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 if (static_cpu_has(X86_FEATURE_PKU) &&
842 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
843 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
844 vcpu->arch.pkru != vcpu->arch.host_pkru)
845 __write_pkru(vcpu->arch.pkru);
847 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
849 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
851 if (static_cpu_has(X86_FEATURE_PKU) &&
852 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
853 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
854 vcpu->arch.pkru = rdpkru();
855 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
856 __write_pkru(vcpu->arch.host_pkru);
859 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
861 if (vcpu->arch.xcr0 != host_xcr0)
862 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
864 if (vcpu->arch.xsaves_enabled &&
865 vcpu->arch.ia32_xss != host_xss)
866 wrmsrl(MSR_IA32_XSS, host_xss);
870 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
872 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
875 u64 old_xcr0 = vcpu->arch.xcr0;
878 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
879 if (index != XCR_XFEATURE_ENABLED_MASK)
881 if (!(xcr0 & XFEATURE_MASK_FP))
883 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
887 * Do not allow the guest to set bits that we do not support
888 * saving. However, xcr0 bit 0 is always set, even if the
889 * emulated CPU does not support XSAVE (see fx_init).
891 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
892 if (xcr0 & ~valid_bits)
895 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
896 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
899 if (xcr0 & XFEATURE_MASK_AVX512) {
900 if (!(xcr0 & XFEATURE_MASK_YMM))
902 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
905 vcpu->arch.xcr0 = xcr0;
907 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
908 kvm_update_cpuid(vcpu);
912 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
914 if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
915 __kvm_set_xcr(vcpu, index, xcr)) {
916 kvm_inject_gp(vcpu, 0);
921 EXPORT_SYMBOL_GPL(kvm_set_xcr);
923 #define __cr4_reserved_bits(__cpu_has, __c) \
925 u64 __reserved_bits = CR4_RESERVED_BITS; \
927 if (!__cpu_has(__c, X86_FEATURE_XSAVE)) \
928 __reserved_bits |= X86_CR4_OSXSAVE; \
929 if (!__cpu_has(__c, X86_FEATURE_SMEP)) \
930 __reserved_bits |= X86_CR4_SMEP; \
931 if (!__cpu_has(__c, X86_FEATURE_SMAP)) \
932 __reserved_bits |= X86_CR4_SMAP; \
933 if (!__cpu_has(__c, X86_FEATURE_FSGSBASE)) \
934 __reserved_bits |= X86_CR4_FSGSBASE; \
935 if (!__cpu_has(__c, X86_FEATURE_PKU)) \
936 __reserved_bits |= X86_CR4_PKE; \
937 if (!__cpu_has(__c, X86_FEATURE_LA57)) \
938 __reserved_bits |= X86_CR4_LA57; \
939 if (!__cpu_has(__c, X86_FEATURE_UMIP)) \
940 __reserved_bits |= X86_CR4_UMIP; \
944 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
946 if (cr4 & cr4_reserved_bits)
949 if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu))
955 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
957 unsigned long old_cr4 = kvm_read_cr4(vcpu);
958 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
959 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
961 if (kvm_valid_cr4(vcpu, cr4))
964 if (is_long_mode(vcpu)) {
965 if (!(cr4 & X86_CR4_PAE))
967 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
968 && ((cr4 ^ old_cr4) & pdptr_bits)
969 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
973 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
974 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
977 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
978 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
982 if (kvm_x86_ops.set_cr4(vcpu, cr4))
985 if (((cr4 ^ old_cr4) & pdptr_bits) ||
986 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
987 kvm_mmu_reset_context(vcpu);
989 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
990 kvm_update_cpuid(vcpu);
994 EXPORT_SYMBOL_GPL(kvm_set_cr4);
996 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
998 bool skip_tlb_flush = false;
1000 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1003 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1004 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1008 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1009 if (!skip_tlb_flush) {
1010 kvm_mmu_sync_roots(vcpu);
1011 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1016 if (is_long_mode(vcpu) &&
1017 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
1019 else if (is_pae_paging(vcpu) &&
1020 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1023 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
1024 vcpu->arch.cr3 = cr3;
1025 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1029 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1031 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1033 if (cr8 & CR8_RESERVED_BITS)
1035 if (lapic_in_kernel(vcpu))
1036 kvm_lapic_set_tpr(vcpu, cr8);
1038 vcpu->arch.cr8 = cr8;
1041 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1043 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1045 if (lapic_in_kernel(vcpu))
1046 return kvm_lapic_get_cr8(vcpu);
1048 return vcpu->arch.cr8;
1050 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1052 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1056 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1057 for (i = 0; i < KVM_NR_DB_REGS; i++)
1058 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1059 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1063 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1067 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1068 dr7 = vcpu->arch.guest_debug_dr7;
1070 dr7 = vcpu->arch.dr7;
1071 kvm_x86_ops.set_dr7(vcpu, dr7);
1072 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1073 if (dr7 & DR7_BP_EN_MASK)
1074 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1077 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1079 u64 fixed = DR6_FIXED_1;
1081 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1086 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1088 size_t size = ARRAY_SIZE(vcpu->arch.db);
1092 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1093 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1094 vcpu->arch.eff_db[dr] = val;
1099 if (val & 0xffffffff00000000ULL)
1100 return -1; /* #GP */
1101 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1106 if (!kvm_dr7_valid(val))
1107 return -1; /* #GP */
1108 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1109 kvm_update_dr7(vcpu);
1116 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1118 if (__kvm_set_dr(vcpu, dr, val)) {
1119 kvm_inject_gp(vcpu, 0);
1124 EXPORT_SYMBOL_GPL(kvm_set_dr);
1126 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1128 size_t size = ARRAY_SIZE(vcpu->arch.db);
1132 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1137 *val = vcpu->arch.dr6;
1142 *val = vcpu->arch.dr7;
1147 EXPORT_SYMBOL_GPL(kvm_get_dr);
1149 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1151 u32 ecx = kvm_rcx_read(vcpu);
1155 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1158 kvm_rax_write(vcpu, (u32)data);
1159 kvm_rdx_write(vcpu, data >> 32);
1162 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1165 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1166 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1168 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1169 * extract the supported MSRs from the related const lists.
1170 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1171 * capabilities of the host cpu. This capabilities test skips MSRs that are
1172 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1173 * may depend on host virtualization features rather than host cpu features.
1176 static const u32 msrs_to_save_all[] = {
1177 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1179 #ifdef CONFIG_X86_64
1180 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1182 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1183 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1185 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1186 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1187 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1188 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1189 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1190 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1191 MSR_IA32_UMWAIT_CONTROL,
1193 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1194 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1195 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1196 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1197 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1198 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1199 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1200 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1201 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1202 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1203 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1204 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1205 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1206 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1207 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1208 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1209 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1210 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1211 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1212 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1213 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1214 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1217 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1218 static unsigned num_msrs_to_save;
1220 static const u32 emulated_msrs_all[] = {
1221 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1222 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1223 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1224 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1225 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1226 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1227 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1229 HV_X64_MSR_VP_INDEX,
1230 HV_X64_MSR_VP_RUNTIME,
1231 HV_X64_MSR_SCONTROL,
1232 HV_X64_MSR_STIMER0_CONFIG,
1233 HV_X64_MSR_VP_ASSIST_PAGE,
1234 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1235 HV_X64_MSR_TSC_EMULATION_STATUS,
1237 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1240 MSR_IA32_TSC_ADJUST,
1241 MSR_IA32_TSCDEADLINE,
1242 MSR_IA32_ARCH_CAPABILITIES,
1243 MSR_IA32_MISC_ENABLE,
1244 MSR_IA32_MCG_STATUS,
1246 MSR_IA32_MCG_EXT_CTL,
1250 MSR_MISC_FEATURES_ENABLES,
1251 MSR_AMD64_VIRT_SPEC_CTRL,
1256 * The following list leaves out MSRs whose values are determined
1257 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1258 * We always support the "true" VMX control MSRs, even if the host
1259 * processor does not, so I am putting these registers here rather
1260 * than in msrs_to_save_all.
1263 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1264 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1265 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1266 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1268 MSR_IA32_VMX_CR0_FIXED0,
1269 MSR_IA32_VMX_CR4_FIXED0,
1270 MSR_IA32_VMX_VMCS_ENUM,
1271 MSR_IA32_VMX_PROCBASED_CTLS2,
1272 MSR_IA32_VMX_EPT_VPID_CAP,
1273 MSR_IA32_VMX_VMFUNC,
1276 MSR_KVM_POLL_CONTROL,
1279 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1280 static unsigned num_emulated_msrs;
1283 * List of msr numbers which are used to expose MSR-based features that
1284 * can be used by a hypervisor to validate requested CPU features.
1286 static const u32 msr_based_features_all[] = {
1288 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1289 MSR_IA32_VMX_PINBASED_CTLS,
1290 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1291 MSR_IA32_VMX_PROCBASED_CTLS,
1292 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1293 MSR_IA32_VMX_EXIT_CTLS,
1294 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1295 MSR_IA32_VMX_ENTRY_CTLS,
1297 MSR_IA32_VMX_CR0_FIXED0,
1298 MSR_IA32_VMX_CR0_FIXED1,
1299 MSR_IA32_VMX_CR4_FIXED0,
1300 MSR_IA32_VMX_CR4_FIXED1,
1301 MSR_IA32_VMX_VMCS_ENUM,
1302 MSR_IA32_VMX_PROCBASED_CTLS2,
1303 MSR_IA32_VMX_EPT_VPID_CAP,
1304 MSR_IA32_VMX_VMFUNC,
1308 MSR_IA32_ARCH_CAPABILITIES,
1311 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1312 static unsigned int num_msr_based_features;
1314 static u64 kvm_get_arch_capabilities(void)
1318 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1319 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1322 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1323 * the nested hypervisor runs with NX huge pages. If it is not,
1324 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1325 * L1 guests, so it need not worry about its own (L2) guests.
1327 data |= ARCH_CAP_PSCHANGE_MC_NO;
1330 * If we're doing cache flushes (either "always" or "cond")
1331 * we will do one whenever the guest does a vmlaunch/vmresume.
1332 * If an outer hypervisor is doing the cache flush for us
1333 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1334 * capability to the guest too, and if EPT is disabled we're not
1335 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1336 * require a nested hypervisor to do a flush of its own.
1338 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1339 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1341 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1342 data |= ARCH_CAP_RDCL_NO;
1343 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1344 data |= ARCH_CAP_SSB_NO;
1345 if (!boot_cpu_has_bug(X86_BUG_MDS))
1346 data |= ARCH_CAP_MDS_NO;
1349 * On TAA affected systems:
1350 * - nothing to do if TSX is disabled on the host.
1351 * - we emulate TSX_CTRL if present on the host.
1352 * This lets the guest use VERW to clear CPU buffers.
1354 if (!boot_cpu_has(X86_FEATURE_RTM))
1355 data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1356 else if (!boot_cpu_has_bug(X86_BUG_TAA))
1357 data |= ARCH_CAP_TAA_NO;
1362 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1364 switch (msr->index) {
1365 case MSR_IA32_ARCH_CAPABILITIES:
1366 msr->data = kvm_get_arch_capabilities();
1368 case MSR_IA32_UCODE_REV:
1369 rdmsrl_safe(msr->index, &msr->data);
1372 if (kvm_x86_ops.get_msr_feature(msr))
1378 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1380 struct kvm_msr_entry msr;
1384 r = kvm_get_msr_feature(&msr);
1393 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1395 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1398 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1401 if (efer & (EFER_LME | EFER_LMA) &&
1402 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1405 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1411 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1413 if (efer & efer_reserved_bits)
1416 return __kvm_valid_efer(vcpu, efer);
1418 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1420 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1422 u64 old_efer = vcpu->arch.efer;
1423 u64 efer = msr_info->data;
1425 if (efer & efer_reserved_bits)
1428 if (!msr_info->host_initiated) {
1429 if (!__kvm_valid_efer(vcpu, efer))
1432 if (is_paging(vcpu) &&
1433 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1438 efer |= vcpu->arch.efer & EFER_LMA;
1440 kvm_x86_ops.set_efer(vcpu, efer);
1442 /* Update reserved bits */
1443 if ((efer ^ old_efer) & EFER_NX)
1444 kvm_mmu_reset_context(vcpu);
1449 void kvm_enable_efer_bits(u64 mask)
1451 efer_reserved_bits &= ~mask;
1453 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1456 * Write @data into the MSR specified by @index. Select MSR specific fault
1457 * checks are bypassed if @host_initiated is %true.
1458 * Returns 0 on success, non-0 otherwise.
1459 * Assumes vcpu_load() was already called.
1461 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1462 bool host_initiated)
1464 struct msr_data msr;
1469 case MSR_KERNEL_GS_BASE:
1472 if (is_noncanonical_address(data, vcpu))
1475 case MSR_IA32_SYSENTER_EIP:
1476 case MSR_IA32_SYSENTER_ESP:
1478 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1479 * non-canonical address is written on Intel but not on
1480 * AMD (which ignores the top 32-bits, because it does
1481 * not implement 64-bit SYSENTER).
1483 * 64-bit code should hence be able to write a non-canonical
1484 * value on AMD. Making the address canonical ensures that
1485 * vmentry does not fail on Intel after writing a non-canonical
1486 * value, and that something deterministic happens if the guest
1487 * invokes 64-bit SYSENTER.
1489 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1494 msr.host_initiated = host_initiated;
1496 return kvm_x86_ops.set_msr(vcpu, &msr);
1500 * Read the MSR specified by @index into @data. Select MSR specific fault
1501 * checks are bypassed if @host_initiated is %true.
1502 * Returns 0 on success, non-0 otherwise.
1503 * Assumes vcpu_load() was already called.
1505 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1506 bool host_initiated)
1508 struct msr_data msr;
1512 msr.host_initiated = host_initiated;
1514 ret = kvm_x86_ops.get_msr(vcpu, &msr);
1520 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1522 return __kvm_get_msr(vcpu, index, data, false);
1524 EXPORT_SYMBOL_GPL(kvm_get_msr);
1526 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1528 return __kvm_set_msr(vcpu, index, data, false);
1530 EXPORT_SYMBOL_GPL(kvm_set_msr);
1532 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1534 u32 ecx = kvm_rcx_read(vcpu);
1537 if (kvm_get_msr(vcpu, ecx, &data)) {
1538 trace_kvm_msr_read_ex(ecx);
1539 kvm_inject_gp(vcpu, 0);
1543 trace_kvm_msr_read(ecx, data);
1545 kvm_rax_write(vcpu, data & -1u);
1546 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1547 return kvm_skip_emulated_instruction(vcpu);
1549 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1551 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1553 u32 ecx = kvm_rcx_read(vcpu);
1554 u64 data = kvm_read_edx_eax(vcpu);
1556 if (kvm_set_msr(vcpu, ecx, data)) {
1557 trace_kvm_msr_write_ex(ecx, data);
1558 kvm_inject_gp(vcpu, 0);
1562 trace_kvm_msr_write(ecx, data);
1563 return kvm_skip_emulated_instruction(vcpu);
1565 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1568 * The fast path for frequent and performance sensitive wrmsr emulation,
1569 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1570 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1571 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1572 * other cases which must be called after interrupts are enabled on the host.
1574 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1576 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1579 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1580 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1581 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1582 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1585 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1586 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1587 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1588 trace_kvm_apic_write(APIC_ICR, (u32)data);
1595 enum exit_fastpath_completion handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1597 u32 msr = kvm_rcx_read(vcpu);
1602 case APIC_BASE_MSR + (APIC_ICR >> 4):
1603 data = kvm_read_edx_eax(vcpu);
1604 ret = handle_fastpath_set_x2apic_icr_irqoff(vcpu, data);
1607 return EXIT_FASTPATH_NONE;
1611 trace_kvm_msr_write(msr, data);
1612 return EXIT_FASTPATH_SKIP_EMUL_INS;
1615 return EXIT_FASTPATH_NONE;
1617 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1620 * Adapt set_msr() to msr_io()'s calling convention
1622 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1624 return __kvm_get_msr(vcpu, index, data, true);
1627 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1629 return __kvm_set_msr(vcpu, index, *data, true);
1632 #ifdef CONFIG_X86_64
1633 struct pvclock_clock {
1643 struct pvclock_gtod_data {
1646 struct pvclock_clock clock; /* extract of a clocksource struct */
1647 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1653 static struct pvclock_gtod_data pvclock_gtod_data;
1655 static void update_pvclock_gtod(struct timekeeper *tk)
1657 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1659 write_seqcount_begin(&vdata->seq);
1661 /* copy pvclock gtod data */
1662 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1663 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1664 vdata->clock.mask = tk->tkr_mono.mask;
1665 vdata->clock.mult = tk->tkr_mono.mult;
1666 vdata->clock.shift = tk->tkr_mono.shift;
1667 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1668 vdata->clock.offset = tk->tkr_mono.base;
1670 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1671 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1672 vdata->raw_clock.mask = tk->tkr_raw.mask;
1673 vdata->raw_clock.mult = tk->tkr_raw.mult;
1674 vdata->raw_clock.shift = tk->tkr_raw.shift;
1675 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1676 vdata->raw_clock.offset = tk->tkr_raw.base;
1678 vdata->wall_time_sec = tk->xtime_sec;
1680 vdata->offs_boot = tk->offs_boot;
1682 write_seqcount_end(&vdata->seq);
1685 static s64 get_kvmclock_base_ns(void)
1687 /* Count up from boot time, but with the frequency of the raw clock. */
1688 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1691 static s64 get_kvmclock_base_ns(void)
1693 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1694 return ktime_get_boottime_ns();
1698 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1700 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1701 kvm_vcpu_kick(vcpu);
1704 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1708 struct pvclock_wall_clock wc;
1714 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1719 ++version; /* first time write, random junk */
1723 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1727 * The guest calculates current wall clock time by adding
1728 * system time (updated by kvm_guest_time_update below) to the
1729 * wall clock specified here. We do the reverse here.
1731 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1733 wc.nsec = do_div(wall_nsec, 1000000000);
1734 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1735 wc.version = version;
1737 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1740 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1743 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1745 do_shl32_div32(dividend, divisor);
1749 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1750 s8 *pshift, u32 *pmultiplier)
1758 scaled64 = scaled_hz;
1759 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1764 tps32 = (uint32_t)tps64;
1765 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1766 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1774 *pmultiplier = div_frac(scaled64, tps32);
1777 #ifdef CONFIG_X86_64
1778 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1781 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1782 static unsigned long max_tsc_khz;
1784 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1786 u64 v = (u64)khz * (1000000 + ppm);
1791 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1795 /* Guest TSC same frequency as host TSC? */
1797 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1801 /* TSC scaling supported? */
1802 if (!kvm_has_tsc_control) {
1803 if (user_tsc_khz > tsc_khz) {
1804 vcpu->arch.tsc_catchup = 1;
1805 vcpu->arch.tsc_always_catchup = 1;
1808 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1813 /* TSC scaling required - calculate ratio */
1814 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1815 user_tsc_khz, tsc_khz);
1817 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1818 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1823 vcpu->arch.tsc_scaling_ratio = ratio;
1827 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1829 u32 thresh_lo, thresh_hi;
1830 int use_scaling = 0;
1832 /* tsc_khz can be zero if TSC calibration fails */
1833 if (user_tsc_khz == 0) {
1834 /* set tsc_scaling_ratio to a safe value */
1835 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1839 /* Compute a scale to convert nanoseconds in TSC cycles */
1840 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1841 &vcpu->arch.virtual_tsc_shift,
1842 &vcpu->arch.virtual_tsc_mult);
1843 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1846 * Compute the variation in TSC rate which is acceptable
1847 * within the range of tolerance and decide if the
1848 * rate being applied is within that bounds of the hardware
1849 * rate. If so, no scaling or compensation need be done.
1851 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1852 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1853 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1854 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1857 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1860 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1862 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1863 vcpu->arch.virtual_tsc_mult,
1864 vcpu->arch.virtual_tsc_shift);
1865 tsc += vcpu->arch.this_tsc_write;
1869 static inline int gtod_is_based_on_tsc(int mode)
1871 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
1874 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1876 #ifdef CONFIG_X86_64
1878 struct kvm_arch *ka = &vcpu->kvm->arch;
1879 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1881 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1882 atomic_read(&vcpu->kvm->online_vcpus));
1885 * Once the masterclock is enabled, always perform request in
1886 * order to update it.
1888 * In order to enable masterclock, the host clocksource must be TSC
1889 * and the vcpus need to have matched TSCs. When that happens,
1890 * perform request to enable masterclock.
1892 if (ka->use_master_clock ||
1893 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1894 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1896 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1897 atomic_read(&vcpu->kvm->online_vcpus),
1898 ka->use_master_clock, gtod->clock.vclock_mode);
1902 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1904 u64 curr_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
1905 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1909 * Multiply tsc by a fixed point number represented by ratio.
1911 * The most significant 64-N bits (mult) of ratio represent the
1912 * integral part of the fixed point number; the remaining N bits
1913 * (frac) represent the fractional part, ie. ratio represents a fixed
1914 * point number (mult + frac * 2^(-N)).
1916 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1918 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1920 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1923 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1926 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1928 if (ratio != kvm_default_tsc_scaling_ratio)
1929 _tsc = __scale_tsc(ratio, tsc);
1933 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1935 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1939 tsc = kvm_scale_tsc(vcpu, rdtsc());
1941 return target_tsc - tsc;
1944 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1946 u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
1948 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1950 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1952 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1954 vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
1957 static inline bool kvm_check_tsc_unstable(void)
1959 #ifdef CONFIG_X86_64
1961 * TSC is marked unstable when we're running on Hyper-V,
1962 * 'TSC page' clocksource is good.
1964 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
1967 return check_tsc_unstable();
1970 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1972 struct kvm *kvm = vcpu->kvm;
1973 u64 offset, ns, elapsed;
1974 unsigned long flags;
1976 bool already_matched;
1977 u64 data = msr->data;
1978 bool synchronizing = false;
1980 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1981 offset = kvm_compute_tsc_offset(vcpu, data);
1982 ns = get_kvmclock_base_ns();
1983 elapsed = ns - kvm->arch.last_tsc_nsec;
1985 if (vcpu->arch.virtual_tsc_khz) {
1986 if (data == 0 && msr->host_initiated) {
1988 * detection of vcpu initialization -- need to sync
1989 * with other vCPUs. This particularly helps to keep
1990 * kvm_clock stable after CPU hotplug
1992 synchronizing = true;
1994 u64 tsc_exp = kvm->arch.last_tsc_write +
1995 nsec_to_cycles(vcpu, elapsed);
1996 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1998 * Special case: TSC write with a small delta (1 second)
1999 * of virtual cycle time against real time is
2000 * interpreted as an attempt to synchronize the CPU.
2002 synchronizing = data < tsc_exp + tsc_hz &&
2003 data + tsc_hz > tsc_exp;
2008 * For a reliable TSC, we can match TSC offsets, and for an unstable
2009 * TSC, we add elapsed time in this computation. We could let the
2010 * compensation code attempt to catch up if we fall behind, but
2011 * it's better to try to match offsets from the beginning.
2013 if (synchronizing &&
2014 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2015 if (!kvm_check_tsc_unstable()) {
2016 offset = kvm->arch.cur_tsc_offset;
2018 u64 delta = nsec_to_cycles(vcpu, elapsed);
2020 offset = kvm_compute_tsc_offset(vcpu, data);
2023 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2026 * We split periods of matched TSC writes into generations.
2027 * For each generation, we track the original measured
2028 * nanosecond time, offset, and write, so if TSCs are in
2029 * sync, we can match exact offset, and if not, we can match
2030 * exact software computation in compute_guest_tsc()
2032 * These values are tracked in kvm->arch.cur_xxx variables.
2034 kvm->arch.cur_tsc_generation++;
2035 kvm->arch.cur_tsc_nsec = ns;
2036 kvm->arch.cur_tsc_write = data;
2037 kvm->arch.cur_tsc_offset = offset;
2042 * We also track th most recent recorded KHZ, write and time to
2043 * allow the matching interval to be extended at each write.
2045 kvm->arch.last_tsc_nsec = ns;
2046 kvm->arch.last_tsc_write = data;
2047 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2049 vcpu->arch.last_guest_tsc = data;
2051 /* Keep track of which generation this VCPU has synchronized to */
2052 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2053 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2054 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2056 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
2057 update_ia32_tsc_adjust_msr(vcpu, offset);
2059 kvm_vcpu_write_tsc_offset(vcpu, offset);
2060 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2062 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2064 kvm->arch.nr_vcpus_matched_tsc = 0;
2065 } else if (!already_matched) {
2066 kvm->arch.nr_vcpus_matched_tsc++;
2069 kvm_track_tsc_matching(vcpu);
2070 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2073 EXPORT_SYMBOL_GPL(kvm_write_tsc);
2075 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2078 u64 tsc_offset = kvm_x86_ops.read_l1_tsc_offset(vcpu);
2079 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2082 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2084 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2085 WARN_ON(adjustment < 0);
2086 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2087 adjust_tsc_offset_guest(vcpu, adjustment);
2090 #ifdef CONFIG_X86_64
2092 static u64 read_tsc(void)
2094 u64 ret = (u64)rdtsc_ordered();
2095 u64 last = pvclock_gtod_data.clock.cycle_last;
2097 if (likely(ret >= last))
2101 * GCC likes to generate cmov here, but this branch is extremely
2102 * predictable (it's just a function of time and the likely is
2103 * very likely) and there's a data dependence, so force GCC
2104 * to generate a branch instead. I don't barrier() because
2105 * we don't actually need a barrier, and if this function
2106 * ever gets inlined it will generate worse code.
