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>
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/isolation.h>
58 #include <linux/mem_encrypt.h>
59 #include <linux/entry-kvm.h>
61 #include <trace/events/kvm.h>
63 #include <asm/debugreg.h>
67 #include <linux/kernel_stat.h>
68 #include <asm/fpu/internal.h> /* Ugh! */
69 #include <asm/pvclock.h>
70 #include <asm/div64.h>
71 #include <asm/irq_remapping.h>
72 #include <asm/mshyperv.h>
73 #include <asm/hypervisor.h>
74 #include <asm/tlbflush.h>
75 #include <asm/intel_pt.h>
76 #include <asm/emulate_prefix.h>
77 #include <clocksource/hyperv_timer.h>
79 #define CREATE_TRACE_POINTS
82 #define MAX_IO_MSRS 256
83 #define KVM_MAX_MCE_BANKS 32
84 u64 __read_mostly kvm_mce_cap_supported
= MCG_CTL_P
| MCG_SER_P
;
85 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported
);
87 #define emul_to_vcpu(ctxt) \
88 ((struct kvm_vcpu *)(ctxt)->vcpu)
91 * - enable syscall per default because its emulated by KVM
92 * - enable LME and LMA per default on 64 bit KVM
96 u64 __read_mostly efer_reserved_bits
= ~((u64
)(EFER_SCE
| EFER_LME
| EFER_LMA
));
98 static u64 __read_mostly efer_reserved_bits
= ~((u64
)EFER_SCE
);
101 static u64 __read_mostly cr4_reserved_bits
= CR4_RESERVED_BITS
;
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 process_smi(struct kvm_vcpu
*vcpu
);
109 static void enter_smm(struct kvm_vcpu
*vcpu
);
110 static void __kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
);
111 static void store_regs(struct kvm_vcpu
*vcpu
);
112 static int sync_regs(struct kvm_vcpu
*vcpu
);
114 struct kvm_x86_ops kvm_x86_ops __read_mostly
;
115 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
117 static bool __read_mostly ignore_msrs
= 0;
118 module_param(ignore_msrs
, bool, S_IRUGO
| S_IWUSR
);
120 static bool __read_mostly report_ignored_msrs
= true;
121 module_param(report_ignored_msrs
, bool, S_IRUGO
| S_IWUSR
);
123 unsigned int min_timer_period_us
= 200;
124 module_param(min_timer_period_us
, uint
, S_IRUGO
| S_IWUSR
);
126 static bool __read_mostly kvmclock_periodic_sync
= true;
127 module_param(kvmclock_periodic_sync
, bool, S_IRUGO
);
129 bool __read_mostly kvm_has_tsc_control
;
130 EXPORT_SYMBOL_GPL(kvm_has_tsc_control
);
131 u32 __read_mostly kvm_max_guest_tsc_khz
;
132 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz
);
133 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits
;
134 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits
);
135 u64 __read_mostly kvm_max_tsc_scaling_ratio
;
136 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio
);
137 u64 __read_mostly kvm_default_tsc_scaling_ratio
;
138 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio
);
140 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
141 static u32 __read_mostly tsc_tolerance_ppm
= 250;
142 module_param(tsc_tolerance_ppm
, uint
, S_IRUGO
| S_IWUSR
);
145 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
146 * adaptive tuning starting from default advancment of 1000ns. '0' disables
147 * advancement entirely. Any other value is used as-is and disables adaptive
148 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
150 static int __read_mostly lapic_timer_advance_ns
= -1;
151 module_param(lapic_timer_advance_ns
, int, S_IRUGO
| S_IWUSR
);
153 static bool __read_mostly vector_hashing
= true;
154 module_param(vector_hashing
, bool, S_IRUGO
);
156 bool __read_mostly enable_vmware_backdoor
= false;
157 module_param(enable_vmware_backdoor
, bool, S_IRUGO
);
158 EXPORT_SYMBOL_GPL(enable_vmware_backdoor
);
160 static bool __read_mostly force_emulation_prefix
= false;
161 module_param(force_emulation_prefix
, bool, S_IRUGO
);
163 int __read_mostly pi_inject_timer
= -1;
164 module_param(pi_inject_timer
, bint
, S_IRUGO
| S_IWUSR
);
167 * Restoring the host value for MSRs that are only consumed when running in
168 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
169 * returns to userspace, i.e. the kernel can run with the guest's value.
171 #define KVM_MAX_NR_USER_RETURN_MSRS 16
173 struct kvm_user_return_msrs_global
{
175 u32 msrs
[KVM_MAX_NR_USER_RETURN_MSRS
];
178 struct kvm_user_return_msrs
{
179 struct user_return_notifier urn
;
181 struct kvm_user_return_msr_values
{
184 } values
[KVM_MAX_NR_USER_RETURN_MSRS
];
187 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global
;
188 static struct kvm_user_return_msrs __percpu
*user_return_msrs
;
190 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
191 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
192 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
193 | XFEATURE_MASK_PKRU)
195 u64 __read_mostly host_efer
;
196 EXPORT_SYMBOL_GPL(host_efer
);
198 bool __read_mostly allow_smaller_maxphyaddr
= 0;
199 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr
);
201 u64 __read_mostly host_xss
;
202 EXPORT_SYMBOL_GPL(host_xss
);
203 u64 __read_mostly supported_xss
;
204 EXPORT_SYMBOL_GPL(supported_xss
);
206 struct kvm_stats_debugfs_item debugfs_entries
[] = {
207 VCPU_STAT("pf_fixed", pf_fixed
),
208 VCPU_STAT("pf_guest", pf_guest
),
209 VCPU_STAT("tlb_flush", tlb_flush
),
210 VCPU_STAT("invlpg", invlpg
),
211 VCPU_STAT("exits", exits
),
212 VCPU_STAT("io_exits", io_exits
),
213 VCPU_STAT("mmio_exits", mmio_exits
),
214 VCPU_STAT("signal_exits", signal_exits
),
215 VCPU_STAT("irq_window", irq_window_exits
),
216 VCPU_STAT("nmi_window", nmi_window_exits
),
217 VCPU_STAT("halt_exits", halt_exits
),
218 VCPU_STAT("halt_successful_poll", halt_successful_poll
),
219 VCPU_STAT("halt_attempted_poll", halt_attempted_poll
),
220 VCPU_STAT("halt_poll_invalid", halt_poll_invalid
),
221 VCPU_STAT("halt_wakeup", halt_wakeup
),
222 VCPU_STAT("hypercalls", hypercalls
),
223 VCPU_STAT("request_irq", request_irq_exits
),
224 VCPU_STAT("irq_exits", irq_exits
),
225 VCPU_STAT("host_state_reload", host_state_reload
),
226 VCPU_STAT("fpu_reload", fpu_reload
),
227 VCPU_STAT("insn_emulation", insn_emulation
),
228 VCPU_STAT("insn_emulation_fail", insn_emulation_fail
),
229 VCPU_STAT("irq_injections", irq_injections
),
230 VCPU_STAT("nmi_injections", nmi_injections
),
231 VCPU_STAT("req_event", req_event
),
232 VCPU_STAT("l1d_flush", l1d_flush
),
233 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns
),
234 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns
),
235 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped
),
236 VM_STAT("mmu_pte_write", mmu_pte_write
),
237 VM_STAT("mmu_pte_updated", mmu_pte_updated
),
238 VM_STAT("mmu_pde_zapped", mmu_pde_zapped
),
239 VM_STAT("mmu_flooded", mmu_flooded
),
240 VM_STAT("mmu_recycled", mmu_recycled
),
241 VM_STAT("mmu_cache_miss", mmu_cache_miss
),
242 VM_STAT("mmu_unsync", mmu_unsync
),
243 VM_STAT("remote_tlb_flush", remote_tlb_flush
),
244 VM_STAT("largepages", lpages
, .mode
= 0444),
245 VM_STAT("nx_largepages_splitted", nx_lpage_splits
, .mode
= 0444),
246 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions
),
250 u64 __read_mostly host_xcr0
;
251 u64 __read_mostly supported_xcr0
;
252 EXPORT_SYMBOL_GPL(supported_xcr0
);
254 static struct kmem_cache
*x86_fpu_cache
;
256 static struct kmem_cache
*x86_emulator_cache
;
259 * When called, it means the previous get/set msr reached an invalid msr.
260 * Return true if we want to ignore/silent this failed msr access.
262 static bool kvm_msr_ignored_check(struct kvm_vcpu
*vcpu
, u32 msr
,
263 u64 data
, bool write
)
265 const char *op
= write
? "wrmsr" : "rdmsr";
268 if (report_ignored_msrs
)
269 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
274 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
280 static struct kmem_cache
*kvm_alloc_emulator_cache(void)
282 unsigned int useroffset
= offsetof(struct x86_emulate_ctxt
, src
);
283 unsigned int size
= sizeof(struct x86_emulate_ctxt
);
285 return kmem_cache_create_usercopy("x86_emulator", size
,
286 __alignof__(struct x86_emulate_ctxt
),
287 SLAB_ACCOUNT
, useroffset
,
288 size
- useroffset
, NULL
);
291 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
);
293 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu
*vcpu
)
296 for (i
= 0; i
< ASYNC_PF_PER_VCPU
; i
++)
297 vcpu
->arch
.apf
.gfns
[i
] = ~0;
300 static void kvm_on_user_return(struct user_return_notifier
*urn
)
303 struct kvm_user_return_msrs
*msrs
304 = container_of(urn
, struct kvm_user_return_msrs
, urn
);
305 struct kvm_user_return_msr_values
*values
;
309 * Disabling irqs at this point since the following code could be
310 * interrupted and executed through kvm_arch_hardware_disable()
312 local_irq_save(flags
);
313 if (msrs
->registered
) {
314 msrs
->registered
= false;
315 user_return_notifier_unregister(urn
);
317 local_irq_restore(flags
);
318 for (slot
= 0; slot
< user_return_msrs_global
.nr
; ++slot
) {
319 values
= &msrs
->values
[slot
];
320 if (values
->host
!= values
->curr
) {
321 wrmsrl(user_return_msrs_global
.msrs
[slot
], values
->host
);
322 values
->curr
= values
->host
;
327 void kvm_define_user_return_msr(unsigned slot
, u32 msr
)
329 BUG_ON(slot
>= KVM_MAX_NR_USER_RETURN_MSRS
);
330 user_return_msrs_global
.msrs
[slot
] = msr
;
331 if (slot
>= user_return_msrs_global
.nr
)
332 user_return_msrs_global
.nr
= slot
+ 1;
334 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr
);
336 static void kvm_user_return_msr_cpu_online(void)
338 unsigned int cpu
= smp_processor_id();
339 struct kvm_user_return_msrs
*msrs
= per_cpu_ptr(user_return_msrs
, cpu
);
343 for (i
= 0; i
< user_return_msrs_global
.nr
; ++i
) {
344 rdmsrl_safe(user_return_msrs_global
.msrs
[i
], &value
);
345 msrs
->values
[i
].host
= value
;
346 msrs
->values
[i
].curr
= value
;
350 int kvm_set_user_return_msr(unsigned slot
, u64 value
, u64 mask
)
352 unsigned int cpu
= smp_processor_id();
353 struct kvm_user_return_msrs
*msrs
= per_cpu_ptr(user_return_msrs
, cpu
);
356 value
= (value
& mask
) | (msrs
->values
[slot
].host
& ~mask
);
357 if (value
== msrs
->values
[slot
].curr
)
359 err
= wrmsrl_safe(user_return_msrs_global
.msrs
[slot
], value
);
363 msrs
->values
[slot
].curr
= value
;
364 if (!msrs
->registered
) {
365 msrs
->urn
.on_user_return
= kvm_on_user_return
;
366 user_return_notifier_register(&msrs
->urn
);
367 msrs
->registered
= true;
371 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr
);
373 static void drop_user_return_notifiers(void)
375 unsigned int cpu
= smp_processor_id();
376 struct kvm_user_return_msrs
*msrs
= per_cpu_ptr(user_return_msrs
, cpu
);
378 if (msrs
->registered
)
379 kvm_on_user_return(&msrs
->urn
);
382 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
384 return vcpu
->arch
.apic_base
;
386 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
388 enum lapic_mode
kvm_get_apic_mode(struct kvm_vcpu
*vcpu
)
390 return kvm_apic_mode(kvm_get_apic_base(vcpu
));
392 EXPORT_SYMBOL_GPL(kvm_get_apic_mode
);
394 int kvm_set_apic_base(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
396 enum lapic_mode old_mode
= kvm_get_apic_mode(vcpu
);
397 enum lapic_mode new_mode
= kvm_apic_mode(msr_info
->data
);
398 u64 reserved_bits
= ((~0ULL) << cpuid_maxphyaddr(vcpu
)) | 0x2ff |
399 (guest_cpuid_has(vcpu
, X86_FEATURE_X2APIC
) ? 0 : X2APIC_ENABLE
);
401 if ((msr_info
->data
& reserved_bits
) != 0 || new_mode
== LAPIC_MODE_INVALID
)
403 if (!msr_info
->host_initiated
) {
404 if (old_mode
== LAPIC_MODE_X2APIC
&& new_mode
== LAPIC_MODE_XAPIC
)
406 if (old_mode
== LAPIC_MODE_DISABLED
&& new_mode
== LAPIC_MODE_X2APIC
)
410 kvm_lapic_set_base(vcpu
, msr_info
->data
);
411 kvm_recalculate_apic_map(vcpu
->kvm
);
414 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
416 asmlinkage __visible noinstr
void kvm_spurious_fault(void)
418 /* Fault while not rebooting. We want the trace. */
419 BUG_ON(!kvm_rebooting
);
421 EXPORT_SYMBOL_GPL(kvm_spurious_fault
);
423 #define EXCPT_BENIGN 0
424 #define EXCPT_CONTRIBUTORY 1
427 static int exception_class(int vector
)
437 return EXCPT_CONTRIBUTORY
;
444 #define EXCPT_FAULT 0
446 #define EXCPT_ABORT 2
447 #define EXCPT_INTERRUPT 3
449 static int exception_type(int vector
)
453 if (WARN_ON(vector
> 31 || vector
== NMI_VECTOR
))
454 return EXCPT_INTERRUPT
;
458 /* #DB is trap, as instruction watchpoints are handled elsewhere */
459 if (mask
& ((1 << DB_VECTOR
) | (1 << BP_VECTOR
) | (1 << OF_VECTOR
)))
462 if (mask
& ((1 << DF_VECTOR
) | (1 << MC_VECTOR
)))
465 /* Reserved exceptions will result in fault */
469 void kvm_deliver_exception_payload(struct kvm_vcpu
*vcpu
)
471 unsigned nr
= vcpu
->arch
.exception
.nr
;
472 bool has_payload
= vcpu
->arch
.exception
.has_payload
;
473 unsigned long payload
= vcpu
->arch
.exception
.payload
;
481 * "Certain debug exceptions may clear bit 0-3. The
482 * remaining contents of the DR6 register are never
483 * cleared by the processor".
485 vcpu
->arch
.dr6
&= ~DR_TRAP_BITS
;
487 * DR6.RTM is set by all #DB exceptions that don't clear it.
489 vcpu
->arch
.dr6
|= DR6_RTM
;
490 vcpu
->arch
.dr6
|= payload
;
492 * Bit 16 should be set in the payload whenever the #DB
493 * exception should clear DR6.RTM. This makes the payload
494 * compatible with the pending debug exceptions under VMX.
495 * Though not currently documented in the SDM, this also
496 * makes the payload compatible with the exit qualification
497 * for #DB exceptions under VMX.
499 vcpu
->arch
.dr6
^= payload
& DR6_RTM
;
502 * The #DB payload is defined as compatible with the 'pending
503 * debug exceptions' field under VMX, not DR6. While bit 12 is
504 * defined in the 'pending debug exceptions' field (enabled
505 * breakpoint), it is reserved and must be zero in DR6.
507 vcpu
->arch
.dr6
&= ~BIT(12);
510 vcpu
->arch
.cr2
= payload
;
514 vcpu
->arch
.exception
.has_payload
= false;
515 vcpu
->arch
.exception
.payload
= 0;
517 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload
);
519 static void kvm_multiple_exception(struct kvm_vcpu
*vcpu
,
520 unsigned nr
, bool has_error
, u32 error_code
,
521 bool has_payload
, unsigned long payload
, bool reinject
)
526 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
528 if (!vcpu
->arch
.exception
.pending
&& !vcpu
->arch
.exception
.injected
) {
530 if (has_error
&& !is_protmode(vcpu
))
534 * On vmentry, vcpu->arch.exception.pending is only
535 * true if an event injection was blocked by
536 * nested_run_pending. In that case, however,
537 * vcpu_enter_guest requests an immediate exit,
538 * and the guest shouldn't proceed far enough to
541 WARN_ON_ONCE(vcpu
->arch
.exception
.pending
);
542 vcpu
->arch
.exception
.injected
= true;
543 if (WARN_ON_ONCE(has_payload
)) {
545 * A reinjected event has already
546 * delivered its payload.
552 vcpu
->arch
.exception
.pending
= true;
553 vcpu
->arch
.exception
.injected
= false;
555 vcpu
->arch
.exception
.has_error_code
= has_error
;
556 vcpu
->arch
.exception
.nr
= nr
;
557 vcpu
->arch
.exception
.error_code
= error_code
;
558 vcpu
->arch
.exception
.has_payload
= has_payload
;
559 vcpu
->arch
.exception
.payload
= payload
;
560 if (!is_guest_mode(vcpu
))
561 kvm_deliver_exception_payload(vcpu
);
565 /* to check exception */
566 prev_nr
= vcpu
->arch
.exception
.nr
;
567 if (prev_nr
== DF_VECTOR
) {
568 /* triple fault -> shutdown */
569 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
572 class1
= exception_class(prev_nr
);
573 class2
= exception_class(nr
);
574 if ((class1
== EXCPT_CONTRIBUTORY
&& class2
== EXCPT_CONTRIBUTORY
)
575 || (class1
== EXCPT_PF
&& class2
!= EXCPT_BENIGN
)) {
577 * Generate double fault per SDM Table 5-5. Set
578 * exception.pending = true so that the double fault
579 * can trigger a nested vmexit.
581 vcpu
->arch
.exception
.pending
= true;
582 vcpu
->arch
.exception
.injected
= false;
583 vcpu
->arch
.exception
.has_error_code
= true;
584 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
585 vcpu
->arch
.exception
.error_code
= 0;
586 vcpu
->arch
.exception
.has_payload
= false;
587 vcpu
->arch
.exception
.payload
= 0;
589 /* replace previous exception with a new one in a hope
590 that instruction re-execution will regenerate lost
595 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
597 kvm_multiple_exception(vcpu
, nr
, false, 0, false, 0, false);
599 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
601 void kvm_requeue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
603 kvm_multiple_exception(vcpu
, nr
, false, 0, false, 0, true);
605 EXPORT_SYMBOL_GPL(kvm_requeue_exception
);
607 void kvm_queue_exception_p(struct kvm_vcpu
*vcpu
, unsigned nr
,
608 unsigned long payload
)
610 kvm_multiple_exception(vcpu
, nr
, false, 0, true, payload
, false);
612 EXPORT_SYMBOL_GPL(kvm_queue_exception_p
);
614 static void kvm_queue_exception_e_p(struct kvm_vcpu
*vcpu
, unsigned nr
,
615 u32 error_code
, unsigned long payload
)
617 kvm_multiple_exception(vcpu
, nr
, true, error_code
,
618 true, payload
, false);
621 int kvm_complete_insn_gp(struct kvm_vcpu
*vcpu
, int err
)
624 kvm_inject_gp(vcpu
, 0);
626 return kvm_skip_emulated_instruction(vcpu
);
630 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp
);
632 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
634 ++vcpu
->stat
.pf_guest
;
635 vcpu
->arch
.exception
.nested_apf
=
636 is_guest_mode(vcpu
) && fault
->async_page_fault
;
637 if (vcpu
->arch
.exception
.nested_apf
) {
638 vcpu
->arch
.apf
.nested_apf_token
= fault
->address
;
639 kvm_queue_exception_e(vcpu
, PF_VECTOR
, fault
->error_code
);
641 kvm_queue_exception_e_p(vcpu
, PF_VECTOR
, fault
->error_code
,
645 EXPORT_SYMBOL_GPL(kvm_inject_page_fault
);
647 bool kvm_inject_emulated_page_fault(struct kvm_vcpu
*vcpu
,
648 struct x86_exception
*fault
)
650 struct kvm_mmu
*fault_mmu
;
651 WARN_ON_ONCE(fault
->vector
!= PF_VECTOR
);
653 fault_mmu
= fault
->nested_page_fault
? vcpu
->arch
.mmu
:
657 * Invalidate the TLB entry for the faulting address, if it exists,
658 * else the access will fault indefinitely (and to emulate hardware).
660 if ((fault
->error_code
& PFERR_PRESENT_MASK
) &&
661 !(fault
->error_code
& PFERR_RSVD_MASK
))
662 kvm_mmu_invalidate_gva(vcpu
, fault_mmu
, fault
->address
,
663 fault_mmu
->root_hpa
);
665 fault_mmu
->inject_page_fault(vcpu
, fault
);
666 return fault
->nested_page_fault
;
668 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault
);
670 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
672 atomic_inc(&vcpu
->arch
.nmi_queued
);
673 kvm_make_request(KVM_REQ_NMI
, vcpu
);
675 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
677 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
679 kvm_multiple_exception(vcpu
, nr
, true, error_code
, false, 0, false);
681 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
683 void kvm_requeue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
685 kvm_multiple_exception(vcpu
, nr
, true, error_code
, false, 0, true);
687 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e
);
690 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
691 * a #GP and return false.
693 bool kvm_require_cpl(struct kvm_vcpu
*vcpu
, int required_cpl
)
695 if (kvm_x86_ops
.get_cpl(vcpu
) <= required_cpl
)
697 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
700 EXPORT_SYMBOL_GPL(kvm_require_cpl
);
702 bool kvm_require_dr(struct kvm_vcpu
*vcpu
, int dr
)
704 if ((dr
!= 4 && dr
!= 5) || !kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
707 kvm_queue_exception(vcpu
, UD_VECTOR
);
710 EXPORT_SYMBOL_GPL(kvm_require_dr
);
713 * This function will be used to read from the physical memory of the currently
714 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
715 * can read from guest physical or from the guest's guest physical memory.
717 int kvm_read_guest_page_mmu(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
,
718 gfn_t ngfn
, void *data
, int offset
, int len
,
721 struct x86_exception exception
;
725 ngpa
= gfn_to_gpa(ngfn
);
726 real_gfn
= mmu
->translate_gpa(vcpu
, ngpa
, access
, &exception
);
727 if (real_gfn
== UNMAPPED_GVA
)
730 real_gfn
= gpa_to_gfn(real_gfn
);
732 return kvm_vcpu_read_guest_page(vcpu
, real_gfn
, data
, offset
, len
);
734 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu
);
736 static int kvm_read_nested_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
737 void *data
, int offset
, int len
, u32 access
)
739 return kvm_read_guest_page_mmu(vcpu
, vcpu
->arch
.walk_mmu
, gfn
,
740 data
, offset
, len
, access
);
743 static inline u64
pdptr_rsvd_bits(struct kvm_vcpu
*vcpu
)
745 return rsvd_bits(cpuid_maxphyaddr(vcpu
), 63) | rsvd_bits(5, 8) |
750 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
752 int load_pdptrs(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
, unsigned long cr3
)
754 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
755 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
758 u64 pdpte
[ARRAY_SIZE(mmu
->pdptrs
)];
760 ret
= kvm_read_guest_page_mmu(vcpu
, mmu
, pdpt_gfn
, pdpte
,
761 offset
* sizeof(u64
), sizeof(pdpte
),
762 PFERR_USER_MASK
|PFERR_WRITE_MASK
);
767 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
768 if ((pdpte
[i
] & PT_PRESENT_MASK
) &&
769 (pdpte
[i
] & pdptr_rsvd_bits(vcpu
))) {
776 memcpy(mmu
->pdptrs
, pdpte
, sizeof(mmu
->pdptrs
));
777 kvm_register_mark_dirty(vcpu
, VCPU_EXREG_PDPTR
);
783 EXPORT_SYMBOL_GPL(load_pdptrs
);
785 bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
787 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.walk_mmu
->pdptrs
)];
792 if (!is_pae_paging(vcpu
))
795 if (!kvm_register_is_available(vcpu
, VCPU_EXREG_PDPTR
))
798 gfn
= (kvm_read_cr3(vcpu
) & 0xffffffe0ul
) >> PAGE_SHIFT
;
799 offset
= (kvm_read_cr3(vcpu
) & 0xffffffe0ul
) & (PAGE_SIZE
- 1);
800 r
= kvm_read_nested_guest_page(vcpu
, gfn
, pdpte
, offset
, sizeof(pdpte
),
801 PFERR_USER_MASK
| PFERR_WRITE_MASK
);
805 return memcmp(pdpte
, vcpu
->arch
.walk_mmu
->pdptrs
, sizeof(pdpte
)) != 0;
807 EXPORT_SYMBOL_GPL(pdptrs_changed
);
809 void kvm_post_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long old_cr0
, unsigned long cr0
)
811 unsigned long update_bits
= X86_CR0_PG
| X86_CR0_WP
;
813 if ((cr0
^ old_cr0
) & X86_CR0_PG
) {
814 kvm_clear_async_pf_completion_queue(vcpu
);
815 kvm_async_pf_hash_reset(vcpu
);
818 if ((cr0
^ old_cr0
) & update_bits
)
819 kvm_mmu_reset_context(vcpu
);
821 if (((cr0
^ old_cr0
) & X86_CR0_CD
) &&
822 kvm_arch_has_noncoherent_dma(vcpu
->kvm
) &&
823 !kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_CD_NW_CLEARED
))
824 kvm_zap_gfn_range(vcpu
->kvm
, 0, ~0ULL);
826 EXPORT_SYMBOL_GPL(kvm_post_set_cr0
);
828 int kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
830 unsigned long old_cr0
= kvm_read_cr0(vcpu
);
831 unsigned long pdptr_bits
= X86_CR0_CD
| X86_CR0_NW
| X86_CR0_PG
;
836 if (cr0
& 0xffffffff00000000UL
)
840 cr0
&= ~CR0_RESERVED_BITS
;
842 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
))
845 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
))
849 if ((vcpu
->arch
.efer
& EFER_LME
) && !is_paging(vcpu
) &&
850 (cr0
& X86_CR0_PG
)) {
855 kvm_x86_ops
.get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
860 if (!(vcpu
->arch
.efer
& EFER_LME
) && (cr0
& X86_CR0_PG
) &&
861 is_pae(vcpu
) && ((cr0
^ old_cr0
) & pdptr_bits
) &&
862 !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
)))
865 if (!(cr0
& X86_CR0_PG
) && kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
))
868 kvm_x86_ops
.set_cr0(vcpu
, cr0
);
870 kvm_post_set_cr0(vcpu
, old_cr0
, cr0
);
874 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
876 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
878 (void)kvm_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~0x0eul
) | (msw
& 0x0f));
880 EXPORT_SYMBOL_GPL(kvm_lmsw
);
882 void kvm_load_guest_xsave_state(struct kvm_vcpu
*vcpu
)
884 if (vcpu
->arch
.guest_state_protected
)
887 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
)) {
889 if (vcpu
->arch
.xcr0
!= host_xcr0
)
890 xsetbv(XCR_XFEATURE_ENABLED_MASK
, vcpu
->arch
.xcr0
);
892 if (vcpu
->arch
.xsaves_enabled
&&
893 vcpu
->arch
.ia32_xss
!= host_xss
)
894 wrmsrl(MSR_IA32_XSS
, vcpu
->arch
.ia32_xss
);
897 if (static_cpu_has(X86_FEATURE_PKU
) &&
898 (kvm_read_cr4_bits(vcpu
, X86_CR4_PKE
) ||
899 (vcpu
->arch
.xcr0
& XFEATURE_MASK_PKRU
)) &&
900 vcpu
->arch
.pkru
!= vcpu
->arch
.host_pkru
)
901 __write_pkru(vcpu
->arch
.pkru
);
903 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state
);
905 void kvm_load_host_xsave_state(struct kvm_vcpu
*vcpu
)
907 if (vcpu
->arch
.guest_state_protected
)
910 if (static_cpu_has(X86_FEATURE_PKU
) &&
911 (kvm_read_cr4_bits(vcpu
, X86_CR4_PKE
) ||
912 (vcpu
->arch
.xcr0
& XFEATURE_MASK_PKRU
))) {
913 vcpu
->arch
.pkru
= rdpkru();
914 if (vcpu
->arch
.pkru
!= vcpu
->arch
.host_pkru
)
915 __write_pkru(vcpu
->arch
.host_pkru
);
918 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
)) {
920 if (vcpu
->arch
.xcr0
!= host_xcr0
)
921 xsetbv(XCR_XFEATURE_ENABLED_MASK
, host_xcr0
);
923 if (vcpu
->arch
.xsaves_enabled
&&
924 vcpu
->arch
.ia32_xss
!= host_xss
)
925 wrmsrl(MSR_IA32_XSS
, host_xss
);
929 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state
);
931 static int __kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
934 u64 old_xcr0
= vcpu
->arch
.xcr0
;
937 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
938 if (index
!= XCR_XFEATURE_ENABLED_MASK
)
940 if (!(xcr0
& XFEATURE_MASK_FP
))
942 if ((xcr0
& XFEATURE_MASK_YMM
) && !(xcr0
& XFEATURE_MASK_SSE
))
946 * Do not allow the guest to set bits that we do not support
947 * saving. However, xcr0 bit 0 is always set, even if the
948 * emulated CPU does not support XSAVE (see fx_init).
950 valid_bits
= vcpu
->arch
.guest_supported_xcr0
| XFEATURE_MASK_FP
;
951 if (xcr0
& ~valid_bits
)
954 if ((!(xcr0
& XFEATURE_MASK_BNDREGS
)) !=
955 (!(xcr0
& XFEATURE_MASK_BNDCSR
)))
958 if (xcr0
& XFEATURE_MASK_AVX512
) {
959 if (!(xcr0
& XFEATURE_MASK_YMM
))
961 if ((xcr0
& XFEATURE_MASK_AVX512
) != XFEATURE_MASK_AVX512
)
964 vcpu
->arch
.xcr0
= xcr0
;
966 if ((xcr0
^ old_xcr0
) & XFEATURE_MASK_EXTEND
)
967 kvm_update_cpuid_runtime(vcpu
);
971 int kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
973 if (kvm_x86_ops
.get_cpl(vcpu
) != 0 ||
974 __kvm_set_xcr(vcpu
, index
, xcr
)) {
975 kvm_inject_gp(vcpu
, 0);
980 EXPORT_SYMBOL_GPL(kvm_set_xcr
);
982 bool kvm_is_valid_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
984 if (cr4
& cr4_reserved_bits
)
987 if (cr4
& vcpu
->arch
.cr4_guest_rsvd_bits
)
990 return kvm_x86_ops
.is_valid_cr4(vcpu
, cr4
);
992 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4
);
994 void kvm_post_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long old_cr4
, unsigned long cr4
)
996 unsigned long mmu_role_bits
= X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PAE
|
997 X86_CR4_SMEP
| X86_CR4_SMAP
| X86_CR4_PKE
;
999 if (((cr4
^ old_cr4
) & mmu_role_bits
) ||
1000 (!(cr4
& X86_CR4_PCIDE
) && (old_cr4
& X86_CR4_PCIDE
)))
1001 kvm_mmu_reset_context(vcpu
);
1003 EXPORT_SYMBOL_GPL(kvm_post_set_cr4
);
1005 int kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
1007 unsigned long old_cr4
= kvm_read_cr4(vcpu
);
1008 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PAE
|
1011 if (!kvm_is_valid_cr4(vcpu
, cr4
))
1014 if (is_long_mode(vcpu
)) {
1015 if (!(cr4
& X86_CR4_PAE
))
1017 if ((cr4
^ old_cr4
) & X86_CR4_LA57
)
1019 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
1020 && ((cr4
^ old_cr4
) & pdptr_bits
)
1021 && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
1022 kvm_read_cr3(vcpu
)))
1025 if ((cr4
& X86_CR4_PCIDE
) && !(old_cr4
& X86_CR4_PCIDE
)) {
1026 if (!guest_cpuid_has(vcpu
, X86_FEATURE_PCID
))
1029 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1030 if ((kvm_read_cr3(vcpu
) & X86_CR3_PCID_MASK
) || !is_long_mode(vcpu
))
1034 kvm_x86_ops
.set_cr4(vcpu
, cr4
);
1036 kvm_post_set_cr4(vcpu
, old_cr4
, cr4
);
1040 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
1042 int kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
1044 bool skip_tlb_flush
= false;
1045 #ifdef CONFIG_X86_64
1046 bool pcid_enabled
= kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
);
1049 skip_tlb_flush
= cr3
& X86_CR3_PCID_NOFLUSH
;
1050 cr3
&= ~X86_CR3_PCID_NOFLUSH
;
1054 if (cr3
== kvm_read_cr3(vcpu
) && !pdptrs_changed(vcpu
)) {
1055 if (!skip_tlb_flush
) {
1056 kvm_mmu_sync_roots(vcpu
);
1057 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT
, vcpu
);
1062 if (is_long_mode(vcpu
) &&
1063 (cr3
& vcpu
->arch
.cr3_lm_rsvd_bits
))
1065 else if (is_pae_paging(vcpu
) &&
1066 !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, cr3
))
1069 kvm_mmu_new_pgd(vcpu
, cr3
, skip_tlb_flush
, skip_tlb_flush
);
1070 vcpu
->arch
.cr3
= cr3
;
1071 kvm_register_mark_available(vcpu
, VCPU_EXREG_CR3
);
1075 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
1077 int kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
1079 if (cr8
& CR8_RESERVED_BITS
)
1081 if (lapic_in_kernel(vcpu
))
1082 kvm_lapic_set_tpr(vcpu
, cr8
);
1084 vcpu
->arch
.cr8
= cr8
;
1087 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
1089 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
1091 if (lapic_in_kernel(vcpu
))
1092 return kvm_lapic_get_cr8(vcpu
);
1094 return vcpu
->arch
.cr8
;
1096 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
1098 static void kvm_update_dr0123(struct kvm_vcpu
*vcpu
)
1102 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)) {
1103 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
1104 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
1105 vcpu
->arch
.switch_db_regs
|= KVM_DEBUGREG_RELOAD
;
1109 void kvm_update_dr7(struct kvm_vcpu
*vcpu
)
1113 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1114 dr7
= vcpu
->arch
.guest_debug_dr7
;
1116 dr7
= vcpu
->arch
.dr7
;
1117 kvm_x86_ops
.set_dr7(vcpu
, dr7
);
1118 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_BP_ENABLED
;
1119 if (dr7
& DR7_BP_EN_MASK
)
1120 vcpu
->arch
.switch_db_regs
|= KVM_DEBUGREG_BP_ENABLED
;
1122 EXPORT_SYMBOL_GPL(kvm_update_dr7
);
1124 static u64
kvm_dr6_fixed(struct kvm_vcpu
*vcpu
)
1126 u64 fixed
= DR6_FIXED_1
;
1128 if (!guest_cpuid_has(vcpu
, X86_FEATURE_RTM
))
1133 static int __kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
1135 size_t size
= ARRAY_SIZE(vcpu
->arch
.db
);
1139 vcpu
->arch
.db
[array_index_nospec(dr
, size
)] = val
;
1140 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
1141 vcpu
->arch
.eff_db
[dr
] = val
;
1145 if (!kvm_dr6_valid(val
))
1146 return -1; /* #GP */
1147 vcpu
->arch
.dr6
= (val
& DR6_VOLATILE
) | kvm_dr6_fixed(vcpu
);
1151 if (!kvm_dr7_valid(val
))
1152 return -1; /* #GP */
1153 vcpu
->arch
.dr7
= (val
& DR7_VOLATILE
) | DR7_FIXED_1
;
1154 kvm_update_dr7(vcpu
);
1161 int kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
1163 if (__kvm_set_dr(vcpu
, dr
, val
)) {
1164 kvm_inject_gp(vcpu
, 0);
1169 EXPORT_SYMBOL_GPL(kvm_set_dr
);
1171 int kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
1173 size_t size
= ARRAY_SIZE(vcpu
->arch
.db
);
1177 *val
= vcpu
->arch
.db
[array_index_nospec(dr
, size
)];
1181 *val
= vcpu
->arch
.dr6
;
1185 *val
= vcpu
->arch
.dr7
;
1190 EXPORT_SYMBOL_GPL(kvm_get_dr
);
1192 bool kvm_rdpmc(struct kvm_vcpu
*vcpu
)
1194 u32 ecx
= kvm_rcx_read(vcpu
);
1198 err
= kvm_pmu_rdpmc(vcpu
, ecx
, &data
);
1201 kvm_rax_write(vcpu
, (u32
)data
);
1202 kvm_rdx_write(vcpu
, data
>> 32);
1205 EXPORT_SYMBOL_GPL(kvm_rdpmc
);
1208 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1209 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1211 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1212 * extract the supported MSRs from the related const lists.
1213 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1214 * capabilities of the host cpu. This capabilities test skips MSRs that are
1215 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1216 * may depend on host virtualization features rather than host cpu features.
1219 static const u32 msrs_to_save_all
[] = {
1220 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
1222 #ifdef CONFIG_X86_64
1223 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
1225 MSR_IA32_TSC
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
,
1226 MSR_IA32_FEAT_CTL
, MSR_IA32_BNDCFGS
, MSR_TSC_AUX
,
1228 MSR_IA32_RTIT_CTL
, MSR_IA32_RTIT_STATUS
, MSR_IA32_RTIT_CR3_MATCH
,
1229 MSR_IA32_RTIT_OUTPUT_BASE
, MSR_IA32_RTIT_OUTPUT_MASK
,
1230 MSR_IA32_RTIT_ADDR0_A
, MSR_IA32_RTIT_ADDR0_B
,
1231 MSR_IA32_RTIT_ADDR1_A
, MSR_IA32_RTIT_ADDR1_B
,
1232 MSR_IA32_RTIT_ADDR2_A
, MSR_IA32_RTIT_ADDR2_B
,
1233 MSR_IA32_RTIT_ADDR3_A
, MSR_IA32_RTIT_ADDR3_B
,
1234 MSR_IA32_UMWAIT_CONTROL
,
1236 MSR_ARCH_PERFMON_FIXED_CTR0
, MSR_ARCH_PERFMON_FIXED_CTR1
,
1237 MSR_ARCH_PERFMON_FIXED_CTR0
+ 2, MSR_ARCH_PERFMON_FIXED_CTR0
+ 3,
1238 MSR_CORE_PERF_FIXED_CTR_CTRL
, MSR_CORE_PERF_GLOBAL_STATUS
,
1239 MSR_CORE_PERF_GLOBAL_CTRL
, MSR_CORE_PERF_GLOBAL_OVF_CTRL
,
1240 MSR_ARCH_PERFMON_PERFCTR0
, MSR_ARCH_PERFMON_PERFCTR1
,
1241 MSR_ARCH_PERFMON_PERFCTR0
+ 2, MSR_ARCH_PERFMON_PERFCTR0
+ 3,
1242 MSR_ARCH_PERFMON_PERFCTR0
+ 4, MSR_ARCH_PERFMON_PERFCTR0
+ 5,
1243 MSR_ARCH_PERFMON_PERFCTR0
+ 6, MSR_ARCH_PERFMON_PERFCTR0
+ 7,
1244 MSR_ARCH_PERFMON_PERFCTR0
+ 8, MSR_ARCH_PERFMON_PERFCTR0
+ 9,
1245 MSR_ARCH_PERFMON_PERFCTR0
+ 10, MSR_ARCH_PERFMON_PERFCTR0
+ 11,
1246 MSR_ARCH_PERFMON_PERFCTR0
+ 12, MSR_ARCH_PERFMON_PERFCTR0
+ 13,
1247 MSR_ARCH_PERFMON_PERFCTR0
+ 14, MSR_ARCH_PERFMON_PERFCTR0
+ 15,
1248 MSR_ARCH_PERFMON_PERFCTR0
+ 16, MSR_ARCH_PERFMON_PERFCTR0
+ 17,
1249 MSR_ARCH_PERFMON_EVENTSEL0
, MSR_ARCH_PERFMON_EVENTSEL1
,
1250 MSR_ARCH_PERFMON_EVENTSEL0
+ 2, MSR_ARCH_PERFMON_EVENTSEL0
+ 3,
1251 MSR_ARCH_PERFMON_EVENTSEL0
+ 4, MSR_ARCH_PERFMON_EVENTSEL0
+ 5,
1252 MSR_ARCH_PERFMON_EVENTSEL0
+ 6, MSR_ARCH_PERFMON_EVENTSEL0
+ 7,
1253 MSR_ARCH_PERFMON_EVENTSEL0
+ 8, MSR_ARCH_PERFMON_EVENTSEL0
+ 9,
1254 MSR_ARCH_PERFMON_EVENTSEL0
+ 10, MSR_ARCH_PERFMON_EVENTSEL0
+ 11,
1255 MSR_ARCH_PERFMON_EVENTSEL0
+ 12, MSR_ARCH_PERFMON_EVENTSEL0
+ 13,
1256 MSR_ARCH_PERFMON_EVENTSEL0
+ 14, MSR_ARCH_PERFMON_EVENTSEL0
+ 15,
1257 MSR_ARCH_PERFMON_EVENTSEL0
+ 16, MSR_ARCH_PERFMON_EVENTSEL0
+ 17,
1260 static u32 msrs_to_save
[ARRAY_SIZE(msrs_to_save_all
)];
1261 static unsigned num_msrs_to_save
;
1263 static const u32 emulated_msrs_all
[] = {
1264 MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
1265 MSR_KVM_SYSTEM_TIME_NEW
, MSR_KVM_WALL_CLOCK_NEW
,
1266 HV_X64_MSR_GUEST_OS_ID
, HV_X64_MSR_HYPERCALL
,
1267 HV_X64_MSR_TIME_REF_COUNT
, HV_X64_MSR_REFERENCE_TSC
,
1268 HV_X64_MSR_TSC_FREQUENCY
, HV_X64_MSR_APIC_FREQUENCY
,
1269 HV_X64_MSR_CRASH_P0
, HV_X64_MSR_CRASH_P1
, HV_X64_MSR_CRASH_P2
,
1270 HV_X64_MSR_CRASH_P3
, HV_X64_MSR_CRASH_P4
, HV_X64_MSR_CRASH_CTL
,
1272 HV_X64_MSR_VP_INDEX
,
1273 HV_X64_MSR_VP_RUNTIME
,
1274 HV_X64_MSR_SCONTROL
,
1275 HV_X64_MSR_STIMER0_CONFIG
,
1276 HV_X64_MSR_VP_ASSIST_PAGE
,
1277 HV_X64_MSR_REENLIGHTENMENT_CONTROL
, HV_X64_MSR_TSC_EMULATION_CONTROL
,
1278 HV_X64_MSR_TSC_EMULATION_STATUS
,
1279 HV_X64_MSR_SYNDBG_OPTIONS
,
1280 HV_X64_MSR_SYNDBG_CONTROL
, HV_X64_MSR_SYNDBG_STATUS
,
1281 HV_X64_MSR_SYNDBG_SEND_BUFFER
, HV_X64_MSR_SYNDBG_RECV_BUFFER
,
1282 HV_X64_MSR_SYNDBG_PENDING_BUFFER
,
1284 MSR_KVM_ASYNC_PF_EN
, MSR_KVM_STEAL_TIME
,
1285 MSR_KVM_PV_EOI_EN
, MSR_KVM_ASYNC_PF_INT
, MSR_KVM_ASYNC_PF_ACK
,
1287 MSR_IA32_TSC_ADJUST
,
1288 MSR_IA32_TSCDEADLINE
,
1289 MSR_IA32_ARCH_CAPABILITIES
,
1290 MSR_IA32_PERF_CAPABILITIES
,
1291 MSR_IA32_MISC_ENABLE
,
1292 MSR_IA32_MCG_STATUS
,
1294 MSR_IA32_MCG_EXT_CTL
,
1298 MSR_MISC_FEATURES_ENABLES
,
1299 MSR_AMD64_VIRT_SPEC_CTRL
,
1304 * The following list leaves out MSRs whose values are determined
1305 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1306 * We always support the "true" VMX control MSRs, even if the host
1307 * processor does not, so I am putting these registers here rather
1308 * than in msrs_to_save_all.
1311 MSR_IA32_VMX_TRUE_PINBASED_CTLS
,
1312 MSR_IA32_VMX_TRUE_PROCBASED_CTLS
,
1313 MSR_IA32_VMX_TRUE_EXIT_CTLS
,
1314 MSR_IA32_VMX_TRUE_ENTRY_CTLS
,
1316 MSR_IA32_VMX_CR0_FIXED0
,
1317 MSR_IA32_VMX_CR4_FIXED0
,
1318 MSR_IA32_VMX_VMCS_ENUM
,
1319 MSR_IA32_VMX_PROCBASED_CTLS2
,
1320 MSR_IA32_VMX_EPT_VPID_CAP
,
1321 MSR_IA32_VMX_VMFUNC
,
1324 MSR_KVM_POLL_CONTROL
,
1327 static u32 emulated_msrs
[ARRAY_SIZE(emulated_msrs_all
)];
1328 static unsigned num_emulated_msrs
;
1331 * List of msr numbers which are used to expose MSR-based features that
1332 * can be used by a hypervisor to validate requested CPU features.
