2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
59 #include <trace/events/kvm.h>
61 #include <asm/debugreg.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
71 #define CREATE_TRACE_POINTS
74 #define MAX_IO_MSRS 256
75 #define KVM_MAX_MCE_BANKS 32
76 u64 __read_mostly kvm_mce_cap_supported
= MCG_CTL_P
| MCG_SER_P
;
77 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported
);
79 #define emul_to_vcpu(ctxt) \
80 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
83 * - enable syscall per default because its emulated by KVM
84 * - enable LME and LMA per default on 64 bit KVM
88 u64 __read_mostly efer_reserved_bits
= ~((u64
)(EFER_SCE
| EFER_LME
| EFER_LMA
));
90 static u64 __read_mostly efer_reserved_bits
= ~((u64
)EFER_SCE
);
93 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
94 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
96 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
97 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
99 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
);
100 static void process_nmi(struct kvm_vcpu
*vcpu
);
101 static void enter_smm(struct kvm_vcpu
*vcpu
);
102 static void __kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
);
104 struct kvm_x86_ops
*kvm_x86_ops __read_mostly
;
105 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
107 static bool __read_mostly ignore_msrs
= 0;
108 module_param(ignore_msrs
, bool, S_IRUGO
| S_IWUSR
);
110 static bool __read_mostly report_ignored_msrs
= true;
111 module_param(report_ignored_msrs
, bool, S_IRUGO
| S_IWUSR
);
113 unsigned int min_timer_period_us
= 200;
114 module_param(min_timer_period_us
, uint
, S_IRUGO
| S_IWUSR
);
116 static bool __read_mostly kvmclock_periodic_sync
= true;
117 module_param(kvmclock_periodic_sync
, bool, S_IRUGO
);
119 bool __read_mostly kvm_has_tsc_control
;
120 EXPORT_SYMBOL_GPL(kvm_has_tsc_control
);
121 u32 __read_mostly kvm_max_guest_tsc_khz
;
122 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz
);
123 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits
;
124 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits
);
125 u64 __read_mostly kvm_max_tsc_scaling_ratio
;
126 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio
);
127 u64 __read_mostly kvm_default_tsc_scaling_ratio
;
128 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio
);
130 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
131 static u32 __read_mostly tsc_tolerance_ppm
= 250;
132 module_param(tsc_tolerance_ppm
, uint
, S_IRUGO
| S_IWUSR
);
134 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
135 unsigned int __read_mostly lapic_timer_advance_ns
= 0;
136 module_param(lapic_timer_advance_ns
, uint
, S_IRUGO
| S_IWUSR
);
138 static bool __read_mostly vector_hashing
= true;
139 module_param(vector_hashing
, bool, S_IRUGO
);
141 #define KVM_NR_SHARED_MSRS 16
143 struct kvm_shared_msrs_global
{
145 u32 msrs
[KVM_NR_SHARED_MSRS
];
148 struct kvm_shared_msrs
{
149 struct user_return_notifier urn
;
151 struct kvm_shared_msr_values
{
154 } values
[KVM_NR_SHARED_MSRS
];
157 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global
;
158 static struct kvm_shared_msrs __percpu
*shared_msrs
;
160 struct kvm_stats_debugfs_item debugfs_entries
[] = {
161 { "pf_fixed", VCPU_STAT(pf_fixed
) },
162 { "pf_guest", VCPU_STAT(pf_guest
) },
163 { "tlb_flush", VCPU_STAT(tlb_flush
) },
164 { "invlpg", VCPU_STAT(invlpg
) },
165 { "exits", VCPU_STAT(exits
) },
166 { "io_exits", VCPU_STAT(io_exits
) },
167 { "mmio_exits", VCPU_STAT(mmio_exits
) },
168 { "signal_exits", VCPU_STAT(signal_exits
) },
169 { "irq_window", VCPU_STAT(irq_window_exits
) },
170 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
171 { "halt_exits", VCPU_STAT(halt_exits
) },
172 { "halt_successful_poll", VCPU_STAT(halt_successful_poll
) },
173 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll
) },
174 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid
) },
175 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
176 { "hypercalls", VCPU_STAT(hypercalls
) },
177 { "request_irq", VCPU_STAT(request_irq_exits
) },
178 { "irq_exits", VCPU_STAT(irq_exits
) },
179 { "host_state_reload", VCPU_STAT(host_state_reload
) },
180 { "efer_reload", VCPU_STAT(efer_reload
) },
181 { "fpu_reload", VCPU_STAT(fpu_reload
) },
182 { "insn_emulation", VCPU_STAT(insn_emulation
) },
183 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
184 { "irq_injections", VCPU_STAT(irq_injections
) },
185 { "nmi_injections", VCPU_STAT(nmi_injections
) },
186 { "req_event", VCPU_STAT(req_event
) },
187 { "l1d_flush", VCPU_STAT(l1d_flush
) },
188 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
189 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
190 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
191 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
192 { "mmu_flooded", VM_STAT(mmu_flooded
) },
193 { "mmu_recycled", VM_STAT(mmu_recycled
) },
194 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
195 { "mmu_unsync", VM_STAT(mmu_unsync
) },
196 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
197 { "largepages", VM_STAT(lpages
) },
198 { "max_mmu_page_hash_collisions",
199 VM_STAT(max_mmu_page_hash_collisions
) },
203 u64 __read_mostly host_xcr0
;
205 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
);
207 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu
*vcpu
)
210 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
); i
++)
211 vcpu
->arch
.apf
.gfns
[i
] = ~0;
214 static void kvm_on_user_return(struct user_return_notifier
*urn
)
217 struct kvm_shared_msrs
*locals
218 = container_of(urn
, struct kvm_shared_msrs
, urn
);
219 struct kvm_shared_msr_values
*values
;
223 * Disabling irqs at this point since the following code could be
224 * interrupted and executed through kvm_arch_hardware_disable()
226 local_irq_save(flags
);
227 if (locals
->registered
) {
228 locals
->registered
= false;
229 user_return_notifier_unregister(urn
);
231 local_irq_restore(flags
);
232 for (slot
= 0; slot
< shared_msrs_global
.nr
; ++slot
) {
233 values
= &locals
->values
[slot
];
234 if (values
->host
!= values
->curr
) {
235 wrmsrl(shared_msrs_global
.msrs
[slot
], values
->host
);
236 values
->curr
= values
->host
;
241 static void shared_msr_update(unsigned slot
, u32 msr
)
244 unsigned int cpu
= smp_processor_id();
245 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
247 /* only read, and nobody should modify it at this time,
248 * so don't need lock */
249 if (slot
>= shared_msrs_global
.nr
) {
250 printk(KERN_ERR
"kvm: invalid MSR slot!");
253 rdmsrl_safe(msr
, &value
);
254 smsr
->values
[slot
].host
= value
;
255 smsr
->values
[slot
].curr
= value
;
258 void kvm_define_shared_msr(unsigned slot
, u32 msr
)
260 BUG_ON(slot
>= KVM_NR_SHARED_MSRS
);
261 shared_msrs_global
.msrs
[slot
] = msr
;
262 if (slot
>= shared_msrs_global
.nr
)
263 shared_msrs_global
.nr
= slot
+ 1;
265 EXPORT_SYMBOL_GPL(kvm_define_shared_msr
);
267 static void kvm_shared_msr_cpu_online(void)
271 for (i
= 0; i
< shared_msrs_global
.nr
; ++i
)
272 shared_msr_update(i
, shared_msrs_global
.msrs
[i
]);
275 int kvm_set_shared_msr(unsigned slot
, u64 value
, u64 mask
)
277 unsigned int cpu
= smp_processor_id();
278 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
281 if (((value
^ smsr
->values
[slot
].curr
) & mask
) == 0)
283 smsr
->values
[slot
].curr
= value
;
284 err
= wrmsrl_safe(shared_msrs_global
.msrs
[slot
], value
);
288 if (!smsr
->registered
) {
289 smsr
->urn
.on_user_return
= kvm_on_user_return
;
290 user_return_notifier_register(&smsr
->urn
);
291 smsr
->registered
= true;
295 EXPORT_SYMBOL_GPL(kvm_set_shared_msr
);
297 static void drop_user_return_notifiers(void)
299 unsigned int cpu
= smp_processor_id();
300 struct kvm_shared_msrs
*smsr
= per_cpu_ptr(shared_msrs
, cpu
);
302 if (smsr
->registered
)
303 kvm_on_user_return(&smsr
->urn
);
306 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
308 return vcpu
->arch
.apic_base
;
310 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
312 int kvm_set_apic_base(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
314 u64 old_state
= vcpu
->arch
.apic_base
&
315 (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
);
316 u64 new_state
= msr_info
->data
&
317 (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
);
318 u64 reserved_bits
= ((~0ULL) << cpuid_maxphyaddr(vcpu
)) | 0x2ff |
319 (guest_cpuid_has(vcpu
, X86_FEATURE_X2APIC
) ? 0 : X2APIC_ENABLE
);
321 if ((msr_info
->data
& reserved_bits
) || new_state
== X2APIC_ENABLE
)
323 if (!msr_info
->host_initiated
&&
324 ((new_state
== MSR_IA32_APICBASE_ENABLE
&&
325 old_state
== (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
)) ||
326 (new_state
== (MSR_IA32_APICBASE_ENABLE
| X2APIC_ENABLE
) &&
330 kvm_lapic_set_base(vcpu
, msr_info
->data
);
333 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
335 asmlinkage __visible
void kvm_spurious_fault(void)
337 /* Fault while not rebooting. We want the trace. */
340 EXPORT_SYMBOL_GPL(kvm_spurious_fault
);
342 #define EXCPT_BENIGN 0
343 #define EXCPT_CONTRIBUTORY 1
346 static int exception_class(int vector
)
356 return EXCPT_CONTRIBUTORY
;
363 #define EXCPT_FAULT 0
365 #define EXCPT_ABORT 2
366 #define EXCPT_INTERRUPT 3
368 static int exception_type(int vector
)
372 if (WARN_ON(vector
> 31 || vector
== NMI_VECTOR
))
373 return EXCPT_INTERRUPT
;
377 /* #DB is trap, as instruction watchpoints are handled elsewhere */
378 if (mask
& ((1 << DB_VECTOR
) | (1 << BP_VECTOR
) | (1 << OF_VECTOR
)))
381 if (mask
& ((1 << DF_VECTOR
) | (1 << MC_VECTOR
)))
384 /* Reserved exceptions will result in fault */
388 static void kvm_multiple_exception(struct kvm_vcpu
*vcpu
,
389 unsigned nr
, bool has_error
, u32 error_code
,
395 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
397 if (!vcpu
->arch
.exception
.pending
&& !vcpu
->arch
.exception
.injected
) {
399 if (has_error
&& !is_protmode(vcpu
))
403 * On vmentry, vcpu->arch.exception.pending is only
404 * true if an event injection was blocked by
405 * nested_run_pending. In that case, however,
406 * vcpu_enter_guest requests an immediate exit,
407 * and the guest shouldn't proceed far enough to
410 WARN_ON_ONCE(vcpu
->arch
.exception
.pending
);
411 vcpu
->arch
.exception
.injected
= true;
413 vcpu
->arch
.exception
.pending
= true;
414 vcpu
->arch
.exception
.injected
= false;
416 vcpu
->arch
.exception
.has_error_code
= has_error
;
417 vcpu
->arch
.exception
.nr
= nr
;
418 vcpu
->arch
.exception
.error_code
= error_code
;
422 /* to check exception */
423 prev_nr
= vcpu
->arch
.exception
.nr
;
424 if (prev_nr
== DF_VECTOR
) {
425 /* triple fault -> shutdown */
426 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
429 class1
= exception_class(prev_nr
);
430 class2
= exception_class(nr
);
431 if ((class1
== EXCPT_CONTRIBUTORY
&& class2
== EXCPT_CONTRIBUTORY
)
432 || (class1
== EXCPT_PF
&& class2
!= EXCPT_BENIGN
)) {
434 * Generate double fault per SDM Table 5-5. Set
435 * exception.pending = true so that the double fault
436 * can trigger a nested vmexit.
438 vcpu
->arch
.exception
.pending
= true;
439 vcpu
->arch
.exception
.injected
= false;
440 vcpu
->arch
.exception
.has_error_code
= true;
441 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
442 vcpu
->arch
.exception
.error_code
= 0;
444 /* replace previous exception with a new one in a hope
445 that instruction re-execution will regenerate lost
450 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
452 kvm_multiple_exception(vcpu
, nr
, false, 0, false);
454 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
456 void kvm_requeue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
458 kvm_multiple_exception(vcpu
, nr
, false, 0, true);
460 EXPORT_SYMBOL_GPL(kvm_requeue_exception
);
462 int kvm_complete_insn_gp(struct kvm_vcpu
*vcpu
, int err
)
465 kvm_inject_gp(vcpu
, 0);
467 return kvm_skip_emulated_instruction(vcpu
);
471 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp
);
473 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
475 ++vcpu
->stat
.pf_guest
;
476 vcpu
->arch
.exception
.nested_apf
=
477 is_guest_mode(vcpu
) && fault
->async_page_fault
;
478 if (vcpu
->arch
.exception
.nested_apf
)
479 vcpu
->arch
.apf
.nested_apf_token
= fault
->address
;
481 vcpu
->arch
.cr2
= fault
->address
;
482 kvm_queue_exception_e(vcpu
, PF_VECTOR
, fault
->error_code
);
484 EXPORT_SYMBOL_GPL(kvm_inject_page_fault
);
486 static bool kvm_propagate_fault(struct kvm_vcpu
*vcpu
, struct x86_exception
*fault
)
488 if (mmu_is_nested(vcpu
) && !fault
->nested_page_fault
)
489 vcpu
->arch
.nested_mmu
.inject_page_fault(vcpu
, fault
);
491 vcpu
->arch
.mmu
.inject_page_fault(vcpu
, fault
);
493 return fault
->nested_page_fault
;
496 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
498 atomic_inc(&vcpu
->arch
.nmi_queued
);
499 kvm_make_request(KVM_REQ_NMI
, vcpu
);
501 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
503 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
505 kvm_multiple_exception(vcpu
, nr
, true, error_code
, false);
507 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
509 void kvm_requeue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
511 kvm_multiple_exception(vcpu
, nr
, true, error_code
, true);
513 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e
);
516 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
517 * a #GP and return false.
519 bool kvm_require_cpl(struct kvm_vcpu
*vcpu
, int required_cpl
)
521 if (kvm_x86_ops
->get_cpl(vcpu
) <= required_cpl
)
523 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
526 EXPORT_SYMBOL_GPL(kvm_require_cpl
);
528 bool kvm_require_dr(struct kvm_vcpu
*vcpu
, int dr
)
530 if ((dr
!= 4 && dr
!= 5) || !kvm_read_cr4_bits(vcpu
, X86_CR4_DE
))
533 kvm_queue_exception(vcpu
, UD_VECTOR
);
536 EXPORT_SYMBOL_GPL(kvm_require_dr
);
539 * This function will be used to read from the physical memory of the currently
540 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
541 * can read from guest physical or from the guest's guest physical memory.
543 int kvm_read_guest_page_mmu(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
,
544 gfn_t ngfn
, void *data
, int offset
, int len
,
547 struct x86_exception exception
;
551 ngpa
= gfn_to_gpa(ngfn
);
552 real_gfn
= mmu
->translate_gpa(vcpu
, ngpa
, access
, &exception
);
553 if (real_gfn
== UNMAPPED_GVA
)
556 real_gfn
= gpa_to_gfn(real_gfn
);
558 return kvm_vcpu_read_guest_page(vcpu
, real_gfn
, data
, offset
, len
);
560 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu
);
562 static int kvm_read_nested_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
563 void *data
, int offset
, int len
, u32 access
)
565 return kvm_read_guest_page_mmu(vcpu
, vcpu
->arch
.walk_mmu
, gfn
,
566 data
, offset
, len
, access
);
570 * Load the pae pdptrs. Return true is they are all valid.
572 int load_pdptrs(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*mmu
, unsigned long cr3
)
574 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
575 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
578 u64 pdpte
[ARRAY_SIZE(mmu
->pdptrs
)];
580 ret
= kvm_read_guest_page_mmu(vcpu
, mmu
, pdpt_gfn
, pdpte
,
581 offset
* sizeof(u64
), sizeof(pdpte
),
582 PFERR_USER_MASK
|PFERR_WRITE_MASK
);
587 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
588 if ((pdpte
[i
] & PT_PRESENT_MASK
) &&
590 vcpu
->arch
.mmu
.guest_rsvd_check
.rsvd_bits_mask
[0][2])) {
597 memcpy(mmu
->pdptrs
, pdpte
, sizeof(mmu
->pdptrs
));
598 __set_bit(VCPU_EXREG_PDPTR
,
599 (unsigned long *)&vcpu
->arch
.regs_avail
);
600 __set_bit(VCPU_EXREG_PDPTR
,
601 (unsigned long *)&vcpu
->arch
.regs_dirty
);
606 EXPORT_SYMBOL_GPL(load_pdptrs
);
608 bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
610 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.walk_mmu
->pdptrs
)];
616 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
619 if (!test_bit(VCPU_EXREG_PDPTR
,
620 (unsigned long *)&vcpu
->arch
.regs_avail
))
623 gfn
= (kvm_read_cr3(vcpu
) & 0xffffffe0ul
) >> PAGE_SHIFT
;
624 offset
= (kvm_read_cr3(vcpu
) & 0xffffffe0ul
) & (PAGE_SIZE
- 1);
625 r
= kvm_read_nested_guest_page(vcpu
, gfn
, pdpte
, offset
, sizeof(pdpte
),
626 PFERR_USER_MASK
| PFERR_WRITE_MASK
);
629 changed
= memcmp(pdpte
, vcpu
->arch
.walk_mmu
->pdptrs
, sizeof(pdpte
)) != 0;
634 EXPORT_SYMBOL_GPL(pdptrs_changed
);
636 int kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
638 unsigned long old_cr0
= kvm_read_cr0(vcpu
);
639 unsigned long update_bits
= X86_CR0_PG
| X86_CR0_WP
;
644 if (cr0
& 0xffffffff00000000UL
)
648 cr0
&= ~CR0_RESERVED_BITS
;
650 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
))
653 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
))
656 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
658 if ((vcpu
->arch
.efer
& EFER_LME
)) {
663 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
668 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
673 if (!(cr0
& X86_CR0_PG
) && kvm_read_cr4_bits(vcpu
, X86_CR4_PCIDE
))
676 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
678 if ((cr0
^ old_cr0
) & X86_CR0_PG
) {
679 kvm_clear_async_pf_completion_queue(vcpu
);
680 kvm_async_pf_hash_reset(vcpu
);
683 if ((cr0
^ old_cr0
) & update_bits
)
684 kvm_mmu_reset_context(vcpu
);
686 if (((cr0
^ old_cr0
) & X86_CR0_CD
) &&
687 kvm_arch_has_noncoherent_dma(vcpu
->kvm
) &&
688 !kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_CD_NW_CLEARED
))
689 kvm_zap_gfn_range(vcpu
->kvm
, 0, ~0ULL);
693 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
695 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
697 (void)kvm_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~0x0eul
) | (msw
& 0x0f));
699 EXPORT_SYMBOL_GPL(kvm_lmsw
);
701 static void kvm_load_guest_xcr0(struct kvm_vcpu
*vcpu
)
703 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
) &&
704 !vcpu
->guest_xcr0_loaded
) {
705 /* kvm_set_xcr() also depends on this */
706 xsetbv(XCR_XFEATURE_ENABLED_MASK
, vcpu
->arch
.xcr0
);
707 vcpu
->guest_xcr0_loaded
= 1;
711 static void kvm_put_guest_xcr0(struct kvm_vcpu
*vcpu
)
713 if (vcpu
->guest_xcr0_loaded
) {
714 if (vcpu
->arch
.xcr0
!= host_xcr0
)
715 xsetbv(XCR_XFEATURE_ENABLED_MASK
, host_xcr0
);
716 vcpu
->guest_xcr0_loaded
= 0;
720 static int __kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
723 u64 old_xcr0
= vcpu
->arch
.xcr0
;
726 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
727 if (index
!= XCR_XFEATURE_ENABLED_MASK
)
729 if (!(xcr0
& XFEATURE_MASK_FP
))
731 if ((xcr0
& XFEATURE_MASK_YMM
) && !(xcr0
& XFEATURE_MASK_SSE
))
735 * Do not allow the guest to set bits that we do not support
736 * saving. However, xcr0 bit 0 is always set, even if the
737 * emulated CPU does not support XSAVE (see fx_init).