2112 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2118 switch (clock->vclock_mode) {
2119 case VDSO_CLOCKMODE_HVCLOCK:
2120 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2122 if (tsc_pg_val != U64_MAX) {
2123 /* TSC page valid */
2124 *mode = VDSO_CLOCKMODE_HVCLOCK;
2125 v = (tsc_pg_val - clock->cycle_last) &
2128 /* TSC page invalid */
2129 *mode = VDSO_CLOCKMODE_NONE;
2132 case VDSO_CLOCKMODE_TSC:
2133 *mode = VDSO_CLOCKMODE_TSC;
2134 *tsc_timestamp = read_tsc();
2135 v = (*tsc_timestamp - clock->cycle_last) &
2139 *mode = VDSO_CLOCKMODE_NONE;
2142 if (*mode == VDSO_CLOCKMODE_NONE)
2143 *tsc_timestamp = v = 0;
2145 return v * clock->mult;
2148 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2150 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2156 seq = read_seqcount_begin(>od->seq);
2157 ns = gtod->raw_clock.base_cycles;
2158 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2159 ns >>= gtod->raw_clock.shift;
2160 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2161 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2167 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2169 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2175 seq = read_seqcount_begin(>od->seq);
2176 ts->tv_sec = gtod->wall_time_sec;
2177 ns = gtod->clock.base_cycles;
2178 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2179 ns >>= gtod->clock.shift;
2180 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2182 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2188 /* returns true if host is using TSC based clocksource */
2189 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2191 /* checked again under seqlock below */
2192 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2195 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2199 /* returns true if host is using TSC based clocksource */
2200 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2203 /* checked again under seqlock below */
2204 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2207 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2213 * Assuming a stable TSC across physical CPUS, and a stable TSC
2214 * across virtual CPUs, the following condition is possible.
2215 * Each numbered line represents an event visible to both
2216 * CPUs at the next numbered event.
2218 * "timespecX" represents host monotonic time. "tscX" represents
2221 * VCPU0 on CPU0 | VCPU1 on CPU1
2223 * 1. read timespec0,tsc0
2224 * 2. | timespec1 = timespec0 + N
2226 * 3. transition to guest | transition to guest
2227 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2228 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2229 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2231 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2234 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2236 * - 0 < N - M => M < N
2238 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2239 * always the case (the difference between two distinct xtime instances
2240 * might be smaller then the difference between corresponding TSC reads,
2241 * when updating guest vcpus pvclock areas).
2243 * To avoid that problem, do not allow visibility of distinct
2244 * system_timestamp/tsc_timestamp values simultaneously: use a master
2245 * copy of host monotonic time values. Update that master copy
2248 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2252 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2254 #ifdef CONFIG_X86_64
2255 struct kvm_arch *ka = &kvm->arch;
2257 bool host_tsc_clocksource, vcpus_matched;
2259 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2260 atomic_read(&kvm->online_vcpus));
2263 * If the host uses TSC clock, then passthrough TSC as stable
2266 host_tsc_clocksource = kvm_get_time_and_clockread(
2267 &ka->master_kernel_ns,
2268 &ka->master_cycle_now);
2270 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2271 && !ka->backwards_tsc_observed
2272 && !ka->boot_vcpu_runs_old_kvmclock;
2274 if (ka->use_master_clock)
2275 atomic_set(&kvm_guest_has_master_clock, 1);
2277 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2278 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2283 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2285 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2288 static void kvm_gen_update_masterclock(struct kvm *kvm)
2290 #ifdef CONFIG_X86_64
2292 struct kvm_vcpu *vcpu;
2293 struct kvm_arch *ka = &kvm->arch;
2295 spin_lock(&ka->pvclock_gtod_sync_lock);
2296 kvm_make_mclock_inprogress_request(kvm);
2297 /* no guest entries from this point */
2298 pvclock_update_vm_gtod_copy(kvm);
2300 kvm_for_each_vcpu(i, vcpu, kvm)
2301 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2303 /* guest entries allowed */
2304 kvm_for_each_vcpu(i, vcpu, kvm)
2305 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2307 spin_unlock(&ka->pvclock_gtod_sync_lock);
2311 u64 get_kvmclock_ns(struct kvm *kvm)
2313 struct kvm_arch *ka = &kvm->arch;
2314 struct pvclock_vcpu_time_info hv_clock;
2317 spin_lock(&ka->pvclock_gtod_sync_lock);
2318 if (!ka->use_master_clock) {
2319 spin_unlock(&ka->pvclock_gtod_sync_lock);
2320 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2323 hv_clock.tsc_timestamp = ka->master_cycle_now;
2324 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2325 spin_unlock(&ka->pvclock_gtod_sync_lock);
2327 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2330 if (__this_cpu_read(cpu_tsc_khz)) {
2331 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2332 &hv_clock.tsc_shift,
2333 &hv_clock.tsc_to_system_mul);
2334 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2336 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2343 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2345 struct kvm_vcpu_arch *vcpu = &v->arch;
2346 struct pvclock_vcpu_time_info guest_hv_clock;
2348 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2349 &guest_hv_clock, sizeof(guest_hv_clock))))
2352 /* This VCPU is paused, but it's legal for a guest to read another
2353 * VCPU's kvmclock, so we really have to follow the specification where
2354 * it says that version is odd if data is being modified, and even after
2357 * Version field updates must be kept separate. This is because
2358 * kvm_write_guest_cached might use a "rep movs" instruction, and
2359 * writes within a string instruction are weakly ordered. So there
2360 * are three writes overall.
2362 * As a small optimization, only write the version field in the first
2363 * and third write. The vcpu->pv_time cache is still valid, because the
2364 * version field is the first in the struct.
2366 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2368 if (guest_hv_clock.version & 1)
2369 ++guest_hv_clock.version; /* first time write, random junk */
2371 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2372 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2374 sizeof(vcpu->hv_clock.version));
2378 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2379 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2381 if (vcpu->pvclock_set_guest_stopped_request) {
2382 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2383 vcpu->pvclock_set_guest_stopped_request = false;
2386 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2388 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2390 sizeof(vcpu->hv_clock));
2394 vcpu->hv_clock.version++;
2395 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2397 sizeof(vcpu->hv_clock.version));
2400 static int kvm_guest_time_update(struct kvm_vcpu *v)
2402 unsigned long flags, tgt_tsc_khz;
2403 struct kvm_vcpu_arch *vcpu = &v->arch;
2404 struct kvm_arch *ka = &v->kvm->arch;
2406 u64 tsc_timestamp, host_tsc;
2408 bool use_master_clock;
2414 * If the host uses TSC clock, then passthrough TSC as stable
2417 spin_lock(&ka->pvclock_gtod_sync_lock);
2418 use_master_clock = ka->use_master_clock;
2419 if (use_master_clock) {
2420 host_tsc = ka->master_cycle_now;
2421 kernel_ns = ka->master_kernel_ns;
2423 spin_unlock(&ka->pvclock_gtod_sync_lock);
2425 /* Keep irq disabled to prevent changes to the clock */
2426 local_irq_save(flags);
2427 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2428 if (unlikely(tgt_tsc_khz == 0)) {
2429 local_irq_restore(flags);
2430 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2433 if (!use_master_clock) {
2435 kernel_ns = get_kvmclock_base_ns();
2438 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2441 * We may have to catch up the TSC to match elapsed wall clock
2442 * time for two reasons, even if kvmclock is used.
2443 * 1) CPU could have been running below the maximum TSC rate
2444 * 2) Broken TSC compensation resets the base at each VCPU
2445 * entry to avoid unknown leaps of TSC even when running
2446 * again on the same CPU. This may cause apparent elapsed
2447 * time to disappear, and the guest to stand still or run
2450 if (vcpu->tsc_catchup) {
2451 u64 tsc = compute_guest_tsc(v, kernel_ns);
2452 if (tsc > tsc_timestamp) {
2453 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2454 tsc_timestamp = tsc;
2458 local_irq_restore(flags);
2460 /* With all the info we got, fill in the values */
2462 if (kvm_has_tsc_control)
2463 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2465 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2466 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2467 &vcpu->hv_clock.tsc_shift,
2468 &vcpu->hv_clock.tsc_to_system_mul);
2469 vcpu->hw_tsc_khz = tgt_tsc_khz;
2472 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2473 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2474 vcpu->last_guest_tsc = tsc_timestamp;
2476 /* If the host uses TSC clocksource, then it is stable */
2478 if (use_master_clock)
2479 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2481 vcpu->hv_clock.flags = pvclock_flags;
2483 if (vcpu->pv_time_enabled)
2484 kvm_setup_pvclock_page(v);
2485 if (v == kvm_get_vcpu(v->kvm, 0))
2486 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2491 * kvmclock updates which are isolated to a given vcpu, such as
2492 * vcpu->cpu migration, should not allow system_timestamp from
2493 * the rest of the vcpus to remain static. Otherwise ntp frequency
2494 * correction applies to one vcpu's system_timestamp but not
2497 * So in those cases, request a kvmclock update for all vcpus.
2498 * We need to rate-limit these requests though, as they can
2499 * considerably slow guests that have a large number of vcpus.
2500 * The time for a remote vcpu to update its kvmclock is bound
2501 * by the delay we use to rate-limit the updates.
2504 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2506 static void kvmclock_update_fn(struct work_struct *work)
2509 struct delayed_work *dwork = to_delayed_work(work);
2510 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2511 kvmclock_update_work);
2512 struct kvm *kvm = container_of(ka, struct kvm, arch);
2513 struct kvm_vcpu *vcpu;
2515 kvm_for_each_vcpu(i, vcpu, kvm) {
2516 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2517 kvm_vcpu_kick(vcpu);
2521 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2523 struct kvm *kvm = v->kvm;
2525 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2526 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2527 KVMCLOCK_UPDATE_DELAY);
2530 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2532 static void kvmclock_sync_fn(struct work_struct *work)
2534 struct delayed_work *dwork = to_delayed_work(work);
2535 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2536 kvmclock_sync_work);
2537 struct kvm *kvm = container_of(ka, struct kvm, arch);
2539 if (!kvmclock_periodic_sync)
2542 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2543 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2544 KVMCLOCK_SYNC_PERIOD);
2548 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2550 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2552 /* McStatusWrEn enabled? */
2553 if (guest_cpuid_is_amd_or_hygon(vcpu))
2554 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2559 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2561 u64 mcg_cap = vcpu->arch.mcg_cap;
2562 unsigned bank_num = mcg_cap & 0xff;
2563 u32 msr = msr_info->index;
2564 u64 data = msr_info->data;
2567 case MSR_IA32_MCG_STATUS:
2568 vcpu->arch.mcg_status = data;
2570 case MSR_IA32_MCG_CTL:
2571 if (!(mcg_cap & MCG_CTL_P) &&
2572 (data || !msr_info->host_initiated))
2574 if (data != 0 && data != ~(u64)0)
2576 vcpu->arch.mcg_ctl = data;
2579 if (msr >= MSR_IA32_MC0_CTL &&
2580 msr < MSR_IA32_MCx_CTL(bank_num)) {
2581 u32 offset = array_index_nospec(
2582 msr - MSR_IA32_MC0_CTL,
2583 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2585 /* only 0 or all 1s can be written to IA32_MCi_CTL
2586 * some Linux kernels though clear bit 10 in bank 4 to
2587 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2588 * this to avoid an uncatched #GP in the guest
2590 if ((offset & 0x3) == 0 &&
2591 data != 0 && (data | (1 << 10)) != ~(u64)0)
2595 if (!msr_info->host_initiated &&
2596 (offset & 0x3) == 1 && data != 0) {
2597 if (!can_set_mci_status(vcpu))
2601 vcpu->arch.mce_banks[offset] = data;
2609 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2611 struct kvm *kvm = vcpu->kvm;
2612 int lm = is_long_mode(vcpu);
2613 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2614 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2615 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2616 : kvm->arch.xen_hvm_config.blob_size_32;
2617 u32 page_num = data & ~PAGE_MASK;
2618 u64 page_addr = data & PAGE_MASK;
2623 if (page_num >= blob_size)
2626 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2631 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2640 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2642 gpa_t gpa = data & ~0x3f;
2644 /* Bits 3:5 are reserved, Should be zero */
2648 vcpu->arch.apf.msr_val = data;
2650 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2651 kvm_clear_async_pf_completion_queue(vcpu);
2652 kvm_async_pf_hash_reset(vcpu);
2656 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2660 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2661 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2662 kvm_async_pf_wakeup_all(vcpu);
2666 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2668 vcpu->arch.pv_time_enabled = false;
2669 vcpu->arch.time = 0;
2672 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2674 ++vcpu->stat.tlb_flush;
2675 kvm_x86_ops.tlb_flush(vcpu, invalidate_gpa);
2678 static void record_steal_time(struct kvm_vcpu *vcpu)
2680 struct kvm_host_map map;
2681 struct kvm_steal_time *st;
2683 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2686 /* -EAGAIN is returned in atomic context so we can just return. */
2687 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2688 &map, &vcpu->arch.st.cache, false))
2692 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2695 * Doing a TLB flush here, on the guest's behalf, can avoid
2698 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2699 st->preempted & KVM_VCPU_FLUSH_TLB);
2700 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2701 kvm_vcpu_flush_tlb(vcpu, false);
2703 vcpu->arch.st.preempted = 0;
2705 if (st->version & 1)
2706 st->version += 1; /* first time write, random junk */
2712 st->steal += current->sched_info.run_delay -
2713 vcpu->arch.st.last_steal;
2714 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2720 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2723 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2726 u32 msr = msr_info->index;
2727 u64 data = msr_info->data;
2730 case MSR_AMD64_NB_CFG:
2731 case MSR_IA32_UCODE_WRITE:
2732 case MSR_VM_HSAVE_PA:
2733 case MSR_AMD64_PATCH_LOADER:
2734 case MSR_AMD64_BU_CFG2:
2735 case MSR_AMD64_DC_CFG:
2736 case MSR_F15H_EX_CFG:
2739 case MSR_IA32_UCODE_REV:
2740 if (msr_info->host_initiated)
2741 vcpu->arch.microcode_version = data;
2743 case MSR_IA32_ARCH_CAPABILITIES:
2744 if (!msr_info->host_initiated)
2746 vcpu->arch.arch_capabilities = data;
2749 return set_efer(vcpu, msr_info);
2751 data &= ~(u64)0x40; /* ignore flush filter disable */
2752 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2753 data &= ~(u64)0x8; /* ignore TLB cache disable */
2755 /* Handle McStatusWrEn */
2756 if (data == BIT_ULL(18)) {
2757 vcpu->arch.msr_hwcr = data;
2758 } else if (data != 0) {
2759 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2764 case MSR_FAM10H_MMIO_CONF_BASE:
2766 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2771 case MSR_IA32_DEBUGCTLMSR:
2773 /* We support the non-activated case already */
2775 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2776 /* Values other than LBR and BTF are vendor-specific,
2777 thus reserved and should throw a #GP */
2780 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2783 case 0x200 ... 0x2ff:
2784 return kvm_mtrr_set_msr(vcpu, msr, data);
2785 case MSR_IA32_APICBASE:
2786 return kvm_set_apic_base(vcpu, msr_info);
2787 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2788 return kvm_x2apic_msr_write(vcpu, msr, data);
2789 case MSR_IA32_TSCDEADLINE:
2790 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2792 case MSR_IA32_TSC_ADJUST:
2793 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2794 if (!msr_info->host_initiated) {
2795 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2796 adjust_tsc_offset_guest(vcpu, adj);
2798 vcpu->arch.ia32_tsc_adjust_msr = data;
2801 case MSR_IA32_MISC_ENABLE:
2802 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2803 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2804 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2806 vcpu->arch.ia32_misc_enable_msr = data;
2807 kvm_update_cpuid(vcpu);
2809 vcpu->arch.ia32_misc_enable_msr = data;
2812 case MSR_IA32_SMBASE:
2813 if (!msr_info->host_initiated)
2815 vcpu->arch.smbase = data;
2817 case MSR_IA32_POWER_CTL:
2818 vcpu->arch.msr_ia32_power_ctl = data;
2821 kvm_write_tsc(vcpu, msr_info);
2824 if (!msr_info->host_initiated &&
2825 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2828 * KVM supports exposing PT to the guest, but does not support
2829 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
2830 * XSAVES/XRSTORS to save/restore PT MSRs.
2832 if (data & ~supported_xss)
2834 vcpu->arch.ia32_xss = data;
2837 if (!msr_info->host_initiated)
2839 vcpu->arch.smi_count = data;
2841 case MSR_KVM_WALL_CLOCK_NEW:
2842 case MSR_KVM_WALL_CLOCK:
2843 vcpu->kvm->arch.wall_clock = data;
2844 kvm_write_wall_clock(vcpu->kvm, data);
2846 case MSR_KVM_SYSTEM_TIME_NEW:
2847 case MSR_KVM_SYSTEM_TIME: {
2848 struct kvm_arch *ka = &vcpu->kvm->arch;
2850 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2851 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2853 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2854 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2856 ka->boot_vcpu_runs_old_kvmclock = tmp;
2859 vcpu->arch.time = data;
2860 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2862 /* we verify if the enable bit is set... */
2863 vcpu->arch.pv_time_enabled = false;
2867 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2868 &vcpu->arch.pv_time, data & ~1ULL,
2869 sizeof(struct pvclock_vcpu_time_info)))
2870 vcpu->arch.pv_time_enabled = true;
2874 case MSR_KVM_ASYNC_PF_EN:
2875 if (kvm_pv_enable_async_pf(vcpu, data))
2878 case MSR_KVM_STEAL_TIME:
2880 if (unlikely(!sched_info_on()))
2883 if (data & KVM_STEAL_RESERVED_MASK)
2886 vcpu->arch.st.msr_val = data;
2888 if (!(data & KVM_MSR_ENABLED))
2891 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2894 case MSR_KVM_PV_EOI_EN:
2895 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2899 case MSR_KVM_POLL_CONTROL:
2900 /* only enable bit supported */
2901 if (data & (-1ULL << 1))
2904 vcpu->arch.msr_kvm_poll_control = data;
2907 case MSR_IA32_MCG_CTL:
2908 case MSR_IA32_MCG_STATUS:
2909 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2910 return set_msr_mce(vcpu, msr_info);
2912 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2913 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2914 pr = true; /* fall through */
2915 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2916 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2917 if (kvm_pmu_is_valid_msr(vcpu, msr))
2918 return kvm_pmu_set_msr(vcpu, msr_info);
2920 if (pr || data != 0)
2921 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2922 "0x%x data 0x%llx\n", msr, data);
2924 case MSR_K7_CLK_CTL:
2926 * Ignore all writes to this no longer documented MSR.
2927 * Writes are only relevant for old K7 processors,
2928 * all pre-dating SVM, but a recommended workaround from
2929 * AMD for these chips. It is possible to specify the
2930 * affected processor models on the command line, hence
2931 * the need to ignore the workaround.
2934 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2935 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2936 case HV_X64_MSR_CRASH_CTL:
2937 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2938 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2939 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2940 case HV_X64_MSR_TSC_EMULATION_STATUS:
2941 return kvm_hv_set_msr_common(vcpu, msr, data,
2942 msr_info->host_initiated);
2943 case MSR_IA32_BBL_CR_CTL3:
2944 /* Drop writes to this legacy MSR -- see rdmsr
2945 * counterpart for further detail.
2947 if (report_ignored_msrs)
2948 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2951 case MSR_AMD64_OSVW_ID_LENGTH:
2952 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2954 vcpu->arch.osvw.length = data;
2956 case MSR_AMD64_OSVW_STATUS:
2957 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2959 vcpu->arch.osvw.status = data;
2961 case MSR_PLATFORM_INFO:
2962 if (!msr_info->host_initiated ||
2963 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2964 cpuid_fault_enabled(vcpu)))
2966 vcpu->arch.msr_platform_info = data;
2968 case MSR_MISC_FEATURES_ENABLES:
2969 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2970 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2971 !supports_cpuid_fault(vcpu)))
2973 vcpu->arch.msr_misc_features_enables = data;
2976 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2977 return xen_hvm_config(vcpu, data);
2978 if (kvm_pmu_is_valid_msr(vcpu, msr))
2979 return kvm_pmu_set_msr(vcpu, msr_info);
2981 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2985 if (report_ignored_msrs)
2987 "ignored wrmsr: 0x%x data 0x%llx\n",
2994 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2996 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2999 u64 mcg_cap = vcpu->arch.mcg_cap;
3000 unsigned bank_num = mcg_cap & 0xff;
3003 case MSR_IA32_P5_MC_ADDR:
3004 case MSR_IA32_P5_MC_TYPE:
3007 case MSR_IA32_MCG_CAP:
3008 data = vcpu->arch.mcg_cap;
3010 case MSR_IA32_MCG_CTL:
3011 if (!(mcg_cap & MCG_CTL_P) && !host)
3013 data = vcpu->arch.mcg_ctl;
3015 case MSR_IA32_MCG_STATUS:
3016 data = vcpu->arch.mcg_status;
3019 if (msr >= MSR_IA32_MC0_CTL &&
3020 msr < MSR_IA32_MCx_CTL(bank_num)) {
3021 u32 offset = array_index_nospec(
3022 msr - MSR_IA32_MC0_CTL,
3023 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3025 data = vcpu->arch.mce_banks[offset];
3034 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3036 switch (msr_info->index) {
3037 case MSR_IA32_PLATFORM_ID:
3038 case MSR_IA32_EBL_CR_POWERON:
3039 case MSR_IA32_DEBUGCTLMSR:
3040 case MSR_IA32_LASTBRANCHFROMIP:
3041 case MSR_IA32_LASTBRANCHTOIP:
3042 case MSR_IA32_LASTINTFROMIP:
3043 case MSR_IA32_LASTINTTOIP:
3045 case MSR_K8_TSEG_ADDR:
3046 case MSR_K8_TSEG_MASK:
3047 case MSR_VM_HSAVE_PA:
3048 case MSR_K8_INT_PENDING_MSG:
3049 case MSR_AMD64_NB_CFG:
3050 case MSR_FAM10H_MMIO_CONF_BASE:
3051 case MSR_AMD64_BU_CFG2:
3052 case MSR_IA32_PERF_CTL:
3053 case MSR_AMD64_DC_CFG:
3054 case MSR_F15H_EX_CFG:
3056 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3057 * limit) MSRs. Just return 0, as we do not want to expose the host
3058 * data here. Do not conditionalize this on CPUID, as KVM does not do
3059 * so for existing CPU-specific MSRs.
3061 case MSR_RAPL_POWER_UNIT:
3062 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3063 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3064 case MSR_PKG_ENERGY_STATUS: /* Total package */
3065 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3068 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3069 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3070 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3071 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3072 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3073 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3074 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3077 case MSR_IA32_UCODE_REV:
3078 msr_info->data = vcpu->arch.microcode_version;
3080 case MSR_IA32_ARCH_CAPABILITIES:
3081 if (!msr_info->host_initiated &&
3082 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3084 msr_info->data = vcpu->arch.arch_capabilities;
3086 case MSR_IA32_POWER_CTL:
3087 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3090 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
3093 case 0x200 ... 0x2ff:
3094 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3095 case 0xcd: /* fsb frequency */
3099 * MSR_EBC_FREQUENCY_ID
3100 * Conservative value valid for even the basic CPU models.
3101 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3102 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3103 * and 266MHz for model 3, or 4. Set Core Clock
3104 * Frequency to System Bus Frequency Ratio to 1 (bits
3105 * 31:24) even though these are only valid for CPU
3106 * models > 2, however guests may end up dividing or
3107 * multiplying by zero otherwise.
3109 case MSR_EBC_FREQUENCY_ID:
3110 msr_info->data = 1 << 24;
3112 case MSR_IA32_APICBASE:
3113 msr_info->data = kvm_get_apic_base(vcpu);
3115 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
3116 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3117 case MSR_IA32_TSCDEADLINE:
3118 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3120 case MSR_IA32_TSC_ADJUST:
3121 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3123 case MSR_IA32_MISC_ENABLE:
3124 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3126 case MSR_IA32_SMBASE:
3127 if (!msr_info->host_initiated)
3129 msr_info->data = vcpu->arch.smbase;
3132 msr_info->data = vcpu->arch.smi_count;
3134 case MSR_IA32_PERF_STATUS:
3135 /* TSC increment by tick */
3136 msr_info->data = 1000ULL;
3137 /* CPU multiplier */
3138 msr_info->data |= (((uint64_t)4ULL) << 40);
3141 msr_info->data = vcpu->arch.efer;
3143 case MSR_KVM_WALL_CLOCK:
3144 case MSR_KVM_WALL_CLOCK_NEW:
3145 msr_info->data = vcpu->kvm->arch.wall_clock;
3147 case MSR_KVM_SYSTEM_TIME:
3148 case MSR_KVM_SYSTEM_TIME_NEW:
3149 msr_info->data = vcpu->arch.time;
3151 case MSR_KVM_ASYNC_PF_EN:
3152 msr_info->data = vcpu->arch.apf.msr_val;
3154 case MSR_KVM_STEAL_TIME:
3155 msr_info->data = vcpu->arch.st.msr_val;
3157 case MSR_KVM_PV_EOI_EN:
3158 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3160 case MSR_KVM_POLL_CONTROL:
3161 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3163 case MSR_IA32_P5_MC_ADDR:
3164 case MSR_IA32_P5_MC_TYPE:
3165 case MSR_IA32_MCG_CAP:
3166 case MSR_IA32_MCG_CTL:
3167 case MSR_IA32_MCG_STATUS:
3168 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3169 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3170 msr_info->host_initiated);
3172 if (!msr_info->host_initiated &&
3173 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3175 msr_info->data = vcpu->arch.ia32_xss;
3177 case MSR_K7_CLK_CTL:
3179 * Provide expected ramp-up count for K7. All other
3180 * are set to zero, indicating minimum divisors for
3183 * This prevents guest kernels on AMD host with CPU
3184 * type 6, model 8 and higher from exploding due to
3185 * the rdmsr failing.