1334 static const u32 msr_based_features_all
[] = {
1336 MSR_IA32_VMX_TRUE_PINBASED_CTLS
,
1337 MSR_IA32_VMX_PINBASED_CTLS
,
1338 MSR_IA32_VMX_TRUE_PROCBASED_CTLS
,
1339 MSR_IA32_VMX_PROCBASED_CTLS
,
1340 MSR_IA32_VMX_TRUE_EXIT_CTLS
,
1341 MSR_IA32_VMX_EXIT_CTLS
,
1342 MSR_IA32_VMX_TRUE_ENTRY_CTLS
,
1343 MSR_IA32_VMX_ENTRY_CTLS
,
1345 MSR_IA32_VMX_CR0_FIXED0
,
1346 MSR_IA32_VMX_CR0_FIXED1
,
1347 MSR_IA32_VMX_CR4_FIXED0
,
1348 MSR_IA32_VMX_CR4_FIXED1
,
1349 MSR_IA32_VMX_VMCS_ENUM
,
1350 MSR_IA32_VMX_PROCBASED_CTLS2
,
1351 MSR_IA32_VMX_EPT_VPID_CAP
,
1352 MSR_IA32_VMX_VMFUNC
,
1356 MSR_IA32_ARCH_CAPABILITIES
,
1357 MSR_IA32_PERF_CAPABILITIES
,
1360 static u32 msr_based_features
[ARRAY_SIZE(msr_based_features_all
)];
1361 static unsigned int num_msr_based_features
;
1363 static u64
kvm_get_arch_capabilities(void)
1367 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES
))
1368 rdmsrl(MSR_IA32_ARCH_CAPABILITIES
, data
);
1371 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1372 * the nested hypervisor runs with NX huge pages. If it is not,
1373 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1374 * L1 guests, so it need not worry about its own (L2) guests.
1376 data
|= ARCH_CAP_PSCHANGE_MC_NO
;
1379 * If we're doing cache flushes (either "always" or "cond")
1380 * we will do one whenever the guest does a vmlaunch/vmresume.
1381 * If an outer hypervisor is doing the cache flush for us
1382 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1383 * capability to the guest too, and if EPT is disabled we're not
1384 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1385 * require a nested hypervisor to do a flush of its own.
1387 if (l1tf_vmx_mitigation
!= VMENTER_L1D_FLUSH_NEVER
)
1388 data
|= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH
;
1390 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN
))
1391 data
|= ARCH_CAP_RDCL_NO
;
1392 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS
))
1393 data
|= ARCH_CAP_SSB_NO
;
1394 if (!boot_cpu_has_bug(X86_BUG_MDS
))
1395 data
|= ARCH_CAP_MDS_NO
;
1397 if (!boot_cpu_has(X86_FEATURE_RTM
)) {
1399 * If RTM=0 because the kernel has disabled TSX, the host might
1400 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1401 * and therefore knows that there cannot be TAA) but keep
1402 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1403 * and we want to allow migrating those guests to tsx=off hosts.
1405 data
&= ~ARCH_CAP_TAA_NO
;
1406 } else if (!boot_cpu_has_bug(X86_BUG_TAA
)) {
1407 data
|= ARCH_CAP_TAA_NO
;
1410 * Nothing to do here; we emulate TSX_CTRL if present on the
1411 * host so the guest can choose between disabling TSX or
1412 * using VERW to clear CPU buffers.
1419 static int kvm_get_msr_feature(struct kvm_msr_entry
*msr
)
1421 switch (msr
->index
) {
1422 case MSR_IA32_ARCH_CAPABILITIES
:
1423 msr
->data
= kvm_get_arch_capabilities();
1425 case MSR_IA32_UCODE_REV
:
1426 rdmsrl_safe(msr
->index
, &msr
->data
);
1429 return kvm_x86_ops
.get_msr_feature(msr
);
1434 static int do_get_msr_feature(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1436 struct kvm_msr_entry msr
;
1440 r
= kvm_get_msr_feature(&msr
);
1442 if (r
== KVM_MSR_RET_INVALID
) {
1443 /* Unconditionally clear the output for simplicity */
1445 if (kvm_msr_ignored_check(vcpu
, index
, 0, false))
1457 static bool __kvm_valid_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1459 if (efer
& EFER_FFXSR
&& !guest_cpuid_has(vcpu
, X86_FEATURE_FXSR_OPT
))
1462 if (efer
& EFER_SVME
&& !guest_cpuid_has(vcpu
, X86_FEATURE_SVM
))
1465 if (efer
& (EFER_LME
| EFER_LMA
) &&
1466 !guest_cpuid_has(vcpu
, X86_FEATURE_LM
))
1469 if (efer
& EFER_NX
&& !guest_cpuid_has(vcpu
, X86_FEATURE_NX
))
1475 bool kvm_valid_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1477 if (efer
& efer_reserved_bits
)
1480 return __kvm_valid_efer(vcpu
, efer
);
1482 EXPORT_SYMBOL_GPL(kvm_valid_efer
);
1484 static int set_efer(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
1486 u64 old_efer
= vcpu
->arch
.efer
;
1487 u64 efer
= msr_info
->data
;
1490 if (efer
& efer_reserved_bits
)
1493 if (!msr_info
->host_initiated
) {
1494 if (!__kvm_valid_efer(vcpu
, efer
))
1497 if (is_paging(vcpu
) &&
1498 (vcpu
->arch
.efer
& EFER_LME
) != (efer
& EFER_LME
))
1503 efer
|= vcpu
->arch
.efer
& EFER_LMA
;
1505 r
= kvm_x86_ops
.set_efer(vcpu
, efer
);
1511 /* Update reserved bits */
1512 if ((efer
^ old_efer
) & EFER_NX
)
1513 kvm_mmu_reset_context(vcpu
);
1518 void kvm_enable_efer_bits(u64 mask
)
1520 efer_reserved_bits
&= ~mask
;
1522 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
1524 bool kvm_msr_allowed(struct kvm_vcpu
*vcpu
, u32 index
, u32 type
)
1526 struct kvm_x86_msr_filter
*msr_filter
;
1527 struct msr_bitmap_range
*ranges
;
1528 struct kvm
*kvm
= vcpu
->kvm
;
1533 /* x2APIC MSRs do not support filtering. */
1534 if (index
>= 0x800 && index
<= 0x8ff)
1537 idx
= srcu_read_lock(&kvm
->srcu
);
1539 msr_filter
= srcu_dereference(kvm
->arch
.msr_filter
, &kvm
->srcu
);
1545 allowed
= msr_filter
->default_allow
;
1546 ranges
= msr_filter
->ranges
;
1548 for (i
= 0; i
< msr_filter
->count
; i
++) {
1549 u32 start
= ranges
[i
].base
;
1550 u32 end
= start
+ ranges
[i
].nmsrs
;
1551 u32 flags
= ranges
[i
].flags
;
1552 unsigned long *bitmap
= ranges
[i
].bitmap
;
1554 if ((index
>= start
) && (index
< end
) && (flags
& type
)) {
1555 allowed
= !!test_bit(index
- start
, bitmap
);
1561 srcu_read_unlock(&kvm
->srcu
, idx
);
1565 EXPORT_SYMBOL_GPL(kvm_msr_allowed
);
1568 * Write @data into the MSR specified by @index. Select MSR specific fault
1569 * checks are bypassed if @host_initiated is %true.
1570 * Returns 0 on success, non-0 otherwise.
1571 * Assumes vcpu_load() was already called.
1573 static int __kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 index
, u64 data
,
1574 bool host_initiated
)
1576 struct msr_data msr
;
1578 if (!host_initiated
&& !kvm_msr_allowed(vcpu
, index
, KVM_MSR_FILTER_WRITE
))
1579 return KVM_MSR_RET_FILTERED
;
1584 case MSR_KERNEL_GS_BASE
:
1587 if (is_noncanonical_address(data
, vcpu
))
1590 case MSR_IA32_SYSENTER_EIP
:
1591 case MSR_IA32_SYSENTER_ESP
:
1593 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1594 * non-canonical address is written on Intel but not on
1595 * AMD (which ignores the top 32-bits, because it does
1596 * not implement 64-bit SYSENTER).
1598 * 64-bit code should hence be able to write a non-canonical
1599 * value on AMD. Making the address canonical ensures that
1600 * vmentry does not fail on Intel after writing a non-canonical
1601 * value, and that something deterministic happens if the guest
1602 * invokes 64-bit SYSENTER.
1604 data
= get_canonical(data
, vcpu_virt_addr_bits(vcpu
));
1609 msr
.host_initiated
= host_initiated
;
1611 return kvm_x86_ops
.set_msr(vcpu
, &msr
);
1614 static int kvm_set_msr_ignored_check(struct kvm_vcpu
*vcpu
,
1615 u32 index
, u64 data
, bool host_initiated
)
1617 int ret
= __kvm_set_msr(vcpu
, index
, data
, host_initiated
);
1619 if (ret
== KVM_MSR_RET_INVALID
)
1620 if (kvm_msr_ignored_check(vcpu
, index
, data
, true))
1627 * Read the MSR specified by @index into @data. Select MSR specific fault
1628 * checks are bypassed if @host_initiated is %true.
1629 * Returns 0 on success, non-0 otherwise.
1630 * Assumes vcpu_load() was already called.
1632 int __kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 index
, u64
*data
,
1633 bool host_initiated
)
1635 struct msr_data msr
;
1638 if (!host_initiated
&& !kvm_msr_allowed(vcpu
, index
, KVM_MSR_FILTER_READ
))
1639 return KVM_MSR_RET_FILTERED
;
1642 msr
.host_initiated
= host_initiated
;
1644 ret
= kvm_x86_ops
.get_msr(vcpu
, &msr
);
1650 static int kvm_get_msr_ignored_check(struct kvm_vcpu
*vcpu
,
1651 u32 index
, u64
*data
, bool host_initiated
)
1653 int ret
= __kvm_get_msr(vcpu
, index
, data
, host_initiated
);
1655 if (ret
== KVM_MSR_RET_INVALID
) {
1656 /* Unconditionally clear *data for simplicity */
1658 if (kvm_msr_ignored_check(vcpu
, index
, 0, false))
1665 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 index
, u64
*data
)
1667 return kvm_get_msr_ignored_check(vcpu
, index
, data
, false);
1669 EXPORT_SYMBOL_GPL(kvm_get_msr
);
1671 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 index
, u64 data
)
1673 return kvm_set_msr_ignored_check(vcpu
, index
, data
, false);
1675 EXPORT_SYMBOL_GPL(kvm_set_msr
);
1677 static int complete_emulated_rdmsr(struct kvm_vcpu
*vcpu
)
1679 int err
= vcpu
->run
->msr
.error
;
1681 kvm_rax_write(vcpu
, (u32
)vcpu
->run
->msr
.data
);
1682 kvm_rdx_write(vcpu
, vcpu
->run
->msr
.data
>> 32);
1685 return kvm_x86_ops
.complete_emulated_msr(vcpu
, err
);
1688 static int complete_emulated_wrmsr(struct kvm_vcpu
*vcpu
)
1690 return kvm_x86_ops
.complete_emulated_msr(vcpu
, vcpu
->run
->msr
.error
);
1693 static u64
kvm_msr_reason(int r
)
1696 case KVM_MSR_RET_INVALID
:
1697 return KVM_MSR_EXIT_REASON_UNKNOWN
;
1698 case KVM_MSR_RET_FILTERED
:
1699 return KVM_MSR_EXIT_REASON_FILTER
;
1701 return KVM_MSR_EXIT_REASON_INVAL
;
1705 static int kvm_msr_user_space(struct kvm_vcpu
*vcpu
, u32 index
,
1706 u32 exit_reason
, u64 data
,
1707 int (*completion
)(struct kvm_vcpu
*vcpu
),
1710 u64 msr_reason
= kvm_msr_reason(r
);
1712 /* Check if the user wanted to know about this MSR fault */
1713 if (!(vcpu
->kvm
->arch
.user_space_msr_mask
& msr_reason
))
1716 vcpu
->run
->exit_reason
= exit_reason
;
1717 vcpu
->run
->msr
.error
= 0;
1718 memset(vcpu
->run
->msr
.pad
, 0, sizeof(vcpu
->run
->msr
.pad
));
1719 vcpu
->run
->msr
.reason
= msr_reason
;
1720 vcpu
->run
->msr
.index
= index
;
1721 vcpu
->run
->msr
.data
= data
;
1722 vcpu
->arch
.complete_userspace_io
= completion
;
1727 static int kvm_get_msr_user_space(struct kvm_vcpu
*vcpu
, u32 index
, int r
)
1729 return kvm_msr_user_space(vcpu
, index
, KVM_EXIT_X86_RDMSR
, 0,
1730 complete_emulated_rdmsr
, r
);
1733 static int kvm_set_msr_user_space(struct kvm_vcpu
*vcpu
, u32 index
, u64 data
, int r
)
1735 return kvm_msr_user_space(vcpu
, index
, KVM_EXIT_X86_WRMSR
, data
,
1736 complete_emulated_wrmsr
, r
);
1739 int kvm_emulate_rdmsr(struct kvm_vcpu
*vcpu
)
1741 u32 ecx
= kvm_rcx_read(vcpu
);
1745 r
= kvm_get_msr(vcpu
, ecx
, &data
);
1747 /* MSR read failed? See if we should ask user space */
1748 if (r
&& kvm_get_msr_user_space(vcpu
, ecx
, r
)) {
1749 /* Bounce to user space */
1754 trace_kvm_msr_read(ecx
, data
);
1756 kvm_rax_write(vcpu
, data
& -1u);
1757 kvm_rdx_write(vcpu
, (data
>> 32) & -1u);
1759 trace_kvm_msr_read_ex(ecx
);
1762 return kvm_x86_ops
.complete_emulated_msr(vcpu
, r
);
1764 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr
);
1766 int kvm_emulate_wrmsr(struct kvm_vcpu
*vcpu
)
1768 u32 ecx
= kvm_rcx_read(vcpu
);
1769 u64 data
= kvm_read_edx_eax(vcpu
);
1772 r
= kvm_set_msr(vcpu
, ecx
, data
);
1774 /* MSR write failed? See if we should ask user space */
1775 if (r
&& kvm_set_msr_user_space(vcpu
, ecx
, data
, r
))
1776 /* Bounce to user space */
1779 /* Signal all other negative errors to userspace */
1784 trace_kvm_msr_write(ecx
, data
);
1786 trace_kvm_msr_write_ex(ecx
, data
);
1788 return kvm_x86_ops
.complete_emulated_msr(vcpu
, r
);
1790 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr
);
1792 bool kvm_vcpu_exit_request(struct kvm_vcpu
*vcpu
)
1794 xfer_to_guest_mode_prepare();
1795 return vcpu
->mode
== EXITING_GUEST_MODE
|| kvm_request_pending(vcpu
) ||
1796 xfer_to_guest_mode_work_pending();
1798 EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request
);
1801 * The fast path for frequent and performance sensitive wrmsr emulation,
1802 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1803 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1804 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1805 * other cases which must be called after interrupts are enabled on the host.
1807 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu
*vcpu
, u64 data
)
1809 if (!lapic_in_kernel(vcpu
) || !apic_x2apic_mode(vcpu
->arch
.apic
))
1812 if (((data
& APIC_SHORT_MASK
) == APIC_DEST_NOSHORT
) &&
1813 ((data
& APIC_DEST_MASK
) == APIC_DEST_PHYSICAL
) &&
1814 ((data
& APIC_MODE_MASK
) == APIC_DM_FIXED
) &&
1815 ((u32
)(data
>> 32) != X2APIC_BROADCAST
)) {
1818 kvm_apic_send_ipi(vcpu
->arch
.apic
, (u32
)data
, (u32
)(data
>> 32));
1819 kvm_lapic_set_reg(vcpu
->arch
.apic
, APIC_ICR2
, (u32
)(data
>> 32));
1820 kvm_lapic_set_reg(vcpu
->arch
.apic
, APIC_ICR
, (u32
)data
);
1821 trace_kvm_apic_write(APIC_ICR
, (u32
)data
);
1828 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu
*vcpu
, u64 data
)
1830 if (!kvm_can_use_hv_timer(vcpu
))
1833 kvm_set_lapic_tscdeadline_msr(vcpu
, data
);
1837 fastpath_t
handle_fastpath_set_msr_irqoff(struct kvm_vcpu
*vcpu
)
1839 u32 msr
= kvm_rcx_read(vcpu
);
1841 fastpath_t ret
= EXIT_FASTPATH_NONE
;
1844 case APIC_BASE_MSR
+ (APIC_ICR
>> 4):
1845 data
= kvm_read_edx_eax(vcpu
);
1846 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu
, data
)) {
1847 kvm_skip_emulated_instruction(vcpu
);
1848 ret
= EXIT_FASTPATH_EXIT_HANDLED
;
1851 case MSR_IA32_TSCDEADLINE
:
1852 data
= kvm_read_edx_eax(vcpu
);
1853 if (!handle_fastpath_set_tscdeadline(vcpu
, data
)) {
1854 kvm_skip_emulated_instruction(vcpu
);
1855 ret
= EXIT_FASTPATH_REENTER_GUEST
;
1862 if (ret
!= EXIT_FASTPATH_NONE
)
1863 trace_kvm_msr_write(msr
, data
);
1867 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff
);
1870 * Adapt set_msr() to msr_io()'s calling convention
1872 static int do_get_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1874 return kvm_get_msr_ignored_check(vcpu
, index
, data
, true);
1877 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1879 return kvm_set_msr_ignored_check(vcpu
, index
, *data
, true);
1882 #ifdef CONFIG_X86_64
1883 struct pvclock_clock
{
1893 struct pvclock_gtod_data
{
1896 struct pvclock_clock clock
; /* extract of a clocksource struct */
1897 struct pvclock_clock raw_clock
; /* extract of a clocksource struct */
1903 static struct pvclock_gtod_data pvclock_gtod_data
;
1905 static void update_pvclock_gtod(struct timekeeper
*tk
)
1907 struct pvclock_gtod_data
*vdata
= &pvclock_gtod_data
;
1909 write_seqcount_begin(&vdata
->seq
);
1911 /* copy pvclock gtod data */
1912 vdata
->clock
.vclock_mode
= tk
->tkr_mono
.clock
->vdso_clock_mode
;
1913 vdata
->clock
.cycle_last
= tk
->tkr_mono
.cycle_last
;
1914 vdata
->clock
.mask
= tk
->tkr_mono
.mask
;
1915 vdata
->clock
.mult
= tk
->tkr_mono
.mult
;
1916 vdata
->clock
.shift
= tk
->tkr_mono
.shift
;
1917 vdata
->clock
.base_cycles
= tk
->tkr_mono
.xtime_nsec
;
1918 vdata
->clock
.offset
= tk
->tkr_mono
.base
;
1920 vdata
->raw_clock
.vclock_mode
= tk
->tkr_raw
.clock
->vdso_clock_mode
;
1921 vdata
->raw_clock
.cycle_last
= tk
->tkr_raw
.cycle_last
;
1922 vdata
->raw_clock
.mask
= tk
->tkr_raw
.mask
;
1923 vdata
->raw_clock
.mult
= tk
->tkr_raw
.mult
;
1924 vdata
->raw_clock
.shift
= tk
->tkr_raw
.shift
;
1925 vdata
->raw_clock
.base_cycles
= tk
->tkr_raw
.xtime_nsec
;
1926 vdata
->raw_clock
.offset
= tk
->tkr_raw
.base
;
1928 vdata
->wall_time_sec
= tk
->xtime_sec
;
1930 vdata
->offs_boot
= tk
->offs_boot
;
1932 write_seqcount_end(&vdata
->seq
);
1935 static s64
get_kvmclock_base_ns(void)
1937 /* Count up from boot time, but with the frequency of the raw clock. */
1938 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data
.offs_boot
));
1941 static s64
get_kvmclock_base_ns(void)
1943 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1944 return ktime_get_boottime_ns();
1948 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
1952 struct pvclock_wall_clock wc
;
1955 kvm
->arch
.wall_clock
= wall_clock
;
1960 r
= kvm_read_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1965 ++version
; /* first time write, random junk */
1969 if (kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
)))
1973 * The guest calculates current wall clock time by adding
1974 * system time (updated by kvm_guest_time_update below) to the
1975 * wall clock specified here. We do the reverse here.
1977 wall_nsec
= ktime_get_real_ns() - get_kvmclock_ns(kvm
);
1979 wc
.nsec
= do_div(wall_nsec
, 1000000000);
1980 wc
.sec
= (u32
)wall_nsec
; /* overflow in 2106 guest time */
1981 wc
.version
= version
;
1983 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
1986 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1989 static void kvm_write_system_time(struct kvm_vcpu
*vcpu
, gpa_t system_time
,
1990 bool old_msr
, bool host_initiated
)
1992 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
1994 if (vcpu
->vcpu_id
== 0 && !host_initiated
) {
1995 if (ka
->boot_vcpu_runs_old_kvmclock
!= old_msr
)
1996 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
1998 ka
->boot_vcpu_runs_old_kvmclock
= old_msr
;
2001 vcpu
->arch
.time
= system_time
;
2002 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
);
2004 /* we verify if the enable bit is set... */
2005 vcpu
->arch
.pv_time_enabled
= false;
2006 if (!(system_time
& 1))
2009 if (!kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
2010 &vcpu
->arch
.pv_time
, system_time
& ~1ULL,
2011 sizeof(struct pvclock_vcpu_time_info
)))
2012 vcpu
->arch
.pv_time_enabled
= true;
2017 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
2019 do_shl32_div32(dividend
, divisor
);
2023 static void kvm_get_time_scale(uint64_t scaled_hz
, uint64_t base_hz
,
2024 s8
*pshift
, u32
*pmultiplier
)
2032 scaled64
= scaled_hz
;
2033 while (tps64
> scaled64
*2 || tps64
& 0xffffffff00000000ULL
) {
2038 tps32
= (uint32_t)tps64
;
2039 while (tps32
<= scaled64
|| scaled64
& 0xffffffff00000000ULL
) {
2040 if (scaled64
& 0xffffffff00000000ULL
|| tps32
& 0x80000000)
2048 *pmultiplier
= div_frac(scaled64
, tps32
);
2051 #ifdef CONFIG_X86_64
2052 static atomic_t kvm_guest_has_master_clock
= ATOMIC_INIT(0);
2055 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
2056 static unsigned long max_tsc_khz
;
2058 static u32
adjust_tsc_khz(u32 khz
, s32 ppm
)
2060 u64 v
= (u64
)khz
* (1000000 + ppm
);
2065 static int set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
, bool scale
)
2069 /* Guest TSC same frequency as host TSC? */
2071 vcpu
->arch
.tsc_scaling_ratio
= kvm_default_tsc_scaling_ratio
;
2075 /* TSC scaling supported? */
2076 if (!kvm_has_tsc_control
) {
2077 if (user_tsc_khz
> tsc_khz
) {
2078 vcpu
->arch
.tsc_catchup
= 1;
2079 vcpu
->arch
.tsc_always_catchup
= 1;
2082 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2087 /* TSC scaling required - calculate ratio */
2088 ratio
= mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits
,
2089 user_tsc_khz
, tsc_khz
);
2091 if (ratio
== 0 || ratio
>= kvm_max_tsc_scaling_ratio
) {
2092 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2097 vcpu
->arch
.tsc_scaling_ratio
= ratio
;
2101 static int kvm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
)
2103 u32 thresh_lo
, thresh_hi
;
2104 int use_scaling
= 0;
2106 /* tsc_khz can be zero if TSC calibration fails */
2107 if (user_tsc_khz
== 0) {
2108 /* set tsc_scaling_ratio to a safe value */
2109 vcpu
->arch
.tsc_scaling_ratio
= kvm_default_tsc_scaling_ratio
;
2113 /* Compute a scale to convert nanoseconds in TSC cycles */
2114 kvm_get_time_scale(user_tsc_khz
* 1000LL, NSEC_PER_SEC
,
2115 &vcpu
->arch
.virtual_tsc_shift
,
2116 &vcpu
->arch
.virtual_tsc_mult
);
2117 vcpu
->arch
.virtual_tsc_khz
= user_tsc_khz
;
2120 * Compute the variation in TSC rate which is acceptable
2121 * within the range of tolerance and decide if the
2122 * rate being applied is within that bounds of the hardware
2123 * rate. If so, no scaling or compensation need be done.
2125 thresh_lo
= adjust_tsc_khz(tsc_khz
, -tsc_tolerance_ppm
);
2126 thresh_hi
= adjust_tsc_khz(tsc_khz
, tsc_tolerance_ppm
);
2127 if (user_tsc_khz
< thresh_lo
|| user_tsc_khz
> thresh_hi
) {
2128 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz
, thresh_lo
, thresh_hi
);
2131 return set_tsc_khz(vcpu
, user_tsc_khz
, use_scaling
);
2134 static u64
compute_guest_tsc(struct kvm_vcpu
*vcpu
, s64 kernel_ns
)
2136 u64 tsc
= pvclock_scale_delta(kernel_ns
-vcpu
->arch
.this_tsc_nsec
,
2137 vcpu
->arch
.virtual_tsc_mult
,
2138 vcpu
->arch
.virtual_tsc_shift
);
2139 tsc
+= vcpu
->arch
.this_tsc_write
;
2143 static inline int gtod_is_based_on_tsc(int mode
)
2145 return mode
== VDSO_CLOCKMODE_TSC
|| mode
== VDSO_CLOCKMODE_HVCLOCK
;
2148 static void kvm_track_tsc_matching(struct kvm_vcpu
*vcpu
)
2150 #ifdef CONFIG_X86_64
2152 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
2153 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
2155 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
2156 atomic_read(&vcpu
->kvm
->online_vcpus
));
2159 * Once the masterclock is enabled, always perform request in
2160 * order to update it.
2162 * In order to enable masterclock, the host clocksource must be TSC
2163 * and the vcpus need to have matched TSCs. When that happens,
2164 * perform request to enable masterclock.
2166 if (ka
->use_master_clock
||
2167 (gtod_is_based_on_tsc(gtod
->clock
.vclock_mode
) && vcpus_matched
))
2168 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
2170 trace_kvm_track_tsc(vcpu
->vcpu_id
, ka
->nr_vcpus_matched_tsc
,
2171 atomic_read(&vcpu
->kvm
->online_vcpus
),
2172 ka
->use_master_clock
, gtod
->clock
.vclock_mode
);
2177 * Multiply tsc by a fixed point number represented by ratio.
2179 * The most significant 64-N bits (mult) of ratio represent the
2180 * integral part of the fixed point number; the remaining N bits
2181 * (frac) represent the fractional part, ie. ratio represents a fixed
2182 * point number (mult + frac * 2^(-N)).
2184 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2186 static inline u64
__scale_tsc(u64 ratio
, u64 tsc
)
2188 return mul_u64_u64_shr(tsc
, ratio
, kvm_tsc_scaling_ratio_frac_bits
);
2191 u64
kvm_scale_tsc(struct kvm_vcpu
*vcpu
, u64 tsc
)
2194 u64 ratio
= vcpu
->arch
.tsc_scaling_ratio
;
2196 if (ratio
!= kvm_default_tsc_scaling_ratio
)
2197 _tsc
= __scale_tsc(ratio
, tsc
);
2201 EXPORT_SYMBOL_GPL(kvm_scale_tsc
);
2203 static u64
kvm_compute_tsc_offset(struct kvm_vcpu
*vcpu
, u64 target_tsc
)
2207 tsc
= kvm_scale_tsc(vcpu
, rdtsc());
2209 return target_tsc
- tsc
;
2212 u64
kvm_read_l1_tsc(struct kvm_vcpu
*vcpu
, u64 host_tsc
)
2214 return vcpu
->arch
.l1_tsc_offset
+ kvm_scale_tsc(vcpu
, host_tsc
);
2216 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc
);
2218 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
2220 vcpu
->arch
.l1_tsc_offset
= offset
;
2221 vcpu
->arch
.tsc_offset
= kvm_x86_ops
.write_l1_tsc_offset(vcpu
, offset
);
2224 static inline bool kvm_check_tsc_unstable(void)
2226 #ifdef CONFIG_X86_64
2228 * TSC is marked unstable when we're running on Hyper-V,
2229 * 'TSC page' clocksource is good.
2231 if (pvclock_gtod_data
.clock
.vclock_mode
== VDSO_CLOCKMODE_HVCLOCK
)
2234 return check_tsc_unstable();
2237 static void kvm_synchronize_tsc(struct kvm_vcpu
*vcpu
, u64 data
)
2239 struct kvm
*kvm
= vcpu
->kvm
;
2240 u64 offset
, ns
, elapsed
;
2241 unsigned long flags
;
2243 bool already_matched
;
2244 bool synchronizing
= false;
2246 raw_spin_lock_irqsave(&kvm
->arch
.tsc_write_lock
, flags
);
2247 offset
= kvm_compute_tsc_offset(vcpu
, data
);
2248 ns
= get_kvmclock_base_ns();
2249 elapsed
= ns
- kvm
->arch
.last_tsc_nsec
;
2251 if (vcpu
->arch
.virtual_tsc_khz
) {
2254 * detection of vcpu initialization -- need to sync
2255 * with other vCPUs. This particularly helps to keep
2256 * kvm_clock stable after CPU hotplug
2258 synchronizing
= true;
2260 u64 tsc_exp
= kvm
->arch
.last_tsc_write
+
2261 nsec_to_cycles(vcpu
, elapsed
);
2262 u64 tsc_hz
= vcpu
->arch
.virtual_tsc_khz
* 1000LL;
2264 * Special case: TSC write with a small delta (1 second)
2265 * of virtual cycle time against real time is
2266 * interpreted as an attempt to synchronize the CPU.
2268 synchronizing
= data
< tsc_exp
+ tsc_hz
&&
2269 data
+ tsc_hz
> tsc_exp
;
2274 * For a reliable TSC, we can match TSC offsets, and for an unstable
2275 * TSC, we add elapsed time in this computation. We could let the
2276 * compensation code attempt to catch up if we fall behind, but
2277 * it's better to try to match offsets from the beginning.
2279 if (synchronizing
&&
2280 vcpu
->arch
.virtual_tsc_khz
== kvm
->arch
.last_tsc_khz
) {
2281 if (!kvm_check_tsc_unstable()) {
2282 offset
= kvm
->arch
.cur_tsc_offset
;
2284 u64 delta
= nsec_to_cycles(vcpu
, elapsed
);
2286 offset
= kvm_compute_tsc_offset(vcpu
, data
);
2289 already_matched
= (vcpu
->arch
.this_tsc_generation
== kvm
->arch
.cur_tsc_generation
);
2292 * We split periods of matched TSC writes into generations.
2293 * For each generation, we track the original measured
2294 * nanosecond time, offset, and write, so if TSCs are in
2295 * sync, we can match exact offset, and if not, we can match
2296 * exact software computation in compute_guest_tsc()
2298 * These values are tracked in kvm->arch.cur_xxx variables.
2300 kvm
->arch
.cur_tsc_generation
++;
2301 kvm
->arch
.cur_tsc_nsec
= ns
;
2302 kvm
->arch
.cur_tsc_write
= data
;
2303 kvm
->arch
.cur_tsc_offset
= offset
;
2308 * We also track th most recent recorded KHZ, write and time to
2309 * allow the matching interval to be extended at each write.
2311 kvm
->arch
.last_tsc_nsec
= ns
;
2312 kvm
->arch
.last_tsc_write
= data
;
2313 kvm
->arch
.last_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
2315 vcpu
->arch
.last_guest_tsc
= data
;
2317 /* Keep track of which generation this VCPU has synchronized to */
2318 vcpu
->arch
.this_tsc_generation
= kvm
->arch
.cur_tsc_generation
;
2319 vcpu
->arch
.this_tsc_nsec
= kvm
->arch
.cur_tsc_nsec
;
2320 vcpu
->arch
.this_tsc_write
= kvm
->arch
.cur_tsc_write
;
2322 kvm_vcpu_write_tsc_offset(vcpu
, offset
);
2323 raw_spin_unlock_irqrestore(&kvm
->arch
.tsc_write_lock
, flags
);
2325 spin_lock(&kvm
->arch
.pvclock_gtod_sync_lock
);
2327 kvm
->arch
.nr_vcpus_matched_tsc
= 0;
2328 } else if (!already_matched
) {
2329 kvm
->arch
.nr_vcpus_matched_tsc
++;
2332 kvm_track_tsc_matching(vcpu
);
2333 spin_unlock(&kvm
->arch
.pvclock_gtod_sync_lock
);
2336 static inline void adjust_tsc_offset_guest(struct kvm_vcpu
*vcpu
,
2339 u64 tsc_offset
= vcpu
->arch
.l1_tsc_offset
;
2340 kvm_vcpu_write_tsc_offset(vcpu
, tsc_offset
+ adjustment
);
2343 static inline void adjust_tsc_offset_host(struct kvm_vcpu
*vcpu
, s64 adjustment
)
2345 if (vcpu
->arch
.tsc_scaling_ratio
!= kvm_default_tsc_scaling_ratio
)
2346 WARN_ON(adjustment
< 0);
2347 adjustment
= kvm_scale_tsc(vcpu
, (u64
) adjustment
);
2348 adjust_tsc_offset_guest(vcpu
, adjustment
);
2351 #ifdef CONFIG_X86_64
2353 static u64
read_tsc(void)
2355 u64 ret
= (u64
)rdtsc_ordered();
2356 u64 last
= pvclock_gtod_data
.clock
.cycle_last
;
2358 if (likely(ret
>= last
))
2362 * GCC likes to generate cmov here, but this branch is extremely
2363 * predictable (it's just a function of time and the likely is
2364 * very likely) and there's a data dependence, so force GCC
2365 * to generate a branch instead. I don't barrier() because
2366 * we don't actually need a barrier, and if this function
2367 * ever gets inlined it will generate worse code.
2373 static inline u64
vgettsc(struct pvclock_clock
*clock
, u64
*tsc_timestamp
,
2379 switch (clock
->vclock_mode
) {
2380 case VDSO_CLOCKMODE_HVCLOCK
:
2381 tsc_pg_val
= hv_read_tsc_page_tsc(hv_get_tsc_page(),
2383 if (tsc_pg_val
!= U64_MAX
) {
2384 /* TSC page valid */
2385 *mode
= VDSO_CLOCKMODE_HVCLOCK
;
2386 v
= (tsc_pg_val
- clock
->cycle_last
) &
2389 /* TSC page invalid */
2390 *mode
= VDSO_CLOCKMODE_NONE
;
2393 case VDSO_CLOCKMODE_TSC
:
2394 *mode
= VDSO_CLOCKMODE_TSC
;
2395 *tsc_timestamp
= read_tsc();
2396 v
= (*tsc_timestamp
- clock
->cycle_last
) &
2400 *mode
= VDSO_CLOCKMODE_NONE
;
2403 if (*mode
== VDSO_CLOCKMODE_NONE
)
2404 *tsc_timestamp
= v
= 0;
2406 return v
* clock
->mult
;
2409 static int do_monotonic_raw(s64
*t
, u64
*tsc_timestamp
)
2411 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
2417 seq
= read_seqcount_begin(>od
->seq
);
2418 ns
= gtod
->raw_clock
.base_cycles
;
2419 ns
+= vgettsc(>od
->raw_clock
, tsc_timestamp
, &mode
);
2420 ns
>>= gtod
->raw_clock
.shift
;
2421 ns
+= ktime_to_ns(ktime_add(gtod
->raw_clock
.offset
, gtod
->offs_boot
));
2422 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
2428 static int do_realtime(struct timespec64
*ts
, u64
*tsc_timestamp
)
2430 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
2436 seq
= read_seqcount_begin(>od
->seq
);
2437 ts
->tv_sec
= gtod
->wall_time_sec
;
2438 ns
= gtod
->clock
.base_cycles
;
2439 ns
+= vgettsc(>od
->clock
, tsc_timestamp
, &mode
);
2440 ns
>>= gtod
->clock
.shift
;
2441 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
2443 ts
->tv_sec
+= __iter_div_u64_rem(ns
, NSEC_PER_SEC
, &ns
);
2449 /* returns true if host is using TSC based clocksource */
2450 static bool kvm_get_time_and_clockread(s64
*kernel_ns
, u64
*tsc_timestamp
)
2452 /* checked again under seqlock below */
2453 if (!gtod_is_based_on_tsc(pvclock_gtod_data
.clock
.vclock_mode
))
2456 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns
,
2460 /* returns true if host is using TSC based clocksource */
2461 static bool kvm_get_walltime_and_clockread(struct timespec64
*ts
,
2464 /* checked again under seqlock below */
2465 if (!gtod_is_based_on_tsc(pvclock_gtod_data
.clock
.vclock_mode
))
2468 return gtod_is_based_on_tsc(do_realtime(ts
, tsc_timestamp
));
2474 * Assuming a stable TSC across physical CPUS, and a stable TSC
2475 * across virtual CPUs, the following condition is possible.
2476 * Each numbered line represents an event visible to both
2477 * CPUs at the next numbered event.
2479 * "timespecX" represents host monotonic time. "tscX" represents
2482 * VCPU0 on CPU0 | VCPU1 on CPU1
2484 * 1. read timespec0,tsc0
2485 * 2. | timespec1 = timespec0 + N
2487 * 3. transition to guest | transition to guest
2488 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2489 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2490 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2492 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2495 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2497 * - 0 < N - M => M < N
2499 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2500 * always the case (the difference between two distinct xtime instances
2501 * might be smaller then the difference between corresponding TSC reads,
2502 * when updating guest vcpus pvclock areas).
2504 * To avoid that problem, do not allow visibility of distinct
2505 * system_timestamp/tsc_timestamp values simultaneously: use a master
2506 * copy of host monotonic time values. Update that master copy
2509 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2513 static void pvclock_update_vm_gtod_copy(struct kvm
*kvm
)
2515 #ifdef CONFIG_X86_64
2516 struct kvm_arch
*ka
= &kvm
->arch
;
2518 bool host_tsc_clocksource
, vcpus_matched
;
2520 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
2521 atomic_read(&kvm
->online_vcpus
));
2524 * If the host uses TSC clock, then passthrough TSC as stable
2527 host_tsc_clocksource
= kvm_get_time_and_clockread(
2528 &ka
->master_kernel_ns
,
2529 &ka
->master_cycle_now
);
2531 ka
->use_master_clock
= host_tsc_clocksource
&& vcpus_matched
2532 && !ka
->backwards_tsc_observed
2533 && !ka
->boot_vcpu_runs_old_kvmclock
;
2535 if (ka
->use_master_clock
)
2536 atomic_set(&kvm_guest_has_master_clock
, 1);
2538 vclock_mode
= pvclock_gtod_data
.clock
.vclock_mode
;
2539 trace_kvm_update_master_clock(ka
->use_master_clock
, vclock_mode
,
2544 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
2546 kvm_make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
2549 static void kvm_gen_update_masterclock(struct kvm
*kvm
)
2551 #ifdef CONFIG_X86_64
2553 struct kvm_vcpu
*vcpu
;
2554 struct kvm_arch
*ka
= &kvm
->arch
;
2556 spin_lock(&ka
->pvclock_gtod_sync_lock
);
2557 kvm_make_mclock_inprogress_request(kvm
);
2558 /* no guest entries from this point */
2559 pvclock_update_vm_gtod_copy(kvm
);
2561 kvm_for_each_vcpu(i
, vcpu
, kvm
)
2562 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2564 /* guest entries allowed */
2565 kvm_for_each_vcpu(i
, vcpu
, kvm
)
2566 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS
, vcpu
);
2568 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
2572 u64
get_kvmclock_ns(struct kvm
*kvm
)
2574 struct kvm_arch
*ka
= &kvm
->arch
;
2575 struct pvclock_vcpu_time_info hv_clock
;
2578 spin_lock(&ka
->pvclock_gtod_sync_lock
);
2579 if (!ka
->use_master_clock
) {
2580 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
2581 return get_kvmclock_base_ns() + ka
->kvmclock_offset
;
2584 hv_clock
.tsc_timestamp
= ka
->master_cycle_now
;
2585 hv_clock
.system_time
= ka
->master_kernel_ns
+ ka
->kvmclock_offset
;
2586 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
2588 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2591 if (__this_cpu_read(cpu_tsc_khz
)) {
2592 kvm_get_time_scale(NSEC_PER_SEC
, __this_cpu_read(cpu_tsc_khz
) * 1000LL,
2593 &hv_clock
.tsc_shift
,
2594 &hv_clock
.tsc_to_system_mul
);
2595 ret
= __pvclock_read_cycles(&hv_clock
, rdtsc());
2597 ret
= get_kvmclock_base_ns() + ka
->kvmclock_offset
;
2604 static void kvm_setup_pvclock_page(struct kvm_vcpu
*v
)
2606 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
2607 struct pvclock_vcpu_time_info guest_hv_clock
;
2609 if (unlikely(kvm_read_guest_cached(v
->kvm
, &vcpu
->pv_time
,
2610 &guest_hv_clock
, sizeof(guest_hv_clock
))))
2613 /* This VCPU is paused, but it's legal for a guest to read another
2614 * VCPU's kvmclock, so we really have to follow the specification where
2615 * it says that version is odd if data is being modified, and even after
2618 * Version field updates must be kept separate. This is because
2619 * kvm_write_guest_cached might use a "rep movs" instruction, and
2620 * writes within a string instruction are weakly ordered. So there
2621 * are three writes overall.
2623 * As a small optimization, only write the version field in the first
2624 * and third write. The vcpu->pv_time cache is still valid, because the
2625 * version field is the first in the struct.
2627 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info
, version
) != 0);
2629 if (guest_hv_clock
.version
& 1)
2630 ++guest_hv_clock
.version
; /* first time write, random junk */
2632 vcpu
->hv_clock
.version
= guest_hv_clock
.version
+ 1;
2633 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
2635 sizeof(vcpu
->hv_clock
.version
));
2639 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2640 vcpu
->hv_clock
.flags
|= (guest_hv_clock
.flags
& PVCLOCK_GUEST_STOPPED
);
2642 if (vcpu
->pvclock_set_guest_stopped_request
) {
2643 vcpu
->hv_clock
.flags
|= PVCLOCK_GUEST_STOPPED
;
2644 vcpu
->pvclock_set_guest_stopped_request
= false;
2647 trace_kvm_pvclock_update(v
->vcpu_id
, &vcpu
->hv_clock
);
2649 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
2651 sizeof(vcpu
->hv_clock
));
2655 vcpu
->hv_clock
.version
++;
2656 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
2658 sizeof(vcpu
->hv_clock
.version
));
2661 static int kvm_guest_time_update(struct kvm_vcpu
*v
)
2663 unsigned long flags
, tgt_tsc_khz
;
2664 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
2665 struct kvm_arch
*ka
= &v
->kvm
->arch
;
2667 u64 tsc_timestamp
, host_tsc
;
2669 bool use_master_clock
;
2675 * If the host uses TSC clock, then passthrough TSC as stable
2678 spin_lock(&ka
->pvclock_gtod_sync_lock
);
2679 use_master_clock
= ka
->use_master_clock
;
2680 if (use_master_clock
) {
2681 host_tsc
= ka
->master_cycle_now
;
2682 kernel_ns
= ka
->master_kernel_ns
;
2684 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
2686 /* Keep irq disabled to prevent changes to the clock */
2687 local_irq_save(flags
);
2688 tgt_tsc_khz
= __this_cpu_read(cpu_tsc_khz
);
2689 if (unlikely(tgt_tsc_khz
== 0)) {
2690 local_irq_restore(flags
);
2691 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
2694 if (!use_master_clock
) {
2696 kernel_ns
= get_kvmclock_base_ns();
2699 tsc_timestamp
= kvm_read_l1_tsc(v
, host_tsc
);
2702 * We may have to catch up the TSC to match elapsed wall clock
2703 * time for two reasons, even if kvmclock is used.
2704 * 1) CPU could have been running below the maximum TSC rate
2705 * 2) Broken TSC compensation resets the base at each VCPU
2706 * entry to avoid unknown leaps of TSC even when running
2707 * again on the same CPU. This may cause apparent elapsed
2708 * time to disappear, and the guest to stand still or run
2711 if (vcpu
->tsc_catchup
) {
2712 u64 tsc
= compute_guest_tsc(v
, kernel_ns
);
2713 if (tsc
> tsc_timestamp
) {
2714 adjust_tsc_offset_guest(v
, tsc
- tsc_timestamp
);
2715 tsc_timestamp
= tsc
;
2719 local_irq_restore(flags
);
2721 /* With all the info we got, fill in the values */
2723 if (kvm_has_tsc_control
)
2724 tgt_tsc_khz
= kvm_scale_tsc(v
, tgt_tsc_khz
);
2726 if (unlikely(vcpu
->hw_tsc_khz
!= tgt_tsc_khz
)) {
2727 kvm_get_time_scale(NSEC_PER_SEC
, tgt_tsc_khz
* 1000LL,
2728 &vcpu
->hv_clock
.tsc_shift
,
2729 &vcpu
->hv_clock
.tsc_to_system_mul
);
2730 vcpu
->hw_tsc_khz
= tgt_tsc_khz
;
2733 vcpu
->hv_clock
.tsc_timestamp
= tsc_timestamp
;
2734 vcpu
->hv_clock
.system_time
= kernel_ns
+ v
->kvm
->arch
.kvmclock_offset
;
2735 vcpu
->last_guest_tsc
= tsc_timestamp
;
2737 /* If the host uses TSC clocksource, then it is stable */
2739 if (use_master_clock
)
2740 pvclock_flags
|= PVCLOCK_TSC_STABLE_BIT
;
2742 vcpu
->hv_clock
.flags
= pvclock_flags
;
2744 if (vcpu
->pv_time_enabled
)
2745 kvm_setup_pvclock_page(v
);
2746 if (v
== kvm_get_vcpu(v
->kvm
, 0))
2747 kvm_hv_setup_tsc_page(v
->kvm
, &vcpu
->hv_clock
);
2752 * kvmclock updates which are isolated to a given vcpu, such as
2753 * vcpu->cpu migration, should not allow system_timestamp from
2754 * the rest of the vcpus to remain static. Otherwise ntp frequency
2755 * correction applies to one vcpu's system_timestamp but not
2758 * So in those cases, request a kvmclock update for all vcpus.