739 valid_bits
= vcpu
->arch
.guest_supported_xcr0
| XFEATURE_MASK_FP
;
740 if (xcr0
& ~valid_bits
)
743 if ((!(xcr0
& XFEATURE_MASK_BNDREGS
)) !=
744 (!(xcr0
& XFEATURE_MASK_BNDCSR
)))
747 if (xcr0
& XFEATURE_MASK_AVX512
) {
748 if (!(xcr0
& XFEATURE_MASK_YMM
))
750 if ((xcr0
& XFEATURE_MASK_AVX512
) != XFEATURE_MASK_AVX512
)
753 vcpu
->arch
.xcr0
= xcr0
;
755 if ((xcr0
^ old_xcr0
) & XFEATURE_MASK_EXTEND
)
756 kvm_update_cpuid(vcpu
);
760 int kvm_set_xcr(struct kvm_vcpu
*vcpu
, u32 index
, u64 xcr
)
762 if (kvm_x86_ops
->get_cpl(vcpu
) != 0 ||
763 __kvm_set_xcr(vcpu
, index
, xcr
)) {
764 kvm_inject_gp(vcpu
, 0);
769 EXPORT_SYMBOL_GPL(kvm_set_xcr
);
771 int kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
773 unsigned long old_cr4
= kvm_read_cr4(vcpu
);
774 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PAE
|
775 X86_CR4_SMEP
| X86_CR4_SMAP
| X86_CR4_PKE
;
777 if (cr4
& CR4_RESERVED_BITS
)
780 if (!guest_cpuid_has(vcpu
, X86_FEATURE_XSAVE
) && (cr4
& X86_CR4_OSXSAVE
))
783 if (!guest_cpuid_has(vcpu
, X86_FEATURE_SMEP
) && (cr4
& X86_CR4_SMEP
))
786 if (!guest_cpuid_has(vcpu
, X86_FEATURE_SMAP
) && (cr4
& X86_CR4_SMAP
))
789 if (!guest_cpuid_has(vcpu
, X86_FEATURE_FSGSBASE
) && (cr4
& X86_CR4_FSGSBASE
))
792 if (!guest_cpuid_has(vcpu
, X86_FEATURE_PKU
) && (cr4
& X86_CR4_PKE
))
795 if (!guest_cpuid_has(vcpu
, X86_FEATURE_LA57
) && (cr4
& X86_CR4_LA57
))
798 if (!guest_cpuid_has(vcpu
, X86_FEATURE_UMIP
) && (cr4
& X86_CR4_UMIP
))
801 if (is_long_mode(vcpu
)) {
802 if (!(cr4
& X86_CR4_PAE
))
804 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
805 && ((cr4
^ old_cr4
) & pdptr_bits
)
806 && !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
,
810 if ((cr4
& X86_CR4_PCIDE
) && !(old_cr4
& X86_CR4_PCIDE
)) {
811 if (!guest_cpuid_has(vcpu
, X86_FEATURE_PCID
))
814 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
815 if ((kvm_read_cr3(vcpu
) & X86_CR3_PCID_MASK
) || !is_long_mode(vcpu
))
819 if (kvm_x86_ops
->set_cr4(vcpu
, cr4
))
822 if (((cr4
^ old_cr4
) & pdptr_bits
) ||
823 (!(cr4
& X86_CR4_PCIDE
) && (old_cr4
& X86_CR4_PCIDE
)))
824 kvm_mmu_reset_context(vcpu
);
826 if ((cr4
^ old_cr4
) & (X86_CR4_OSXSAVE
| X86_CR4_PKE
))
827 kvm_update_cpuid(vcpu
);
831 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
833 int kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
836 cr3
&= ~CR3_PCID_INVD
;
839 if (cr3
== kvm_read_cr3(vcpu
) && !pdptrs_changed(vcpu
)) {
840 kvm_mmu_sync_roots(vcpu
);
841 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
845 if (is_long_mode(vcpu
) &&
846 (cr3
& rsvd_bits(cpuid_maxphyaddr(vcpu
), 63)))
848 else if (is_pae(vcpu
) && is_paging(vcpu
) &&
849 !load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, cr3
))
852 vcpu
->arch
.cr3
= cr3
;
853 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
854 kvm_mmu_new_cr3(vcpu
);
857 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
859 int kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
861 if (cr8
& CR8_RESERVED_BITS
)
863 if (lapic_in_kernel(vcpu
))
864 kvm_lapic_set_tpr(vcpu
, cr8
);
866 vcpu
->arch
.cr8
= cr8
;
869 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
871 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
873 if (lapic_in_kernel(vcpu
))
874 return kvm_lapic_get_cr8(vcpu
);
876 return vcpu
->arch
.cr8
;
878 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
880 static void kvm_update_dr0123(struct kvm_vcpu
*vcpu
)
884 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)) {
885 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
886 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
887 vcpu
->arch
.switch_db_regs
|= KVM_DEBUGREG_RELOAD
;
891 static void kvm_update_dr6(struct kvm_vcpu
*vcpu
)
893 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
894 kvm_x86_ops
->set_dr6(vcpu
, vcpu
->arch
.dr6
);
897 static void kvm_update_dr7(struct kvm_vcpu
*vcpu
)
901 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
902 dr7
= vcpu
->arch
.guest_debug_dr7
;
904 dr7
= vcpu
->arch
.dr7
;
905 kvm_x86_ops
->set_dr7(vcpu
, dr7
);
906 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_BP_ENABLED
;
907 if (dr7
& DR7_BP_EN_MASK
)
908 vcpu
->arch
.switch_db_regs
|= KVM_DEBUGREG_BP_ENABLED
;
911 static u64
kvm_dr6_fixed(struct kvm_vcpu
*vcpu
)
913 u64 fixed
= DR6_FIXED_1
;
915 if (!guest_cpuid_has(vcpu
, X86_FEATURE_RTM
))
920 static int __kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
924 vcpu
->arch
.db
[dr
] = val
;
925 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
))
926 vcpu
->arch
.eff_db
[dr
] = val
;
931 if (val
& 0xffffffff00000000ULL
)
933 vcpu
->arch
.dr6
= (val
& DR6_VOLATILE
) | kvm_dr6_fixed(vcpu
);
934 kvm_update_dr6(vcpu
);
939 if (val
& 0xffffffff00000000ULL
)
941 vcpu
->arch
.dr7
= (val
& DR7_VOLATILE
) | DR7_FIXED_1
;
942 kvm_update_dr7(vcpu
);
949 int kvm_set_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long val
)
951 if (__kvm_set_dr(vcpu
, dr
, val
)) {
952 kvm_inject_gp(vcpu
, 0);
957 EXPORT_SYMBOL_GPL(kvm_set_dr
);
959 int kvm_get_dr(struct kvm_vcpu
*vcpu
, int dr
, unsigned long *val
)
963 *val
= vcpu
->arch
.db
[dr
];
968 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
969 *val
= vcpu
->arch
.dr6
;
971 *val
= kvm_x86_ops
->get_dr6(vcpu
);
976 *val
= vcpu
->arch
.dr7
;
981 EXPORT_SYMBOL_GPL(kvm_get_dr
);
983 bool kvm_rdpmc(struct kvm_vcpu
*vcpu
)
985 u32 ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
989 err
= kvm_pmu_rdpmc(vcpu
, ecx
, &data
);
992 kvm_register_write(vcpu
, VCPU_REGS_RAX
, (u32
)data
);
993 kvm_register_write(vcpu
, VCPU_REGS_RDX
, data
>> 32);
996 EXPORT_SYMBOL_GPL(kvm_rdpmc
);
999 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1000 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1002 * This list is modified at module load time to reflect the
1003 * capabilities of the host cpu. This capabilities test skips MSRs that are
1004 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1005 * may depend on host virtualization features rather than host cpu features.
1008 static u32 msrs_to_save
[] = {
1009 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
1011 #ifdef CONFIG_X86_64
1012 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
1014 MSR_IA32_TSC
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
,
1015 MSR_IA32_FEATURE_CONTROL
, MSR_IA32_BNDCFGS
, MSR_TSC_AUX
,
1016 MSR_IA32_SPEC_CTRL
, MSR_IA32_ARCH_CAPABILITIES
1019 static unsigned num_msrs_to_save
;
1021 static u32 emulated_msrs
[] = {
1022 MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
1023 MSR_KVM_SYSTEM_TIME_NEW
, MSR_KVM_WALL_CLOCK_NEW
,
1024 HV_X64_MSR_GUEST_OS_ID
, HV_X64_MSR_HYPERCALL
,
1025 HV_X64_MSR_TIME_REF_COUNT
, HV_X64_MSR_REFERENCE_TSC
,
1026 HV_X64_MSR_TSC_FREQUENCY
, HV_X64_MSR_APIC_FREQUENCY
,
1027 HV_X64_MSR_CRASH_P0
, HV_X64_MSR_CRASH_P1
, HV_X64_MSR_CRASH_P2
,
1028 HV_X64_MSR_CRASH_P3
, HV_X64_MSR_CRASH_P4
, HV_X64_MSR_CRASH_CTL
,
1030 HV_X64_MSR_VP_INDEX
,
1031 HV_X64_MSR_VP_RUNTIME
,
1032 HV_X64_MSR_SCONTROL
,
1033 HV_X64_MSR_STIMER0_CONFIG
,
1034 HV_X64_MSR_APIC_ASSIST_PAGE
, MSR_KVM_ASYNC_PF_EN
, MSR_KVM_STEAL_TIME
,
1037 MSR_IA32_TSC_ADJUST
,
1038 MSR_IA32_TSCDEADLINE
,
1039 MSR_IA32_MISC_ENABLE
,
1040 MSR_IA32_MCG_STATUS
,
1042 MSR_IA32_MCG_EXT_CTL
,
1045 MSR_MISC_FEATURES_ENABLES
,
1046 MSR_AMD64_VIRT_SPEC_CTRL
,
1049 static unsigned num_emulated_msrs
;
1052 * List of msr numbers which are used to expose MSR-based features that
1053 * can be used by a hypervisor to validate requested CPU features.
1055 static u32 msr_based_features
[] = {
1056 MSR_IA32_ARCH_CAPABILITIES
,
1059 static unsigned int num_msr_based_features
;
1061 u64
kvm_get_arch_capabilities(void)
1065 rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES
, &data
);
1068 * If we're doing cache flushes (either "always" or "cond")
1069 * we will do one whenever the guest does a vmlaunch/vmresume.
1070 * If an outer hypervisor is doing the cache flush for us
1071 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1072 * capability to the guest too, and if EPT is disabled we're not
1073 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1074 * require a nested hypervisor to do a flush of its own.
1076 if (l1tf_vmx_mitigation
!= VMENTER_L1D_FLUSH_NEVER
)
1077 data
|= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH
;
1081 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities
);
1083 static int kvm_get_msr_feature(struct kvm_msr_entry
*msr
)
1085 switch (msr
->index
) {
1086 case MSR_IA32_ARCH_CAPABILITIES
:
1087 msr
->data
= kvm_get_arch_capabilities();
1090 if (kvm_x86_ops
->get_msr_feature(msr
))
1096 static int do_get_msr_feature(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1098 struct kvm_msr_entry msr
;
1102 r
= kvm_get_msr_feature(&msr
);
1111 bool kvm_valid_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1113 if (efer
& efer_reserved_bits
)
1116 if (efer
& EFER_FFXSR
&& !guest_cpuid_has(vcpu
, X86_FEATURE_FXSR_OPT
))
1119 if (efer
& EFER_SVME
&& !guest_cpuid_has(vcpu
, X86_FEATURE_SVM
))
1124 EXPORT_SYMBOL_GPL(kvm_valid_efer
);
1126 static int set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1128 u64 old_efer
= vcpu
->arch
.efer
;
1130 if (!kvm_valid_efer(vcpu
, efer
))
1134 && (vcpu
->arch
.efer
& EFER_LME
) != (efer
& EFER_LME
))
1138 efer
|= vcpu
->arch
.efer
& EFER_LMA
;
1140 kvm_x86_ops
->set_efer(vcpu
, efer
);
1142 /* Update reserved bits */
1143 if ((efer
^ old_efer
) & EFER_NX
)
1144 kvm_mmu_reset_context(vcpu
);
1149 void kvm_enable_efer_bits(u64 mask
)
1151 efer_reserved_bits
&= ~mask
;
1153 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
1156 * Writes msr value into into the appropriate "register".
1157 * Returns 0 on success, non-0 otherwise.
1158 * Assumes vcpu_load() was already called.
1160 int kvm_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1162 switch (msr
->index
) {
1165 case MSR_KERNEL_GS_BASE
:
1168 if (is_noncanonical_address(msr
->data
, vcpu
))
1171 case MSR_IA32_SYSENTER_EIP
:
1172 case MSR_IA32_SYSENTER_ESP
:
1174 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1175 * non-canonical address is written on Intel but not on
1176 * AMD (which ignores the top 32-bits, because it does
1177 * not implement 64-bit SYSENTER).
1179 * 64-bit code should hence be able to write a non-canonical
1180 * value on AMD. Making the address canonical ensures that
1181 * vmentry does not fail on Intel after writing a non-canonical
1182 * value, and that something deterministic happens if the guest
1183 * invokes 64-bit SYSENTER.
1185 msr
->data
= get_canonical(msr
->data
, vcpu_virt_addr_bits(vcpu
));
1187 return kvm_x86_ops
->set_msr(vcpu
, msr
);
1189 EXPORT_SYMBOL_GPL(kvm_set_msr
);
1192 * Adapt set_msr() to msr_io()'s calling convention
1194 static int do_get_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1196 struct msr_data msr
;
1200 msr
.host_initiated
= true;
1201 r
= kvm_get_msr(vcpu
, &msr
);
1209 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
1211 struct msr_data msr
;
1215 msr
.host_initiated
= true;
1216 return kvm_set_msr(vcpu
, &msr
);
1219 #ifdef CONFIG_X86_64
1220 struct pvclock_gtod_data
{
1223 struct { /* extract of a clocksource struct */
1236 static struct pvclock_gtod_data pvclock_gtod_data
;
1238 static void update_pvclock_gtod(struct timekeeper
*tk
)
1240 struct pvclock_gtod_data
*vdata
= &pvclock_gtod_data
;
1243 boot_ns
= ktime_to_ns(ktime_add(tk
->tkr_mono
.base
, tk
->offs_boot
));
1245 write_seqcount_begin(&vdata
->seq
);
1247 /* copy pvclock gtod data */
1248 vdata
->clock
.vclock_mode
= tk
->tkr_mono
.clock
->archdata
.vclock_mode
;
1249 vdata
->clock
.cycle_last
= tk
->tkr_mono
.cycle_last
;
1250 vdata
->clock
.mask
= tk
->tkr_mono
.mask
;
1251 vdata
->clock
.mult
= tk
->tkr_mono
.mult
;
1252 vdata
->clock
.shift
= tk
->tkr_mono
.shift
;
1254 vdata
->boot_ns
= boot_ns
;
1255 vdata
->nsec_base
= tk
->tkr_mono
.xtime_nsec
;
1257 vdata
->wall_time_sec
= tk
->xtime_sec
;
1259 write_seqcount_end(&vdata
->seq
);
1263 void kvm_set_pending_timer(struct kvm_vcpu
*vcpu
)
1266 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1267 * vcpu_enter_guest. This function is only called from
1268 * the physical CPU that is running vcpu.
1270 kvm_make_request(KVM_REQ_PENDING_TIMER
, vcpu
);
1273 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
1277 struct pvclock_wall_clock wc
;
1278 struct timespec64 boot
;
1283 r
= kvm_read_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1288 ++version
; /* first time write, random junk */
1292 if (kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
)))
1296 * The guest calculates current wall clock time by adding
1297 * system time (updated by kvm_guest_time_update below) to the
1298 * wall clock specified here. guest system time equals host
1299 * system time for us, thus we must fill in host boot time here.
