if (mutex_lock_killable(&vcpu->mutex))
9411 msr.index = MSR_IA32_TSC;
9412 msr.host_initiated = true;
9413 kvm_write_tsc(vcpu, &msr);
9416 /* poll control enabled by default */
9417 vcpu->arch.msr_kvm_poll_control = 1;
9419 mutex_unlock(&vcpu->mutex);
9421 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9422 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9423 KVMCLOCK_SYNC_PERIOD);
9426 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9428 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9431 kvm_release_pfn(cache->pfn, cache->dirty, cache);
9433 kvmclock_reset(vcpu);
9435 kvm_x86_ops.vcpu_free(vcpu);
9437 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9438 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9439 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9440 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9442 kvm_hv_vcpu_uninit(vcpu);
9443 kvm_pmu_destroy(vcpu);
9444 kfree(vcpu->arch.mce_banks);
9445 kvm_free_lapic(vcpu);
9446 idx = srcu_read_lock(&vcpu->kvm->srcu);
9447 kvm_mmu_destroy(vcpu);
9448 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9449 free_page((unsigned long)vcpu->arch.pio_data);
9450 if (!lapic_in_kernel(vcpu))
9451 static_key_slow_dec(&kvm_no_apic_vcpu);
9454 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9456 kvm_lapic_reset(vcpu, init_event);
9458 vcpu->arch.hflags = 0;
9460 vcpu->arch.smi_pending = 0;
9461 vcpu->arch.smi_count = 0;
9462 atomic_set(&vcpu->arch.nmi_queued, 0);
9463 vcpu->arch.nmi_pending = 0;
9464 vcpu->arch.nmi_injected = false;
9465 kvm_clear_interrupt_queue(vcpu);
9466 kvm_clear_exception_queue(vcpu);
9468 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9469 kvm_update_dr0123(vcpu);
9470 vcpu->arch.dr6 = DR6_INIT;
9471 vcpu->arch.dr7 = DR7_FIXED_1;
9472 kvm_update_dr7(vcpu);
9476 kvm_make_request(KVM_REQ_EVENT, vcpu);
9477 vcpu->arch.apf.msr_val = 0;
9478 vcpu->arch.st.msr_val = 0;
9480 kvmclock_reset(vcpu);
9482 kvm_clear_async_pf_completion_queue(vcpu);
9483 kvm_async_pf_hash_reset(vcpu);
9484 vcpu->arch.apf.halted = false;
9486 if (kvm_mpx_supported()) {
9487 void *mpx_state_buffer;
9490 * To avoid have the INIT path from kvm_apic_has_events() that be
9491 * called with loaded FPU and does not let userspace fix the state.
9494 kvm_put_guest_fpu(vcpu);
9495 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9497 if (mpx_state_buffer)
9498 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9499 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9501 if (mpx_state_buffer)
9502 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9504 kvm_load_guest_fpu(vcpu);
9508 kvm_pmu_reset(vcpu);
9509 vcpu->arch.smbase = 0x30000;
9511 vcpu->arch.msr_misc_features_enables = 0;
9513 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9516 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9517 vcpu->arch.regs_avail = ~0;
9518 vcpu->arch.regs_dirty = ~0;
9520 vcpu->arch.ia32_xss = 0;
9522 kvm_x86_ops.vcpu_reset(vcpu, init_event);
9525 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9527 struct kvm_segment cs;
9529 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9530 cs.selector = vector << 8;
9531 cs.base = vector << 12;
9532 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9533 kvm_rip_write(vcpu, 0);
9536 int kvm_arch_hardware_enable(void)
9539 struct kvm_vcpu *vcpu;
9544 bool stable, backwards_tsc = false;
9546 kvm_shared_msr_cpu_online();
9547 ret = kvm_x86_ops.hardware_enable();
9551 local_tsc = rdtsc();
9552 stable = !kvm_check_tsc_unstable();
9553 list_for_each_entry(kvm, &vm_list, vm_list) {
9554 kvm_for_each_vcpu(i, vcpu, kvm) {
9555 if (!stable && vcpu->cpu == smp_processor_id())
9556 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9557 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9558 backwards_tsc = true;
9559 if (vcpu->arch.last_host_tsc > max_tsc)
9560 max_tsc = vcpu->arch.last_host_tsc;
9566 * Sometimes, even reliable TSCs go backwards. This happens on
9567 * platforms that reset TSC during suspend or hibernate actions, but
9568 * maintain synchronization. We must compensate. Fortunately, we can
9569 * detect that condition here, which happens early in CPU bringup,
9570 * before any KVM threads can be running. Unfortunately, we can't
9571 * bring the TSCs fully up to date with real time, as we aren't yet far
9572 * enough into CPU bringup that we know how much real time has actually
9573 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9574 * variables that haven't been updated yet.