2759 * We need to rate-limit these requests though, as they can
2760 * considerably slow guests that have a large number of vcpus.
2761 * The time for a remote vcpu to update its kvmclock is bound
2762 * by the delay we use to rate-limit the updates.
2765 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2767 static void kvmclock_update_fn(struct work_struct
*work
)
2770 struct delayed_work
*dwork
= to_delayed_work(work
);
2771 struct kvm_arch
*ka
= container_of(dwork
, struct kvm_arch
,
2772 kvmclock_update_work
);
2773 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
2774 struct kvm_vcpu
*vcpu
;
2776 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2777 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2778 kvm_vcpu_kick(vcpu
);
2782 static void kvm_gen_kvmclock_update(struct kvm_vcpu
*v
)
2784 struct kvm
*kvm
= v
->kvm
;
2786 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
2787 schedule_delayed_work(&kvm
->arch
.kvmclock_update_work
,
2788 KVMCLOCK_UPDATE_DELAY
);
2791 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2793 static void kvmclock_sync_fn(struct work_struct
*work
)
2795 struct delayed_work
*dwork
= to_delayed_work(work
);
2796 struct kvm_arch
*ka
= container_of(dwork
, struct kvm_arch
,
2797 kvmclock_sync_work
);
2798 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
2800 if (!kvmclock_periodic_sync
)
2803 schedule_delayed_work(&kvm
->arch
.kvmclock_update_work
, 0);
2804 schedule_delayed_work(&kvm
->arch
.kvmclock_sync_work
,
2805 KVMCLOCK_SYNC_PERIOD
);
2809 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2811 static bool can_set_mci_status(struct kvm_vcpu
*vcpu
)
2813 /* McStatusWrEn enabled? */
2814 if (guest_cpuid_is_amd_or_hygon(vcpu
))
2815 return !!(vcpu
->arch
.msr_hwcr
& BIT_ULL(18));
2820 static int set_msr_mce(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2822 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2823 unsigned bank_num
= mcg_cap
& 0xff;
2824 u32 msr
= msr_info
->index
;
2825 u64 data
= msr_info
->data
;
2828 case MSR_IA32_MCG_STATUS
:
2829 vcpu
->arch
.mcg_status
= data
;
2831 case MSR_IA32_MCG_CTL
:
2832 if (!(mcg_cap
& MCG_CTL_P
) &&
2833 (data
|| !msr_info
->host_initiated
))
2835 if (data
!= 0 && data
!= ~(u64
)0)
2837 vcpu
->arch
.mcg_ctl
= data
;
2840 if (msr
>= MSR_IA32_MC0_CTL
&&
2841 msr
< MSR_IA32_MCx_CTL(bank_num
)) {
2842 u32 offset
= array_index_nospec(
2843 msr
- MSR_IA32_MC0_CTL
,
2844 MSR_IA32_MCx_CTL(bank_num
) - MSR_IA32_MC0_CTL
);
2846 /* only 0 or all 1s can be written to IA32_MCi_CTL
2847 * some Linux kernels though clear bit 10 in bank 4 to
2848 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2849 * this to avoid an uncatched #GP in the guest
2851 if ((offset
& 0x3) == 0 &&
2852 data
!= 0 && (data
| (1 << 10)) != ~(u64
)0)
2856 if (!msr_info
->host_initiated
&&
2857 (offset
& 0x3) == 1 && data
!= 0) {
2858 if (!can_set_mci_status(vcpu
))
2862 vcpu
->arch
.mce_banks
[offset
] = data
;
2870 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
2872 struct kvm
*kvm
= vcpu
->kvm
;
2873 int lm
= is_long_mode(vcpu
);
2874 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
2875 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
2876 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
2877 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
2878 u32 page_num
= data
& ~PAGE_MASK
;
2879 u64 page_addr
= data
& PAGE_MASK
;
2882 if (page_num
>= blob_size
)
2885 page
= memdup_user(blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
);
2887 return PTR_ERR(page
);
2889 if (kvm_vcpu_write_guest(vcpu
, page_addr
, page
, PAGE_SIZE
)) {
2896 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu
*vcpu
)
2898 u64 mask
= KVM_ASYNC_PF_ENABLED
| KVM_ASYNC_PF_DELIVERY_AS_INT
;
2900 return (vcpu
->arch
.apf
.msr_en_val
& mask
) == mask
;
2903 static int kvm_pv_enable_async_pf(struct kvm_vcpu
*vcpu
, u64 data
)
2905 gpa_t gpa
= data
& ~0x3f;
2907 /* Bits 4:5 are reserved, Should be zero */
2911 if (!guest_pv_has(vcpu
, KVM_FEATURE_ASYNC_PF_VMEXIT
) &&
2912 (data
& KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT
))
2915 if (!guest_pv_has(vcpu
, KVM_FEATURE_ASYNC_PF_INT
) &&
2916 (data
& KVM_ASYNC_PF_DELIVERY_AS_INT
))
2919 if (!lapic_in_kernel(vcpu
))
2920 return data
? 1 : 0;
2922 vcpu
->arch
.apf
.msr_en_val
= data
;
2924 if (!kvm_pv_async_pf_enabled(vcpu
)) {
2925 kvm_clear_async_pf_completion_queue(vcpu
);
2926 kvm_async_pf_hash_reset(vcpu
);
2930 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, gpa
,
2934 vcpu
->arch
.apf
.send_user_only
= !(data
& KVM_ASYNC_PF_SEND_ALWAYS
);
2935 vcpu
->arch
.apf
.delivery_as_pf_vmexit
= data
& KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT
;
2937 kvm_async_pf_wakeup_all(vcpu
);
2942 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu
*vcpu
, u64 data
)
2944 /* Bits 8-63 are reserved */
2948 if (!lapic_in_kernel(vcpu
))
2951 vcpu
->arch
.apf
.msr_int_val
= data
;
2953 vcpu
->arch
.apf
.vec
= data
& KVM_ASYNC_PF_VEC_MASK
;
2958 static void kvmclock_reset(struct kvm_vcpu
*vcpu
)
2960 vcpu
->arch
.pv_time_enabled
= false;
2961 vcpu
->arch
.time
= 0;
2964 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu
*vcpu
)
2966 ++vcpu
->stat
.tlb_flush
;
2967 kvm_x86_ops
.tlb_flush_all(vcpu
);
2970 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu
*vcpu
)
2972 ++vcpu
->stat
.tlb_flush
;
2973 kvm_x86_ops
.tlb_flush_guest(vcpu
);
2976 static void record_steal_time(struct kvm_vcpu
*vcpu
)
2978 struct kvm_host_map map
;
2979 struct kvm_steal_time
*st
;
2981 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
2984 /* -EAGAIN is returned in atomic context so we can just return. */
2985 if (kvm_map_gfn(vcpu
, vcpu
->arch
.st
.msr_val
>> PAGE_SHIFT
,
2986 &map
, &vcpu
->arch
.st
.cache
, false))
2990 offset_in_page(vcpu
->arch
.st
.msr_val
& KVM_STEAL_VALID_BITS
);
2993 * Doing a TLB flush here, on the guest's behalf, can avoid
2996 if (guest_pv_has(vcpu
, KVM_FEATURE_PV_TLB_FLUSH
)) {
2997 trace_kvm_pv_tlb_flush(vcpu
->vcpu_id
,
2998 st
->preempted
& KVM_VCPU_FLUSH_TLB
);
2999 if (xchg(&st
->preempted
, 0) & KVM_VCPU_FLUSH_TLB
)
3000 kvm_vcpu_flush_tlb_guest(vcpu
);
3003 vcpu
->arch
.st
.preempted
= 0;
3005 if (st
->version
& 1)
3006 st
->version
+= 1; /* first time write, random junk */
3012 st
->steal
+= current
->sched_info
.run_delay
-
3013 vcpu
->arch
.st
.last_steal
;
3014 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
3020 kvm_unmap_gfn(vcpu
, &map
, &vcpu
->arch
.st
.cache
, true, false);
3023 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
3026 u32 msr
= msr_info
->index
;
3027 u64 data
= msr_info
->data
;
3030 case MSR_AMD64_NB_CFG
:
3031 case MSR_IA32_UCODE_WRITE
:
3032 case MSR_VM_HSAVE_PA
:
3033 case MSR_AMD64_PATCH_LOADER
:
3034 case MSR_AMD64_BU_CFG2
:
3035 case MSR_AMD64_DC_CFG
:
3036 case MSR_F15H_EX_CFG
:
3039 case MSR_IA32_UCODE_REV
:
3040 if (msr_info
->host_initiated
)
3041 vcpu
->arch
.microcode_version
= data
;
3043 case MSR_IA32_ARCH_CAPABILITIES
:
3044 if (!msr_info
->host_initiated
)
3046 vcpu
->arch
.arch_capabilities
= data
;
3048 case MSR_IA32_PERF_CAPABILITIES
: {
3049 struct kvm_msr_entry msr_ent
= {.index
= msr
, .data
= 0};
3051 if (!msr_info
->host_initiated
)
3053 if (guest_cpuid_has(vcpu
, X86_FEATURE_PDCM
) && kvm_get_msr_feature(&msr_ent
))
3055 if (data
& ~msr_ent
.data
)
3058 vcpu
->arch
.perf_capabilities
= data
;
3063 return set_efer(vcpu
, msr_info
);
3065 data
&= ~(u64
)0x40; /* ignore flush filter disable */
3066 data
&= ~(u64
)0x100; /* ignore ignne emulation enable */
3067 data
&= ~(u64
)0x8; /* ignore TLB cache disable */
3069 /* Handle McStatusWrEn */
3070 if (data
== BIT_ULL(18)) {
3071 vcpu
->arch
.msr_hwcr
= data
;
3072 } else if (data
!= 0) {
3073 vcpu_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
3078 case MSR_FAM10H_MMIO_CONF_BASE
:
3080 vcpu_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
3085 case MSR_IA32_DEBUGCTLMSR
:
3087 /* We support the non-activated case already */
3089 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
3090 /* Values other than LBR and BTF are vendor-specific,
3091 thus reserved and should throw a #GP */
3093 } else if (report_ignored_msrs
)
3094 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
3097 case 0x200 ... 0x2ff:
3098 return kvm_mtrr_set_msr(vcpu
, msr
, data
);
3099 case MSR_IA32_APICBASE
:
3100 return kvm_set_apic_base(vcpu
, msr_info
);
3101 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0xff:
3102 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
3103 case MSR_IA32_TSCDEADLINE
:
3104 kvm_set_lapic_tscdeadline_msr(vcpu
, data
);
3106 case MSR_IA32_TSC_ADJUST
:
3107 if (guest_cpuid_has(vcpu
, X86_FEATURE_TSC_ADJUST
)) {
3108 if (!msr_info
->host_initiated
) {
3109 s64 adj
= data
- vcpu
->arch
.ia32_tsc_adjust_msr
;
3110 adjust_tsc_offset_guest(vcpu
, adj
);
3112 vcpu
->arch
.ia32_tsc_adjust_msr
= data
;
3115 case MSR_IA32_MISC_ENABLE
:
3116 if (!kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT
) &&
3117 ((vcpu
->arch
.ia32_misc_enable_msr
^ data
) & MSR_IA32_MISC_ENABLE_MWAIT
)) {
3118 if (!guest_cpuid_has(vcpu
, X86_FEATURE_XMM3
))
3120 vcpu
->arch
.ia32_misc_enable_msr
= data
;
3121 kvm_update_cpuid_runtime(vcpu
);
3123 vcpu
->arch
.ia32_misc_enable_msr
= data
;
3126 case MSR_IA32_SMBASE
:
3127 if (!msr_info
->host_initiated
)
3129 vcpu
->arch
.smbase
= data
;
3131 case MSR_IA32_POWER_CTL
:
3132 vcpu
->arch
.msr_ia32_power_ctl
= data
;
3135 if (msr_info
->host_initiated
) {
3136 kvm_synchronize_tsc(vcpu
, data
);
3138 u64 adj
= kvm_compute_tsc_offset(vcpu
, data
) - vcpu
->arch
.l1_tsc_offset
;
3139 adjust_tsc_offset_guest(vcpu
, adj
);
3140 vcpu
->arch
.ia32_tsc_adjust_msr
+= adj
;
3144 if (!msr_info
->host_initiated
&&
3145 !guest_cpuid_has(vcpu
, X86_FEATURE_XSAVES
))
3148 * KVM supports exposing PT to the guest, but does not support
3149 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3150 * XSAVES/XRSTORS to save/restore PT MSRs.
3152 if (data
& ~supported_xss
)
3154 vcpu
->arch
.ia32_xss
= data
;
3157 if (!msr_info
->host_initiated
)
3159 vcpu
->arch
.smi_count
= data
;
3161 case MSR_KVM_WALL_CLOCK_NEW
:
3162 if (!guest_pv_has(vcpu
, KVM_FEATURE_CLOCKSOURCE2
))
3165 kvm_write_wall_clock(vcpu
->kvm
, data
);
3167 case MSR_KVM_WALL_CLOCK
:
3168 if (!guest_pv_has(vcpu
, KVM_FEATURE_CLOCKSOURCE
))
3171 kvm_write_wall_clock(vcpu
->kvm
, data
);
3173 case MSR_KVM_SYSTEM_TIME_NEW
:
3174 if (!guest_pv_has(vcpu
, KVM_FEATURE_CLOCKSOURCE2
))
3177 kvm_write_system_time(vcpu
, data
, false, msr_info
->host_initiated
);
3179 case MSR_KVM_SYSTEM_TIME
:
3180 if (!guest_pv_has(vcpu
, KVM_FEATURE_CLOCKSOURCE
))
3183 kvm_write_system_time(vcpu
, data
, true, msr_info
->host_initiated
);
3185 case MSR_KVM_ASYNC_PF_EN
:
3186 if (!guest_pv_has(vcpu
, KVM_FEATURE_ASYNC_PF
))
3189 if (kvm_pv_enable_async_pf(vcpu
, data
))
3192 case MSR_KVM_ASYNC_PF_INT
:
3193 if (!guest_pv_has(vcpu
, KVM_FEATURE_ASYNC_PF_INT
))
3196 if (kvm_pv_enable_async_pf_int(vcpu
, data
))
3199 case MSR_KVM_ASYNC_PF_ACK
:
3200 if (!guest_pv_has(vcpu
, KVM_FEATURE_ASYNC_PF
))
3203 vcpu
->arch
.apf
.pageready_pending
= false;
3204 kvm_check_async_pf_completion(vcpu
);
3207 case MSR_KVM_STEAL_TIME
:
3208 if (!guest_pv_has(vcpu
, KVM_FEATURE_STEAL_TIME
))
3211 if (unlikely(!sched_info_on()))
3214 if (data
& KVM_STEAL_RESERVED_MASK
)
3217 vcpu
->arch
.st
.msr_val
= data
;
3219 if (!(data
& KVM_MSR_ENABLED
))
3222 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
3225 case MSR_KVM_PV_EOI_EN
:
3226 if (!guest_pv_has(vcpu
, KVM_FEATURE_PV_EOI
))
3229 if (kvm_lapic_enable_pv_eoi(vcpu
, data
, sizeof(u8
)))
3233 case MSR_KVM_POLL_CONTROL
:
3234 if (!guest_pv_has(vcpu
, KVM_FEATURE_POLL_CONTROL
))
3237 /* only enable bit supported */
3238 if (data
& (-1ULL << 1))
3241 vcpu
->arch
.msr_kvm_poll_control
= data
;
3244 case MSR_IA32_MCG_CTL
:
3245 case MSR_IA32_MCG_STATUS
:
3246 case MSR_IA32_MC0_CTL
... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS
) - 1:
3247 return set_msr_mce(vcpu
, msr_info
);
3249 case MSR_K7_PERFCTR0
... MSR_K7_PERFCTR3
:
3250 case MSR_P6_PERFCTR0
... MSR_P6_PERFCTR1
:
3253 case MSR_K7_EVNTSEL0
... MSR_K7_EVNTSEL3
:
3254 case MSR_P6_EVNTSEL0
... MSR_P6_EVNTSEL1
:
3255 if (kvm_pmu_is_valid_msr(vcpu
, msr
))
3256 return kvm_pmu_set_msr(vcpu
, msr_info
);
3258 if (pr
|| data
!= 0)
3259 vcpu_unimpl(vcpu
, "disabled perfctr wrmsr: "
3260 "0x%x data 0x%llx\n", msr
, data
);
3262 case MSR_K7_CLK_CTL
:
3264 * Ignore all writes to this no longer documented MSR.
3265 * Writes are only relevant for old K7 processors,
3266 * all pre-dating SVM, but a recommended workaround from
3267 * AMD for these chips. It is possible to specify the
3268 * affected processor models on the command line, hence
3269 * the need to ignore the workaround.
3272 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
3273 case HV_X64_MSR_SYNDBG_CONTROL
... HV_X64_MSR_SYNDBG_PENDING_BUFFER
:
3274 case HV_X64_MSR_SYNDBG_OPTIONS
:
3275 case HV_X64_MSR_CRASH_P0
... HV_X64_MSR_CRASH_P4
:
3276 case HV_X64_MSR_CRASH_CTL
:
3277 case HV_X64_MSR_STIMER0_CONFIG
... HV_X64_MSR_STIMER3_COUNT
:
3278 case HV_X64_MSR_REENLIGHTENMENT_CONTROL
:
3279 case HV_X64_MSR_TSC_EMULATION_CONTROL
:
3280 case HV_X64_MSR_TSC_EMULATION_STATUS
:
3281 return kvm_hv_set_msr_common(vcpu
, msr
, data
,
3282 msr_info
->host_initiated
);
3283 case MSR_IA32_BBL_CR_CTL3
:
3284 /* Drop writes to this legacy MSR -- see rdmsr
3285 * counterpart for further detail.
3287 if (report_ignored_msrs
)
3288 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data 0x%llx\n",
3291 case MSR_AMD64_OSVW_ID_LENGTH
:
3292 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
3294 vcpu
->arch
.osvw
.length
= data
;
3296 case MSR_AMD64_OSVW_STATUS
:
3297 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
3299 vcpu
->arch
.osvw
.status
= data
;
3301 case MSR_PLATFORM_INFO
:
3302 if (!msr_info
->host_initiated
||
3303 (!(data
& MSR_PLATFORM_INFO_CPUID_FAULT
) &&
3304 cpuid_fault_enabled(vcpu
)))
3306 vcpu
->arch
.msr_platform_info
= data
;
3308 case MSR_MISC_FEATURES_ENABLES
:
3309 if (data
& ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT
||
3310 (data
& MSR_MISC_FEATURES_ENABLES_CPUID_FAULT
&&
3311 !supports_cpuid_fault(vcpu
)))
3313 vcpu
->arch
.msr_misc_features_enables
= data
;
3316 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
3317 return xen_hvm_config(vcpu
, data
);
3318 if (kvm_pmu_is_valid_msr(vcpu
, msr
))
3319 return kvm_pmu_set_msr(vcpu
, msr_info
);
3320 return KVM_MSR_RET_INVALID
;
3324 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
3326 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
, bool host
)
3329 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
3330 unsigned bank_num
= mcg_cap
& 0xff;
3333 case MSR_IA32_P5_MC_ADDR
:
3334 case MSR_IA32_P5_MC_TYPE
:
3337 case MSR_IA32_MCG_CAP
:
3338 data
= vcpu
->arch
.mcg_cap
;
3340 case MSR_IA32_MCG_CTL
:
3341 if (!(mcg_cap
& MCG_CTL_P
) && !host
)
3343 data
= vcpu
->arch
.mcg_ctl
;
3345 case MSR_IA32_MCG_STATUS
:
3346 data
= vcpu
->arch
.mcg_status
;
3349 if (msr
>= MSR_IA32_MC0_CTL
&&
3350 msr
< MSR_IA32_MCx_CTL(bank_num
)) {
3351 u32 offset
= array_index_nospec(
3352 msr
- MSR_IA32_MC0_CTL
,
3353 MSR_IA32_MCx_CTL(bank_num
) - MSR_IA32_MC0_CTL
);
3355 data
= vcpu
->arch
.mce_banks
[offset
];
3364 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
3366 switch (msr_info
->index
) {
3367 case MSR_IA32_PLATFORM_ID
:
3368 case MSR_IA32_EBL_CR_POWERON
:
3369 case MSR_IA32_DEBUGCTLMSR
:
3370 case MSR_IA32_LASTBRANCHFROMIP
:
3371 case MSR_IA32_LASTBRANCHTOIP
:
3372 case MSR_IA32_LASTINTFROMIP
:
3373 case MSR_IA32_LASTINTTOIP
:
3375 case MSR_K8_TSEG_ADDR
:
3376 case MSR_K8_TSEG_MASK
:
3377 case MSR_VM_HSAVE_PA
:
3378 case MSR_K8_INT_PENDING_MSG
:
3379 case MSR_AMD64_NB_CFG
:
3380 case MSR_FAM10H_MMIO_CONF_BASE
:
3381 case MSR_AMD64_BU_CFG2
:
3382 case MSR_IA32_PERF_CTL
:
3383 case MSR_AMD64_DC_CFG
:
3384 case MSR_F15H_EX_CFG
:
3386 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3387 * limit) MSRs. Just return 0, as we do not want to expose the host
3388 * data here. Do not conditionalize this on CPUID, as KVM does not do
3389 * so for existing CPU-specific MSRs.
3391 case MSR_RAPL_POWER_UNIT
:
3392 case MSR_PP0_ENERGY_STATUS
: /* Power plane 0 (core) */
3393 case MSR_PP1_ENERGY_STATUS
: /* Power plane 1 (graphics uncore) */
3394 case MSR_PKG_ENERGY_STATUS
: /* Total package */
3395 case MSR_DRAM_ENERGY_STATUS
: /* DRAM controller */
3398 case MSR_F15H_PERF_CTL0
... MSR_F15H_PERF_CTR5
:
3399 case MSR_K7_EVNTSEL0
... MSR_K7_EVNTSEL3
:
3400 case MSR_K7_PERFCTR0
... MSR_K7_PERFCTR3
:
3401 case MSR_P6_PERFCTR0
... MSR_P6_PERFCTR1
:
3402 case MSR_P6_EVNTSEL0
... MSR_P6_EVNTSEL1
:
3403 if (kvm_pmu_is_valid_msr(vcpu
, msr_info
->index
))
3404 return kvm_pmu_get_msr(vcpu
, msr_info
);
3407 case MSR_IA32_UCODE_REV
:
3408 msr_info
->data
= vcpu
->arch
.microcode_version
;
3410 case MSR_IA32_ARCH_CAPABILITIES
:
3411 if (!msr_info
->host_initiated
&&
3412 !guest_cpuid_has(vcpu
, X86_FEATURE_ARCH_CAPABILITIES
))
3414 msr_info
->data
= vcpu
->arch
.arch_capabilities
;
3416 case MSR_IA32_PERF_CAPABILITIES
:
3417 if (!msr_info
->host_initiated
&&
3418 !guest_cpuid_has(vcpu
, X86_FEATURE_PDCM
))
3420 msr_info
->data
= vcpu
->arch
.perf_capabilities
;
3422 case MSR_IA32_POWER_CTL
:
3423 msr_info
->data
= vcpu
->arch
.msr_ia32_power_ctl
;
3425 case MSR_IA32_TSC
: {
3427 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3428 * even when not intercepted. AMD manual doesn't explicitly
3429 * state this but appears to behave the same.
3431 * On userspace reads and writes, however, we unconditionally
3432 * return L1's TSC value to ensure backwards-compatible
3433 * behavior for migration.
3435 u64 tsc_offset
= msr_info
->host_initiated
? vcpu
->arch
.l1_tsc_offset
:
3436 vcpu
->arch
.tsc_offset
;
3438 msr_info
->data
= kvm_scale_tsc(vcpu
, rdtsc()) + tsc_offset
;
3442 case 0x200 ... 0x2ff:
3443 return kvm_mtrr_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
3444 case 0xcd: /* fsb frequency */
3448 * MSR_EBC_FREQUENCY_ID
3449 * Conservative value valid for even the basic CPU models.
3450 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3451 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3452 * and 266MHz for model 3, or 4. Set Core Clock
3453 * Frequency to System Bus Frequency Ratio to 1 (bits
3454 * 31:24) even though these are only valid for CPU
3455 * models > 2, however guests may end up dividing or
3456 * multiplying by zero otherwise.
3458 case MSR_EBC_FREQUENCY_ID
:
3459 msr_info
->data
= 1 << 24;
3461 case MSR_IA32_APICBASE
:
3462 msr_info
->data
= kvm_get_apic_base(vcpu
);
3464 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0xff:
3465 return kvm_x2apic_msr_read(vcpu
, msr_info
->index
, &msr_info
->data
);
3466 case MSR_IA32_TSCDEADLINE
:
3467 msr_info
->data
= kvm_get_lapic_tscdeadline_msr(vcpu
);
3469 case MSR_IA32_TSC_ADJUST
:
3470 msr_info
->data
= (u64
)vcpu
->arch
.ia32_tsc_adjust_msr
;
3472 case MSR_IA32_MISC_ENABLE
:
3473 msr_info
->data
= vcpu
->arch
.ia32_misc_enable_msr
;
3475 case MSR_IA32_SMBASE
:
3476 if (!msr_info
->host_initiated
)
3478 msr_info
->data
= vcpu
->arch
.smbase
;
3481 msr_info
->data
= vcpu
->arch
.smi_count
;
3483 case MSR_IA32_PERF_STATUS
:
3484 /* TSC increment by tick */
3485 msr_info
->data
= 1000ULL;
3486 /* CPU multiplier */
3487 msr_info
->data
|= (((uint64_t)4ULL) << 40);
3490 msr_info
->data
= vcpu
->arch
.efer
;
3492 case MSR_KVM_WALL_CLOCK
:
3493 if (!guest_pv_has(vcpu
, KVM_FEATURE_CLOCKSOURCE
))
3496 msr_info
->data
= vcpu
->kvm
->arch
.wall_clock
;
3498 case MSR_KVM_WALL_CLOCK_NEW
:
3499 if (!guest_pv_has(vcpu
, KVM_FEATURE_CLOCKSOURCE2
))
3502 msr_info
->data
= vcpu
->kvm
->arch
.wall_clock
;
3504 case MSR_KVM_SYSTEM_TIME
:
3505 if (!guest_pv_has(vcpu
, KVM_FEATURE_CLOCKSOURCE
))
3508 msr_info
->data
= vcpu
->arch
.time
;
3510 case MSR_KVM_SYSTEM_TIME_NEW
:
3511 if (!guest_pv_has(vcpu
, KVM_FEATURE_CLOCKSOURCE2
))
3514 msr_info
->data
= vcpu
->arch
.time
;
3516 case MSR_KVM_ASYNC_PF_EN
:
3517 if (!guest_pv_has(vcpu
, KVM_FEATURE_ASYNC_PF
))
3520 msr_info
->data
= vcpu
->arch
.apf
.msr_en_val
;
3522 case MSR_KVM_ASYNC_PF_INT
:
3523 if (!guest_pv_has(vcpu
, KVM_FEATURE_ASYNC_PF_INT
))
3526 msr_info
->data
= vcpu
->arch
.apf
.msr_int_val
;
3528 case MSR_KVM_ASYNC_PF_ACK
:
3529 if (!guest_pv_has(vcpu
, KVM_FEATURE_ASYNC_PF
))
3534 case MSR_KVM_STEAL_TIME
:
3535 if (!guest_pv_has(vcpu
, KVM_FEATURE_STEAL_TIME
))
3538 msr_info
->data
= vcpu
->arch
.st
.msr_val
;
3540 case MSR_KVM_PV_EOI_EN
:
3541 if (!guest_pv_has(vcpu
, KVM_FEATURE_PV_EOI
))
3544 msr_info
->data
= vcpu
->arch
.pv_eoi
.msr_val
;
3546 case MSR_KVM_POLL_CONTROL
:
3547 if (!guest_pv_has(vcpu
, KVM_FEATURE_POLL_CONTROL
))
3550 msr_info
->data
= vcpu
->arch
.msr_kvm_poll_control
;
3552 case MSR_IA32_P5_MC_ADDR
:
3553 case MSR_IA32_P5_MC_TYPE
:
3554 case MSR_IA32_MCG_CAP
:
3555 case MSR_IA32_MCG_CTL
:
3556 case MSR_IA32_MCG_STATUS
:
3557 case MSR_IA32_MC0_CTL
... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS
) - 1:
3558 return get_msr_mce(vcpu
, msr_info
->index
, &msr_info
->data
,
3559 msr_info
->host_initiated
);
3561 if (!msr_info
->host_initiated
&&
3562 !guest_cpuid_has(vcpu
, X86_FEATURE_XSAVES
))
3564 msr_info
->data
= vcpu
->arch
.ia32_xss
;
3566 case MSR_K7_CLK_CTL
:
3568 * Provide expected ramp-up count for K7. All other
3569 * are set to zero, indicating minimum divisors for
3572 * This prevents guest kernels on AMD host with CPU
3573 * type 6, model 8 and higher from exploding due to
3574 * the rdmsr failing.
3576 msr_info
->data
= 0x20000000;
3578 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
3579 case HV_X64_MSR_SYNDBG_CONTROL
... HV_X64_MSR_SYNDBG_PENDING_BUFFER
:
3580 case HV_X64_MSR_SYNDBG_OPTIONS
:
3581 case HV_X64_MSR_CRASH_P0
... HV_X64_MSR_CRASH_P4
:
3582 case HV_X64_MSR_CRASH_CTL
:
3583 case HV_X64_MSR_STIMER0_CONFIG
... HV_X64_MSR_STIMER3_COUNT
:
3584 case HV_X64_MSR_REENLIGHTENMENT_CONTROL
:
3585 case HV_X64_MSR_TSC_EMULATION_CONTROL
:
3586 case HV_X64_MSR_TSC_EMULATION_STATUS
:
3587 return kvm_hv_get_msr_common(vcpu
,
3588 msr_info
->index
, &msr_info
->data
,
3589 msr_info
->host_initiated
);
3590 case MSR_IA32_BBL_CR_CTL3
:
3591 /* This legacy MSR exists but isn't fully documented in current
3592 * silicon. It is however accessed by winxp in very narrow
3593 * scenarios where it sets bit #19, itself documented as
3594 * a "reserved" bit. Best effort attempt to source coherent
3595 * read data here should the balance of the register be
3596 * interpreted by the guest:
3598 * L2 cache control register 3: 64GB range, 256KB size,
3599 * enabled, latency 0x1, configured
3601 msr_info
->data
= 0xbe702111;
3603 case MSR_AMD64_OSVW_ID_LENGTH
:
3604 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
3606 msr_info
->data
= vcpu
->arch
.osvw
.length
;
3608 case MSR_AMD64_OSVW_STATUS
:
3609 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
3611 msr_info
->data
= vcpu
->arch
.osvw
.status
;
3613 case MSR_PLATFORM_INFO
:
3614 if (!msr_info
->host_initiated
&&
3615 !vcpu
->kvm
->arch
.guest_can_read_msr_platform_info
)
3617 msr_info
->data
= vcpu
->arch
.msr_platform_info
;
3619 case MSR_MISC_FEATURES_ENABLES
:
3620 msr_info
->data
= vcpu
->arch
.msr_misc_features_enables
;
3623 msr_info
->data
= vcpu
->arch
.msr_hwcr
;
3626 if (kvm_pmu_is_valid_msr(vcpu
, msr_info
->index
))
3627 return kvm_pmu_get_msr(vcpu
, msr_info
);
3628 return KVM_MSR_RET_INVALID
;
3632 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
3635 * Read or write a bunch of msrs. All parameters are kernel addresses.
3637 * @return number of msrs set successfully.
3639 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
3640 struct kvm_msr_entry
*entries
,
3641 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
3642 unsigned index
, u64
*data
))
3646 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
3647 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
3654 * Read or write a bunch of msrs. Parameters are user addresses.
3656 * @return number of msrs set successfully.
3658 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
3659 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
3660 unsigned index
, u64
*data
),
3663 struct kvm_msrs msrs
;
3664 struct kvm_msr_entry
*entries
;
3669 if (copy_from_user(&msrs
, user_msrs
, sizeof(msrs
)))
3673 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
3676 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
3677 entries
= memdup_user(user_msrs
->entries
, size
);
3678 if (IS_ERR(entries
)) {
3679 r
= PTR_ERR(entries
);
3683 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
3688 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
3699 static inline bool kvm_can_mwait_in_guest(void)
3701 return boot_cpu_has(X86_FEATURE_MWAIT
) &&
3702 !boot_cpu_has_bug(X86_BUG_MONITOR
) &&
3703 boot_cpu_has(X86_FEATURE_ARAT
);
3706 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu
*vcpu
,
3707 struct kvm_cpuid2 __user
*cpuid_arg
)
3709 struct kvm_cpuid2 cpuid
;
3713 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof(cpuid
)))
3716 r
= kvm_get_hv_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
3721 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof(cpuid
)))
3727 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
3732 case KVM_CAP_IRQCHIP
:
3734 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
3735 case KVM_CAP_SET_TSS_ADDR
:
3736 case KVM_CAP_EXT_CPUID
:
3737 case KVM_CAP_EXT_EMUL_CPUID
:
3738 case KVM_CAP_CLOCKSOURCE
:
3740 case KVM_CAP_NOP_IO_DELAY
:
3741 case KVM_CAP_MP_STATE
:
3742 case KVM_CAP_SYNC_MMU
:
3743 case KVM_CAP_USER_NMI
:
3744 case KVM_CAP_REINJECT_CONTROL
:
3745 case KVM_CAP_IRQ_INJECT_STATUS
:
3746 case KVM_CAP_IOEVENTFD
:
3747 case KVM_CAP_IOEVENTFD_NO_LENGTH
:
3749 case KVM_CAP_PIT_STATE2
:
3750 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
3751 case KVM_CAP_XEN_HVM
:
3752 case KVM_CAP_VCPU_EVENTS
:
3753 case KVM_CAP_HYPERV
:
3754 case KVM_CAP_HYPERV_VAPIC
:
3755 case KVM_CAP_HYPERV_SPIN
:
3756 case KVM_CAP_HYPERV_SYNIC
:
3757 case KVM_CAP_HYPERV_SYNIC2
:
3758 case KVM_CAP_HYPERV_VP_INDEX
:
3759 case KVM_CAP_HYPERV_EVENTFD
:
3760 case KVM_CAP_HYPERV_TLBFLUSH
:
3761 case KVM_CAP_HYPERV_SEND_IPI
:
3762 case KVM_CAP_HYPERV_CPUID
:
3763 case KVM_CAP_SYS_HYPERV_CPUID
:
3764 case KVM_CAP_PCI_SEGMENT
:
3765 case KVM_CAP_DEBUGREGS
:
3766 case KVM_CAP_X86_ROBUST_SINGLESTEP
:
3768 case KVM_CAP_ASYNC_PF
:
3769 case KVM_CAP_ASYNC_PF_INT
:
3770 case KVM_CAP_GET_TSC_KHZ
:
3771 case KVM_CAP_KVMCLOCK_CTRL
:
3772 case KVM_CAP_READONLY_MEM
:
3773 case KVM_CAP_HYPERV_TIME
:
3774 case KVM_CAP_IOAPIC_POLARITY_IGNORED
:
3775 case KVM_CAP_TSC_DEADLINE_TIMER
:
3776 case KVM_CAP_DISABLE_QUIRKS
:
3777 case KVM_CAP_SET_BOOT_CPU_ID
:
3778 case KVM_CAP_SPLIT_IRQCHIP
:
3779 case KVM_CAP_IMMEDIATE_EXIT
:
3780 case KVM_CAP_PMU_EVENT_FILTER
:
3781 case KVM_CAP_GET_MSR_FEATURES
:
3782 case KVM_CAP_MSR_PLATFORM_INFO
:
3783 case KVM_CAP_EXCEPTION_PAYLOAD
:
3784 case KVM_CAP_SET_GUEST_DEBUG
:
3785 case KVM_CAP_LAST_CPU
:
3786 case KVM_CAP_X86_USER_SPACE_MSR
:
3787 case KVM_CAP_X86_MSR_FILTER
:
3788 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID
:
3791 case KVM_CAP_SYNC_REGS
:
3792 r
= KVM_SYNC_X86_VALID_FIELDS
;
3794 case KVM_CAP_ADJUST_CLOCK
:
3795 r
= KVM_CLOCK_TSC_STABLE
;
3797 case KVM_CAP_X86_DISABLE_EXITS
:
3798 r
|= KVM_X86_DISABLE_EXITS_HLT
| KVM_X86_DISABLE_EXITS_PAUSE
|
3799 KVM_X86_DISABLE_EXITS_CSTATE
;
3800 if(kvm_can_mwait_in_guest())
3801 r
|= KVM_X86_DISABLE_EXITS_MWAIT
;
3803 case KVM_CAP_X86_SMM
:
3804 /* SMBASE is usually relocated above 1M on modern chipsets,
3805 * and SMM handlers might indeed rely on 4G segment limits,
3806 * so do not report SMM to be available if real mode is
3807 * emulated via vm86 mode. Still, do not go to great lengths
3808 * to avoid userspace's usage of the feature, because it is a
3809 * fringe case that is not enabled except via specific settings
3810 * of the module parameters.
3812 r
= kvm_x86_ops
.has_emulated_msr(kvm
, MSR_IA32_SMBASE
);
3815 r
= !kvm_x86_ops
.cpu_has_accelerated_tpr();
3817 case KVM_CAP_NR_VCPUS
:
3818 r
= KVM_SOFT_MAX_VCPUS
;
3820 case KVM_CAP_MAX_VCPUS
:
3823 case KVM_CAP_MAX_VCPU_ID
:
3824 r
= KVM_MAX_VCPU_ID
;
3826 case KVM_CAP_PV_MMU
: /* obsolete */
3830 r
= KVM_MAX_MCE_BANKS
;
3833 r
= boot_cpu_has(X86_FEATURE_XSAVE
);
3835 case KVM_CAP_TSC_CONTROL
:
3836 r
= kvm_has_tsc_control
;
3838 case KVM_CAP_X2APIC_API
:
3839 r
= KVM_X2APIC_API_VALID_FLAGS
;
3841 case KVM_CAP_NESTED_STATE
:
3842 r
= kvm_x86_ops
.nested_ops
->get_state
?
3843 kvm_x86_ops
.nested_ops
->get_state(NULL
, NULL
, 0) : 0;
3845 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH
:
3846 r
= kvm_x86_ops
.enable_direct_tlbflush
!= NULL
;
3848 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS
:
3849 r
= kvm_x86_ops
.nested_ops
->enable_evmcs
!= NULL
;
3851 case KVM_CAP_SMALLER_MAXPHYADDR
:
3852 r
= (int) allow_smaller_maxphyaddr
;
3854 case KVM_CAP_STEAL_TIME
:
3855 r
= sched_info_on();
3864 long kvm_arch_dev_ioctl(struct file
*filp
,
3865 unsigned int ioctl
, unsigned long arg
)
3867 void __user
*argp
= (void __user
*)arg
;
3871 case KVM_GET_MSR_INDEX_LIST
: {
3872 struct kvm_msr_list __user
*user_msr_list
= argp
;
3873 struct kvm_msr_list msr_list
;
3877 if (copy_from_user(&msr_list
, user_msr_list
, sizeof(msr_list
)))
3880 msr_list
.nmsrs
= num_msrs_to_save
+ num_emulated_msrs
;
3881 if (copy_to_user(user_msr_list
, &msr_list
, sizeof(msr_list
)))
3884 if (n
< msr_list
.nmsrs
)
3887 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
3888 num_msrs_to_save
* sizeof(u32
)))
3890 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
3892 num_emulated_msrs
* sizeof(u32
)))
3897 case KVM_GET_SUPPORTED_CPUID
:
3898 case KVM_GET_EMULATED_CPUID
: {
3899 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3900 struct kvm_cpuid2 cpuid
;
3903 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof(cpuid
)))
3906 r
= kvm_dev_ioctl_get_cpuid(&cpuid
, cpuid_arg
->entries
,
3912 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof(cpuid
)))
3917 case KVM_X86_GET_MCE_CAP_SUPPORTED
:
3919 if (copy_to_user(argp
, &kvm_mce_cap_supported
,
3920 sizeof(kvm_mce_cap_supported
)))
3924 case KVM_GET_MSR_FEATURE_INDEX_LIST
: {
3925 struct kvm_msr_list __user
*user_msr_list
= argp
;
3926 struct kvm_msr_list msr_list
;
3930 if (copy_from_user(&msr_list
, user_msr_list
, sizeof(msr_list
)))
3933 msr_list
.nmsrs
= num_msr_based_features
;
3934 if (copy_to_user(user_msr_list
, &msr_list
, sizeof(msr_list
)))
3937 if (n
< msr_list
.nmsrs
)
3940 if (copy_to_user(user_msr_list
->indices
, &msr_based_features
,
3941 num_msr_based_features
* sizeof(u32
)))
3947 r
= msr_io(NULL
, argp
, do_get_msr_feature
, 1);
3949 case KVM_GET_SUPPORTED_HV_CPUID
:
3950 r
= kvm_ioctl_get_supported_hv_cpuid(NULL
, argp
);
3960 static void wbinvd_ipi(void *garbage
)
3965 static bool need_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
3967 return kvm_arch_has_noncoherent_dma(vcpu
->kvm
);
3970 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
3972 /* Address WBINVD may be executed by guest */
3973 if (need_emulate_wbinvd(vcpu
)) {
3974 if (kvm_x86_ops
.has_wbinvd_exit())
3975 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
3976 else if (vcpu
->cpu
!= -1 && vcpu
->cpu
!= cpu
)
3977 smp_call_function_single(vcpu
->cpu
,
3978 wbinvd_ipi
, NULL
, 1);
3981 kvm_x86_ops
.vcpu_load(vcpu
, cpu
);
3983 /* Save host pkru register if supported */
3984 vcpu
->arch
.host_pkru
= read_pkru();
3986 /* Apply any externally detected TSC adjustments (due to suspend) */
3987 if (unlikely(vcpu
->arch
.tsc_offset_adjustment
)) {
3988 adjust_tsc_offset_host(vcpu
, vcpu
->arch
.tsc_offset_adjustment
);
3989 vcpu
->arch
.tsc_offset_adjustment
= 0;
3990 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
3993 if (unlikely(vcpu
->cpu
!= cpu
) || kvm_check_tsc_unstable()) {
3994 s64 tsc_delta
= !vcpu
->arch
.last_host_tsc
? 0 :
3995 rdtsc() - vcpu
->arch
.last_host_tsc
;
3997 mark_tsc_unstable("KVM discovered backwards TSC");
3999 if (kvm_check_tsc_unstable()) {
4000 u64 offset
= kvm_compute_tsc_offset(vcpu
,
4001 vcpu
->arch
.last_guest_tsc
);
4002 kvm_vcpu_write_tsc_offset(vcpu
, offset
);
4003 vcpu
->arch
.tsc_catchup
= 1;
4006 if (kvm_lapic_hv_timer_in_use(vcpu
))
4007 kvm_lapic_restart_hv_timer(vcpu
);
4010 * On a host with synchronized TSC, there is no need to update
4011 * kvmclock on vcpu->cpu migration
4013 if (!vcpu
->kvm
->arch
.use_master_clock
|| vcpu
->cpu
== -1)
4014 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
);
4015 if (vcpu
->cpu
!= cpu
)
4016 kvm_make_request(KVM_REQ_MIGRATE_TIMER
, vcpu
);
4020 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
4023 static void kvm_steal_time_set_preempted(struct kvm_vcpu
*vcpu
)
4025 struct kvm_host_map map
;
4026 struct kvm_steal_time
*st
;
4028 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
4031 if (vcpu
->arch
.st
.preempted
)
4034 if (kvm_map_gfn(vcpu
, vcpu
->arch
.st
.msr_val
>> PAGE_SHIFT
, &map
,
4035 &vcpu
->arch
.st
.cache
, true))
4039 offset_in_page(vcpu
->arch
.st
.msr_val
& KVM_STEAL_VALID_BITS
);
4041 st
->preempted
= vcpu
->arch
.st
.preempted
= KVM_VCPU_PREEMPTED
;
4043 kvm_unmap_gfn(vcpu
, &map
, &vcpu
->arch
.st
.cache
, true, true);
4046 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
4050 if (vcpu
->preempted
&& !vcpu
->arch
.guest_state_protected
)
4051 vcpu
->arch
.preempted_in_kernel
= !kvm_x86_ops
.get_cpl(vcpu
);
4054 * Disable page faults because we're in atomic context here.
4055 * kvm_write_guest_offset_cached() would call might_fault()
4056 * that relies on pagefault_disable() to tell if there's a
4057 * bug. NOTE: the write to guest memory may not go through if
4058 * during postcopy live migration or if there's heavy guest
4061 pagefault_disable();
4063 * kvm_memslots() will be called by
4064 * kvm_write_guest_offset_cached() so take the srcu lock.
4066 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
4067 kvm_steal_time_set_preempted(vcpu
);
4068 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
4070 kvm_x86_ops
.vcpu_put(vcpu
);
4071 vcpu
->arch
.last_host_tsc
= rdtsc();
4073 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4074 * on every vmexit, but if not, we might have a stale dr6 from the
4075 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4080 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
4081 struct kvm_lapic_state
*s
)
4083 if (vcpu
->arch
.apicv_active
)
4084 kvm_x86_ops
.sync_pir_to_irr(vcpu
);
4086 return kvm_apic_get_state(vcpu
, s
);
4089 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
4090 struct kvm_lapic_state
*s
)
4094 r
= kvm_apic_set_state(vcpu
, s
);
4097 update_cr8_intercept(vcpu
);
4102 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu
*vcpu
)
4105 * We can accept userspace's request for interrupt injection
4106 * as long as we have a place to store the interrupt number.