1301 getboottime64(&boot
);
1303 if (kvm
->arch
.kvmclock_offset
) {
1304 struct timespec64 ts
= ns_to_timespec64(kvm
->arch
.kvmclock_offset
);
1305 boot
= timespec64_sub(boot
, ts
);
1307 wc
.sec
= (u32
)boot
.tv_sec
; /* overflow in 2106 guest time */
1308 wc
.nsec
= boot
.tv_nsec
;
1309 wc
.version
= version
;
1311 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
1314 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
1317 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
1319 do_shl32_div32(dividend
, divisor
);
1323 static void kvm_get_time_scale(uint64_t scaled_hz
, uint64_t base_hz
,
1324 s8
*pshift
, u32
*pmultiplier
)
1332 scaled64
= scaled_hz
;
1333 while (tps64
> scaled64
*2 || tps64
& 0xffffffff00000000ULL
) {
1338 tps32
= (uint32_t)tps64
;
1339 while (tps32
<= scaled64
|| scaled64
& 0xffffffff00000000ULL
) {
1340 if (scaled64
& 0xffffffff00000000ULL
|| tps32
& 0x80000000)
1348 *pmultiplier
= div_frac(scaled64
, tps32
);
1350 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1351 __func__
, base_hz
, scaled_hz
, shift
, *pmultiplier
);
1354 #ifdef CONFIG_X86_64
1355 static atomic_t kvm_guest_has_master_clock
= ATOMIC_INIT(0);
1358 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
1359 static unsigned long max_tsc_khz
;
1361 static u32
adjust_tsc_khz(u32 khz
, s32 ppm
)
1363 u64 v
= (u64
)khz
* (1000000 + ppm
);
1368 static int set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
, bool scale
)
1372 /* Guest TSC same frequency as host TSC? */
1374 vcpu
->arch
.tsc_scaling_ratio
= kvm_default_tsc_scaling_ratio
;
1378 /* TSC scaling supported? */
1379 if (!kvm_has_tsc_control
) {
1380 if (user_tsc_khz
> tsc_khz
) {
1381 vcpu
->arch
.tsc_catchup
= 1;
1382 vcpu
->arch
.tsc_always_catchup
= 1;
1385 WARN(1, "user requested TSC rate below hardware speed\n");
1390 /* TSC scaling required - calculate ratio */
1391 ratio
= mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits
,
1392 user_tsc_khz
, tsc_khz
);
1394 if (ratio
== 0 || ratio
>= kvm_max_tsc_scaling_ratio
) {
1395 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1400 vcpu
->arch
.tsc_scaling_ratio
= ratio
;
1404 static int kvm_set_tsc_khz(struct kvm_vcpu
*vcpu
, u32 user_tsc_khz
)
1406 u32 thresh_lo
, thresh_hi
;
1407 int use_scaling
= 0;
1409 /* tsc_khz can be zero if TSC calibration fails */
1410 if (user_tsc_khz
== 0) {
1411 /* set tsc_scaling_ratio to a safe value */
1412 vcpu
->arch
.tsc_scaling_ratio
= kvm_default_tsc_scaling_ratio
;
1416 /* Compute a scale to convert nanoseconds in TSC cycles */
1417 kvm_get_time_scale(user_tsc_khz
* 1000LL, NSEC_PER_SEC
,
1418 &vcpu
->arch
.virtual_tsc_shift
,
1419 &vcpu
->arch
.virtual_tsc_mult
);
1420 vcpu
->arch
.virtual_tsc_khz
= user_tsc_khz
;
1423 * Compute the variation in TSC rate which is acceptable
1424 * within the range of tolerance and decide if the
1425 * rate being applied is within that bounds of the hardware
1426 * rate. If so, no scaling or compensation need be done.
1428 thresh_lo
= adjust_tsc_khz(tsc_khz
, -tsc_tolerance_ppm
);
1429 thresh_hi
= adjust_tsc_khz(tsc_khz
, tsc_tolerance_ppm
);
1430 if (user_tsc_khz
< thresh_lo
|| user_tsc_khz
> thresh_hi
) {
1431 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz
, thresh_lo
, thresh_hi
);
1434 return set_tsc_khz(vcpu
, user_tsc_khz
, use_scaling
);
1437 static u64
compute_guest_tsc(struct kvm_vcpu
*vcpu
, s64 kernel_ns
)
1439 u64 tsc
= pvclock_scale_delta(kernel_ns
-vcpu
->arch
.this_tsc_nsec
,
1440 vcpu
->arch
.virtual_tsc_mult
,
1441 vcpu
->arch
.virtual_tsc_shift
);
1442 tsc
+= vcpu
->arch
.this_tsc_write
;
1446 static void kvm_track_tsc_matching(struct kvm_vcpu
*vcpu
)
1448 #ifdef CONFIG_X86_64
1450 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
1451 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1453 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1454 atomic_read(&vcpu
->kvm
->online_vcpus
));
1457 * Once the masterclock is enabled, always perform request in
1458 * order to update it.
1460 * In order to enable masterclock, the host clocksource must be TSC
1461 * and the vcpus need to have matched TSCs. When that happens,
1462 * perform request to enable masterclock.
1464 if (ka
->use_master_clock
||
1465 (gtod
->clock
.vclock_mode
== VCLOCK_TSC
&& vcpus_matched
))
1466 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
1468 trace_kvm_track_tsc(vcpu
->vcpu_id
, ka
->nr_vcpus_matched_tsc
,
1469 atomic_read(&vcpu
->kvm
->online_vcpus
),
1470 ka
->use_master_clock
, gtod
->clock
.vclock_mode
);
1474 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu
*vcpu
, s64 offset
)
1476 u64 curr_offset
= kvm_x86_ops
->read_l1_tsc_offset(vcpu
);
1477 vcpu
->arch
.ia32_tsc_adjust_msr
+= offset
- curr_offset
;
1481 * Multiply tsc by a fixed point number represented by ratio.
1483 * The most significant 64-N bits (mult) of ratio represent the
1484 * integral part of the fixed point number; the remaining N bits
1485 * (frac) represent the fractional part, ie. ratio represents a fixed
1486 * point number (mult + frac * 2^(-N)).
1488 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1490 static inline u64
__scale_tsc(u64 ratio
, u64 tsc
)
1492 return mul_u64_u64_shr(tsc
, ratio
, kvm_tsc_scaling_ratio_frac_bits
);
1495 u64
kvm_scale_tsc(struct kvm_vcpu
*vcpu
, u64 tsc
)
1498 u64 ratio
= vcpu
->arch
.tsc_scaling_ratio
;
1500 if (ratio
!= kvm_default_tsc_scaling_ratio
)
1501 _tsc
= __scale_tsc(ratio
, tsc
);
1505 EXPORT_SYMBOL_GPL(kvm_scale_tsc
);
1507 static u64
kvm_compute_tsc_offset(struct kvm_vcpu
*vcpu
, u64 target_tsc
)
1511 tsc
= kvm_scale_tsc(vcpu
, rdtsc());
1513 return target_tsc
- tsc
;
1516 u64
kvm_read_l1_tsc(struct kvm_vcpu
*vcpu
, u64 host_tsc
)
1518 u64 tsc_offset
= kvm_x86_ops
->read_l1_tsc_offset(vcpu
);
1520 return tsc_offset
+ kvm_scale_tsc(vcpu
, host_tsc
);
1522 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc
);
1524 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
1526 kvm_x86_ops
->write_tsc_offset(vcpu
, offset
);
1527 vcpu
->arch
.tsc_offset
= offset
;
1530 void kvm_write_tsc(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
1532 struct kvm
*kvm
= vcpu
->kvm
;
1533 u64 offset
, ns
, elapsed
;
1534 unsigned long flags
;
1536 bool already_matched
;
1537 u64 data
= msr
->data
;
1538 bool synchronizing
= false;
1540 raw_spin_lock_irqsave(&kvm
->arch
.tsc_write_lock
, flags
);
1541 offset
= kvm_compute_tsc_offset(vcpu
, data
);
1542 ns
= ktime_get_boot_ns();
1543 elapsed
= ns
- kvm
->arch
.last_tsc_nsec
;
1545 if (vcpu
->arch
.virtual_tsc_khz
) {
1546 if (data
== 0 && msr
->host_initiated
) {
1548 * detection of vcpu initialization -- need to sync
1549 * with other vCPUs. This particularly helps to keep
1550 * kvm_clock stable after CPU hotplug
1552 synchronizing
= true;
1554 u64 tsc_exp
= kvm
->arch
.last_tsc_write
+
1555 nsec_to_cycles(vcpu
, elapsed
);
1556 u64 tsc_hz
= vcpu
->arch
.virtual_tsc_khz
* 1000LL;
1558 * Special case: TSC write with a small delta (1 second)
1559 * of virtual cycle time against real time is
1560 * interpreted as an attempt to synchronize the CPU.
1562 synchronizing
= data
< tsc_exp
+ tsc_hz
&&
1563 data
+ tsc_hz
> tsc_exp
;
1568 * For a reliable TSC, we can match TSC offsets, and for an unstable
1569 * TSC, we add elapsed time in this computation. We could let the
1570 * compensation code attempt to catch up if we fall behind, but
1571 * it's better to try to match offsets from the beginning.
1573 if (synchronizing
&&
1574 vcpu
->arch
.virtual_tsc_khz
== kvm
->arch
.last_tsc_khz
) {
1575 if (!check_tsc_unstable()) {
1576 offset
= kvm
->arch
.cur_tsc_offset
;
1577 pr_debug("kvm: matched tsc offset for %llu\n", data
);
1579 u64 delta
= nsec_to_cycles(vcpu
, elapsed
);
1581 offset
= kvm_compute_tsc_offset(vcpu
, data
);
1582 pr_debug("kvm: adjusted tsc offset by %llu\n", delta
);
1585 already_matched
= (vcpu
->arch
.this_tsc_generation
== kvm
->arch
.cur_tsc_generation
);
1588 * We split periods of matched TSC writes into generations.
1589 * For each generation, we track the original measured
1590 * nanosecond time, offset, and write, so if TSCs are in
1591 * sync, we can match exact offset, and if not, we can match
1592 * exact software computation in compute_guest_tsc()
1594 * These values are tracked in kvm->arch.cur_xxx variables.
1596 kvm
->arch
.cur_tsc_generation
++;
1597 kvm
->arch
.cur_tsc_nsec
= ns
;
1598 kvm
->arch
.cur_tsc_write
= data
;
1599 kvm
->arch
.cur_tsc_offset
= offset
;
1601 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1602 kvm
->arch
.cur_tsc_generation
, data
);
1606 * We also track th most recent recorded KHZ, write and time to
1607 * allow the matching interval to be extended at each write.
1609 kvm
->arch
.last_tsc_nsec
= ns
;
1610 kvm
->arch
.last_tsc_write
= data
;
1611 kvm
->arch
.last_tsc_khz
= vcpu
->arch
.virtual_tsc_khz
;
1613 vcpu
->arch
.last_guest_tsc
= data
;
1615 /* Keep track of which generation this VCPU has synchronized to */
1616 vcpu
->arch
.this_tsc_generation
= kvm
->arch
.cur_tsc_generation
;
1617 vcpu
->arch
.this_tsc_nsec
= kvm
->arch
.cur_tsc_nsec
;
1618 vcpu
->arch
.this_tsc_write
= kvm
->arch
.cur_tsc_write
;
1620 if (!msr
->host_initiated
&& guest_cpuid_has(vcpu
, X86_FEATURE_TSC_ADJUST
))
1621 update_ia32_tsc_adjust_msr(vcpu
, offset
);
1623 kvm_vcpu_write_tsc_offset(vcpu
, offset
);
1624 raw_spin_unlock_irqrestore(&kvm
->arch
.tsc_write_lock
, flags
);
1626 spin_lock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1628 kvm
->arch
.nr_vcpus_matched_tsc
= 0;
1629 } else if (!already_matched
) {
1630 kvm
->arch
.nr_vcpus_matched_tsc
++;
1633 kvm_track_tsc_matching(vcpu
);
1634 spin_unlock(&kvm
->arch
.pvclock_gtod_sync_lock
);
1637 EXPORT_SYMBOL_GPL(kvm_write_tsc
);
1639 static inline void adjust_tsc_offset_guest(struct kvm_vcpu
*vcpu
,
1642 kvm_vcpu_write_tsc_offset(vcpu
, vcpu
->arch
.tsc_offset
+ adjustment
);
1645 static inline void adjust_tsc_offset_host(struct kvm_vcpu
*vcpu
, s64 adjustment
)
1647 if (vcpu
->arch
.tsc_scaling_ratio
!= kvm_default_tsc_scaling_ratio
)
1648 WARN_ON(adjustment
< 0);
1649 adjustment
= kvm_scale_tsc(vcpu
, (u64
) adjustment
);
1650 adjust_tsc_offset_guest(vcpu
, adjustment
);
1653 #ifdef CONFIG_X86_64
1655 static u64
read_tsc(void)
1657 u64 ret
= (u64
)rdtsc_ordered();
1658 u64 last
= pvclock_gtod_data
.clock
.cycle_last
;
1660 if (likely(ret
>= last
))
1664 * GCC likes to generate cmov here, but this branch is extremely
1665 * predictable (it's just a function of time and the likely is
1666 * very likely) and there's a data dependence, so force GCC
1667 * to generate a branch instead. I don't barrier() because
1668 * we don't actually need a barrier, and if this function
1669 * ever gets inlined it will generate worse code.
1675 static inline u64
vgettsc(u64
*cycle_now
)
1678 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1680 *cycle_now
= read_tsc();
1682 v
= (*cycle_now
- gtod
->clock
.cycle_last
) & gtod
->clock
.mask
;
1683 return v
* gtod
->clock
.mult
;
1686 static int do_monotonic_boot(s64
*t
, u64
*cycle_now
)
1688 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1694 seq
= read_seqcount_begin(>od
->seq
);
1695 mode
= gtod
->clock
.vclock_mode
;
1696 ns
= gtod
->nsec_base
;
1697 ns
+= vgettsc(cycle_now
);
1698 ns
>>= gtod
->clock
.shift
;
1699 ns
+= gtod
->boot_ns
;
1700 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
1706 static int do_realtime(struct timespec
*ts
, u64
*cycle_now
)
1708 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
1714 seq
= read_seqcount_begin(>od
->seq
);
1715 mode
= gtod
->clock
.vclock_mode
;
1716 ts
->tv_sec
= gtod
->wall_time_sec
;
1717 ns
= gtod
->nsec_base
;
1718 ns
+= vgettsc(cycle_now
);
1719 ns
>>= gtod
->clock
.shift
;
1720 } while (unlikely(read_seqcount_retry(>od
->seq
, seq
)));
1722 ts
->tv_sec
+= __iter_div_u64_rem(ns
, NSEC_PER_SEC
, &ns
);
1728 /* returns true if host is using tsc clocksource */
1729 static bool kvm_get_time_and_clockread(s64
*kernel_ns
, u64
*cycle_now
)
1731 /* checked again under seqlock below */
1732 if (pvclock_gtod_data
.clock
.vclock_mode
!= VCLOCK_TSC
)
1735 return do_monotonic_boot(kernel_ns
, cycle_now
) == VCLOCK_TSC
;
1738 /* returns true if host is using tsc clocksource */
1739 static bool kvm_get_walltime_and_clockread(struct timespec
*ts
,
1742 /* checked again under seqlock below */
1743 if (pvclock_gtod_data
.clock
.vclock_mode
!= VCLOCK_TSC
)
1746 return do_realtime(ts
, cycle_now
) == VCLOCK_TSC
;
1752 * Assuming a stable TSC across physical CPUS, and a stable TSC
1753 * across virtual CPUs, the following condition is possible.
1754 * Each numbered line represents an event visible to both
1755 * CPUs at the next numbered event.
1757 * "timespecX" represents host monotonic time. "tscX" represents
1760 * VCPU0 on CPU0 | VCPU1 on CPU1
1762 * 1. read timespec0,tsc0
1763 * 2. | timespec1 = timespec0 + N
1765 * 3. transition to guest | transition to guest
1766 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1767 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1768 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1770 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1773 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1775 * - 0 < N - M => M < N
1777 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1778 * always the case (the difference between two distinct xtime instances
1779 * might be smaller then the difference between corresponding TSC reads,
1780 * when updating guest vcpus pvclock areas).
1782 * To avoid that problem, do not allow visibility of distinct
1783 * system_timestamp/tsc_timestamp values simultaneously: use a master
1784 * copy of host monotonic time values. Update that master copy
1787 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1791 static void pvclock_update_vm_gtod_copy(struct kvm
*kvm
)
1793 #ifdef CONFIG_X86_64
1794 struct kvm_arch
*ka
= &kvm
->arch
;
1796 bool host_tsc_clocksource
, vcpus_matched
;
1798 vcpus_matched
= (ka
->nr_vcpus_matched_tsc
+ 1 ==
1799 atomic_read(&kvm
->online_vcpus
));
1802 * If the host uses TSC clock, then passthrough TSC as stable
1805 host_tsc_clocksource
= kvm_get_time_and_clockread(
1806 &ka
->master_kernel_ns
,
1807 &ka
->master_cycle_now
);
1809 ka
->use_master_clock
= host_tsc_clocksource
&& vcpus_matched
1810 && !ka
->backwards_tsc_observed
1811 && !ka
->boot_vcpu_runs_old_kvmclock
;
1813 if (ka
->use_master_clock
)
1814 atomic_set(&kvm_guest_has_master_clock
, 1);
1816 vclock_mode
= pvclock_gtod_data
.clock
.vclock_mode
;
1817 trace_kvm_update_master_clock(ka
->use_master_clock
, vclock_mode
,
1822 void kvm_make_mclock_inprogress_request(struct kvm
*kvm
)
1824 kvm_make_all_cpus_request(kvm
, KVM_REQ_MCLOCK_INPROGRESS
);
1827 static void kvm_gen_update_masterclock(struct kvm
*kvm
)
1829 #ifdef CONFIG_X86_64
1831 struct kvm_vcpu
*vcpu
;
1832 struct kvm_arch
*ka
= &kvm
->arch
;
1834 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1835 kvm_make_mclock_inprogress_request(kvm
);
1836 /* no guest entries from this point */
1837 pvclock_update_vm_gtod_copy(kvm
);
1839 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1840 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
1842 /* guest entries allowed */
1843 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1844 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS
, vcpu
);
1846 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1850 u64
get_kvmclock_ns(struct kvm
*kvm
)
1852 struct kvm_arch
*ka
= &kvm
->arch
;
1853 struct pvclock_vcpu_time_info hv_clock
;
1856 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1857 if (!ka
->use_master_clock
) {
1858 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1859 return ktime_get_boot_ns() + ka
->kvmclock_offset
;
1862 hv_clock
.tsc_timestamp
= ka
->master_cycle_now
;
1863 hv_clock
.system_time
= ka
->master_kernel_ns
+ ka
->kvmclock_offset
;
1864 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1866 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1869 if (__this_cpu_read(cpu_tsc_khz
)) {
1870 kvm_get_time_scale(NSEC_PER_SEC
, __this_cpu_read(cpu_tsc_khz
) * 1000LL,
1871 &hv_clock
.tsc_shift
,
1872 &hv_clock
.tsc_to_system_mul
);
1873 ret
= __pvclock_read_cycles(&hv_clock
, rdtsc());
1875 ret
= ktime_get_boot_ns() + ka
->kvmclock_offset
;
1882 static void kvm_setup_pvclock_page(struct kvm_vcpu
*v
)
1884 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1885 struct pvclock_vcpu_time_info guest_hv_clock
;
1887 if (unlikely(kvm_read_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1888 &guest_hv_clock
, sizeof(guest_hv_clock
))))
1891 /* This VCPU is paused, but it's legal for a guest to read another
1892 * VCPU's kvmclock, so we really have to follow the specification where
1893 * it says that version is odd if data is being modified, and even after
1896 * Version field updates must be kept separate. This is because
1897 * kvm_write_guest_cached might use a "rep movs" instruction, and
1898 * writes within a string instruction are weakly ordered. So there
1899 * are three writes overall.
1901 * As a small optimization, only write the version field in the first
1902 * and third write. The vcpu->pv_time cache is still valid, because the
1903 * version field is the first in the struct.