3187 msr_info->data = 0x20000000;
3189 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3190 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3191 case HV_X64_MSR_CRASH_CTL:
3192 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3193 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3194 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3195 case HV_X64_MSR_TSC_EMULATION_STATUS:
3196 return kvm_hv_get_msr_common(vcpu,
3197 msr_info->index, &msr_info->data,
3198 msr_info->host_initiated);
3199 case MSR_IA32_BBL_CR_CTL3:
3200 /* This legacy MSR exists but isn't fully documented in current
3201 * silicon. It is however accessed by winxp in very narrow
3202 * scenarios where it sets bit #19, itself documented as
3203 * a "reserved" bit. Best effort attempt to source coherent
3204 * read data here should the balance of the register be
3205 * interpreted by the guest:
3207 * L2 cache control register 3: 64GB range, 256KB size,
3208 * enabled, latency 0x1, configured
3210 msr_info->data = 0xbe702111;
3212 case MSR_AMD64_OSVW_ID_LENGTH:
3213 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3215 msr_info->data = vcpu->arch.osvw.length;
3217 case MSR_AMD64_OSVW_STATUS:
3218 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3220 msr_info->data = vcpu->arch.osvw.status;
3222 case MSR_PLATFORM_INFO:
3223 if (!msr_info->host_initiated &&
3224 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3226 msr_info->data = vcpu->arch.msr_platform_info;
3228 case MSR_MISC_FEATURES_ENABLES:
3229 msr_info->data = vcpu->arch.msr_misc_features_enables;
3232 msr_info->data = vcpu->arch.msr_hwcr;
3235 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3236 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3238 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3242 if (report_ignored_msrs)
3243 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3251 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3254 * Read or write a bunch of msrs. All parameters are kernel addresses.
3256 * @return number of msrs set successfully.
3258 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3259 struct kvm_msr_entry *entries,
3260 int (*do_msr)(struct kvm_vcpu *vcpu,
3261 unsigned index, u64 *data))
3265 for (i = 0; i < msrs->nmsrs; ++i)
3266 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3273 * Read or write a bunch of msrs. Parameters are user addresses.
3275 * @return number of msrs set successfully.
3277 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3278 int (*do_msr)(struct kvm_vcpu *vcpu,
3279 unsigned index, u64 *data),
3282 struct kvm_msrs msrs;
3283 struct kvm_msr_entry *entries;
3288 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3292 if (msrs.nmsrs >= MAX_IO_MSRS)
3295 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3296 entries = memdup_user(user_msrs->entries, size);
3297 if (IS_ERR(entries)) {
3298 r = PTR_ERR(entries);
3302 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3307 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3318 static inline bool kvm_can_mwait_in_guest(void)
3320 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3321 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3322 boot_cpu_has(X86_FEATURE_ARAT);
3325 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3330 case KVM_CAP_IRQCHIP:
3332 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3333 case KVM_CAP_SET_TSS_ADDR:
3334 case KVM_CAP_EXT_CPUID:
3335 case KVM_CAP_EXT_EMUL_CPUID:
3336 case KVM_CAP_CLOCKSOURCE:
3338 case KVM_CAP_NOP_IO_DELAY:
3339 case KVM_CAP_MP_STATE:
3340 case KVM_CAP_SYNC_MMU:
3341 case KVM_CAP_USER_NMI:
3342 case KVM_CAP_REINJECT_CONTROL:
3343 case KVM_CAP_IRQ_INJECT_STATUS:
3344 case KVM_CAP_IOEVENTFD:
3345 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3347 case KVM_CAP_PIT_STATE2:
3348 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3349 case KVM_CAP_XEN_HVM:
3350 case KVM_CAP_VCPU_EVENTS:
3351 case KVM_CAP_HYPERV:
3352 case KVM_CAP_HYPERV_VAPIC:
3353 case KVM_CAP_HYPERV_SPIN:
3354 case KVM_CAP_HYPERV_SYNIC:
3355 case KVM_CAP_HYPERV_SYNIC2:
3356 case KVM_CAP_HYPERV_VP_INDEX:
3357 case KVM_CAP_HYPERV_EVENTFD:
3358 case KVM_CAP_HYPERV_TLBFLUSH:
3359 case KVM_CAP_HYPERV_SEND_IPI:
3360 case KVM_CAP_HYPERV_CPUID:
3361 case KVM_CAP_PCI_SEGMENT:
3362 case KVM_CAP_DEBUGREGS:
3363 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3365 case KVM_CAP_ASYNC_PF:
3366 case KVM_CAP_GET_TSC_KHZ:
3367 case KVM_CAP_KVMCLOCK_CTRL:
3368 case KVM_CAP_READONLY_MEM:
3369 case KVM_CAP_HYPERV_TIME:
3370 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3371 case KVM_CAP_TSC_DEADLINE_TIMER:
3372 case KVM_CAP_DISABLE_QUIRKS:
3373 case KVM_CAP_SET_BOOT_CPU_ID:
3374 case KVM_CAP_SPLIT_IRQCHIP:
3375 case KVM_CAP_IMMEDIATE_EXIT:
3376 case KVM_CAP_PMU_EVENT_FILTER:
3377 case KVM_CAP_GET_MSR_FEATURES:
3378 case KVM_CAP_MSR_PLATFORM_INFO:
3379 case KVM_CAP_EXCEPTION_PAYLOAD:
3380 case KVM_CAP_SET_GUEST_DEBUG:
3383 case KVM_CAP_SYNC_REGS:
3384 r = KVM_SYNC_X86_VALID_FIELDS;
3386 case KVM_CAP_ADJUST_CLOCK:
3387 r = KVM_CLOCK_TSC_STABLE;
3389 case KVM_CAP_X86_DISABLE_EXITS:
3390 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3391 KVM_X86_DISABLE_EXITS_CSTATE;
3392 if(kvm_can_mwait_in_guest())
3393 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3395 case KVM_CAP_X86_SMM:
3396 /* SMBASE is usually relocated above 1M on modern chipsets,
3397 * and SMM handlers might indeed rely on 4G segment limits,
3398 * so do not report SMM to be available if real mode is
3399 * emulated via vm86 mode. Still, do not go to great lengths
3400 * to avoid userspace's usage of the feature, because it is a
3401 * fringe case that is not enabled except via specific settings
3402 * of the module parameters.
3404 r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3407 r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3409 case KVM_CAP_NR_VCPUS:
3410 r = KVM_SOFT_MAX_VCPUS;
3412 case KVM_CAP_MAX_VCPUS:
3415 case KVM_CAP_MAX_VCPU_ID:
3416 r = KVM_MAX_VCPU_ID;
3418 case KVM_CAP_PV_MMU: /* obsolete */
3422 r = KVM_MAX_MCE_BANKS;
3425 r = boot_cpu_has(X86_FEATURE_XSAVE);
3427 case KVM_CAP_TSC_CONTROL:
3428 r = kvm_has_tsc_control;
3430 case KVM_CAP_X2APIC_API:
3431 r = KVM_X2APIC_API_VALID_FLAGS;
3433 case KVM_CAP_NESTED_STATE:
3434 r = kvm_x86_ops.get_nested_state ?
3435 kvm_x86_ops.get_nested_state(NULL, NULL, 0) : 0;
3437 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3438 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3440 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3441 r = kvm_x86_ops.nested_enable_evmcs != NULL;
3450 long kvm_arch_dev_ioctl(struct file *filp,
3451 unsigned int ioctl, unsigned long arg)
3453 void __user *argp = (void __user *)arg;
3457 case KVM_GET_MSR_INDEX_LIST: {
3458 struct kvm_msr_list __user *user_msr_list = argp;
3459 struct kvm_msr_list msr_list;
3463 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3466 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3467 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3470 if (n < msr_list.nmsrs)
3473 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3474 num_msrs_to_save * sizeof(u32)))
3476 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3478 num_emulated_msrs * sizeof(u32)))
3483 case KVM_GET_SUPPORTED_CPUID:
3484 case KVM_GET_EMULATED_CPUID: {
3485 struct kvm_cpuid2 __user *cpuid_arg = argp;
3486 struct kvm_cpuid2 cpuid;
3489 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3492 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3498 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3503 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3505 if (copy_to_user(argp, &kvm_mce_cap_supported,
3506 sizeof(kvm_mce_cap_supported)))
3510 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3511 struct kvm_msr_list __user *user_msr_list = argp;
3512 struct kvm_msr_list msr_list;
3516 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3519 msr_list.nmsrs = num_msr_based_features;
3520 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3523 if (n < msr_list.nmsrs)
3526 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3527 num_msr_based_features * sizeof(u32)))
3533 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3543 static void wbinvd_ipi(void *garbage)
3548 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3550 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3553 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3555 /* Address WBINVD may be executed by guest */
3556 if (need_emulate_wbinvd(vcpu)) {
3557 if (kvm_x86_ops.has_wbinvd_exit())
3558 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3559 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3560 smp_call_function_single(vcpu->cpu,
3561 wbinvd_ipi, NULL, 1);
3564 kvm_x86_ops.vcpu_load(vcpu, cpu);
3566 /* Save host pkru register if supported */
3567 vcpu->arch.host_pkru = read_pkru();
3569 /* Apply any externally detected TSC adjustments (due to suspend) */
3570 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3571 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3572 vcpu->arch.tsc_offset_adjustment = 0;
3573 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3576 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3577 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3578 rdtsc() - vcpu->arch.last_host_tsc;
3580 mark_tsc_unstable("KVM discovered backwards TSC");
3582 if (kvm_check_tsc_unstable()) {
3583 u64 offset = kvm_compute_tsc_offset(vcpu,
3584 vcpu->arch.last_guest_tsc);
3585 kvm_vcpu_write_tsc_offset(vcpu, offset);
3586 vcpu->arch.tsc_catchup = 1;
3589 if (kvm_lapic_hv_timer_in_use(vcpu))
3590 kvm_lapic_restart_hv_timer(vcpu);
3593 * On a host with synchronized TSC, there is no need to update
3594 * kvmclock on vcpu->cpu migration
3596 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3597 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3598 if (vcpu->cpu != cpu)
3599 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3603 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3606 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3608 struct kvm_host_map map;
3609 struct kvm_steal_time *st;
3611 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3614 if (vcpu->arch.st.preempted)
3617 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3618 &vcpu->arch.st.cache, true))
3622 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3624 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3626 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3629 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3633 if (vcpu->preempted)
3634 vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
3637 * Disable page faults because we're in atomic context here.
3638 * kvm_write_guest_offset_cached() would call might_fault()
3639 * that relies on pagefault_disable() to tell if there's a
3640 * bug. NOTE: the write to guest memory may not go through if
3641 * during postcopy live migration or if there's heavy guest
3644 pagefault_disable();
3646 * kvm_memslots() will be called by
3647 * kvm_write_guest_offset_cached() so take the srcu lock.
3649 idx = srcu_read_lock(&vcpu->kvm->srcu);
3650 kvm_steal_time_set_preempted(vcpu);
3651 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3653 kvm_x86_ops.vcpu_put(vcpu);
3654 vcpu->arch.last_host_tsc = rdtsc();
3656 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3657 * on every vmexit, but if not, we might have a stale dr6 from the
3658 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3663 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3664 struct kvm_lapic_state *s)
3666 if (vcpu->arch.apicv_active)
3667 kvm_x86_ops.sync_pir_to_irr(vcpu);
3669 return kvm_apic_get_state(vcpu, s);
3672 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3673 struct kvm_lapic_state *s)
3677 r = kvm_apic_set_state(vcpu, s);
3680 update_cr8_intercept(vcpu);
3685 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3687 return (!lapic_in_kernel(vcpu) ||
3688 kvm_apic_accept_pic_intr(vcpu));
3692 * if userspace requested an interrupt window, check that the
3693 * interrupt window is open.
3695 * No need to exit to userspace if we already have an interrupt queued.
3697 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3699 return kvm_arch_interrupt_allowed(vcpu) &&
3700 !kvm_cpu_has_interrupt(vcpu) &&
3701 !kvm_event_needs_reinjection(vcpu) &&
3702 kvm_cpu_accept_dm_intr(vcpu);
3705 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3706 struct kvm_interrupt *irq)
3708 if (irq->irq >= KVM_NR_INTERRUPTS)
3711 if (!irqchip_in_kernel(vcpu->kvm)) {
3712 kvm_queue_interrupt(vcpu, irq->irq, false);
3713 kvm_make_request(KVM_REQ_EVENT, vcpu);
3718 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3719 * fail for in-kernel 8259.
3721 if (pic_in_kernel(vcpu->kvm))
3724 if (vcpu->arch.pending_external_vector != -1)
3727 vcpu->arch.pending_external_vector = irq->irq;
3728 kvm_make_request(KVM_REQ_EVENT, vcpu);
3732 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3734 kvm_inject_nmi(vcpu);
3739 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3741 kvm_make_request(KVM_REQ_SMI, vcpu);
3746 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3747 struct kvm_tpr_access_ctl *tac)
3751 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3755 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3759 unsigned bank_num = mcg_cap & 0xff, bank;
3762 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
3764 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3767 vcpu->arch.mcg_cap = mcg_cap;
3768 /* Init IA32_MCG_CTL to all 1s */
3769 if (mcg_cap & MCG_CTL_P)
3770 vcpu->arch.mcg_ctl = ~(u64)0;
3771 /* Init IA32_MCi_CTL to all 1s */
3772 for (bank = 0; bank < bank_num; bank++)
3773 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3775 kvm_x86_ops.setup_mce(vcpu);
3780 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3781 struct kvm_x86_mce *mce)
3783 u64 mcg_cap = vcpu->arch.mcg_cap;
3784 unsigned bank_num = mcg_cap & 0xff;
3785 u64 *banks = vcpu->arch.mce_banks;
3787 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3790 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3791 * reporting is disabled
3793 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3794 vcpu->arch.mcg_ctl != ~(u64)0)
3796 banks += 4 * mce->bank;
3798 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3799 * reporting is disabled for the bank
3801 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3803 if (mce->status & MCI_STATUS_UC) {
3804 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3805 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3806 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3809 if (banks[1] & MCI_STATUS_VAL)
3810 mce->status |= MCI_STATUS_OVER;
3811 banks[2] = mce->addr;
3812 banks[3] = mce->misc;
3813 vcpu->arch.mcg_status = mce->mcg_status;
3814 banks[1] = mce->status;
3815 kvm_queue_exception(vcpu, MC_VECTOR);
3816 } else if (!(banks[1] & MCI_STATUS_VAL)
3817 || !(banks[1] & MCI_STATUS_UC)) {
3818 if (banks[1] & MCI_STATUS_VAL)
3819 mce->status |= MCI_STATUS_OVER;
3820 banks[2] = mce->addr;
3821 banks[3] = mce->misc;
3822 banks[1] = mce->status;
3824 banks[1] |= MCI_STATUS_OVER;
3828 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3829 struct kvm_vcpu_events *events)
3834 * In guest mode, payload delivery should be deferred,
3835 * so that the L1 hypervisor can intercept #PF before
3836 * CR2 is modified (or intercept #DB before DR6 is
3837 * modified under nVMX). Unless the per-VM capability,
3838 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
3839 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
3840 * opportunistically defer the exception payload, deliver it if the
3841 * capability hasn't been requested before processing a
3842 * KVM_GET_VCPU_EVENTS.
3844 if (!vcpu->kvm->arch.exception_payload_enabled &&
3845 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
3846 kvm_deliver_exception_payload(vcpu);
3849 * The API doesn't provide the instruction length for software
3850 * exceptions, so don't report them. As long as the guest RIP
3851 * isn't advanced, we should expect to encounter the exception
3854 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3855 events->exception.injected = 0;
3856 events->exception.pending = 0;
3858 events->exception.injected = vcpu->arch.exception.injected;
3859 events->exception.pending = vcpu->arch.exception.pending;
3861 * For ABI compatibility, deliberately conflate
3862 * pending and injected exceptions when
3863 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3865 if (!vcpu->kvm->arch.exception_payload_enabled)
3866 events->exception.injected |=
3867 vcpu->arch.exception.pending;
3869 events->exception.nr = vcpu->arch.exception.nr;
3870 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3871 events->exception.error_code = vcpu->arch.exception.error_code;
3872 events->exception_has_payload = vcpu->arch.exception.has_payload;
3873 events->exception_payload = vcpu->arch.exception.payload;
3875 events->interrupt.injected =
3876 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3877 events->interrupt.nr = vcpu->arch.interrupt.nr;
3878 events->interrupt.soft = 0;
3879 events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
3881 events->nmi.injected = vcpu->arch.nmi_injected;
3882 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3883 events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
3884 events->nmi.pad = 0;
3886 events->sipi_vector = 0; /* never valid when reporting to user space */
3888 events->smi.smm = is_smm(vcpu);
3889 events->smi.pending = vcpu->arch.smi_pending;
3890 events->smi.smm_inside_nmi =
3891 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3892 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3894 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3895 | KVM_VCPUEVENT_VALID_SHADOW
3896 | KVM_VCPUEVENT_VALID_SMM);
3897 if (vcpu->kvm->arch.exception_payload_enabled)
3898 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3900 memset(&events->reserved, 0, sizeof(events->reserved));
3903 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3905 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3906 struct kvm_vcpu_events *events)
3908 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3909 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3910 | KVM_VCPUEVENT_VALID_SHADOW
3911 | KVM_VCPUEVENT_VALID_SMM
3912 | KVM_VCPUEVENT_VALID_PAYLOAD))
3915 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3916 if (!vcpu->kvm->arch.exception_payload_enabled)
3918 if (events->exception.pending)
3919 events->exception.injected = 0;
3921 events->exception_has_payload = 0;
3923 events->exception.pending = 0;
3924 events->exception_has_payload = 0;
3927 if ((events->exception.injected || events->exception.pending) &&
3928 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3931 /* INITs are latched while in SMM */
3932 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3933 (events->smi.smm || events->smi.pending) &&
3934 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3938 vcpu->arch.exception.injected = events->exception.injected;
3939 vcpu->arch.exception.pending = events->exception.pending;
3940 vcpu->arch.exception.nr = events->exception.nr;
3941 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3942 vcpu->arch.exception.error_code = events->exception.error_code;
3943 vcpu->arch.exception.has_payload = events->exception_has_payload;
3944 vcpu->arch.exception.payload = events->exception_payload;
3946 vcpu->arch.interrupt.injected = events->interrupt.injected;
3947 vcpu->arch.interrupt.nr = events->interrupt.nr;
3948 vcpu->arch.interrupt.soft = events->interrupt.soft;
3949 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3950 kvm_x86_ops.set_interrupt_shadow(vcpu,
3951 events->interrupt.shadow);
3953 vcpu->arch.nmi_injected = events->nmi.injected;
3954 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3955 vcpu->arch.nmi_pending = events->nmi.pending;
3956 kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
3958 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3959 lapic_in_kernel(vcpu))
3960 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3962 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3963 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3964 if (events->smi.smm)
3965 vcpu->arch.hflags |= HF_SMM_MASK;
3967 vcpu->arch.hflags &= ~HF_SMM_MASK;
3968 kvm_smm_changed(vcpu);
3971 vcpu->arch.smi_pending = events->smi.pending;
3973 if (events->smi.smm) {
3974 if (events->smi.smm_inside_nmi)
3975 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3977 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3980 if (lapic_in_kernel(vcpu)) {
3981 if (events->smi.latched_init)
3982 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3984 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3988 kvm_make_request(KVM_REQ_EVENT, vcpu);
3993 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3994 struct kvm_debugregs *dbgregs)
3998 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3999 kvm_get_dr(vcpu, 6, &val);
4001 dbgregs->dr7 = vcpu->arch.dr7;
4003 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4006 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4007 struct kvm_debugregs *dbgregs)
4012 if (dbgregs->dr6 & ~0xffffffffull)
4014 if (dbgregs->dr7 & ~0xffffffffull)
4017 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4018 kvm_update_dr0123(vcpu);
4019 vcpu->arch.dr6 = dbgregs->dr6;
4020 vcpu->arch.dr7 = dbgregs->dr7;
4021 kvm_update_dr7(vcpu);
4026 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4028 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4030 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4031 u64 xstate_bv = xsave->header.xfeatures;
4035 * Copy legacy XSAVE area, to avoid complications with CPUID
4036 * leaves 0 and 1 in the loop below.
4038 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4041 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4042 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4045 * Copy each region from the possibly compacted offset to the
4046 * non-compacted offset.
4048 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4050 u64 xfeature_mask = valid & -valid;
4051 int xfeature_nr = fls64(xfeature_mask) - 1;
4052 void *src = get_xsave_addr(xsave, xfeature_nr);
4055 u32 size, offset, ecx, edx;
4056 cpuid_count(XSTATE_CPUID, xfeature_nr,
4057 &size, &offset, &ecx, &edx);
4058 if (xfeature_nr == XFEATURE_PKRU)
4059 memcpy(dest + offset, &vcpu->arch.pkru,
4060 sizeof(vcpu->arch.pkru));
4062 memcpy(dest + offset, src, size);
4066 valid -= xfeature_mask;
4070 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4072 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4073 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4077 * Copy legacy XSAVE area, to avoid complications with CPUID
4078 * leaves 0 and 1 in the loop below.
4080 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4082 /* Set XSTATE_BV and possibly XCOMP_BV. */
4083 xsave->header.xfeatures = xstate_bv;
4084 if (boot_cpu_has(X86_FEATURE_XSAVES))
4085 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4088 * Copy each region from the non-compacted offset to the
4089 * possibly compacted offset.
4091 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4093 u64 xfeature_mask = valid & -valid;
4094 int xfeature_nr = fls64(xfeature_mask) - 1;
4095 void *dest = get_xsave_addr(xsave, xfeature_nr);
4098 u32 size, offset, ecx, edx;
4099 cpuid_count(XSTATE_CPUID, xfeature_nr,
4100 &size, &offset, &ecx, &edx);
4101 if (xfeature_nr == XFEATURE_PKRU)
4102 memcpy(&vcpu->arch.pkru, src + offset,
4103 sizeof(vcpu->arch.pkru));
4105 memcpy(dest, src + offset, size);
4108 valid -= xfeature_mask;
4112 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4113 struct kvm_xsave *guest_xsave)
4115 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4116 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4117 fill_xsave((u8 *) guest_xsave->region, vcpu);
4119 memcpy(guest_xsave->region,
4120 &vcpu->arch.guest_fpu->state.fxsave,
4121 sizeof(struct fxregs_state));
4122 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4123 XFEATURE_MASK_FPSSE;
4127 #define XSAVE_MXCSR_OFFSET 24
4129 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4130 struct kvm_xsave *guest_xsave)
4133 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4134 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4136 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4138 * Here we allow setting states that are not present in
4139 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4140 * with old userspace.
4142 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4144 load_xsave(vcpu, (u8 *)guest_xsave->region);
4146 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4147 mxcsr & ~mxcsr_feature_mask)
4149 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4150 guest_xsave->region, sizeof(struct fxregs_state));
4155 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4156 struct kvm_xcrs *guest_xcrs)
4158 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4159 guest_xcrs->nr_xcrs = 0;
4163 guest_xcrs->nr_xcrs = 1;
4164 guest_xcrs->flags = 0;
4165 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4166 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4169 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4170 struct kvm_xcrs *guest_xcrs)
4174 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4177 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4180 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4181 /* Only support XCR0 currently */
4182 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4183 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4184 guest_xcrs->xcrs[i].value);
4193 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4194 * stopped by the hypervisor. This function will be called from the host only.
4195 * EINVAL is returned when the host attempts to set the flag for a guest that
4196 * does not support pv clocks.