4107 * The actual injection will happen when the CPU is able to
4108 * deliver the interrupt.
4110 if (kvm_cpu_has_extint(vcpu
))
4113 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4114 return (!lapic_in_kernel(vcpu
) ||
4115 kvm_apic_accept_pic_intr(vcpu
));
4118 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu
*vcpu
)
4120 return kvm_arch_interrupt_allowed(vcpu
) &&
4121 kvm_cpu_accept_dm_intr(vcpu
);
4124 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
4125 struct kvm_interrupt
*irq
)
4127 if (irq
->irq
>= KVM_NR_INTERRUPTS
)
4130 if (!irqchip_in_kernel(vcpu
->kvm
)) {
4131 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
4132 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
4137 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4138 * fail for in-kernel 8259.
4140 if (pic_in_kernel(vcpu
->kvm
))
4143 if (vcpu
->arch
.pending_external_vector
!= -1)
4146 vcpu
->arch
.pending_external_vector
= irq
->irq
;
4147 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
4151 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
4153 kvm_inject_nmi(vcpu
);
4158 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu
*vcpu
)
4160 kvm_make_request(KVM_REQ_SMI
, vcpu
);
4165 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
4166 struct kvm_tpr_access_ctl
*tac
)
4170 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
4174 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
4178 unsigned bank_num
= mcg_cap
& 0xff, bank
;
4181 if (!bank_num
|| bank_num
> KVM_MAX_MCE_BANKS
)
4183 if (mcg_cap
& ~(kvm_mce_cap_supported
| 0xff | 0xff0000))
4186 vcpu
->arch
.mcg_cap
= mcg_cap
;
4187 /* Init IA32_MCG_CTL to all 1s */
4188 if (mcg_cap
& MCG_CTL_P
)
4189 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
4190 /* Init IA32_MCi_CTL to all 1s */
4191 for (bank
= 0; bank
< bank_num
; bank
++)
4192 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
4194 kvm_x86_ops
.setup_mce(vcpu
);
4199 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
4200 struct kvm_x86_mce
*mce
)
4202 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
4203 unsigned bank_num
= mcg_cap
& 0xff;
4204 u64
*banks
= vcpu
->arch
.mce_banks
;
4206 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
4209 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4210 * reporting is disabled
4212 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
4213 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
4215 banks
+= 4 * mce
->bank
;
4217 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4218 * reporting is disabled for the bank
4220 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
4222 if (mce
->status
& MCI_STATUS_UC
) {
4223 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
4224 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
4225 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
4228 if (banks
[1] & MCI_STATUS_VAL
)
4229 mce
->status
|= MCI_STATUS_OVER
;
4230 banks
[2] = mce
->addr
;
4231 banks
[3] = mce
->misc
;
4232 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
4233 banks
[1] = mce
->status
;
4234 kvm_queue_exception(vcpu
, MC_VECTOR
);
4235 } else if (!(banks
[1] & MCI_STATUS_VAL
)
4236 || !(banks
[1] & MCI_STATUS_UC
)) {
4237 if (banks
[1] & MCI_STATUS_VAL
)
4238 mce
->status
|= MCI_STATUS_OVER
;
4239 banks
[2] = mce
->addr
;
4240 banks
[3] = mce
->misc
;
4241 banks
[1] = mce
->status
;
4243 banks
[1] |= MCI_STATUS_OVER
;
4247 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
4248 struct kvm_vcpu_events
*events
)
4252 if (kvm_check_request(KVM_REQ_SMI
, vcpu
))
4256 * In guest mode, payload delivery should be deferred,
4257 * so that the L1 hypervisor can intercept #PF before
4258 * CR2 is modified (or intercept #DB before DR6 is
4259 * modified under nVMX). Unless the per-VM capability,
4260 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4261 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4262 * opportunistically defer the exception payload, deliver it if the
4263 * capability hasn't been requested before processing a
4264 * KVM_GET_VCPU_EVENTS.
4266 if (!vcpu
->kvm
->arch
.exception_payload_enabled
&&
4267 vcpu
->arch
.exception
.pending
&& vcpu
->arch
.exception
.has_payload
)
4268 kvm_deliver_exception_payload(vcpu
);
4271 * The API doesn't provide the instruction length for software
4272 * exceptions, so don't report them. As long as the guest RIP
4273 * isn't advanced, we should expect to encounter the exception
4276 if (kvm_exception_is_soft(vcpu
->arch
.exception
.nr
)) {
4277 events
->exception
.injected
= 0;
4278 events
->exception
.pending
= 0;
4280 events
->exception
.injected
= vcpu
->arch
.exception
.injected
;
4281 events
->exception
.pending
= vcpu
->arch
.exception
.pending
;
4283 * For ABI compatibility, deliberately conflate
4284 * pending and injected exceptions when
4285 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4287 if (!vcpu
->kvm
->arch
.exception_payload_enabled
)
4288 events
->exception
.injected
|=
4289 vcpu
->arch
.exception
.pending
;
4291 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
4292 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
4293 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
4294 events
->exception_has_payload
= vcpu
->arch
.exception
.has_payload
;
4295 events
->exception_payload
= vcpu
->arch
.exception
.payload
;
4297 events
->interrupt
.injected
=
4298 vcpu
->arch
.interrupt
.injected
&& !vcpu
->arch
.interrupt
.soft
;
4299 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
4300 events
->interrupt
.soft
= 0;
4301 events
->interrupt
.shadow
= kvm_x86_ops
.get_interrupt_shadow(vcpu
);
4303 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
4304 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
!= 0;
4305 events
->nmi
.masked
= kvm_x86_ops
.get_nmi_mask(vcpu
);
4306 events
->nmi
.pad
= 0;
4308 events
->sipi_vector
= 0; /* never valid when reporting to user space */
4310 events
->smi
.smm
= is_smm(vcpu
);
4311 events
->smi
.pending
= vcpu
->arch
.smi_pending
;
4312 events
->smi
.smm_inside_nmi
=
4313 !!(vcpu
->arch
.hflags
& HF_SMM_INSIDE_NMI_MASK
);
4314 events
->smi
.latched_init
= kvm_lapic_latched_init(vcpu
);
4316 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
4317 | KVM_VCPUEVENT_VALID_SHADOW
4318 | KVM_VCPUEVENT_VALID_SMM
);
4319 if (vcpu
->kvm
->arch
.exception_payload_enabled
)
4320 events
->flags
|= KVM_VCPUEVENT_VALID_PAYLOAD
;
4322 memset(&events
->reserved
, 0, sizeof(events
->reserved
));
4325 static void kvm_smm_changed(struct kvm_vcpu
*vcpu
);
4327 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
4328 struct kvm_vcpu_events
*events
)
4330 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4331 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4332 | KVM_VCPUEVENT_VALID_SHADOW
4333 | KVM_VCPUEVENT_VALID_SMM
4334 | KVM_VCPUEVENT_VALID_PAYLOAD
))
4337 if (events
->flags
& KVM_VCPUEVENT_VALID_PAYLOAD
) {
4338 if (!vcpu
->kvm
->arch
.exception_payload_enabled
)
4340 if (events
->exception
.pending
)
4341 events
->exception
.injected
= 0;
4343 events
->exception_has_payload
= 0;
4345 events
->exception
.pending
= 0;
4346 events
->exception_has_payload
= 0;
4349 if ((events
->exception
.injected
|| events
->exception
.pending
) &&
4350 (events
->exception
.nr
> 31 || events
->exception
.nr
== NMI_VECTOR
))
4353 /* INITs are latched while in SMM */
4354 if (events
->flags
& KVM_VCPUEVENT_VALID_SMM
&&
4355 (events
->smi
.smm
|| events
->smi
.pending
) &&
4356 vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
)
4360 vcpu
->arch
.exception
.injected
= events
->exception
.injected
;
4361 vcpu
->arch
.exception
.pending
= events
->exception
.pending
;
4362 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
4363 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
4364 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
4365 vcpu
->arch
.exception
.has_payload
= events
->exception_has_payload
;
4366 vcpu
->arch
.exception
.payload
= events
->exception_payload
;
4368 vcpu
->arch
.interrupt
.injected
= events
->interrupt
.injected
;
4369 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
4370 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
4371 if (events
->flags
& KVM_VCPUEVENT_VALID_SHADOW
)
4372 kvm_x86_ops
.set_interrupt_shadow(vcpu
,
4373 events
->interrupt
.shadow
);
4375 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
4376 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
4377 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
4378 kvm_x86_ops
.set_nmi_mask(vcpu
, events
->nmi
.masked
);
4380 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
&&
4381 lapic_in_kernel(vcpu
))
4382 vcpu
->arch
.apic
->sipi_vector
= events
->sipi_vector
;
4384 if (events
->flags
& KVM_VCPUEVENT_VALID_SMM
) {
4385 if (!!(vcpu
->arch
.hflags
& HF_SMM_MASK
) != events
->smi
.smm
) {
4386 if (events
->smi
.smm
)
4387 vcpu
->arch
.hflags
|= HF_SMM_MASK
;
4389 vcpu
->arch
.hflags
&= ~HF_SMM_MASK
;
4390 kvm_smm_changed(vcpu
);
4393 vcpu
->arch
.smi_pending
= events
->smi
.pending
;
4395 if (events
->smi
.smm
) {
4396 if (events
->smi
.smm_inside_nmi
)
4397 vcpu
->arch
.hflags
|= HF_SMM_INSIDE_NMI_MASK
;
4399 vcpu
->arch
.hflags
&= ~HF_SMM_INSIDE_NMI_MASK
;
4402 if (lapic_in_kernel(vcpu
)) {
4403 if (events
->smi
.latched_init
)
4404 set_bit(KVM_APIC_INIT
, &vcpu
->arch
.apic
->pending_events
);
4406 clear_bit(KVM_APIC_INIT
, &vcpu
->arch
.apic
->pending_events
);
4410 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
4415 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu
*vcpu
,
4416 struct kvm_debugregs
*dbgregs
)
4420 memcpy(dbgregs
->db
, vcpu
->arch
.db
, sizeof(vcpu
->arch
.db
));
4421 kvm_get_dr(vcpu
, 6, &val
);
4423 dbgregs
->dr7
= vcpu
->arch
.dr7
;
4425 memset(&dbgregs
->reserved
, 0, sizeof(dbgregs
->reserved
));
4428 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu
*vcpu
,
4429 struct kvm_debugregs
*dbgregs
)
4434 if (dbgregs
->dr6
& ~0xffffffffull
)
4436 if (dbgregs
->dr7
& ~0xffffffffull
)
4439 memcpy(vcpu
->arch
.db
, dbgregs
->db
, sizeof(vcpu
->arch
.db
));
4440 kvm_update_dr0123(vcpu
);
4441 vcpu
->arch
.dr6
= dbgregs
->dr6
;
4442 vcpu
->arch
.dr7
= dbgregs
->dr7
;
4443 kvm_update_dr7(vcpu
);
4448 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4450 static void fill_xsave(u8
*dest
, struct kvm_vcpu
*vcpu
)
4452 struct xregs_state
*xsave
= &vcpu
->arch
.guest_fpu
->state
.xsave
;
4453 u64 xstate_bv
= xsave
->header
.xfeatures
;
4457 * Copy legacy XSAVE area, to avoid complications with CPUID
4458 * leaves 0 and 1 in the loop below.
4460 memcpy(dest
, xsave
, XSAVE_HDR_OFFSET
);
4463 xstate_bv
&= vcpu
->arch
.guest_supported_xcr0
| XFEATURE_MASK_FPSSE
;
4464 *(u64
*)(dest
+ XSAVE_HDR_OFFSET
) = xstate_bv
;
4467 * Copy each region from the possibly compacted offset to the
4468 * non-compacted offset.
4470 valid
= xstate_bv
& ~XFEATURE_MASK_FPSSE
;
4472 u64 xfeature_mask
= valid
& -valid
;
4473 int xfeature_nr
= fls64(xfeature_mask
) - 1;
4474 void *src
= get_xsave_addr(xsave
, xfeature_nr
);
4477 u32 size
, offset
, ecx
, edx
;
4478 cpuid_count(XSTATE_CPUID
, xfeature_nr
,
4479 &size
, &offset
, &ecx
, &edx
);
4480 if (xfeature_nr
== XFEATURE_PKRU
)
4481 memcpy(dest
+ offset
, &vcpu
->arch
.pkru
,
4482 sizeof(vcpu
->arch
.pkru
));
4484 memcpy(dest
+ offset
, src
, size
);
4488 valid
-= xfeature_mask
;
4492 static void load_xsave(struct kvm_vcpu
*vcpu
, u8
*src
)
4494 struct xregs_state
*xsave
= &vcpu
->arch
.guest_fpu
->state
.xsave
;
4495 u64 xstate_bv
= *(u64
*)(src
+ XSAVE_HDR_OFFSET
);
4499 * Copy legacy XSAVE area, to avoid complications with CPUID
4500 * leaves 0 and 1 in the loop below.
4502 memcpy(xsave
, src
, XSAVE_HDR_OFFSET
);
4504 /* Set XSTATE_BV and possibly XCOMP_BV. */
4505 xsave
->header
.xfeatures
= xstate_bv
;
4506 if (boot_cpu_has(X86_FEATURE_XSAVES
))
4507 xsave
->header
.xcomp_bv
= host_xcr0
| XSTATE_COMPACTION_ENABLED
;
4510 * Copy each region from the non-compacted offset to the
4511 * possibly compacted offset.
4513 valid
= xstate_bv
& ~XFEATURE_MASK_FPSSE
;
4515 u64 xfeature_mask
= valid
& -valid
;
4516 int xfeature_nr
= fls64(xfeature_mask
) - 1;
4517 void *dest
= get_xsave_addr(xsave
, xfeature_nr
);
4520 u32 size
, offset
, ecx
, edx
;
4521 cpuid_count(XSTATE_CPUID
, xfeature_nr
,
4522 &size
, &offset
, &ecx
, &edx
);
4523 if (xfeature_nr
== XFEATURE_PKRU
)
4524 memcpy(&vcpu
->arch
.pkru
, src
+ offset
,
4525 sizeof(vcpu
->arch
.pkru
));
4527 memcpy(dest
, src
+ offset
, size
);
4530 valid
-= xfeature_mask
;
4534 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu
*vcpu
,
4535 struct kvm_xsave
*guest_xsave
)
4537 if (!vcpu
->arch
.guest_fpu
)
4540 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
4541 memset(guest_xsave
, 0, sizeof(struct kvm_xsave
));
4542 fill_xsave((u8
*) guest_xsave
->region
, vcpu
);
4544 memcpy(guest_xsave
->region
,
4545 &vcpu
->arch
.guest_fpu
->state
.fxsave
,
4546 sizeof(struct fxregs_state
));
4547 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)] =
4548 XFEATURE_MASK_FPSSE
;
4552 #define XSAVE_MXCSR_OFFSET 24
4554 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu
*vcpu
,
4555 struct kvm_xsave
*guest_xsave
)
4560 if (!vcpu
->arch
.guest_fpu
)
4563 xstate_bv
= *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)];
4564 mxcsr
= *(u32
*)&guest_xsave
->region
[XSAVE_MXCSR_OFFSET
/ sizeof(u32
)];
4566 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
4568 * Here we allow setting states that are not present in
4569 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4570 * with old userspace.
4572 if (xstate_bv
& ~supported_xcr0
|| mxcsr
& ~mxcsr_feature_mask
)
4574 load_xsave(vcpu
, (u8
*)guest_xsave
->region
);
4576 if (xstate_bv
& ~XFEATURE_MASK_FPSSE
||
4577 mxcsr
& ~mxcsr_feature_mask
)
4579 memcpy(&vcpu
->arch
.guest_fpu
->state
.fxsave
,
4580 guest_xsave
->region
, sizeof(struct fxregs_state
));
4585 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu
*vcpu
,
4586 struct kvm_xcrs
*guest_xcrs
)
4588 if (!boot_cpu_has(X86_FEATURE_XSAVE
)) {
4589 guest_xcrs
->nr_xcrs
= 0;
4593 guest_xcrs
->nr_xcrs
= 1;
4594 guest_xcrs
->flags
= 0;
4595 guest_xcrs
->xcrs
[0].xcr
= XCR_XFEATURE_ENABLED_MASK
;
4596 guest_xcrs
->xcrs
[0].value
= vcpu
->arch
.xcr0
;
4599 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu
*vcpu
,
4600 struct kvm_xcrs
*guest_xcrs
)
4604 if (!boot_cpu_has(X86_FEATURE_XSAVE
))
4607 if (guest_xcrs
->nr_xcrs
> KVM_MAX_XCRS
|| guest_xcrs
->flags
)
4610 for (i
= 0; i
< guest_xcrs
->nr_xcrs
; i
++)
4611 /* Only support XCR0 currently */
4612 if (guest_xcrs
->xcrs
[i
].xcr
== XCR_XFEATURE_ENABLED_MASK
) {
4613 r
= __kvm_set_xcr(vcpu
, XCR_XFEATURE_ENABLED_MASK
,
4614 guest_xcrs
->xcrs
[i
].value
);
4623 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4624 * stopped by the hypervisor. This function will be called from the host only.
4625 * EINVAL is returned when the host attempts to set the flag for a guest that
4626 * does not support pv clocks.
4628 static int kvm_set_guest_paused(struct kvm_vcpu
*vcpu
)
4630 if (!vcpu
->arch
.pv_time_enabled
)
4632 vcpu
->arch
.pvclock_set_guest_stopped_request
= true;
4633 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
4637 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu
*vcpu
,
4638 struct kvm_enable_cap
*cap
)
4641 uint16_t vmcs_version
;
4642 void __user
*user_ptr
;
4648 case KVM_CAP_HYPERV_SYNIC2
:
4653 case KVM_CAP_HYPERV_SYNIC
:
4654 if (!irqchip_in_kernel(vcpu
->kvm
))
4656 return kvm_hv_activate_synic(vcpu
, cap
->cap
==
4657 KVM_CAP_HYPERV_SYNIC2
);
4658 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS
:
4659 if (!kvm_x86_ops
.nested_ops
->enable_evmcs
)
4661 r
= kvm_x86_ops
.nested_ops
->enable_evmcs(vcpu
, &vmcs_version
);
4663 user_ptr
= (void __user
*)(uintptr_t)cap
->args
[0];
4664 if (copy_to_user(user_ptr
, &vmcs_version
,
4665 sizeof(vmcs_version
)))
4669 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH
:
4670 if (!kvm_x86_ops
.enable_direct_tlbflush
)
4673 return kvm_x86_ops
.enable_direct_tlbflush(vcpu
);
4675 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID
:
4676 vcpu
->arch
.pv_cpuid
.enforce
= cap
->args
[0];
4677 if (vcpu
->arch
.pv_cpuid
.enforce
)
4678 kvm_update_pv_runtime(vcpu
);
4687 long kvm_arch_vcpu_ioctl(struct file
*filp
,
4688 unsigned int ioctl
, unsigned long arg
)
4690 struct kvm_vcpu
*vcpu
= filp
->private_data
;
4691 void __user
*argp
= (void __user
*)arg
;
4694 struct kvm_lapic_state
*lapic
;
4695 struct kvm_xsave
*xsave
;
4696 struct kvm_xcrs
*xcrs
;
4704 case KVM_GET_LAPIC
: {
4706 if (!lapic_in_kernel(vcpu
))
4708 u
.lapic
= kzalloc(sizeof(struct kvm_lapic_state
),
4709 GFP_KERNEL_ACCOUNT
);
4714 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, u
.lapic
);
4718 if (copy_to_user(argp
, u
.lapic
, sizeof(struct kvm_lapic_state
)))
4723 case KVM_SET_LAPIC
: {
4725 if (!lapic_in_kernel(vcpu
))
4727 u
.lapic
= memdup_user(argp
, sizeof(*u
.lapic
));
4728 if (IS_ERR(u
.lapic
)) {
4729 r
= PTR_ERR(u
.lapic
);
4733 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, u
.lapic
);
4736 case KVM_INTERRUPT
: {
4737 struct kvm_interrupt irq
;
4740 if (copy_from_user(&irq
, argp
, sizeof(irq
)))
4742 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
4746 r
= kvm_vcpu_ioctl_nmi(vcpu
);
4750 r
= kvm_vcpu_ioctl_smi(vcpu
);
4753 case KVM_SET_CPUID
: {
4754 struct kvm_cpuid __user
*cpuid_arg
= argp
;
4755 struct kvm_cpuid cpuid
;
4758 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof(cpuid
)))
4760 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
4763 case KVM_SET_CPUID2
: {
4764 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
4765 struct kvm_cpuid2 cpuid
;
4768 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof(cpuid
)))
4770 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
4771 cpuid_arg
->entries
);
4774 case KVM_GET_CPUID2
: {
4775 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
4776 struct kvm_cpuid2 cpuid
;
4779 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof(cpuid
)))
4781 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
4782 cpuid_arg
->entries
);
4786 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof(cpuid
)))
4791 case KVM_GET_MSRS
: {
4792 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
4793 r
= msr_io(vcpu
, argp
, do_get_msr
, 1);
4794 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
4797 case KVM_SET_MSRS
: {
4798 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
4799 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
4800 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
4803 case KVM_TPR_ACCESS_REPORTING
: {
4804 struct kvm_tpr_access_ctl tac
;
4807 if (copy_from_user(&tac
, argp
, sizeof(tac
)))
4809 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
4813 if (copy_to_user(argp
, &tac
, sizeof(tac
)))
4818 case KVM_SET_VAPIC_ADDR
: {
4819 struct kvm_vapic_addr va
;
4823 if (!lapic_in_kernel(vcpu
))
4826 if (copy_from_user(&va
, argp
, sizeof(va
)))
4828 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
4829 r
= kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
4830 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
4833 case KVM_X86_SETUP_MCE
: {
4837 if (copy_from_user(&mcg_cap
, argp
, sizeof(mcg_cap
)))
4839 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
4842 case KVM_X86_SET_MCE
: {
4843 struct kvm_x86_mce mce
;
4846 if (copy_from_user(&mce
, argp
, sizeof(mce
)))
4848 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
4851 case KVM_GET_VCPU_EVENTS
: {
4852 struct kvm_vcpu_events events
;
4854 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
4857 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
4862 case KVM_SET_VCPU_EVENTS
: {
4863 struct kvm_vcpu_events events
;
4866 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
4869 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
4872 case KVM_GET_DEBUGREGS
: {
4873 struct kvm_debugregs dbgregs
;
4875 kvm_vcpu_ioctl_x86_get_debugregs(vcpu
, &dbgregs
);
4878 if (copy_to_user(argp
, &dbgregs
,
4879 sizeof(struct kvm_debugregs
)))
4884 case KVM_SET_DEBUGREGS
: {
4885 struct kvm_debugregs dbgregs
;
4888 if (copy_from_user(&dbgregs
, argp
,
4889 sizeof(struct kvm_debugregs
)))
4892 r
= kvm_vcpu_ioctl_x86_set_debugregs(vcpu
, &dbgregs
);
4895 case KVM_GET_XSAVE
: {
4896 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL_ACCOUNT
);
4901 kvm_vcpu_ioctl_x86_get_xsave(vcpu
, u
.xsave
);
4904 if (copy_to_user(argp
, u
.xsave
, sizeof(struct kvm_xsave
)))
4909 case KVM_SET_XSAVE
: {
4910 u
.xsave
= memdup_user(argp
, sizeof(*u
.xsave
));
4911 if (IS_ERR(u
.xsave
)) {
4912 r
= PTR_ERR(u
.xsave
);
4916 r
= kvm_vcpu_ioctl_x86_set_xsave(vcpu
, u
.xsave
);
4919 case KVM_GET_XCRS
: {
4920 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL_ACCOUNT
);
4925 kvm_vcpu_ioctl_x86_get_xcrs(vcpu
, u
.xcrs
);
4928 if (copy_to_user(argp
, u
.xcrs
,
4929 sizeof(struct kvm_xcrs
)))
4934 case KVM_SET_XCRS
: {
4935 u
.xcrs
= memdup_user(argp
, sizeof(*u
.xcrs
));
4936 if (IS_ERR(u
.xcrs
)) {
4937 r
= PTR_ERR(u
.xcrs
);
4941 r
= kvm_vcpu_ioctl_x86_set_xcrs(vcpu
, u
.xcrs
);
4944 case KVM_SET_TSC_KHZ
: {
4948 user_tsc_khz
= (u32
)arg
;
4950 if (kvm_has_tsc_control
&&
4951 user_tsc_khz
>= kvm_max_guest_tsc_khz
)
4954 if (user_tsc_khz
== 0)
4955 user_tsc_khz
= tsc_khz
;
4957 if (!kvm_set_tsc_khz(vcpu
, user_tsc_khz
))
4962 case KVM_GET_TSC_KHZ
: {
4963 r
= vcpu
->arch
.virtual_tsc_khz
;
4966 case KVM_KVMCLOCK_CTRL
: {
4967 r
= kvm_set_guest_paused(vcpu
);
4970 case KVM_ENABLE_CAP
: {
4971 struct kvm_enable_cap cap
;
4974 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
4976 r
= kvm_vcpu_ioctl_enable_cap(vcpu
, &cap
);
4979 case KVM_GET_NESTED_STATE
: {
4980 struct kvm_nested_state __user
*user_kvm_nested_state
= argp
;
4984 if (!kvm_x86_ops
.nested_ops
->get_state
)
4987 BUILD_BUG_ON(sizeof(user_data_size
) != sizeof(user_kvm_nested_state
->size
));
4989 if (get_user(user_data_size
, &user_kvm_nested_state
->size
))
4992 r
= kvm_x86_ops
.nested_ops
->get_state(vcpu
, user_kvm_nested_state
,
4997 if (r
> user_data_size
) {
4998 if (put_user(r
, &user_kvm_nested_state
->size
))
5008 case KVM_SET_NESTED_STATE
: {
5009 struct kvm_nested_state __user
*user_kvm_nested_state
= argp
;
5010 struct kvm_nested_state kvm_state
;
5014 if (!kvm_x86_ops
.nested_ops
->set_state
)
5018 if (copy_from_user(&kvm_state
, user_kvm_nested_state
, sizeof(kvm_state
)))
5022 if (kvm_state
.size
< sizeof(kvm_state
))
5025 if (kvm_state
.flags
&
5026 ~(KVM_STATE_NESTED_RUN_PENDING
| KVM_STATE_NESTED_GUEST_MODE
5027 | KVM_STATE_NESTED_EVMCS
| KVM_STATE_NESTED_MTF_PENDING
5028 | KVM_STATE_NESTED_GIF_SET
))
5031 /* nested_run_pending implies guest_mode. */
5032 if ((kvm_state
.flags
& KVM_STATE_NESTED_RUN_PENDING
)
5033 && !(kvm_state
.flags
& KVM_STATE_NESTED_GUEST_MODE
))
5036 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
5037 r
= kvm_x86_ops
.nested_ops
->set_state(vcpu
, user_kvm_nested_state
, &kvm_state
);
5038 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
5041 case KVM_GET_SUPPORTED_HV_CPUID
:
5042 r
= kvm_ioctl_get_supported_hv_cpuid(vcpu
, argp
);
5054 vm_fault_t
kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
5056 return VM_FAULT_SIGBUS
;
5059 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
5063 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
5065 ret
= kvm_x86_ops
.set_tss_addr(kvm
, addr
);
5069 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
5072 return kvm_x86_ops
.set_identity_map_addr(kvm
, ident_addr
);
5075 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
5076 unsigned long kvm_nr_mmu_pages
)
5078 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
5081 mutex_lock(&kvm
->slots_lock
);
5083 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
5084 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
5086 mutex_unlock(&kvm
->slots_lock
);
5090 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
5092 return kvm
->arch
.n_max_mmu_pages
;
5095 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
5097 struct kvm_pic
*pic
= kvm
->arch
.vpic
;
5101 switch (chip
->chip_id
) {
5102 case KVM_IRQCHIP_PIC_MASTER
:
5103 memcpy(&chip
->chip
.pic
, &pic
->pics
[0],
5104 sizeof(struct kvm_pic_state
));
5106 case KVM_IRQCHIP_PIC_SLAVE
:
5107 memcpy(&chip
->chip
.pic
, &pic
->pics
[1],
5108 sizeof(struct kvm_pic_state
));
5110 case KVM_IRQCHIP_IOAPIC
:
5111 kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
5120 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
5122 struct kvm_pic
*pic
= kvm
->arch
.vpic
;
5126 switch (chip
->chip_id
) {
5127 case KVM_IRQCHIP_PIC_MASTER
:
5128 spin_lock(&pic
->lock
);
5129 memcpy(&pic
->pics
[0], &chip
->chip
.pic
,
5130 sizeof(struct kvm_pic_state
));
5131 spin_unlock(&pic
->lock
);
5133 case KVM_IRQCHIP_PIC_SLAVE
:
5134 spin_lock(&pic
->lock
);
5135 memcpy(&pic
->pics
[1], &chip
->chip
.pic
,
5136 sizeof(struct kvm_pic_state
));
5137 spin_unlock(&pic
->lock
);
5139 case KVM_IRQCHIP_IOAPIC
:
5140 kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
5146 kvm_pic_update_irq(pic
);
5150 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
5152 struct kvm_kpit_state
*kps
= &kvm
->arch
.vpit
->pit_state
;
5154 BUILD_BUG_ON(sizeof(*ps
) != sizeof(kps
->channels
));
5156 mutex_lock(&kps
->lock
);
5157 memcpy(ps
, &kps
->channels
, sizeof(*ps
));
5158 mutex_unlock(&kps
->lock
);
5162 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
5165 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
5167 mutex_lock(&pit
->pit_state
.lock
);
5168 memcpy(&pit
->pit_state
.channels
, ps
, sizeof(*ps
));
5169 for (i
= 0; i
< 3; i
++)
5170 kvm_pit_load_count(pit
, i
, ps
->channels
[i
].count
, 0);
5171 mutex_unlock(&pit
->pit_state
.lock
);
5175 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
5177 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
5178 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
5179 sizeof(ps
->channels
));
5180 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
5181 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
5182 memset(&ps
->reserved
, 0, sizeof(ps
->reserved
));
5186 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
5190 u32 prev_legacy
, cur_legacy
;
5191 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
5193 mutex_lock(&pit
->pit_state
.lock
);
5194 prev_legacy
= pit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
5195 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
5196 if (!prev_legacy
&& cur_legacy
)
5198 memcpy(&pit
->pit_state
.channels
, &ps
->channels
,
5199 sizeof(pit
->pit_state
.channels
));
5200 pit
->pit_state
.flags
= ps
->flags
;
5201 for (i
= 0; i
< 3; i
++)
5202 kvm_pit_load_count(pit
, i
, pit
->pit_state
.channels
[i
].count
,
5204 mutex_unlock(&pit
->pit_state
.lock
);
5208 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
5209 struct kvm_reinject_control
*control
)
5211 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
5213 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5214 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5215 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5217 mutex_lock(&pit
->pit_state
.lock
);
5218 kvm_pit_set_reinject(pit
, control
->pit_reinject
);
5219 mutex_unlock(&pit
->pit_state
.lock
);
5224 void kvm_arch_sync_dirty_log(struct kvm
*kvm
, struct kvm_memory_slot
*memslot
)
5227 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
5229 if (kvm_x86_ops
.flush_log_dirty
)
5230 kvm_x86_ops
.flush_log_dirty(kvm
);
5233 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
,
5236 if (!irqchip_in_kernel(kvm
))
5239 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
5240 irq_event
->irq
, irq_event
->level
,
5245 int kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
5246 struct kvm_enable_cap
*cap
)
5254 case KVM_CAP_DISABLE_QUIRKS
:
5255 kvm
->arch
.disabled_quirks
= cap
->args
[0];
5258 case KVM_CAP_SPLIT_IRQCHIP
: {
5259 mutex_lock(&kvm
->lock
);
5261 if (cap
->args
[0] > MAX_NR_RESERVED_IOAPIC_PINS
)
5262 goto split_irqchip_unlock
;
5264 if (irqchip_in_kernel(kvm
))
5265 goto split_irqchip_unlock
;
5266 if (kvm
->created_vcpus
)
5267 goto split_irqchip_unlock
;
5268 r
= kvm_setup_empty_irq_routing(kvm
);
5270 goto split_irqchip_unlock
;
5271 /* Pairs with irqchip_in_kernel. */
5273 kvm
->arch
.irqchip_mode
= KVM_IRQCHIP_SPLIT
;
5274 kvm
->arch
.nr_reserved_ioapic_pins
= cap
->args
[0];
5276 split_irqchip_unlock
:
5277 mutex_unlock(&kvm
->lock
);
5280 case KVM_CAP_X2APIC_API
:
5282 if (cap
->args
[0] & ~KVM_X2APIC_API_VALID_FLAGS
)
5285 if (cap
->args
[0] & KVM_X2APIC_API_USE_32BIT_IDS
)
5286 kvm
->arch
.x2apic_format
= true;
5287 if (cap
->args
[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK
)
5288 kvm
->arch
.x2apic_broadcast_quirk_disabled
= true;
5292 case KVM_CAP_X86_DISABLE_EXITS
:
5294 if (cap
->args
[0] & ~KVM_X86_DISABLE_VALID_EXITS
)
5297 if ((cap
->args
[0] & KVM_X86_DISABLE_EXITS_MWAIT
) &&
5298 kvm_can_mwait_in_guest())
5299 kvm
->arch
.mwait_in_guest
= true;
5300 if (cap
->args
[0] & KVM_X86_DISABLE_EXITS_HLT
)
5301 kvm
->arch
.hlt_in_guest
= true;
5302 if (cap
->args
[0] & KVM_X86_DISABLE_EXITS_PAUSE
)
5303 kvm
->arch
.pause_in_guest
= true;
5304 if (cap
->args
[0] & KVM_X86_DISABLE_EXITS_CSTATE
)
5305 kvm
->arch
.cstate_in_guest
= true;
5308 case KVM_CAP_MSR_PLATFORM_INFO
:
5309 kvm
->arch
.guest_can_read_msr_platform_info
= cap
->args
[0];
5312 case KVM_CAP_EXCEPTION_PAYLOAD
:
5313 kvm
->arch
.exception_payload_enabled
= cap
->args
[0];
5316 case KVM_CAP_X86_USER_SPACE_MSR
:
5317 kvm
->arch
.user_space_msr_mask
= cap
->args
[0];
5327 static struct kvm_x86_msr_filter
*kvm_alloc_msr_filter(bool default_allow
)
5329 struct kvm_x86_msr_filter
*msr_filter
;
5331 msr_filter
= kzalloc(sizeof(*msr_filter
), GFP_KERNEL_ACCOUNT
);
5335 msr_filter
->default_allow
= default_allow
;
5339 static void kvm_free_msr_filter(struct kvm_x86_msr_filter
*msr_filter
)
5346 for (i
= 0; i
< msr_filter
->count
; i
++)
5347 kfree(msr_filter
->ranges
[i
].bitmap
);
5352 static int kvm_add_msr_filter(struct kvm_x86_msr_filter
*msr_filter
,
5353 struct kvm_msr_filter_range
*user_range
)
5355 struct msr_bitmap_range range
;
5356 unsigned long *bitmap
= NULL
;
5360 if (!user_range
->nmsrs
)
5363 bitmap_size
= BITS_TO_LONGS(user_range
->nmsrs
) * sizeof(long);
5364 if (!bitmap_size
|| bitmap_size
> KVM_MSR_FILTER_MAX_BITMAP_SIZE
)
5367 bitmap
= memdup_user((__user u8
*)user_range
->bitmap
, bitmap_size
);
5369 return PTR_ERR(bitmap
);
5371 range
= (struct msr_bitmap_range
) {
5372 .flags
= user_range
->flags
,
5373 .base
= user_range
->base
,
5374 .nmsrs
= user_range
->nmsrs
,
5378 if (range
.flags
& ~(KVM_MSR_FILTER_READ
| KVM_MSR_FILTER_WRITE
)) {
5388 /* Everything ok, add this range identifier. */
5389 msr_filter
->ranges
[msr_filter
->count
] = range
;
5390 msr_filter
->count
++;
5398 static int kvm_vm_ioctl_set_msr_filter(struct kvm
*kvm
, void __user
*argp
)
5400 struct kvm_msr_filter __user
*user_msr_filter
= argp
;
5401 struct kvm_x86_msr_filter
*new_filter
, *old_filter
;
5402 struct kvm_msr_filter filter
;
5408 if (copy_from_user(&filter
, user_msr_filter
, sizeof(filter
)))
5411 for (i
= 0; i
< ARRAY_SIZE(filter
.ranges
); i
++)
5412 empty
&= !filter
.ranges
[i
].nmsrs
;
5414 default_allow
= !(filter
.flags
& KVM_MSR_FILTER_DEFAULT_DENY
);
5415 if (empty
&& !default_allow
)
5418 new_filter
= kvm_alloc_msr_filter(default_allow
);
5422 for (i
= 0; i
< ARRAY_SIZE(filter
.ranges
); i
++) {
5423 r
= kvm_add_msr_filter(new_filter
, &filter
.ranges
[i
]);
5425 kvm_free_msr_filter(new_filter
);
5430 mutex_lock(&kvm
->lock
);
5432 /* The per-VM filter is protected by kvm->lock... */
5433 old_filter
= srcu_dereference_check(kvm
->arch
.msr_filter
, &kvm
->srcu
, 1);
5435 rcu_assign_pointer(kvm
->arch
.msr_filter
, new_filter
);
5436 synchronize_srcu(&kvm
->srcu
);
5438 kvm_free_msr_filter(old_filter
);
5440 kvm_make_all_cpus_request(kvm
, KVM_REQ_MSR_FILTER_CHANGED
);
5441 mutex_unlock(&kvm
->lock
);
5446 long kvm_arch_vm_ioctl(struct file
*filp
,
5447 unsigned int ioctl
, unsigned long arg
)
5449 struct kvm
*kvm
= filp
->private_data
;
5450 void __user
*argp
= (void __user
*)arg
;
5453 * This union makes it completely explicit to gcc-3.x
5454 * that these two variables' stack usage should be
5455 * combined, not added together.
5458 struct kvm_pit_state ps
;
5459 struct kvm_pit_state2 ps2
;
5460 struct kvm_pit_config pit_config
;
5464 case KVM_SET_TSS_ADDR
:
5465 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
5467 case KVM_SET_IDENTITY_MAP_ADDR
: {
5470 mutex_lock(&kvm
->lock
);
5472 if (kvm
->created_vcpus
)
5473 goto set_identity_unlock
;
5475 if (copy_from_user(&ident_addr
, argp
, sizeof(ident_addr
)))
5476 goto set_identity_unlock
;
5477 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
5478 set_identity_unlock
:
5479 mutex_unlock(&kvm
->lock
);
5482 case KVM_SET_NR_MMU_PAGES
:
5483 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
5485 case KVM_GET_NR_MMU_PAGES
:
5486 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
5488 case KVM_CREATE_IRQCHIP
: {
5489 mutex_lock(&kvm
->lock
);
5492 if (irqchip_in_kernel(kvm
))
5493 goto create_irqchip_unlock
;
5496 if (kvm
->created_vcpus
)
5497 goto create_irqchip_unlock
;
5499 r
= kvm_pic_init(kvm
);
5501 goto create_irqchip_unlock
;
5503 r
= kvm_ioapic_init(kvm
);
5505 kvm_pic_destroy(kvm
);
5506 goto create_irqchip_unlock
;
5509 r
= kvm_setup_default_irq_routing(kvm
);
5511 kvm_ioapic_destroy(kvm
);
5512 kvm_pic_destroy(kvm
);
5513 goto create_irqchip_unlock
;
5515 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5517 kvm
->arch
.irqchip_mode
= KVM_IRQCHIP_KERNEL
;
5518 create_irqchip_unlock
:
5519 mutex_unlock(&kvm
->lock
);
5522 case KVM_CREATE_PIT
:
5523 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
5525 case KVM_CREATE_PIT2
:
5527 if (copy_from_user(&u
.pit_config
, argp
,
5528 sizeof(struct kvm_pit_config
)))
5531 mutex_lock(&kvm
->lock
);
5534 goto create_pit_unlock
;
5536 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
5540 mutex_unlock(&kvm
->lock
);
5542 case KVM_GET_IRQCHIP
: {
5543 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5544 struct kvm_irqchip
*chip
;
5546 chip
= memdup_user(argp
, sizeof(*chip
));
5553 if (!irqchip_kernel(kvm
))
5554 goto get_irqchip_out
;
5555 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
5557 goto get_irqchip_out
;
5559 if (copy_to_user(argp
, chip
, sizeof(*chip
)))
5560 goto get_irqchip_out
;
5566 case KVM_SET_IRQCHIP
: {
5567 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5568 struct kvm_irqchip
*chip
;
5570 chip
= memdup_user(argp
, sizeof(*chip
));
5577 if (!irqchip_kernel(kvm
))
5578 goto set_irqchip_out
;
5579 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
5586 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
5589 if (!kvm
->arch
.vpit
)
5591 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
5595 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
5602 if (copy_from_user(&u
.ps
, argp
, sizeof(u
.ps
)))
5604 mutex_lock(&kvm
->lock
);
5606 if (!kvm
->arch
.vpit
)
5608 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
5610 mutex_unlock(&kvm
->lock
);
5613 case KVM_GET_PIT2
: {
5615 if (!kvm
->arch
.vpit
)
5617 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
5621 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
5626 case KVM_SET_PIT2
: {
5628 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
5630 mutex_lock(&kvm
->lock
);
5632 if (!kvm
->arch
.vpit
)
5634 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
5636 mutex_unlock(&kvm
->lock
);
5639 case KVM_REINJECT_CONTROL
: {
5640 struct kvm_reinject_control control
;
5642 if (copy_from_user(&control
, argp
, sizeof(control
)))
5645 if (!kvm
->arch
.vpit
)
5647 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
5650 case KVM_SET_BOOT_CPU_ID
:
5652 mutex_lock(&kvm
->lock
);
5653 if (kvm
->created_vcpus
)
5656 kvm
->arch
.bsp_vcpu_id
= arg
;
5657 mutex_unlock(&kvm
->lock
);
5659 case KVM_XEN_HVM_CONFIG
: {
5660 struct kvm_xen_hvm_config xhc
;
5662 if (copy_from_user(&xhc
, argp
, sizeof(xhc
)))
5667 memcpy(&kvm
->arch
.xen_hvm_config
, &xhc
, sizeof(xhc
));
5671 case KVM_SET_CLOCK
: {
5672 struct kvm_clock_data user_ns
;
5676 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
5685 * TODO: userspace has to take care of races with VCPU_RUN, so
5686 * kvm_gen_update_masterclock() can be cut down to locked
5687 * pvclock_update_vm_gtod_copy().
5689 kvm_gen_update_masterclock(kvm
);
5690 now_ns
= get_kvmclock_ns(kvm
);
5691 kvm
->arch
.kvmclock_offset
+= user_ns
.clock
- now_ns
;
5692 kvm_make_all_cpus_request(kvm
, KVM_REQ_CLOCK_UPDATE
);
5695 case KVM_GET_CLOCK
: {
5696 struct kvm_clock_data user_ns
;
5699 now_ns
= get_kvmclock_ns(kvm
);
5700 user_ns
.clock
= now_ns
;
5701 user_ns
.flags
= kvm
->arch
.use_master_clock
? KVM_CLOCK_TSC_STABLE
: 0;
5702 memset(&user_ns
.pad
, 0, sizeof(user_ns
.pad
));
5705 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
5710 case KVM_MEMORY_ENCRYPT_OP
: {
5712 if (kvm_x86_ops
.mem_enc_op
)
5713 r
= kvm_x86_ops
.mem_enc_op(kvm
, argp
);
5716 case KVM_MEMORY_ENCRYPT_REG_REGION
: {
5717 struct kvm_enc_region region
;
5720 if (copy_from_user(®ion
, argp
, sizeof(region
)))
5724 if (kvm_x86_ops
.mem_enc_reg_region
)
5725 r
= kvm_x86_ops
.mem_enc_reg_region(kvm
, ®ion
);
5728 case KVM_MEMORY_ENCRYPT_UNREG_REGION
: {
5729 struct kvm_enc_region region
;
5732 if (copy_from_user(®ion
, argp
, sizeof(region
)))
5736 if (kvm_x86_ops
.mem_enc_unreg_region
)
5737 r
= kvm_x86_ops
.mem_enc_unreg_region(kvm
, ®ion
);
5740 case KVM_HYPERV_EVENTFD
: {
5741 struct kvm_hyperv_eventfd hvevfd
;
5744 if (copy_from_user(&hvevfd
, argp
, sizeof(hvevfd
)))
5746 r
= kvm_vm_ioctl_hv_eventfd(kvm
, &hvevfd
);
5749 case KVM_SET_PMU_EVENT_FILTER
:
5750 r
= kvm_vm_ioctl_set_pmu_event_filter(kvm
, argp
);
5752 case KVM_X86_SET_MSR_FILTER
:
5753 r
= kvm_vm_ioctl_set_msr_filter(kvm
, argp
);
5762 static void kvm_init_msr_list(void)
5764 struct x86_pmu_capability x86_pmu
;
5768 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED
!= 4,
5769 "Please update the fixed PMCs in msrs_to_saved_all[]");
5771 perf_get_x86_pmu_capability(&x86_pmu
);
5773 num_msrs_to_save
= 0;
5774 num_emulated_msrs
= 0;
5775 num_msr_based_features
= 0;
5777 for (i
= 0; i
< ARRAY_SIZE(msrs_to_save_all
); i
++) {
5778 if (rdmsr_safe(msrs_to_save_all
[i
], &dummy
[0], &dummy
[1]) < 0)
5782 * Even MSRs that are valid in the host may not be exposed
5783 * to the guests in some cases.