1905 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info
, version
) != 0);
1907 if (guest_hv_clock
.version
& 1)
1908 ++guest_hv_clock
.version
; /* first time write, random junk */
1910 vcpu
->hv_clock
.version
= guest_hv_clock
.version
+ 1;
1911 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1913 sizeof(vcpu
->hv_clock
.version
));
1917 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1918 vcpu
->hv_clock
.flags
|= (guest_hv_clock
.flags
& PVCLOCK_GUEST_STOPPED
);
1920 if (vcpu
->pvclock_set_guest_stopped_request
) {
1921 vcpu
->hv_clock
.flags
|= PVCLOCK_GUEST_STOPPED
;
1922 vcpu
->pvclock_set_guest_stopped_request
= false;
1925 trace_kvm_pvclock_update(v
->vcpu_id
, &vcpu
->hv_clock
);
1927 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1929 sizeof(vcpu
->hv_clock
));
1933 vcpu
->hv_clock
.version
++;
1934 kvm_write_guest_cached(v
->kvm
, &vcpu
->pv_time
,
1936 sizeof(vcpu
->hv_clock
.version
));
1939 static int kvm_guest_time_update(struct kvm_vcpu
*v
)
1941 unsigned long flags
, tgt_tsc_khz
;
1942 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
1943 struct kvm_arch
*ka
= &v
->kvm
->arch
;
1945 u64 tsc_timestamp
, host_tsc
;
1947 bool use_master_clock
;
1953 * If the host uses TSC clock, then passthrough TSC as stable
1956 spin_lock(&ka
->pvclock_gtod_sync_lock
);
1957 use_master_clock
= ka
->use_master_clock
;
1958 if (use_master_clock
) {
1959 host_tsc
= ka
->master_cycle_now
;
1960 kernel_ns
= ka
->master_kernel_ns
;
1962 spin_unlock(&ka
->pvclock_gtod_sync_lock
);
1964 /* Keep irq disabled to prevent changes to the clock */
1965 local_irq_save(flags
);
1966 tgt_tsc_khz
= __this_cpu_read(cpu_tsc_khz
);
1967 if (unlikely(tgt_tsc_khz
== 0)) {
1968 local_irq_restore(flags
);
1969 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
1972 if (!use_master_clock
) {
1974 kernel_ns
= ktime_get_boot_ns();
1977 tsc_timestamp
= kvm_read_l1_tsc(v
, host_tsc
);
1980 * We may have to catch up the TSC to match elapsed wall clock
1981 * time for two reasons, even if kvmclock is used.
1982 * 1) CPU could have been running below the maximum TSC rate
1983 * 2) Broken TSC compensation resets the base at each VCPU
1984 * entry to avoid unknown leaps of TSC even when running
1985 * again on the same CPU. This may cause apparent elapsed
1986 * time to disappear, and the guest to stand still or run
1989 if (vcpu
->tsc_catchup
) {
1990 u64 tsc
= compute_guest_tsc(v
, kernel_ns
);
1991 if (tsc
> tsc_timestamp
) {
1992 adjust_tsc_offset_guest(v
, tsc
- tsc_timestamp
);
1993 tsc_timestamp
= tsc
;
1997 local_irq_restore(flags
);
1999 /* With all the info we got, fill in the values */
2001 if (kvm_has_tsc_control
)
2002 tgt_tsc_khz
= kvm_scale_tsc(v
, tgt_tsc_khz
);
2004 if (unlikely(vcpu
->hw_tsc_khz
!= tgt_tsc_khz
)) {
2005 kvm_get_time_scale(NSEC_PER_SEC
, tgt_tsc_khz
* 1000LL,
2006 &vcpu
->hv_clock
.tsc_shift
,
2007 &vcpu
->hv_clock
.tsc_to_system_mul
);
2008 vcpu
->hw_tsc_khz
= tgt_tsc_khz
;
2011 vcpu
->hv_clock
.tsc_timestamp
= tsc_timestamp
;
2012 vcpu
->hv_clock
.system_time
= kernel_ns
+ v
->kvm
->arch
.kvmclock_offset
;
2013 vcpu
->last_guest_tsc
= tsc_timestamp
;
2015 /* If the host uses TSC clocksource, then it is stable */
2017 if (use_master_clock
)
2018 pvclock_flags
|= PVCLOCK_TSC_STABLE_BIT
;
2020 vcpu
->hv_clock
.flags
= pvclock_flags
;
2022 if (vcpu
->pv_time_enabled
)
2023 kvm_setup_pvclock_page(v
);
2024 if (v
== kvm_get_vcpu(v
->kvm
, 0))
2025 kvm_hv_setup_tsc_page(v
->kvm
, &vcpu
->hv_clock
);
2030 * kvmclock updates which are isolated to a given vcpu, such as
2031 * vcpu->cpu migration, should not allow system_timestamp from
2032 * the rest of the vcpus to remain static. Otherwise ntp frequency
2033 * correction applies to one vcpu's system_timestamp but not
2036 * So in those cases, request a kvmclock update for all vcpus.
2037 * We need to rate-limit these requests though, as they can
2038 * considerably slow guests that have a large number of vcpus.
2039 * The time for a remote vcpu to update its kvmclock is bound
2040 * by the delay we use to rate-limit the updates.
2043 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2045 static void kvmclock_update_fn(struct work_struct
*work
)
2048 struct delayed_work
*dwork
= to_delayed_work(work
);
2049 struct kvm_arch
*ka
= container_of(dwork
, struct kvm_arch
,
2050 kvmclock_update_work
);
2051 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
2052 struct kvm_vcpu
*vcpu
;
2054 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2055 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2056 kvm_vcpu_kick(vcpu
);
2060 static void kvm_gen_kvmclock_update(struct kvm_vcpu
*v
)
2062 struct kvm
*kvm
= v
->kvm
;
2064 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, v
);
2065 schedule_delayed_work(&kvm
->arch
.kvmclock_update_work
,
2066 KVMCLOCK_UPDATE_DELAY
);
2069 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2071 static void kvmclock_sync_fn(struct work_struct
*work
)
2073 struct delayed_work
*dwork
= to_delayed_work(work
);
2074 struct kvm_arch
*ka
= container_of(dwork
, struct kvm_arch
,
2075 kvmclock_sync_work
);
2076 struct kvm
*kvm
= container_of(ka
, struct kvm
, arch
);
2078 if (!kvmclock_periodic_sync
)
2081 schedule_delayed_work(&kvm
->arch
.kvmclock_update_work
, 0);
2082 schedule_delayed_work(&kvm
->arch
.kvmclock_sync_work
,
2083 KVMCLOCK_SYNC_PERIOD
);
2086 static int set_msr_mce(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2088 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2089 unsigned bank_num
= mcg_cap
& 0xff;
2090 u32 msr
= msr_info
->index
;
2091 u64 data
= msr_info
->data
;
2094 case MSR_IA32_MCG_STATUS
:
2095 vcpu
->arch
.mcg_status
= data
;
2097 case MSR_IA32_MCG_CTL
:
2098 if (!(mcg_cap
& MCG_CTL_P
))
2100 if (data
!= 0 && data
!= ~(u64
)0)
2102 vcpu
->arch
.mcg_ctl
= data
;
2105 if (msr
>= MSR_IA32_MC0_CTL
&&
2106 msr
< MSR_IA32_MCx_CTL(bank_num
)) {
2107 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
2108 /* only 0 or all 1s can be written to IA32_MCi_CTL
2109 * some Linux kernels though clear bit 10 in bank 4 to
2110 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2111 * this to avoid an uncatched #GP in the guest
2113 if ((offset
& 0x3) == 0 &&
2114 data
!= 0 && (data
| (1 << 10)) != ~(u64
)0)
2116 if (!msr_info
->host_initiated
&&
2117 (offset
& 0x3) == 1 && data
!= 0)
2119 vcpu
->arch
.mce_banks
[offset
] = data
;
2127 static int xen_hvm_config(struct kvm_vcpu
*vcpu
, u64 data
)
2129 struct kvm
*kvm
= vcpu
->kvm
;
2130 int lm
= is_long_mode(vcpu
);
2131 u8
*blob_addr
= lm
? (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_64
2132 : (u8
*)(long)kvm
->arch
.xen_hvm_config
.blob_addr_32
;
2133 u8 blob_size
= lm
? kvm
->arch
.xen_hvm_config
.blob_size_64
2134 : kvm
->arch
.xen_hvm_config
.blob_size_32
;
2135 u32 page_num
= data
& ~PAGE_MASK
;
2136 u64 page_addr
= data
& PAGE_MASK
;
2141 if (page_num
>= blob_size
)
2144 page
= memdup_user(blob_addr
+ (page_num
* PAGE_SIZE
), PAGE_SIZE
);
2149 if (kvm_vcpu_write_guest(vcpu
, page_addr
, page
, PAGE_SIZE
))
2158 static int kvm_pv_enable_async_pf(struct kvm_vcpu
*vcpu
, u64 data
)
2160 gpa_t gpa
= data
& ~0x3f;
2162 /* Bits 3:5 are reserved, Should be zero */
2166 vcpu
->arch
.apf
.msr_val
= data
;
2168 if (!(data
& KVM_ASYNC_PF_ENABLED
)) {
2169 kvm_clear_async_pf_completion_queue(vcpu
);
2170 kvm_async_pf_hash_reset(vcpu
);
2174 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, gpa
,
2178 vcpu
->arch
.apf
.send_user_only
= !(data
& KVM_ASYNC_PF_SEND_ALWAYS
);
2179 vcpu
->arch
.apf
.delivery_as_pf_vmexit
= data
& KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT
;
2180 kvm_async_pf_wakeup_all(vcpu
);
2184 static void kvmclock_reset(struct kvm_vcpu
*vcpu
)
2186 vcpu
->arch
.pv_time_enabled
= false;
2189 static void record_steal_time(struct kvm_vcpu
*vcpu
)
2191 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
2194 if (unlikely(kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2195 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
))))
2198 vcpu
->arch
.st
.steal
.preempted
= 0;
2200 if (vcpu
->arch
.st
.steal
.version
& 1)
2201 vcpu
->arch
.st
.steal
.version
+= 1; /* first time write, random junk */
2203 vcpu
->arch
.st
.steal
.version
+= 1;
2205 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2206 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2210 vcpu
->arch
.st
.steal
.steal
+= current
->sched_info
.run_delay
-
2211 vcpu
->arch
.st
.last_steal
;
2212 vcpu
->arch
.st
.last_steal
= current
->sched_info
.run_delay
;
2214 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2215 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2219 vcpu
->arch
.st
.steal
.version
+= 1;
2221 kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2222 &vcpu
->arch
.st
.steal
, sizeof(struct kvm_steal_time
));
2225 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2228 u32 msr
= msr_info
->index
;
2229 u64 data
= msr_info
->data
;
2232 case MSR_AMD64_NB_CFG
:
2233 case MSR_IA32_UCODE_REV
:
2234 case MSR_IA32_UCODE_WRITE
:
2235 case MSR_VM_HSAVE_PA
:
2236 case MSR_AMD64_PATCH_LOADER
:
2237 case MSR_AMD64_BU_CFG2
:
2238 case MSR_AMD64_DC_CFG
:
2242 return set_efer(vcpu
, data
);
2244 data
&= ~(u64
)0x40; /* ignore flush filter disable */
2245 data
&= ~(u64
)0x100; /* ignore ignne emulation enable */
2246 data
&= ~(u64
)0x8; /* ignore TLB cache disable */
2247 data
&= ~(u64
)0x40000; /* ignore Mc status write enable */
2249 vcpu_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
2254 case MSR_FAM10H_MMIO_CONF_BASE
:
2256 vcpu_unimpl(vcpu
, "unimplemented MMIO_CONF_BASE wrmsr: "
2261 case MSR_IA32_DEBUGCTLMSR
:
2263 /* We support the non-activated case already */
2265 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
2266 /* Values other than LBR and BTF are vendor-specific,
2267 thus reserved and should throw a #GP */
2270 vcpu_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2273 case 0x200 ... 0x2ff:
2274 return kvm_mtrr_set_msr(vcpu
, msr
, data
);
2275 case MSR_IA32_APICBASE
:
2276 return kvm_set_apic_base(vcpu
, msr_info
);
2277 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2278 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
2279 case MSR_IA32_TSCDEADLINE
:
2280 kvm_set_lapic_tscdeadline_msr(vcpu
, data
);
2282 case MSR_IA32_TSC_ADJUST
:
2283 if (guest_cpuid_has(vcpu
, X86_FEATURE_TSC_ADJUST
)) {
2284 if (!msr_info
->host_initiated
) {
2285 s64 adj
= data
- vcpu
->arch
.ia32_tsc_adjust_msr
;
2286 adjust_tsc_offset_guest(vcpu
, adj
);
2288 vcpu
->arch
.ia32_tsc_adjust_msr
= data
;
2291 case MSR_IA32_MISC_ENABLE
:
2292 vcpu
->arch
.ia32_misc_enable_msr
= data
;
2294 case MSR_IA32_SMBASE
:
2295 if (!msr_info
->host_initiated
)
2297 vcpu
->arch
.smbase
= data
;
2300 kvm_write_tsc(vcpu
, msr_info
);
2302 case MSR_KVM_WALL_CLOCK_NEW
:
2303 case MSR_KVM_WALL_CLOCK
:
2304 vcpu
->kvm
->arch
.wall_clock
= data
;
2305 kvm_write_wall_clock(vcpu
->kvm
, data
);
2307 case MSR_KVM_SYSTEM_TIME_NEW
:
2308 case MSR_KVM_SYSTEM_TIME
: {
2309 struct kvm_arch
*ka
= &vcpu
->kvm
->arch
;
2311 kvmclock_reset(vcpu
);
2313 if (vcpu
->vcpu_id
== 0 && !msr_info
->host_initiated
) {
2314 bool tmp
= (msr
== MSR_KVM_SYSTEM_TIME
);
2316 if (ka
->boot_vcpu_runs_old_kvmclock
!= tmp
)
2317 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
2319 ka
->boot_vcpu_runs_old_kvmclock
= tmp
;
2322 vcpu
->arch
.time
= data
;
2323 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
);
2325 /* we verify if the enable bit is set... */
2329 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
,
2330 &vcpu
->arch
.pv_time
, data
& ~1ULL,
2331 sizeof(struct pvclock_vcpu_time_info
)))
2332 vcpu
->arch
.pv_time_enabled
= false;
2334 vcpu
->arch
.pv_time_enabled
= true;
2338 case MSR_KVM_ASYNC_PF_EN
:
2339 if (kvm_pv_enable_async_pf(vcpu
, data
))
2342 case MSR_KVM_STEAL_TIME
:
2344 if (unlikely(!sched_info_on()))
2347 if (data
& KVM_STEAL_RESERVED_MASK
)
2350 if (kvm_gfn_to_hva_cache_init(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
2351 data
& KVM_STEAL_VALID_BITS
,
2352 sizeof(struct kvm_steal_time
)))
2355 vcpu
->arch
.st
.msr_val
= data
;
2357 if (!(data
& KVM_MSR_ENABLED
))
2360 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
2363 case MSR_KVM_PV_EOI_EN
:
2364 if (kvm_lapic_enable_pv_eoi(vcpu
, data
))
2368 case MSR_IA32_MCG_CTL
:
2369 case MSR_IA32_MCG_STATUS
:
2370 case MSR_IA32_MC0_CTL
... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS
) - 1:
2371 return set_msr_mce(vcpu
, msr_info
);
2373 case MSR_K7_PERFCTR0
... MSR_K7_PERFCTR3
:
2374 case MSR_P6_PERFCTR0
... MSR_P6_PERFCTR1
:
2375 pr
= true; /* fall through */
2376 case MSR_K7_EVNTSEL0
... MSR_K7_EVNTSEL3
:
2377 case MSR_P6_EVNTSEL0
... MSR_P6_EVNTSEL1
:
2378 if (kvm_pmu_is_valid_msr(vcpu
, msr
))
2379 return kvm_pmu_set_msr(vcpu
, msr_info
);
2381 if (pr
|| data
!= 0)
2382 vcpu_unimpl(vcpu
, "disabled perfctr wrmsr: "
2383 "0x%x data 0x%llx\n", msr
, data
);
2385 case MSR_K7_CLK_CTL
:
2387 * Ignore all writes to this no longer documented MSR.
2388 * Writes are only relevant for old K7 processors,
2389 * all pre-dating SVM, but a recommended workaround from
2390 * AMD for these chips. It is possible to specify the
2391 * affected processor models on the command line, hence
2392 * the need to ignore the workaround.
2395 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2396 case HV_X64_MSR_CRASH_P0
... HV_X64_MSR_CRASH_P4
:
2397 case HV_X64_MSR_CRASH_CTL
:
2398 case HV_X64_MSR_STIMER0_CONFIG
... HV_X64_MSR_STIMER3_COUNT
:
2399 return kvm_hv_set_msr_common(vcpu
, msr
, data
,
2400 msr_info
->host_initiated
);
2401 case MSR_IA32_BBL_CR_CTL3
:
2402 /* Drop writes to this legacy MSR -- see rdmsr
2403 * counterpart for further detail.
2405 if (report_ignored_msrs
)
2406 vcpu_unimpl(vcpu
, "ignored wrmsr: 0x%x data 0x%llx\n",
2409 case MSR_AMD64_OSVW_ID_LENGTH
:
2410 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2412 vcpu
->arch
.osvw
.length
= data
;
2414 case MSR_AMD64_OSVW_STATUS
:
2415 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2417 vcpu
->arch
.osvw
.status
= data
;
2419 case MSR_PLATFORM_INFO
:
2420 if (!msr_info
->host_initiated
||
2421 data
& ~MSR_PLATFORM_INFO_CPUID_FAULT
||
2422 (!(data
& MSR_PLATFORM_INFO_CPUID_FAULT
) &&
2423 cpuid_fault_enabled(vcpu
)))
2425 vcpu
->arch
.msr_platform_info
= data
;
2427 case MSR_MISC_FEATURES_ENABLES
:
2428 if (data
& ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT
||
2429 (data
& MSR_MISC_FEATURES_ENABLES_CPUID_FAULT
&&
2430 !supports_cpuid_fault(vcpu
)))
2432 vcpu
->arch
.msr_misc_features_enables
= data
;
2435 if (msr
&& (msr
== vcpu
->kvm
->arch
.xen_hvm_config
.msr
))
2436 return xen_hvm_config(vcpu
, data
);
2437 if (kvm_pmu_is_valid_msr(vcpu
, msr
))
2438 return kvm_pmu_set_msr(vcpu
, msr_info
);
2440 vcpu_debug_ratelimited(vcpu
, "unhandled wrmsr: 0x%x data 0x%llx\n",
2444 if (report_ignored_msrs
)
2446 "ignored wrmsr: 0x%x data 0x%llx\n",
2453 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
2457 * Reads an msr value (of 'msr_index') into 'pdata'.
2458 * Returns 0 on success, non-0 otherwise.
2459 * Assumes vcpu_load() was already called.