4198 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4200 if (!vcpu->arch.pv_time_enabled)
4202 vcpu->arch.pvclock_set_guest_stopped_request = true;
4203 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4207 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4208 struct kvm_enable_cap *cap)
4211 uint16_t vmcs_version;
4212 void __user *user_ptr;
4218 case KVM_CAP_HYPERV_SYNIC2:
4223 case KVM_CAP_HYPERV_SYNIC:
4224 if (!irqchip_in_kernel(vcpu->kvm))
4226 return kvm_hv_activate_synic(vcpu, cap->cap ==
4227 KVM_CAP_HYPERV_SYNIC2);
4228 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4229 if (!kvm_x86_ops.nested_enable_evmcs)
4231 r = kvm_x86_ops.nested_enable_evmcs(vcpu, &vmcs_version);
4233 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4234 if (copy_to_user(user_ptr, &vmcs_version,
4235 sizeof(vmcs_version)))
4239 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4240 if (!kvm_x86_ops.enable_direct_tlbflush)
4243 return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4250 long kvm_arch_vcpu_ioctl(struct file *filp,
4251 unsigned int ioctl, unsigned long arg)
4253 struct kvm_vcpu *vcpu = filp->private_data;
4254 void __user *argp = (void __user *)arg;
4257 struct kvm_lapic_state *lapic;
4258 struct kvm_xsave *xsave;
4259 struct kvm_xcrs *xcrs;
4267 case KVM_GET_LAPIC: {
4269 if (!lapic_in_kernel(vcpu))
4271 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4272 GFP_KERNEL_ACCOUNT);
4277 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4281 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4286 case KVM_SET_LAPIC: {
4288 if (!lapic_in_kernel(vcpu))
4290 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4291 if (IS_ERR(u.lapic)) {
4292 r = PTR_ERR(u.lapic);
4296 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4299 case KVM_INTERRUPT: {
4300 struct kvm_interrupt irq;
4303 if (copy_from_user(&irq, argp, sizeof(irq)))
4305 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4309 r = kvm_vcpu_ioctl_nmi(vcpu);
4313 r = kvm_vcpu_ioctl_smi(vcpu);
4316 case KVM_SET_CPUID: {
4317 struct kvm_cpuid __user *cpuid_arg = argp;
4318 struct kvm_cpuid cpuid;
4321 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4323 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4326 case KVM_SET_CPUID2: {
4327 struct kvm_cpuid2 __user *cpuid_arg = argp;
4328 struct kvm_cpuid2 cpuid;
4331 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4333 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4334 cpuid_arg->entries);
4337 case KVM_GET_CPUID2: {
4338 struct kvm_cpuid2 __user *cpuid_arg = argp;
4339 struct kvm_cpuid2 cpuid;
4342 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4344 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4345 cpuid_arg->entries);
4349 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4354 case KVM_GET_MSRS: {
4355 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4356 r = msr_io(vcpu, argp, do_get_msr, 1);
4357 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4360 case KVM_SET_MSRS: {
4361 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4362 r = msr_io(vcpu, argp, do_set_msr, 0);
4363 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4366 case KVM_TPR_ACCESS_REPORTING: {
4367 struct kvm_tpr_access_ctl tac;
4370 if (copy_from_user(&tac, argp, sizeof(tac)))
4372 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4376 if (copy_to_user(argp, &tac, sizeof(tac)))
4381 case KVM_SET_VAPIC_ADDR: {
4382 struct kvm_vapic_addr va;
4386 if (!lapic_in_kernel(vcpu))
4389 if (copy_from_user(&va, argp, sizeof(va)))
4391 idx = srcu_read_lock(&vcpu->kvm->srcu);
4392 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4393 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4396 case KVM_X86_SETUP_MCE: {
4400 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4402 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4405 case KVM_X86_SET_MCE: {
4406 struct kvm_x86_mce mce;
4409 if (copy_from_user(&mce, argp, sizeof(mce)))
4411 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4414 case KVM_GET_VCPU_EVENTS: {
4415 struct kvm_vcpu_events events;
4417 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4420 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4425 case KVM_SET_VCPU_EVENTS: {
4426 struct kvm_vcpu_events events;
4429 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4432 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4435 case KVM_GET_DEBUGREGS: {
4436 struct kvm_debugregs dbgregs;
4438 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4441 if (copy_to_user(argp, &dbgregs,
4442 sizeof(struct kvm_debugregs)))
4447 case KVM_SET_DEBUGREGS: {
4448 struct kvm_debugregs dbgregs;
4451 if (copy_from_user(&dbgregs, argp,
4452 sizeof(struct kvm_debugregs)))
4455 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4458 case KVM_GET_XSAVE: {
4459 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4464 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4467 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4472 case KVM_SET_XSAVE: {
4473 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4474 if (IS_ERR(u.xsave)) {
4475 r = PTR_ERR(u.xsave);
4479 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4482 case KVM_GET_XCRS: {
4483 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4488 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4491 if (copy_to_user(argp, u.xcrs,
4492 sizeof(struct kvm_xcrs)))
4497 case KVM_SET_XCRS: {
4498 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4499 if (IS_ERR(u.xcrs)) {
4500 r = PTR_ERR(u.xcrs);
4504 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4507 case KVM_SET_TSC_KHZ: {
4511 user_tsc_khz = (u32)arg;
4513 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4516 if (user_tsc_khz == 0)
4517 user_tsc_khz = tsc_khz;
4519 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4524 case KVM_GET_TSC_KHZ: {
4525 r = vcpu->arch.virtual_tsc_khz;
4528 case KVM_KVMCLOCK_CTRL: {
4529 r = kvm_set_guest_paused(vcpu);
4532 case KVM_ENABLE_CAP: {
4533 struct kvm_enable_cap cap;
4536 if (copy_from_user(&cap, argp, sizeof(cap)))
4538 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4541 case KVM_GET_NESTED_STATE: {
4542 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4546 if (!kvm_x86_ops.get_nested_state)
4549 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4551 if (get_user(user_data_size, &user_kvm_nested_state->size))
4554 r = kvm_x86_ops.get_nested_state(vcpu, user_kvm_nested_state,
4559 if (r > user_data_size) {
4560 if (put_user(r, &user_kvm_nested_state->size))
4570 case KVM_SET_NESTED_STATE: {
4571 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4572 struct kvm_nested_state kvm_state;
4576 if (!kvm_x86_ops.set_nested_state)
4580 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4584 if (kvm_state.size < sizeof(kvm_state))
4587 if (kvm_state.flags &
4588 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4589 | KVM_STATE_NESTED_EVMCS))
4592 /* nested_run_pending implies guest_mode. */
4593 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4594 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4597 idx = srcu_read_lock(&vcpu->kvm->srcu);
4598 r = kvm_x86_ops.set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4599 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4602 case KVM_GET_SUPPORTED_HV_CPUID: {
4603 struct kvm_cpuid2 __user *cpuid_arg = argp;
4604 struct kvm_cpuid2 cpuid;
4607 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4610 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4611 cpuid_arg->entries);
4616 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4631 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4633 return VM_FAULT_SIGBUS;
4636 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4640 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4642 ret = kvm_x86_ops.set_tss_addr(kvm, addr);
4646 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4649 return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
4652 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4653 unsigned long kvm_nr_mmu_pages)
4655 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4658 mutex_lock(&kvm->slots_lock);
4660 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4661 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4663 mutex_unlock(&kvm->slots_lock);
4667 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4669 return kvm->arch.n_max_mmu_pages;
4672 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4674 struct kvm_pic *pic = kvm->arch.vpic;
4678 switch (chip->chip_id) {
4679 case KVM_IRQCHIP_PIC_MASTER:
4680 memcpy(&chip->chip.pic, &pic->pics[0],
4681 sizeof(struct kvm_pic_state));
4683 case KVM_IRQCHIP_PIC_SLAVE:
4684 memcpy(&chip->chip.pic, &pic->pics[1],
4685 sizeof(struct kvm_pic_state));
4687 case KVM_IRQCHIP_IOAPIC:
4688 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4697 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4699 struct kvm_pic *pic = kvm->arch.vpic;
4703 switch (chip->chip_id) {
4704 case KVM_IRQCHIP_PIC_MASTER:
4705 spin_lock(&pic->lock);
4706 memcpy(&pic->pics[0], &chip->chip.pic,
4707 sizeof(struct kvm_pic_state));
4708 spin_unlock(&pic->lock);
4710 case KVM_IRQCHIP_PIC_SLAVE:
4711 spin_lock(&pic->lock);
4712 memcpy(&pic->pics[1], &chip->chip.pic,
4713 sizeof(struct kvm_pic_state));
4714 spin_unlock(&pic->lock);
4716 case KVM_IRQCHIP_IOAPIC:
4717 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4723 kvm_pic_update_irq(pic);
4727 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4729 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4731 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4733 mutex_lock(&kps->lock);
4734 memcpy(ps, &kps->channels, sizeof(*ps));
4735 mutex_unlock(&kps->lock);
4739 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4742 struct kvm_pit *pit = kvm->arch.vpit;
4744 mutex_lock(&pit->pit_state.lock);
4745 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4746 for (i = 0; i < 3; i++)
4747 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4748 mutex_unlock(&pit->pit_state.lock);
4752 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4754 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4755 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4756 sizeof(ps->channels));
4757 ps->flags = kvm->arch.vpit->pit_state.flags;
4758 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4759 memset(&ps->reserved, 0, sizeof(ps->reserved));
4763 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4767 u32 prev_legacy, cur_legacy;
4768 struct kvm_pit *pit = kvm->arch.vpit;
4770 mutex_lock(&pit->pit_state.lock);
4771 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4772 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4773 if (!prev_legacy && cur_legacy)
4775 memcpy(&pit->pit_state.channels, &ps->channels,
4776 sizeof(pit->pit_state.channels));
4777 pit->pit_state.flags = ps->flags;
4778 for (i = 0; i < 3; i++)
4779 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4781 mutex_unlock(&pit->pit_state.lock);
4785 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4786 struct kvm_reinject_control *control)
4788 struct kvm_pit *pit = kvm->arch.vpit;
4790 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4791 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4792 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4794 mutex_lock(&pit->pit_state.lock);
4795 kvm_pit_set_reinject(pit, control->pit_reinject);
4796 mutex_unlock(&pit->pit_state.lock);
4801 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
4804 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4806 if (kvm_x86_ops.flush_log_dirty)
4807 kvm_x86_ops.flush_log_dirty(kvm);
4810 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4813 if (!irqchip_in_kernel(kvm))
4816 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4817 irq_event->irq, irq_event->level,
4822 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4823 struct kvm_enable_cap *cap)
4831 case KVM_CAP_DISABLE_QUIRKS:
4832 kvm->arch.disabled_quirks = cap->args[0];
4835 case KVM_CAP_SPLIT_IRQCHIP: {
4836 mutex_lock(&kvm->lock);
4838 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4839 goto split_irqchip_unlock;
4841 if (irqchip_in_kernel(kvm))
4842 goto split_irqchip_unlock;
4843 if (kvm->created_vcpus)
4844 goto split_irqchip_unlock;
4845 r = kvm_setup_empty_irq_routing(kvm);
4847 goto split_irqchip_unlock;
4848 /* Pairs with irqchip_in_kernel. */
4850 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4851 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4853 split_irqchip_unlock:
4854 mutex_unlock(&kvm->lock);
4857 case KVM_CAP_X2APIC_API:
4859 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4862 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4863 kvm->arch.x2apic_format = true;
4864 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4865 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4869 case KVM_CAP_X86_DISABLE_EXITS:
4871 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4874 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4875 kvm_can_mwait_in_guest())
4876 kvm->arch.mwait_in_guest = true;
4877 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4878 kvm->arch.hlt_in_guest = true;
4879 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4880 kvm->arch.pause_in_guest = true;
4881 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4882 kvm->arch.cstate_in_guest = true;
4885 case KVM_CAP_MSR_PLATFORM_INFO:
4886 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4889 case KVM_CAP_EXCEPTION_PAYLOAD:
4890 kvm->arch.exception_payload_enabled = cap->args[0];
4900 long kvm_arch_vm_ioctl(struct file *filp,
4901 unsigned int ioctl, unsigned long arg)
4903 struct kvm *kvm = filp->private_data;
4904 void __user *argp = (void __user *)arg;
4907 * This union makes it completely explicit to gcc-3.x
4908 * that these two variables' stack usage should be
4909 * combined, not added together.
4912 struct kvm_pit_state ps;
4913 struct kvm_pit_state2 ps2;
4914 struct kvm_pit_config pit_config;
4918 case KVM_SET_TSS_ADDR:
4919 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4921 case KVM_SET_IDENTITY_MAP_ADDR: {
4924 mutex_lock(&kvm->lock);
4926 if (kvm->created_vcpus)
4927 goto set_identity_unlock;
4929 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4930 goto set_identity_unlock;
4931 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4932 set_identity_unlock:
4933 mutex_unlock(&kvm->lock);
4936 case KVM_SET_NR_MMU_PAGES:
4937 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4939 case KVM_GET_NR_MMU_PAGES:
4940 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4942 case KVM_CREATE_IRQCHIP: {
4943 mutex_lock(&kvm->lock);
4946 if (irqchip_in_kernel(kvm))
4947 goto create_irqchip_unlock;
4950 if (kvm->created_vcpus)
4951 goto create_irqchip_unlock;
4953 r = kvm_pic_init(kvm);
4955 goto create_irqchip_unlock;
4957 r = kvm_ioapic_init(kvm);
4959 kvm_pic_destroy(kvm);
4960 goto create_irqchip_unlock;
4963 r = kvm_setup_default_irq_routing(kvm);
4965 kvm_ioapic_destroy(kvm);
4966 kvm_pic_destroy(kvm);
4967 goto create_irqchip_unlock;
4969 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4971 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4972 create_irqchip_unlock:
4973 mutex_unlock(&kvm->lock);
4976 case KVM_CREATE_PIT:
4977 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4979 case KVM_CREATE_PIT2:
4981 if (copy_from_user(&u.pit_config, argp,
4982 sizeof(struct kvm_pit_config)))
4985 mutex_lock(&kvm->lock);
4988 goto create_pit_unlock;
4990 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4994 mutex_unlock(&kvm->lock);
4996 case KVM_GET_IRQCHIP: {
4997 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4998 struct kvm_irqchip *chip;
5000 chip = memdup_user(argp, sizeof(*chip));
5007 if (!irqchip_kernel(kvm))
5008 goto get_irqchip_out;
5009 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5011 goto get_irqchip_out;
5013 if (copy_to_user(argp, chip, sizeof(*chip)))
5014 goto get_irqchip_out;
5020 case KVM_SET_IRQCHIP: {
5021 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5022 struct kvm_irqchip *chip;
5024 chip = memdup_user(argp, sizeof(*chip));
5031 if (!irqchip_kernel(kvm))
5032 goto set_irqchip_out;
5033 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5040 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5043 if (!kvm->arch.vpit)
5045 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5049 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5056 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5058 mutex_lock(&kvm->lock);
5060 if (!kvm->arch.vpit)
5062 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5064 mutex_unlock(&kvm->lock);
5067 case KVM_GET_PIT2: {
5069 if (!kvm->arch.vpit)
5071 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5075 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5080 case KVM_SET_PIT2: {
5082 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5084 mutex_lock(&kvm->lock);
5086 if (!kvm->arch.vpit)
5088 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5090 mutex_unlock(&kvm->lock);
5093 case KVM_REINJECT_CONTROL: {
5094 struct kvm_reinject_control control;
5096 if (copy_from_user(&control, argp, sizeof(control)))
5099 if (!kvm->arch.vpit)
5101 r = kvm_vm_ioctl_reinject(kvm, &control);
5104 case KVM_SET_BOOT_CPU_ID:
5106 mutex_lock(&kvm->lock);
5107 if (kvm->created_vcpus)
5110 kvm->arch.bsp_vcpu_id = arg;
5111 mutex_unlock(&kvm->lock);
5113 case KVM_XEN_HVM_CONFIG: {
5114 struct kvm_xen_hvm_config xhc;
5116 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5121 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5125 case KVM_SET_CLOCK: {
5126 struct kvm_clock_data user_ns;
5130 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5139 * TODO: userspace has to take care of races with VCPU_RUN, so
5140 * kvm_gen_update_masterclock() can be cut down to locked
5141 * pvclock_update_vm_gtod_copy().
5143 kvm_gen_update_masterclock(kvm);
5144 now_ns = get_kvmclock_ns(kvm);
5145 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5146 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5149 case KVM_GET_CLOCK: {
5150 struct kvm_clock_data user_ns;
5153 now_ns = get_kvmclock_ns(kvm);
5154 user_ns.clock = now_ns;
5155 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5156 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5159 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5164 case KVM_MEMORY_ENCRYPT_OP: {
5166 if (kvm_x86_ops.mem_enc_op)
5167 r = kvm_x86_ops.mem_enc_op(kvm, argp);
5170 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5171 struct kvm_enc_region region;
5174 if (copy_from_user(®ion, argp, sizeof(region)))
5178 if (kvm_x86_ops.mem_enc_reg_region)
5179 r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion);
5182 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5183 struct kvm_enc_region region;
5186 if (copy_from_user(®ion, argp, sizeof(region)))
5190 if (kvm_x86_ops.mem_enc_unreg_region)
5191 r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion);
5194 case KVM_HYPERV_EVENTFD: {
5195 struct kvm_hyperv_eventfd hvevfd;
5198 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5200 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5203 case KVM_SET_PMU_EVENT_FILTER:
5204 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5213 static void kvm_init_msr_list(void)
5215 struct x86_pmu_capability x86_pmu;
5219 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5220 "Please update the fixed PMCs in msrs_to_saved_all[]");
5222 perf_get_x86_pmu_capability(&x86_pmu);
5224 num_msrs_to_save = 0;
5225 num_emulated_msrs = 0;
5226 num_msr_based_features = 0;
5228 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5229 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5233 * Even MSRs that are valid in the host may not be exposed
5234 * to the guests in some cases.
5236 switch (msrs_to_save_all[i]) {
5237 case MSR_IA32_BNDCFGS:
5238 if (!kvm_mpx_supported())
5242 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5245 case MSR_IA32_RTIT_CTL:
5246 case MSR_IA32_RTIT_STATUS:
5247 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5250 case MSR_IA32_RTIT_CR3_MATCH:
5251 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5252 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5255 case MSR_IA32_RTIT_OUTPUT_BASE:
5256 case MSR_IA32_RTIT_OUTPUT_MASK:
5257 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5258 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5259 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5262 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5263 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5264 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5265 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5268 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5269 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5270 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5273 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5274 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5275 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5282 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5285 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5286 if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5289 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5292 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5293 struct kvm_msr_entry msr;
5295 msr.index = msr_based_features_all[i];
5296 if (kvm_get_msr_feature(&msr))
5299 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5303 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5311 if (!(lapic_in_kernel(vcpu) &&
5312 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5313 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5324 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5331 if (!(lapic_in_kernel(vcpu) &&
5332 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5334 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5336 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5346 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5347 struct kvm_segment *var, int seg)
5349 kvm_x86_ops.set_segment(vcpu, var, seg);
5352 void kvm_get_segment(struct kvm_vcpu *vcpu,
5353 struct kvm_segment *var, int seg)
5355 kvm_x86_ops.get_segment(vcpu, var, seg);
5358 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5359 struct x86_exception *exception)
5363 BUG_ON(!mmu_is_nested(vcpu));
5365 /* NPT walks are always user-walks */
5366 access |= PFERR_USER_MASK;
5367 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5372 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5373 struct x86_exception *exception)
5375 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5376 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5379 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5380 struct x86_exception *exception)
5382 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5383 access |= PFERR_FETCH_MASK;
5384 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5387 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5388 struct x86_exception *exception)
5390 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5391 access |= PFERR_WRITE_MASK;
5392 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5395 /* uses this to access any guest's mapped memory without checking CPL */
5396 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5397 struct x86_exception *exception)
5399 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5402 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5403 struct kvm_vcpu *vcpu, u32 access,
5404 struct x86_exception *exception)
5407 int r = X86EMUL_CONTINUE;
5410 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5412 unsigned offset = addr & (PAGE_SIZE-1);
5413 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5416 if (gpa == UNMAPPED_GVA)
5417 return X86EMUL_PROPAGATE_FAULT;
5418 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5421 r = X86EMUL_IO_NEEDED;
5433 /* used for instruction fetching */
5434 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5435 gva_t addr, void *val, unsigned int bytes,
5436 struct x86_exception *exception)
5438 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5439 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5443 /* Inline kvm_read_guest_virt_helper for speed. */
5444 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5446 if (unlikely(gpa == UNMAPPED_GVA))
5447 return X86EMUL_PROPAGATE_FAULT;
5449 offset = addr & (PAGE_SIZE-1);
5450 if (WARN_ON(offset + bytes > PAGE_SIZE))
5451 bytes = (unsigned)PAGE_SIZE - offset;
5452 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5454 if (unlikely(ret < 0))
5455 return X86EMUL_IO_NEEDED;
5457 return X86EMUL_CONTINUE;
5460 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5461 gva_t addr, void *val, unsigned int bytes,
5462 struct x86_exception *exception)
5464 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5467 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5468 * is returned, but our callers are not ready for that and they blindly
5469 * call kvm_inject_page_fault. Ensure that they at least do not leak
5470 * uninitialized kernel stack memory into cr2 and error code.
5472 memset(exception, 0, sizeof(*exception));
5473 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5476 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5478 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5479 gva_t addr, void *val, unsigned int bytes,
5480 struct x86_exception *exception, bool system)
5482 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5485 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5486 access |= PFERR_USER_MASK;
5488 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5491 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5492 unsigned long addr, void *val, unsigned int bytes)
5494 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5495 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5497 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5500 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5501 struct kvm_vcpu *vcpu, u32 access,
5502 struct x86_exception *exception)
5505 int r = X86EMUL_CONTINUE;
5508 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5511 unsigned offset = addr & (PAGE_SIZE-1);
5512 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5515 if (gpa == UNMAPPED_GVA)
5516 return X86EMUL_PROPAGATE_FAULT;
5517 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5519 r = X86EMUL_IO_NEEDED;
5531 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5532 unsigned int bytes, struct x86_exception *exception,
5535 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5536 u32 access = PFERR_WRITE_MASK;
5538 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5539 access |= PFERR_USER_MASK;
5541 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5545 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5546 unsigned int bytes, struct x86_exception *exception)
5548 /* kvm_write_guest_virt_system can pull in tons of pages. */
5549 vcpu->arch.l1tf_flush_l1d = true;
5552 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5553 * is returned, but our callers are not ready for that and they blindly
5554 * call kvm_inject_page_fault. Ensure that they at least do not leak
5555 * uninitialized kernel stack memory into cr2 and error code.
5557 memset(exception, 0, sizeof(*exception));
5558 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5559 PFERR_WRITE_MASK, exception);
5561 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5563 int handle_ud(struct kvm_vcpu *vcpu)
5565 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5566 int emul_type = EMULTYPE_TRAP_UD;
5567 char sig[5]; /* ud2; .ascii "kvm" */
5568 struct x86_exception e;
5570 if (force_emulation_prefix &&
5571 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5572 sig, sizeof(sig), &e) == 0 &&
5573 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5574 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5575 emul_type = EMULTYPE_TRAP_UD_FORCED;
5578 return kvm_emulate_instruction(vcpu, emul_type);
5580 EXPORT_SYMBOL_GPL(handle_ud);
5582 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5583 gpa_t gpa, bool write)
5585 /* For APIC access vmexit */
5586 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5589 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5590 trace_vcpu_match_mmio(gva, gpa, write, true);
5597 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5598 gpa_t *gpa, struct x86_exception *exception,
5601 u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5602 | (write ? PFERR_WRITE_MASK : 0);
5605 * currently PKRU is only applied to ept enabled guest so
5606 * there is no pkey in EPT page table for L1 guest or EPT
5607 * shadow page table for L2 guest.
5609 if (vcpu_match_mmio_gva(vcpu, gva)
5610 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5611 vcpu->arch.mmio_access, 0, access)) {
5612 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5613 (gva & (PAGE_SIZE - 1));
5614 trace_vcpu_match_mmio(gva, *gpa, write, false);
5618 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5620 if (*gpa == UNMAPPED_GVA)
5623 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5626 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5627 const void *val, int bytes)
5631 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5634 kvm_page_track_write(vcpu, gpa, val, bytes);
5638 struct read_write_emulator_ops {
5639 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5641 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5642 void *val, int bytes);
5643 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5644 int bytes, void *val);
5645 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5646 void *val, int bytes);
5650 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5652 if (vcpu->mmio_read_completed) {
5653 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5654 vcpu->mmio_fragments[0].gpa, val);
5655 vcpu->mmio_read_completed = 0;
5662 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5663 void *val, int bytes)
5665 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5668 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5669 void *val, int bytes)
5671 return emulator_write_phys(vcpu, gpa, val, bytes);
5674 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5676 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5677 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5680 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5681 void *val, int bytes)
5683 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5684 return X86EMUL_IO_NEEDED;
5687 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5688 void *val, int bytes)
5690 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5692 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5693 return X86EMUL_CONTINUE;
5696 static const struct read_write_emulator_ops read_emultor = {
5697 .read_write_prepare = read_prepare,
5698 .read_write_emulate = read_emulate,
5699 .read_write_mmio = vcpu_mmio_read,
5700 .read_write_exit_mmio = read_exit_mmio,
5703 static const struct read_write_emulator_ops write_emultor = {
5704 .read_write_emulate = write_emulate,
5705 .read_write_mmio = write_mmio,
5706 .read_write_exit_mmio = write_exit_mmio,
5710 static int emulator_read_write_onepage(unsigned long addr, void *val,
5712 struct x86_exception *exception,
5713 struct kvm_vcpu *vcpu,
5714 const struct read_write_emulator_ops *ops)
5718 bool write = ops->write;
5719 struct kvm_mmio_fragment *frag;
5720 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
5723 * If the exit was due to a NPF we may already have a GPA.
5724 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5725 * Note, this cannot be used on string operations since string
5726 * operation using rep will only have the initial GPA from the NPF
5729 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
5730 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
5731 gpa = ctxt->gpa_val;
5732 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5734 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5736 return X86EMUL_PROPAGATE_FAULT;
5739 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5740 return X86EMUL_CONTINUE;
5743 * Is this MMIO handled locally?