9576 * So we simply find the maximum observed TSC above, then record the
9577 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9578 * the adjustment will be applied. Note that we accumulate
9579 * adjustments, in case multiple suspend cycles happen before some VCPU
9580 * gets a chance to run again. In the event that no KVM threads get a
9581 * chance to run, we will miss the entire elapsed period, as we'll have
9582 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9583 * loose cycle time. This isn't too big a deal, since the loss will be
9584 * uniform across all VCPUs (not to mention the scenario is extremely
9585 * unlikely). It is possible that a second hibernate recovery happens
9586 * much faster than a first, causing the observed TSC here to be
9587 * smaller; this would require additional padding adjustment, which is
9588 * why we set last_host_tsc to the local tsc observed here.
9590 * N.B. - this code below runs only on platforms with reliable TSC,
9591 * as that is the only way backwards_tsc is set above. Also note
9592 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9593 * have the same delta_cyc adjustment applied if backwards_tsc
9594 * is detected. Note further, this adjustment is only done once,
9595 * as we reset last_host_tsc on all VCPUs to stop this from being
9596 * called multiple times (one for each physical CPU bringup).
9598 * Platforms with unreliable TSCs don't have to deal with this, they
9599 * will be compensated by the logic in vcpu_load, which sets the TSC to
9600 * catchup mode. This will catchup all VCPUs to real time, but cannot
9601 * guarantee that they stay in perfect synchronization.
9603 if (backwards_tsc) {
9604 u64 delta_cyc = max_tsc - local_tsc;
9605 list_for_each_entry(kvm, &vm_list, vm_list) {
9606 kvm->arch.backwards_tsc_observed = true;
9607 kvm_for_each_vcpu(i, vcpu, kvm) {
9608 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9609 vcpu->arch.last_host_tsc = local_tsc;
9610 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9614 * We have to disable TSC offset matching.. if you were
9615 * booting a VM while issuing an S4 host suspend....
9616 * you may have some problem. Solving this issue is
9617 * left as an exercise to the reader.
9619 kvm->arch.last_tsc_nsec = 0;
9620 kvm->arch.last_tsc_write = 0;
9627 void kvm_arch_hardware_disable(void)
9629 kvm_x86_ops.hardware_disable();
9630 drop_user_return_notifiers();
9633 int kvm_arch_hardware_setup(void *opaque)
9635 struct kvm_x86_init_ops *ops = opaque;
9638 rdmsrl_safe(MSR_EFER, &host_efer);
9640 if (boot_cpu_has(X86_FEATURE_XSAVES))
9641 rdmsrl(MSR_IA32_XSS, host_xss);
9643 r = ops->hardware_setup();
9647 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9649 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9652 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
9653 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
9654 #undef __kvm_cpu_cap_has
9656 if (kvm_has_tsc_control) {
9658 * Make sure the user can only configure tsc_khz values that
9659 * fit into a signed integer.
9660 * A min value is not calculated because it will always
9661 * be 1 on all machines.