2461 int kvm_get_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr
)
2463 return kvm_x86_ops
->get_msr(vcpu
, msr
);
2465 EXPORT_SYMBOL_GPL(kvm_get_msr
);
2467 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
2470 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
2471 unsigned bank_num
= mcg_cap
& 0xff;
2474 case MSR_IA32_P5_MC_ADDR
:
2475 case MSR_IA32_P5_MC_TYPE
:
2478 case MSR_IA32_MCG_CAP
:
2479 data
= vcpu
->arch
.mcg_cap
;
2481 case MSR_IA32_MCG_CTL
:
2482 if (!(mcg_cap
& MCG_CTL_P
))
2484 data
= vcpu
->arch
.mcg_ctl
;
2486 case MSR_IA32_MCG_STATUS
:
2487 data
= vcpu
->arch
.mcg_status
;
2490 if (msr
>= MSR_IA32_MC0_CTL
&&
2491 msr
< MSR_IA32_MCx_CTL(bank_num
)) {
2492 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
2493 data
= vcpu
->arch
.mce_banks
[offset
];
2502 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
2504 switch (msr_info
->index
) {
2505 case MSR_IA32_PLATFORM_ID
:
2506 case MSR_IA32_EBL_CR_POWERON
:
2507 case MSR_IA32_DEBUGCTLMSR
:
2508 case MSR_IA32_LASTBRANCHFROMIP
:
2509 case MSR_IA32_LASTBRANCHTOIP
:
2510 case MSR_IA32_LASTINTFROMIP
:
2511 case MSR_IA32_LASTINTTOIP
:
2513 case MSR_K8_TSEG_ADDR
:
2514 case MSR_K8_TSEG_MASK
:
2516 case MSR_VM_HSAVE_PA
:
2517 case MSR_K8_INT_PENDING_MSG
:
2518 case MSR_AMD64_NB_CFG
:
2519 case MSR_FAM10H_MMIO_CONF_BASE
:
2520 case MSR_AMD64_BU_CFG2
:
2521 case MSR_IA32_PERF_CTL
:
2522 case MSR_AMD64_DC_CFG
:
2525 case MSR_K7_EVNTSEL0
... MSR_K7_EVNTSEL3
:
2526 case MSR_K7_PERFCTR0
... MSR_K7_PERFCTR3
:
2527 case MSR_P6_PERFCTR0
... MSR_P6_PERFCTR1
:
2528 case MSR_P6_EVNTSEL0
... MSR_P6_EVNTSEL1
:
2529 if (kvm_pmu_is_valid_msr(vcpu
, msr_info
->index
))
2530 return kvm_pmu_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2533 case MSR_IA32_UCODE_REV
:
2534 msr_info
->data
= 0x100000000ULL
;
2537 msr_info
->data
= kvm_scale_tsc(vcpu
, rdtsc()) + vcpu
->arch
.tsc_offset
;
2540 case 0x200 ... 0x2ff:
2541 return kvm_mtrr_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2542 case 0xcd: /* fsb frequency */
2546 * MSR_EBC_FREQUENCY_ID
2547 * Conservative value valid for even the basic CPU models.
2548 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2549 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2550 * and 266MHz for model 3, or 4. Set Core Clock
2551 * Frequency to System Bus Frequency Ratio to 1 (bits
2552 * 31:24) even though these are only valid for CPU
2553 * models > 2, however guests may end up dividing or
2554 * multiplying by zero otherwise.
2556 case MSR_EBC_FREQUENCY_ID
:
2557 msr_info
->data
= 1 << 24;
2559 case MSR_IA32_APICBASE
:
2560 msr_info
->data
= kvm_get_apic_base(vcpu
);
2562 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
2563 return kvm_x2apic_msr_read(vcpu
, msr_info
->index
, &msr_info
->data
);
2565 case MSR_IA32_TSCDEADLINE
:
2566 msr_info
->data
= kvm_get_lapic_tscdeadline_msr(vcpu
);
2568 case MSR_IA32_TSC_ADJUST
:
2569 msr_info
->data
= (u64
)vcpu
->arch
.ia32_tsc_adjust_msr
;
2571 case MSR_IA32_MISC_ENABLE
:
2572 msr_info
->data
= vcpu
->arch
.ia32_misc_enable_msr
;
2574 case MSR_IA32_SMBASE
:
2575 if (!msr_info
->host_initiated
)
2577 msr_info
->data
= vcpu
->arch
.smbase
;
2579 case MSR_IA32_PERF_STATUS
:
2580 /* TSC increment by tick */
2581 msr_info
->data
= 1000ULL;
2582 /* CPU multiplier */
2583 msr_info
->data
|= (((uint64_t)4ULL) << 40);
2586 msr_info
->data
= vcpu
->arch
.efer
;
2588 case MSR_KVM_WALL_CLOCK
:
2589 case MSR_KVM_WALL_CLOCK_NEW
:
2590 msr_info
->data
= vcpu
->kvm
->arch
.wall_clock
;
2592 case MSR_KVM_SYSTEM_TIME
:
2593 case MSR_KVM_SYSTEM_TIME_NEW
:
2594 msr_info
->data
= vcpu
->arch
.time
;
2596 case MSR_KVM_ASYNC_PF_EN
:
2597 msr_info
->data
= vcpu
->arch
.apf
.msr_val
;
2599 case MSR_KVM_STEAL_TIME
:
2600 msr_info
->data
= vcpu
->arch
.st
.msr_val
;
2602 case MSR_KVM_PV_EOI_EN
:
2603 msr_info
->data
= vcpu
->arch
.pv_eoi
.msr_val
;
2605 case MSR_IA32_P5_MC_ADDR
:
2606 case MSR_IA32_P5_MC_TYPE
:
2607 case MSR_IA32_MCG_CAP
:
2608 case MSR_IA32_MCG_CTL
:
2609 case MSR_IA32_MCG_STATUS
:
2610 case MSR_IA32_MC0_CTL
... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS
) - 1:
2611 return get_msr_mce(vcpu
, msr_info
->index
, &msr_info
->data
);
2612 case MSR_K7_CLK_CTL
:
2614 * Provide expected ramp-up count for K7. All other
2615 * are set to zero, indicating minimum divisors for
2618 * This prevents guest kernels on AMD host with CPU
2619 * type 6, model 8 and higher from exploding due to
2620 * the rdmsr failing.
2622 msr_info
->data
= 0x20000000;
2624 case HV_X64_MSR_GUEST_OS_ID
... HV_X64_MSR_SINT15
:
2625 case HV_X64_MSR_CRASH_P0
... HV_X64_MSR_CRASH_P4
:
2626 case HV_X64_MSR_CRASH_CTL
:
2627 case HV_X64_MSR_STIMER0_CONFIG
... HV_X64_MSR_STIMER3_COUNT
:
2628 return kvm_hv_get_msr_common(vcpu
,
2629 msr_info
->index
, &msr_info
->data
);
2631 case MSR_IA32_BBL_CR_CTL3
:
2632 /* This legacy MSR exists but isn't fully documented in current
2633 * silicon. It is however accessed by winxp in very narrow
2634 * scenarios where it sets bit #19, itself documented as
2635 * a "reserved" bit. Best effort attempt to source coherent
2636 * read data here should the balance of the register be
2637 * interpreted by the guest:
2639 * L2 cache control register 3: 64GB range, 256KB size,
2640 * enabled, latency 0x1, configured
2642 msr_info
->data
= 0xbe702111;
2644 case MSR_AMD64_OSVW_ID_LENGTH
:
2645 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2647 msr_info
->data
= vcpu
->arch
.osvw
.length
;
2649 case MSR_AMD64_OSVW_STATUS
:
2650 if (!guest_cpuid_has(vcpu
, X86_FEATURE_OSVW
))
2652 msr_info
->data
= vcpu
->arch
.osvw
.status
;
2654 case MSR_PLATFORM_INFO
:
2655 msr_info
->data
= vcpu
->arch
.msr_platform_info
;
2657 case MSR_MISC_FEATURES_ENABLES
:
2658 msr_info
->data
= vcpu
->arch
.msr_misc_features_enables
;
2661 if (kvm_pmu_is_valid_msr(vcpu
, msr_info
->index
))
2662 return kvm_pmu_get_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
2664 vcpu_debug_ratelimited(vcpu
, "unhandled rdmsr: 0x%x\n",
2668 if (report_ignored_msrs
)
2669 vcpu_unimpl(vcpu
, "ignored rdmsr: 0x%x\n",
2677 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
2680 * Read or write a bunch of msrs. All parameters are kernel addresses.
2682 * @return number of msrs set successfully.
2684 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
2685 struct kvm_msr_entry
*entries
,
2686 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2687 unsigned index
, u64
*data
))
2691 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
2692 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
2699 * Read or write a bunch of msrs. Parameters are user addresses.
2701 * @return number of msrs set successfully.
2703 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
2704 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
2705 unsigned index
, u64
*data
),
2708 struct kvm_msrs msrs
;
2709 struct kvm_msr_entry
*entries
;
2714 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
2718 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
2721 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
2722 entries
= memdup_user(user_msrs
->entries
, size
);
2723 if (IS_ERR(entries
)) {
2724 r
= PTR_ERR(entries
);
2728 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
2733 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
2744 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
2749 case KVM_CAP_IRQCHIP
:
2751 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
2752 case KVM_CAP_SET_TSS_ADDR
:
2753 case KVM_CAP_EXT_CPUID
:
2754 case KVM_CAP_EXT_EMUL_CPUID
:
2755 case KVM_CAP_CLOCKSOURCE
:
2757 case KVM_CAP_NOP_IO_DELAY
:
2758 case KVM_CAP_MP_STATE
:
2759 case KVM_CAP_SYNC_MMU
:
2760 case KVM_CAP_USER_NMI
:
2761 case KVM_CAP_REINJECT_CONTROL
:
2762 case KVM_CAP_IRQ_INJECT_STATUS
:
2763 case KVM_CAP_IOEVENTFD
:
2764 case KVM_CAP_IOEVENTFD_NO_LENGTH
:
2766 case KVM_CAP_PIT_STATE2
:
2767 case KVM_CAP_SET_IDENTITY_MAP_ADDR
:
2768 case KVM_CAP_XEN_HVM
:
2769 case KVM_CAP_VCPU_EVENTS
:
2770 case KVM_CAP_HYPERV
:
2771 case KVM_CAP_HYPERV_VAPIC
:
2772 case KVM_CAP_HYPERV_SPIN
:
2773 case KVM_CAP_HYPERV_SYNIC
:
2774 case KVM_CAP_HYPERV_SYNIC2
:
2775 case KVM_CAP_HYPERV_VP_INDEX
:
2776 case KVM_CAP_PCI_SEGMENT
:
2777 case KVM_CAP_DEBUGREGS
:
2778 case KVM_CAP_X86_ROBUST_SINGLESTEP
:
2780 case KVM_CAP_ASYNC_PF
:
2781 case KVM_CAP_GET_TSC_KHZ
:
2782 case KVM_CAP_KVMCLOCK_CTRL
:
2783 case KVM_CAP_READONLY_MEM
:
2784 case KVM_CAP_HYPERV_TIME
:
2785 case KVM_CAP_IOAPIC_POLARITY_IGNORED
:
2786 case KVM_CAP_TSC_DEADLINE_TIMER
:
2787 case KVM_CAP_ENABLE_CAP_VM
:
2788 case KVM_CAP_DISABLE_QUIRKS
:
2789 case KVM_CAP_SET_BOOT_CPU_ID
:
2790 case KVM_CAP_SPLIT_IRQCHIP
:
2791 case KVM_CAP_IMMEDIATE_EXIT
:
2792 case KVM_CAP_GET_MSR_FEATURES
:
2795 case KVM_CAP_ADJUST_CLOCK
:
2796 r
= KVM_CLOCK_TSC_STABLE
;
2798 case KVM_CAP_X86_GUEST_MWAIT
:
2799 r
= kvm_mwait_in_guest();
2801 case KVM_CAP_X86_SMM
:
2802 /* SMBASE is usually relocated above 1M on modern chipsets,
2803 * and SMM handlers might indeed rely on 4G segment limits,
2804 * so do not report SMM to be available if real mode is
2805 * emulated via vm86 mode. Still, do not go to great lengths
2806 * to avoid userspace's usage of the feature, because it is a
2807 * fringe case that is not enabled except via specific settings
2808 * of the module parameters.
2810 r
= kvm_x86_ops
->has_emulated_msr(MSR_IA32_SMBASE
);
2813 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
2815 case KVM_CAP_NR_VCPUS
:
2816 r
= KVM_SOFT_MAX_VCPUS
;
2818 case KVM_CAP_MAX_VCPUS
:
2821 case KVM_CAP_NR_MEMSLOTS
:
2822 r
= KVM_USER_MEM_SLOTS
;
2824 case KVM_CAP_PV_MMU
: /* obsolete */
2828 r
= KVM_MAX_MCE_BANKS
;
2831 r
= boot_cpu_has(X86_FEATURE_XSAVE
);
2833 case KVM_CAP_TSC_CONTROL
:
2834 r
= kvm_has_tsc_control
;
2836 case KVM_CAP_X2APIC_API
:
2837 r
= KVM_X2APIC_API_VALID_FLAGS
;
2847 long kvm_arch_dev_ioctl(struct file
*filp
,
2848 unsigned int ioctl
, unsigned long arg
)
2850 void __user
*argp
= (void __user
*)arg
;
2854 case KVM_GET_MSR_INDEX_LIST
: {
2855 struct kvm_msr_list __user
*user_msr_list
= argp
;
2856 struct kvm_msr_list msr_list
;
2860 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
2863 msr_list
.nmsrs
= num_msrs_to_save
+ num_emulated_msrs
;
2864 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
2867 if (n
< msr_list
.nmsrs
)
2870 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
2871 num_msrs_to_save
* sizeof(u32
)))
2873 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
2875 num_emulated_msrs
* sizeof(u32
)))
2880 case KVM_GET_SUPPORTED_CPUID
:
2881 case KVM_GET_EMULATED_CPUID
: {
2882 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
2883 struct kvm_cpuid2 cpuid
;
2886 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
2889 r
= kvm_dev_ioctl_get_cpuid(&cpuid
, cpuid_arg
->entries
,
2895 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
2900 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
2902 if (copy_to_user(argp
, &kvm_mce_cap_supported
,
2903 sizeof(kvm_mce_cap_supported
)))
2907 case KVM_GET_MSR_FEATURE_INDEX_LIST
: {
2908 struct kvm_msr_list __user
*user_msr_list
= argp
;
2909 struct kvm_msr_list msr_list
;
2913 if (copy_from_user(&msr_list
, user_msr_list
, sizeof(msr_list
)))
2916 msr_list
.nmsrs
= num_msr_based_features
;
2917 if (copy_to_user(user_msr_list
, &msr_list
, sizeof(msr_list
)))
2920 if (n
< msr_list
.nmsrs
)
2923 if (copy_to_user(user_msr_list
->indices
, &msr_based_features
,
2924 num_msr_based_features
* sizeof(u32
)))
2930 r
= msr_io(NULL
, argp
, do_get_msr_feature
, 1);
2940 static void wbinvd_ipi(void *garbage
)
2945 static bool need_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
2947 return kvm_arch_has_noncoherent_dma(vcpu
->kvm
);
2950 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
2952 /* Address WBINVD may be executed by guest */
2953 if (need_emulate_wbinvd(vcpu
)) {
2954 if (kvm_x86_ops
->has_wbinvd_exit())
2955 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
2956 else if (vcpu
->cpu
!= -1 && vcpu
->cpu
!= cpu
)
2957 smp_call_function_single(vcpu
->cpu
,
2958 wbinvd_ipi
, NULL
, 1);
2961 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
2963 /* Apply any externally detected TSC adjustments (due to suspend) */
2964 if (unlikely(vcpu
->arch
.tsc_offset_adjustment
)) {
2965 adjust_tsc_offset_host(vcpu
, vcpu
->arch
.tsc_offset_adjustment
);
2966 vcpu
->arch
.tsc_offset_adjustment
= 0;
2967 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
2970 if (unlikely(vcpu
->cpu
!= cpu
) || check_tsc_unstable()) {
2971 s64 tsc_delta
= !vcpu
->arch
.last_host_tsc
? 0 :
2972 rdtsc() - vcpu
->arch
.last_host_tsc
;
2974 mark_tsc_unstable("KVM discovered backwards TSC");
2976 if (check_tsc_unstable()) {
2977 u64 offset
= kvm_compute_tsc_offset(vcpu
,
2978 vcpu
->arch
.last_guest_tsc
);
2979 kvm_vcpu_write_tsc_offset(vcpu
, offset
);
2980 vcpu
->arch
.tsc_catchup
= 1;
2983 if (kvm_lapic_hv_timer_in_use(vcpu
))
2984 kvm_lapic_restart_hv_timer(vcpu
);
2987 * On a host with synchronized TSC, there is no need to update
2988 * kvmclock on vcpu->cpu migration
2990 if (!vcpu
->kvm
->arch
.use_master_clock
|| vcpu
->cpu
== -1)
2991 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
);
2992 if (vcpu
->cpu
!= cpu
)
2993 kvm_make_request(KVM_REQ_MIGRATE_TIMER
, vcpu
);
2997 kvm_make_request(KVM_REQ_STEAL_UPDATE
, vcpu
);
3000 static void kvm_steal_time_set_preempted(struct kvm_vcpu
*vcpu
)
3002 if (!(vcpu
->arch
.st
.msr_val
& KVM_MSR_ENABLED
))
3005 vcpu
->arch
.st
.steal
.preempted
= 1;
3007 kvm_write_guest_offset_cached(vcpu
->kvm
, &vcpu
->arch
.st
.stime
,
3008 &vcpu
->arch
.st
.steal
.preempted
,
3009 offsetof(struct kvm_steal_time
, preempted
),
3010 sizeof(vcpu
->arch
.st
.steal
.preempted
));
3013 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
3017 if (vcpu
->preempted
)
3018 vcpu
->arch
.preempted_in_kernel
= !kvm_x86_ops
->get_cpl(vcpu
);
3021 * Disable page faults because we're in atomic context here.
3022 * kvm_write_guest_offset_cached() would call might_fault()
3023 * that relies on pagefault_disable() to tell if there's a
3024 * bug. NOTE: the write to guest memory may not go through if
3025 * during postcopy live migration or if there's heavy guest
3028 pagefault_disable();
3030 * kvm_memslots() will be called by
3031 * kvm_write_guest_offset_cached() so take the srcu lock.