5745 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5746 if (handled == bytes)
5747 return X86EMUL_CONTINUE;
5753 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5754 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5758 return X86EMUL_CONTINUE;
5761 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5763 void *val, unsigned int bytes,
5764 struct x86_exception *exception,
5765 const struct read_write_emulator_ops *ops)
5767 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5771 if (ops->read_write_prepare &&
5772 ops->read_write_prepare(vcpu, val, bytes))
5773 return X86EMUL_CONTINUE;
5775 vcpu->mmio_nr_fragments = 0;
5777 /* Crossing a page boundary? */
5778 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5781 now = -addr & ~PAGE_MASK;
5782 rc = emulator_read_write_onepage(addr, val, now, exception,
5785 if (rc != X86EMUL_CONTINUE)
5788 if (ctxt->mode != X86EMUL_MODE_PROT64)
5794 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5796 if (rc != X86EMUL_CONTINUE)
5799 if (!vcpu->mmio_nr_fragments)
5802 gpa = vcpu->mmio_fragments[0].gpa;
5804 vcpu->mmio_needed = 1;
5805 vcpu->mmio_cur_fragment = 0;
5807 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5808 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5809 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5810 vcpu->run->mmio.phys_addr = gpa;
5812 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5815 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5819 struct x86_exception *exception)
5821 return emulator_read_write(ctxt, addr, val, bytes,
5822 exception, &read_emultor);
5825 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5829 struct x86_exception *exception)
5831 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5832 exception, &write_emultor);
5835 #define CMPXCHG_TYPE(t, ptr, old, new) \
5836 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5838 #ifdef CONFIG_X86_64
5839 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5841 # define CMPXCHG64(ptr, old, new) \
5842 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5845 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5850 struct x86_exception *exception)
5852 struct kvm_host_map map;
5853 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5859 /* guests cmpxchg8b have to be emulated atomically */
5860 if (bytes > 8 || (bytes & (bytes - 1)))
5863 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5865 if (gpa == UNMAPPED_GVA ||
5866 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5870 * Emulate the atomic as a straight write to avoid #AC if SLD is
5871 * enabled in the host and the access splits a cache line.
5873 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
5874 page_line_mask = ~(cache_line_size() - 1);
5876 page_line_mask = PAGE_MASK;
5878 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
5881 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5884 kaddr = map.hva + offset_in_page(gpa);
5888 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5891 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5894 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5897 exchanged = CMPXCHG64(kaddr, old, new);
5903 kvm_vcpu_unmap(vcpu, &map, true);
5906 return X86EMUL_CMPXCHG_FAILED;
5908 kvm_page_track_write(vcpu, gpa, new, bytes);
5910 return X86EMUL_CONTINUE;
5913 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5915 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5918 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5922 for (i = 0; i < vcpu->arch.pio.count; i++) {
5923 if (vcpu->arch.pio.in)
5924 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5925 vcpu->arch.pio.size, pd);
5927 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5928 vcpu->arch.pio.port, vcpu->arch.pio.size,
5932 pd += vcpu->arch.pio.size;
5937 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5938 unsigned short port, void *val,
5939 unsigned int count, bool in)
5941 vcpu->arch.pio.port = port;
5942 vcpu->arch.pio.in = in;
5943 vcpu->arch.pio.count = count;
5944 vcpu->arch.pio.size = size;
5946 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5947 vcpu->arch.pio.count = 0;
5951 vcpu->run->exit_reason = KVM_EXIT_IO;
5952 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5953 vcpu->run->io.size = size;
5954 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5955 vcpu->run->io.count = count;
5956 vcpu->run->io.port = port;
5961 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
5962 unsigned short port, void *val, unsigned int count)
5966 if (vcpu->arch.pio.count)
5969 memset(vcpu->arch.pio_data, 0, size * count);
5971 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5974 memcpy(val, vcpu->arch.pio_data, size * count);
5975 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5976 vcpu->arch.pio.count = 0;
5983 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5984 int size, unsigned short port, void *val,
5987 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
5991 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
5992 unsigned short port, const void *val,
5995 memcpy(vcpu->arch.pio_data, val, size * count);
5996 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5997 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6000 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6001 int size, unsigned short port,
6002 const void *val, unsigned int count)
6004 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6007 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6009 return kvm_x86_ops.get_segment_base(vcpu, seg);
6012 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6014 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6017 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6019 if (!need_emulate_wbinvd(vcpu))
6020 return X86EMUL_CONTINUE;
6022 if (kvm_x86_ops.has_wbinvd_exit()) {
6023 int cpu = get_cpu();
6025 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6026 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6027 wbinvd_ipi, NULL, 1);
6029 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6032 return X86EMUL_CONTINUE;
6035 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6037 kvm_emulate_wbinvd_noskip(vcpu);
6038 return kvm_skip_emulated_instruction(vcpu);
6040 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6044 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6046 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6049 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6050 unsigned long *dest)
6052 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6055 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6056 unsigned long value)
6059 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6062 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6064 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6067 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6069 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6070 unsigned long value;
6074 value = kvm_read_cr0(vcpu);
6077 value = vcpu->arch.cr2;
6080 value = kvm_read_cr3(vcpu);
6083 value = kvm_read_cr4(vcpu);
6086 value = kvm_get_cr8(vcpu);
6089 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6096 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6098 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6103 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6106 vcpu->arch.cr2 = val;
6109 res = kvm_set_cr3(vcpu, val);
6112 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6115 res = kvm_set_cr8(vcpu, val);
6118 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6125 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6127 return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6130 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6132 kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6135 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6137 kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6140 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6142 kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6145 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6147 kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6150 static unsigned long emulator_get_cached_segment_base(
6151 struct x86_emulate_ctxt *ctxt, int seg)
6153 return get_segment_base(emul_to_vcpu(ctxt), seg);
6156 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6157 struct desc_struct *desc, u32 *base3,
6160 struct kvm_segment var;
6162 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6163 *selector = var.selector;
6166 memset(desc, 0, sizeof(*desc));
6174 set_desc_limit(desc, var.limit);
6175 set_desc_base(desc, (unsigned long)var.base);
6176 #ifdef CONFIG_X86_64
6178 *base3 = var.base >> 32;
6180 desc->type = var.type;
6182 desc->dpl = var.dpl;
6183 desc->p = var.present;
6184 desc->avl = var.avl;
6192 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6193 struct desc_struct *desc, u32 base3,
6196 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6197 struct kvm_segment var;
6199 var.selector = selector;
6200 var.base = get_desc_base(desc);
6201 #ifdef CONFIG_X86_64
6202 var.base |= ((u64)base3) << 32;
6204 var.limit = get_desc_limit(desc);
6206 var.limit = (var.limit << 12) | 0xfff;
6207 var.type = desc->type;
6208 var.dpl = desc->dpl;
6213 var.avl = desc->avl;
6214 var.present = desc->p;
6215 var.unusable = !var.present;
6218 kvm_set_segment(vcpu, &var, seg);
6222 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6223 u32 msr_index, u64 *pdata)
6225 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6228 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6229 u32 msr_index, u64 data)
6231 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6234 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6236 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6238 return vcpu->arch.smbase;
6241 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6243 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6245 vcpu->arch.smbase = smbase;
6248 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6251 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6254 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6255 u32 pmc, u64 *pdata)
6257 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6260 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6262 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6265 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6266 struct x86_instruction_info *info,
6267 enum x86_intercept_stage stage)
6269 return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6273 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6274 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6277 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6280 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6282 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6285 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6287 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6290 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6292 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6295 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6297 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6300 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6302 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6305 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6307 kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6310 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6312 return emul_to_vcpu(ctxt)->arch.hflags;
6315 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6317 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6320 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6321 const char *smstate)
6323 return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6326 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6328 kvm_smm_changed(emul_to_vcpu(ctxt));
6331 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6333 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6336 static const struct x86_emulate_ops emulate_ops = {
6337 .read_gpr = emulator_read_gpr,
6338 .write_gpr = emulator_write_gpr,
6339 .read_std = emulator_read_std,
6340 .write_std = emulator_write_std,
6341 .read_phys = kvm_read_guest_phys_system,
6342 .fetch = kvm_fetch_guest_virt,
6343 .read_emulated = emulator_read_emulated,
6344 .write_emulated = emulator_write_emulated,
6345 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6346 .invlpg = emulator_invlpg,
6347 .pio_in_emulated = emulator_pio_in_emulated,
6348 .pio_out_emulated = emulator_pio_out_emulated,
6349 .get_segment = emulator_get_segment,
6350 .set_segment = emulator_set_segment,
6351 .get_cached_segment_base = emulator_get_cached_segment_base,
6352 .get_gdt = emulator_get_gdt,
6353 .get_idt = emulator_get_idt,
6354 .set_gdt = emulator_set_gdt,
6355 .set_idt = emulator_set_idt,
6356 .get_cr = emulator_get_cr,
6357 .set_cr = emulator_set_cr,
6358 .cpl = emulator_get_cpl,
6359 .get_dr = emulator_get_dr,
6360 .set_dr = emulator_set_dr,
6361 .get_smbase = emulator_get_smbase,
6362 .set_smbase = emulator_set_smbase,
6363 .set_msr = emulator_set_msr,
6364 .get_msr = emulator_get_msr,
6365 .check_pmc = emulator_check_pmc,
6366 .read_pmc = emulator_read_pmc,
6367 .halt = emulator_halt,
6368 .wbinvd = emulator_wbinvd,
6369 .fix_hypercall = emulator_fix_hypercall,
6370 .intercept = emulator_intercept,
6371 .get_cpuid = emulator_get_cpuid,
6372 .guest_has_long_mode = emulator_guest_has_long_mode,
6373 .guest_has_movbe = emulator_guest_has_movbe,
6374 .guest_has_fxsr = emulator_guest_has_fxsr,
6375 .set_nmi_mask = emulator_set_nmi_mask,
6376 .get_hflags = emulator_get_hflags,
6377 .set_hflags = emulator_set_hflags,
6378 .pre_leave_smm = emulator_pre_leave_smm,
6379 .post_leave_smm = emulator_post_leave_smm,
6380 .set_xcr = emulator_set_xcr,
6383 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6385 u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6387 * an sti; sti; sequence only disable interrupts for the first
6388 * instruction. So, if the last instruction, be it emulated or
6389 * not, left the system with the INT_STI flag enabled, it
6390 * means that the last instruction is an sti. We should not
6391 * leave the flag on in this case. The same goes for mov ss
6393 if (int_shadow & mask)
6395 if (unlikely(int_shadow || mask)) {
6396 kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6398 kvm_make_request(KVM_REQ_EVENT, vcpu);
6402 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6404 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6405 if (ctxt->exception.vector == PF_VECTOR)
6406 return kvm_propagate_fault(vcpu, &ctxt->exception);
6408 if (ctxt->exception.error_code_valid)
6409 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6410 ctxt->exception.error_code);
6412 kvm_queue_exception(vcpu, ctxt->exception.vector);
6416 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6418 struct x86_emulate_ctxt *ctxt;
6420 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6422 pr_err("kvm: failed to allocate vcpu's emulator\n");
6427 ctxt->ops = &emulate_ops;
6428 vcpu->arch.emulate_ctxt = ctxt;
6433 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6435 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6438 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6440 ctxt->gpa_available = false;
6441 ctxt->eflags = kvm_get_rflags(vcpu);
6442 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6444 ctxt->eip = kvm_rip_read(vcpu);
6445 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6446 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6447 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6448 cs_db ? X86EMUL_MODE_PROT32 :
6449 X86EMUL_MODE_PROT16;
6450 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6451 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6452 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6454 init_decode_cache(ctxt);
6455 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6458 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6460 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6463 init_emulate_ctxt(vcpu);
6467 ctxt->_eip = ctxt->eip + inc_eip;
6468 ret = emulate_int_real(ctxt, irq);
6470 if (ret != X86EMUL_CONTINUE) {
6471 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6473 ctxt->eip = ctxt->_eip;
6474 kvm_rip_write(vcpu, ctxt->eip);
6475 kvm_set_rflags(vcpu, ctxt->eflags);
6478 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6480 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6482 ++vcpu->stat.insn_emulation_fail;
6483 trace_kvm_emulate_insn_failed(vcpu);
6485 if (emulation_type & EMULTYPE_VMWARE_GP) {
6486 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6490 if (emulation_type & EMULTYPE_SKIP) {
6491 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6492 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6493 vcpu->run->internal.ndata = 0;
6497 kvm_queue_exception(vcpu, UD_VECTOR);
6499 if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
6500 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6501 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6502 vcpu->run->internal.ndata = 0;
6509 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6510 bool write_fault_to_shadow_pgtable,
6513 gpa_t gpa = cr2_or_gpa;
6516 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6519 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6520 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6523 if (!vcpu->arch.mmu->direct_map) {
6525 * Write permission should be allowed since only
6526 * write access need to be emulated.
6528 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6531 * If the mapping is invalid in guest, let cpu retry
6532 * it to generate fault.
6534 if (gpa == UNMAPPED_GVA)
6539 * Do not retry the unhandleable instruction if it faults on the
6540 * readonly host memory, otherwise it will goto a infinite loop:
6541 * retry instruction -> write #PF -> emulation fail -> retry
6542 * instruction -> ...
6544 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6547 * If the instruction failed on the error pfn, it can not be fixed,
6548 * report the error to userspace.
6550 if (is_error_noslot_pfn(pfn))
6553 kvm_release_pfn_clean(pfn);
6555 /* The instructions are well-emulated on direct mmu. */
6556 if (vcpu->arch.mmu->direct_map) {
6557 unsigned int indirect_shadow_pages;
6559 spin_lock(&vcpu->kvm->mmu_lock);
6560 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6561 spin_unlock(&vcpu->kvm->mmu_lock);
6563 if (indirect_shadow_pages)
6564 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6570 * if emulation was due to access to shadowed page table
6571 * and it failed try to unshadow page and re-enter the
6572 * guest to let CPU execute the instruction.
6574 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6577 * If the access faults on its page table, it can not
6578 * be fixed by unprotecting shadow page and it should
6579 * be reported to userspace.
6581 return !write_fault_to_shadow_pgtable;
6584 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6585 gpa_t cr2_or_gpa, int emulation_type)
6587 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6588 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
6590 last_retry_eip = vcpu->arch.last_retry_eip;
6591 last_retry_addr = vcpu->arch.last_retry_addr;
6594 * If the emulation is caused by #PF and it is non-page_table
6595 * writing instruction, it means the VM-EXIT is caused by shadow
6596 * page protected, we can zap the shadow page and retry this
6597 * instruction directly.
6599 * Note: if the guest uses a non-page-table modifying instruction
6600 * on the PDE that points to the instruction, then we will unmap
6601 * the instruction and go to an infinite loop. So, we cache the
6602 * last retried eip and the last fault address, if we meet the eip
6603 * and the address again, we can break out of the potential infinite
6606 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6608 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
6611 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
6612 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
6615 if (x86_page_table_writing_insn(ctxt))
6618 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
6621 vcpu->arch.last_retry_eip = ctxt->eip;
6622 vcpu->arch.last_retry_addr = cr2_or_gpa;
6624 if (!vcpu->arch.mmu->direct_map)
6625 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6627 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6632 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6633 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6635 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6637 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6638 /* This is a good place to trace that we are exiting SMM. */
6639 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6641 /* Process a latched INIT or SMI, if any. */
6642 kvm_make_request(KVM_REQ_EVENT, vcpu);
6645 kvm_mmu_reset_context(vcpu);
6648 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6657 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6658 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6663 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6665 struct kvm_run *kvm_run = vcpu->run;
6667 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6668 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6669 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6670 kvm_run->debug.arch.exception = DB_VECTOR;
6671 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6674 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6678 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6680 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6683 r = kvm_x86_ops.skip_emulated_instruction(vcpu);
6688 * rflags is the old, "raw" value of the flags. The new value has
6689 * not been saved yet.
6691 * This is correct even for TF set by the guest, because "the
6692 * processor will not generate this exception after the instruction
6693 * that sets the TF flag".
6695 if (unlikely(rflags & X86_EFLAGS_TF))
6696 r = kvm_vcpu_do_singlestep(vcpu);
6699 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6701 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6703 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6704 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6705 struct kvm_run *kvm_run = vcpu->run;
6706 unsigned long eip = kvm_get_linear_rip(vcpu);
6707 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6708 vcpu->arch.guest_debug_dr7,
6712 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6713 kvm_run->debug.arch.pc = eip;
6714 kvm_run->debug.arch.exception = DB_VECTOR;
6715 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6721 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6722 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6723 unsigned long eip = kvm_get_linear_rip(vcpu);
6724 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6729 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
6738 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6740 switch (ctxt->opcode_len) {
6747 case 0xe6: /* OUT */
6751 case 0x6c: /* INS */
6753 case 0x6e: /* OUTS */
6760 case 0x33: /* RDPMC */
6769 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6770 int emulation_type, void *insn, int insn_len)
6773 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6774 bool writeback = true;
6775 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6777 vcpu->arch.l1tf_flush_l1d = true;
6780 * Clear write_fault_to_shadow_pgtable here to ensure it is
6783 vcpu->arch.write_fault_to_shadow_pgtable = false;
6784 kvm_clear_exception_queue(vcpu);
6786 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6787 init_emulate_ctxt(vcpu);
6790 * We will reenter on the same instruction since
6791 * we do not set complete_userspace_io. This does not
6792 * handle watchpoints yet, those would be handled in
6795 if (!(emulation_type & EMULTYPE_SKIP) &&
6796 kvm_vcpu_check_breakpoint(vcpu, &r))
6799 ctxt->interruptibility = 0;
6800 ctxt->have_exception = false;
6801 ctxt->exception.vector = -1;
6802 ctxt->perm_ok = false;
6804 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6806 r = x86_decode_insn(ctxt, insn, insn_len);
6808 trace_kvm_emulate_insn_start(vcpu);
6809 ++vcpu->stat.insn_emulation;
6810 if (r != EMULATION_OK) {
6811 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6812 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6813 kvm_queue_exception(vcpu, UD_VECTOR);
6816 if (reexecute_instruction(vcpu, cr2_or_gpa,
6820 if (ctxt->have_exception) {
6822 * #UD should result in just EMULATION_FAILED, and trap-like
6823 * exception should not be encountered during decode.
6825 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6826 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6827 inject_emulated_exception(vcpu);
6830 return handle_emulation_failure(vcpu, emulation_type);
6834 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6835 !is_vmware_backdoor_opcode(ctxt)) {
6836 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6841 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6842 * for kvm_skip_emulated_instruction(). The caller is responsible for
6843 * updating interruptibility state and injecting single-step #DBs.
6845 if (emulation_type & EMULTYPE_SKIP) {
6846 kvm_rip_write(vcpu, ctxt->_eip);
6847 if (ctxt->eflags & X86_EFLAGS_RF)
6848 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6852 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
6855 /* this is needed for vmware backdoor interface to work since it
6856 changes registers values during IO operation */
6857 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6858 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6859 emulator_invalidate_register_cache(ctxt);
6863 if (emulation_type & EMULTYPE_PF) {
6864 /* Save the faulting GPA (cr2) in the address field */
6865 ctxt->exception.address = cr2_or_gpa;
6867 /* With shadow page tables, cr2 contains a GVA or nGPA. */
6868 if (vcpu->arch.mmu->direct_map) {
6869 ctxt->gpa_available = true;
6870 ctxt->gpa_val = cr2_or_gpa;
6873 /* Sanitize the address out of an abundance of paranoia. */
6874 ctxt->exception.address = 0;
6877 r = x86_emulate_insn(ctxt);
6879 if (r == EMULATION_INTERCEPTED)
6882 if (r == EMULATION_FAILED) {
6883 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
6887 return handle_emulation_failure(vcpu, emulation_type);
6890 if (ctxt->have_exception) {
6892 if (inject_emulated_exception(vcpu))
6894 } else if (vcpu->arch.pio.count) {
6895 if (!vcpu->arch.pio.in) {
6896 /* FIXME: return into emulator if single-stepping. */
6897 vcpu->arch.pio.count = 0;
6900 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6903 } else if (vcpu->mmio_needed) {
6904 ++vcpu->stat.mmio_exits;
6906 if (!vcpu->mmio_is_write)
6909 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6910 } else if (r == EMULATION_RESTART)
6916 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
6917 toggle_interruptibility(vcpu, ctxt->interruptibility);
6918 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6919 if (!ctxt->have_exception ||
6920 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6921 kvm_rip_write(vcpu, ctxt->eip);
6923 r = kvm_vcpu_do_singlestep(vcpu);
6924 if (kvm_x86_ops.update_emulated_instruction)
6925 kvm_x86_ops.update_emulated_instruction(vcpu);
6926 __kvm_set_rflags(vcpu, ctxt->eflags);
6930 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6931 * do nothing, and it will be requested again as soon as
6932 * the shadow expires. But we still need to check here,
6933 * because POPF has no interrupt shadow.
6935 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6936 kvm_make_request(KVM_REQ_EVENT, vcpu);
6938 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6943 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6945 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6947 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6949 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6950 void *insn, int insn_len)
6952 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6954 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6956 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6958 vcpu->arch.pio.count = 0;
6962 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6964 vcpu->arch.pio.count = 0;
6966 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6969 return kvm_skip_emulated_instruction(vcpu);
6972 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6973 unsigned short port)
6975 unsigned long val = kvm_rax_read(vcpu);
6976 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
6982 * Workaround userspace that relies on old KVM behavior of %rip being
6983 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6986 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6987 vcpu->arch.complete_userspace_io =
6988 complete_fast_pio_out_port_0x7e;
6989 kvm_skip_emulated_instruction(vcpu);
6991 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6992 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6997 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7001 /* We should only ever be called with arch.pio.count equal to 1 */
7002 BUG_ON(vcpu->arch.pio.count != 1);
7004 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7005 vcpu->arch.pio.count = 0;
7009 /* For size less than 4 we merge, else we zero extend */
7010 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7013 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7014 * the copy and tracing
7016 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7017 kvm_rax_write(vcpu, val);
7019 return kvm_skip_emulated_instruction(vcpu);
7022 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7023 unsigned short port)
7028 /* For size less than 4 we merge, else we zero extend */
7029 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7031 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7033 kvm_rax_write(vcpu, val);
7037 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7038 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7043 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7048 ret = kvm_fast_pio_in(vcpu, size, port);
7050 ret = kvm_fast_pio_out(vcpu, size, port);
7051 return ret && kvm_skip_emulated_instruction(vcpu);
7053 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7055 static int kvmclock_cpu_down_prep(unsigned int cpu)
7057 __this_cpu_write(cpu_tsc_khz, 0);
7061 static void tsc_khz_changed(void *data)
7063 struct cpufreq_freqs *freq = data;
7064 unsigned long khz = 0;
7068 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7069 khz = cpufreq_quick_get(raw_smp_processor_id());
7072 __this_cpu_write(cpu_tsc_khz, khz);
7075 #ifdef CONFIG_X86_64
7076 static void kvm_hyperv_tsc_notifier(void)
7079 struct kvm_vcpu *vcpu;
7082 mutex_lock(&kvm_lock);
7083 list_for_each_entry(kvm, &vm_list, vm_list)
7084 kvm_make_mclock_inprogress_request(kvm);
7086 hyperv_stop_tsc_emulation();
7088 /* TSC frequency always matches when on Hyper-V */
7089 for_each_present_cpu(cpu)
7090 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7091 kvm_max_guest_tsc_khz = tsc_khz;
7093 list_for_each_entry(kvm, &vm_list, vm_list) {
7094 struct kvm_arch *ka = &kvm->arch;
7096 spin_lock(&ka->pvclock_gtod_sync_lock);
7098 pvclock_update_vm_gtod_copy(kvm);
7100 kvm_for_each_vcpu(cpu, vcpu, kvm)
7101 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7103 kvm_for_each_vcpu(cpu, vcpu, kvm)
7104 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7106 spin_unlock(&ka->pvclock_gtod_sync_lock);
7108 mutex_unlock(&kvm_lock);
7112 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7115 struct kvm_vcpu *vcpu;
7116 int i, send_ipi = 0;
7119 * We allow guests to temporarily run on slowing clocks,
7120 * provided we notify them after, or to run on accelerating
7121 * clocks, provided we notify them before. Thus time never
7124 * However, we have a problem. We can't atomically update
7125 * the frequency of a given CPU from this function; it is
7126 * merely a notifier, which can be called from any CPU.
7127 * Changing the TSC frequency at arbitrary points in time
7128 * requires a recomputation of local variables related to
7129 * the TSC for each VCPU. We must flag these local variables
7130 * to be updated and be sure the update takes place with the
7131 * new frequency before any guests proceed.
7133 * Unfortunately, the combination of hotplug CPU and frequency
7134 * change creates an intractable locking scenario; the order
7135 * of when these callouts happen is undefined with respect to
7136 * CPU hotplug, and they can race with each other. As such,
7137 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7138 * undefined; you can actually have a CPU frequency change take
7139 * place in between the computation of X and the setting of the
7140 * variable. To protect against this problem, all updates of
7141 * the per_cpu tsc_khz variable are done in an interrupt
7142 * protected IPI, and all callers wishing to update the value
7143 * must wait for a synchronous IPI to complete (which is trivial
7144 * if the caller is on the CPU already). This establishes the
7145 * necessary total order on variable updates.
7147 * Note that because a guest time update may take place
7148 * anytime after the setting of the VCPU's request bit, the
7149 * correct TSC value must be set before the request. However,
7150 * to ensure the update actually makes it to any guest which
7151 * starts running in hardware virtualization between the set
7152 * and the acquisition of the spinlock, we must also ping the
7153 * CPU after setting the request bit.