5785 switch (msrs_to_save_all
[i
]) {
5786 case MSR_IA32_BNDCFGS
:
5787 if (!kvm_mpx_supported())
5791 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP
))
5794 case MSR_IA32_UMWAIT_CONTROL
:
5795 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG
))
5798 case MSR_IA32_RTIT_CTL
:
5799 case MSR_IA32_RTIT_STATUS
:
5800 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT
))
5803 case MSR_IA32_RTIT_CR3_MATCH
:
5804 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT
) ||
5805 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering
))
5808 case MSR_IA32_RTIT_OUTPUT_BASE
:
5809 case MSR_IA32_RTIT_OUTPUT_MASK
:
5810 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT
) ||
5811 (!intel_pt_validate_hw_cap(PT_CAP_topa_output
) &&
5812 !intel_pt_validate_hw_cap(PT_CAP_single_range_output
)))
5815 case MSR_IA32_RTIT_ADDR0_A
... MSR_IA32_RTIT_ADDR3_B
:
5816 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT
) ||
5817 msrs_to_save_all
[i
] - MSR_IA32_RTIT_ADDR0_A
>=
5818 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges
) * 2)
5821 case MSR_ARCH_PERFMON_PERFCTR0
... MSR_ARCH_PERFMON_PERFCTR0
+ 17:
5822 if (msrs_to_save_all
[i
] - MSR_ARCH_PERFMON_PERFCTR0
>=
5823 min(INTEL_PMC_MAX_GENERIC
, x86_pmu
.num_counters_gp
))
5826 case MSR_ARCH_PERFMON_EVENTSEL0
... MSR_ARCH_PERFMON_EVENTSEL0
+ 17:
5827 if (msrs_to_save_all
[i
] - MSR_ARCH_PERFMON_EVENTSEL0
>=
5828 min(INTEL_PMC_MAX_GENERIC
, x86_pmu
.num_counters_gp
))
5835 msrs_to_save
[num_msrs_to_save
++] = msrs_to_save_all
[i
];
5838 for (i
= 0; i
< ARRAY_SIZE(emulated_msrs_all
); i
++) {
5839 if (!kvm_x86_ops
.has_emulated_msr(NULL
, emulated_msrs_all
[i
]))
5842 emulated_msrs
[num_emulated_msrs
++] = emulated_msrs_all
[i
];
5845 for (i
= 0; i
< ARRAY_SIZE(msr_based_features_all
); i
++) {
5846 struct kvm_msr_entry msr
;
5848 msr
.index
= msr_based_features_all
[i
];
5849 if (kvm_get_msr_feature(&msr
))
5852 msr_based_features
[num_msr_based_features
++] = msr_based_features_all
[i
];
5856 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
5864 if (!(lapic_in_kernel(vcpu
) &&
5865 !kvm_iodevice_write(vcpu
, &vcpu
->arch
.apic
->dev
, addr
, n
, v
))
5866 && kvm_io_bus_write(vcpu
, KVM_MMIO_BUS
, addr
, n
, v
))
5877 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
5884 if (!(lapic_in_kernel(vcpu
) &&
5885 !kvm_iodevice_read(vcpu
, &vcpu
->arch
.apic
->dev
,
5887 && kvm_io_bus_read(vcpu
, KVM_MMIO_BUS
, addr
, n
, v
))
5889 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, n
, addr
, v
);
5899 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
5900 struct kvm_segment
*var
, int seg
)
5902 kvm_x86_ops
.set_segment(vcpu
, var
, seg
);
5905 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
5906 struct kvm_segment
*var
, int seg
)
5908 kvm_x86_ops
.get_segment(vcpu
, var
, seg
);
5911 gpa_t
translate_nested_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
,
5912 struct x86_exception
*exception
)
5916 BUG_ON(!mmu_is_nested(vcpu
));
5918 /* NPT walks are always user-walks */
5919 access
|= PFERR_USER_MASK
;
5920 t_gpa
= vcpu
->arch
.mmu
->gva_to_gpa(vcpu
, gpa
, access
, exception
);
5925 gpa_t
kvm_mmu_gva_to_gpa_read(struct kvm_vcpu
*vcpu
, gva_t gva
,
5926 struct x86_exception
*exception
)
5928 u32 access
= (kvm_x86_ops
.get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
5929 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
5932 gpa_t
kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu
*vcpu
, gva_t gva
,
5933 struct x86_exception
*exception
)
5935 u32 access
= (kvm_x86_ops
.get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
5936 access
|= PFERR_FETCH_MASK
;
5937 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
5940 gpa_t
kvm_mmu_gva_to_gpa_write(struct kvm_vcpu
*vcpu
, gva_t gva
,
5941 struct x86_exception
*exception
)
5943 u32 access
= (kvm_x86_ops
.get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
5944 access
|= PFERR_WRITE_MASK
;
5945 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
5948 /* uses this to access any guest's mapped memory without checking CPL */
5949 gpa_t
kvm_mmu_gva_to_gpa_system(struct kvm_vcpu
*vcpu
, gva_t gva
,
5950 struct x86_exception
*exception
)
5952 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, 0, exception
);
5955 static int kvm_read_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
5956 struct kvm_vcpu
*vcpu
, u32 access
,
5957 struct x86_exception
*exception
)
5960 int r
= X86EMUL_CONTINUE
;
5963 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
,
5965 unsigned offset
= addr
& (PAGE_SIZE
-1);
5966 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
5969 if (gpa
== UNMAPPED_GVA
)
5970 return X86EMUL_PROPAGATE_FAULT
;
5971 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa
>> PAGE_SHIFT
, data
,
5974 r
= X86EMUL_IO_NEEDED
;
5986 /* used for instruction fetching */
5987 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt
*ctxt
,
5988 gva_t addr
, void *val
, unsigned int bytes
,
5989 struct x86_exception
*exception
)
5991 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5992 u32 access
= (kvm_x86_ops
.get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
5996 /* Inline kvm_read_guest_virt_helper for speed. */
5997 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
|PFERR_FETCH_MASK
,
5999 if (unlikely(gpa
== UNMAPPED_GVA
))
6000 return X86EMUL_PROPAGATE_FAULT
;
6002 offset
= addr
& (PAGE_SIZE
-1);
6003 if (WARN_ON(offset
+ bytes
> PAGE_SIZE
))
6004 bytes
= (unsigned)PAGE_SIZE
- offset
;
6005 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa
>> PAGE_SHIFT
, val
,
6007 if (unlikely(ret
< 0))
6008 return X86EMUL_IO_NEEDED
;
6010 return X86EMUL_CONTINUE
;
6013 int kvm_read_guest_virt(struct kvm_vcpu
*vcpu
,
6014 gva_t addr
, void *val
, unsigned int bytes
,
6015 struct x86_exception
*exception
)
6017 u32 access
= (kvm_x86_ops
.get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
6020 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6021 * is returned, but our callers are not ready for that and they blindly
6022 * call kvm_inject_page_fault. Ensure that they at least do not leak
6023 * uninitialized kernel stack memory into cr2 and error code.
6025 memset(exception
, 0, sizeof(*exception
));
6026 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
,
6029 EXPORT_SYMBOL_GPL(kvm_read_guest_virt
);
6031 static int emulator_read_std(struct x86_emulate_ctxt
*ctxt
,
6032 gva_t addr
, void *val
, unsigned int bytes
,
6033 struct x86_exception
*exception
, bool system
)
6035 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6038 if (!system
&& kvm_x86_ops
.get_cpl(vcpu
) == 3)
6039 access
|= PFERR_USER_MASK
;
6041 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
, exception
);
6044 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt
*ctxt
,
6045 unsigned long addr
, void *val
, unsigned int bytes
)
6047 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6048 int r
= kvm_vcpu_read_guest(vcpu
, addr
, val
, bytes
);
6050 return r
< 0 ? X86EMUL_IO_NEEDED
: X86EMUL_CONTINUE
;
6053 static int kvm_write_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
6054 struct kvm_vcpu
*vcpu
, u32 access
,
6055 struct x86_exception
*exception
)
6058 int r
= X86EMUL_CONTINUE
;
6061 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
,
6064 unsigned offset
= addr
& (PAGE_SIZE
-1);
6065 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
6068 if (gpa
== UNMAPPED_GVA
)
6069 return X86EMUL_PROPAGATE_FAULT
;
6070 ret
= kvm_vcpu_write_guest(vcpu
, gpa
, data
, towrite
);
6072 r
= X86EMUL_IO_NEEDED
;
6084 static int emulator_write_std(struct x86_emulate_ctxt
*ctxt
, gva_t addr
, void *val
,
6085 unsigned int bytes
, struct x86_exception
*exception
,
6088 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6089 u32 access
= PFERR_WRITE_MASK
;
6091 if (!system
&& kvm_x86_ops
.get_cpl(vcpu
) == 3)
6092 access
|= PFERR_USER_MASK
;
6094 return kvm_write_guest_virt_helper(addr
, val
, bytes
, vcpu
,
6098 int kvm_write_guest_virt_system(struct kvm_vcpu
*vcpu
, gva_t addr
, void *val
,
6099 unsigned int bytes
, struct x86_exception
*exception
)
6101 /* kvm_write_guest_virt_system can pull in tons of pages. */
6102 vcpu
->arch
.l1tf_flush_l1d
= true;
6104 return kvm_write_guest_virt_helper(addr
, val
, bytes
, vcpu
,
6105 PFERR_WRITE_MASK
, exception
);
6107 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system
);
6109 int handle_ud(struct kvm_vcpu
*vcpu
)
6111 static const char kvm_emulate_prefix
[] = { __KVM_EMULATE_PREFIX
};
6112 int emul_type
= EMULTYPE_TRAP_UD
;
6113 char sig
[5]; /* ud2; .ascii "kvm" */
6114 struct x86_exception e
;
6116 if (unlikely(!kvm_x86_ops
.can_emulate_instruction(vcpu
, NULL
, 0)))
6119 if (force_emulation_prefix
&&
6120 kvm_read_guest_virt(vcpu
, kvm_get_linear_rip(vcpu
),
6121 sig
, sizeof(sig
), &e
) == 0 &&
6122 memcmp(sig
, kvm_emulate_prefix
, sizeof(sig
)) == 0) {
6123 kvm_rip_write(vcpu
, kvm_rip_read(vcpu
) + sizeof(sig
));
6124 emul_type
= EMULTYPE_TRAP_UD_FORCED
;
6127 return kvm_emulate_instruction(vcpu
, emul_type
);
6129 EXPORT_SYMBOL_GPL(handle_ud
);
6131 static int vcpu_is_mmio_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
6132 gpa_t gpa
, bool write
)
6134 /* For APIC access vmexit */
6135 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
6138 if (vcpu_match_mmio_gpa(vcpu
, gpa
)) {
6139 trace_vcpu_match_mmio(gva
, gpa
, write
, true);
6146 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
6147 gpa_t
*gpa
, struct x86_exception
*exception
,
6150 u32 access
= ((kvm_x86_ops
.get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0)
6151 | (write
? PFERR_WRITE_MASK
: 0);
6154 * currently PKRU is only applied to ept enabled guest so
6155 * there is no pkey in EPT page table for L1 guest or EPT
6156 * shadow page table for L2 guest.
6158 if (vcpu_match_mmio_gva(vcpu
, gva
)
6159 && !permission_fault(vcpu
, vcpu
->arch
.walk_mmu
,
6160 vcpu
->arch
.mmio_access
, 0, access
)) {
6161 *gpa
= vcpu
->arch
.mmio_gfn
<< PAGE_SHIFT
|
6162 (gva
& (PAGE_SIZE
- 1));
6163 trace_vcpu_match_mmio(gva
, *gpa
, write
, false);
6167 *gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
6169 if (*gpa
== UNMAPPED_GVA
)
6172 return vcpu_is_mmio_gpa(vcpu
, gva
, *gpa
, write
);
6175 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
6176 const void *val
, int bytes
)
6180 ret
= kvm_vcpu_write_guest(vcpu
, gpa
, val
, bytes
);
6183 kvm_page_track_write(vcpu
, gpa
, val
, bytes
);
6187 struct read_write_emulator_ops
{
6188 int (*read_write_prepare
)(struct kvm_vcpu
*vcpu
, void *val
,
6190 int (*read_write_emulate
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
6191 void *val
, int bytes
);
6192 int (*read_write_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
6193 int bytes
, void *val
);
6194 int (*read_write_exit_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
6195 void *val
, int bytes
);
6199 static int read_prepare(struct kvm_vcpu
*vcpu
, void *val
, int bytes
)
6201 if (vcpu
->mmio_read_completed
) {
6202 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
6203 vcpu
->mmio_fragments
[0].gpa
, val
);
6204 vcpu
->mmio_read_completed
= 0;
6211 static int read_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
6212 void *val
, int bytes
)
6214 return !kvm_vcpu_read_guest(vcpu
, gpa
, val
, bytes
);
6217 static int write_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
6218 void *val
, int bytes
)
6220 return emulator_write_phys(vcpu
, gpa
, val
, bytes
);
6223 static int write_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
, void *val
)
6225 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, val
);
6226 return vcpu_mmio_write(vcpu
, gpa
, bytes
, val
);
6229 static int read_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
6230 void *val
, int bytes
)
6232 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, NULL
);
6233 return X86EMUL_IO_NEEDED
;
6236 static int write_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
6237 void *val
, int bytes
)
6239 struct kvm_mmio_fragment
*frag
= &vcpu
->mmio_fragments
[0];
6241 memcpy(vcpu
->run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
6242 return X86EMUL_CONTINUE
;
6245 static const struct read_write_emulator_ops read_emultor
= {
6246 .read_write_prepare
= read_prepare
,
6247 .read_write_emulate
= read_emulate
,
6248 .read_write_mmio
= vcpu_mmio_read
,
6249 .read_write_exit_mmio
= read_exit_mmio
,
6252 static const struct read_write_emulator_ops write_emultor
= {
6253 .read_write_emulate
= write_emulate
,
6254 .read_write_mmio
= write_mmio
,
6255 .read_write_exit_mmio
= write_exit_mmio
,
6259 static int emulator_read_write_onepage(unsigned long addr
, void *val
,
6261 struct x86_exception
*exception
,
6262 struct kvm_vcpu
*vcpu
,
6263 const struct read_write_emulator_ops
*ops
)
6267 bool write
= ops
->write
;
6268 struct kvm_mmio_fragment
*frag
;
6269 struct x86_emulate_ctxt
*ctxt
= vcpu
->arch
.emulate_ctxt
;
6272 * If the exit was due to a NPF we may already have a GPA.
6273 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6274 * Note, this cannot be used on string operations since string
6275 * operation using rep will only have the initial GPA from the NPF
6278 if (ctxt
->gpa_available
&& emulator_can_use_gpa(ctxt
) &&
6279 (addr
& ~PAGE_MASK
) == (ctxt
->gpa_val
& ~PAGE_MASK
)) {
6280 gpa
= ctxt
->gpa_val
;
6281 ret
= vcpu_is_mmio_gpa(vcpu
, addr
, gpa
, write
);
6283 ret
= vcpu_mmio_gva_to_gpa(vcpu
, addr
, &gpa
, exception
, write
);
6285 return X86EMUL_PROPAGATE_FAULT
;
6288 if (!ret
&& ops
->read_write_emulate(vcpu
, gpa
, val
, bytes
))
6289 return X86EMUL_CONTINUE
;
6292 * Is this MMIO handled locally?
6294 handled
= ops
->read_write_mmio(vcpu
, gpa
, bytes
, val
);
6295 if (handled
== bytes
)
6296 return X86EMUL_CONTINUE
;
6302 WARN_ON(vcpu
->mmio_nr_fragments
>= KVM_MAX_MMIO_FRAGMENTS
);
6303 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_nr_fragments
++];
6307 return X86EMUL_CONTINUE
;
6310 static int emulator_read_write(struct x86_emulate_ctxt
*ctxt
,
6312 void *val
, unsigned int bytes
,
6313 struct x86_exception
*exception
,
6314 const struct read_write_emulator_ops
*ops
)
6316 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6320 if (ops
->read_write_prepare
&&
6321 ops
->read_write_prepare(vcpu
, val
, bytes
))
6322 return X86EMUL_CONTINUE
;
6324 vcpu
->mmio_nr_fragments
= 0;
6326 /* Crossing a page boundary? */
6327 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
6330 now
= -addr
& ~PAGE_MASK
;
6331 rc
= emulator_read_write_onepage(addr
, val
, now
, exception
,
6334 if (rc
!= X86EMUL_CONTINUE
)
6337 if (ctxt
->mode
!= X86EMUL_MODE_PROT64
)
6343 rc
= emulator_read_write_onepage(addr
, val
, bytes
, exception
,
6345 if (rc
!= X86EMUL_CONTINUE
)
6348 if (!vcpu
->mmio_nr_fragments
)
6351 gpa
= vcpu
->mmio_fragments
[0].gpa
;
6353 vcpu
->mmio_needed
= 1;
6354 vcpu
->mmio_cur_fragment
= 0;
6356 vcpu
->run
->mmio
.len
= min(8u, vcpu
->mmio_fragments
[0].len
);
6357 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= ops
->write
;
6358 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
6359 vcpu
->run
->mmio
.phys_addr
= gpa
;
6361 return ops
->read_write_exit_mmio(vcpu
, gpa
, val
, bytes
);
6364 static int emulator_read_emulated(struct x86_emulate_ctxt
*ctxt
,
6368 struct x86_exception
*exception
)
6370 return emulator_read_write(ctxt
, addr
, val
, bytes
,
6371 exception
, &read_emultor
);
6374 static int emulator_write_emulated(struct x86_emulate_ctxt
*ctxt
,
6378 struct x86_exception
*exception
)
6380 return emulator_read_write(ctxt
, addr
, (void *)val
, bytes
,
6381 exception
, &write_emultor
);
6384 #define CMPXCHG_TYPE(t, ptr, old, new) \
6385 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6387 #ifdef CONFIG_X86_64
6388 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6390 # define CMPXCHG64(ptr, old, new) \
6391 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6394 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt
*ctxt
,
6399 struct x86_exception
*exception
)
6401 struct kvm_host_map map
;
6402 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6408 /* guests cmpxchg8b have to be emulated atomically */
6409 if (bytes
> 8 || (bytes
& (bytes
- 1)))
6412 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, NULL
);
6414 if (gpa
== UNMAPPED_GVA
||
6415 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
6419 * Emulate the atomic as a straight write to avoid #AC if SLD is
6420 * enabled in the host and the access splits a cache line.
6422 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT
))
6423 page_line_mask
= ~(cache_line_size() - 1);
6425 page_line_mask
= PAGE_MASK
;
6427 if (((gpa
+ bytes
- 1) & page_line_mask
) != (gpa
& page_line_mask
))
6430 if (kvm_vcpu_map(vcpu
, gpa_to_gfn(gpa
), &map
))
6433 kaddr
= map
.hva
+ offset_in_page(gpa
);
6437 exchanged
= CMPXCHG_TYPE(u8
, kaddr
, old
, new);
6440 exchanged
= CMPXCHG_TYPE(u16
, kaddr
, old
, new);
6443 exchanged
= CMPXCHG_TYPE(u32
, kaddr
, old
, new);
6446 exchanged
= CMPXCHG64(kaddr
, old
, new);
6452 kvm_vcpu_unmap(vcpu
, &map
, true);
6455 return X86EMUL_CMPXCHG_FAILED
;
6457 kvm_page_track_write(vcpu
, gpa
, new, bytes
);
6459 return X86EMUL_CONTINUE
;
6462 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
6464 return emulator_write_emulated(ctxt
, addr
, new, bytes
, exception
);
6467 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
6471 for (i
= 0; i
< vcpu
->arch
.pio
.count
; i
++) {
6472 if (vcpu
->arch
.pio
.in
)
6473 r
= kvm_io_bus_read(vcpu
, KVM_PIO_BUS
, vcpu
->arch
.pio
.port
,
6474 vcpu
->arch
.pio
.size
, pd
);
6476 r
= kvm_io_bus_write(vcpu
, KVM_PIO_BUS
,
6477 vcpu
->arch
.pio
.port
, vcpu
->arch
.pio
.size
,
6481 pd
+= vcpu
->arch
.pio
.size
;
6486 static int emulator_pio_in_out(struct kvm_vcpu
*vcpu
, int size
,
6487 unsigned short port
, void *val
,
6488 unsigned int count
, bool in
)
6490 vcpu
->arch
.pio
.port
= port
;
6491 vcpu
->arch
.pio
.in
= in
;
6492 vcpu
->arch
.pio
.count
= count
;
6493 vcpu
->arch
.pio
.size
= size
;
6495 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
6496 vcpu
->arch
.pio
.count
= 0;
6500 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
6501 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
6502 vcpu
->run
->io
.size
= size
;
6503 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
6504 vcpu
->run
->io
.count
= count
;
6505 vcpu
->run
->io
.port
= port
;
6510 static int emulator_pio_in(struct kvm_vcpu
*vcpu
, int size
,
6511 unsigned short port
, void *val
, unsigned int count
)
6515 if (vcpu
->arch
.pio
.count
)
6518 memset(vcpu
->arch
.pio_data
, 0, size
* count
);
6520 ret
= emulator_pio_in_out(vcpu
, size
, port
, val
, count
, true);
6523 memcpy(val
, vcpu
->arch
.pio_data
, size
* count
);
6524 trace_kvm_pio(KVM_PIO_IN
, port
, size
, count
, vcpu
->arch
.pio_data
);
6525 vcpu
->arch
.pio
.count
= 0;
6532 static int emulator_pio_in_emulated(struct x86_emulate_ctxt
*ctxt
,
6533 int size
, unsigned short port
, void *val
,
6536 return emulator_pio_in(emul_to_vcpu(ctxt
), size
, port
, val
, count
);
6540 static int emulator_pio_out(struct kvm_vcpu
*vcpu
, int size
,
6541 unsigned short port
, const void *val
,
6544 memcpy(vcpu
->arch
.pio_data
, val
, size
* count
);
6545 trace_kvm_pio(KVM_PIO_OUT
, port
, size
, count
, vcpu
->arch
.pio_data
);
6546 return emulator_pio_in_out(vcpu
, size
, port
, (void *)val
, count
, false);
6549 static int emulator_pio_out_emulated(struct x86_emulate_ctxt
*ctxt
,
6550 int size
, unsigned short port
,
6551 const void *val
, unsigned int count
)
6553 return emulator_pio_out(emul_to_vcpu(ctxt
), size
, port
, val
, count
);
6556 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
6558 return kvm_x86_ops
.get_segment_base(vcpu
, seg
);
6561 static void emulator_invlpg(struct x86_emulate_ctxt
*ctxt
, ulong address
)
6563 kvm_mmu_invlpg(emul_to_vcpu(ctxt
), address
);
6566 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu
*vcpu
)
6568 if (!need_emulate_wbinvd(vcpu
))
6569 return X86EMUL_CONTINUE
;
6571 if (kvm_x86_ops
.has_wbinvd_exit()) {
6572 int cpu
= get_cpu();
6574 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
6575 smp_call_function_many(vcpu
->arch
.wbinvd_dirty_mask
,
6576 wbinvd_ipi
, NULL
, 1);
6578 cpumask_clear(vcpu
->arch
.wbinvd_dirty_mask
);
6581 return X86EMUL_CONTINUE
;
6584 int kvm_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
6586 kvm_emulate_wbinvd_noskip(vcpu
);
6587 return kvm_skip_emulated_instruction(vcpu
);
6589 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd
);
6593 static void emulator_wbinvd(struct x86_emulate_ctxt
*ctxt
)
6595 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt
));
6598 static int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
,
6599 unsigned long *dest
)
6601 return kvm_get_dr(emul_to_vcpu(ctxt
), dr
, dest
);
6604 static int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
,
6605 unsigned long value
)
6608 return __kvm_set_dr(emul_to_vcpu(ctxt
), dr
, value
);
6611 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
6613 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
6616 static unsigned long emulator_get_cr(struct x86_emulate_ctxt
*ctxt
, int cr
)
6618 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6619 unsigned long value
;
6623 value
= kvm_read_cr0(vcpu
);
6626 value
= vcpu
->arch
.cr2
;
6629 value
= kvm_read_cr3(vcpu
);
6632 value
= kvm_read_cr4(vcpu
);
6635 value
= kvm_get_cr8(vcpu
);
6638 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
6645 static int emulator_set_cr(struct x86_emulate_ctxt
*ctxt
, int cr
, ulong val
)
6647 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6652 res
= kvm_set_cr0(vcpu
, mk_cr_64(kvm_read_cr0(vcpu
), val
));
6655 vcpu
->arch
.cr2
= val
;
6658 res
= kvm_set_cr3(vcpu
, val
);
6661 res
= kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
6664 res
= kvm_set_cr8(vcpu
, val
);
6667 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
6674 static int emulator_get_cpl(struct x86_emulate_ctxt
*ctxt
)
6676 return kvm_x86_ops
.get_cpl(emul_to_vcpu(ctxt
));
6679 static void emulator_get_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
6681 kvm_x86_ops
.get_gdt(emul_to_vcpu(ctxt
), dt
);
6684 static void emulator_get_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
6686 kvm_x86_ops
.get_idt(emul_to_vcpu(ctxt
), dt
);
6689 static void emulator_set_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
6691 kvm_x86_ops
.set_gdt(emul_to_vcpu(ctxt
), dt
);
6694 static void emulator_set_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
6696 kvm_x86_ops
.set_idt(emul_to_vcpu(ctxt
), dt
);
6699 static unsigned long emulator_get_cached_segment_base(
6700 struct x86_emulate_ctxt
*ctxt
, int seg
)
6702 return get_segment_base(emul_to_vcpu(ctxt
), seg
);
6705 static bool emulator_get_segment(struct x86_emulate_ctxt
*ctxt
, u16
*selector
,
6706 struct desc_struct
*desc
, u32
*base3
,
6709 struct kvm_segment var
;
6711 kvm_get_segment(emul_to_vcpu(ctxt
), &var
, seg
);
6712 *selector
= var
.selector
;
6715 memset(desc
, 0, sizeof(*desc
));
6723 set_desc_limit(desc
, var
.limit
);
6724 set_desc_base(desc
, (unsigned long)var
.base
);
6725 #ifdef CONFIG_X86_64
6727 *base3
= var
.base
>> 32;
6729 desc
->type
= var
.type
;
6731 desc
->dpl
= var
.dpl
;
6732 desc
->p
= var
.present
;
6733 desc
->avl
= var
.avl
;
6741 static void emulator_set_segment(struct x86_emulate_ctxt
*ctxt
, u16 selector
,
6742 struct desc_struct
*desc
, u32 base3
,
6745 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6746 struct kvm_segment var
;
6748 var
.selector
= selector
;
6749 var
.base
= get_desc_base(desc
);
6750 #ifdef CONFIG_X86_64
6751 var
.base
|= ((u64
)base3
) << 32;
6753 var
.limit
= get_desc_limit(desc
);
6755 var
.limit
= (var
.limit
<< 12) | 0xfff;
6756 var
.type
= desc
->type
;
6757 var
.dpl
= desc
->dpl
;
6762 var
.avl
= desc
->avl
;
6763 var
.present
= desc
->p
;
6764 var
.unusable
= !var
.present
;
6767 kvm_set_segment(vcpu
, &var
, seg
);
6771 static int emulator_get_msr(struct x86_emulate_ctxt
*ctxt
,
6772 u32 msr_index
, u64
*pdata
)
6774 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6777 r
= kvm_get_msr(vcpu
, msr_index
, pdata
);
6779 if (r
&& kvm_get_msr_user_space(vcpu
, msr_index
, r
)) {
6780 /* Bounce to user space */
6781 return X86EMUL_IO_NEEDED
;
6787 static int emulator_set_msr(struct x86_emulate_ctxt
*ctxt
,
6788 u32 msr_index
, u64 data
)
6790 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6793 r
= kvm_set_msr(vcpu
, msr_index
, data
);
6795 if (r
&& kvm_set_msr_user_space(vcpu
, msr_index
, data
, r
)) {
6796 /* Bounce to user space */
6797 return X86EMUL_IO_NEEDED
;
6803 static u64
emulator_get_smbase(struct x86_emulate_ctxt
*ctxt
)
6805 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6807 return vcpu
->arch
.smbase
;
6810 static void emulator_set_smbase(struct x86_emulate_ctxt
*ctxt
, u64 smbase
)
6812 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6814 vcpu
->arch
.smbase
= smbase
;
6817 static int emulator_check_pmc(struct x86_emulate_ctxt
*ctxt
,
6820 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt
), pmc
);
6823 static int emulator_read_pmc(struct x86_emulate_ctxt
*ctxt
,
6824 u32 pmc
, u64
*pdata
)
6826 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt
), pmc
, pdata
);
6829 static void emulator_halt(struct x86_emulate_ctxt
*ctxt
)
6831 emul_to_vcpu(ctxt
)->arch
.halt_request
= 1;
6834 static int emulator_intercept(struct x86_emulate_ctxt
*ctxt
,
6835 struct x86_instruction_info
*info
,
6836 enum x86_intercept_stage stage
)
6838 return kvm_x86_ops
.check_intercept(emul_to_vcpu(ctxt
), info
, stage
,
6842 static bool emulator_get_cpuid(struct x86_emulate_ctxt
*ctxt
,
6843 u32
*eax
, u32
*ebx
, u32
*ecx
, u32
*edx
,
6846 return kvm_cpuid(emul_to_vcpu(ctxt
), eax
, ebx
, ecx
, edx
, exact_only
);
6849 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt
*ctxt
)
6851 return guest_cpuid_has(emul_to_vcpu(ctxt
), X86_FEATURE_LM
);
6854 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt
*ctxt
)
6856 return guest_cpuid_has(emul_to_vcpu(ctxt
), X86_FEATURE_MOVBE
);
6859 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt
*ctxt
)
6861 return guest_cpuid_has(emul_to_vcpu(ctxt
), X86_FEATURE_FXSR
);
6864 static ulong
emulator_read_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
)
6866 return kvm_register_read(emul_to_vcpu(ctxt
), reg
);
6869 static void emulator_write_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
, ulong val
)
6871 kvm_register_write(emul_to_vcpu(ctxt
), reg
, val
);
6874 static void emulator_set_nmi_mask(struct x86_emulate_ctxt
*ctxt
, bool masked
)
6876 kvm_x86_ops
.set_nmi_mask(emul_to_vcpu(ctxt
), masked
);
6879 static unsigned emulator_get_hflags(struct x86_emulate_ctxt
*ctxt
)
6881 return emul_to_vcpu(ctxt
)->arch
.hflags
;
6884 static void emulator_set_hflags(struct x86_emulate_ctxt
*ctxt
, unsigned emul_flags
)
6886 emul_to_vcpu(ctxt
)->arch
.hflags
= emul_flags
;
6889 static int emulator_pre_leave_smm(struct x86_emulate_ctxt
*ctxt
,
6890 const char *smstate
)
6892 return kvm_x86_ops
.pre_leave_smm(emul_to_vcpu(ctxt
), smstate
);
6895 static void emulator_post_leave_smm(struct x86_emulate_ctxt
*ctxt
)
6897 kvm_smm_changed(emul_to_vcpu(ctxt
));
6900 static int emulator_set_xcr(struct x86_emulate_ctxt
*ctxt
, u32 index
, u64 xcr
)
6902 return __kvm_set_xcr(emul_to_vcpu(ctxt
), index
, xcr
);
6905 static const struct x86_emulate_ops emulate_ops
= {
6906 .read_gpr
= emulator_read_gpr
,
6907 .write_gpr
= emulator_write_gpr
,
6908 .read_std
= emulator_read_std
,
6909 .write_std
= emulator_write_std
,
6910 .read_phys
= kvm_read_guest_phys_system
,
6911 .fetch
= kvm_fetch_guest_virt
,
6912 .read_emulated
= emulator_read_emulated
,
6913 .write_emulated
= emulator_write_emulated
,
6914 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
6915 .invlpg
= emulator_invlpg
,
6916 .pio_in_emulated
= emulator_pio_in_emulated
,
6917 .pio_out_emulated
= emulator_pio_out_emulated
,
6918 .get_segment
= emulator_get_segment
,
6919 .set_segment
= emulator_set_segment
,
6920 .get_cached_segment_base
= emulator_get_cached_segment_base
,
6921 .get_gdt
= emulator_get_gdt
,
6922 .get_idt
= emulator_get_idt
,
6923 .set_gdt
= emulator_set_gdt
,
6924 .set_idt
= emulator_set_idt
,
6925 .get_cr
= emulator_get_cr
,
6926 .set_cr
= emulator_set_cr
,
6927 .cpl
= emulator_get_cpl
,
6928 .get_dr
= emulator_get_dr
,
6929 .set_dr
= emulator_set_dr
,
6930 .get_smbase
= emulator_get_smbase
,
6931 .set_smbase
= emulator_set_smbase
,
6932 .set_msr
= emulator_set_msr
,
6933 .get_msr
= emulator_get_msr
,
6934 .check_pmc
= emulator_check_pmc
,
6935 .read_pmc
= emulator_read_pmc
,
6936 .halt
= emulator_halt
,
6937 .wbinvd
= emulator_wbinvd
,
6938 .fix_hypercall
= emulator_fix_hypercall
,
6939 .intercept
= emulator_intercept
,
6940 .get_cpuid
= emulator_get_cpuid
,
6941 .guest_has_long_mode
= emulator_guest_has_long_mode
,
6942 .guest_has_movbe
= emulator_guest_has_movbe
,
6943 .guest_has_fxsr
= emulator_guest_has_fxsr
,
6944 .set_nmi_mask
= emulator_set_nmi_mask
,
6945 .get_hflags
= emulator_get_hflags
,
6946 .set_hflags
= emulator_set_hflags
,
6947 .pre_leave_smm
= emulator_pre_leave_smm
,
6948 .post_leave_smm
= emulator_post_leave_smm
,
6949 .set_xcr
= emulator_set_xcr
,
6952 static void toggle_interruptibility(struct kvm_vcpu
*vcpu
, u32 mask
)
6954 u32 int_shadow
= kvm_x86_ops
.get_interrupt_shadow(vcpu
);
6956 * an sti; sti; sequence only disable interrupts for the first
6957 * instruction. So, if the last instruction, be it emulated or
6958 * not, left the system with the INT_STI flag enabled, it
6959 * means that the last instruction is an sti. We should not
6960 * leave the flag on in this case. The same goes for mov ss
6962 if (int_shadow
& mask
)
6964 if (unlikely(int_shadow
|| mask
)) {
6965 kvm_x86_ops
.set_interrupt_shadow(vcpu
, mask
);
6967 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6971 static bool inject_emulated_exception(struct kvm_vcpu
*vcpu
)
6973 struct x86_emulate_ctxt
*ctxt
= vcpu
->arch
.emulate_ctxt
;
6974 if (ctxt
->exception
.vector
== PF_VECTOR
)
6975 return kvm_inject_emulated_page_fault(vcpu
, &ctxt
->exception
);
6977 if (ctxt
->exception
.error_code_valid
)
6978 kvm_queue_exception_e(vcpu
, ctxt
->exception
.vector
,
6979 ctxt
->exception
.error_code
);
6981 kvm_queue_exception(vcpu
, ctxt
->exception
.vector
);
6985 static struct x86_emulate_ctxt
*alloc_emulate_ctxt(struct kvm_vcpu
*vcpu
)
6987 struct x86_emulate_ctxt
*ctxt
;
6989 ctxt
= kmem_cache_zalloc(x86_emulator_cache
, GFP_KERNEL_ACCOUNT
);
6991 pr_err("kvm: failed to allocate vcpu's emulator\n");
6996 ctxt
->ops
= &emulate_ops
;
6997 vcpu
->arch
.emulate_ctxt
= ctxt
;
7002 static void init_emulate_ctxt(struct kvm_vcpu
*vcpu
)
7004 struct x86_emulate_ctxt
*ctxt
= vcpu
->arch
.emulate_ctxt
;
7007 kvm_x86_ops
.get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
7009 ctxt
->gpa_available
= false;
7010 ctxt
->eflags
= kvm_get_rflags(vcpu
);
7011 ctxt
->tf
= (ctxt
->eflags
& X86_EFLAGS_TF
) != 0;
7013 ctxt
->eip
= kvm_rip_read(vcpu
);
7014 ctxt
->mode
= (!is_protmode(vcpu
)) ? X86EMUL_MODE_REAL
:
7015 (ctxt
->eflags
& X86_EFLAGS_VM
) ? X86EMUL_MODE_VM86
:
7016 (cs_l
&& is_long_mode(vcpu
)) ? X86EMUL_MODE_PROT64
:
7017 cs_db
? X86EMUL_MODE_PROT32
:
7018 X86EMUL_MODE_PROT16
;
7019 BUILD_BUG_ON(HF_GUEST_MASK
!= X86EMUL_GUEST_MASK
);
7020 BUILD_BUG_ON(HF_SMM_MASK
!= X86EMUL_SMM_MASK
);
7021 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK
!= X86EMUL_SMM_INSIDE_NMI_MASK
);
7023 init_decode_cache(ctxt
);
7024 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
7027 void kvm_inject_realmode_interrupt(struct kvm_vcpu
*vcpu
, int irq
, int inc_eip
)
7029 struct x86_emulate_ctxt
*ctxt
= vcpu
->arch
.emulate_ctxt
;
7032 init_emulate_ctxt(vcpu
);
7036 ctxt
->_eip
= ctxt
->eip
+ inc_eip
;
7037 ret
= emulate_int_real(ctxt
, irq
);
7039 if (ret
!= X86EMUL_CONTINUE
) {
7040 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
7042 ctxt
->eip
= ctxt
->_eip
;
7043 kvm_rip_write(vcpu
, ctxt
->eip
);
7044 kvm_set_rflags(vcpu
, ctxt
->eflags
);
7047 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt
);
7049 static int handle_emulation_failure(struct kvm_vcpu
*vcpu
, int emulation_type
)
7051 ++vcpu
->stat
.insn_emulation_fail
;
7052 trace_kvm_emulate_insn_failed(vcpu
);
7054 if (emulation_type
& EMULTYPE_VMWARE_GP
) {
7055 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
7059 if (emulation_type
& EMULTYPE_SKIP
) {
7060 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
7061 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
7062 vcpu
->run
->internal
.ndata
= 0;
7066 kvm_queue_exception(vcpu
, UD_VECTOR
);
7068 if (!is_guest_mode(vcpu
) && kvm_x86_ops
.get_cpl(vcpu
) == 0) {
7069 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
7070 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
7071 vcpu
->run
->internal
.ndata
= 0;
7078 static bool reexecute_instruction(struct kvm_vcpu
*vcpu
, gpa_t cr2_or_gpa
,
7079 bool write_fault_to_shadow_pgtable
,
7082 gpa_t gpa
= cr2_or_gpa
;
7085 if (!(emulation_type
& EMULTYPE_ALLOW_RETRY_PF
))
7088 if (WARN_ON_ONCE(is_guest_mode(vcpu
)) ||
7089 WARN_ON_ONCE(!(emulation_type
& EMULTYPE_PF
)))
7092 if (!vcpu
->arch
.mmu
->direct_map
) {
7094 * Write permission should be allowed since only
7095 * write access need to be emulated.
7097 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2_or_gpa
, NULL
);
7100 * If the mapping is invalid in guest, let cpu retry
7101 * it to generate fault.
7103 if (gpa
== UNMAPPED_GVA
)
7108 * Do not retry the unhandleable instruction if it faults on the
7109 * readonly host memory, otherwise it will goto a infinite loop:
7110 * retry instruction -> write #PF -> emulation fail -> retry
7111 * instruction -> ...
7113 pfn
= gfn_to_pfn(vcpu
->kvm
, gpa_to_gfn(gpa
));
7116 * If the instruction failed on the error pfn, it can not be fixed,
7117 * report the error to userspace.
7119 if (is_error_noslot_pfn(pfn
))
7122 kvm_release_pfn_clean(pfn
);
7124 /* The instructions are well-emulated on direct mmu. */
7125 if (vcpu
->arch
.mmu
->direct_map
) {
7126 unsigned int indirect_shadow_pages
;
7128 spin_lock(&vcpu
->kvm
->mmu_lock
);
7129 indirect_shadow_pages
= vcpu
->kvm
->arch
.indirect_shadow_pages
;
7130 spin_unlock(&vcpu
->kvm
->mmu_lock
);
7132 if (indirect_shadow_pages
)
7133 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
7139 * if emulation was due to access to shadowed page table
7140 * and it failed try to unshadow page and re-enter the
7141 * guest to let CPU execute the instruction.
7143 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
7146 * If the access faults on its page table, it can not
7147 * be fixed by unprotecting shadow page and it should
7148 * be reported to userspace.
7150 return !write_fault_to_shadow_pgtable
;
7153 static bool retry_instruction(struct x86_emulate_ctxt
*ctxt
,
7154 gpa_t cr2_or_gpa
, int emulation_type
)
7156 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
7157 unsigned long last_retry_eip
, last_retry_addr
, gpa
= cr2_or_gpa
;
7159 last_retry_eip
= vcpu
->arch
.last_retry_eip
;
7160 last_retry_addr
= vcpu
->arch
.last_retry_addr
;
7163 * If the emulation is caused by #PF and it is non-page_table
7164 * writing instruction, it means the VM-EXIT is caused by shadow
7165 * page protected, we can zap the shadow page and retry this
7166 * instruction directly.
7168 * Note: if the guest uses a non-page-table modifying instruction
7169 * on the PDE that points to the instruction, then we will unmap
7170 * the instruction and go to an infinite loop. So, we cache the
7171 * last retried eip and the last fault address, if we meet the eip
7172 * and the address again, we can break out of the potential infinite
7175 vcpu
->arch
.last_retry_eip
= vcpu
->arch
.last_retry_addr
= 0;
7177 if (!(emulation_type
& EMULTYPE_ALLOW_RETRY_PF
))
7180 if (WARN_ON_ONCE(is_guest_mode(vcpu
)) ||
7181 WARN_ON_ONCE(!(emulation_type
& EMULTYPE_PF
)))
7184 if (x86_page_table_writing_insn(ctxt
))
7187 if (ctxt
->eip
== last_retry_eip
&& last_retry_addr
== cr2_or_gpa
)
7190 vcpu
->arch
.last_retry_eip
= ctxt
->eip
;
7191 vcpu
->arch
.last_retry_addr
= cr2_or_gpa
;
7193 if (!vcpu
->arch
.mmu
->direct_map
)
7194 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2_or_gpa
, NULL
);
7196 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
7201 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
);
7202 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
);
7204 static void kvm_smm_changed(struct kvm_vcpu
*vcpu
)
7206 if (!(vcpu
->arch
.hflags
& HF_SMM_MASK
)) {
7207 /* This is a good place to trace that we are exiting SMM. */
7208 trace_kvm_enter_smm(vcpu
->vcpu_id
, vcpu
->arch
.smbase
, false);
7210 /* Process a latched INIT or SMI, if any. */
7211 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7214 kvm_mmu_reset_context(vcpu
);
7217 static int kvm_vcpu_check_hw_bp(unsigned long addr
, u32 type
, u32 dr7
,
7226 for (i
= 0; i
< 4; i
++, enable
>>= 2, rwlen
>>= 4)
7227 if ((enable
& 3) && (rwlen
& 15) == type
&& db
[i
] == addr
)
7232 static int kvm_vcpu_do_singlestep(struct kvm_vcpu
*vcpu
)
7234 struct kvm_run
*kvm_run
= vcpu
->run
;
7236 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
) {
7237 kvm_run
->debug
.arch
.dr6
= DR6_BS
| DR6_FIXED_1
| DR6_RTM
;
7238 kvm_run
->debug
.arch
.pc
= kvm_get_linear_rip(vcpu
);
7239 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
7240 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
7243 kvm_queue_exception_p(vcpu
, DB_VECTOR
, DR6_BS
);
7247 int kvm_skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
7249 unsigned long rflags
= kvm_x86_ops
.get_rflags(vcpu
);
7252 r
= kvm_x86_ops
.skip_emulated_instruction(vcpu
);
7257 * rflags is the old, "raw" value of the flags. The new value has
7258 * not been saved yet.
7260 * This is correct even for TF set by the guest, because "the
7261 * processor will not generate this exception after the instruction
7262 * that sets the TF flag".