3033 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3034 kvm_steal_time_set_preempted(vcpu
);
3035 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3037 kvm_x86_ops
->vcpu_put(vcpu
);
3038 vcpu
->arch
.last_host_tsc
= rdtsc();
3040 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3041 * on every vmexit, but if not, we might have a stale dr6 from the
3042 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3047 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
3048 struct kvm_lapic_state
*s
)
3050 if (kvm_x86_ops
->sync_pir_to_irr
&& vcpu
->arch
.apicv_active
)
3051 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
3053 return kvm_apic_get_state(vcpu
, s
);
3056 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
3057 struct kvm_lapic_state
*s
)
3061 r
= kvm_apic_set_state(vcpu
, s
);
3064 update_cr8_intercept(vcpu
);
3069 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu
*vcpu
)
3071 return (!lapic_in_kernel(vcpu
) ||
3072 kvm_apic_accept_pic_intr(vcpu
));
3076 * if userspace requested an interrupt window, check that the
3077 * interrupt window is open.
3079 * No need to exit to userspace if we already have an interrupt queued.
3081 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu
*vcpu
)
3083 return kvm_arch_interrupt_allowed(vcpu
) &&
3084 !kvm_cpu_has_interrupt(vcpu
) &&
3085 !kvm_event_needs_reinjection(vcpu
) &&
3086 kvm_cpu_accept_dm_intr(vcpu
);
3089 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
3090 struct kvm_interrupt
*irq
)
3092 if (irq
->irq
>= KVM_NR_INTERRUPTS
)
3095 if (!irqchip_in_kernel(vcpu
->kvm
)) {
3096 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
3097 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
3102 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3103 * fail for in-kernel 8259.
3105 if (pic_in_kernel(vcpu
->kvm
))
3108 if (vcpu
->arch
.pending_external_vector
!= -1)
3111 vcpu
->arch
.pending_external_vector
= irq
->irq
;
3112 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
3116 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
3118 kvm_inject_nmi(vcpu
);
3123 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu
*vcpu
)
3125 kvm_make_request(KVM_REQ_SMI
, vcpu
);
3130 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
3131 struct kvm_tpr_access_ctl
*tac
)
3135 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
3139 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
3143 unsigned bank_num
= mcg_cap
& 0xff, bank
;
3146 if (!bank_num
|| bank_num
>= KVM_MAX_MCE_BANKS
)
3148 if (mcg_cap
& ~(kvm_mce_cap_supported
| 0xff | 0xff0000))
3151 vcpu
->arch
.mcg_cap
= mcg_cap
;
3152 /* Init IA32_MCG_CTL to all 1s */
3153 if (mcg_cap
& MCG_CTL_P
)
3154 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
3155 /* Init IA32_MCi_CTL to all 1s */
3156 for (bank
= 0; bank
< bank_num
; bank
++)
3157 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
3159 if (kvm_x86_ops
->setup_mce
)
3160 kvm_x86_ops
->setup_mce(vcpu
);
3165 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
3166 struct kvm_x86_mce
*mce
)
3168 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
3169 unsigned bank_num
= mcg_cap
& 0xff;
3170 u64
*banks
= vcpu
->arch
.mce_banks
;
3172 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
3175 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3176 * reporting is disabled
3178 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
3179 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
3181 banks
+= 4 * mce
->bank
;
3183 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3184 * reporting is disabled for the bank
3186 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
3188 if (mce
->status
& MCI_STATUS_UC
) {
3189 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
3190 !kvm_read_cr4_bits(vcpu
, X86_CR4_MCE
)) {
3191 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
3194 if (banks
[1] & MCI_STATUS_VAL
)
3195 mce
->status
|= MCI_STATUS_OVER
;
3196 banks
[2] = mce
->addr
;
3197 banks
[3] = mce
->misc
;
3198 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
3199 banks
[1] = mce
->status
;
3200 kvm_queue_exception(vcpu
, MC_VECTOR
);
3201 } else if (!(banks
[1] & MCI_STATUS_VAL
)
3202 || !(banks
[1] & MCI_STATUS_UC
)) {
3203 if (banks
[1] & MCI_STATUS_VAL
)
3204 mce
->status
|= MCI_STATUS_OVER
;
3205 banks
[2] = mce
->addr
;
3206 banks
[3] = mce
->misc
;
3207 banks
[1] = mce
->status
;
3209 banks
[1] |= MCI_STATUS_OVER
;
3213 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu
*vcpu
,
3214 struct kvm_vcpu_events
*events
)
3218 * FIXME: pass injected and pending separately. This is only
3219 * needed for nested virtualization, whose state cannot be
3220 * migrated yet. For now we can combine them.
3222 events
->exception
.injected
=
3223 (vcpu
->arch
.exception
.pending
||
3224 vcpu
->arch
.exception
.injected
) &&
3225 !kvm_exception_is_soft(vcpu
->arch
.exception
.nr
);
3226 events
->exception
.nr
= vcpu
->arch
.exception
.nr
;
3227 events
->exception
.has_error_code
= vcpu
->arch
.exception
.has_error_code
;
3228 events
->exception
.pad
= 0;
3229 events
->exception
.error_code
= vcpu
->arch
.exception
.error_code
;
3231 events
->interrupt
.injected
=
3232 vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
;
3233 events
->interrupt
.nr
= vcpu
->arch
.interrupt
.nr
;
3234 events
->interrupt
.soft
= 0;
3235 events
->interrupt
.shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
);
3237 events
->nmi
.injected
= vcpu
->arch
.nmi_injected
;
3238 events
->nmi
.pending
= vcpu
->arch
.nmi_pending
!= 0;
3239 events
->nmi
.masked
= kvm_x86_ops
->get_nmi_mask(vcpu
);
3240 events
->nmi
.pad
= 0;
3242 events
->sipi_vector
= 0; /* never valid when reporting to user space */
3244 events
->smi
.smm
= is_smm(vcpu
);
3245 events
->smi
.pending
= vcpu
->arch
.smi_pending
;
3246 events
->smi
.smm_inside_nmi
=
3247 !!(vcpu
->arch
.hflags
& HF_SMM_INSIDE_NMI_MASK
);
3248 events
->smi
.latched_init
= kvm_lapic_latched_init(vcpu
);
3250 events
->flags
= (KVM_VCPUEVENT_VALID_NMI_PENDING
3251 | KVM_VCPUEVENT_VALID_SHADOW
3252 | KVM_VCPUEVENT_VALID_SMM
);
3253 memset(&events
->reserved
, 0, sizeof(events
->reserved
));
3256 static void kvm_set_hflags(struct kvm_vcpu
*vcpu
, unsigned emul_flags
);
3258 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu
*vcpu
,
3259 struct kvm_vcpu_events
*events
)
3261 if (events
->flags
& ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3262 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3263 | KVM_VCPUEVENT_VALID_SHADOW
3264 | KVM_VCPUEVENT_VALID_SMM
))
3267 if (events
->exception
.injected
&&
3268 (events
->exception
.nr
> 31 || events
->exception
.nr
== NMI_VECTOR
||
3269 is_guest_mode(vcpu
)))
3272 /* INITs are latched while in SMM */
3273 if (events
->flags
& KVM_VCPUEVENT_VALID_SMM
&&
3274 (events
->smi
.smm
|| events
->smi
.pending
) &&
3275 vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
)
3279 vcpu
->arch
.exception
.injected
= false;
3280 vcpu
->arch
.exception
.pending
= events
->exception
.injected
;
3281 vcpu
->arch
.exception
.nr
= events
->exception
.nr
;
3282 vcpu
->arch
.exception
.has_error_code
= events
->exception
.has_error_code
;
3283 vcpu
->arch
.exception
.error_code
= events
->exception
.error_code
;
3285 vcpu
->arch
.interrupt
.pending
= events
->interrupt
.injected
;
3286 vcpu
->arch
.interrupt
.nr
= events
->interrupt
.nr
;
3287 vcpu
->arch
.interrupt
.soft
= events
->interrupt
.soft
;
3288 if (events
->flags
& KVM_VCPUEVENT_VALID_SHADOW
)
3289 kvm_x86_ops
->set_interrupt_shadow(vcpu
,
3290 events
->interrupt
.shadow
);
3292 vcpu
->arch
.nmi_injected
= events
->nmi
.injected
;
3293 if (events
->flags
& KVM_VCPUEVENT_VALID_NMI_PENDING
)
3294 vcpu
->arch
.nmi_pending
= events
->nmi
.pending
;
3295 kvm_x86_ops
->set_nmi_mask(vcpu
, events
->nmi
.masked
);
3297 if (events
->flags
& KVM_VCPUEVENT_VALID_SIPI_VECTOR
&&
3298 lapic_in_kernel(vcpu
))
3299 vcpu
->arch
.apic
->sipi_vector
= events
->sipi_vector
;
3301 if (events
->flags
& KVM_VCPUEVENT_VALID_SMM
) {
3302 u32 hflags
= vcpu
->arch
.hflags
;
3303 if (events
->smi
.smm
)
3304 hflags
|= HF_SMM_MASK
;
3306 hflags
&= ~HF_SMM_MASK
;
3307 kvm_set_hflags(vcpu
, hflags
);
3309 vcpu
->arch
.smi_pending
= events
->smi
.pending
;
3311 if (events
->smi
.smm
) {
3312 if (events
->smi
.smm_inside_nmi
)
3313 vcpu
->arch
.hflags
|= HF_SMM_INSIDE_NMI_MASK
;
3315 vcpu
->arch
.hflags
&= ~HF_SMM_INSIDE_NMI_MASK
;
3316 if (lapic_in_kernel(vcpu
)) {
3317 if (events
->smi
.latched_init
)
3318 set_bit(KVM_APIC_INIT
, &vcpu
->arch
.apic
->pending_events
);
3320 clear_bit(KVM_APIC_INIT
, &vcpu
->arch
.apic
->pending_events
);
3325 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
3330 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu
*vcpu
,
3331 struct kvm_debugregs
*dbgregs
)
3335 memcpy(dbgregs
->db
, vcpu
->arch
.db
, sizeof(vcpu
->arch
.db
));
3336 kvm_get_dr(vcpu
, 6, &val
);
3338 dbgregs
->dr7
= vcpu
->arch
.dr7
;
3340 memset(&dbgregs
->reserved
, 0, sizeof(dbgregs
->reserved
));
3343 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu
*vcpu
,
3344 struct kvm_debugregs
*dbgregs
)
3349 if (dbgregs
->dr6
& ~0xffffffffull
)
3351 if (dbgregs
->dr7
& ~0xffffffffull
)
3354 memcpy(vcpu
->arch
.db
, dbgregs
->db
, sizeof(vcpu
->arch
.db
));
3355 kvm_update_dr0123(vcpu
);
3356 vcpu
->arch
.dr6
= dbgregs
->dr6
;
3357 kvm_update_dr6(vcpu
);
3358 vcpu
->arch
.dr7
= dbgregs
->dr7
;
3359 kvm_update_dr7(vcpu
);
3364 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3366 static void fill_xsave(u8
*dest
, struct kvm_vcpu
*vcpu
)
3368 struct xregs_state
*xsave
= &vcpu
->arch
.guest_fpu
.state
.xsave
;
3369 u64 xstate_bv
= xsave
->header
.xfeatures
;
3373 * Copy legacy XSAVE area, to avoid complications with CPUID
3374 * leaves 0 and 1 in the loop below.
3376 memcpy(dest
, xsave
, XSAVE_HDR_OFFSET
);
3379 xstate_bv
&= vcpu
->arch
.guest_supported_xcr0
| XFEATURE_MASK_FPSSE
;
3380 *(u64
*)(dest
+ XSAVE_HDR_OFFSET
) = xstate_bv
;
3383 * Copy each region from the possibly compacted offset to the
3384 * non-compacted offset.
3386 valid
= xstate_bv
& ~XFEATURE_MASK_FPSSE
;
3388 u64 feature
= valid
& -valid
;
3389 int index
= fls64(feature
) - 1;
3390 void *src
= get_xsave_addr(xsave
, feature
);
3393 u32 size
, offset
, ecx
, edx
;
3394 cpuid_count(XSTATE_CPUID
, index
,
3395 &size
, &offset
, &ecx
, &edx
);
3396 if (feature
== XFEATURE_MASK_PKRU
)
3397 memcpy(dest
+ offset
, &vcpu
->arch
.pkru
,
3398 sizeof(vcpu
->arch
.pkru
));
3400 memcpy(dest
+ offset
, src
, size
);
3408 static void load_xsave(struct kvm_vcpu
*vcpu
, u8
*src
)
3410 struct xregs_state
*xsave
= &vcpu
->arch
.guest_fpu
.state
.xsave
;
3411 u64 xstate_bv
= *(u64
*)(src
+ XSAVE_HDR_OFFSET
);
3415 * Copy legacy XSAVE area, to avoid complications with CPUID
3416 * leaves 0 and 1 in the loop below.
3418 memcpy(xsave
, src
, XSAVE_HDR_OFFSET
);
3420 /* Set XSTATE_BV and possibly XCOMP_BV. */
3421 xsave
->header
.xfeatures
= xstate_bv
;
3422 if (boot_cpu_has(X86_FEATURE_XSAVES
))
3423 xsave
->header
.xcomp_bv
= host_xcr0
| XSTATE_COMPACTION_ENABLED
;
3426 * Copy each region from the non-compacted offset to the
3427 * possibly compacted offset.
3429 valid
= xstate_bv
& ~XFEATURE_MASK_FPSSE
;
3431 u64 feature
= valid
& -valid
;
3432 int index
= fls64(feature
) - 1;
3433 void *dest
= get_xsave_addr(xsave
, feature
);
3436 u32 size
, offset
, ecx
, edx
;
3437 cpuid_count(XSTATE_CPUID
, index
,
3438 &size
, &offset
, &ecx
, &edx
);
3439 if (feature
== XFEATURE_MASK_PKRU
)
3440 memcpy(&vcpu
->arch
.pkru
, src
+ offset
,
3441 sizeof(vcpu
->arch
.pkru
));
3443 memcpy(dest
, src
+ offset
, size
);
3450 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu
*vcpu
,
3451 struct kvm_xsave
*guest_xsave
)
3453 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
3454 memset(guest_xsave
, 0, sizeof(struct kvm_xsave
));
3455 fill_xsave((u8
*) guest_xsave
->region
, vcpu
);
3457 memcpy(guest_xsave
->region
,
3458 &vcpu
->arch
.guest_fpu
.state
.fxsave
,
3459 sizeof(struct fxregs_state
));
3460 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)] =
3461 XFEATURE_MASK_FPSSE
;
3465 #define XSAVE_MXCSR_OFFSET 24
3467 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu
*vcpu
,
3468 struct kvm_xsave
*guest_xsave
)
3471 *(u64
*)&guest_xsave
->region
[XSAVE_HDR_OFFSET
/ sizeof(u32
)];
3472 u32 mxcsr
= *(u32
*)&guest_xsave
->region
[XSAVE_MXCSR_OFFSET
/ sizeof(u32
)];
3474 if (boot_cpu_has(X86_FEATURE_XSAVE
)) {
3476 * Here we allow setting states that are not present in
3477 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3478 * with old userspace.
3480 if (xstate_bv
& ~kvm_supported_xcr0() ||
3481 mxcsr
& ~mxcsr_feature_mask
)
3483 load_xsave(vcpu
, (u8
*)guest_xsave
->region
);
3485 if (xstate_bv
& ~XFEATURE_MASK_FPSSE
||
3486 mxcsr
& ~mxcsr_feature_mask
)
3488 memcpy(&vcpu
->arch
.guest_fpu
.state
.fxsave
,
3489 guest_xsave
->region
, sizeof(struct fxregs_state
));
3494 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu
*vcpu
,
3495 struct kvm_xcrs
*guest_xcrs
)
3497 if (!boot_cpu_has(X86_FEATURE_XSAVE
)) {
3498 guest_xcrs
->nr_xcrs
= 0;
3502 guest_xcrs
->nr_xcrs
= 1;
3503 guest_xcrs
->flags
= 0;
3504 guest_xcrs
->xcrs
[0].xcr
= XCR_XFEATURE_ENABLED_MASK
;
3505 guest_xcrs
->xcrs
[0].value
= vcpu
->arch
.xcr0
;
3508 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu
*vcpu
,
3509 struct kvm_xcrs
*guest_xcrs
)
3513 if (!boot_cpu_has(X86_FEATURE_XSAVE
))
3516 if (guest_xcrs
->nr_xcrs
> KVM_MAX_XCRS
|| guest_xcrs
->flags
)
3519 for (i
= 0; i
< guest_xcrs
->nr_xcrs
; i
++)
3520 /* Only support XCR0 currently */
3521 if (guest_xcrs
->xcrs
[i
].xcr
== XCR_XFEATURE_ENABLED_MASK
) {
3522 r
= __kvm_set_xcr(vcpu
, XCR_XFEATURE_ENABLED_MASK
,
3523 guest_xcrs
->xcrs
[i
].value
);
3532 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3533 * stopped by the hypervisor. This function will be called from the host only.
3534 * EINVAL is returned when the host attempts to set the flag for a guest that
3535 * does not support pv clocks.