7157 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7159 mutex_lock(&kvm_lock);
7160 list_for_each_entry(kvm, &vm_list, vm_list) {
7161 kvm_for_each_vcpu(i, vcpu, kvm) {
7162 if (vcpu->cpu != cpu)
7164 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7165 if (vcpu->cpu != raw_smp_processor_id())
7169 mutex_unlock(&kvm_lock);
7171 if (freq->old < freq->new && send_ipi) {
7173 * We upscale the frequency. Must make the guest
7174 * doesn't see old kvmclock values while running with
7175 * the new frequency, otherwise we risk the guest sees
7176 * time go backwards.
7178 * In case we update the frequency for another cpu
7179 * (which might be in guest context) send an interrupt
7180 * to kick the cpu out of guest context. Next time
7181 * guest context is entered kvmclock will be updated,
7182 * so the guest will not see stale values.
7184 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7188 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7191 struct cpufreq_freqs *freq = data;
7194 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7196 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7199 for_each_cpu(cpu, freq->policy->cpus)
7200 __kvmclock_cpufreq_notifier(freq, cpu);
7205 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7206 .notifier_call = kvmclock_cpufreq_notifier
7209 static int kvmclock_cpu_online(unsigned int cpu)
7211 tsc_khz_changed(NULL);
7215 static void kvm_timer_init(void)
7217 max_tsc_khz = tsc_khz;
7219 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7220 #ifdef CONFIG_CPU_FREQ
7221 struct cpufreq_policy *policy;
7225 policy = cpufreq_cpu_get(cpu);
7227 if (policy->cpuinfo.max_freq)
7228 max_tsc_khz = policy->cpuinfo.max_freq;
7229 cpufreq_cpu_put(policy);
7233 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7234 CPUFREQ_TRANSITION_NOTIFIER);
7237 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7238 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7241 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7242 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7244 int kvm_is_in_guest(void)
7246 return __this_cpu_read(current_vcpu) != NULL;
7249 static int kvm_is_user_mode(void)
7253 if (__this_cpu_read(current_vcpu))
7254 user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7256 return user_mode != 0;
7259 static unsigned long kvm_get_guest_ip(void)
7261 unsigned long ip = 0;
7263 if (__this_cpu_read(current_vcpu))
7264 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7269 static void kvm_handle_intel_pt_intr(void)
7271 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7273 kvm_make_request(KVM_REQ_PMI, vcpu);
7274 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7275 (unsigned long *)&vcpu->arch.pmu.global_status);
7278 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7279 .is_in_guest = kvm_is_in_guest,
7280 .is_user_mode = kvm_is_user_mode,
7281 .get_guest_ip = kvm_get_guest_ip,
7282 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7285 #ifdef CONFIG_X86_64
7286 static void pvclock_gtod_update_fn(struct work_struct *work)
7290 struct kvm_vcpu *vcpu;
7293 mutex_lock(&kvm_lock);
7294 list_for_each_entry(kvm, &vm_list, vm_list)
7295 kvm_for_each_vcpu(i, vcpu, kvm)
7296 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7297 atomic_set(&kvm_guest_has_master_clock, 0);
7298 mutex_unlock(&kvm_lock);
7301 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7304 * Notification about pvclock gtod data update.
7306 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7309 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7310 struct timekeeper *tk = priv;
7312 update_pvclock_gtod(tk);
7314 /* disable master clock if host does not trust, or does not
7315 * use, TSC based clocksource.
7317 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7318 atomic_read(&kvm_guest_has_master_clock) != 0)
7319 queue_work(system_long_wq, &pvclock_gtod_work);
7324 static struct notifier_block pvclock_gtod_notifier = {
7325 .notifier_call = pvclock_gtod_notify,
7329 int kvm_arch_init(void *opaque)
7331 struct kvm_x86_init_ops *ops = opaque;
7334 if (kvm_x86_ops.hardware_enable) {
7335 printk(KERN_ERR "kvm: already loaded the other module\n");
7340 if (!ops->cpu_has_kvm_support()) {
7341 pr_err_ratelimited("kvm: no hardware support\n");
7345 if (ops->disabled_by_bios()) {
7346 pr_err_ratelimited("kvm: disabled by bios\n");
7352 * KVM explicitly assumes that the guest has an FPU and
7353 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7354 * vCPU's FPU state as a fxregs_state struct.
7356 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7357 printk(KERN_ERR "kvm: inadequate fpu\n");
7363 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7364 __alignof__(struct fpu), SLAB_ACCOUNT,
7366 if (!x86_fpu_cache) {
7367 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7371 x86_emulator_cache = kvm_alloc_emulator_cache();
7372 if (!x86_emulator_cache) {
7373 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7374 goto out_free_x86_fpu_cache;
7377 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7379 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7380 goto out_free_x86_emulator_cache;
7383 r = kvm_mmu_module_init();
7385 goto out_free_percpu;
7387 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7388 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7389 PT_PRESENT_MASK, 0, sme_me_mask);
7392 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7394 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7395 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7396 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7400 if (pi_inject_timer == -1)
7401 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7402 #ifdef CONFIG_X86_64
7403 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7405 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7406 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7412 free_percpu(shared_msrs);
7413 out_free_x86_emulator_cache:
7414 kmem_cache_destroy(x86_emulator_cache);
7415 out_free_x86_fpu_cache:
7416 kmem_cache_destroy(x86_fpu_cache);
7421 void kvm_arch_exit(void)
7423 #ifdef CONFIG_X86_64
7424 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7425 clear_hv_tscchange_cb();
7428 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7430 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7431 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7432 CPUFREQ_TRANSITION_NOTIFIER);
7433 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7434 #ifdef CONFIG_X86_64
7435 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7437 kvm_x86_ops.hardware_enable = NULL;
7438 kvm_mmu_module_exit();
7439 free_percpu(shared_msrs);
7440 kmem_cache_destroy(x86_fpu_cache);
7443 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7445 ++vcpu->stat.halt_exits;
7446 if (lapic_in_kernel(vcpu)) {
7447 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7450 vcpu->run->exit_reason = KVM_EXIT_HLT;
7454 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7456 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7458 int ret = kvm_skip_emulated_instruction(vcpu);
7460 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7461 * KVM_EXIT_DEBUG here.
7463 return kvm_vcpu_halt(vcpu) && ret;
7465 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7467 #ifdef CONFIG_X86_64
7468 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7469 unsigned long clock_type)
7471 struct kvm_clock_pairing clock_pairing;
7472 struct timespec64 ts;
7476 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7477 return -KVM_EOPNOTSUPP;
7479 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7480 return -KVM_EOPNOTSUPP;
7482 clock_pairing.sec = ts.tv_sec;
7483 clock_pairing.nsec = ts.tv_nsec;
7484 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7485 clock_pairing.flags = 0;
7486 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7489 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7490 sizeof(struct kvm_clock_pairing)))
7498 * kvm_pv_kick_cpu_op: Kick a vcpu.
7500 * @apicid - apicid of vcpu to be kicked.
7502 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7504 struct kvm_lapic_irq lapic_irq;
7506 lapic_irq.shorthand = APIC_DEST_NOSHORT;
7507 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7508 lapic_irq.level = 0;
7509 lapic_irq.dest_id = apicid;
7510 lapic_irq.msi_redir_hint = false;
7512 lapic_irq.delivery_mode = APIC_DM_REMRD;
7513 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7516 bool kvm_apicv_activated(struct kvm *kvm)
7518 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
7520 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
7522 void kvm_apicv_init(struct kvm *kvm, bool enable)
7525 clear_bit(APICV_INHIBIT_REASON_DISABLE,
7526 &kvm->arch.apicv_inhibit_reasons);
7528 set_bit(APICV_INHIBIT_REASON_DISABLE,
7529 &kvm->arch.apicv_inhibit_reasons);
7531 EXPORT_SYMBOL_GPL(kvm_apicv_init);
7533 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7535 struct kvm_vcpu *target = NULL;
7536 struct kvm_apic_map *map;
7539 map = rcu_dereference(kvm->arch.apic_map);
7541 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7542 target = map->phys_map[dest_id]->vcpu;
7546 if (target && READ_ONCE(target->ready))
7547 kvm_vcpu_yield_to(target);
7550 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7552 unsigned long nr, a0, a1, a2, a3, ret;
7555 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7556 return kvm_hv_hypercall(vcpu);
7558 nr = kvm_rax_read(vcpu);
7559 a0 = kvm_rbx_read(vcpu);
7560 a1 = kvm_rcx_read(vcpu);
7561 a2 = kvm_rdx_read(vcpu);
7562 a3 = kvm_rsi_read(vcpu);
7564 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7566 op_64_bit = is_64_bit_mode(vcpu);
7575 if (kvm_x86_ops.get_cpl(vcpu) != 0) {
7581 case KVM_HC_VAPIC_POLL_IRQ:
7584 case KVM_HC_KICK_CPU:
7585 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7586 kvm_sched_yield(vcpu->kvm, a1);
7589 #ifdef CONFIG_X86_64
7590 case KVM_HC_CLOCK_PAIRING:
7591 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7594 case KVM_HC_SEND_IPI:
7595 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7597 case KVM_HC_SCHED_YIELD:
7598 kvm_sched_yield(vcpu->kvm, a0);
7608 kvm_rax_write(vcpu, ret);
7610 ++vcpu->stat.hypercalls;
7611 return kvm_skip_emulated_instruction(vcpu);
7613 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7615 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7617 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7618 char instruction[3];
7619 unsigned long rip = kvm_rip_read(vcpu);
7621 kvm_x86_ops.patch_hypercall(vcpu, instruction);
7623 return emulator_write_emulated(ctxt, rip, instruction, 3,
7627 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7629 return vcpu->run->request_interrupt_window &&
7630 likely(!pic_in_kernel(vcpu->kvm));
7633 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7635 struct kvm_run *kvm_run = vcpu->run;
7637 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7638 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7639 kvm_run->cr8 = kvm_get_cr8(vcpu);
7640 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7641 kvm_run->ready_for_interrupt_injection =
7642 pic_in_kernel(vcpu->kvm) ||
7643 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7646 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7650 if (!kvm_x86_ops.update_cr8_intercept)
7653 if (!lapic_in_kernel(vcpu))
7656 if (vcpu->arch.apicv_active)
7659 if (!vcpu->arch.apic->vapic_addr)
7660 max_irr = kvm_lapic_find_highest_irr(vcpu);
7667 tpr = kvm_lapic_get_cr8(vcpu);
7669 kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
7672 static int inject_pending_event(struct kvm_vcpu *vcpu)
7676 /* try to reinject previous events if any */
7678 if (vcpu->arch.exception.injected)
7679 kvm_x86_ops.queue_exception(vcpu);
7681 * Do not inject an NMI or interrupt if there is a pending
7682 * exception. Exceptions and interrupts are recognized at
7683 * instruction boundaries, i.e. the start of an instruction.
7684 * Trap-like exceptions, e.g. #DB, have higher priority than
7685 * NMIs and interrupts, i.e. traps are recognized before an
7686 * NMI/interrupt that's pending on the same instruction.
7687 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7688 * priority, but are only generated (pended) during instruction
7689 * execution, i.e. a pending fault-like exception means the
7690 * fault occurred on the *previous* instruction and must be
7691 * serviced prior to recognizing any new events in order to
7692 * fully complete the previous instruction.
7694 else if (!vcpu->arch.exception.pending) {
7695 if (vcpu->arch.nmi_injected)
7696 kvm_x86_ops.set_nmi(vcpu);
7697 else if (vcpu->arch.interrupt.injected)
7698 kvm_x86_ops.set_irq(vcpu);
7702 * Call check_nested_events() even if we reinjected a previous event
7703 * in order for caller to determine if it should require immediate-exit
7704 * from L2 to L1 due to pending L1 events which require exit
7707 if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) {
7708 r = kvm_x86_ops.check_nested_events(vcpu);
7713 /* try to inject new event if pending */
7714 if (vcpu->arch.exception.pending) {
7715 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7716 vcpu->arch.exception.has_error_code,
7717 vcpu->arch.exception.error_code);
7719 WARN_ON_ONCE(vcpu->arch.exception.injected);
7720 vcpu->arch.exception.pending = false;
7721 vcpu->arch.exception.injected = true;
7723 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7724 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7727 if (vcpu->arch.exception.nr == DB_VECTOR) {
7729 * This code assumes that nSVM doesn't use
7730 * check_nested_events(). If it does, the
7731 * DR6/DR7 changes should happen before L1
7732 * gets a #VMEXIT for an intercepted #DB in
7733 * L2. (Under VMX, on the other hand, the
7734 * DR6/DR7 changes should not happen in the
7735 * event of a VM-exit to L1 for an intercepted
7738 kvm_deliver_exception_payload(vcpu);
7739 if (vcpu->arch.dr7 & DR7_GD) {
7740 vcpu->arch.dr7 &= ~DR7_GD;
7741 kvm_update_dr7(vcpu);
7745 kvm_x86_ops.queue_exception(vcpu);
7748 /* Don't consider new event if we re-injected an event */
7749 if (kvm_event_needs_reinjection(vcpu))
7752 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7753 kvm_x86_ops.smi_allowed(vcpu)) {
7754 vcpu->arch.smi_pending = false;
7755 ++vcpu->arch.smi_count;
7757 } else if (vcpu->arch.nmi_pending && kvm_x86_ops.nmi_allowed(vcpu)) {
7758 --vcpu->arch.nmi_pending;
7759 vcpu->arch.nmi_injected = true;
7760 kvm_x86_ops.set_nmi(vcpu);
7761 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7763 * Because interrupts can be injected asynchronously, we are
7764 * calling check_nested_events again here to avoid a race condition.
7765 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7766 * proposal and current concerns. Perhaps we should be setting
7767 * KVM_REQ_EVENT only on certain events and not unconditionally?
7769 if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events) {
7770 r = kvm_x86_ops.check_nested_events(vcpu);
7774 if (kvm_x86_ops.interrupt_allowed(vcpu)) {
7775 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7777 kvm_x86_ops.set_irq(vcpu);
7784 static void process_nmi(struct kvm_vcpu *vcpu)
7789 * x86 is limited to one NMI running, and one NMI pending after it.
7790 * If an NMI is already in progress, limit further NMIs to just one.
7791 * Otherwise, allow two (and we'll inject the first one immediately).
7793 if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7796 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7797 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7798 kvm_make_request(KVM_REQ_EVENT, vcpu);
7801 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7804 flags |= seg->g << 23;
7805 flags |= seg->db << 22;
7806 flags |= seg->l << 21;
7807 flags |= seg->avl << 20;
7808 flags |= seg->present << 15;
7809 flags |= seg->dpl << 13;
7810 flags |= seg->s << 12;
7811 flags |= seg->type << 8;
7815 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7817 struct kvm_segment seg;
7820 kvm_get_segment(vcpu, &seg, n);
7821 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7824 offset = 0x7f84 + n * 12;
7826 offset = 0x7f2c + (n - 3) * 12;
7828 put_smstate(u32, buf, offset + 8, seg.base);
7829 put_smstate(u32, buf, offset + 4, seg.limit);
7830 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7833 #ifdef CONFIG_X86_64
7834 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7836 struct kvm_segment seg;
7840 kvm_get_segment(vcpu, &seg, n);
7841 offset = 0x7e00 + n * 16;
7843 flags = enter_smm_get_segment_flags(&seg) >> 8;
7844 put_smstate(u16, buf, offset, seg.selector);
7845 put_smstate(u16, buf, offset + 2, flags);
7846 put_smstate(u32, buf, offset + 4, seg.limit);
7847 put_smstate(u64, buf, offset + 8, seg.base);
7851 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7854 struct kvm_segment seg;
7858 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7859 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7860 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7861 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7863 for (i = 0; i < 8; i++)
7864 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7866 kvm_get_dr(vcpu, 6, &val);
7867 put_smstate(u32, buf, 0x7fcc, (u32)val);
7868 kvm_get_dr(vcpu, 7, &val);
7869 put_smstate(u32, buf, 0x7fc8, (u32)val);
7871 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7872 put_smstate(u32, buf, 0x7fc4, seg.selector);
7873 put_smstate(u32, buf, 0x7f64, seg.base);
7874 put_smstate(u32, buf, 0x7f60, seg.limit);
7875 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7877 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7878 put_smstate(u32, buf, 0x7fc0, seg.selector);
7879 put_smstate(u32, buf, 0x7f80, seg.base);
7880 put_smstate(u32, buf, 0x7f7c, seg.limit);
7881 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7883 kvm_x86_ops.get_gdt(vcpu, &dt);
7884 put_smstate(u32, buf, 0x7f74, dt.address);
7885 put_smstate(u32, buf, 0x7f70, dt.size);
7887 kvm_x86_ops.get_idt(vcpu, &dt);
7888 put_smstate(u32, buf, 0x7f58, dt.address);
7889 put_smstate(u32, buf, 0x7f54, dt.size);
7891 for (i = 0; i < 6; i++)
7892 enter_smm_save_seg_32(vcpu, buf, i);
7894 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7897 put_smstate(u32, buf, 0x7efc, 0x00020000);
7898 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7901 #ifdef CONFIG_X86_64
7902 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7905 struct kvm_segment seg;
7909 for (i = 0; i < 16; i++)
7910 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7912 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7913 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7915 kvm_get_dr(vcpu, 6, &val);
7916 put_smstate(u64, buf, 0x7f68, val);
7917 kvm_get_dr(vcpu, 7, &val);
7918 put_smstate(u64, buf, 0x7f60, val);
7920 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7921 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7922 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7924 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7927 put_smstate(u32, buf, 0x7efc, 0x00020064);
7929 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7931 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7932 put_smstate(u16, buf, 0x7e90, seg.selector);
7933 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7934 put_smstate(u32, buf, 0x7e94, seg.limit);
7935 put_smstate(u64, buf, 0x7e98, seg.base);
7937 kvm_x86_ops.get_idt(vcpu, &dt);
7938 put_smstate(u32, buf, 0x7e84, dt.size);
7939 put_smstate(u64, buf, 0x7e88, dt.address);
7941 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7942 put_smstate(u16, buf, 0x7e70, seg.selector);
7943 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7944 put_smstate(u32, buf, 0x7e74, seg.limit);
7945 put_smstate(u64, buf, 0x7e78, seg.base);
7947 kvm_x86_ops.get_gdt(vcpu, &dt);
7948 put_smstate(u32, buf, 0x7e64, dt.size);
7949 put_smstate(u64, buf, 0x7e68, dt.address);
7951 for (i = 0; i < 6; i++)
7952 enter_smm_save_seg_64(vcpu, buf, i);
7956 static void enter_smm(struct kvm_vcpu *vcpu)
7958 struct kvm_segment cs, ds;
7963 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7964 memset(buf, 0, 512);
7965 #ifdef CONFIG_X86_64
7966 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7967 enter_smm_save_state_64(vcpu, buf);
7970 enter_smm_save_state_32(vcpu, buf);
7973 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7974 * vCPU state (e.g. leave guest mode) after we've saved the state into
7975 * the SMM state-save area.
7977 kvm_x86_ops.pre_enter_smm(vcpu, buf);
7979 vcpu->arch.hflags |= HF_SMM_MASK;
7980 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7982 if (kvm_x86_ops.get_nmi_mask(vcpu))
7983 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7985 kvm_x86_ops.set_nmi_mask(vcpu, true);
7987 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7988 kvm_rip_write(vcpu, 0x8000);
7990 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7991 kvm_x86_ops.set_cr0(vcpu, cr0);
7992 vcpu->arch.cr0 = cr0;
7994 kvm_x86_ops.set_cr4(vcpu, 0);
7996 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7997 dt.address = dt.size = 0;
7998 kvm_x86_ops.set_idt(vcpu, &dt);
8000 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8002 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8003 cs.base = vcpu->arch.smbase;
8008 cs.limit = ds.limit = 0xffffffff;
8009 cs.type = ds.type = 0x3;
8010 cs.dpl = ds.dpl = 0;
8015 cs.avl = ds.avl = 0;
8016 cs.present = ds.present = 1;
8017 cs.unusable = ds.unusable = 0;
8018 cs.padding = ds.padding = 0;
8020 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8021 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8022 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8023 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8024 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8025 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8027 #ifdef CONFIG_X86_64
8028 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8029 kvm_x86_ops.set_efer(vcpu, 0);
8032 kvm_update_cpuid(vcpu);
8033 kvm_mmu_reset_context(vcpu);
8036 static void process_smi(struct kvm_vcpu *vcpu)
8038 vcpu->arch.smi_pending = true;
8039 kvm_make_request(KVM_REQ_EVENT, vcpu);
8042 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8043 unsigned long *vcpu_bitmap)
8047 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8049 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8050 NULL, vcpu_bitmap, cpus);
8052 free_cpumask_var(cpus);
8055 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8057 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8060 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8062 if (!lapic_in_kernel(vcpu))
8065 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8066 kvm_apic_update_apicv(vcpu);
8067 kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8069 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8072 * NOTE: Do not hold any lock prior to calling this.
8074 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8075 * locked, because it calls __x86_set_memory_region() which does
8076 * synchronize_srcu(&kvm->srcu).
8078 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8080 struct kvm_vcpu *except;
8081 unsigned long old, new, expected;
8083 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8084 !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8087 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8089 expected = new = old;
8091 __clear_bit(bit, &new);
8093 __set_bit(bit, &new);
8096 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8097 } while (old != expected);
8102 trace_kvm_apicv_update_request(activate, bit);
8103 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8104 kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8107 * Sending request to update APICV for all other vcpus,
8108 * while update the calling vcpu immediately instead of
8109 * waiting for another #VMEXIT to handle the request.
8111 except = kvm_get_running_vcpu();
8112 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8115 kvm_vcpu_update_apicv(except);
8117 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8119 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8121 if (!kvm_apic_present(vcpu))
8124 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8126 if (irqchip_split(vcpu->kvm))
8127 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8129 if (vcpu->arch.apicv_active)
8130 kvm_x86_ops.sync_pir_to_irr(vcpu);
8131 if (ioapic_in_kernel(vcpu->kvm))
8132 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8135 if (is_guest_mode(vcpu))
8136 vcpu->arch.load_eoi_exitmap_pending = true;
8138 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8141 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8143 u64 eoi_exit_bitmap[4];
8145 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8148 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8149 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8150 kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8153 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8154 unsigned long start, unsigned long end,
8157 unsigned long apic_address;
8160 * The physical address of apic access page is stored in the VMCS.
8161 * Update it when it becomes invalid.
8163 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8164 if (start <= apic_address && apic_address < end)
8165 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8170 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8172 struct page *page = NULL;
8174 if (!lapic_in_kernel(vcpu))
8177 if (!kvm_x86_ops.set_apic_access_page_addr)
8180 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8181 if (is_error_page(page))
8183 kvm_x86_ops.set_apic_access_page_addr(vcpu, page_to_phys(page));
8186 * Do not pin apic access page in memory, the MMU notifier
8187 * will call us again if it is migrated or swapped out.
8192 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8194 smp_send_reschedule(vcpu->cpu);
8196 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8199 * Returns 1 to let vcpu_run() continue the guest execution loop without
8200 * exiting to the userspace. Otherwise, the value will be returned to the
8203 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8207 dm_request_for_irq_injection(vcpu) &&
8208 kvm_cpu_accept_dm_intr(vcpu);
8209 enum exit_fastpath_completion exit_fastpath = EXIT_FASTPATH_NONE;
8211 bool req_immediate_exit = false;
8213 if (kvm_request_pending(vcpu)) {
8214 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8215 if (unlikely(!kvm_x86_ops.get_vmcs12_pages(vcpu))) {
8220 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8221 kvm_mmu_unload(vcpu);
8222 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8223 __kvm_migrate_timers(vcpu);
8224 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8225 kvm_gen_update_masterclock(vcpu->kvm);
8226 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8227 kvm_gen_kvmclock_update(vcpu);
8228 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8229 r = kvm_guest_time_update(vcpu);
8233 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8234 kvm_mmu_sync_roots(vcpu);
8235 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8236 kvm_mmu_load_pgd(vcpu);
8237 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
8238 kvm_vcpu_flush_tlb(vcpu, true);
8239 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8240 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8244 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8245 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8246 vcpu->mmio_needed = 0;
8250 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8251 /* Page is swapped out. Do synthetic halt */
8252 vcpu->arch.apf.halted = true;
8256 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8257 record_steal_time(vcpu);
8258 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8260 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8262 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8263 kvm_pmu_handle_event(vcpu);
8264 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8265 kvm_pmu_deliver_pmi(vcpu);
8266 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8267 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8268 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8269 vcpu->arch.ioapic_handled_vectors)) {
8270 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8271 vcpu->run->eoi.vector =
8272 vcpu->arch.pending_ioapic_eoi;
8277 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8278 vcpu_scan_ioapic(vcpu);
8279 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8280 vcpu_load_eoi_exitmap(vcpu);
8281 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8282 kvm_vcpu_reload_apic_access_page(vcpu);
8283 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8284 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8285 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8289 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8290 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8291 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8295 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8296 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8297 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8303 * KVM_REQ_HV_STIMER has to be processed after
8304 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8305 * depend on the guest clock being up-to-date
8307 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8308 kvm_hv_process_stimers(vcpu);
8309 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8310 kvm_vcpu_update_apicv(vcpu);
8313 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8314 ++vcpu->stat.req_event;
8315 kvm_apic_accept_events(vcpu);
8316 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8321 if (inject_pending_event(vcpu) != 0)
8322 req_immediate_exit = true;
8324 /* Enable SMI/NMI/IRQ window open exits if needed.