7264 if (unlikely(rflags
& X86_EFLAGS_TF
))
7265 r
= kvm_vcpu_do_singlestep(vcpu
);
7268 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction
);
7270 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu
*vcpu
, int *r
)
7272 if (unlikely(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) &&
7273 (vcpu
->arch
.guest_debug_dr7
& DR7_BP_EN_MASK
)) {
7274 struct kvm_run
*kvm_run
= vcpu
->run
;
7275 unsigned long eip
= kvm_get_linear_rip(vcpu
);
7276 u32 dr6
= kvm_vcpu_check_hw_bp(eip
, 0,
7277 vcpu
->arch
.guest_debug_dr7
,
7281 kvm_run
->debug
.arch
.dr6
= dr6
| DR6_FIXED_1
| DR6_RTM
;
7282 kvm_run
->debug
.arch
.pc
= eip
;
7283 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
7284 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
7290 if (unlikely(vcpu
->arch
.dr7
& DR7_BP_EN_MASK
) &&
7291 !(kvm_get_rflags(vcpu
) & X86_EFLAGS_RF
)) {
7292 unsigned long eip
= kvm_get_linear_rip(vcpu
);
7293 u32 dr6
= kvm_vcpu_check_hw_bp(eip
, 0,
7298 kvm_queue_exception_p(vcpu
, DB_VECTOR
, dr6
);
7307 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt
*ctxt
)
7309 switch (ctxt
->opcode_len
) {
7316 case 0xe6: /* OUT */
7320 case 0x6c: /* INS */
7322 case 0x6e: /* OUTS */
7329 case 0x33: /* RDPMC */
7338 int x86_emulate_instruction(struct kvm_vcpu
*vcpu
, gpa_t cr2_or_gpa
,
7339 int emulation_type
, void *insn
, int insn_len
)
7342 struct x86_emulate_ctxt
*ctxt
= vcpu
->arch
.emulate_ctxt
;
7343 bool writeback
= true;
7344 bool write_fault_to_spt
;
7346 if (unlikely(!kvm_x86_ops
.can_emulate_instruction(vcpu
, insn
, insn_len
)))
7349 vcpu
->arch
.l1tf_flush_l1d
= true;
7352 * Clear write_fault_to_shadow_pgtable here to ensure it is
7355 write_fault_to_spt
= vcpu
->arch
.write_fault_to_shadow_pgtable
;
7356 vcpu
->arch
.write_fault_to_shadow_pgtable
= false;
7357 kvm_clear_exception_queue(vcpu
);
7359 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
7360 init_emulate_ctxt(vcpu
);
7363 * We will reenter on the same instruction since
7364 * we do not set complete_userspace_io. This does not
7365 * handle watchpoints yet, those would be handled in
7368 if (!(emulation_type
& EMULTYPE_SKIP
) &&
7369 kvm_vcpu_check_breakpoint(vcpu
, &r
))
7372 ctxt
->interruptibility
= 0;
7373 ctxt
->have_exception
= false;
7374 ctxt
->exception
.vector
= -1;
7375 ctxt
->perm_ok
= false;
7377 ctxt
->ud
= emulation_type
& EMULTYPE_TRAP_UD
;
7379 r
= x86_decode_insn(ctxt
, insn
, insn_len
);
7381 trace_kvm_emulate_insn_start(vcpu
);
7382 ++vcpu
->stat
.insn_emulation
;
7383 if (r
!= EMULATION_OK
) {
7384 if ((emulation_type
& EMULTYPE_TRAP_UD
) ||
7385 (emulation_type
& EMULTYPE_TRAP_UD_FORCED
)) {
7386 kvm_queue_exception(vcpu
, UD_VECTOR
);
7389 if (reexecute_instruction(vcpu
, cr2_or_gpa
,
7393 if (ctxt
->have_exception
) {
7395 * #UD should result in just EMULATION_FAILED, and trap-like
7396 * exception should not be encountered during decode.
7398 WARN_ON_ONCE(ctxt
->exception
.vector
== UD_VECTOR
||
7399 exception_type(ctxt
->exception
.vector
) == EXCPT_TRAP
);
7400 inject_emulated_exception(vcpu
);
7403 return handle_emulation_failure(vcpu
, emulation_type
);
7407 if ((emulation_type
& EMULTYPE_VMWARE_GP
) &&
7408 !is_vmware_backdoor_opcode(ctxt
)) {
7409 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
7414 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7415 * for kvm_skip_emulated_instruction(). The caller is responsible for
7416 * updating interruptibility state and injecting single-step #DBs.
7418 if (emulation_type
& EMULTYPE_SKIP
) {
7419 kvm_rip_write(vcpu
, ctxt
->_eip
);
7420 if (ctxt
->eflags
& X86_EFLAGS_RF
)
7421 kvm_set_rflags(vcpu
, ctxt
->eflags
& ~X86_EFLAGS_RF
);
7425 if (retry_instruction(ctxt
, cr2_or_gpa
, emulation_type
))
7428 /* this is needed for vmware backdoor interface to work since it
7429 changes registers values during IO operation */
7430 if (vcpu
->arch
.emulate_regs_need_sync_from_vcpu
) {
7431 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
7432 emulator_invalidate_register_cache(ctxt
);
7436 if (emulation_type
& EMULTYPE_PF
) {
7437 /* Save the faulting GPA (cr2) in the address field */
7438 ctxt
->exception
.address
= cr2_or_gpa
;
7440 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7441 if (vcpu
->arch
.mmu
->direct_map
) {
7442 ctxt
->gpa_available
= true;
7443 ctxt
->gpa_val
= cr2_or_gpa
;
7446 /* Sanitize the address out of an abundance of paranoia. */
7447 ctxt
->exception
.address
= 0;
7450 r
= x86_emulate_insn(ctxt
);
7452 if (r
== EMULATION_INTERCEPTED
)
7455 if (r
== EMULATION_FAILED
) {
7456 if (reexecute_instruction(vcpu
, cr2_or_gpa
, write_fault_to_spt
,
7460 return handle_emulation_failure(vcpu
, emulation_type
);
7463 if (ctxt
->have_exception
) {
7465 if (inject_emulated_exception(vcpu
))
7467 } else if (vcpu
->arch
.pio
.count
) {
7468 if (!vcpu
->arch
.pio
.in
) {
7469 /* FIXME: return into emulator if single-stepping. */
7470 vcpu
->arch
.pio
.count
= 0;
7473 vcpu
->arch
.complete_userspace_io
= complete_emulated_pio
;
7476 } else if (vcpu
->mmio_needed
) {
7477 ++vcpu
->stat
.mmio_exits
;
7479 if (!vcpu
->mmio_is_write
)
7482 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
7483 } else if (r
== EMULATION_RESTART
)
7489 unsigned long rflags
= kvm_x86_ops
.get_rflags(vcpu
);
7490 toggle_interruptibility(vcpu
, ctxt
->interruptibility
);
7491 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
7492 if (!ctxt
->have_exception
||
7493 exception_type(ctxt
->exception
.vector
) == EXCPT_TRAP
) {
7494 kvm_rip_write(vcpu
, ctxt
->eip
);
7495 if (r
&& (ctxt
->tf
|| (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)))
7496 r
= kvm_vcpu_do_singlestep(vcpu
);
7497 if (kvm_x86_ops
.update_emulated_instruction
)
7498 kvm_x86_ops
.update_emulated_instruction(vcpu
);
7499 __kvm_set_rflags(vcpu
, ctxt
->eflags
);
7503 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7504 * do nothing, and it will be requested again as soon as
7505 * the shadow expires. But we still need to check here,
7506 * because POPF has no interrupt shadow.
7508 if (unlikely((ctxt
->eflags
& ~rflags
) & X86_EFLAGS_IF
))
7509 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7511 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= true;
7516 int kvm_emulate_instruction(struct kvm_vcpu
*vcpu
, int emulation_type
)
7518 return x86_emulate_instruction(vcpu
, 0, emulation_type
, NULL
, 0);
7520 EXPORT_SYMBOL_GPL(kvm_emulate_instruction
);
7522 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu
*vcpu
,
7523 void *insn
, int insn_len
)
7525 return x86_emulate_instruction(vcpu
, 0, 0, insn
, insn_len
);
7527 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer
);
7529 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu
*vcpu
)
7531 vcpu
->arch
.pio
.count
= 0;
7535 static int complete_fast_pio_out(struct kvm_vcpu
*vcpu
)
7537 vcpu
->arch
.pio
.count
= 0;
7539 if (unlikely(!kvm_is_linear_rip(vcpu
, vcpu
->arch
.pio
.linear_rip
)))
7542 return kvm_skip_emulated_instruction(vcpu
);
7545 static int kvm_fast_pio_out(struct kvm_vcpu
*vcpu
, int size
,
7546 unsigned short port
)
7548 unsigned long val
= kvm_rax_read(vcpu
);
7549 int ret
= emulator_pio_out(vcpu
, size
, port
, &val
, 1);
7555 * Workaround userspace that relies on old KVM behavior of %rip being
7556 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7559 kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_OUT_7E_INC_RIP
)) {
7560 vcpu
->arch
.complete_userspace_io
=
7561 complete_fast_pio_out_port_0x7e
;
7562 kvm_skip_emulated_instruction(vcpu
);
7564 vcpu
->arch
.pio
.linear_rip
= kvm_get_linear_rip(vcpu
);
7565 vcpu
->arch
.complete_userspace_io
= complete_fast_pio_out
;
7570 static int complete_fast_pio_in(struct kvm_vcpu
*vcpu
)
7574 /* We should only ever be called with arch.pio.count equal to 1 */
7575 BUG_ON(vcpu
->arch
.pio
.count
!= 1);
7577 if (unlikely(!kvm_is_linear_rip(vcpu
, vcpu
->arch
.pio
.linear_rip
))) {
7578 vcpu
->arch
.pio
.count
= 0;
7582 /* For size less than 4 we merge, else we zero extend */
7583 val
= (vcpu
->arch
.pio
.size
< 4) ? kvm_rax_read(vcpu
) : 0;
7586 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7587 * the copy and tracing
7589 emulator_pio_in(vcpu
, vcpu
->arch
.pio
.size
, vcpu
->arch
.pio
.port
, &val
, 1);
7590 kvm_rax_write(vcpu
, val
);
7592 return kvm_skip_emulated_instruction(vcpu
);
7595 static int kvm_fast_pio_in(struct kvm_vcpu
*vcpu
, int size
,
7596 unsigned short port
)
7601 /* For size less than 4 we merge, else we zero extend */
7602 val
= (size
< 4) ? kvm_rax_read(vcpu
) : 0;
7604 ret
= emulator_pio_in(vcpu
, size
, port
, &val
, 1);
7606 kvm_rax_write(vcpu
, val
);
7610 vcpu
->arch
.pio
.linear_rip
= kvm_get_linear_rip(vcpu
);
7611 vcpu
->arch
.complete_userspace_io
= complete_fast_pio_in
;
7616 int kvm_fast_pio(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
, int in
)
7621 ret
= kvm_fast_pio_in(vcpu
, size
, port
);
7623 ret
= kvm_fast_pio_out(vcpu
, size
, port
);
7624 return ret
&& kvm_skip_emulated_instruction(vcpu
);
7626 EXPORT_SYMBOL_GPL(kvm_fast_pio
);
7628 static int kvmclock_cpu_down_prep(unsigned int cpu
)
7630 __this_cpu_write(cpu_tsc_khz
, 0);
7634 static void tsc_khz_changed(void *data
)
7636 struct cpufreq_freqs
*freq
= data
;
7637 unsigned long khz
= 0;
7641 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
7642 khz
= cpufreq_quick_get(raw_smp_processor_id());
7645 __this_cpu_write(cpu_tsc_khz
, khz
);
7648 #ifdef CONFIG_X86_64
7649 static void kvm_hyperv_tsc_notifier(void)
7652 struct kvm_vcpu
*vcpu
;
7655 mutex_lock(&kvm_lock
);
7656 list_for_each_entry(kvm
, &vm_list
, vm_list
)
7657 kvm_make_mclock_inprogress_request(kvm
);
7659 hyperv_stop_tsc_emulation();
7661 /* TSC frequency always matches when on Hyper-V */
7662 for_each_present_cpu(cpu
)
7663 per_cpu(cpu_tsc_khz
, cpu
) = tsc_khz
;
7664 kvm_max_guest_tsc_khz
= tsc_khz
;
7666 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
7667 struct kvm_arch
*ka
= &kvm
->arch
;
7669 spin_lock(&ka
->pvclock_gtod_sync_lock
);
7671 pvclock_update_vm_gtod_copy(kvm
);
7673 kvm_for_each_vcpu(cpu
, vcpu
, kvm
)
7674 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
7676 kvm_for_each_vcpu(cpu
, vcpu
, kvm
)
7677 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS
, vcpu
);
7679 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
7681 mutex_unlock(&kvm_lock
);
7685 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs
*freq
, int cpu
)
7688 struct kvm_vcpu
*vcpu
;
7689 int i
, send_ipi
= 0;
7692 * We allow guests to temporarily run on slowing clocks,
7693 * provided we notify them after, or to run on accelerating
7694 * clocks, provided we notify them before. Thus time never
7697 * However, we have a problem. We can't atomically update
7698 * the frequency of a given CPU from this function; it is
7699 * merely a notifier, which can be called from any CPU.
7700 * Changing the TSC frequency at arbitrary points in time
7701 * requires a recomputation of local variables related to
7702 * the TSC for each VCPU. We must flag these local variables
7703 * to be updated and be sure the update takes place with the
7704 * new frequency before any guests proceed.
7706 * Unfortunately, the combination of hotplug CPU and frequency
7707 * change creates an intractable locking scenario; the order
7708 * of when these callouts happen is undefined with respect to
7709 * CPU hotplug, and they can race with each other. As such,
7710 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7711 * undefined; you can actually have a CPU frequency change take
7712 * place in between the computation of X and the setting of the
7713 * variable. To protect against this problem, all updates of
7714 * the per_cpu tsc_khz variable are done in an interrupt
7715 * protected IPI, and all callers wishing to update the value
7716 * must wait for a synchronous IPI to complete (which is trivial
7717 * if the caller is on the CPU already). This establishes the
7718 * necessary total order on variable updates.
7720 * Note that because a guest time update may take place
7721 * anytime after the setting of the VCPU's request bit, the
7722 * correct TSC value must be set before the request. However,
7723 * to ensure the update actually makes it to any guest which
7724 * starts running in hardware virtualization between the set
7725 * and the acquisition of the spinlock, we must also ping the
7726 * CPU after setting the request bit.
7730 smp_call_function_single(cpu
, tsc_khz_changed
, freq
, 1);
7732 mutex_lock(&kvm_lock
);
7733 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
7734 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
7735 if (vcpu
->cpu
!= cpu
)
7737 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
7738 if (vcpu
->cpu
!= raw_smp_processor_id())
7742 mutex_unlock(&kvm_lock
);
7744 if (freq
->old
< freq
->new && send_ipi
) {
7746 * We upscale the frequency. Must make the guest
7747 * doesn't see old kvmclock values while running with
7748 * the new frequency, otherwise we risk the guest sees
7749 * time go backwards.
7751 * In case we update the frequency for another cpu
7752 * (which might be in guest context) send an interrupt
7753 * to kick the cpu out of guest context. Next time
7754 * guest context is entered kvmclock will be updated,
7755 * so the guest will not see stale values.
7757 smp_call_function_single(cpu
, tsc_khz_changed
, freq
, 1);
7761 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
7764 struct cpufreq_freqs
*freq
= data
;
7767 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
7769 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
7772 for_each_cpu(cpu
, freq
->policy
->cpus
)
7773 __kvmclock_cpufreq_notifier(freq
, cpu
);
7778 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
7779 .notifier_call
= kvmclock_cpufreq_notifier
7782 static int kvmclock_cpu_online(unsigned int cpu
)
7784 tsc_khz_changed(NULL
);
7788 static void kvm_timer_init(void)
7790 max_tsc_khz
= tsc_khz
;
7792 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
7793 #ifdef CONFIG_CPU_FREQ
7794 struct cpufreq_policy
*policy
;
7798 policy
= cpufreq_cpu_get(cpu
);
7800 if (policy
->cpuinfo
.max_freq
)
7801 max_tsc_khz
= policy
->cpuinfo
.max_freq
;
7802 cpufreq_cpu_put(policy
);
7806 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
7807 CPUFREQ_TRANSITION_NOTIFIER
);
7810 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE
, "x86/kvm/clk:online",
7811 kvmclock_cpu_online
, kvmclock_cpu_down_prep
);
7814 DEFINE_PER_CPU(struct kvm_vcpu
*, current_vcpu
);
7815 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu
);
7817 int kvm_is_in_guest(void)
7819 return __this_cpu_read(current_vcpu
) != NULL
;
7822 static int kvm_is_user_mode(void)
7826 if (__this_cpu_read(current_vcpu
))
7827 user_mode
= kvm_x86_ops
.get_cpl(__this_cpu_read(current_vcpu
));
7829 return user_mode
!= 0;
7832 static unsigned long kvm_get_guest_ip(void)
7834 unsigned long ip
= 0;
7836 if (__this_cpu_read(current_vcpu
))
7837 ip
= kvm_rip_read(__this_cpu_read(current_vcpu
));
7842 static void kvm_handle_intel_pt_intr(void)
7844 struct kvm_vcpu
*vcpu
= __this_cpu_read(current_vcpu
);
7846 kvm_make_request(KVM_REQ_PMI
, vcpu
);
7847 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT
,
7848 (unsigned long *)&vcpu
->arch
.pmu
.global_status
);
7851 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
7852 .is_in_guest
= kvm_is_in_guest
,
7853 .is_user_mode
= kvm_is_user_mode
,
7854 .get_guest_ip
= kvm_get_guest_ip
,
7855 .handle_intel_pt_intr
= kvm_handle_intel_pt_intr
,
7858 #ifdef CONFIG_X86_64
7859 static void pvclock_gtod_update_fn(struct work_struct
*work
)
7863 struct kvm_vcpu
*vcpu
;
7866 mutex_lock(&kvm_lock
);
7867 list_for_each_entry(kvm
, &vm_list
, vm_list
)
7868 kvm_for_each_vcpu(i
, vcpu
, kvm
)
7869 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
7870 atomic_set(&kvm_guest_has_master_clock
, 0);
7871 mutex_unlock(&kvm_lock
);
7874 static DECLARE_WORK(pvclock_gtod_work
, pvclock_gtod_update_fn
);
7877 * Notification about pvclock gtod data update.
7879 static int pvclock_gtod_notify(struct notifier_block
*nb
, unsigned long unused
,
7882 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
7883 struct timekeeper
*tk
= priv
;
7885 update_pvclock_gtod(tk
);
7887 /* disable master clock if host does not trust, or does not
7888 * use, TSC based clocksource.
7890 if (!gtod_is_based_on_tsc(gtod
->clock
.vclock_mode
) &&
7891 atomic_read(&kvm_guest_has_master_clock
) != 0)
7892 queue_work(system_long_wq
, &pvclock_gtod_work
);
7897 static struct notifier_block pvclock_gtod_notifier
= {
7898 .notifier_call
= pvclock_gtod_notify
,
7902 int kvm_arch_init(void *opaque
)
7904 struct kvm_x86_init_ops
*ops
= opaque
;
7907 if (kvm_x86_ops
.hardware_enable
) {
7908 printk(KERN_ERR
"kvm: already loaded the other module\n");
7913 if (!ops
->cpu_has_kvm_support()) {
7914 pr_err_ratelimited("kvm: no hardware support\n");
7918 if (ops
->disabled_by_bios()) {
7919 pr_warn_ratelimited("kvm: disabled by bios\n");
7925 * KVM explicitly assumes that the guest has an FPU and
7926 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7927 * vCPU's FPU state as a fxregs_state struct.
7929 if (!boot_cpu_has(X86_FEATURE_FPU
) || !boot_cpu_has(X86_FEATURE_FXSR
)) {
7930 printk(KERN_ERR
"kvm: inadequate fpu\n");
7936 x86_fpu_cache
= kmem_cache_create("x86_fpu", sizeof(struct fpu
),
7937 __alignof__(struct fpu
), SLAB_ACCOUNT
,
7939 if (!x86_fpu_cache
) {
7940 printk(KERN_ERR
"kvm: failed to allocate cache for x86 fpu\n");
7944 x86_emulator_cache
= kvm_alloc_emulator_cache();
7945 if (!x86_emulator_cache
) {
7946 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7947 goto out_free_x86_fpu_cache
;
7950 user_return_msrs
= alloc_percpu(struct kvm_user_return_msrs
);
7951 if (!user_return_msrs
) {
7952 printk(KERN_ERR
"kvm: failed to allocate percpu kvm_user_return_msrs\n");
7953 goto out_free_x86_emulator_cache
;
7956 r
= kvm_mmu_module_init();
7958 goto out_free_percpu
;
7960 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
7961 PT_DIRTY_MASK
, PT64_NX_MASK
, 0,
7962 PT_PRESENT_MASK
, 0, sme_me_mask
);
7965 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
7967 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
7968 host_xcr0
= xgetbv(XCR_XFEATURE_ENABLED_MASK
);
7969 supported_xcr0
= host_xcr0
& KVM_SUPPORTED_XCR0
;
7973 if (pi_inject_timer
== -1)
7974 pi_inject_timer
= housekeeping_enabled(HK_FLAG_TIMER
);
7975 #ifdef CONFIG_X86_64
7976 pvclock_gtod_register_notifier(&pvclock_gtod_notifier
);
7978 if (hypervisor_is_type(X86_HYPER_MS_HYPERV
))
7979 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier
);
7985 free_percpu(user_return_msrs
);
7986 out_free_x86_emulator_cache
:
7987 kmem_cache_destroy(x86_emulator_cache
);
7988 out_free_x86_fpu_cache
:
7989 kmem_cache_destroy(x86_fpu_cache
);
7994 void kvm_arch_exit(void)
7996 #ifdef CONFIG_X86_64
7997 if (hypervisor_is_type(X86_HYPER_MS_HYPERV
))
7998 clear_hv_tscchange_cb();
8001 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
8003 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
8004 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
8005 CPUFREQ_TRANSITION_NOTIFIER
);
8006 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE
);
8007 #ifdef CONFIG_X86_64
8008 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier
);
8010 kvm_x86_ops
.hardware_enable
= NULL
;
8011 kvm_mmu_module_exit();
8012 free_percpu(user_return_msrs
);
8013 kmem_cache_destroy(x86_fpu_cache
);
8016 static int __kvm_vcpu_halt(struct kvm_vcpu
*vcpu
, int state
, int reason
)
8018 ++vcpu
->stat
.halt_exits
;
8019 if (lapic_in_kernel(vcpu
)) {
8020 vcpu
->arch
.mp_state
= state
;
8023 vcpu
->run
->exit_reason
= reason
;
8028 int kvm_vcpu_halt(struct kvm_vcpu
*vcpu
)
8030 return __kvm_vcpu_halt(vcpu
, KVM_MP_STATE_HALTED
, KVM_EXIT_HLT
);
8032 EXPORT_SYMBOL_GPL(kvm_vcpu_halt
);
8034 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
8036 int ret
= kvm_skip_emulated_instruction(vcpu
);
8038 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8039 * KVM_EXIT_DEBUG here.
8041 return kvm_vcpu_halt(vcpu
) && ret
;
8043 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
8045 int kvm_emulate_ap_reset_hold(struct kvm_vcpu
*vcpu
)
8047 int ret
= kvm_skip_emulated_instruction(vcpu
);
8049 return __kvm_vcpu_halt(vcpu
, KVM_MP_STATE_AP_RESET_HOLD
, KVM_EXIT_AP_RESET_HOLD
) && ret
;
8051 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold
);
8053 #ifdef CONFIG_X86_64
8054 static int kvm_pv_clock_pairing(struct kvm_vcpu
*vcpu
, gpa_t paddr
,
8055 unsigned long clock_type
)
8057 struct kvm_clock_pairing clock_pairing
;
8058 struct timespec64 ts
;
8062 if (clock_type
!= KVM_CLOCK_PAIRING_WALLCLOCK
)
8063 return -KVM_EOPNOTSUPP
;
8065 if (kvm_get_walltime_and_clockread(&ts
, &cycle
) == false)
8066 return -KVM_EOPNOTSUPP
;
8068 clock_pairing
.sec
= ts
.tv_sec
;
8069 clock_pairing
.nsec
= ts
.tv_nsec
;
8070 clock_pairing
.tsc
= kvm_read_l1_tsc(vcpu
, cycle
);
8071 clock_pairing
.flags
= 0;
8072 memset(&clock_pairing
.pad
, 0, sizeof(clock_pairing
.pad
));
8075 if (kvm_write_guest(vcpu
->kvm
, paddr
, &clock_pairing
,
8076 sizeof(struct kvm_clock_pairing
)))
8084 * kvm_pv_kick_cpu_op: Kick a vcpu.
8086 * @apicid - apicid of vcpu to be kicked.
8088 static void kvm_pv_kick_cpu_op(struct kvm
*kvm
, unsigned long flags
, int apicid
)
8090 struct kvm_lapic_irq lapic_irq
;
8092 lapic_irq
.shorthand
= APIC_DEST_NOSHORT
;
8093 lapic_irq
.dest_mode
= APIC_DEST_PHYSICAL
;
8094 lapic_irq
.level
= 0;
8095 lapic_irq
.dest_id
= apicid
;
8096 lapic_irq
.msi_redir_hint
= false;
8098 lapic_irq
.delivery_mode
= APIC_DM_REMRD
;
8099 kvm_irq_delivery_to_apic(kvm
, NULL
, &lapic_irq
, NULL
);
8102 bool kvm_apicv_activated(struct kvm
*kvm
)
8104 return (READ_ONCE(kvm
->arch
.apicv_inhibit_reasons
) == 0);
8106 EXPORT_SYMBOL_GPL(kvm_apicv_activated
);
8108 void kvm_apicv_init(struct kvm
*kvm
, bool enable
)
8111 clear_bit(APICV_INHIBIT_REASON_DISABLE
,
8112 &kvm
->arch
.apicv_inhibit_reasons
);
8114 set_bit(APICV_INHIBIT_REASON_DISABLE
,
8115 &kvm
->arch
.apicv_inhibit_reasons
);
8117 EXPORT_SYMBOL_GPL(kvm_apicv_init
);
8119 static void kvm_sched_yield(struct kvm
*kvm
, unsigned long dest_id
)
8121 struct kvm_vcpu
*target
= NULL
;
8122 struct kvm_apic_map
*map
;
8125 map
= rcu_dereference(kvm
->arch
.apic_map
);
8127 if (likely(map
) && dest_id
<= map
->max_apic_id
&& map
->phys_map
[dest_id
])
8128 target
= map
->phys_map
[dest_id
]->vcpu
;
8132 if (target
&& READ_ONCE(target
->ready
))
8133 kvm_vcpu_yield_to(target
);
8136 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
8138 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
8141 if (kvm_hv_hypercall_enabled(vcpu
->kvm
))
8142 return kvm_hv_hypercall(vcpu
);
8144 nr
= kvm_rax_read(vcpu
);
8145 a0
= kvm_rbx_read(vcpu
);
8146 a1
= kvm_rcx_read(vcpu
);
8147 a2
= kvm_rdx_read(vcpu
);
8148 a3
= kvm_rsi_read(vcpu
);
8150 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
8152 op_64_bit
= is_64_bit_mode(vcpu
);
8161 if (kvm_x86_ops
.get_cpl(vcpu
) != 0) {
8169 case KVM_HC_VAPIC_POLL_IRQ
:
8172 case KVM_HC_KICK_CPU
:
8173 if (!guest_pv_has(vcpu
, KVM_FEATURE_PV_UNHALT
))
8176 kvm_pv_kick_cpu_op(vcpu
->kvm
, a0
, a1
);
8177 kvm_sched_yield(vcpu
->kvm
, a1
);
8180 #ifdef CONFIG_X86_64
8181 case KVM_HC_CLOCK_PAIRING
:
8182 ret
= kvm_pv_clock_pairing(vcpu
, a0
, a1
);
8185 case KVM_HC_SEND_IPI
:
8186 if (!guest_pv_has(vcpu
, KVM_FEATURE_PV_SEND_IPI
))
8189 ret
= kvm_pv_send_ipi(vcpu
->kvm
, a0
, a1
, a2
, a3
, op_64_bit
);
8191 case KVM_HC_SCHED_YIELD
:
8192 if (!guest_pv_has(vcpu
, KVM_FEATURE_PV_SCHED_YIELD
))
8195 kvm_sched_yield(vcpu
->kvm
, a0
);
8205 kvm_rax_write(vcpu
, ret
);
8207 ++vcpu
->stat
.hypercalls
;
8208 return kvm_skip_emulated_instruction(vcpu
);
8210 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
8212 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
)
8214 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
8215 char instruction
[3];
8216 unsigned long rip
= kvm_rip_read(vcpu
);
8218 kvm_x86_ops
.patch_hypercall(vcpu
, instruction
);
8220 return emulator_write_emulated(ctxt
, rip
, instruction
, 3,
8224 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
8226 return vcpu
->run
->request_interrupt_window
&&
8227 likely(!pic_in_kernel(vcpu
->kvm
));
8230 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
8232 struct kvm_run
*kvm_run
= vcpu
->run
;
8235 * if_flag is obsolete and useless, so do not bother
8236 * setting it for SEV-ES guests. Userspace can just
8237 * use kvm_run->ready_for_interrupt_injection.
8239 kvm_run
->if_flag
= !vcpu
->arch
.guest_state_protected
8240 && (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
8242 kvm_run
->flags
= is_smm(vcpu
) ? KVM_RUN_X86_SMM
: 0;
8243 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
8244 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
8245 kvm_run
->ready_for_interrupt_injection
=
8246 pic_in_kernel(vcpu
->kvm
) ||
8247 kvm_vcpu_ready_for_interrupt_injection(vcpu
);
8250 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
8254 if (!kvm_x86_ops
.update_cr8_intercept
)
8257 if (!lapic_in_kernel(vcpu
))
8260 if (vcpu
->arch
.apicv_active
)
8263 if (!vcpu
->arch
.apic
->vapic_addr
)
8264 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
8271 tpr
= kvm_lapic_get_cr8(vcpu
);
8273 kvm_x86_ops
.update_cr8_intercept(vcpu
, tpr
, max_irr
);
8276 static void inject_pending_event(struct kvm_vcpu
*vcpu
, bool *req_immediate_exit
)
8279 bool can_inject
= true;
8281 /* try to reinject previous events if any */
8283 if (vcpu
->arch
.exception
.injected
) {
8284 kvm_x86_ops
.queue_exception(vcpu
);
8288 * Do not inject an NMI or interrupt if there is a pending
8289 * exception. Exceptions and interrupts are recognized at
8290 * instruction boundaries, i.e. the start of an instruction.
8291 * Trap-like exceptions, e.g. #DB, have higher priority than
8292 * NMIs and interrupts, i.e. traps are recognized before an
8293 * NMI/interrupt that's pending on the same instruction.
8294 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8295 * priority, but are only generated (pended) during instruction
8296 * execution, i.e. a pending fault-like exception means the
8297 * fault occurred on the *previous* instruction and must be
8298 * serviced prior to recognizing any new events in order to
8299 * fully complete the previous instruction.
8301 else if (!vcpu
->arch
.exception
.pending
) {
8302 if (vcpu
->arch
.nmi_injected
) {
8303 kvm_x86_ops
.set_nmi(vcpu
);
8305 } else if (vcpu
->arch
.interrupt
.injected
) {
8306 kvm_x86_ops
.set_irq(vcpu
);
8311 WARN_ON_ONCE(vcpu
->arch
.exception
.injected
&&
8312 vcpu
->arch
.exception
.pending
);
8315 * Call check_nested_events() even if we reinjected a previous event
8316 * in order for caller to determine if it should require immediate-exit
8317 * from L2 to L1 due to pending L1 events which require exit
8320 if (is_guest_mode(vcpu
)) {
8321 r
= kvm_x86_ops
.nested_ops
->check_events(vcpu
);
8326 /* try to inject new event if pending */
8327 if (vcpu
->arch
.exception
.pending
) {
8328 trace_kvm_inj_exception(vcpu
->arch
.exception
.nr
,
8329 vcpu
->arch
.exception
.has_error_code
,
8330 vcpu
->arch
.exception
.error_code
);
8332 vcpu
->arch
.exception
.pending
= false;
8333 vcpu
->arch
.exception
.injected
= true;
8335 if (exception_type(vcpu
->arch
.exception
.nr
) == EXCPT_FAULT
)
8336 __kvm_set_rflags(vcpu
, kvm_get_rflags(vcpu
) |
8339 if (vcpu
->arch
.exception
.nr
== DB_VECTOR
) {
8340 kvm_deliver_exception_payload(vcpu
);
8341 if (vcpu
->arch
.dr7
& DR7_GD
) {
8342 vcpu
->arch
.dr7
&= ~DR7_GD
;
8343 kvm_update_dr7(vcpu
);
8347 kvm_x86_ops
.queue_exception(vcpu
);
8352 * Finally, inject interrupt events. If an event cannot be injected
8353 * due to architectural conditions (e.g. IF=0) a window-open exit
8354 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8355 * and can architecturally be injected, but we cannot do it right now:
8356 * an interrupt could have arrived just now and we have to inject it
8357 * as a vmexit, or there could already an event in the queue, which is
8358 * indicated by can_inject. In that case we request an immediate exit
8359 * in order to make progress and get back here for another iteration.
8360 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8362 if (vcpu
->arch
.smi_pending
) {
8363 r
= can_inject
? kvm_x86_ops
.smi_allowed(vcpu
, true) : -EBUSY
;
8367 vcpu
->arch
.smi_pending
= false;
8368 ++vcpu
->arch
.smi_count
;
8372 kvm_x86_ops
.enable_smi_window(vcpu
);
8375 if (vcpu
->arch
.nmi_pending
) {
8376 r
= can_inject
? kvm_x86_ops
.nmi_allowed(vcpu
, true) : -EBUSY
;
8380 --vcpu
->arch
.nmi_pending
;
8381 vcpu
->arch
.nmi_injected
= true;
8382 kvm_x86_ops
.set_nmi(vcpu
);
8384 WARN_ON(kvm_x86_ops
.nmi_allowed(vcpu
, true) < 0);
8386 if (vcpu
->arch
.nmi_pending
)
8387 kvm_x86_ops
.enable_nmi_window(vcpu
);
8390 if (kvm_cpu_has_injectable_intr(vcpu
)) {
8391 r
= can_inject
? kvm_x86_ops
.interrupt_allowed(vcpu
, true) : -EBUSY
;
8395 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
), false);
8396 kvm_x86_ops
.set_irq(vcpu
);
8397 WARN_ON(kvm_x86_ops
.interrupt_allowed(vcpu
, true) < 0);
8399 if (kvm_cpu_has_injectable_intr(vcpu
))
8400 kvm_x86_ops
.enable_irq_window(vcpu
);
8403 if (is_guest_mode(vcpu
) &&
8404 kvm_x86_ops
.nested_ops
->hv_timer_pending
&&
8405 kvm_x86_ops
.nested_ops
->hv_timer_pending(vcpu
))
8406 *req_immediate_exit
= true;
8408 WARN_ON(vcpu
->arch
.exception
.pending
);
8412 *req_immediate_exit
= true;
8416 static void process_nmi(struct kvm_vcpu
*vcpu
)
8421 * x86 is limited to one NMI running, and one NMI pending after it.
8422 * If an NMI is already in progress, limit further NMIs to just one.
8423 * Otherwise, allow two (and we'll inject the first one immediately).
8425 if (kvm_x86_ops
.get_nmi_mask(vcpu
) || vcpu
->arch
.nmi_injected
)
8428 vcpu
->arch
.nmi_pending
+= atomic_xchg(&vcpu
->arch
.nmi_queued
, 0);
8429 vcpu
->arch
.nmi_pending
= min(vcpu
->arch
.nmi_pending
, limit
);
8430 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
8433 static u32
enter_smm_get_segment_flags(struct kvm_segment
*seg
)
8436 flags
|= seg
->g
<< 23;
8437 flags
|= seg
->db
<< 22;
8438 flags
|= seg
->l
<< 21;
8439 flags
|= seg
->avl
<< 20;
8440 flags
|= seg
->present
<< 15;
8441 flags
|= seg
->dpl
<< 13;
8442 flags
|= seg
->s
<< 12;
8443 flags
|= seg
->type
<< 8;
8447 static void enter_smm_save_seg_32(struct kvm_vcpu
*vcpu
, char *buf
, int n
)
8449 struct kvm_segment seg
;
8452 kvm_get_segment(vcpu
, &seg
, n
);
8453 put_smstate(u32
, buf
, 0x7fa8 + n
* 4, seg
.selector
);
8456 offset
= 0x7f84 + n
* 12;
8458 offset
= 0x7f2c + (n
- 3) * 12;
8460 put_smstate(u32
, buf
, offset
+ 8, seg
.base
);
8461 put_smstate(u32
, buf
, offset
+ 4, seg
.limit
);
8462 put_smstate(u32
, buf
, offset
, enter_smm_get_segment_flags(&seg
));
8465 #ifdef CONFIG_X86_64
8466 static void enter_smm_save_seg_64(struct kvm_vcpu
*vcpu
, char *buf
, int n
)
8468 struct kvm_segment seg
;
8472 kvm_get_segment(vcpu
, &seg
, n
);
8473 offset
= 0x7e00 + n
* 16;
8475 flags
= enter_smm_get_segment_flags(&seg
) >> 8;
8476 put_smstate(u16
, buf
, offset
, seg
.selector
);
8477 put_smstate(u16
, buf
, offset
+ 2, flags
);
8478 put_smstate(u32
, buf
, offset
+ 4, seg
.limit
);
8479 put_smstate(u64
, buf
, offset
+ 8, seg
.base
);
8483 static void enter_smm_save_state_32(struct kvm_vcpu
*vcpu
, char *buf
)
8486 struct kvm_segment seg
;
8490 put_smstate(u32
, buf
, 0x7ffc, kvm_read_cr0(vcpu
));
8491 put_smstate(u32
, buf
, 0x7ff8, kvm_read_cr3(vcpu
));
8492 put_smstate(u32
, buf
, 0x7ff4, kvm_get_rflags(vcpu
));
8493 put_smstate(u32
, buf
, 0x7ff0, kvm_rip_read(vcpu
));
8495 for (i
= 0; i
< 8; i
++)
8496 put_smstate(u32
, buf
, 0x7fd0 + i
* 4, kvm_register_read(vcpu
, i
));
8498 kvm_get_dr(vcpu
, 6, &val
);
8499 put_smstate(u32
, buf
, 0x7fcc, (u32
)val
);
8500 kvm_get_dr(vcpu
, 7, &val
);
8501 put_smstate(u32
, buf
, 0x7fc8, (u32
)val
);
8503 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_TR
);
8504 put_smstate(u32
, buf
, 0x7fc4, seg
.selector
);
8505 put_smstate(u32
, buf
, 0x7f64, seg
.base
);
8506 put_smstate(u32
, buf
, 0x7f60, seg
.limit
);
8507 put_smstate(u32
, buf
, 0x7f5c, enter_smm_get_segment_flags(&seg
));
8509 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_LDTR
);
8510 put_smstate(u32
, buf
, 0x7fc0, seg
.selector
);
8511 put_smstate(u32
, buf
, 0x7f80, seg
.base
);
8512 put_smstate(u32
, buf
, 0x7f7c, seg
.limit
);
8513 put_smstate(u32
, buf
, 0x7f78, enter_smm_get_segment_flags(&seg
));
8515 kvm_x86_ops
.get_gdt(vcpu
, &dt
);
8516 put_smstate(u32
, buf
, 0x7f74, dt
.address
);
8517 put_smstate(u32
, buf
, 0x7f70, dt
.size
);
8519 kvm_x86_ops
.get_idt(vcpu
, &dt
);
8520 put_smstate(u32
, buf
, 0x7f58, dt
.address
);
8521 put_smstate(u32
, buf
, 0x7f54, dt
.size
);
8523 for (i
= 0; i
< 6; i
++)
8524 enter_smm_save_seg_32(vcpu
, buf
, i
);
8526 put_smstate(u32
, buf
, 0x7f14, kvm_read_cr4(vcpu
));
8529 put_smstate(u32
, buf
, 0x7efc, 0x00020000);
8530 put_smstate(u32
, buf
, 0x7ef8, vcpu
->arch
.smbase
);
8533 #ifdef CONFIG_X86_64
8534 static void enter_smm_save_state_64(struct kvm_vcpu
*vcpu
, char *buf
)
8537 struct kvm_segment seg
;
8541 for (i
= 0; i
< 16; i
++)
8542 put_smstate(u64
, buf
, 0x7ff8 - i
* 8, kvm_register_read(vcpu
, i
));
8544 put_smstate(u64
, buf
, 0x7f78, kvm_rip_read(vcpu
));
8545 put_smstate(u32
, buf
, 0x7f70, kvm_get_rflags(vcpu
));
8547 kvm_get_dr(vcpu
, 6, &val
);
8548 put_smstate(u64
, buf
, 0x7f68, val
);
8549 kvm_get_dr(vcpu
, 7, &val
);
8550 put_smstate(u64
, buf
, 0x7f60, val
);
8552 put_smstate(u64
, buf
, 0x7f58, kvm_read_cr0(vcpu
));
8553 put_smstate(u64
, buf
, 0x7f50, kvm_read_cr3(vcpu
));
8554 put_smstate(u64
, buf
, 0x7f48, kvm_read_cr4(vcpu
));
8556 put_smstate(u32
, buf
, 0x7f00, vcpu
->arch
.smbase
);
8559 put_smstate(u32
, buf
, 0x7efc, 0x00020064);
8561 put_smstate(u64
, buf
, 0x7ed0, vcpu
->arch
.efer
);
8563 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_TR
);
8564 put_smstate(u16
, buf
, 0x7e90, seg
.selector
);
8565 put_smstate(u16
, buf
, 0x7e92, enter_smm_get_segment_flags(&seg
) >> 8);
8566 put_smstate(u32
, buf
, 0x7e94, seg
.limit
);
8567 put_smstate(u64
, buf
, 0x7e98, seg
.base
);
8569 kvm_x86_ops
.get_idt(vcpu
, &dt
);
8570 put_smstate(u32
, buf
, 0x7e84, dt
.size
);
8571 put_smstate(u64
, buf
, 0x7e88, dt
.address
);
8573 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_LDTR
);
8574 put_smstate(u16
, buf
, 0x7e70, seg
.selector
);
8575 put_smstate(u16
, buf
, 0x7e72, enter_smm_get_segment_flags(&seg
) >> 8);
8576 put_smstate(u32
, buf
, 0x7e74, seg
.limit
);
8577 put_smstate(u64
, buf
, 0x7e78, seg
.base
);
8579 kvm_x86_ops
.get_gdt(vcpu
, &dt
);
8580 put_smstate(u32
, buf
, 0x7e64, dt
.size
);
8581 put_smstate(u64
, buf
, 0x7e68, dt
.address
);
8583 for (i
= 0; i
< 6; i
++)
8584 enter_smm_save_seg_64(vcpu
, buf
, i
);
8588 static void enter_smm(struct kvm_vcpu
*vcpu
)
8590 struct kvm_segment cs
, ds
;
8595 trace_kvm_enter_smm(vcpu
->vcpu_id
, vcpu
->arch
.smbase
, true);
8596 memset(buf
, 0, 512);
8597 #ifdef CONFIG_X86_64
8598 if (guest_cpuid_has(vcpu
, X86_FEATURE_LM
))
8599 enter_smm_save_state_64(vcpu
, buf
);
8602 enter_smm_save_state_32(vcpu
, buf
);
8605 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8606 * vCPU state (e.g. leave guest mode) after we've saved the state into
8607 * the SMM state-save area.
8609 kvm_x86_ops
.pre_enter_smm(vcpu
, buf
);
8611 vcpu
->arch
.hflags
|= HF_SMM_MASK
;
8612 kvm_vcpu_write_guest(vcpu
, vcpu
->arch
.smbase
+ 0xfe00, buf
, sizeof(buf
));
8614 if (kvm_x86_ops
.get_nmi_mask(vcpu
))
8615 vcpu
->arch
.hflags
|= HF_SMM_INSIDE_NMI_MASK
;
8617 kvm_x86_ops
.set_nmi_mask(vcpu
, true);
8619 kvm_set_rflags(vcpu
, X86_EFLAGS_FIXED
);
8620 kvm_rip_write(vcpu
, 0x8000);
8622 cr0
= vcpu
->arch
.cr0
& ~(X86_CR0_PE
| X86_CR0_EM
| X86_CR0_TS
| X86_CR0_PG
);
8623 kvm_x86_ops
.set_cr0(vcpu
, cr0
);
8624 vcpu
->arch
.cr0
= cr0
;
8626 kvm_x86_ops
.set_cr4(vcpu
, 0);
8628 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8629 dt
.address
= dt
.size
= 0;
8630 kvm_x86_ops
.set_idt(vcpu
, &dt
);
8632 __kvm_set_dr(vcpu
, 7, DR7_FIXED_1
);
8634 cs
.selector
= (vcpu
->arch
.smbase
>> 4) & 0xffff;
8635 cs
.base
= vcpu
->arch
.smbase
;
8640 cs
.limit
= ds
.limit
= 0xffffffff;
8641 cs
.type
= ds
.type
= 0x3;
8642 cs
.dpl
= ds
.dpl
= 0;
8647 cs
.avl
= ds
.avl
= 0;
8648 cs
.present
= ds
.present
= 1;
8649 cs
.unusable
= ds
.unusable
= 0;
8650 cs
.padding
= ds
.padding
= 0;
8652 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
8653 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_DS
);
8654 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_ES
);
8655 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_FS
);
8656 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_GS
);
8657 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_SS
);
8659 #ifdef CONFIG_X86_64
8660 if (guest_cpuid_has(vcpu
, X86_FEATURE_LM
))
8661 kvm_x86_ops
.set_efer(vcpu
, 0);
8664 kvm_update_cpuid_runtime(vcpu
);
8665 kvm_mmu_reset_context(vcpu
);
8668 static void process_smi(struct kvm_vcpu
*vcpu
)
8670 vcpu
->arch
.smi_pending
= true;
8671 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
8674 void kvm_make_scan_ioapic_request_mask(struct kvm
*kvm
,
8675 unsigned long *vcpu_bitmap
)
8679 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
8681 kvm_make_vcpus_request_mask(kvm
, KVM_REQ_SCAN_IOAPIC
,
8682 NULL
, vcpu_bitmap
, cpus
);
8684 free_cpumask_var(cpus
);
8687 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
8689 kvm_make_all_cpus_request(kvm
, KVM_REQ_SCAN_IOAPIC
);
8692 void kvm_vcpu_update_apicv(struct kvm_vcpu
*vcpu
)
8694 if (!lapic_in_kernel(vcpu
))
8697 vcpu
->arch
.apicv_active
= kvm_apicv_activated(vcpu
->kvm
);
8698 kvm_apic_update_apicv(vcpu
);
8699 kvm_x86_ops
.refresh_apicv_exec_ctrl(vcpu
);
8701 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv
);
8704 * NOTE: Do not hold any lock prior to calling this.
8706 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8707 * locked, because it calls __x86_set_memory_region() which does
8708 * synchronize_srcu(&kvm->srcu).