9663 u64 max = min(0x7fffffffULL,
9664 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9665 kvm_max_guest_tsc_khz = max;
9667 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9670 kvm_init_msr_list();
9674 void kvm_arch_hardware_unsetup(void)
9676 kvm_x86_ops.hardware_unsetup();
9679 int kvm_arch_check_processor_compat(void *opaque)
9681 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
9682 struct kvm_x86_init_ops *ops = opaque;
9684 WARN_ON(!irqs_disabled());
9686 if (__cr4_reserved_bits(cpu_has, c) !=
9687 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
9690 return ops->check_processor_compatibility();
9693 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9695 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9697 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9699 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9701 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9704 struct static_key kvm_no_apic_vcpu __read_mostly;
9705 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9707 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9709 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9711 vcpu->arch.l1tf_flush_l1d = true;
9712 if (pmu->version && unlikely(pmu->event_count)) {
9713 pmu->need_cleanup = true;
9714 kvm_make_request(KVM_REQ_PMU, vcpu);
9716 kvm_x86_ops.sched_in(vcpu, cpu);
9719 void kvm_arch_free_vm(struct kvm *kvm)
9721 kfree(kvm->arch.hyperv.hv_pa_pg);
9726 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9731 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9732 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9733 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9734 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9735 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9736 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9738 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9739 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9740 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9741 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9742 &kvm->arch.irq_sources_bitmap);
9744 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9745 mutex_init(&kvm->arch.apic_map_lock);
9746 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9748 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
9749 pvclock_update_vm_gtod_copy(kvm);
9751 kvm->arch.guest_can_read_msr_platform_info = true;
9753 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9754 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9756 kvm_hv_init_vm(kvm);
9757 kvm_page_track_init(kvm);
9758 kvm_mmu_init_vm(kvm);
9760 return kvm_x86_ops.vm_init(kvm);
9763 int kvm_arch_post_init_vm(struct kvm *kvm)
9765 return kvm_mmu_post_init_vm(kvm);
9768 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9771 kvm_mmu_unload(vcpu);
9775 static void kvm_free_vcpus(struct kvm *kvm)
9778 struct kvm_vcpu *vcpu;
9781 * Unpin any mmu pages first.
9783 kvm_for_each_vcpu(i, vcpu, kvm) {
9784 kvm_clear_async_pf_completion_queue(vcpu);
9785 kvm_unload_vcpu_mmu(vcpu);
9787 kvm_for_each_vcpu(i, vcpu, kvm)
9788 kvm_vcpu_destroy(vcpu);
9790 mutex_lock(&kvm->lock);
9791 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9792 kvm->vcpus[i] = NULL;
9794 atomic_set(&kvm->online_vcpus, 0);
9795 mutex_unlock(&kvm->lock);
9798 void kvm_arch_sync_events(struct kvm *kvm)
9800 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9801 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9805 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9808 unsigned long hva, uninitialized_var(old_npages);
9809 struct kvm_memslots *slots = kvm_memslots(kvm);
9810 struct kvm_memory_slot *slot;
9812 /* Called with kvm->slots_lock held. */
9813 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9816 slot = id_to_memslot(slots, id);
9818 if (slot && slot->npages)
9822 * MAP_SHARED to prevent internal slot pages from being moved
9825 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9826 MAP_SHARED | MAP_ANONYMOUS, 0);
9827 if (IS_ERR((void *)hva))
9828 return PTR_ERR((void *)hva);
9830 if (!slot || !slot->npages)
9834 * Stuff a non-canonical value to catch use-after-delete. This
9835 * ends up being 0 on 32-bit KVM, but there's no better
9838 hva = (unsigned long)(0xdeadull << 48);
9839 old_npages = slot->npages;
9842 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9843 struct kvm_userspace_memory_region m;
9845 m.slot = id | (i << 16);
9847 m.guest_phys_addr = gpa;
9848 m.userspace_addr = hva;