8326 * SMIs have three cases:
8327 * 1) They can be nested, and then there is nothing to
8328 * do here because RSM will cause a vmexit anyway.
8329 * 2) There is an ISA-specific reason why SMI cannot be
8330 * injected, and the moment when this changes can be
8332 * 3) Or the SMI can be pending because
8333 * inject_pending_event has completed the injection
8334 * of an IRQ or NMI from the previous vmexit, and
8335 * then we request an immediate exit to inject the
8338 if (vcpu->arch.smi_pending && !is_smm(vcpu))
8339 if (!kvm_x86_ops.enable_smi_window(vcpu))
8340 req_immediate_exit = true;
8341 if (vcpu->arch.nmi_pending)
8342 kvm_x86_ops.enable_nmi_window(vcpu);
8343 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8344 kvm_x86_ops.enable_irq_window(vcpu);
8345 WARN_ON(vcpu->arch.exception.pending);
8348 if (kvm_lapic_enabled(vcpu)) {
8349 update_cr8_intercept(vcpu);
8350 kvm_lapic_sync_to_vapic(vcpu);
8354 r = kvm_mmu_reload(vcpu);
8356 goto cancel_injection;
8361 kvm_x86_ops.prepare_guest_switch(vcpu);
8364 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8365 * IPI are then delayed after guest entry, which ensures that they
8366 * result in virtual interrupt delivery.
8368 local_irq_disable();
8369 vcpu->mode = IN_GUEST_MODE;
8371 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8374 * 1) We should set ->mode before checking ->requests. Please see
8375 * the comment in kvm_vcpu_exiting_guest_mode().
8377 * 2) For APICv, we should set ->mode before checking PID.ON. This
8378 * pairs with the memory barrier implicit in pi_test_and_set_on
8379 * (see vmx_deliver_posted_interrupt).
8381 * 3) This also orders the write to mode from any reads to the page
8382 * tables done while the VCPU is running. Please see the comment
8383 * in kvm_flush_remote_tlbs.
8385 smp_mb__after_srcu_read_unlock();
8388 * This handles the case where a posted interrupt was
8389 * notified with kvm_vcpu_kick.
8391 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8392 kvm_x86_ops.sync_pir_to_irr(vcpu);
8394 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8395 || need_resched() || signal_pending(current)) {
8396 vcpu->mode = OUTSIDE_GUEST_MODE;
8400 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8402 goto cancel_injection;
8405 if (req_immediate_exit) {
8406 kvm_make_request(KVM_REQ_EVENT, vcpu);
8407 kvm_x86_ops.request_immediate_exit(vcpu);
8410 trace_kvm_entry(vcpu->vcpu_id);
8411 guest_enter_irqoff();
8413 fpregs_assert_state_consistent();
8414 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8415 switch_fpu_return();
8417 if (unlikely(vcpu->arch.switch_db_regs)) {
8419 set_debugreg(vcpu->arch.eff_db[0], 0);
8420 set_debugreg(vcpu->arch.eff_db[1], 1);
8421 set_debugreg(vcpu->arch.eff_db[2], 2);
8422 set_debugreg(vcpu->arch.eff_db[3], 3);
8423 set_debugreg(vcpu->arch.dr6, 6);
8424 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8427 kvm_x86_ops.run(vcpu);
8430 * Do this here before restoring debug registers on the host. And
8431 * since we do this before handling the vmexit, a DR access vmexit
8432 * can (a) read the correct value of the debug registers, (b) set
8433 * KVM_DEBUGREG_WONT_EXIT again.
8435 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8436 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8437 kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8438 kvm_update_dr0123(vcpu);
8439 kvm_update_dr7(vcpu);
8440 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8444 * If the guest has used debug registers, at least dr7
8445 * will be disabled while returning to the host.
8446 * If we don't have active breakpoints in the host, we don't
8447 * care about the messed up debug address registers. But if
8448 * we have some of them active, restore the old state.
8450 if (hw_breakpoint_active())
8451 hw_breakpoint_restore();
8453 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8455 vcpu->mode = OUTSIDE_GUEST_MODE;
8458 kvm_x86_ops.handle_exit_irqoff(vcpu, &exit_fastpath);
8461 * Consume any pending interrupts, including the possible source of
8462 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8463 * An instruction is required after local_irq_enable() to fully unblock
8464 * interrupts on processors that implement an interrupt shadow, the
8465 * stat.exits increment will do nicely.
8467 kvm_before_interrupt(vcpu);
8470 local_irq_disable();
8471 kvm_after_interrupt(vcpu);
8473 guest_exit_irqoff();
8474 if (lapic_in_kernel(vcpu)) {
8475 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8476 if (delta != S64_MIN) {
8477 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8478 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8485 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8488 * Profile KVM exit RIPs:
8490 if (unlikely(prof_on == KVM_PROFILING)) {
8491 unsigned long rip = kvm_rip_read(vcpu);
8492 profile_hit(KVM_PROFILING, (void *)rip);
8495 if (unlikely(vcpu->arch.tsc_always_catchup))
8496 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8498 if (vcpu->arch.apic_attention)
8499 kvm_lapic_sync_from_vapic(vcpu);
8501 r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
8505 kvm_x86_ops.cancel_injection(vcpu);
8506 if (unlikely(vcpu->arch.apic_attention))
8507 kvm_lapic_sync_from_vapic(vcpu);
8512 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8514 if (!kvm_arch_vcpu_runnable(vcpu) &&
8515 (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
8516 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8517 kvm_vcpu_block(vcpu);
8518 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8520 if (kvm_x86_ops.post_block)
8521 kvm_x86_ops.post_block(vcpu);
8523 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8527 kvm_apic_accept_events(vcpu);
8528 switch(vcpu->arch.mp_state) {
8529 case KVM_MP_STATE_HALTED:
8530 vcpu->arch.pv.pv_unhalted = false;
8531 vcpu->arch.mp_state =
8532 KVM_MP_STATE_RUNNABLE;
8534 case KVM_MP_STATE_RUNNABLE:
8535 vcpu->arch.apf.halted = false;
8537 case KVM_MP_STATE_INIT_RECEIVED:
8545 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8547 if (is_guest_mode(vcpu) && kvm_x86_ops.check_nested_events)
8548 kvm_x86_ops.check_nested_events(vcpu);
8550 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8551 !vcpu->arch.apf.halted);
8554 static int vcpu_run(struct kvm_vcpu *vcpu)
8557 struct kvm *kvm = vcpu->kvm;
8559 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8560 vcpu->arch.l1tf_flush_l1d = true;
8563 if (kvm_vcpu_running(vcpu)) {
8564 r = vcpu_enter_guest(vcpu);
8566 r = vcpu_block(kvm, vcpu);
8572 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8573 if (kvm_cpu_has_pending_timer(vcpu))
8574 kvm_inject_pending_timer_irqs(vcpu);
8576 if (dm_request_for_irq_injection(vcpu) &&
8577 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8579 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8580 ++vcpu->stat.request_irq_exits;
8584 kvm_check_async_pf_completion(vcpu);
8586 if (signal_pending(current)) {
8588 vcpu->run->exit_reason = KVM_EXIT_INTR;
8589 ++vcpu->stat.signal_exits;
8592 if (need_resched()) {
8593 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8595 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8599 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8604 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8608 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8609 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8610 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8614 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8616 BUG_ON(!vcpu->arch.pio.count);
8618 return complete_emulated_io(vcpu);
8622 * Implements the following, as a state machine:
8626 * for each mmio piece in the fragment
8634 * for each mmio piece in the fragment
8639 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8641 struct kvm_run *run = vcpu->run;
8642 struct kvm_mmio_fragment *frag;
8645 BUG_ON(!vcpu->mmio_needed);
8647 /* Complete previous fragment */
8648 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8649 len = min(8u, frag->len);
8650 if (!vcpu->mmio_is_write)
8651 memcpy(frag->data, run->mmio.data, len);
8653 if (frag->len <= 8) {
8654 /* Switch to the next fragment. */
8656 vcpu->mmio_cur_fragment++;
8658 /* Go forward to the next mmio piece. */
8664 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8665 vcpu->mmio_needed = 0;
8667 /* FIXME: return into emulator if single-stepping. */
8668 if (vcpu->mmio_is_write)
8670 vcpu->mmio_read_completed = 1;
8671 return complete_emulated_io(vcpu);
8674 run->exit_reason = KVM_EXIT_MMIO;
8675 run->mmio.phys_addr = frag->gpa;
8676 if (vcpu->mmio_is_write)
8677 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8678 run->mmio.len = min(8u, frag->len);
8679 run->mmio.is_write = vcpu->mmio_is_write;
8680 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8684 static void kvm_save_current_fpu(struct fpu *fpu)
8687 * If the target FPU state is not resident in the CPU registers, just
8688 * memcpy() from current, else save CPU state directly to the target.
8690 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8691 memcpy(&fpu->state, ¤t->thread.fpu.state,
8692 fpu_kernel_xstate_size);
8694 copy_fpregs_to_fpstate(fpu);
8697 /* Swap (qemu) user FPU context for the guest FPU context. */
8698 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8702 kvm_save_current_fpu(vcpu->arch.user_fpu);
8704 /* PKRU is separately restored in kvm_x86_ops.run. */
8705 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8706 ~XFEATURE_MASK_PKRU);
8708 fpregs_mark_activate();
8714 /* When vcpu_run ends, restore user space FPU context. */
8715 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8719 kvm_save_current_fpu(vcpu->arch.guest_fpu);
8721 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8723 fpregs_mark_activate();
8726 ++vcpu->stat.fpu_reload;
8730 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8735 kvm_sigset_activate(vcpu);
8736 kvm_load_guest_fpu(vcpu);
8738 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8739 if (kvm_run->immediate_exit) {
8743 kvm_vcpu_block(vcpu);
8744 kvm_apic_accept_events(vcpu);
8745 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8747 if (signal_pending(current)) {
8749 vcpu->run->exit_reason = KVM_EXIT_INTR;
8750 ++vcpu->stat.signal_exits;
8755 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8760 if (vcpu->run->kvm_dirty_regs) {
8761 r = sync_regs(vcpu);
8766 /* re-sync apic's tpr */
8767 if (!lapic_in_kernel(vcpu)) {
8768 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8774 if (unlikely(vcpu->arch.complete_userspace_io)) {
8775 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8776 vcpu->arch.complete_userspace_io = NULL;
8781 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8783 if (kvm_run->immediate_exit)
8789 kvm_put_guest_fpu(vcpu);
8790 if (vcpu->run->kvm_valid_regs)
8792 post_kvm_run_save(vcpu);
8793 kvm_sigset_deactivate(vcpu);
8799 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8801 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8803 * We are here if userspace calls get_regs() in the middle of
8804 * instruction emulation. Registers state needs to be copied
8805 * back from emulation context to vcpu. Userspace shouldn't do
8806 * that usually, but some bad designed PV devices (vmware
8807 * backdoor interface) need this to work
8809 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
8810 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8812 regs->rax = kvm_rax_read(vcpu);
8813 regs->rbx = kvm_rbx_read(vcpu);
8814 regs->rcx = kvm_rcx_read(vcpu);
8815 regs->rdx = kvm_rdx_read(vcpu);
8816 regs->rsi = kvm_rsi_read(vcpu);
8817 regs->rdi = kvm_rdi_read(vcpu);
8818 regs->rsp = kvm_rsp_read(vcpu);
8819 regs->rbp = kvm_rbp_read(vcpu);
8820 #ifdef CONFIG_X86_64
8821 regs->r8 = kvm_r8_read(vcpu);
8822 regs->r9 = kvm_r9_read(vcpu);
8823 regs->r10 = kvm_r10_read(vcpu);
8824 regs->r11 = kvm_r11_read(vcpu);
8825 regs->r12 = kvm_r12_read(vcpu);
8826 regs->r13 = kvm_r13_read(vcpu);
8827 regs->r14 = kvm_r14_read(vcpu);
8828 regs->r15 = kvm_r15_read(vcpu);
8831 regs->rip = kvm_rip_read(vcpu);
8832 regs->rflags = kvm_get_rflags(vcpu);
8835 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8838 __get_regs(vcpu, regs);
8843 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8845 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8846 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8848 kvm_rax_write(vcpu, regs->rax);
8849 kvm_rbx_write(vcpu, regs->rbx);
8850 kvm_rcx_write(vcpu, regs->rcx);
8851 kvm_rdx_write(vcpu, regs->rdx);
8852 kvm_rsi_write(vcpu, regs->rsi);
8853 kvm_rdi_write(vcpu, regs->rdi);
8854 kvm_rsp_write(vcpu, regs->rsp);
8855 kvm_rbp_write(vcpu, regs->rbp);
8856 #ifdef CONFIG_X86_64
8857 kvm_r8_write(vcpu, regs->r8);
8858 kvm_r9_write(vcpu, regs->r9);
8859 kvm_r10_write(vcpu, regs->r10);
8860 kvm_r11_write(vcpu, regs->r11);
8861 kvm_r12_write(vcpu, regs->r12);
8862 kvm_r13_write(vcpu, regs->r13);
8863 kvm_r14_write(vcpu, regs->r14);
8864 kvm_r15_write(vcpu, regs->r15);
8867 kvm_rip_write(vcpu, regs->rip);
8868 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8870 vcpu->arch.exception.pending = false;
8872 kvm_make_request(KVM_REQ_EVENT, vcpu);
8875 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8878 __set_regs(vcpu, regs);
8883 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8885 struct kvm_segment cs;
8887 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8891 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8893 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8897 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8898 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8899 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8900 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8901 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8902 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8904 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8905 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8907 kvm_x86_ops.get_idt(vcpu, &dt);
8908 sregs->idt.limit = dt.size;
8909 sregs->idt.base = dt.address;
8910 kvm_x86_ops.get_gdt(vcpu, &dt);
8911 sregs->gdt.limit = dt.size;
8912 sregs->gdt.base = dt.address;
8914 sregs->cr0 = kvm_read_cr0(vcpu);
8915 sregs->cr2 = vcpu->arch.cr2;
8916 sregs->cr3 = kvm_read_cr3(vcpu);
8917 sregs->cr4 = kvm_read_cr4(vcpu);
8918 sregs->cr8 = kvm_get_cr8(vcpu);
8919 sregs->efer = vcpu->arch.efer;
8920 sregs->apic_base = kvm_get_apic_base(vcpu);
8922 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8924 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8925 set_bit(vcpu->arch.interrupt.nr,
8926 (unsigned long *)sregs->interrupt_bitmap);
8929 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8930 struct kvm_sregs *sregs)
8933 __get_sregs(vcpu, sregs);
8938 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8939 struct kvm_mp_state *mp_state)
8942 if (kvm_mpx_supported())
8943 kvm_load_guest_fpu(vcpu);
8945 kvm_apic_accept_events(vcpu);
8946 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8947 vcpu->arch.pv.pv_unhalted)
8948 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8950 mp_state->mp_state = vcpu->arch.mp_state;
8952 if (kvm_mpx_supported())
8953 kvm_put_guest_fpu(vcpu);
8958 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8959 struct kvm_mp_state *mp_state)
8965 if (!lapic_in_kernel(vcpu) &&
8966 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8970 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
8971 * INIT state; latched init should be reported using
8972 * KVM_SET_VCPU_EVENTS, so reject it here.
8974 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
8975 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8976 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8979 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8980 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8981 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8983 vcpu->arch.mp_state = mp_state->mp_state;
8984 kvm_make_request(KVM_REQ_EVENT, vcpu);
8992 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8993 int reason, bool has_error_code, u32 error_code)
8995 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8998 init_emulate_ctxt(vcpu);
9000 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9001 has_error_code, error_code);
9003 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9004 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9005 vcpu->run->internal.ndata = 0;
9009 kvm_rip_write(vcpu, ctxt->eip);
9010 kvm_set_rflags(vcpu, ctxt->eflags);
9013 EXPORT_SYMBOL_GPL(kvm_task_switch);
9015 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9017 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9019 * When EFER.LME and CR0.PG are set, the processor is in
9020 * 64-bit mode (though maybe in a 32-bit code segment).
9021 * CR4.PAE and EFER.LMA must be set.
9023 if (!(sregs->cr4 & X86_CR4_PAE)
9024 || !(sregs->efer & EFER_LMA))
9028 * Not in 64-bit mode: EFER.LMA is clear and the code
9029 * segment cannot be 64-bit.
9031 if (sregs->efer & EFER_LMA || sregs->cs.l)
9035 return kvm_valid_cr4(vcpu, sregs->cr4);
9038 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9040 struct msr_data apic_base_msr;
9041 int mmu_reset_needed = 0;
9042 int cpuid_update_needed = 0;
9043 int pending_vec, max_bits, idx;
9047 if (kvm_valid_sregs(vcpu, sregs))
9050 apic_base_msr.data = sregs->apic_base;
9051 apic_base_msr.host_initiated = true;
9052 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9055 dt.size = sregs->idt.limit;
9056 dt.address = sregs->idt.base;
9057 kvm_x86_ops.set_idt(vcpu, &dt);
9058 dt.size = sregs->gdt.limit;
9059 dt.address = sregs->gdt.base;
9060 kvm_x86_ops.set_gdt(vcpu, &dt);
9062 vcpu->arch.cr2 = sregs->cr2;
9063 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9064 vcpu->arch.cr3 = sregs->cr3;
9065 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9067 kvm_set_cr8(vcpu, sregs->cr8);
9069 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9070 kvm_x86_ops.set_efer(vcpu, sregs->efer);
9072 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9073 kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9074 vcpu->arch.cr0 = sregs->cr0;
9076 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9077 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9078 (X86_CR4_OSXSAVE | X86_CR4_PKE));
9079 kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9080 if (cpuid_update_needed)
9081 kvm_update_cpuid(vcpu);
9083 idx = srcu_read_lock(&vcpu->kvm->srcu);
9084 if (is_pae_paging(vcpu)) {
9085 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9086 mmu_reset_needed = 1;
9088 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9090 if (mmu_reset_needed)
9091 kvm_mmu_reset_context(vcpu);
9093 max_bits = KVM_NR_INTERRUPTS;
9094 pending_vec = find_first_bit(
9095 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9096 if (pending_vec < max_bits) {
9097 kvm_queue_interrupt(vcpu, pending_vec, false);
9098 pr_debug("Set back pending irq %d\n", pending_vec);
9101 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9102 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9103 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9104 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9105 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9106 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9108 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9109 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9111 update_cr8_intercept(vcpu);
9113 /* Older userspace won't unhalt the vcpu on reset. */
9114 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9115 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9117 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9119 kvm_make_request(KVM_REQ_EVENT, vcpu);
9126 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9127 struct kvm_sregs *sregs)
9132 ret = __set_sregs(vcpu, sregs);
9137 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9138 struct kvm_guest_debug *dbg)
9140 unsigned long rflags;
9145 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9147 if (vcpu->arch.exception.pending)
9149 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9150 kvm_queue_exception(vcpu, DB_VECTOR);
9152 kvm_queue_exception(vcpu, BP_VECTOR);
9156 * Read rflags as long as potentially injected trace flags are still
9159 rflags = kvm_get_rflags(vcpu);
9161 vcpu->guest_debug = dbg->control;
9162 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9163 vcpu->guest_debug = 0;
9165 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9166 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9167 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9168 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9170 for (i = 0; i < KVM_NR_DB_REGS; i++)
9171 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9173 kvm_update_dr7(vcpu);
9175 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9176 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9177 get_segment_base(vcpu, VCPU_SREG_CS);
9180 * Trigger an rflags update that will inject or remove the trace
9183 kvm_set_rflags(vcpu, rflags);
9185 kvm_x86_ops.update_bp_intercept(vcpu);
9195 * Translate a guest virtual address to a guest physical address.
9197 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9198 struct kvm_translation *tr)
9200 unsigned long vaddr = tr->linear_address;
9206 idx = srcu_read_lock(&vcpu->kvm->srcu);
9207 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9208 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9209 tr->physical_address = gpa;
9210 tr->valid = gpa != UNMAPPED_GVA;
9218 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9220 struct fxregs_state *fxsave;
9224 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9225 memcpy(fpu->fpr, fxsave->st_space, 128);
9226 fpu->fcw = fxsave->cwd;
9227 fpu->fsw = fxsave->swd;
9228 fpu->ftwx = fxsave->twd;
9229 fpu->last_opcode = fxsave->fop;
9230 fpu->last_ip = fxsave->rip;
9231 fpu->last_dp = fxsave->rdp;
9232 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9238 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9240 struct fxregs_state *fxsave;
9244 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9246 memcpy(fxsave->st_space, fpu->fpr, 128);
9247 fxsave->cwd = fpu->fcw;
9248 fxsave->swd = fpu->fsw;
9249 fxsave->twd = fpu->ftwx;
9250 fxsave->fop = fpu->last_opcode;
9251 fxsave->rip = fpu->last_ip;
9252 fxsave->rdp = fpu->last_dp;
9253 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9259 static void store_regs(struct kvm_vcpu *vcpu)
9261 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9263 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9264 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9266 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9267 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9269 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9270 kvm_vcpu_ioctl_x86_get_vcpu_events(
9271 vcpu, &vcpu->run->s.regs.events);
9274 static int sync_regs(struct kvm_vcpu *vcpu)
9276 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9279 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9280 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9281 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9283 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9284 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9286 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9288 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9289 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9290 vcpu, &vcpu->run->s.regs.events))
9292 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9298 static void fx_init(struct kvm_vcpu *vcpu)
9300 fpstate_init(&vcpu->arch.guest_fpu->state);
9301 if (boot_cpu_has(X86_FEATURE_XSAVES))
9302 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9303 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9306 * Ensure guest xcr0 is valid for loading
9308 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9310 vcpu->arch.cr0 |= X86_CR0_ET;
9313 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9315 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9316 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9317 "guest TSC will not be reliable\n");
9322 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9327 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9328 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9330 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9332 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9334 r = kvm_mmu_create(vcpu);
9338 if (irqchip_in_kernel(vcpu->kvm)) {
9339 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9341 goto fail_mmu_destroy;
9342 if (kvm_apicv_activated(vcpu->kvm))
9343 vcpu->arch.apicv_active = true;
9345 static_key_slow_inc(&kvm_no_apic_vcpu);
9349 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9351 goto fail_free_lapic;
9352 vcpu->arch.pio_data = page_address(page);
9354 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9355 GFP_KERNEL_ACCOUNT);
9356 if (!vcpu->arch.mce_banks)
9357 goto fail_free_pio_data;
9358 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9360 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9361 GFP_KERNEL_ACCOUNT))
9362 goto fail_free_mce_banks;
9364 if (!alloc_emulate_ctxt(vcpu))
9365 goto free_wbinvd_dirty_mask;
9367 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9368 GFP_KERNEL_ACCOUNT);
9369 if (!vcpu->arch.user_fpu) {
9370 pr_err("kvm: failed to allocate userspace's fpu\n");
9371 goto free_emulate_ctxt;
9374 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9375 GFP_KERNEL_ACCOUNT);
9376 if (!vcpu->arch.guest_fpu) {
9377 pr_err("kvm: failed to allocate vcpu's fpu\n");
9382 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9384 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9386 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9388 kvm_async_pf_hash_reset(vcpu);
9391 vcpu->arch.pending_external_vector = -1;
9392 vcpu->arch.preempted_in_kernel = false;
9394 kvm_hv_vcpu_init(vcpu);
9396 r = kvm_x86_ops.vcpu_create(vcpu);
9398 goto free_guest_fpu;
9400 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9401 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9402 kvm_vcpu_mtrr_init(vcpu);
9404 kvm_vcpu_reset(vcpu, false);
9405 kvm_init_mmu(vcpu, false);
9410 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9412 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9414 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9415 free_wbinvd_dirty_mask:
9416 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9417 fail_free_mce_banks:
9418 kfree(vcpu->arch.mce_banks);
9420 free_page((unsigned long)vcpu->arch.pio_data);
9422 kvm_free_lapic(vcpu);
9424 kvm_mmu_destroy(vcpu);
9428 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9430 struct msr_data msr;
9431 struct kvm *kvm = vcpu->kvm;
9433 kvm_hv_vcpu_postcreate(vcpu);
9435 if (mutex_lock_killable(&vcpu->mutex))
9439 msr.index = MSR_IA32_TSC;
9440 msr.host_initiated = true;
9441 kvm_write_tsc(vcpu, &msr);
9444 /* poll control enabled by default */
9445 vcpu->arch.msr_kvm_poll_control = 1;
9447 mutex_unlock(&vcpu->mutex);
9449 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9450 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9451 KVMCLOCK_SYNC_PERIOD);
9454 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9456 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9459 kvm_release_pfn(cache->pfn, cache->dirty, cache);
9461 kvmclock_reset(vcpu);
9463 kvm_x86_ops.vcpu_free(vcpu);
9465 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9466 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9467 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9468 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9470 kvm_hv_vcpu_uninit(vcpu);
9471 kvm_pmu_destroy(vcpu);
9472 kfree(vcpu->arch.mce_banks);
9473 kvm_free_lapic(vcpu);
9474 idx = srcu_read_lock(&vcpu->kvm->srcu);
9475 kvm_mmu_destroy(vcpu);
9476 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9477 free_page((unsigned long)vcpu->arch.pio_data);
9478 if (!lapic_in_kernel(vcpu))
9479 static_key_slow_dec(&kvm_no_apic_vcpu);
9482 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9484 kvm_lapic_reset(vcpu, init_event);
9486 vcpu->arch.hflags = 0;
9488 vcpu->arch.smi_pending = 0;
9489 vcpu->arch.smi_count = 0;
9490 atomic_set(&vcpu->arch.nmi_queued, 0);
9491 vcpu->arch.nmi_pending = 0;
9492 vcpu->arch.nmi_injected = false;
9493 kvm_clear_interrupt_queue(vcpu);
9494 kvm_clear_exception_queue(vcpu);
9496 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9497 kvm_update_dr0123(vcpu);
9498 vcpu->arch.dr6 = DR6_INIT;
9499 vcpu->arch.dr7 = DR7_FIXED_1;
9500 kvm_update_dr7(vcpu);
9504 kvm_make_request(KVM_REQ_EVENT, vcpu);
9505 vcpu->arch.apf.msr_val = 0;
9506 vcpu->arch.st.msr_val = 0;
9508 kvmclock_reset(vcpu);
9510 kvm_clear_async_pf_completion_queue(vcpu);
9511 kvm_async_pf_hash_reset(vcpu);
9512 vcpu->arch.apf.halted = false;
9514 if (kvm_mpx_supported()) {
9515 void *mpx_state_buffer;
9518 * To avoid have the INIT path from kvm_apic_has_events() that be
9519 * called with loaded FPU and does not let userspace fix the state.