8710 void kvm_request_apicv_update(struct kvm
*kvm
, bool activate
, ulong bit
)
8712 struct kvm_vcpu
*except
;
8713 unsigned long old
, new, expected
;
8715 if (!kvm_x86_ops
.check_apicv_inhibit_reasons
||
8716 !kvm_x86_ops
.check_apicv_inhibit_reasons(bit
))
8719 old
= READ_ONCE(kvm
->arch
.apicv_inhibit_reasons
);
8721 expected
= new = old
;
8723 __clear_bit(bit
, &new);
8725 __set_bit(bit
, &new);
8728 old
= cmpxchg(&kvm
->arch
.apicv_inhibit_reasons
, expected
, new);
8729 } while (old
!= expected
);
8734 trace_kvm_apicv_update_request(activate
, bit
);
8735 if (kvm_x86_ops
.pre_update_apicv_exec_ctrl
)
8736 kvm_x86_ops
.pre_update_apicv_exec_ctrl(kvm
, activate
);
8739 * Sending request to update APICV for all other vcpus,
8740 * while update the calling vcpu immediately instead of
8741 * waiting for another #VMEXIT to handle the request.
8743 except
= kvm_get_running_vcpu();
8744 kvm_make_all_cpus_request_except(kvm
, KVM_REQ_APICV_UPDATE
,
8747 kvm_vcpu_update_apicv(except
);
8749 EXPORT_SYMBOL_GPL(kvm_request_apicv_update
);
8751 static void vcpu_scan_ioapic(struct kvm_vcpu
*vcpu
)
8753 if (!kvm_apic_present(vcpu
))
8756 bitmap_zero(vcpu
->arch
.ioapic_handled_vectors
, 256);
8758 if (irqchip_split(vcpu
->kvm
))
8759 kvm_scan_ioapic_routes(vcpu
, vcpu
->arch
.ioapic_handled_vectors
);
8761 if (vcpu
->arch
.apicv_active
)
8762 kvm_x86_ops
.sync_pir_to_irr(vcpu
);
8763 if (ioapic_in_kernel(vcpu
->kvm
))
8764 kvm_ioapic_scan_entry(vcpu
, vcpu
->arch
.ioapic_handled_vectors
);
8767 if (is_guest_mode(vcpu
))
8768 vcpu
->arch
.load_eoi_exitmap_pending
= true;
8770 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP
, vcpu
);
8773 static void vcpu_load_eoi_exitmap(struct kvm_vcpu
*vcpu
)
8775 u64 eoi_exit_bitmap
[4];
8777 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
8780 bitmap_or((ulong
*)eoi_exit_bitmap
, vcpu
->arch
.ioapic_handled_vectors
,
8781 vcpu_to_synic(vcpu
)->vec_bitmap
, 256);
8782 kvm_x86_ops
.load_eoi_exitmap(vcpu
, eoi_exit_bitmap
);
8785 void kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
8786 unsigned long start
, unsigned long end
)
8788 unsigned long apic_address
;
8791 * The physical address of apic access page is stored in the VMCS.
8792 * Update it when it becomes invalid.
8794 apic_address
= gfn_to_hva(kvm
, APIC_DEFAULT_PHYS_BASE
>> PAGE_SHIFT
);
8795 if (start
<= apic_address
&& apic_address
< end
)
8796 kvm_make_all_cpus_request(kvm
, KVM_REQ_APIC_PAGE_RELOAD
);
8799 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu
*vcpu
)
8801 if (!lapic_in_kernel(vcpu
))
8804 if (!kvm_x86_ops
.set_apic_access_page_addr
)
8807 kvm_x86_ops
.set_apic_access_page_addr(vcpu
);
8810 void __kvm_request_immediate_exit(struct kvm_vcpu
*vcpu
)
8812 smp_send_reschedule(vcpu
->cpu
);
8814 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit
);
8817 * Returns 1 to let vcpu_run() continue the guest execution loop without
8818 * exiting to the userspace. Otherwise, the value will be returned to the
8821 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
8825 dm_request_for_irq_injection(vcpu
) &&
8826 kvm_cpu_accept_dm_intr(vcpu
);
8827 fastpath_t exit_fastpath
;
8829 bool req_immediate_exit
= false;
8831 /* Forbid vmenter if vcpu dirty ring is soft-full */
8832 if (unlikely(vcpu
->kvm
->dirty_ring_size
&&
8833 kvm_dirty_ring_soft_full(&vcpu
->dirty_ring
))) {
8834 vcpu
->run
->exit_reason
= KVM_EXIT_DIRTY_RING_FULL
;
8835 trace_kvm_dirty_ring_exit(vcpu
);
8840 if (kvm_request_pending(vcpu
)) {
8841 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES
, vcpu
)) {
8842 if (unlikely(!kvm_x86_ops
.nested_ops
->get_nested_state_pages(vcpu
))) {
8847 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
))
8848 kvm_mmu_unload(vcpu
);
8849 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER
, vcpu
))
8850 __kvm_migrate_timers(vcpu
);
8851 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
))
8852 kvm_gen_update_masterclock(vcpu
->kvm
);
8853 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
))
8854 kvm_gen_kvmclock_update(vcpu
);
8855 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE
, vcpu
)) {
8856 r
= kvm_guest_time_update(vcpu
);
8860 if (kvm_check_request(KVM_REQ_MMU_SYNC
, vcpu
))
8861 kvm_mmu_sync_roots(vcpu
);
8862 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD
, vcpu
))
8863 kvm_mmu_load_pgd(vcpu
);
8864 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
)) {
8865 kvm_vcpu_flush_tlb_all(vcpu
);
8867 /* Flushing all ASIDs flushes the current ASID... */
8868 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT
, vcpu
);
8870 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT
, vcpu
))
8871 kvm_vcpu_flush_tlb_current(vcpu
);
8872 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH
, vcpu
))
8873 kvm_vcpu_flush_tlb_guest(vcpu
);
8875 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
)) {
8876 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
8880 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT
, vcpu
)) {
8881 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
8882 vcpu
->mmio_needed
= 0;
8886 if (kvm_check_request(KVM_REQ_APF_HALT
, vcpu
)) {
8887 /* Page is swapped out. Do synthetic halt */
8888 vcpu
->arch
.apf
.halted
= true;
8892 if (kvm_check_request(KVM_REQ_STEAL_UPDATE
, vcpu
))
8893 record_steal_time(vcpu
);
8894 if (kvm_check_request(KVM_REQ_SMI
, vcpu
))
8896 if (kvm_check_request(KVM_REQ_NMI
, vcpu
))
8898 if (kvm_check_request(KVM_REQ_PMU
, vcpu
))
8899 kvm_pmu_handle_event(vcpu
);
8900 if (kvm_check_request(KVM_REQ_PMI
, vcpu
))
8901 kvm_pmu_deliver_pmi(vcpu
);
8902 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT
, vcpu
)) {
8903 BUG_ON(vcpu
->arch
.pending_ioapic_eoi
> 255);
8904 if (test_bit(vcpu
->arch
.pending_ioapic_eoi
,
8905 vcpu
->arch
.ioapic_handled_vectors
)) {
8906 vcpu
->run
->exit_reason
= KVM_EXIT_IOAPIC_EOI
;
8907 vcpu
->run
->eoi
.vector
=
8908 vcpu
->arch
.pending_ioapic_eoi
;
8913 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC
, vcpu
))
8914 vcpu_scan_ioapic(vcpu
);
8915 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP
, vcpu
))
8916 vcpu_load_eoi_exitmap(vcpu
);
8917 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD
, vcpu
))
8918 kvm_vcpu_reload_apic_access_page(vcpu
);
8919 if (kvm_check_request(KVM_REQ_HV_CRASH
, vcpu
)) {
8920 vcpu
->run
->exit_reason
= KVM_EXIT_SYSTEM_EVENT
;
8921 vcpu
->run
->system_event
.type
= KVM_SYSTEM_EVENT_CRASH
;
8925 if (kvm_check_request(KVM_REQ_HV_RESET
, vcpu
)) {
8926 vcpu
->run
->exit_reason
= KVM_EXIT_SYSTEM_EVENT
;
8927 vcpu
->run
->system_event
.type
= KVM_SYSTEM_EVENT_RESET
;
8931 if (kvm_check_request(KVM_REQ_HV_EXIT
, vcpu
)) {
8932 vcpu
->run
->exit_reason
= KVM_EXIT_HYPERV
;
8933 vcpu
->run
->hyperv
= vcpu
->arch
.hyperv
.exit
;
8939 * KVM_REQ_HV_STIMER has to be processed after
8940 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8941 * depend on the guest clock being up-to-date
8943 if (kvm_check_request(KVM_REQ_HV_STIMER
, vcpu
))
8944 kvm_hv_process_stimers(vcpu
);
8945 if (kvm_check_request(KVM_REQ_APICV_UPDATE
, vcpu
))
8946 kvm_vcpu_update_apicv(vcpu
);
8947 if (kvm_check_request(KVM_REQ_APF_READY
, vcpu
))
8948 kvm_check_async_pf_completion(vcpu
);
8949 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED
, vcpu
))
8950 kvm_x86_ops
.msr_filter_changed(vcpu
);
8953 if (kvm_check_request(KVM_REQ_EVENT
, vcpu
) || req_int_win
) {
8954 ++vcpu
->stat
.req_event
;
8955 kvm_apic_accept_events(vcpu
);
8956 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
8961 inject_pending_event(vcpu
, &req_immediate_exit
);
8963 kvm_x86_ops
.enable_irq_window(vcpu
);
8965 if (kvm_lapic_enabled(vcpu
)) {
8966 update_cr8_intercept(vcpu
);
8967 kvm_lapic_sync_to_vapic(vcpu
);
8971 r
= kvm_mmu_reload(vcpu
);
8973 goto cancel_injection
;
8978 kvm_x86_ops
.prepare_guest_switch(vcpu
);
8981 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8982 * IPI are then delayed after guest entry, which ensures that they
8983 * result in virtual interrupt delivery.
8985 local_irq_disable();
8986 vcpu
->mode
= IN_GUEST_MODE
;
8988 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
8991 * 1) We should set ->mode before checking ->requests. Please see
8992 * the comment in kvm_vcpu_exiting_guest_mode().
8994 * 2) For APICv, we should set ->mode before checking PID.ON. This
8995 * pairs with the memory barrier implicit in pi_test_and_set_on
8996 * (see vmx_deliver_posted_interrupt).
8998 * 3) This also orders the write to mode from any reads to the page
8999 * tables done while the VCPU is running. Please see the comment
9000 * in kvm_flush_remote_tlbs.
9002 smp_mb__after_srcu_read_unlock();
9005 * This handles the case where a posted interrupt was
9006 * notified with kvm_vcpu_kick.
9008 if (kvm_lapic_enabled(vcpu
) && vcpu
->arch
.apicv_active
)
9009 kvm_x86_ops
.sync_pir_to_irr(vcpu
);
9011 if (kvm_vcpu_exit_request(vcpu
)) {
9012 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
9016 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
9018 goto cancel_injection
;
9021 if (req_immediate_exit
) {
9022 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
9023 kvm_x86_ops
.request_immediate_exit(vcpu
);
9026 fpregs_assert_state_consistent();
9027 if (test_thread_flag(TIF_NEED_FPU_LOAD
))
9028 switch_fpu_return();
9030 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
9032 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
9033 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
9034 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
9035 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
9036 set_debugreg(vcpu
->arch
.dr6
, 6);
9037 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_RELOAD
;
9040 exit_fastpath
= kvm_x86_ops
.run(vcpu
);
9043 * Do this here before restoring debug registers on the host. And
9044 * since we do this before handling the vmexit, a DR access vmexit
9045 * can (a) read the correct value of the debug registers, (b) set
9046 * KVM_DEBUGREG_WONT_EXIT again.
9048 if (unlikely(vcpu
->arch
.switch_db_regs
& KVM_DEBUGREG_WONT_EXIT
)) {
9049 WARN_ON(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
);
9050 kvm_x86_ops
.sync_dirty_debug_regs(vcpu
);
9051 kvm_update_dr0123(vcpu
);
9052 kvm_update_dr7(vcpu
);
9053 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_RELOAD
;
9057 * If the guest has used debug registers, at least dr7
9058 * will be disabled while returning to the host.
9059 * If we don't have active breakpoints in the host, we don't
9060 * care about the messed up debug address registers. But if
9061 * we have some of them active, restore the old state.
9063 if (hw_breakpoint_active())
9064 hw_breakpoint_restore();
9066 vcpu
->arch
.last_vmentry_cpu
= vcpu
->cpu
;
9067 vcpu
->arch
.last_guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
9069 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
9072 kvm_x86_ops
.handle_exit_irqoff(vcpu
);
9075 * Consume any pending interrupts, including the possible source of
9076 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9077 * An instruction is required after local_irq_enable() to fully unblock
9078 * interrupts on processors that implement an interrupt shadow, the
9079 * stat.exits increment will do nicely.
9081 kvm_before_interrupt(vcpu
);
9084 local_irq_disable();
9085 kvm_after_interrupt(vcpu
);
9087 if (lapic_in_kernel(vcpu
)) {
9088 s64 delta
= vcpu
->arch
.apic
->lapic_timer
.advance_expire_delta
;
9089 if (delta
!= S64_MIN
) {
9090 trace_kvm_wait_lapic_expire(vcpu
->vcpu_id
, delta
);
9091 vcpu
->arch
.apic
->lapic_timer
.advance_expire_delta
= S64_MIN
;
9098 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
9101 * Profile KVM exit RIPs:
9103 if (unlikely(prof_on
== KVM_PROFILING
)) {
9104 unsigned long rip
= kvm_rip_read(vcpu
);
9105 profile_hit(KVM_PROFILING
, (void *)rip
);
9108 if (unlikely(vcpu
->arch
.tsc_always_catchup
))
9109 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
9111 if (vcpu
->arch
.apic_attention
)
9112 kvm_lapic_sync_from_vapic(vcpu
);
9114 r
= kvm_x86_ops
.handle_exit(vcpu
, exit_fastpath
);
9118 if (req_immediate_exit
)
9119 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
9120 kvm_x86_ops
.cancel_injection(vcpu
);
9121 if (unlikely(vcpu
->arch
.apic_attention
))
9122 kvm_lapic_sync_from_vapic(vcpu
);
9127 static inline int vcpu_block(struct kvm
*kvm
, struct kvm_vcpu
*vcpu
)
9129 if (!kvm_arch_vcpu_runnable(vcpu
) &&
9130 (!kvm_x86_ops
.pre_block
|| kvm_x86_ops
.pre_block(vcpu
) == 0)) {
9131 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
9132 kvm_vcpu_block(vcpu
);
9133 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
9135 if (kvm_x86_ops
.post_block
)
9136 kvm_x86_ops
.post_block(vcpu
);
9138 if (!kvm_check_request(KVM_REQ_UNHALT
, vcpu
))
9142 kvm_apic_accept_events(vcpu
);
9143 switch(vcpu
->arch
.mp_state
) {
9144 case KVM_MP_STATE_HALTED
:
9145 case KVM_MP_STATE_AP_RESET_HOLD
:
9146 vcpu
->arch
.pv
.pv_unhalted
= false;
9147 vcpu
->arch
.mp_state
=
9148 KVM_MP_STATE_RUNNABLE
;
9150 case KVM_MP_STATE_RUNNABLE
:
9151 vcpu
->arch
.apf
.halted
= false;
9153 case KVM_MP_STATE_INIT_RECEIVED
:
9161 static inline bool kvm_vcpu_running(struct kvm_vcpu
*vcpu
)
9163 if (is_guest_mode(vcpu
))
9164 kvm_x86_ops
.nested_ops
->check_events(vcpu
);
9166 return (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
9167 !vcpu
->arch
.apf
.halted
);
9170 static int vcpu_run(struct kvm_vcpu
*vcpu
)
9173 struct kvm
*kvm
= vcpu
->kvm
;
9175 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
9176 vcpu
->arch
.l1tf_flush_l1d
= true;
9179 if (kvm_vcpu_running(vcpu
)) {
9180 r
= vcpu_enter_guest(vcpu
);
9182 r
= vcpu_block(kvm
, vcpu
);
9188 kvm_clear_request(KVM_REQ_PENDING_TIMER
, vcpu
);
9189 if (kvm_cpu_has_pending_timer(vcpu
))
9190 kvm_inject_pending_timer_irqs(vcpu
);
9192 if (dm_request_for_irq_injection(vcpu
) &&
9193 kvm_vcpu_ready_for_interrupt_injection(vcpu
)) {
9195 vcpu
->run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
9196 ++vcpu
->stat
.request_irq_exits
;
9200 if (__xfer_to_guest_mode_work_pending()) {
9201 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
9202 r
= xfer_to_guest_mode_handle_work(vcpu
);
9205 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
9209 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
9214 static inline int complete_emulated_io(struct kvm_vcpu
*vcpu
)
9218 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
9219 r
= kvm_emulate_instruction(vcpu
, EMULTYPE_NO_DECODE
);
9220 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
9224 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
)
9226 BUG_ON(!vcpu
->arch
.pio
.count
);
9228 return complete_emulated_io(vcpu
);
9232 * Implements the following, as a state machine:
9236 * for each mmio piece in the fragment
9244 * for each mmio piece in the fragment
9249 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
)
9251 struct kvm_run
*run
= vcpu
->run
;
9252 struct kvm_mmio_fragment
*frag
;
9255 BUG_ON(!vcpu
->mmio_needed
);
9257 /* Complete previous fragment */
9258 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_cur_fragment
];
9259 len
= min(8u, frag
->len
);
9260 if (!vcpu
->mmio_is_write
)
9261 memcpy(frag
->data
, run
->mmio
.data
, len
);
9263 if (frag
->len
<= 8) {
9264 /* Switch to the next fragment. */
9266 vcpu
->mmio_cur_fragment
++;
9268 /* Go forward to the next mmio piece. */
9274 if (vcpu
->mmio_cur_fragment
>= vcpu
->mmio_nr_fragments
) {
9275 vcpu
->mmio_needed
= 0;
9277 /* FIXME: return into emulator if single-stepping. */
9278 if (vcpu
->mmio_is_write
)
9280 vcpu
->mmio_read_completed
= 1;
9281 return complete_emulated_io(vcpu
);
9284 run
->exit_reason
= KVM_EXIT_MMIO
;
9285 run
->mmio
.phys_addr
= frag
->gpa
;
9286 if (vcpu
->mmio_is_write
)
9287 memcpy(run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
9288 run
->mmio
.len
= min(8u, frag
->len
);
9289 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
9290 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
9294 static void kvm_save_current_fpu(struct fpu
*fpu
)
9297 * If the target FPU state is not resident in the CPU registers, just
9298 * memcpy() from current, else save CPU state directly to the target.
9300 if (test_thread_flag(TIF_NEED_FPU_LOAD
))
9301 memcpy(&fpu
->state
, ¤t
->thread
.fpu
.state
,
9302 fpu_kernel_xstate_size
);
9304 copy_fpregs_to_fpstate(fpu
);
9307 /* Swap (qemu) user FPU context for the guest FPU context. */
9308 static void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
9312 kvm_save_current_fpu(vcpu
->arch
.user_fpu
);
9315 * Guests with protected state can't have it set by the hypervisor,
9316 * so skip trying to set it.
9318 if (vcpu
->arch
.guest_fpu
)
9319 /* PKRU is separately restored in kvm_x86_ops.run. */
9320 __copy_kernel_to_fpregs(&vcpu
->arch
.guest_fpu
->state
,
9321 ~XFEATURE_MASK_PKRU
);
9323 fpregs_mark_activate();
9329 /* When vcpu_run ends, restore user space FPU context. */
9330 static void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
9335 * Guests with protected state can't have it read by the hypervisor,
9336 * so skip trying to save it.
9338 if (vcpu
->arch
.guest_fpu
)
9339 kvm_save_current_fpu(vcpu
->arch
.guest_fpu
);
9341 copy_kernel_to_fpregs(&vcpu
->arch
.user_fpu
->state
);
9343 fpregs_mark_activate();
9346 ++vcpu
->stat
.fpu_reload
;
9350 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
)
9352 struct kvm_run
*kvm_run
= vcpu
->run
;
9356 kvm_sigset_activate(vcpu
);
9357 kvm_load_guest_fpu(vcpu
);
9359 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
9360 if (kvm_run
->immediate_exit
) {
9364 kvm_vcpu_block(vcpu
);
9365 kvm_apic_accept_events(vcpu
);
9366 kvm_clear_request(KVM_REQ_UNHALT
, vcpu
);
9368 if (signal_pending(current
)) {
9370 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
9371 ++vcpu
->stat
.signal_exits
;
9376 if (kvm_run
->kvm_valid_regs
& ~KVM_SYNC_X86_VALID_FIELDS
) {
9381 if (kvm_run
->kvm_dirty_regs
) {
9382 r
= sync_regs(vcpu
);
9387 /* re-sync apic's tpr */
9388 if (!lapic_in_kernel(vcpu
)) {
9389 if (kvm_set_cr8(vcpu
, kvm_run
->cr8
) != 0) {
9395 if (unlikely(vcpu
->arch
.complete_userspace_io
)) {
9396 int (*cui
)(struct kvm_vcpu
*) = vcpu
->arch
.complete_userspace_io
;
9397 vcpu
->arch
.complete_userspace_io
= NULL
;
9402 WARN_ON(vcpu
->arch
.pio
.count
|| vcpu
->mmio_needed
);
9404 if (kvm_run
->immediate_exit
)
9410 kvm_put_guest_fpu(vcpu
);
9411 if (kvm_run
->kvm_valid_regs
)
9413 post_kvm_run_save(vcpu
);
9414 kvm_sigset_deactivate(vcpu
);
9420 static void __get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
9422 if (vcpu
->arch
.emulate_regs_need_sync_to_vcpu
) {
9424 * We are here if userspace calls get_regs() in the middle of
9425 * instruction emulation. Registers state needs to be copied
9426 * back from emulation context to vcpu. Userspace shouldn't do
9427 * that usually, but some bad designed PV devices (vmware
9428 * backdoor interface) need this to work
9430 emulator_writeback_register_cache(vcpu
->arch
.emulate_ctxt
);
9431 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
9433 regs
->rax
= kvm_rax_read(vcpu
);
9434 regs
->rbx
= kvm_rbx_read(vcpu
);
9435 regs
->rcx
= kvm_rcx_read(vcpu
);
9436 regs
->rdx
= kvm_rdx_read(vcpu
);
9437 regs
->rsi
= kvm_rsi_read(vcpu
);
9438 regs
->rdi
= kvm_rdi_read(vcpu
);
9439 regs
->rsp
= kvm_rsp_read(vcpu
);
9440 regs
->rbp
= kvm_rbp_read(vcpu
);
9441 #ifdef CONFIG_X86_64
9442 regs
->r8
= kvm_r8_read(vcpu
);
9443 regs
->r9
= kvm_r9_read(vcpu
);
9444 regs
->r10
= kvm_r10_read(vcpu
);
9445 regs
->r11
= kvm_r11_read(vcpu
);
9446 regs
->r12
= kvm_r12_read(vcpu
);
9447 regs
->r13
= kvm_r13_read(vcpu
);
9448 regs
->r14
= kvm_r14_read(vcpu
);
9449 regs
->r15
= kvm_r15_read(vcpu
);
9452 regs
->rip
= kvm_rip_read(vcpu
);
9453 regs
->rflags
= kvm_get_rflags(vcpu
);
9456 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
9459 __get_regs(vcpu
, regs
);
9464 static void __set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
9466 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= true;
9467 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
9469 kvm_rax_write(vcpu
, regs
->rax
);
9470 kvm_rbx_write(vcpu
, regs
->rbx
);
9471 kvm_rcx_write(vcpu
, regs
->rcx
);
9472 kvm_rdx_write(vcpu
, regs
->rdx
);
9473 kvm_rsi_write(vcpu
, regs
->rsi
);
9474 kvm_rdi_write(vcpu
, regs
->rdi
);
9475 kvm_rsp_write(vcpu
, regs
->rsp
);
9476 kvm_rbp_write(vcpu
, regs
->rbp
);
9477 #ifdef CONFIG_X86_64
9478 kvm_r8_write(vcpu
, regs
->r8
);
9479 kvm_r9_write(vcpu
, regs
->r9
);
9480 kvm_r10_write(vcpu
, regs
->r10
);
9481 kvm_r11_write(vcpu
, regs
->r11
);
9482 kvm_r12_write(vcpu
, regs
->r12
);
9483 kvm_r13_write(vcpu
, regs
->r13
);
9484 kvm_r14_write(vcpu
, regs
->r14
);
9485 kvm_r15_write(vcpu
, regs
->r15
);
9488 kvm_rip_write(vcpu
, regs
->rip
);
9489 kvm_set_rflags(vcpu
, regs
->rflags
| X86_EFLAGS_FIXED
);
9491 vcpu
->arch
.exception
.pending
= false;
9493 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
9496 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
9499 __set_regs(vcpu
, regs
);
9504 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
9506 struct kvm_segment cs
;
9508 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
9512 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
9514 static void __get_sregs(struct kvm_vcpu
*vcpu
, struct kvm_sregs
*sregs
)
9518 if (vcpu
->arch
.guest_state_protected
)
9519 goto skip_protected_regs
;
9521 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
9522 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
9523 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
9524 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
9525 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
9526 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
9528 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
9529 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
9531 kvm_x86_ops
.get_idt(vcpu
, &dt
);
9532 sregs
->idt
.limit
= dt
.size
;
9533 sregs
->idt
.base
= dt
.address
;
9534 kvm_x86_ops
.get_gdt(vcpu
, &dt
);
9535 sregs
->gdt
.limit
= dt
.size
;
9536 sregs
->gdt
.base
= dt
.address
;
9538 sregs
->cr2
= vcpu
->arch
.cr2
;
9539 sregs
->cr3
= kvm_read_cr3(vcpu
);
9541 skip_protected_regs
:
9542 sregs
->cr0
= kvm_read_cr0(vcpu
);
9543 sregs
->cr4
= kvm_read_cr4(vcpu
);
9544 sregs
->cr8
= kvm_get_cr8(vcpu
);
9545 sregs
->efer
= vcpu
->arch
.efer
;
9546 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
9548 memset(sregs
->interrupt_bitmap
, 0, sizeof(sregs
->interrupt_bitmap
));
9550 if (vcpu
->arch
.interrupt
.injected
&& !vcpu
->arch
.interrupt
.soft
)
9551 set_bit(vcpu
->arch
.interrupt
.nr
,
9552 (unsigned long *)sregs
->interrupt_bitmap
);
9555 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
9556 struct kvm_sregs
*sregs
)
9559 __get_sregs(vcpu
, sregs
);
9564 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
9565 struct kvm_mp_state
*mp_state
)
9568 if (kvm_mpx_supported())
9569 kvm_load_guest_fpu(vcpu
);
9571 kvm_apic_accept_events(vcpu
);
9572 if ((vcpu
->arch
.mp_state
== KVM_MP_STATE_HALTED
||
9573 vcpu
->arch
.mp_state
== KVM_MP_STATE_AP_RESET_HOLD
) &&
9574 vcpu
->arch
.pv
.pv_unhalted
)
9575 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
9577 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
9579 if (kvm_mpx_supported())
9580 kvm_put_guest_fpu(vcpu
);
9585 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
9586 struct kvm_mp_state
*mp_state
)
9592 if (!lapic_in_kernel(vcpu
) &&
9593 mp_state
->mp_state
!= KVM_MP_STATE_RUNNABLE
)
9597 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9598 * INIT state; latched init should be reported using
9599 * KVM_SET_VCPU_EVENTS, so reject it here.
9601 if ((kvm_vcpu_latch_init(vcpu
) || vcpu
->arch
.smi_pending
) &&
9602 (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
||
9603 mp_state
->mp_state
== KVM_MP_STATE_INIT_RECEIVED
))
9606 if (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
) {
9607 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
9608 set_bit(KVM_APIC_SIPI
, &vcpu
->arch
.apic
->pending_events
);
9610 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
9611 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
9619 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int idt_index
,
9620 int reason
, bool has_error_code
, u32 error_code
)
9622 struct x86_emulate_ctxt
*ctxt
= vcpu
->arch
.emulate_ctxt
;
9625 init_emulate_ctxt(vcpu
);
9627 ret
= emulator_task_switch(ctxt
, tss_selector
, idt_index
, reason
,
9628 has_error_code
, error_code
);
9630 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
9631 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
9632 vcpu
->run
->internal
.ndata
= 0;
9636 kvm_rip_write(vcpu
, ctxt
->eip
);
9637 kvm_set_rflags(vcpu
, ctxt
->eflags
);
9640 EXPORT_SYMBOL_GPL(kvm_task_switch
);
9642 static bool kvm_is_valid_sregs(struct kvm_vcpu
*vcpu
, struct kvm_sregs
*sregs
)
9644 if ((sregs
->efer
& EFER_LME
) && (sregs
->cr0
& X86_CR0_PG
)) {
9646 * When EFER.LME and CR0.PG are set, the processor is in
9647 * 64-bit mode (though maybe in a 32-bit code segment).
9648 * CR4.PAE and EFER.LMA must be set.
9650 if (!(sregs
->cr4
& X86_CR4_PAE
) || !(sregs
->efer
& EFER_LMA
))
9652 if (sregs
->cr3
& vcpu
->arch
.cr3_lm_rsvd_bits
)
9656 * Not in 64-bit mode: EFER.LMA is clear and the code
9657 * segment cannot be 64-bit.
9659 if (sregs
->efer
& EFER_LMA
|| sregs
->cs
.l
)
9663 return kvm_is_valid_cr4(vcpu
, sregs
->cr4
);
9666 static int __set_sregs(struct kvm_vcpu
*vcpu
, struct kvm_sregs
*sregs
)
9668 struct msr_data apic_base_msr
;
9669 int mmu_reset_needed
= 0;
9670 int pending_vec
, max_bits
, idx
;
9674 if (!kvm_is_valid_sregs(vcpu
, sregs
))
9677 apic_base_msr
.data
= sregs
->apic_base
;
9678 apic_base_msr
.host_initiated
= true;
9679 if (kvm_set_apic_base(vcpu
, &apic_base_msr
))
9682 if (vcpu
->arch
.guest_state_protected
)
9683 goto skip_protected_regs
;
9685 dt
.size
= sregs
->idt
.limit
;
9686 dt
.address
= sregs
->idt
.base
;
9687 kvm_x86_ops
.set_idt(vcpu
, &dt
);
9688 dt
.size
= sregs
->gdt
.limit
;
9689 dt
.address
= sregs
->gdt
.base
;
9690 kvm_x86_ops
.set_gdt(vcpu
, &dt
);
9692 vcpu
->arch
.cr2
= sregs
->cr2
;
9693 mmu_reset_needed
|= kvm_read_cr3(vcpu
) != sregs
->cr3
;
9694 vcpu
->arch
.cr3
= sregs
->cr3
;
9695 kvm_register_mark_available(vcpu
, VCPU_EXREG_CR3
);
9697 kvm_set_cr8(vcpu
, sregs
->cr8
);
9699 mmu_reset_needed
|= vcpu
->arch
.efer
!= sregs
->efer
;
9700 kvm_x86_ops
.set_efer(vcpu
, sregs
->efer
);
9702 mmu_reset_needed
|= kvm_read_cr0(vcpu
) != sregs
->cr0
;
9703 kvm_x86_ops
.set_cr0(vcpu
, sregs
->cr0
);
9704 vcpu
->arch
.cr0
= sregs
->cr0
;
9706 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
9707 kvm_x86_ops
.set_cr4(vcpu
, sregs
->cr4
);
9709 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
9710 if (is_pae_paging(vcpu
)) {
9711 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
9712 mmu_reset_needed
= 1;
9714 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
9716 if (mmu_reset_needed
)
9717 kvm_mmu_reset_context(vcpu
);
9719 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
9720 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
9721 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
9722 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
9723 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
9724 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
9726 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
9727 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
9729 update_cr8_intercept(vcpu
);
9731 /* Older userspace won't unhalt the vcpu on reset. */
9732 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
9733 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
9735 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
9737 skip_protected_regs
:
9738 max_bits
= KVM_NR_INTERRUPTS
;
9739 pending_vec
= find_first_bit(
9740 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
9741 if (pending_vec
< max_bits
) {
9742 kvm_queue_interrupt(vcpu
, pending_vec
, false);
9743 pr_debug("Set back pending irq %d\n", pending_vec
);
9746 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
9753 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
9754 struct kvm_sregs
*sregs
)
9759 ret
= __set_sregs(vcpu
, sregs
);
9764 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
9765 struct kvm_guest_debug
*dbg
)
9767 unsigned long rflags
;
9770 if (vcpu
->arch
.guest_state_protected
)
9775 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
9777 if (vcpu
->arch
.exception
.pending
)
9779 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
9780 kvm_queue_exception(vcpu
, DB_VECTOR
);
9782 kvm_queue_exception(vcpu
, BP_VECTOR
);
9786 * Read rflags as long as potentially injected trace flags are still
9789 rflags
= kvm_get_rflags(vcpu
);
9791 vcpu
->guest_debug
= dbg
->control
;
9792 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
9793 vcpu
->guest_debug
= 0;
9795 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
9796 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
9797 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
9798 vcpu
->arch
.guest_debug_dr7
= dbg
->arch
.debugreg
[7];
9800 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
9801 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
9803 kvm_update_dr7(vcpu
);
9805 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
9806 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
) +
9807 get_segment_base(vcpu
, VCPU_SREG_CS
);
9810 * Trigger an rflags update that will inject or remove the trace
9813 kvm_set_rflags(vcpu
, rflags
);
9815 kvm_x86_ops
.update_exception_bitmap(vcpu
);
9825 * Translate a guest virtual address to a guest physical address.
9827 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
9828 struct kvm_translation
*tr
)
9830 unsigned long vaddr
= tr
->linear_address
;
9836 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
9837 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, vaddr
, NULL
);
9838 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
9839 tr
->physical_address
= gpa
;
9840 tr
->valid
= gpa
!= UNMAPPED_GVA
;
9848 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
9850 struct fxregs_state
*fxsave
;
9852 if (!vcpu
->arch
.guest_fpu
)
9857 fxsave
= &vcpu
->arch
.guest_fpu
->state
.fxsave
;
9858 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
9859 fpu
->fcw
= fxsave
->cwd
;
9860 fpu
->fsw
= fxsave
->swd
;
9861 fpu
->ftwx
= fxsave
->twd
;
9862 fpu
->last_opcode
= fxsave
->fop
;
9863 fpu
->last_ip
= fxsave
->rip
;
9864 fpu
->last_dp
= fxsave
->rdp
;
9865 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof(fxsave
->xmm_space
));
9871 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
9873 struct fxregs_state
*fxsave
;
9875 if (!vcpu
->arch
.guest_fpu
)
9880 fxsave
= &vcpu
->arch
.guest_fpu
->state
.fxsave
;
9882 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
9883 fxsave
->cwd
= fpu
->fcw
;
9884 fxsave
->swd
= fpu
->fsw
;
9885 fxsave
->twd
= fpu
->ftwx
;
9886 fxsave
->fop
= fpu
->last_opcode
;
9887 fxsave
->rip
= fpu
->last_ip
;
9888 fxsave
->rdp
= fpu
->last_dp
;
9889 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof(fxsave
->xmm_space
));
9895 static void store_regs(struct kvm_vcpu
*vcpu
)
9897 BUILD_BUG_ON(sizeof(struct kvm_sync_regs
) > SYNC_REGS_SIZE_BYTES
);
9899 if (vcpu
->run
->kvm_valid_regs
& KVM_SYNC_X86_REGS
)
9900 __get_regs(vcpu
, &vcpu
->run
->s
.regs
.regs
);
9902 if (vcpu
->run
->kvm_valid_regs
& KVM_SYNC_X86_SREGS
)
9903 __get_sregs(vcpu
, &vcpu
->run
->s
.regs
.sregs
);
9905 if (vcpu
->run
->kvm_valid_regs
& KVM_SYNC_X86_EVENTS
)
9906 kvm_vcpu_ioctl_x86_get_vcpu_events(
9907 vcpu
, &vcpu
->run
->s
.regs
.events
);
9910 static int sync_regs(struct kvm_vcpu
*vcpu
)
9912 if (vcpu
->run
->kvm_dirty_regs
& ~KVM_SYNC_X86_VALID_FIELDS
)
9915 if (vcpu
->run
->kvm_dirty_regs
& KVM_SYNC_X86_REGS
) {
9916 __set_regs(vcpu
, &vcpu
->run
->s
.regs
.regs
);
9917 vcpu
->run
->kvm_dirty_regs
&= ~KVM_SYNC_X86_REGS
;
9919 if (vcpu
->run
->kvm_dirty_regs
& KVM_SYNC_X86_SREGS
) {
9920 if (__set_sregs(vcpu
, &vcpu
->run
->s
.regs
.sregs
))
9922 vcpu
->run
->kvm_dirty_regs
&= ~KVM_SYNC_X86_SREGS
;
9924 if (vcpu
->run
->kvm_dirty_regs
& KVM_SYNC_X86_EVENTS
) {
9925 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9926 vcpu
, &vcpu
->run
->s
.regs
.events
))
9928 vcpu
->run
->kvm_dirty_regs
&= ~KVM_SYNC_X86_EVENTS
;
9934 static void fx_init(struct kvm_vcpu
*vcpu
)
9936 if (!vcpu
->arch
.guest_fpu
)
9939 fpstate_init(&vcpu
->arch
.guest_fpu
->state
);
9940 if (boot_cpu_has(X86_FEATURE_XSAVES
))
9941 vcpu
->arch
.guest_fpu
->state
.xsave
.header
.xcomp_bv
=
9942 host_xcr0
| XSTATE_COMPACTION_ENABLED
;
9945 * Ensure guest xcr0 is valid for loading
9947 vcpu
->arch
.xcr0
= XFEATURE_MASK_FP
;
9949 vcpu
->arch
.cr0
|= X86_CR0_ET
;
9952 void kvm_free_guest_fpu(struct kvm_vcpu
*vcpu
)
9954 if (vcpu
->arch
.guest_fpu
) {
9955 kmem_cache_free(x86_fpu_cache
, vcpu
->arch
.guest_fpu
);
9956 vcpu
->arch
.guest_fpu
= NULL
;
9959 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu
);
9961 int kvm_arch_vcpu_precreate(struct kvm
*kvm
, unsigned int id
)
9963 if (kvm_check_tsc_unstable() && atomic_read(&kvm
->online_vcpus
) != 0)
9964 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9965 "guest TSC will not be reliable\n");
9970 int kvm_arch_vcpu_create(struct kvm_vcpu
*vcpu
)
9975 if (!irqchip_in_kernel(vcpu
->kvm
) || kvm_vcpu_is_reset_bsp(vcpu
))
9976 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
9978 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
9980 kvm_set_tsc_khz(vcpu
, max_tsc_khz
);
9982 r
= kvm_mmu_create(vcpu
);
9986 if (irqchip_in_kernel(vcpu
->kvm
)) {
9987 r
= kvm_create_lapic(vcpu
, lapic_timer_advance_ns
);
9989 goto fail_mmu_destroy
;
9990 if (kvm_apicv_activated(vcpu
->kvm
))
9991 vcpu
->arch
.apicv_active
= true;
9993 static_key_slow_inc(&kvm_no_apic_vcpu
);
9997 page
= alloc_page(GFP_KERNEL_ACCOUNT
| __GFP_ZERO
);
9999 goto fail_free_lapic
;
10000 vcpu
->arch
.pio_data
= page_address(page
);
10002 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
10003 GFP_KERNEL_ACCOUNT
);
10004 if (!vcpu
->arch
.mce_banks
)
10005 goto fail_free_pio_data
;
10006 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
10008 if (!zalloc_cpumask_var(&vcpu
->arch
.wbinvd_dirty_mask
,
10009 GFP_KERNEL_ACCOUNT
))
10010 goto fail_free_mce_banks
;
10012 if (!alloc_emulate_ctxt(vcpu
))
10013 goto free_wbinvd_dirty_mask
;
10015 vcpu
->arch
.user_fpu
= kmem_cache_zalloc(x86_fpu_cache
,
10016 GFP_KERNEL_ACCOUNT
);
10017 if (!vcpu
->arch
.user_fpu
) {
10018 pr_err("kvm: failed to allocate userspace's fpu\n");
10019 goto free_emulate_ctxt
;
10022 vcpu
->arch
.guest_fpu
= kmem_cache_zalloc(x86_fpu_cache
,
10023 GFP_KERNEL_ACCOUNT
);
10024 if (!vcpu
->arch
.guest_fpu
) {
10025 pr_err("kvm: failed to allocate vcpu's fpu\n");
10026 goto free_user_fpu
;
10030 vcpu
->arch
.maxphyaddr
= cpuid_query_maxphyaddr(vcpu
);
10031 vcpu
->arch
.cr3_lm_rsvd_bits
= rsvd_bits(cpuid_maxphyaddr(vcpu
), 63);
10033 vcpu
->arch
.pat
= MSR_IA32_CR_PAT_DEFAULT
;
10035 kvm_async_pf_hash_reset(vcpu
);
10036 kvm_pmu_init(vcpu
);
10038 vcpu
->arch
.pending_external_vector
= -1;
10039 vcpu
->arch
.preempted_in_kernel
= false;
10041 kvm_hv_vcpu_init(vcpu
);
10043 r
= kvm_x86_ops
.vcpu_create(vcpu
);
10045 goto free_guest_fpu
;
10047 vcpu
->arch
.arch_capabilities
= kvm_get_arch_capabilities();
10048 vcpu
->arch
.msr_platform_info
= MSR_PLATFORM_INFO_CPUID_FAULT
;
10049 kvm_vcpu_mtrr_init(vcpu
);
10051 kvm_vcpu_reset(vcpu
, false);
10052 kvm_init_mmu(vcpu
, false);
10057 kvm_free_guest_fpu(vcpu
);
10059 kmem_cache_free(x86_fpu_cache
, vcpu
->arch
.user_fpu
);
10061 kmem_cache_free(x86_emulator_cache
, vcpu
->arch
.emulate_ctxt
);
10062 free_wbinvd_dirty_mask
:
10063 free_cpumask_var(vcpu
->arch
.wbinvd_dirty_mask
);
10064 fail_free_mce_banks
:
10065 kfree(vcpu
->arch
.mce_banks
);
10066 fail_free_pio_data
:
10067 free_page((unsigned long)vcpu
->arch
.pio_data
);
10069 kvm_free_lapic(vcpu
);
10071 kvm_mmu_destroy(vcpu
);
10075 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
10077 struct kvm
*kvm
= vcpu
->kvm
;
10079 kvm_hv_vcpu_postcreate(vcpu
);
10081 if (mutex_lock_killable(&vcpu
->mutex
))
10084 kvm_synchronize_tsc(vcpu
, 0);
10087 /* poll control enabled by default */
10088 vcpu
->arch
.msr_kvm_poll_control
= 1;
10090 mutex_unlock(&vcpu
->mutex
);
10092 if (kvmclock_periodic_sync
&& vcpu
->vcpu_idx
== 0)
10093 schedule_delayed_work(&kvm
->arch
.kvmclock_sync_work
,
10094 KVMCLOCK_SYNC_PERIOD
);
10097 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
10099 struct gfn_to_pfn_cache
*cache
= &vcpu
->arch
.st
.cache
;
10102 kvm_release_pfn(cache
->pfn
, cache
->dirty
, cache
);
10104 kvmclock_reset(vcpu
);
10106 kvm_x86_ops
.vcpu_free(vcpu
);
10108 kmem_cache_free(x86_emulator_cache
, vcpu
->arch
.emulate_ctxt
);
10109 free_cpumask_var(vcpu
->arch
.wbinvd_dirty_mask
);
10110 kmem_cache_free(x86_fpu_cache
, vcpu
->arch
.user_fpu
);
10111 kvm_free_guest_fpu(vcpu
);
10113 kvm_hv_vcpu_uninit(vcpu
);
10114 kvm_pmu_destroy(vcpu
);
10115 kfree(vcpu
->arch
.mce_banks
);
10116 kvm_free_lapic(vcpu
);
10117 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
10118 kvm_mmu_destroy(vcpu
);
10119 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
10120 free_page((unsigned long)vcpu
->arch
.pio_data
);
10121 kvfree(vcpu
->arch
.cpuid_entries
);
10122 if (!lapic_in_kernel(vcpu
))
10123 static_key_slow_dec(&kvm_no_apic_vcpu
);
10126 void kvm_vcpu_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
10128 kvm_lapic_reset(vcpu
, init_event
);
10130 vcpu
->arch
.hflags
= 0;
10132 vcpu
->arch
.smi_pending
= 0;
10133 vcpu
->arch
.smi_count
= 0;
10134 atomic_set(&vcpu
->arch
.nmi_queued
, 0);
10135 vcpu
->arch
.nmi_pending
= 0;
10136 vcpu
->arch
.nmi_injected
= false;
10137 kvm_clear_interrupt_queue(vcpu
);
10138 kvm_clear_exception_queue(vcpu
);
10140 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
10141 kvm_update_dr0123(vcpu
);
10142 vcpu
->arch
.dr6
= DR6_INIT
;
10143 vcpu
->arch
.dr7
= DR7_FIXED_1
;
10144 kvm_update_dr7(vcpu
);
10146 vcpu
->arch
.cr2
= 0;
10148 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
10149 vcpu
->arch
.apf
.msr_en_val
= 0;
10150 vcpu
->arch
.apf
.msr_int_val
= 0;
10151 vcpu
->arch
.st
.msr_val
= 0;
10153 kvmclock_reset(vcpu
);
10155 kvm_clear_async_pf_completion_queue(vcpu
);
10156 kvm_async_pf_hash_reset(vcpu
);
10157 vcpu
->arch
.apf
.halted
= false;
10159 if (vcpu
->arch
.guest_fpu
&& kvm_mpx_supported()) {
10160 void *mpx_state_buffer
;
10163 * To avoid have the INIT path from kvm_apic_has_events() that be
10164 * called with loaded FPU and does not let userspace fix the state.