3537 static int kvm_set_guest_paused(struct kvm_vcpu
*vcpu
)
3539 if (!vcpu
->arch
.pv_time_enabled
)
3541 vcpu
->arch
.pvclock_set_guest_stopped_request
= true;
3542 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
3546 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu
*vcpu
,
3547 struct kvm_enable_cap
*cap
)
3553 case KVM_CAP_HYPERV_SYNIC2
:
3556 case KVM_CAP_HYPERV_SYNIC
:
3557 if (!irqchip_in_kernel(vcpu
->kvm
))
3559 return kvm_hv_activate_synic(vcpu
, cap
->cap
==
3560 KVM_CAP_HYPERV_SYNIC2
);
3566 long kvm_arch_vcpu_ioctl(struct file
*filp
,
3567 unsigned int ioctl
, unsigned long arg
)
3569 struct kvm_vcpu
*vcpu
= filp
->private_data
;
3570 void __user
*argp
= (void __user
*)arg
;
3573 struct kvm_lapic_state
*lapic
;
3574 struct kvm_xsave
*xsave
;
3575 struct kvm_xcrs
*xcrs
;
3581 case KVM_GET_LAPIC
: {
3583 if (!lapic_in_kernel(vcpu
))
3585 u
.lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
3590 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, u
.lapic
);
3594 if (copy_to_user(argp
, u
.lapic
, sizeof(struct kvm_lapic_state
)))
3599 case KVM_SET_LAPIC
: {
3601 if (!lapic_in_kernel(vcpu
))
3603 u
.lapic
= memdup_user(argp
, sizeof(*u
.lapic
));
3604 if (IS_ERR(u
.lapic
))
3605 return PTR_ERR(u
.lapic
);
3607 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, u
.lapic
);
3610 case KVM_INTERRUPT
: {
3611 struct kvm_interrupt irq
;
3614 if (copy_from_user(&irq
, argp
, sizeof irq
))
3616 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
3620 r
= kvm_vcpu_ioctl_nmi(vcpu
);
3624 r
= kvm_vcpu_ioctl_smi(vcpu
);
3627 case KVM_SET_CPUID
: {
3628 struct kvm_cpuid __user
*cpuid_arg
= argp
;
3629 struct kvm_cpuid cpuid
;
3632 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3634 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
3637 case KVM_SET_CPUID2
: {
3638 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3639 struct kvm_cpuid2 cpuid
;
3642 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3644 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
3645 cpuid_arg
->entries
);
3648 case KVM_GET_CPUID2
: {
3649 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
3650 struct kvm_cpuid2 cpuid
;
3653 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
3655 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
3656 cpuid_arg
->entries
);
3660 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
3665 case KVM_GET_MSRS
: {
3666 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3667 r
= msr_io(vcpu
, argp
, do_get_msr
, 1);
3668 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3671 case KVM_SET_MSRS
: {
3672 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3673 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
3674 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3677 case KVM_TPR_ACCESS_REPORTING
: {
3678 struct kvm_tpr_access_ctl tac
;
3681 if (copy_from_user(&tac
, argp
, sizeof tac
))
3683 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
3687 if (copy_to_user(argp
, &tac
, sizeof tac
))
3692 case KVM_SET_VAPIC_ADDR
: {
3693 struct kvm_vapic_addr va
;
3697 if (!lapic_in_kernel(vcpu
))
3700 if (copy_from_user(&va
, argp
, sizeof va
))
3702 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
3703 r
= kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
3704 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
3707 case KVM_X86_SETUP_MCE
: {
3711 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
3713 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
3716 case KVM_X86_SET_MCE
: {
3717 struct kvm_x86_mce mce
;
3720 if (copy_from_user(&mce
, argp
, sizeof mce
))
3722 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
3725 case KVM_GET_VCPU_EVENTS
: {
3726 struct kvm_vcpu_events events
;
3728 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu
, &events
);
3731 if (copy_to_user(argp
, &events
, sizeof(struct kvm_vcpu_events
)))
3736 case KVM_SET_VCPU_EVENTS
: {
3737 struct kvm_vcpu_events events
;
3740 if (copy_from_user(&events
, argp
, sizeof(struct kvm_vcpu_events
)))
3743 r
= kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu
, &events
);
3746 case KVM_GET_DEBUGREGS
: {
3747 struct kvm_debugregs dbgregs
;
3749 kvm_vcpu_ioctl_x86_get_debugregs(vcpu
, &dbgregs
);
3752 if (copy_to_user(argp
, &dbgregs
,
3753 sizeof(struct kvm_debugregs
)))
3758 case KVM_SET_DEBUGREGS
: {
3759 struct kvm_debugregs dbgregs
;
3762 if (copy_from_user(&dbgregs
, argp
,
3763 sizeof(struct kvm_debugregs
)))
3766 r
= kvm_vcpu_ioctl_x86_set_debugregs(vcpu
, &dbgregs
);
3769 case KVM_GET_XSAVE
: {
3770 u
.xsave
= kzalloc(sizeof(struct kvm_xsave
), GFP_KERNEL
);
3775 kvm_vcpu_ioctl_x86_get_xsave(vcpu
, u
.xsave
);
3778 if (copy_to_user(argp
, u
.xsave
, sizeof(struct kvm_xsave
)))
3783 case KVM_SET_XSAVE
: {
3784 u
.xsave
= memdup_user(argp
, sizeof(*u
.xsave
));
3785 if (IS_ERR(u
.xsave
))
3786 return PTR_ERR(u
.xsave
);
3788 r
= kvm_vcpu_ioctl_x86_set_xsave(vcpu
, u
.xsave
);
3791 case KVM_GET_XCRS
: {
3792 u
.xcrs
= kzalloc(sizeof(struct kvm_xcrs
), GFP_KERNEL
);
3797 kvm_vcpu_ioctl_x86_get_xcrs(vcpu
, u
.xcrs
);
3800 if (copy_to_user(argp
, u
.xcrs
,
3801 sizeof(struct kvm_xcrs
)))
3806 case KVM_SET_XCRS
: {
3807 u
.xcrs
= memdup_user(argp
, sizeof(*u
.xcrs
));
3809 return PTR_ERR(u
.xcrs
);
3811 r
= kvm_vcpu_ioctl_x86_set_xcrs(vcpu
, u
.xcrs
);
3814 case KVM_SET_TSC_KHZ
: {
3818 user_tsc_khz
= (u32
)arg
;
3820 if (user_tsc_khz
>= kvm_max_guest_tsc_khz
)
3823 if (user_tsc_khz
== 0)
3824 user_tsc_khz
= tsc_khz
;
3826 if (!kvm_set_tsc_khz(vcpu
, user_tsc_khz
))
3831 case KVM_GET_TSC_KHZ
: {
3832 r
= vcpu
->arch
.virtual_tsc_khz
;
3835 case KVM_KVMCLOCK_CTRL
: {
3836 r
= kvm_set_guest_paused(vcpu
);
3839 case KVM_ENABLE_CAP
: {
3840 struct kvm_enable_cap cap
;
3843 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
3845 r
= kvm_vcpu_ioctl_enable_cap(vcpu
, &cap
);
3856 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
3858 return VM_FAULT_SIGBUS
;
3861 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
3865 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
3867 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
3871 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm
*kvm
,
3874 kvm
->arch
.ept_identity_map_addr
= ident_addr
;
3878 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
3879 u32 kvm_nr_mmu_pages
)
3881 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
3884 mutex_lock(&kvm
->slots_lock
);
3886 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
3887 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
3889 mutex_unlock(&kvm
->slots_lock
);
3893 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
3895 return kvm
->arch
.n_max_mmu_pages
;
3898 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3900 struct kvm_pic
*pic
= kvm
->arch
.vpic
;
3904 switch (chip
->chip_id
) {
3905 case KVM_IRQCHIP_PIC_MASTER
:
3906 memcpy(&chip
->chip
.pic
, &pic
->pics
[0],
3907 sizeof(struct kvm_pic_state
));
3909 case KVM_IRQCHIP_PIC_SLAVE
:
3910 memcpy(&chip
->chip
.pic
, &pic
->pics
[1],
3911 sizeof(struct kvm_pic_state
));
3913 case KVM_IRQCHIP_IOAPIC
:
3914 kvm_get_ioapic(kvm
, &chip
->chip
.ioapic
);
3923 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
3925 struct kvm_pic
*pic
= kvm
->arch
.vpic
;
3929 switch (chip
->chip_id
) {
3930 case KVM_IRQCHIP_PIC_MASTER
:
3931 spin_lock(&pic
->lock
);
3932 memcpy(&pic
->pics
[0], &chip
->chip
.pic
,
3933 sizeof(struct kvm_pic_state
));
3934 spin_unlock(&pic
->lock
);
3936 case KVM_IRQCHIP_PIC_SLAVE
:
3937 spin_lock(&pic
->lock
);
3938 memcpy(&pic
->pics
[1], &chip
->chip
.pic
,
3939 sizeof(struct kvm_pic_state
));
3940 spin_unlock(&pic
->lock
);
3942 case KVM_IRQCHIP_IOAPIC
:
3943 kvm_set_ioapic(kvm
, &chip
->chip
.ioapic
);
3949 kvm_pic_update_irq(pic
);
3953 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3955 struct kvm_kpit_state
*kps
= &kvm
->arch
.vpit
->pit_state
;
3957 BUILD_BUG_ON(sizeof(*ps
) != sizeof(kps
->channels
));
3959 mutex_lock(&kps
->lock
);
3960 memcpy(ps
, &kps
->channels
, sizeof(*ps
));
3961 mutex_unlock(&kps
->lock
);
3965 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
3968 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3970 mutex_lock(&pit
->pit_state
.lock
);
3971 memcpy(&pit
->pit_state
.channels
, ps
, sizeof(*ps
));
3972 for (i
= 0; i
< 3; i
++)
3973 kvm_pit_load_count(pit
, i
, ps
->channels
[i
].count
, 0);
3974 mutex_unlock(&pit
->pit_state
.lock
);
3978 static int kvm_vm_ioctl_get_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3980 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
3981 memcpy(ps
->channels
, &kvm
->arch
.vpit
->pit_state
.channels
,
3982 sizeof(ps
->channels
));
3983 ps
->flags
= kvm
->arch
.vpit
->pit_state
.flags
;
3984 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
3985 memset(&ps
->reserved
, 0, sizeof(ps
->reserved
));
3989 static int kvm_vm_ioctl_set_pit2(struct kvm
*kvm
, struct kvm_pit_state2
*ps
)
3993 u32 prev_legacy
, cur_legacy
;
3994 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
3996 mutex_lock(&pit
->pit_state
.lock
);
3997 prev_legacy
= pit
->pit_state
.flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3998 cur_legacy
= ps
->flags
& KVM_PIT_FLAGS_HPET_LEGACY
;
3999 if (!prev_legacy
&& cur_legacy
)
4001 memcpy(&pit
->pit_state
.channels
, &ps
->channels
,
4002 sizeof(pit
->pit_state
.channels
));
4003 pit
->pit_state
.flags
= ps
->flags
;
4004 for (i
= 0; i
< 3; i
++)
4005 kvm_pit_load_count(pit
, i
, pit
->pit_state
.channels
[i
].count
,
4007 mutex_unlock(&pit
->pit_state
.lock
);
4011 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
4012 struct kvm_reinject_control
*control
)
4014 struct kvm_pit
*pit
= kvm
->arch
.vpit
;
4019 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4020 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4021 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4023 mutex_lock(&pit
->pit_state
.lock
);
4024 kvm_pit_set_reinject(pit
, control
->pit_reinject
);
4025 mutex_unlock(&pit
->pit_state
.lock
);
4031 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4032 * @kvm: kvm instance
4033 * @log: slot id and address to which we copy the log
4035 * Steps 1-4 below provide general overview of dirty page logging. See
4036 * kvm_get_dirty_log_protect() function description for additional details.
4038 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4039 * always flush the TLB (step 4) even if previous step failed and the dirty
4040 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4041 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4042 * writes will be marked dirty for next log read.
4044 * 1. Take a snapshot of the bit and clear it if needed.
4045 * 2. Write protect the corresponding page.
4046 * 3. Copy the snapshot to the userspace.
4047 * 4. Flush TLB's if needed.
4049 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
4051 bool is_dirty
= false;
4054 mutex_lock(&kvm
->slots_lock
);
4057 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4059 if (kvm_x86_ops
->flush_log_dirty
)
4060 kvm_x86_ops
->flush_log_dirty(kvm
);
4062 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
4065 * All the TLBs can be flushed out of mmu lock, see the comments in
4066 * kvm_mmu_slot_remove_write_access().
4068 lockdep_assert_held(&kvm
->slots_lock
);
4070 kvm_flush_remote_tlbs(kvm
);
4072 mutex_unlock(&kvm
->slots_lock
);
4076 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
,
4079 if (!irqchip_in_kernel(kvm
))
4082 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
4083 irq_event
->irq
, irq_event
->level
,
4088 static int kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
4089 struct kvm_enable_cap
*cap
)
4097 case KVM_CAP_DISABLE_QUIRKS
:
4098 kvm
->arch
.disabled_quirks
= cap
->args
[0];
4101 case KVM_CAP_SPLIT_IRQCHIP
: {
4102 mutex_lock(&kvm
->lock
);
4104 if (cap
->args
[0] > MAX_NR_RESERVED_IOAPIC_PINS
)
4105 goto split_irqchip_unlock
;
4107 if (irqchip_in_kernel(kvm
))
4108 goto split_irqchip_unlock
;
4109 if (kvm
->created_vcpus
)
4110 goto split_irqchip_unlock
;
4111 r
= kvm_setup_empty_irq_routing(kvm
);
4113 goto split_irqchip_unlock
;
4114 /* Pairs with irqchip_in_kernel. */
4116 kvm
->arch
.irqchip_mode
= KVM_IRQCHIP_SPLIT
;
4117 kvm
->arch
.nr_reserved_ioapic_pins
= cap
->args
[0];
4119 split_irqchip_unlock
:
4120 mutex_unlock(&kvm
->lock
);
4123 case KVM_CAP_X2APIC_API
:
4125 if (cap
->args
[0] & ~KVM_X2APIC_API_VALID_FLAGS
)
4128 if (cap
->args
[0] & KVM_X2APIC_API_USE_32BIT_IDS
)
4129 kvm
->arch
.x2apic_format
= true;
4130 if (cap
->args
[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK
)
4131 kvm
->arch
.x2apic_broadcast_quirk_disabled
= true;
4142 long kvm_arch_vm_ioctl(struct file
*filp
,
4143 unsigned int ioctl
, unsigned long arg
)
4145 struct kvm
*kvm
= filp
->private_data
;
4146 void __user
*argp
= (void __user
*)arg
;
4149 * This union makes it completely explicit to gcc-3.x
4150 * that these two variables' stack usage should be
4151 * combined, not added together.
4154 struct kvm_pit_state ps
;
4155 struct kvm_pit_state2 ps2
;
4156 struct kvm_pit_config pit_config
;
4160 case KVM_SET_TSS_ADDR
:
4161 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
4163 case KVM_SET_IDENTITY_MAP_ADDR
: {
4166 mutex_lock(&kvm
->lock
);
4168 if (kvm
->created_vcpus
)
4169 goto set_identity_unlock
;
4171 if (copy_from_user(&ident_addr
, argp
, sizeof ident_addr
))
4172 goto set_identity_unlock
;
4173 r
= kvm_vm_ioctl_set_identity_map_addr(kvm
, ident_addr
);
4174 set_identity_unlock
:
4175 mutex_unlock(&kvm
->lock
);
4178 case KVM_SET_NR_MMU_PAGES
:
4179 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
4181 case KVM_GET_NR_MMU_PAGES
:
4182 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
4184 case KVM_CREATE_IRQCHIP
: {
4185 mutex_lock(&kvm
->lock
);
4188 if (irqchip_in_kernel(kvm
))
4189 goto create_irqchip_unlock
;
4192 if (kvm
->created_vcpus
)
4193 goto create_irqchip_unlock
;
4195 r
= kvm_pic_init(kvm
);
4197 goto create_irqchip_unlock
;
4199 r
= kvm_ioapic_init(kvm
);
4201 kvm_pic_destroy(kvm
);
4202 goto create_irqchip_unlock
;
4205 r
= kvm_setup_default_irq_routing(kvm
);
4207 kvm_ioapic_destroy(kvm
);
4208 kvm_pic_destroy(kvm
);
4209 goto create_irqchip_unlock
;
4211 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4213 kvm
->arch
.irqchip_mode
= KVM_IRQCHIP_KERNEL
;
4214 create_irqchip_unlock
:
4215 mutex_unlock(&kvm
->lock
);
4218 case KVM_CREATE_PIT
:
4219 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
4221 case KVM_CREATE_PIT2
:
4223 if (copy_from_user(&u
.pit_config
, argp
,
4224 sizeof(struct kvm_pit_config
)))
4227 mutex_lock(&kvm
->lock
);
4230 goto create_pit_unlock
;
4232 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
4236 mutex_unlock(&kvm
->lock
);
4238 case KVM_GET_IRQCHIP
: {
4239 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4240 struct kvm_irqchip
*chip
;
4242 chip
= memdup_user(argp
, sizeof(*chip
));
4249 if (!irqchip_kernel(kvm
))
4250 goto get_irqchip_out
;
4251 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
4253 goto get_irqchip_out
;
4255 if (copy_to_user(argp
, chip
, sizeof *chip
))
4256 goto get_irqchip_out
;
4262 case KVM_SET_IRQCHIP
: {
4263 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4264 struct kvm_irqchip
*chip
;
4266 chip
= memdup_user(argp
, sizeof(*chip
));
4273 if (!irqchip_kernel(kvm
))
4274 goto set_irqchip_out
;
4275 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
4277 goto set_irqchip_out
;
4285 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
4288 if (!kvm
->arch
.vpit
)
4290 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
4294 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
4301 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
4304 if (!kvm
->arch
.vpit
)
4306 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
4309 case KVM_GET_PIT2
: {
4311 if (!kvm
->arch
.vpit
)
4313 r
= kvm_vm_ioctl_get_pit2(kvm
, &u
.ps2
);
4317 if (copy_to_user(argp
, &u
.ps2
, sizeof(u
.ps2
)))
4322 case KVM_SET_PIT2
: {
4324 if (copy_from_user(&u
.ps2
, argp
, sizeof(u
.ps2
)))
4327 if (!kvm
->arch
.vpit
)
4329 r
= kvm_vm_ioctl_set_pit2(kvm
, &u
.ps2
);
4332 case KVM_REINJECT_CONTROL
: {
4333 struct kvm_reinject_control control
;
4335 if (copy_from_user(&control
, argp
, sizeof(control
)))
4337 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
4340 case KVM_SET_BOOT_CPU_ID
:
4342 mutex_lock(&kvm
->lock
);
4343 if (kvm
->created_vcpus
)
4346 kvm
->arch
.bsp_vcpu_id
= arg
;
4347 mutex_unlock(&kvm
->lock
);
4349 case KVM_XEN_HVM_CONFIG
: {
4350 struct kvm_xen_hvm_config xhc
;
4352 if (copy_from_user(&xhc
, argp
, sizeof(xhc
)))
4357 memcpy(&kvm
->arch
.xen_hvm_config
, &xhc
, sizeof(xhc
));
4361 case KVM_SET_CLOCK
: {
4362 struct kvm_clock_data user_ns
;
4366 if (copy_from_user(&user_ns
, argp
, sizeof(user_ns
)))
4375 * TODO: userspace has to take care of races with VCPU_RUN, so
4376 * kvm_gen_update_masterclock() can be cut down to locked
4377 * pvclock_update_vm_gtod_copy().
4379 kvm_gen_update_masterclock(kvm
);
4380 now_ns
= get_kvmclock_ns(kvm
);
4381 kvm
->arch
.kvmclock_offset
+= user_ns
.clock
- now_ns
;
4382 kvm_make_all_cpus_request(kvm
, KVM_REQ_CLOCK_UPDATE
);
4385 case KVM_GET_CLOCK
: {
4386 struct kvm_clock_data user_ns
;
4389 now_ns
= get_kvmclock_ns(kvm
);
4390 user_ns
.clock
= now_ns
;
4391 user_ns
.flags
= kvm
->arch
.use_master_clock
? KVM_CLOCK_TSC_STABLE
: 0;
4392 memset(&user_ns
.pad
, 0, sizeof(user_ns
.pad
));
4395 if (copy_to_user(argp
, &user_ns
, sizeof(user_ns
)))
4400 case KVM_ENABLE_CAP
: {
4401 struct kvm_enable_cap cap
;
4404 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
4406 r
= kvm_vm_ioctl_enable_cap(kvm
, &cap
);
4416 static void kvm_init_msr_list(void)
4421 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
4422 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
4426 * Even MSRs that are valid in the host may not be exposed
4427 * to the guests in some cases.