9522 kvm_put_guest_fpu(vcpu);
9523 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9525 if (mpx_state_buffer)
9526 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9527 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9529 if (mpx_state_buffer)
9530 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9532 kvm_load_guest_fpu(vcpu);
9536 kvm_pmu_reset(vcpu);
9537 vcpu->arch.smbase = 0x30000;
9539 vcpu->arch.msr_misc_features_enables = 0;
9541 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9544 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9545 vcpu->arch.regs_avail = ~0;
9546 vcpu->arch.regs_dirty = ~0;
9548 vcpu->arch.ia32_xss = 0;
9550 kvm_x86_ops.vcpu_reset(vcpu, init_event);
9553 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9555 struct kvm_segment cs;
9557 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9558 cs.selector = vector << 8;
9559 cs.base = vector << 12;
9560 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9561 kvm_rip_write(vcpu, 0);
9564 int kvm_arch_hardware_enable(void)
9567 struct kvm_vcpu *vcpu;
9572 bool stable, backwards_tsc = false;
9574 kvm_shared_msr_cpu_online();
9575 ret = kvm_x86_ops.hardware_enable();
9579 local_tsc = rdtsc();
9580 stable = !kvm_check_tsc_unstable();
9581 list_for_each_entry(kvm, &vm_list, vm_list) {
9582 kvm_for_each_vcpu(i, vcpu, kvm) {
9583 if (!stable && vcpu->cpu == smp_processor_id())
9584 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9585 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9586 backwards_tsc = true;
9587 if (vcpu->arch.last_host_tsc > max_tsc)
9588 max_tsc = vcpu->arch.last_host_tsc;
9594 * Sometimes, even reliable TSCs go backwards. This happens on
9595 * platforms that reset TSC during suspend or hibernate actions, but
9596 * maintain synchronization. We must compensate. Fortunately, we can
9597 * detect that condition here, which happens early in CPU bringup,
9598 * before any KVM threads can be running. Unfortunately, we can't
9599 * bring the TSCs fully up to date with real time, as we aren't yet far
9600 * enough into CPU bringup that we know how much real time has actually
9601 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9602 * variables that haven't been updated yet.
9604 * So we simply find the maximum observed TSC above, then record the
9605 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9606 * the adjustment will be applied. Note that we accumulate
9607 * adjustments, in case multiple suspend cycles happen before some VCPU
9608 * gets a chance to run again. In the event that no KVM threads get a
9609 * chance to run, we will miss the entire elapsed period, as we'll have
9610 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9611 * loose cycle time. This isn't too big a deal, since the loss will be
9612 * uniform across all VCPUs (not to mention the scenario is extremely
9613 * unlikely). It is possible that a second hibernate recovery happens
9614 * much faster than a first, causing the observed TSC here to be
9615 * smaller; this would require additional padding adjustment, which is
9616 * why we set last_host_tsc to the local tsc observed here.
9618 * N.B. - this code below runs only on platforms with reliable TSC,
9619 * as that is the only way backwards_tsc is set above. Also note
9620 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9621 * have the same delta_cyc adjustment applied if backwards_tsc
9622 * is detected. Note further, this adjustment is only done once,
9623 * as we reset last_host_tsc on all VCPUs to stop this from being
9624 * called multiple times (one for each physical CPU bringup).
9626 * Platforms with unreliable TSCs don't have to deal with this, they
9627 * will be compensated by the logic in vcpu_load, which sets the TSC to
9628 * catchup mode. This will catchup all VCPUs to real time, but cannot
9629 * guarantee that they stay in perfect synchronization.
9631 if (backwards_tsc) {
9632 u64 delta_cyc = max_tsc - local_tsc;
9633 list_for_each_entry(kvm, &vm_list, vm_list) {
9634 kvm->arch.backwards_tsc_observed = true;
9635 kvm_for_each_vcpu(i, vcpu, kvm) {
9636 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9637 vcpu->arch.last_host_tsc = local_tsc;
9638 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9642 * We have to disable TSC offset matching.. if you were
9643 * booting a VM while issuing an S4 host suspend....
9644 * you may have some problem. Solving this issue is
9645 * left as an exercise to the reader.
9647 kvm->arch.last_tsc_nsec = 0;
9648 kvm->arch.last_tsc_write = 0;
9655 void kvm_arch_hardware_disable(void)
9657 kvm_x86_ops.hardware_disable();
9658 drop_user_return_notifiers();
9661 int kvm_arch_hardware_setup(void *opaque)
9663 struct kvm_x86_init_ops *ops = opaque;
9666 rdmsrl_safe(MSR_EFER, &host_efer);
9668 if (boot_cpu_has(X86_FEATURE_XSAVES))
9669 rdmsrl(MSR_IA32_XSS, host_xss);
9671 r = ops->hardware_setup();
9675 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9677 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9680 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
9681 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
9682 #undef __kvm_cpu_cap_has
9684 if (kvm_has_tsc_control) {
9686 * Make sure the user can only configure tsc_khz values that
9687 * fit into a signed integer.
9688 * A min value is not calculated because it will always
9689 * be 1 on all machines.
9691 u64 max = min(0x7fffffffULL,
9692 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9693 kvm_max_guest_tsc_khz = max;
9695 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9698 kvm_init_msr_list();
9702 void kvm_arch_hardware_unsetup(void)
9704 kvm_x86_ops.hardware_unsetup();
9707 int kvm_arch_check_processor_compat(void *opaque)
9709 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
9710 struct kvm_x86_init_ops *ops = opaque;
9712 WARN_ON(!irqs_disabled());
9714 if (__cr4_reserved_bits(cpu_has, c) !=
9715 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
9718 return ops->check_processor_compatibility();
9721 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9723 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9725 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9727 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9729 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9732 struct static_key kvm_no_apic_vcpu __read_mostly;
9733 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9735 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9737 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9739 vcpu->arch.l1tf_flush_l1d = true;
9740 if (pmu->version && unlikely(pmu->event_count)) {
9741 pmu->need_cleanup = true;
9742 kvm_make_request(KVM_REQ_PMU, vcpu);
9744 kvm_x86_ops.sched_in(vcpu, cpu);
9747 void kvm_arch_free_vm(struct kvm *kvm)
9749 kfree(kvm->arch.hyperv.hv_pa_pg);
9754 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9759 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9760 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9761 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9762 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9763 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9764 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9766 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9767 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9768 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9769 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9770 &kvm->arch.irq_sources_bitmap);
9772 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9773 mutex_init(&kvm->arch.apic_map_lock);
9774 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9776 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
9777 pvclock_update_vm_gtod_copy(kvm);
9779 kvm->arch.guest_can_read_msr_platform_info = true;
9781 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9782 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9784 kvm_hv_init_vm(kvm);
9785 kvm_page_track_init(kvm);
9786 kvm_mmu_init_vm(kvm);
9788 return kvm_x86_ops.vm_init(kvm);
9791 int kvm_arch_post_init_vm(struct kvm *kvm)
9793 return kvm_mmu_post_init_vm(kvm);
9796 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9799 kvm_mmu_unload(vcpu);
9803 static void kvm_free_vcpus(struct kvm *kvm)
9806 struct kvm_vcpu *vcpu;
9809 * Unpin any mmu pages first.
9811 kvm_for_each_vcpu(i, vcpu, kvm) {
9812 kvm_clear_async_pf_completion_queue(vcpu);
9813 kvm_unload_vcpu_mmu(vcpu);
9815 kvm_for_each_vcpu(i, vcpu, kvm)
9816 kvm_vcpu_destroy(vcpu);
9818 mutex_lock(&kvm->lock);
9819 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9820 kvm->vcpus[i] = NULL;
9822 atomic_set(&kvm->online_vcpus, 0);
9823 mutex_unlock(&kvm->lock);
9826 void kvm_arch_sync_events(struct kvm *kvm)
9828 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9829 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9833 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9836 unsigned long hva, uninitialized_var(old_npages);
9837 struct kvm_memslots *slots = kvm_memslots(kvm);
9838 struct kvm_memory_slot *slot;
9840 /* Called with kvm->slots_lock held. */
9841 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9844 slot = id_to_memslot(slots, id);
9846 if (slot && slot->npages)
9850 * MAP_SHARED to prevent internal slot pages from being moved
9853 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9854 MAP_SHARED | MAP_ANONYMOUS, 0);
9855 if (IS_ERR((void *)hva))
9856 return PTR_ERR((void *)hva);
9858 if (!slot || !slot->npages)
9862 * Stuff a non-canonical value to catch use-after-delete. This
9863 * ends up being 0 on 32-bit KVM, but there's no better
9866 hva = (unsigned long)(0xdeadull << 48);
9867 old_npages = slot->npages;
9870 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9871 struct kvm_userspace_memory_region m;
9873 m.slot = id | (i << 16);
9875 m.guest_phys_addr = gpa;
9876 m.userspace_addr = hva;
9877 m.memory_size = size;
9878 r = __kvm_set_memory_region(kvm, &m);
9884 vm_munmap(hva, old_npages * PAGE_SIZE);
9888 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9890 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9892 kvm_mmu_pre_destroy_vm(kvm);
9895 void kvm_arch_destroy_vm(struct kvm *kvm)
9897 if (current->mm == kvm->mm) {
9899 * Free memory regions allocated on behalf of userspace,
9900 * unless the the memory map has changed due to process exit
9903 mutex_lock(&kvm->slots_lock);
9904 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
9906 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
9908 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9909 mutex_unlock(&kvm->slots_lock);
9911 if (kvm_x86_ops.vm_destroy)
9912 kvm_x86_ops.vm_destroy(kvm);
9913 kvm_pic_destroy(kvm);
9914 kvm_ioapic_destroy(kvm);
9915 kvm_free_vcpus(kvm);
9916 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9917 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9918 kvm_mmu_uninit_vm(kvm);
9919 kvm_page_track_cleanup(kvm);
9920 kvm_hv_destroy_vm(kvm);
9923 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
9927 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9928 kvfree(slot->arch.rmap[i]);
9929 slot->arch.rmap[i] = NULL;
9934 kvfree(slot->arch.lpage_info[i - 1]);
9935 slot->arch.lpage_info[i - 1] = NULL;
9938 kvm_page_track_free_memslot(slot);
9941 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
9942 unsigned long npages)
9947 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
9948 * old arrays will be freed by __kvm_set_memory_region() if installing
9949 * the new memslot is successful.
9951 memset(&slot->arch, 0, sizeof(slot->arch));
9953 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9954 struct kvm_lpage_info *linfo;
9959 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9960 slot->base_gfn, level) + 1;
9962 slot->arch.rmap[i] =
9963 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9964 GFP_KERNEL_ACCOUNT);
9965 if (!slot->arch.rmap[i])
9970 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9974 slot->arch.lpage_info[i - 1] = linfo;
9976 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9977 linfo[0].disallow_lpage = 1;
9978 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9979 linfo[lpages - 1].disallow_lpage = 1;
9980 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9982 * If the gfn and userspace address are not aligned wrt each
9983 * other, disable large page support for this slot.
9985 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
9988 for (j = 0; j < lpages; ++j)
9989 linfo[j].disallow_lpage = 1;
9993 if (kvm_page_track_create_memslot(slot, npages))
9999 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10000 kvfree(slot->arch.rmap[i]);
10001 slot->arch.rmap[i] = NULL;
10005 kvfree(slot->arch.lpage_info[i - 1]);
10006 slot->arch.lpage_info[i - 1] = NULL;
10011 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10013 struct kvm_vcpu *vcpu;
10017 * memslots->generation has been incremented.
10018 * mmio generation may have reached its maximum value.
10020 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10022 /* Force re-initialization of steal_time cache */
10023 kvm_for_each_vcpu(i, vcpu, kvm)
10024 kvm_vcpu_kick(vcpu);
10027 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10028 struct kvm_memory_slot *memslot,
10029 const struct kvm_userspace_memory_region *mem,
10030 enum kvm_mr_change change)
10032 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10033 return kvm_alloc_memslot_metadata(memslot,
10034 mem->memory_size >> PAGE_SHIFT);
10038 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10039 struct kvm_memory_slot *new)
10041 /* Still write protect RO slot */
10042 if (new->flags & KVM_MEM_READONLY) {
10043 kvm_mmu_slot_remove_write_access(kvm, new, PT_PAGE_TABLE_LEVEL);
10048 * Call kvm_x86_ops dirty logging hooks when they are valid.
10050 * kvm_x86_ops.slot_disable_log_dirty is called when:
10052 * - KVM_MR_CREATE with dirty logging is disabled
10053 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
10055 * The reason is, in case of PML, we need to set D-bit for any slots
10056 * with dirty logging disabled in order to eliminate unnecessary GPA
10057 * logging in PML buffer (and potential PML buffer full VMEXIT). This
10058 * guarantees leaving PML enabled during guest's lifetime won't have
10059 * any additional overhead from PML when guest is running with dirty
10060 * logging disabled for memory slots.
10062 * kvm_x86_ops.slot_enable_log_dirty is called when switching new slot
10063 * to dirty logging mode.
10065 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
10067 * In case of write protect:
10069 * Write protect all pages for dirty logging.
10071 * All the sptes including the large sptes which point to this
10072 * slot are set to readonly. We can not create any new large
10073 * spte on this slot until the end of the logging.
10075 * See the comments in fast_page_fault().
10077 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10078 if (kvm_x86_ops.slot_enable_log_dirty) {
10079 kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10082 kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10083 PT_DIRECTORY_LEVEL : PT_PAGE_TABLE_LEVEL;
10086 * If we're with initial-all-set, we don't need
10087 * to write protect any small page because
10088 * they're reported as dirty already. However
10089 * we still need to write-protect huge pages
10090 * so that the page split can happen lazily on
10091 * the first write to the huge page.
10093 kvm_mmu_slot_remove_write_access(kvm, new, level);
10096 if (kvm_x86_ops.slot_disable_log_dirty)
10097 kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10101 void kvm_arch_commit_memory_region(struct kvm *kvm,
10102 const struct kvm_userspace_memory_region *mem,
10103 struct kvm_memory_slot *old,
10104 const struct kvm_memory_slot *new,
10105 enum kvm_mr_change change)
10107 if (!kvm->arch.n_requested_mmu_pages)
10108 kvm_mmu_change_mmu_pages(kvm,
10109 kvm_mmu_calculate_default_mmu_pages(kvm));
10112 * Dirty logging tracks sptes in 4k granularity, meaning that large
10113 * sptes have to be split. If live migration is successful, the guest
10114 * in the source machine will be destroyed and large sptes will be
10115 * created in the destination. However, if the guest continues to run
10116 * in the source machine (for example if live migration fails), small
10117 * sptes will remain around and cause bad performance.
10119 * Scan sptes if dirty logging has been stopped, dropping those
10120 * which can be collapsed into a single large-page spte. Later
10121 * page faults will create the large-page sptes.
10123 * There is no need to do this in any of the following cases:
10124 * CREATE: No dirty mappings will already exist.
10125 * MOVE/DELETE: The old mappings will already have been cleaned up by
10126 * kvm_arch_flush_shadow_memslot()
10128 if (change == KVM_MR_FLAGS_ONLY &&
10129 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10130 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10131 kvm_mmu_zap_collapsible_sptes(kvm, new);
10134 * Set up write protection and/or dirty logging for the new slot.
10136 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
10137 * been zapped so no dirty logging staff is needed for old slot. For
10138 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
10139 * new and it's also covered when dealing with the new slot.
10141 * FIXME: const-ify all uses of struct kvm_memory_slot.
10143 if (change != KVM_MR_DELETE)
10144 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
10146 /* Free the arrays associated with the old memslot. */
10147 if (change == KVM_MR_MOVE)
10148 kvm_arch_free_memslot(kvm, old);
10151 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10153 kvm_mmu_zap_all(kvm);
10156 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10157 struct kvm_memory_slot *slot)
10159 kvm_page_track_flush_slot(kvm, slot);
10162 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10164 return (is_guest_mode(vcpu) &&
10165 kvm_x86_ops.guest_apic_has_interrupt &&
10166 kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10169 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10171 if (!list_empty_careful(&vcpu->async_pf.done))
10174 if (kvm_apic_has_events(vcpu))
10177 if (vcpu->arch.pv.pv_unhalted)
10180 if (vcpu->arch.exception.pending)
10183 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10184 (vcpu->arch.nmi_pending &&
10185 kvm_x86_ops.nmi_allowed(vcpu)))
10188 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10189 (vcpu->arch.smi_pending && !is_smm(vcpu)))
10192 if (kvm_arch_interrupt_allowed(vcpu) &&
10193 (kvm_cpu_has_interrupt(vcpu) ||
10194 kvm_guest_apic_has_interrupt(vcpu)))
10197 if (kvm_hv_has_stimer_pending(vcpu))
10203 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10205 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10208 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10210 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10213 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10214 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10215 kvm_test_request(KVM_REQ_EVENT, vcpu))
10218 if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10224 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10226 return vcpu->arch.preempted_in_kernel;
10229 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10231 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10234 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10236 return kvm_x86_ops.interrupt_allowed(vcpu);
10239 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10241 if (is_64_bit_mode(vcpu))
10242 return kvm_rip_read(vcpu);
10243 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10244 kvm_rip_read(vcpu));
10246 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10248 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10250 return kvm_get_linear_rip(vcpu) == linear_rip;
10252 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10254 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10256 unsigned long rflags;
10258 rflags = kvm_x86_ops.get_rflags(vcpu);
10259 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10260 rflags &= ~X86_EFLAGS_TF;
10263 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10265 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10267 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10268 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10269 rflags |= X86_EFLAGS_TF;
10270 kvm_x86_ops.set_rflags(vcpu, rflags);
10273 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10275 __kvm_set_rflags(vcpu, rflags);
10276 kvm_make_request(KVM_REQ_EVENT, vcpu);
10278 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10280 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10284 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10288 r = kvm_mmu_reload(vcpu);
10292 if (!vcpu->arch.mmu->direct_map &&
10293 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10296 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10299 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10301 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10304 static inline u32 kvm_async_pf_next_probe(u32 key)
10306 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
10309 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10311 u32 key = kvm_async_pf_hash_fn(gfn);
10313 while (vcpu->arch.apf.gfns[key] != ~0)
10314 key = kvm_async_pf_next_probe(key);
10316 vcpu->arch.apf.gfns[key] = gfn;
10319 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10322 u32 key = kvm_async_pf_hash_fn(gfn);
10324 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
10325 (vcpu->arch.apf.gfns[key] != gfn &&
10326 vcpu->arch.apf.gfns[key] != ~0); i++)
10327 key = kvm_async_pf_next_probe(key);
10332 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10334 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10337 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10341 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10343 vcpu->arch.apf.gfns[i] = ~0;
10345 j = kvm_async_pf_next_probe(j);
10346 if (vcpu->arch.apf.gfns[j] == ~0)
10348 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10350 * k lies cyclically in ]i,j]
10352 * |....j i.k.| or |.k..j i...|
10354 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10355 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10360 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10363 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10367 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10370 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10374 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10376 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10379 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10380 (vcpu->arch.apf.send_user_only &&
10381 kvm_x86_ops.get_cpl(vcpu) == 0))
10387 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10389 if (unlikely(!lapic_in_kernel(vcpu) ||
10390 kvm_event_needs_reinjection(vcpu) ||
10391 vcpu->arch.exception.pending))
10394 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10398 * If interrupts are off we cannot even use an artificial
10401 return kvm_x86_ops.interrupt_allowed(vcpu);
10404 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10405 struct kvm_async_pf *work)
10407 struct x86_exception fault;
10409 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10410 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10412 if (kvm_can_deliver_async_pf(vcpu) &&
10413 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10414 fault.vector = PF_VECTOR;
10415 fault.error_code_valid = true;
10416 fault.error_code = 0;
10417 fault.nested_page_fault = false;
10418 fault.address = work->arch.token;
10419 fault.async_page_fault = true;
10420 kvm_inject_page_fault(vcpu, &fault);
10423 * It is not possible to deliver a paravirtualized asynchronous
10424 * page fault, but putting the guest in an artificial halt state
10425 * can be beneficial nevertheless: if an interrupt arrives, we
10426 * can deliver it timely and perhaps the guest will schedule
10427 * another process. When the instruction that triggered a page
10428 * fault is retried, hopefully the page will be ready in the host.
10430 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10434 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10435 struct kvm_async_pf *work)
10437 struct x86_exception fault;
10440 if (work->wakeup_all)
10441 work->arch.token = ~0; /* broadcast wakeup */
10443 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10444 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10446 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10447 !apf_get_user(vcpu, &val)) {
10448 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10449 vcpu->arch.exception.pending &&
10450 vcpu->arch.exception.nr == PF_VECTOR &&
10451 !apf_put_user(vcpu, 0)) {
10452 vcpu->arch.exception.injected = false;
10453 vcpu->arch.exception.pending = false;
10454 vcpu->arch.exception.nr = 0;
10455 vcpu->arch.exception.has_error_code = false;
10456 vcpu->arch.exception.error_code = 0;
10457 vcpu->arch.exception.has_payload = false;
10458 vcpu->arch.exception.payload = 0;
10459 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10460 fault.vector = PF_VECTOR;
10461 fault.error_code_valid = true;
10462 fault.error_code = 0;
10463 fault.nested_page_fault = false;
10464 fault.address = work->arch.token;
10465 fault.async_page_fault = true;
10466 kvm_inject_page_fault(vcpu, &fault);
10469 vcpu->arch.apf.halted = false;
10470 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10473 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10475 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10478 return kvm_can_do_async_pf(vcpu);
10481 void kvm_arch_start_assignment(struct kvm *kvm)
10483 atomic_inc(&kvm->arch.assigned_device_count);
10485 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10487 void kvm_arch_end_assignment(struct kvm *kvm)
10489 atomic_dec(&kvm->arch.assigned_device_count);
10491 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10493 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10495 return atomic_read(&kvm->arch.assigned_device_count);
10497 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10499 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10501 atomic_inc(&kvm->arch.noncoherent_dma_count);
10503 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10505 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10507 atomic_dec(&kvm->arch.noncoherent_dma_count);
10509 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10511 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10513 return atomic_read(&kvm->arch.noncoherent_dma_count);
10515 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10517 bool kvm_arch_has_irq_bypass(void)
10522 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10523 struct irq_bypass_producer *prod)
10525 struct kvm_kernel_irqfd *irqfd =
10526 container_of(cons, struct kvm_kernel_irqfd, consumer);
10528 irqfd->producer = prod;
10530 return kvm_x86_ops.update_pi_irte(irqfd->kvm,
10531 prod->irq, irqfd->gsi, 1);
10534 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10535 struct irq_bypass_producer *prod)
10538 struct kvm_kernel_irqfd *irqfd =
10539 container_of(cons, struct kvm_kernel_irqfd, consumer);
10541 WARN_ON(irqfd->producer != prod);
10542 irqfd->producer = NULL;
10545 * When producer of consumer is unregistered, we change back to
10546 * remapped mode, so we can re-use the current implementation
10547 * when the irq is masked/disabled or the consumer side (KVM
10548 * int this case doesn't want to receive the interrupts.
10550 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10552 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10553 " fails: %d\n", irqfd->consumer.token, ret);
10556 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10557 uint32_t guest_irq, bool set)
10559 return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
10562 bool kvm_vector_hashing_enabled(void)
10564 return vector_hashing;
10567 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10569 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10571 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10573 u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu)
10575 uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD;
10577 /* The STIBP bit doesn't fault even if it's not advertised */
10578 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
10579 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
10580 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10581 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
10582 !boot_cpu_has(X86_FEATURE_AMD_IBRS))
10583 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10585 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) &&
10586 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
10587 bits &= ~SPEC_CTRL_SSBD;
10588 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
10589 !boot_cpu_has(X86_FEATURE_AMD_SSBD))
10590 bits &= ~SPEC_CTRL_SSBD;
10594 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits);
10596 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10597 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10598 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10599 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10600 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10601 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10602 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10603 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10604 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10605 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10606 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10607 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10608 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10609 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10610 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10611 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10612 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10613 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10614 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10615 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
10616 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
10617 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);