10167 kvm_put_guest_fpu(vcpu
);
10168 mpx_state_buffer
= get_xsave_addr(&vcpu
->arch
.guest_fpu
->state
.xsave
,
10170 if (mpx_state_buffer
)
10171 memset(mpx_state_buffer
, 0, sizeof(struct mpx_bndreg_state
));
10172 mpx_state_buffer
= get_xsave_addr(&vcpu
->arch
.guest_fpu
->state
.xsave
,
10174 if (mpx_state_buffer
)
10175 memset(mpx_state_buffer
, 0, sizeof(struct mpx_bndcsr
));
10177 kvm_load_guest_fpu(vcpu
);
10181 kvm_pmu_reset(vcpu
);
10182 vcpu
->arch
.smbase
= 0x30000;
10184 vcpu
->arch
.msr_misc_features_enables
= 0;
10186 vcpu
->arch
.xcr0
= XFEATURE_MASK_FP
;
10189 memset(vcpu
->arch
.regs
, 0, sizeof(vcpu
->arch
.regs
));
10190 vcpu
->arch
.regs_avail
= ~0;
10191 vcpu
->arch
.regs_dirty
= ~0;
10193 vcpu
->arch
.ia32_xss
= 0;
10195 kvm_x86_ops
.vcpu_reset(vcpu
, init_event
);
10198 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu
*vcpu
, u8 vector
)
10200 struct kvm_segment cs
;
10202 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
10203 cs
.selector
= vector
<< 8;
10204 cs
.base
= vector
<< 12;
10205 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
10206 kvm_rip_write(vcpu
, 0);
10208 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector
);
10210 int kvm_arch_hardware_enable(void)
10213 struct kvm_vcpu
*vcpu
;
10218 bool stable
, backwards_tsc
= false;
10220 kvm_user_return_msr_cpu_online();
10221 ret
= kvm_x86_ops
.hardware_enable();
10225 local_tsc
= rdtsc();
10226 stable
= !kvm_check_tsc_unstable();
10227 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
10228 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
10229 if (!stable
&& vcpu
->cpu
== smp_processor_id())
10230 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
10231 if (stable
&& vcpu
->arch
.last_host_tsc
> local_tsc
) {
10232 backwards_tsc
= true;
10233 if (vcpu
->arch
.last_host_tsc
> max_tsc
)
10234 max_tsc
= vcpu
->arch
.last_host_tsc
;
10240 * Sometimes, even reliable TSCs go backwards. This happens on
10241 * platforms that reset TSC during suspend or hibernate actions, but
10242 * maintain synchronization. We must compensate. Fortunately, we can
10243 * detect that condition here, which happens early in CPU bringup,
10244 * before any KVM threads can be running. Unfortunately, we can't
10245 * bring the TSCs fully up to date with real time, as we aren't yet far
10246 * enough into CPU bringup that we know how much real time has actually
10247 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10248 * variables that haven't been updated yet.
10250 * So we simply find the maximum observed TSC above, then record the
10251 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10252 * the adjustment will be applied. Note that we accumulate
10253 * adjustments, in case multiple suspend cycles happen before some VCPU
10254 * gets a chance to run again. In the event that no KVM threads get a
10255 * chance to run, we will miss the entire elapsed period, as we'll have
10256 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10257 * loose cycle time. This isn't too big a deal, since the loss will be
10258 * uniform across all VCPUs (not to mention the scenario is extremely
10259 * unlikely). It is possible that a second hibernate recovery happens
10260 * much faster than a first, causing the observed TSC here to be
10261 * smaller; this would require additional padding adjustment, which is
10262 * why we set last_host_tsc to the local tsc observed here.
10264 * N.B. - this code below runs only on platforms with reliable TSC,
10265 * as that is the only way backwards_tsc is set above. Also note
10266 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10267 * have the same delta_cyc adjustment applied if backwards_tsc
10268 * is detected. Note further, this adjustment is only done once,
10269 * as we reset last_host_tsc on all VCPUs to stop this from being
10270 * called multiple times (one for each physical CPU bringup).
10272 * Platforms with unreliable TSCs don't have to deal with this, they
10273 * will be compensated by the logic in vcpu_load, which sets the TSC to
10274 * catchup mode. This will catchup all VCPUs to real time, but cannot
10275 * guarantee that they stay in perfect synchronization.
10277 if (backwards_tsc
) {
10278 u64 delta_cyc
= max_tsc
- local_tsc
;
10279 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
10280 kvm
->arch
.backwards_tsc_observed
= true;
10281 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
10282 vcpu
->arch
.tsc_offset_adjustment
+= delta_cyc
;
10283 vcpu
->arch
.last_host_tsc
= local_tsc
;
10284 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
10288 * We have to disable TSC offset matching.. if you were
10289 * booting a VM while issuing an S4 host suspend....
10290 * you may have some problem. Solving this issue is
10291 * left as an exercise to the reader.
10293 kvm
->arch
.last_tsc_nsec
= 0;
10294 kvm
->arch
.last_tsc_write
= 0;
10301 void kvm_arch_hardware_disable(void)
10303 kvm_x86_ops
.hardware_disable();
10304 drop_user_return_notifiers();
10307 int kvm_arch_hardware_setup(void *opaque
)
10309 struct kvm_x86_init_ops
*ops
= opaque
;
10312 rdmsrl_safe(MSR_EFER
, &host_efer
);
10314 if (boot_cpu_has(X86_FEATURE_XSAVES
))
10315 rdmsrl(MSR_IA32_XSS
, host_xss
);
10317 r
= ops
->hardware_setup();
10321 memcpy(&kvm_x86_ops
, ops
->runtime_ops
, sizeof(kvm_x86_ops
));
10323 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES
))
10326 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10327 cr4_reserved_bits
= __cr4_reserved_bits(__kvm_cpu_cap_has
, UNUSED_
);
10328 #undef __kvm_cpu_cap_has
10330 if (kvm_has_tsc_control
) {
10332 * Make sure the user can only configure tsc_khz values that
10333 * fit into a signed integer.
10334 * A min value is not calculated because it will always
10335 * be 1 on all machines.
10337 u64 max
= min(0x7fffffffULL
,
10338 __scale_tsc(kvm_max_tsc_scaling_ratio
, tsc_khz
));
10339 kvm_max_guest_tsc_khz
= max
;
10341 kvm_default_tsc_scaling_ratio
= 1ULL << kvm_tsc_scaling_ratio_frac_bits
;
10344 kvm_init_msr_list();
10348 void kvm_arch_hardware_unsetup(void)
10350 kvm_x86_ops
.hardware_unsetup();
10353 int kvm_arch_check_processor_compat(void *opaque
)
10355 struct cpuinfo_x86
*c
= &cpu_data(smp_processor_id());
10356 struct kvm_x86_init_ops
*ops
= opaque
;
10358 WARN_ON(!irqs_disabled());
10360 if (__cr4_reserved_bits(cpu_has
, c
) !=
10361 __cr4_reserved_bits(cpu_has
, &boot_cpu_data
))
10364 return ops
->check_processor_compatibility();
10367 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu
*vcpu
)
10369 return vcpu
->kvm
->arch
.bsp_vcpu_id
== vcpu
->vcpu_id
;
10371 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp
);
10373 bool kvm_vcpu_is_bsp(struct kvm_vcpu
*vcpu
)
10375 return (vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_BSP
) != 0;
10378 struct static_key kvm_no_apic_vcpu __read_mostly
;
10379 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu
);
10381 void kvm_arch_sched_in(struct kvm_vcpu
*vcpu
, int cpu
)
10383 struct kvm_pmu
*pmu
= vcpu_to_pmu(vcpu
);
10385 vcpu
->arch
.l1tf_flush_l1d
= true;
10386 if (pmu
->version
&& unlikely(pmu
->event_count
)) {
10387 pmu
->need_cleanup
= true;
10388 kvm_make_request(KVM_REQ_PMU
, vcpu
);
10390 kvm_x86_ops
.sched_in(vcpu
, cpu
);
10393 void kvm_arch_free_vm(struct kvm
*kvm
)
10395 kfree(kvm
->arch
.hyperv
.hv_pa_pg
);
10400 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
10405 INIT_HLIST_HEAD(&kvm
->arch
.mask_notifier_list
);
10406 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
10407 INIT_LIST_HEAD(&kvm
->arch
.zapped_obsolete_pages
);
10408 INIT_LIST_HEAD(&kvm
->arch
.lpage_disallowed_mmu_pages
);
10409 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
10410 atomic_set(&kvm
->arch
.noncoherent_dma_count
, 0);
10412 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10413 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
10414 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10415 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID
,
10416 &kvm
->arch
.irq_sources_bitmap
);
10418 raw_spin_lock_init(&kvm
->arch
.tsc_write_lock
);
10419 mutex_init(&kvm
->arch
.apic_map_lock
);
10420 spin_lock_init(&kvm
->arch
.pvclock_gtod_sync_lock
);
10422 kvm
->arch
.kvmclock_offset
= -get_kvmclock_base_ns();
10423 pvclock_update_vm_gtod_copy(kvm
);
10425 kvm
->arch
.guest_can_read_msr_platform_info
= true;
10427 INIT_DELAYED_WORK(&kvm
->arch
.kvmclock_update_work
, kvmclock_update_fn
);
10428 INIT_DELAYED_WORK(&kvm
->arch
.kvmclock_sync_work
, kvmclock_sync_fn
);
10430 kvm_hv_init_vm(kvm
);
10431 kvm_page_track_init(kvm
);
10432 kvm_mmu_init_vm(kvm
);
10434 return kvm_x86_ops
.vm_init(kvm
);
10437 int kvm_arch_post_init_vm(struct kvm
*kvm
)
10439 return kvm_mmu_post_init_vm(kvm
);
10442 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
10445 kvm_mmu_unload(vcpu
);
10449 static void kvm_free_vcpus(struct kvm
*kvm
)
10452 struct kvm_vcpu
*vcpu
;
10455 * Unpin any mmu pages first.
10457 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
10458 kvm_clear_async_pf_completion_queue(vcpu
);
10459 kvm_unload_vcpu_mmu(vcpu
);
10461 kvm_for_each_vcpu(i
, vcpu
, kvm
)
10462 kvm_vcpu_destroy(vcpu
);
10464 mutex_lock(&kvm
->lock
);
10465 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
10466 kvm
->vcpus
[i
] = NULL
;
10468 atomic_set(&kvm
->online_vcpus
, 0);
10469 mutex_unlock(&kvm
->lock
);
10472 void kvm_arch_sync_events(struct kvm
*kvm
)
10474 cancel_delayed_work_sync(&kvm
->arch
.kvmclock_sync_work
);
10475 cancel_delayed_work_sync(&kvm
->arch
.kvmclock_update_work
);
10479 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
10482 * __x86_set_memory_region: Setup KVM internal memory slot
10484 * @kvm: the kvm pointer to the VM.
10485 * @id: the slot ID to setup.
10486 * @gpa: the GPA to install the slot (unused when @size == 0).
10487 * @size: the size of the slot. Set to zero to uninstall a slot.
10489 * This function helps to setup a KVM internal memory slot. Specify
10490 * @size > 0 to install a new slot, while @size == 0 to uninstall a
10491 * slot. The return code can be one of the following:
10493 * HVA: on success (uninstall will return a bogus HVA)
10496 * The caller should always use IS_ERR() to check the return value
10497 * before use. Note, the KVM internal memory slots are guaranteed to
10498 * remain valid and unchanged until the VM is destroyed, i.e., the
10499 * GPA->HVA translation will not change. However, the HVA is a user
10500 * address, i.e. its accessibility is not guaranteed, and must be
10501 * accessed via __copy_{to,from}_user().
10503 void __user
* __x86_set_memory_region(struct kvm
*kvm
, int id
, gpa_t gpa
,
10507 unsigned long hva
, old_npages
;
10508 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
10509 struct kvm_memory_slot
*slot
;
10511 /* Called with kvm->slots_lock held. */
10512 if (WARN_ON(id
>= KVM_MEM_SLOTS_NUM
))
10513 return ERR_PTR_USR(-EINVAL
);
10515 slot
= id_to_memslot(slots
, id
);
10517 if (slot
&& slot
->npages
)
10518 return ERR_PTR_USR(-EEXIST
);
10521 * MAP_SHARED to prevent internal slot pages from being moved
10524 hva
= vm_mmap(NULL
, 0, size
, PROT_READ
| PROT_WRITE
,
10525 MAP_SHARED
| MAP_ANONYMOUS
, 0);
10526 if (IS_ERR((void *)hva
))
10527 return (void __user
*)hva
;
10529 if (!slot
|| !slot
->npages
)
10532 old_npages
= slot
->npages
;
10533 hva
= slot
->userspace_addr
;
10536 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
10537 struct kvm_userspace_memory_region m
;
10539 m
.slot
= id
| (i
<< 16);
10541 m
.guest_phys_addr
= gpa
;
10542 m
.userspace_addr
= hva
;
10543 m
.memory_size
= size
;
10544 r
= __kvm_set_memory_region(kvm
, &m
);
10546 return ERR_PTR_USR(r
);
10550 vm_munmap(hva
, old_npages
* PAGE_SIZE
);
10552 return (void __user
*)hva
;
10554 EXPORT_SYMBOL_GPL(__x86_set_memory_region
);
10556 void kvm_arch_pre_destroy_vm(struct kvm
*kvm
)
10558 kvm_mmu_pre_destroy_vm(kvm
);
10561 void kvm_arch_destroy_vm(struct kvm
*kvm
)
10563 if (current
->mm
== kvm
->mm
) {
10565 * Free memory regions allocated on behalf of userspace,
10566 * unless the the memory map has changed due to process exit
10569 mutex_lock(&kvm
->slots_lock
);
10570 __x86_set_memory_region(kvm
, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
,
10572 __x86_set_memory_region(kvm
, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT
,
10574 __x86_set_memory_region(kvm
, TSS_PRIVATE_MEMSLOT
, 0, 0);
10575 mutex_unlock(&kvm
->slots_lock
);
10577 if (kvm_x86_ops
.vm_destroy
)
10578 kvm_x86_ops
.vm_destroy(kvm
);
10579 kvm_free_msr_filter(srcu_dereference_check(kvm
->arch
.msr_filter
, &kvm
->srcu
, 1));
10580 kvm_pic_destroy(kvm
);
10581 kvm_ioapic_destroy(kvm
);
10582 kvm_free_vcpus(kvm
);
10583 kvfree(rcu_dereference_check(kvm
->arch
.apic_map
, 1));
10584 kfree(srcu_dereference_check(kvm
->arch
.pmu_event_filter
, &kvm
->srcu
, 1));
10585 kvm_mmu_uninit_vm(kvm
);
10586 kvm_page_track_cleanup(kvm
);
10587 kvm_hv_destroy_vm(kvm
);
10590 void kvm_arch_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
)
10594 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
10595 kvfree(slot
->arch
.rmap
[i
]);
10596 slot
->arch
.rmap
[i
] = NULL
;
10601 kvfree(slot
->arch
.lpage_info
[i
- 1]);
10602 slot
->arch
.lpage_info
[i
- 1] = NULL
;
10605 kvm_page_track_free_memslot(slot
);
10608 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot
*slot
,
10609 unsigned long npages
)
10614 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10615 * old arrays will be freed by __kvm_set_memory_region() if installing
10616 * the new memslot is successful.
10618 memset(&slot
->arch
, 0, sizeof(slot
->arch
));
10620 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
10621 struct kvm_lpage_info
*linfo
;
10622 unsigned long ugfn
;
10626 lpages
= gfn_to_index(slot
->base_gfn
+ npages
- 1,
10627 slot
->base_gfn
, level
) + 1;
10629 slot
->arch
.rmap
[i
] =
10630 kvcalloc(lpages
, sizeof(*slot
->arch
.rmap
[i
]),
10631 GFP_KERNEL_ACCOUNT
);
10632 if (!slot
->arch
.rmap
[i
])
10637 linfo
= kvcalloc(lpages
, sizeof(*linfo
), GFP_KERNEL_ACCOUNT
);
10641 slot
->arch
.lpage_info
[i
- 1] = linfo
;
10643 if (slot
->base_gfn
& (KVM_PAGES_PER_HPAGE(level
) - 1))
10644 linfo
[0].disallow_lpage
= 1;
10645 if ((slot
->base_gfn
+ npages
) & (KVM_PAGES_PER_HPAGE(level
) - 1))
10646 linfo
[lpages
- 1].disallow_lpage
= 1;
10647 ugfn
= slot
->userspace_addr
>> PAGE_SHIFT
;
10649 * If the gfn and userspace address are not aligned wrt each
10650 * other, disable large page support for this slot.
10652 if ((slot
->base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE(level
) - 1)) {
10655 for (j
= 0; j
< lpages
; ++j
)
10656 linfo
[j
].disallow_lpage
= 1;
10660 if (kvm_page_track_create_memslot(slot
, npages
))
10666 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
10667 kvfree(slot
->arch
.rmap
[i
]);
10668 slot
->arch
.rmap
[i
] = NULL
;
10672 kvfree(slot
->arch
.lpage_info
[i
- 1]);
10673 slot
->arch
.lpage_info
[i
- 1] = NULL
;
10678 void kvm_arch_memslots_updated(struct kvm
*kvm
, u64 gen
)
10680 struct kvm_vcpu
*vcpu
;
10684 * memslots->generation has been incremented.
10685 * mmio generation may have reached its maximum value.
10687 kvm_mmu_invalidate_mmio_sptes(kvm
, gen
);
10689 /* Force re-initialization of steal_time cache */
10690 kvm_for_each_vcpu(i
, vcpu
, kvm
)
10691 kvm_vcpu_kick(vcpu
);
10694 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
10695 struct kvm_memory_slot
*memslot
,
10696 const struct kvm_userspace_memory_region
*mem
,
10697 enum kvm_mr_change change
)
10699 if (change
== KVM_MR_CREATE
|| change
== KVM_MR_MOVE
)
10700 return kvm_alloc_memslot_metadata(memslot
,
10701 mem
->memory_size
>> PAGE_SHIFT
);
10705 static void kvm_mmu_slot_apply_flags(struct kvm
*kvm
,
10706 struct kvm_memory_slot
*old
,
10707 struct kvm_memory_slot
*new,
10708 enum kvm_mr_change change
)
10711 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
10712 * See comments below.
10714 if ((change
!= KVM_MR_FLAGS_ONLY
) || (new->flags
& KVM_MEM_READONLY
))
10718 * Dirty logging tracks sptes in 4k granularity, meaning that large
10719 * sptes have to be split. If live migration is successful, the guest
10720 * in the source machine will be destroyed and large sptes will be
10721 * created in the destination. However, if the guest continues to run
10722 * in the source machine (for example if live migration fails), small
10723 * sptes will remain around and cause bad performance.
10725 * Scan sptes if dirty logging has been stopped, dropping those
10726 * which can be collapsed into a single large-page spte. Later
10727 * page faults will create the large-page sptes.
10729 * There is no need to do this in any of the following cases:
10730 * CREATE: No dirty mappings will already exist.
10731 * MOVE/DELETE: The old mappings will already have been cleaned up by
10732 * kvm_arch_flush_shadow_memslot()
10734 if ((old
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) &&
10735 !(new->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
10736 kvm_mmu_zap_collapsible_sptes(kvm
, new);
10739 * Enable or disable dirty logging for the slot.
10741 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old
10742 * slot have been zapped so no dirty logging updates are needed for
10744 * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible
10745 * any mappings that might be created in it will consume the
10746 * properties of the new slot and do not need to be updated here.
10748 * When PML is enabled, the kvm_x86_ops dirty logging hooks are
10749 * called to enable/disable dirty logging.
10751 * When disabling dirty logging with PML enabled, the D-bit is set
10752 * for sptes in the slot in order to prevent unnecessary GPA
10753 * logging in the PML buffer (and potential PML buffer full VMEXIT).
10754 * This guarantees leaving PML enabled for the guest's lifetime
10755 * won't have any additional overhead from PML when the guest is
10756 * running with dirty logging disabled.
10758 * When enabling dirty logging, large sptes are write-protected
10759 * so they can be split on first write. New large sptes cannot
10760 * be created for this slot until the end of the logging.
10761 * See the comments in fast_page_fault().
10762 * For small sptes, nothing is done if the dirty log is in the
10763 * initial-all-set state. Otherwise, depending on whether pml
10764 * is enabled the D-bit or the W-bit will be cleared.
10766 if (new->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
10767 if (kvm_x86_ops
.slot_enable_log_dirty
) {
10768 kvm_x86_ops
.slot_enable_log_dirty(kvm
, new);
10771 kvm_dirty_log_manual_protect_and_init_set(kvm
) ?
10772 PG_LEVEL_2M
: PG_LEVEL_4K
;
10775 * If we're with initial-all-set, we don't need
10776 * to write protect any small page because
10777 * they're reported as dirty already. However
10778 * we still need to write-protect huge pages
10779 * so that the page split can happen lazily on
10780 * the first write to the huge page.
10782 kvm_mmu_slot_remove_write_access(kvm
, new, level
);
10785 if (kvm_x86_ops
.slot_disable_log_dirty
)
10786 kvm_x86_ops
.slot_disable_log_dirty(kvm
, new);
10790 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
10791 const struct kvm_userspace_memory_region
*mem
,
10792 struct kvm_memory_slot
*old
,
10793 const struct kvm_memory_slot
*new,
10794 enum kvm_mr_change change
)
10796 if (!kvm
->arch
.n_requested_mmu_pages
)
10797 kvm_mmu_change_mmu_pages(kvm
,
10798 kvm_mmu_calculate_default_mmu_pages(kvm
));
10801 * FIXME: const-ify all uses of struct kvm_memory_slot.
10803 kvm_mmu_slot_apply_flags(kvm
, old
, (struct kvm_memory_slot
*) new, change
);
10805 /* Free the arrays associated with the old memslot. */
10806 if (change
== KVM_MR_MOVE
)
10807 kvm_arch_free_memslot(kvm
, old
);
10810 void kvm_arch_flush_shadow_all(struct kvm
*kvm
)
10812 kvm_mmu_zap_all(kvm
);
10815 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
10816 struct kvm_memory_slot
*slot
)
10818 kvm_page_track_flush_slot(kvm
, slot
);
10821 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu
*vcpu
)
10823 return (is_guest_mode(vcpu
) &&
10824 kvm_x86_ops
.guest_apic_has_interrupt
&&
10825 kvm_x86_ops
.guest_apic_has_interrupt(vcpu
));
10828 static inline bool kvm_vcpu_has_events(struct kvm_vcpu
*vcpu
)
10830 if (!list_empty_careful(&vcpu
->async_pf
.done
))
10833 if (kvm_apic_has_events(vcpu
))
10836 if (vcpu
->arch
.pv
.pv_unhalted
)
10839 if (vcpu
->arch
.exception
.pending
)
10842 if (kvm_test_request(KVM_REQ_NMI
, vcpu
) ||
10843 (vcpu
->arch
.nmi_pending
&&
10844 kvm_x86_ops
.nmi_allowed(vcpu
, false)))
10847 if (kvm_test_request(KVM_REQ_SMI
, vcpu
) ||
10848 (vcpu
->arch
.smi_pending
&&
10849 kvm_x86_ops
.smi_allowed(vcpu
, false)))
10852 if (kvm_arch_interrupt_allowed(vcpu
) &&
10853 (kvm_cpu_has_interrupt(vcpu
) ||
10854 kvm_guest_apic_has_interrupt(vcpu
)))
10857 if (kvm_hv_has_stimer_pending(vcpu
))
10860 if (is_guest_mode(vcpu
) &&
10861 kvm_x86_ops
.nested_ops
->hv_timer_pending
&&
10862 kvm_x86_ops
.nested_ops
->hv_timer_pending(vcpu
))
10868 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
10870 return kvm_vcpu_running(vcpu
) || kvm_vcpu_has_events(vcpu
);
10873 bool kvm_arch_dy_runnable(struct kvm_vcpu
*vcpu
)
10875 if (READ_ONCE(vcpu
->arch
.pv
.pv_unhalted
))
10878 if (kvm_test_request(KVM_REQ_NMI
, vcpu
) ||
10879 kvm_test_request(KVM_REQ_SMI
, vcpu
) ||
10880 kvm_test_request(KVM_REQ_EVENT
, vcpu
))
10883 if (vcpu
->arch
.apicv_active
&& kvm_x86_ops
.dy_apicv_has_pending_interrupt(vcpu
))
10889 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu
*vcpu
)
10891 return vcpu
->arch
.preempted_in_kernel
;
10894 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
10896 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
10899 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
10901 return kvm_x86_ops
.interrupt_allowed(vcpu
, false);
10904 unsigned long kvm_get_linear_rip(struct kvm_vcpu
*vcpu
)
10906 /* Can't read the RIP when guest state is protected, just return 0 */
10907 if (vcpu
->arch
.guest_state_protected
)
10910 if (is_64_bit_mode(vcpu
))
10911 return kvm_rip_read(vcpu
);
10912 return (u32
)(get_segment_base(vcpu
, VCPU_SREG_CS
) +
10913 kvm_rip_read(vcpu
));
10915 EXPORT_SYMBOL_GPL(kvm_get_linear_rip
);
10917 bool kvm_is_linear_rip(struct kvm_vcpu
*vcpu
, unsigned long linear_rip
)
10919 return kvm_get_linear_rip(vcpu
) == linear_rip
;
10921 EXPORT_SYMBOL_GPL(kvm_is_linear_rip
);
10923 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
10925 unsigned long rflags
;
10927 rflags
= kvm_x86_ops
.get_rflags(vcpu
);
10928 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
10929 rflags
&= ~X86_EFLAGS_TF
;
10932 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
10934 static void __kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
10936 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
10937 kvm_is_linear_rip(vcpu
, vcpu
->arch
.singlestep_rip
))
10938 rflags
|= X86_EFLAGS_TF
;
10939 kvm_x86_ops
.set_rflags(vcpu
, rflags
);
10942 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
10944 __kvm_set_rflags(vcpu
, rflags
);
10945 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
10947 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
10949 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
, struct kvm_async_pf
*work
)
10953 if ((vcpu
->arch
.mmu
->direct_map
!= work
->arch
.direct_map
) ||
10957 r
= kvm_mmu_reload(vcpu
);
10961 if (!vcpu
->arch
.mmu
->direct_map
&&
10962 work
->arch
.cr3
!= vcpu
->arch
.mmu
->get_guest_pgd(vcpu
))
10965 kvm_mmu_do_page_fault(vcpu
, work
->cr2_or_gpa
, 0, true);
10968 static inline u32
kvm_async_pf_hash_fn(gfn_t gfn
)
10970 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU
));
10972 return hash_32(gfn
& 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU
));
10975 static inline u32
kvm_async_pf_next_probe(u32 key
)
10977 return (key
+ 1) & (ASYNC_PF_PER_VCPU
- 1);
10980 static void kvm_add_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
10982 u32 key
= kvm_async_pf_hash_fn(gfn
);
10984 while (vcpu
->arch
.apf
.gfns
[key
] != ~0)
10985 key
= kvm_async_pf_next_probe(key
);
10987 vcpu
->arch
.apf
.gfns
[key
] = gfn
;
10990 static u32
kvm_async_pf_gfn_slot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
10993 u32 key
= kvm_async_pf_hash_fn(gfn
);
10995 for (i
= 0; i
< ASYNC_PF_PER_VCPU
&&
10996 (vcpu
->arch
.apf
.gfns
[key
] != gfn
&&
10997 vcpu
->arch
.apf
.gfns
[key
] != ~0); i
++)
10998 key
= kvm_async_pf_next_probe(key
);
11003 bool kvm_find_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
11005 return vcpu
->arch
.apf
.gfns
[kvm_async_pf_gfn_slot(vcpu
, gfn
)] == gfn
;
11008 static void kvm_del_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
11012 i
= j
= kvm_async_pf_gfn_slot(vcpu
, gfn
);
11014 if (WARN_ON_ONCE(vcpu
->arch
.apf
.gfns
[i
] != gfn
))
11018 vcpu
->arch
.apf
.gfns
[i
] = ~0;
11020 j
= kvm_async_pf_next_probe(j
);
11021 if (vcpu
->arch
.apf
.gfns
[j
] == ~0)
11023 k
= kvm_async_pf_hash_fn(vcpu
->arch
.apf
.gfns
[j
]);
11025 * k lies cyclically in ]i,j]
11027 * |....j i.k.| or |.k..j i...|
11029 } while ((i
<= j
) ? (i
< k
&& k
<= j
) : (i
< k
|| k
<= j
));
11030 vcpu
->arch
.apf
.gfns
[i
] = vcpu
->arch
.apf
.gfns
[j
];
11035 static inline int apf_put_user_notpresent(struct kvm_vcpu
*vcpu
)
11037 u32 reason
= KVM_PV_REASON_PAGE_NOT_PRESENT
;
11039 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, &reason
,
11043 static inline int apf_put_user_ready(struct kvm_vcpu
*vcpu
, u32 token
)
11045 unsigned int offset
= offsetof(struct kvm_vcpu_pv_apf_data
, token
);
11047 return kvm_write_guest_offset_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
,
11048 &token
, offset
, sizeof(token
));
11051 static inline bool apf_pageready_slot_free(struct kvm_vcpu
*vcpu
)
11053 unsigned int offset
= offsetof(struct kvm_vcpu_pv_apf_data
, token
);
11056 if (kvm_read_guest_offset_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
,
11057 &val
, offset
, sizeof(val
)))
11063 static bool kvm_can_deliver_async_pf(struct kvm_vcpu
*vcpu
)
11065 if (!vcpu
->arch
.apf
.delivery_as_pf_vmexit
&& is_guest_mode(vcpu
))
11068 if (!kvm_pv_async_pf_enabled(vcpu
) ||
11069 (vcpu
->arch
.apf
.send_user_only
&& kvm_x86_ops
.get_cpl(vcpu
) == 0))
11075 bool kvm_can_do_async_pf(struct kvm_vcpu
*vcpu
)
11077 if (unlikely(!lapic_in_kernel(vcpu
) ||
11078 kvm_event_needs_reinjection(vcpu
) ||
11079 vcpu
->arch
.exception
.pending
))
11082 if (kvm_hlt_in_guest(vcpu
->kvm
) && !kvm_can_deliver_async_pf(vcpu
))
11086 * If interrupts are off we cannot even use an artificial
11089 return kvm_arch_interrupt_allowed(vcpu
);
11092 bool kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
11093 struct kvm_async_pf
*work
)
11095 struct x86_exception fault
;
11097 trace_kvm_async_pf_not_present(work
->arch
.token
, work
->cr2_or_gpa
);
11098 kvm_add_async_pf_gfn(vcpu
, work
->arch
.gfn
);
11100 if (kvm_can_deliver_async_pf(vcpu
) &&
11101 !apf_put_user_notpresent(vcpu
)) {
11102 fault
.vector
= PF_VECTOR
;
11103 fault
.error_code_valid
= true;
11104 fault
.error_code
= 0;
11105 fault
.nested_page_fault
= false;
11106 fault
.address
= work
->arch
.token
;
11107 fault
.async_page_fault
= true;
11108 kvm_inject_page_fault(vcpu
, &fault
);
11112 * It is not possible to deliver a paravirtualized asynchronous
11113 * page fault, but putting the guest in an artificial halt state
11114 * can be beneficial nevertheless: if an interrupt arrives, we
11115 * can deliver it timely and perhaps the guest will schedule
11116 * another process. When the instruction that triggered a page
11117 * fault is retried, hopefully the page will be ready in the host.
11119 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
11124 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
11125 struct kvm_async_pf
*work
)
11127 struct kvm_lapic_irq irq
= {
11128 .delivery_mode
= APIC_DM_FIXED
,
11129 .vector
= vcpu
->arch
.apf
.vec
11132 if (work
->wakeup_all
)
11133 work
->arch
.token
= ~0; /* broadcast wakeup */
11135 kvm_del_async_pf_gfn(vcpu
, work
->arch
.gfn
);
11136 trace_kvm_async_pf_ready(work
->arch
.token
, work
->cr2_or_gpa
);
11138 if ((work
->wakeup_all
|| work
->notpresent_injected
) &&
11139 kvm_pv_async_pf_enabled(vcpu
) &&
11140 !apf_put_user_ready(vcpu
, work
->arch
.token
)) {
11141 vcpu
->arch
.apf
.pageready_pending
= true;
11142 kvm_apic_set_irq(vcpu
, &irq
, NULL
);
11145 vcpu
->arch
.apf
.halted
= false;
11146 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
11149 void kvm_arch_async_page_present_queued(struct kvm_vcpu
*vcpu
)
11151 kvm_make_request(KVM_REQ_APF_READY
, vcpu
);
11152 if (!vcpu
->arch
.apf
.pageready_pending
)
11153 kvm_vcpu_kick(vcpu
);
11156 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu
*vcpu
)
11158 if (!kvm_pv_async_pf_enabled(vcpu
))
11161 return apf_pageready_slot_free(vcpu
);
11164 void kvm_arch_start_assignment(struct kvm
*kvm
)
11166 atomic_inc(&kvm
->arch
.assigned_device_count
);
11168 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment
);
11170 void kvm_arch_end_assignment(struct kvm
*kvm
)
11172 atomic_dec(&kvm
->arch
.assigned_device_count
);
11174 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment
);
11176 bool kvm_arch_has_assigned_device(struct kvm
*kvm
)
11178 return atomic_read(&kvm
->arch
.assigned_device_count
);
11180 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device
);
11182 void kvm_arch_register_noncoherent_dma(struct kvm
*kvm
)
11184 atomic_inc(&kvm
->arch
.noncoherent_dma_count
);
11186 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma
);
11188 void kvm_arch_unregister_noncoherent_dma(struct kvm
*kvm
)
11190 atomic_dec(&kvm
->arch
.noncoherent_dma_count
);
11192 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma
);
11194 bool kvm_arch_has_noncoherent_dma(struct kvm
*kvm
)
11196 return atomic_read(&kvm
->arch
.noncoherent_dma_count
);
11198 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma
);
11200 bool kvm_arch_has_irq_bypass(void)
11205 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer
*cons
,
11206 struct irq_bypass_producer
*prod
)
11208 struct kvm_kernel_irqfd
*irqfd
=
11209 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
11212 irqfd
->producer
= prod
;
11213 kvm_arch_start_assignment(irqfd
->kvm
);
11214 ret
= kvm_x86_ops
.update_pi_irte(irqfd
->kvm
,
11215 prod
->irq
, irqfd
->gsi
, 1);
11218 kvm_arch_end_assignment(irqfd
->kvm
);
11223 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer
*cons
,
11224 struct irq_bypass_producer
*prod
)
11227 struct kvm_kernel_irqfd
*irqfd
=
11228 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
11230 WARN_ON(irqfd
->producer
!= prod
);
11231 irqfd
->producer
= NULL
;
11234 * When producer of consumer is unregistered, we change back to
11235 * remapped mode, so we can re-use the current implementation
11236 * when the irq is masked/disabled or the consumer side (KVM
11237 * int this case doesn't want to receive the interrupts.
11239 ret
= kvm_x86_ops
.update_pi_irte(irqfd
->kvm
, prod
->irq
, irqfd
->gsi
, 0);
11241 printk(KERN_INFO
"irq bypass consumer (token %p) unregistration"
11242 " fails: %d\n", irqfd
->consumer
.token
, ret
);
11244 kvm_arch_end_assignment(irqfd
->kvm
);
11247 int kvm_arch_update_irqfd_routing(struct kvm
*kvm
, unsigned int host_irq
,
11248 uint32_t guest_irq
, bool set
)
11250 return kvm_x86_ops
.update_pi_irte(kvm
, host_irq
, guest_irq
, set
);
11253 bool kvm_vector_hashing_enabled(void)
11255 return vector_hashing
;
11258 bool kvm_arch_no_poll(struct kvm_vcpu
*vcpu
)
11260 return (vcpu
->arch
.msr_kvm_poll_control
& 1) == 0;
11262 EXPORT_SYMBOL_GPL(kvm_arch_no_poll
);
11265 int kvm_spec_ctrl_test_value(u64 value
)
11268 * test that setting IA32_SPEC_CTRL to given value
11269 * is allowed by the host processor
11273 unsigned long flags
;
11276 local_irq_save(flags
);
11278 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL
, &saved_value
))
11280 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL
, value
))
11283 wrmsrl(MSR_IA32_SPEC_CTRL
, saved_value
);
11285 local_irq_restore(flags
);
11289 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value
);
11291 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu
*vcpu
, gva_t gva
, u16 error_code
)
11293 struct x86_exception fault
;
11294 u32 access
= error_code
&
11295 (PFERR_WRITE_MASK
| PFERR_FETCH_MASK
| PFERR_USER_MASK
);
11297 if (!(error_code
& PFERR_PRESENT_MASK
) ||
11298 vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, &fault
) != UNMAPPED_GVA
) {
11300 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11301 * tables probably do not match the TLB. Just proceed
11302 * with the error code that the processor gave.
11304 fault
.vector
= PF_VECTOR
;
11305 fault
.error_code_valid
= true;
11306 fault
.error_code
= error_code
;
11307 fault
.nested_page_fault
= false;
11308 fault
.address
= gva
;
11310 vcpu
->arch
.walk_mmu
->inject_page_fault(vcpu
, &fault
);
11312 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error
);
11315 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11316 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11317 * indicates whether exit to userspace is needed.
11319 int kvm_handle_memory_failure(struct kvm_vcpu
*vcpu
, int r
,
11320 struct x86_exception
*e
)
11322 if (r
== X86EMUL_PROPAGATE_FAULT
) {
11323 kvm_inject_emulated_page_fault(vcpu
, e
);
11328 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11329 * while handling a VMX instruction KVM could've handled the request
11330 * correctly by exiting to userspace and performing I/O but there
11331 * doesn't seem to be a real use-case behind such requests, just return
11332 * KVM_EXIT_INTERNAL_ERROR for now.
11334 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
11335 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
11336 vcpu
->run
->internal
.ndata
= 0;
11340 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure
);
11342 int kvm_handle_invpcid(struct kvm_vcpu
*vcpu
, unsigned long type
, gva_t gva
)
11345 struct x86_exception e
;
11347 unsigned long roots_to_free
= 0;
11354 r
= kvm_read_guest_virt(vcpu
, gva
, &operand
, sizeof(operand
), &e
);
11355 if (r
!= X86EMUL_CONTINUE
)
11356 return kvm_handle_memory_failure(vcpu
, r
, &e
);
11358 if (operand
.pcid
>> 12 != 0) {
11359 kvm_inject_gp(vcpu
, 0);
11363 pcid_enabled
= kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
);
11366 case INVPCID_TYPE_INDIV_ADDR
:
11367 if ((!pcid_enabled
&& (operand
.pcid
!= 0)) ||
11368 is_noncanonical_address(operand
.gla
, vcpu
)) {
11369 kvm_inject_gp(vcpu
, 0);
11372 kvm_mmu_invpcid_gva(vcpu
, operand
.gla
, operand
.pcid
);
11373 return kvm_skip_emulated_instruction(vcpu
);
11375 case INVPCID_TYPE_SINGLE_CTXT
:
11376 if (!pcid_enabled
&& (operand
.pcid
!= 0)) {
11377 kvm_inject_gp(vcpu
, 0);
11381 if (kvm_get_active_pcid(vcpu
) == operand
.pcid
) {
11382 kvm_mmu_sync_roots(vcpu
);
11383 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT
, vcpu
);
11386 for (i
= 0; i
< KVM_MMU_NUM_PREV_ROOTS
; i
++)
11387 if (kvm_get_pcid(vcpu
, vcpu
->arch
.mmu
->prev_roots
[i
].pgd
)
11389 roots_to_free
|= KVM_MMU_ROOT_PREVIOUS(i
);
11391 kvm_mmu_free_roots(vcpu
, vcpu
->arch
.mmu
, roots_to_free
);
11393 * If neither the current cr3 nor any of the prev_roots use the
11394 * given PCID, then nothing needs to be done here because a
11395 * resync will happen anyway before switching to any other CR3.
11398 return kvm_skip_emulated_instruction(vcpu
);
11400 case INVPCID_TYPE_ALL_NON_GLOBAL
:
11402 * Currently, KVM doesn't mark global entries in the shadow
11403 * page tables, so a non-global flush just degenerates to a
11404 * global flush. If needed, we could optimize this later by
11405 * keeping track of global entries in shadow page tables.
11409 case INVPCID_TYPE_ALL_INCL_GLOBAL
:
11410 kvm_mmu_unload(vcpu
);
11411 return kvm_skip_emulated_instruction(vcpu
);
11414 BUG(); /* We have already checked above that type <= 3 */
11417 EXPORT_SYMBOL_GPL(kvm_handle_invpcid
);
11419 static int complete_sev_es_emulated_mmio(struct kvm_vcpu
*vcpu
)
11421 struct kvm_run
*run
= vcpu
->run
;
11422 struct kvm_mmio_fragment
*frag
;
11425 BUG_ON(!vcpu
->mmio_needed
);
11427 /* Complete previous fragment */
11428 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_cur_fragment
];
11429 len
= min(8u, frag
->len
);
11430 if (!vcpu
->mmio_is_write
)
11431 memcpy(frag
->data
, run
->mmio
.data
, len
);
11433 if (frag
->len
<= 8) {
11434 /* Switch to the next fragment. */
11436 vcpu
->mmio_cur_fragment
++;
11438 /* Go forward to the next mmio piece. */
11444 if (vcpu
->mmio_cur_fragment
>= vcpu
->mmio_nr_fragments
) {
11445 vcpu
->mmio_needed
= 0;
11447 // VMG change, at this point, we're always done
11448 // RIP has already been advanced
11452 // More MMIO is needed
11453 run
->mmio
.phys_addr
= frag
->gpa
;
11454 run
->mmio
.len
= min(8u, frag
->len
);
11455 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
11456 if (run
->mmio
.is_write
)
11457 memcpy(run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
11458 run
->exit_reason
= KVM_EXIT_MMIO
;
11460 vcpu
->arch
.complete_userspace_io
= complete_sev_es_emulated_mmio
;
11465 int kvm_sev_es_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
, unsigned int bytes
,
11469 struct kvm_mmio_fragment
*frag
;
11474 handled
= write_emultor
.read_write_mmio(vcpu
, gpa
, bytes
, data
);
11475 if (handled
== bytes
)
11482 /*TODO: Check if need to increment number of frags */
11483 frag
= vcpu
->mmio_fragments
;
11484 vcpu
->mmio_nr_fragments
= 1;
11489 vcpu
->mmio_needed
= 1;
11490 vcpu
->mmio_cur_fragment
= 0;
11492 vcpu
->run
->mmio
.phys_addr
= gpa
;
11493 vcpu
->run
->mmio
.len
= min(8u, frag
->len
);
11494 vcpu
->run
->mmio
.is_write
= 1;
11495 memcpy(vcpu
->run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
11496 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
11498 vcpu
->arch
.complete_userspace_io
= complete_sev_es_emulated_mmio
;
11502 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write
);
11504 int kvm_sev_es_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t gpa
, unsigned int bytes
,
11508 struct kvm_mmio_fragment
*frag
;
11513 handled
= read_emultor
.read_write_mmio(vcpu
, gpa
, bytes
, data
);
11514 if (handled
== bytes
)
11521 /*TODO: Check if need to increment number of frags */
11522 frag
= vcpu
->mmio_fragments
;
11523 vcpu
->mmio_nr_fragments
= 1;
11528 vcpu
->mmio_needed
= 1;
11529 vcpu
->mmio_cur_fragment
= 0;
11531 vcpu
->run
->mmio
.phys_addr
= gpa
;
11532 vcpu
->run
->mmio
.len
= min(8u, frag
->len
);
11533 vcpu
->run
->mmio
.is_write
= 0;
11534 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
11536 vcpu
->arch
.complete_userspace_io
= complete_sev_es_emulated_mmio
;
11540 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read
);
11542 static int complete_sev_es_emulated_ins(struct kvm_vcpu
*vcpu
)
11544 memcpy(vcpu
->arch
.guest_ins_data
, vcpu
->arch
.pio_data
,
11545 vcpu
->arch
.pio
.count
* vcpu
->arch
.pio
.size
);
11546 vcpu
->arch
.pio
.count
= 0;
11551 static int kvm_sev_es_outs(struct kvm_vcpu
*vcpu
, unsigned int size
,
11552 unsigned int port
, void *data
, unsigned int count
)
11556 ret
= emulator_pio_out_emulated(vcpu
->arch
.emulate_ctxt
, size
, port
,
11561 vcpu
->arch
.pio
.count
= 0;
11566 static int kvm_sev_es_ins(struct kvm_vcpu
*vcpu
, unsigned int size
,
11567 unsigned int port
, void *data
, unsigned int count
)
11571 ret
= emulator_pio_in_emulated(vcpu
->arch
.emulate_ctxt
, size
, port
,
11574 vcpu
->arch
.pio
.count
= 0;
11576 vcpu
->arch
.guest_ins_data
= data
;
11577 vcpu
->arch
.complete_userspace_io
= complete_sev_es_emulated_ins
;
11583 int kvm_sev_es_string_io(struct kvm_vcpu
*vcpu
, unsigned int size
,
11584 unsigned int port
, void *data
, unsigned int count
,
11587 return in
? kvm_sev_es_ins(vcpu
, size
, port
, data
, count
)
11588 : kvm_sev_es_outs(vcpu
, size
, port
, data
, count
);
11590 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io
);
11592 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry
);
11593 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
11594 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio
);
11595 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
11596 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
11597 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
11598 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
11599 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
11600 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
11601 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
11602 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
11603 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed
);
11604 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
11605 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);
11606 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts
);
11607 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset
);
11608 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update
);
11609 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full
);
11610 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update
);
11611 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access
);
11612 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi
);
11613 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log
);
11614 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request
);
11615 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter
);
11616 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit
);
11617 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter
);
11618 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit
);