9849 m.memory_size = size;
9850 r = __kvm_set_memory_region(kvm, &m);
9856 vm_munmap(hva, old_npages * PAGE_SIZE);
9860 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9862 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9864 kvm_mmu_pre_destroy_vm(kvm);
9867 void kvm_arch_destroy_vm(struct kvm *kvm)
9869 if (current->mm == kvm->mm) {
9871 * Free memory regions allocated on behalf of userspace,
9872 * unless the the memory map has changed due to process exit
9875 mutex_lock(&kvm->slots_lock);
9876 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
9878 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
9880 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9881 mutex_unlock(&kvm->slots_lock);
9883 if (kvm_x86_ops.vm_destroy)
9884 kvm_x86_ops.vm_destroy(kvm);
9885 kvm_pic_destroy(kvm);
9886 kvm_ioapic_destroy(kvm);
9887 kvm_free_vcpus(kvm);
9888 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9889 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9890 kvm_mmu_uninit_vm(kvm);
9891 kvm_page_track_cleanup(kvm);
9892 kvm_hv_destroy_vm(kvm);
9895 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
9899 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9900 kvfree(slot->arch.rmap[i]);
9901 slot->arch.rmap[i] = NULL;
9906 kvfree(slot->arch.lpage_info[i - 1]);
9907 slot->arch.lpage_info[i - 1] = NULL;
9910 kvm_page_track_free_memslot(slot);
9913 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
9914 unsigned long npages)
9919 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
9920 * old arrays will be freed by __kvm_set_memory_region() if installing
9921 * the new memslot is successful.
9923 memset(&slot->arch, 0, sizeof(slot->arch));
9925 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9926 struct kvm_lpage_info *linfo;
9931 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9932 slot->base_gfn, level) + 1;
9934 slot->arch.rmap[i] =
9935 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9936 GFP_KERNEL_ACCOUNT);
9937 if (!slot->arch.rmap[i])
9942 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9946 slot->arch.lpage_info[i - 1] = linfo;
9948 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9949 linfo[0].disallow_lpage = 1;
9950 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9951 linfo[lpages - 1].disallow_lpage = 1;
9952 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9954 * If the gfn and userspace address are not aligned wrt each
9955 * other, disable large page support for this slot.
9957 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
9960 for (j = 0; j < lpages; ++j)
9961 linfo[j].disallow_lpage = 1;
9965 if (kvm_page_track_create_memslot(slot, npages))
9971 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9972 kvfree(slot->arch.rmap[i]);
9973 slot->arch.rmap[i] = NULL;
9977 kvfree(slot->arch.lpage_info[i - 1]);
9978 slot->arch.lpage_info[i - 1] = NULL;
9983 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9985 struct kvm_vcpu *vcpu;
9989 * memslots->generation has been incremented.
9990 * mmio generation may have reached its maximum value.
9992 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9994 /* Force re-initialization of steal_time cache */
9995 kvm_for_each_vcpu(i, vcpu, kvm)
9996 kvm_vcpu_kick(vcpu);
9999 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10000 struct kvm_memory_slot *memslot,
10001 const struct kvm_userspace_memory_region *mem,
10002 enum kvm_mr_change change)
10004 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10005 return kvm_alloc_memslot_metadata(memslot,
10006 mem->memory_size >> PAGE_SHIFT);
10010 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10011 struct kvm_memory_slot *new)
10013 /* Still write protect RO slot */
10014 if (new->flags & KVM_MEM_READONLY) {
10015 kvm_mmu_slot_remove_write_access(kvm, new, PT_PAGE_TABLE_LEVEL);
10020 * Call kvm_x86_ops dirty logging hooks when they are valid.
10022 * kvm_x86_ops.slot_disable_log_dirty is called when:
10024 * - KVM_MR_CREATE with dirty logging is disabled
10025 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
10027 * The reason is, in case of PML, we need to set D-bit for any slots
10028 * with dirty logging disabled in order to eliminate unnecessary GPA
10029 * logging in PML buffer (and potential PML buffer full VMEXIT). This
10030 * guarantees leaving PML enabled during guest's lifetime won't have
10031 * any additional overhead from PML when guest is running with dirty
10032 * logging disabled for memory slots.
10034 * kvm_x86_ops.slot_enable_log_dirty is called when switching new slot
10035 * to dirty logging mode.
10037 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
10039 * In case of write protect:
10041 * Write protect all pages for dirty logging.
10043 * All the sptes including the large sptes which point to this
10044 * slot are set to readonly. We can not create any new large
10045 * spte on this slot until the end of the logging.
10047 * See the comments in fast_page_fault().
10049 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10050 if (kvm_x86_ops.slot_enable_log_dirty) {
10051 kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10054 kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10055 PT_DIRECTORY_LEVEL : PT_PAGE_TABLE_LEVEL;
10058 * If we're with initial-all-set, we don't need
10059 * to write protect any small page because
10060 * they're reported as dirty already. However
10061 * we still need to write-protect huge pages
10062 * so that the page split can happen lazily on
10063 * the first write to the huge page.
10065 kvm_mmu_slot_remove_write_access(kvm, new, level);
10068 if (kvm_x86_ops.slot_disable_log_dirty)
10069 kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10073 void kvm_arch_commit_memory_region(struct kvm *kvm,
10074 const struct kvm_userspace_memory_region *mem,
10075 struct kvm_memory_slot *old,
10076 const struct kvm_memory_slot *new,
10077 enum kvm_mr_change change)
10079 if (!kvm->arch.n_requested_mmu_pages)
10080 kvm_mmu_change_mmu_pages(kvm,
10081 kvm_mmu_calculate_default_mmu_pages(kvm));
10084 * Dirty logging tracks sptes in 4k granularity, meaning that large
10085 * sptes have to be split. If live migration is successful, the guest
10086 * in the source machine will be destroyed and large sptes will be
10087 * created in the destination. However, if the guest continues to run
10088 * in the source machine (for example if live migration fails), small
10089 * sptes will remain around and cause bad performance.
10091 * Scan sptes if dirty logging has been stopped, dropping those
10092 * which can be collapsed into a single large-page spte. Later
10093 * page faults will create the large-page sptes.
10095 * There is no need to do this in any of the following cases:
10096 * CREATE: No dirty mappings will already exist.
10097 * MOVE/DELETE: The old mappings will already have been cleaned up by
10098 * kvm_arch_flush_shadow_memslot()
10100 if (change == KVM_MR_FLAGS_ONLY &&
10101 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10102 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10103 kvm_mmu_zap_collapsible_sptes(kvm, new);
10106 * Set up write protection and/or dirty logging for the new slot.
10108 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
10109 * been zapped so no dirty logging staff is needed for old slot. For
10110 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
10111 * new and it's also covered when dealing with the new slot.
10113 * FIXME: const-ify all uses of struct kvm_memory_slot.
10115 if (change != KVM_MR_DELETE)
10116 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
10118 /* Free the arrays associated with the old memslot. */
10119 if (change == KVM_MR_MOVE)
10120 kvm_arch_free_memslot(kvm, old);
10123 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10125 kvm_mmu_zap_all(kvm);
10128 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10129 struct kvm_memory_slot *slot)
10131 kvm_page_track_flush_slot(kvm, slot);
10134 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10136 return (is_guest_mode(vcpu) &&
10137 kvm_x86_ops.guest_apic_has_interrupt &&
10138 kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10141 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10143 if (!list_empty_careful(&vcpu->async_pf.done))
10146 if (kvm_apic_has_events(vcpu))
10149 if (vcpu->arch.pv.pv_unhalted)
10152 if (vcpu->arch.exception.pending)
10155 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10156 (vcpu->arch.nmi_pending &&
10157 kvm_x86_ops.nmi_allowed(vcpu)))
10160 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10161 (vcpu->arch.smi_pending && !is_smm(vcpu)))
10164 if (kvm_arch_interrupt_allowed(vcpu) &&
10165 (kvm_cpu_has_interrupt(vcpu) ||
10166 kvm_guest_apic_has_interrupt(vcpu)))
10169 if (kvm_hv_has_stimer_pending(vcpu))
10175 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10177 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10180 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10182 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10185 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10186 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10187 kvm_test_request(KVM_REQ_EVENT, vcpu))
10190 if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10196 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10198 return vcpu->arch.preempted_in_kernel;
10201 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10203 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10206 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10208 return kvm_x86_ops.interrupt_allowed(vcpu);
10211 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10213 if (is_64_bit_mode(vcpu))
10214 return kvm_rip_read(vcpu);
10215 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10216 kvm_rip_read(vcpu));
10218 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10220 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10222 return kvm_get_linear_rip(vcpu) == linear_rip;
10224 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10226 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10228 unsigned long rflags;
10230 rflags = kvm_x86_ops.get_rflags(vcpu);
10231 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10232 rflags &= ~X86_EFLAGS_TF;
10235 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10237 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10239 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10240 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10241 rflags |= X86_EFLAGS_TF;
10242 kvm_x86_ops.set_rflags(vcpu, rflags);
10245 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10247 __kvm_set_rflags(vcpu, rflags);
10248 kvm_make_request(KVM_REQ_EVENT, vcpu);
10250 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10252 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10256 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10260 r = kvm_mmu_reload(vcpu);
10264 if (!vcpu->arch.mmu->direct_map &&
10265 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10268 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10271 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10273 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10276 static inline u32 kvm_async_pf_next_probe(u32 key)
10278 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
10281 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10283 u32 key = kvm_async_pf_hash_fn(gfn);
10285 while (vcpu->arch.apf.gfns[key] != ~0)
10286 key = kvm_async_pf_next_probe(key);
10288 vcpu->arch.apf.gfns[key] = gfn;
10291 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10294 u32 key = kvm_async_pf_hash_fn(gfn);
10296 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
10297 (vcpu->arch.apf.gfns[key] != gfn &&
10298 vcpu->arch.apf.gfns[key] != ~0); i++)
10299 key = kvm_async_pf_next_probe(key);
10304 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10306 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10309 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10313 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10315 vcpu->arch.apf.gfns[i] = ~0;
10317 j = kvm_async_pf_next_probe(j);
10318 if (vcpu->arch.apf.gfns[j] == ~0)
10320 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10322 * k lies cyclically in ]i,j]
10324 * |....j i.k.| or |.k..j i...|
10326 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10327 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10332 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10335 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10339 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10342 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10346 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10348 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10351 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10352 (vcpu->arch.apf.send_user_only &&
10353 kvm_x86_ops.get_cpl(vcpu) == 0))
10359 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10361 if (unlikely(!lapic_in_kernel(vcpu) ||
10362 kvm_event_needs_reinjection(vcpu) ||
10363 vcpu->arch.exception.pending))
10366 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10370 * If interrupts are off we cannot even use an artificial
10373 return kvm_x86_ops.interrupt_allowed(vcpu);
10376 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10377 struct kvm_async_pf *work)
10379 struct x86_exception fault;
10381 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10382 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10384 if (kvm_can_deliver_async_pf(vcpu) &&
10385 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10386 fault.vector = PF_VECTOR;
10387 fault.error_code_valid = true;
10388 fault.error_code = 0;
10389 fault.nested_page_fault = false;
10390 fault.address = work->arch.token;
10391 fault.async_page_fault = true;
10392 kvm_inject_page_fault(vcpu, &fault);
10395 * It is not possible to deliver a paravirtualized asynchronous
10396 * page fault, but putting the guest in an artificial halt state
10397 * can be beneficial nevertheless: if an interrupt arrives, we
10398 * can deliver it timely and perhaps the guest will schedule
10399 * another process. When the instruction that triggered a page
10400 * fault is retried, hopefully the page will be ready in the host.
10402 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10406 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10407 struct kvm_async_pf *work)
10409 struct x86_exception fault;
10412 if (work->wakeup_all)
10413 work->arch.token = ~0; /* broadcast wakeup */
10415 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10416 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10418 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10419 !apf_get_user(vcpu, &val)) {
10420 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10421 vcpu->arch.exception.pending &&
10422 vcpu->arch.exception.nr == PF_VECTOR &&
10423 !apf_put_user(vcpu, 0)) {
10424 vcpu->arch.exception.injected = false;
10425 vcpu->arch.exception.pending = false;
10426 vcpu->arch.exception.nr = 0;
10427 vcpu->arch.exception.has_error_code = false;
10428 vcpu->arch.exception.error_code = 0;
10429 vcpu->arch.exception.has_payload = false;
10430 vcpu->arch.exception.payload = 0;
10431 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10432 fault.vector = PF_VECTOR;
10433 fault.error_code_valid = true;
10434 fault.error_code = 0;
10435 fault.nested_page_fault = false;
10436 fault.address = work->arch.token;
10437 fault.async_page_fault = true;
10438 kvm_inject_page_fault(vcpu, &fault);
10441 vcpu->arch.apf.halted = false;
10442 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10445 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10447 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10450 return kvm_can_do_async_pf(vcpu);
10453 void kvm_arch_start_assignment(struct kvm *kvm)
10455 atomic_inc(&kvm->arch.assigned_device_count);
10457 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10459 void kvm_arch_end_assignment(struct kvm *kvm)
10461 atomic_dec(&kvm->arch.assigned_device_count);
10463 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10465 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10467 return atomic_read(&kvm->arch.assigned_device_count);
10469 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10471 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10473 atomic_inc(&kvm->arch.noncoherent_dma_count);
10475 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10477 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10479 atomic_dec(&kvm->arch.noncoherent_dma_count);
10481 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10483 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10485 return atomic_read(&kvm->arch.noncoherent_dma_count);
10487 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10489 bool kvm_arch_has_irq_bypass(void)
10494 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10495 struct irq_bypass_producer *prod)
10497 struct kvm_kernel_irqfd *irqfd =
10498 container_of(cons, struct kvm_kernel_irqfd, consumer);
10500 irqfd->producer = prod;
10502 return kvm_x86_ops.update_pi_irte(irqfd->kvm,
10503 prod->irq, irqfd->gsi, 1);
10506 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10507 struct irq_bypass_producer *prod)
10510 struct kvm_kernel_irqfd *irqfd =
10511 container_of(cons, struct kvm_kernel_irqfd, consumer);
10513 WARN_ON(irqfd->producer != prod);
10514 irqfd->producer = NULL;
10517 * When producer of consumer is unregistered, we change back to
10518 * remapped mode, so we can re-use the current implementation
10519 * when the irq is masked/disabled or the consumer side (KVM
10520 * int this case doesn't want to receive the interrupts.
10522 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10524 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10525 " fails: %d\n", irqfd->consumer.token, ret);
10528 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10529 uint32_t guest_irq, bool set)
10531 return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
10534 bool kvm_vector_hashing_enabled(void)
10536 return vector_hashing;
10539 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10541 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10543 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10545 u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu)
10547 uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD;
10549 /* The STIBP bit doesn't fault even if it's not advertised */
10550 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
10551 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
10552 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10553 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
10554 !boot_cpu_has(X86_FEATURE_AMD_IBRS))
10555 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10557 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) &&
10558 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
10559 bits &= ~SPEC_CTRL_SSBD;
10560 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
10561 !boot_cpu_has(X86_FEATURE_AMD_SSBD))
10562 bits &= ~SPEC_CTRL_SSBD;
10566 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits);
10568 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10569 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10570 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10571 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10572 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10573 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10574 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10575 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10576 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10577 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10578 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10579 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10580 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10581 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10582 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10583 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10584 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10585 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10586 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10587 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10588 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
10589 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);