4429 switch (msrs_to_save
[i
]) {
4430 case MSR_IA32_BNDCFGS
:
4431 if (!kvm_x86_ops
->mpx_supported())
4435 if (!kvm_x86_ops
->rdtscp_supported())
4443 msrs_to_save
[j
] = msrs_to_save
[i
];
4446 num_msrs_to_save
= j
;
4448 for (i
= j
= 0; i
< ARRAY_SIZE(emulated_msrs
); i
++) {
4449 if (!kvm_x86_ops
->has_emulated_msr(emulated_msrs
[i
]))
4453 emulated_msrs
[j
] = emulated_msrs
[i
];
4456 num_emulated_msrs
= j
;
4458 for (i
= j
= 0; i
< ARRAY_SIZE(msr_based_features
); i
++) {
4459 struct kvm_msr_entry msr
;
4461 msr
.index
= msr_based_features
[i
];
4462 if (kvm_get_msr_feature(&msr
))
4466 msr_based_features
[j
] = msr_based_features
[i
];
4469 num_msr_based_features
= j
;
4472 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
4480 if (!(lapic_in_kernel(vcpu
) &&
4481 !kvm_iodevice_write(vcpu
, &vcpu
->arch
.apic
->dev
, addr
, n
, v
))
4482 && kvm_io_bus_write(vcpu
, KVM_MMIO_BUS
, addr
, n
, v
))
4493 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
4500 if (!(lapic_in_kernel(vcpu
) &&
4501 !kvm_iodevice_read(vcpu
, &vcpu
->arch
.apic
->dev
,
4503 && kvm_io_bus_read(vcpu
, KVM_MMIO_BUS
, addr
, n
, v
))
4505 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, n
, addr
, v
);
4515 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
4516 struct kvm_segment
*var
, int seg
)
4518 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
4521 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
4522 struct kvm_segment
*var
, int seg
)
4524 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
4527 gpa_t
translate_nested_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
, u32 access
,
4528 struct x86_exception
*exception
)
4532 BUG_ON(!mmu_is_nested(vcpu
));
4534 /* NPT walks are always user-walks */
4535 access
|= PFERR_USER_MASK
;
4536 t_gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gpa
, access
, exception
);
4541 gpa_t
kvm_mmu_gva_to_gpa_read(struct kvm_vcpu
*vcpu
, gva_t gva
,
4542 struct x86_exception
*exception
)
4544 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4545 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4548 gpa_t
kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu
*vcpu
, gva_t gva
,
4549 struct x86_exception
*exception
)
4551 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4552 access
|= PFERR_FETCH_MASK
;
4553 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4556 gpa_t
kvm_mmu_gva_to_gpa_write(struct kvm_vcpu
*vcpu
, gva_t gva
,
4557 struct x86_exception
*exception
)
4559 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4560 access
|= PFERR_WRITE_MASK
;
4561 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4564 /* uses this to access any guest's mapped memory without checking CPL */
4565 gpa_t
kvm_mmu_gva_to_gpa_system(struct kvm_vcpu
*vcpu
, gva_t gva
,
4566 struct x86_exception
*exception
)
4568 return vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, 0, exception
);
4571 static int kvm_read_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
4572 struct kvm_vcpu
*vcpu
, u32 access
,
4573 struct x86_exception
*exception
)
4576 int r
= X86EMUL_CONTINUE
;
4579 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
,
4581 unsigned offset
= addr
& (PAGE_SIZE
-1);
4582 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
4585 if (gpa
== UNMAPPED_GVA
)
4586 return X86EMUL_PROPAGATE_FAULT
;
4587 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa
>> PAGE_SHIFT
, data
,
4590 r
= X86EMUL_IO_NEEDED
;
4602 /* used for instruction fetching */
4603 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt
*ctxt
,
4604 gva_t addr
, void *val
, unsigned int bytes
,
4605 struct x86_exception
*exception
)
4607 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4608 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4612 /* Inline kvm_read_guest_virt_helper for speed. */
4613 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
, access
|PFERR_FETCH_MASK
,
4615 if (unlikely(gpa
== UNMAPPED_GVA
))
4616 return X86EMUL_PROPAGATE_FAULT
;
4618 offset
= addr
& (PAGE_SIZE
-1);
4619 if (WARN_ON(offset
+ bytes
> PAGE_SIZE
))
4620 bytes
= (unsigned)PAGE_SIZE
- offset
;
4621 ret
= kvm_vcpu_read_guest_page(vcpu
, gpa
>> PAGE_SHIFT
, val
,
4623 if (unlikely(ret
< 0))
4624 return X86EMUL_IO_NEEDED
;
4626 return X86EMUL_CONTINUE
;
4629 int kvm_read_guest_virt(struct kvm_vcpu
*vcpu
,
4630 gva_t addr
, void *val
, unsigned int bytes
,
4631 struct x86_exception
*exception
)
4633 u32 access
= (kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0;
4636 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
4637 * is returned, but our callers are not ready for that and they blindly
4638 * call kvm_inject_page_fault. Ensure that they at least do not leak
4639 * uninitialized kernel stack memory into cr2 and error code.
4641 memset(exception
, 0, sizeof(*exception
));
4642 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
,
4645 EXPORT_SYMBOL_GPL(kvm_read_guest_virt
);
4647 static int emulator_read_std(struct x86_emulate_ctxt
*ctxt
,
4648 gva_t addr
, void *val
, unsigned int bytes
,
4649 struct x86_exception
*exception
, bool system
)
4651 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4654 if (!system
&& kvm_x86_ops
->get_cpl(vcpu
) == 3)
4655 access
|= PFERR_USER_MASK
;
4657 return kvm_read_guest_virt_helper(addr
, val
, bytes
, vcpu
, access
, exception
);
4660 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt
*ctxt
,
4661 unsigned long addr
, void *val
, unsigned int bytes
)
4663 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4664 int r
= kvm_vcpu_read_guest(vcpu
, addr
, val
, bytes
);
4666 return r
< 0 ? X86EMUL_IO_NEEDED
: X86EMUL_CONTINUE
;
4669 static int kvm_write_guest_virt_helper(gva_t addr
, void *val
, unsigned int bytes
,
4670 struct kvm_vcpu
*vcpu
, u32 access
,
4671 struct x86_exception
*exception
)
4674 int r
= X86EMUL_CONTINUE
;
4676 /* kvm_write_guest_virt_system can pull in tons of pages. */
4677 vcpu
->arch
.l1tf_flush_l1d
= true;
4680 gpa_t gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, addr
,
4683 unsigned offset
= addr
& (PAGE_SIZE
-1);
4684 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
4687 if (gpa
== UNMAPPED_GVA
)
4688 return X86EMUL_PROPAGATE_FAULT
;
4689 ret
= kvm_vcpu_write_guest(vcpu
, gpa
, data
, towrite
);
4691 r
= X86EMUL_IO_NEEDED
;
4703 static int emulator_write_std(struct x86_emulate_ctxt
*ctxt
, gva_t addr
, void *val
,
4704 unsigned int bytes
, struct x86_exception
*exception
,
4707 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4708 u32 access
= PFERR_WRITE_MASK
;
4710 if (!system
&& kvm_x86_ops
->get_cpl(vcpu
) == 3)
4711 access
|= PFERR_USER_MASK
;
4713 return kvm_write_guest_virt_helper(addr
, val
, bytes
, vcpu
,
4717 int kvm_write_guest_virt_system(struct kvm_vcpu
*vcpu
, gva_t addr
, void *val
,
4718 unsigned int bytes
, struct x86_exception
*exception
)
4720 return kvm_write_guest_virt_helper(addr
, val
, bytes
, vcpu
,
4721 PFERR_WRITE_MASK
, exception
);
4723 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system
);
4725 static int vcpu_is_mmio_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
4726 gpa_t gpa
, bool write
)
4728 /* For APIC access vmexit */
4729 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
4732 if (vcpu_match_mmio_gpa(vcpu
, gpa
)) {
4733 trace_vcpu_match_mmio(gva
, gpa
, write
, true);
4740 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu
*vcpu
, unsigned long gva
,
4741 gpa_t
*gpa
, struct x86_exception
*exception
,
4744 u32 access
= ((kvm_x86_ops
->get_cpl(vcpu
) == 3) ? PFERR_USER_MASK
: 0)
4745 | (write
? PFERR_WRITE_MASK
: 0);
4748 * currently PKRU is only applied to ept enabled guest so
4749 * there is no pkey in EPT page table for L1 guest or EPT
4750 * shadow page table for L2 guest.
4752 if (vcpu_match_mmio_gva(vcpu
, gva
)
4753 && !permission_fault(vcpu
, vcpu
->arch
.walk_mmu
,
4754 vcpu
->arch
.access
, 0, access
)) {
4755 *gpa
= vcpu
->arch
.mmio_gfn
<< PAGE_SHIFT
|
4756 (gva
& (PAGE_SIZE
- 1));
4757 trace_vcpu_match_mmio(gva
, *gpa
, write
, false);
4761 *gpa
= vcpu
->arch
.walk_mmu
->gva_to_gpa(vcpu
, gva
, access
, exception
);
4763 if (*gpa
== UNMAPPED_GVA
)
4766 return vcpu_is_mmio_gpa(vcpu
, gva
, *gpa
, write
);
4769 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4770 const void *val
, int bytes
)
4774 ret
= kvm_vcpu_write_guest(vcpu
, gpa
, val
, bytes
);
4777 kvm_page_track_write(vcpu
, gpa
, val
, bytes
);
4781 struct read_write_emulator_ops
{
4782 int (*read_write_prepare
)(struct kvm_vcpu
*vcpu
, void *val
,
4784 int (*read_write_emulate
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4785 void *val
, int bytes
);
4786 int (*read_write_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4787 int bytes
, void *val
);
4788 int (*read_write_exit_mmio
)(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4789 void *val
, int bytes
);
4793 static int read_prepare(struct kvm_vcpu
*vcpu
, void *val
, int bytes
)
4795 if (vcpu
->mmio_read_completed
) {
4796 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
4797 vcpu
->mmio_fragments
[0].gpa
, val
);
4798 vcpu
->mmio_read_completed
= 0;
4805 static int read_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4806 void *val
, int bytes
)
4808 return !kvm_vcpu_read_guest(vcpu
, gpa
, val
, bytes
);
4811 static int write_emulate(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4812 void *val
, int bytes
)
4814 return emulator_write_phys(vcpu
, gpa
, val
, bytes
);
4817 static int write_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
, void *val
)
4819 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, val
);
4820 return vcpu_mmio_write(vcpu
, gpa
, bytes
, val
);
4823 static int read_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4824 void *val
, int bytes
)
4826 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, NULL
);
4827 return X86EMUL_IO_NEEDED
;
4830 static int write_exit_mmio(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
4831 void *val
, int bytes
)
4833 struct kvm_mmio_fragment
*frag
= &vcpu
->mmio_fragments
[0];
4835 memcpy(vcpu
->run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
4836 return X86EMUL_CONTINUE
;
4839 static const struct read_write_emulator_ops read_emultor
= {
4840 .read_write_prepare
= read_prepare
,
4841 .read_write_emulate
= read_emulate
,
4842 .read_write_mmio
= vcpu_mmio_read
,
4843 .read_write_exit_mmio
= read_exit_mmio
,
4846 static const struct read_write_emulator_ops write_emultor
= {
4847 .read_write_emulate
= write_emulate
,
4848 .read_write_mmio
= write_mmio
,
4849 .read_write_exit_mmio
= write_exit_mmio
,
4853 static int emulator_read_write_onepage(unsigned long addr
, void *val
,
4855 struct x86_exception
*exception
,
4856 struct kvm_vcpu
*vcpu
,
4857 const struct read_write_emulator_ops
*ops
)
4861 bool write
= ops
->write
;
4862 struct kvm_mmio_fragment
*frag
;
4863 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
4866 * If the exit was due to a NPF we may already have a GPA.
4867 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4868 * Note, this cannot be used on string operations since string
4869 * operation using rep will only have the initial GPA from the NPF
4872 if (vcpu
->arch
.gpa_available
&&
4873 emulator_can_use_gpa(ctxt
) &&
4874 (addr
& ~PAGE_MASK
) == (vcpu
->arch
.gpa_val
& ~PAGE_MASK
)) {
4875 gpa
= vcpu
->arch
.gpa_val
;
4876 ret
= vcpu_is_mmio_gpa(vcpu
, addr
, gpa
, write
);
4878 ret
= vcpu_mmio_gva_to_gpa(vcpu
, addr
, &gpa
, exception
, write
);
4880 return X86EMUL_PROPAGATE_FAULT
;
4883 if (!ret
&& ops
->read_write_emulate(vcpu
, gpa
, val
, bytes
))
4884 return X86EMUL_CONTINUE
;
4887 * Is this MMIO handled locally?
4889 handled
= ops
->read_write_mmio(vcpu
, gpa
, bytes
, val
);
4890 if (handled
== bytes
)
4891 return X86EMUL_CONTINUE
;
4897 WARN_ON(vcpu
->mmio_nr_fragments
>= KVM_MAX_MMIO_FRAGMENTS
);
4898 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_nr_fragments
++];
4902 return X86EMUL_CONTINUE
;
4905 static int emulator_read_write(struct x86_emulate_ctxt
*ctxt
,
4907 void *val
, unsigned int bytes
,
4908 struct x86_exception
*exception
,
4909 const struct read_write_emulator_ops
*ops
)
4911 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
4915 if (ops
->read_write_prepare
&&
4916 ops
->read_write_prepare(vcpu
, val
, bytes
))
4917 return X86EMUL_CONTINUE
;
4919 vcpu
->mmio_nr_fragments
= 0;
4921 /* Crossing a page boundary? */
4922 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
4925 now
= -addr
& ~PAGE_MASK
;
4926 rc
= emulator_read_write_onepage(addr
, val
, now
, exception
,
4929 if (rc
!= X86EMUL_CONTINUE
)
4932 if (ctxt
->mode
!= X86EMUL_MODE_PROT64
)
4938 rc
= emulator_read_write_onepage(addr
, val
, bytes
, exception
,
4940 if (rc
!= X86EMUL_CONTINUE
)
4943 if (!vcpu
->mmio_nr_fragments
)
4946 gpa
= vcpu
->mmio_fragments
[0].gpa
;
4948 vcpu
->mmio_needed
= 1;
4949 vcpu
->mmio_cur_fragment
= 0;
4951 vcpu
->run
->mmio
.len
= min(8u, vcpu
->mmio_fragments
[0].len
);
4952 vcpu
->run
->mmio
.is_write
= vcpu
->mmio_is_write
= ops
->write
;
4953 vcpu
->run
->exit_reason
= KVM_EXIT_MMIO
;
4954 vcpu
->run
->mmio
.phys_addr
= gpa
;
4956 return ops
->read_write_exit_mmio(vcpu
, gpa
, val
, bytes
);
4959 static int emulator_read_emulated(struct x86_emulate_ctxt
*ctxt
,
4963 struct x86_exception
*exception
)
4965 return emulator_read_write(ctxt
, addr
, val
, bytes
,
4966 exception
, &read_emultor
);
4969 static int emulator_write_emulated(struct x86_emulate_ctxt
*ctxt
,
4973 struct x86_exception
*exception
)
4975 return emulator_read_write(ctxt
, addr
, (void *)val
, bytes
,
4976 exception
, &write_emultor
);
4979 #define CMPXCHG_TYPE(t, ptr, old, new) \
4980 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4982 #ifdef CONFIG_X86_64
4983 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4985 # define CMPXCHG64(ptr, old, new) \
4986 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4989 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt
*ctxt
,
4994 struct x86_exception
*exception
)
4996 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5002 /* guests cmpxchg8b have to be emulated atomically */
5003 if (bytes
> 8 || (bytes
& (bytes
- 1)))
5006 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, addr
, NULL
);
5008 if (gpa
== UNMAPPED_GVA
||
5009 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
5012 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
5015 page
= kvm_vcpu_gfn_to_page(vcpu
, gpa
>> PAGE_SHIFT
);
5016 if (is_error_page(page
))
5019 kaddr
= kmap_atomic(page
);
5020 kaddr
+= offset_in_page(gpa
);
5023 exchanged
= CMPXCHG_TYPE(u8
, kaddr
, old
, new);
5026 exchanged
= CMPXCHG_TYPE(u16
, kaddr
, old
, new);
5029 exchanged
= CMPXCHG_TYPE(u32
, kaddr
, old
, new);
5032 exchanged
= CMPXCHG64(kaddr
, old
, new);
5037 kunmap_atomic(kaddr
);
5038 kvm_release_page_dirty(page
);
5041 return X86EMUL_CMPXCHG_FAILED
;
5043 kvm_vcpu_mark_page_dirty(vcpu
, gpa
>> PAGE_SHIFT
);
5044 kvm_page_track_write(vcpu
, gpa
, new, bytes
);
5046 return X86EMUL_CONTINUE
;
5049 printk_once(KERN_WARNING
"kvm: emulating exchange as write\n");
5051 return emulator_write_emulated(ctxt
, addr
, new, bytes
, exception
);
5054 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
5058 for (i
= 0; i
< vcpu
->arch
.pio
.count
; i
++) {
5059 if (vcpu
->arch
.pio
.in
)
5060 r
= kvm_io_bus_read(vcpu
, KVM_PIO_BUS
, vcpu
->arch
.pio
.port
,
5061 vcpu
->arch
.pio
.size
, pd
);
5063 r
= kvm_io_bus_write(vcpu
, KVM_PIO_BUS
,
5064 vcpu
->arch
.pio
.port
, vcpu
->arch
.pio
.size
,
5068 pd
+= vcpu
->arch
.pio
.size
;
5073 static int emulator_pio_in_out(struct kvm_vcpu
*vcpu
, int size
,
5074 unsigned short port
, void *val
,
5075 unsigned int count
, bool in
)
5077 vcpu
->arch
.pio
.port
= port
;
5078 vcpu
->arch
.pio
.in
= in
;
5079 vcpu
->arch
.pio
.count
= count
;
5080 vcpu
->arch
.pio
.size
= size
;
5082 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
5083 vcpu
->arch
.pio
.count
= 0;
5087 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
5088 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
5089 vcpu
->run
->io
.size
= size
;
5090 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
5091 vcpu
->run
->io
.count
= count
;
5092 vcpu
->run
->io
.port
= port
;
5097 static int emulator_pio_in_emulated(struct x86_emulate_ctxt
*ctxt
,
5098 int size
, unsigned short port
, void *val
,
5101 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5104 if (vcpu
->arch
.pio
.count
)
5107 memset(vcpu
->arch
.pio_data
, 0, size
* count
);
5109 ret
= emulator_pio_in_out(vcpu
, size
, port
, val
, count
, true);
5112 memcpy(val
, vcpu
->arch
.pio_data
, size
* count
);
5113 trace_kvm_pio(KVM_PIO_IN
, port
, size
, count
, vcpu
->arch
.pio_data
);
5114 vcpu
->arch
.pio
.count
= 0;
5121 static int emulator_pio_out_emulated(struct x86_emulate_ctxt
*ctxt
,
5122 int size
, unsigned short port
,
5123 const void *val
, unsigned int count
)
5125 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5127 memcpy(vcpu
->arch
.pio_data
, val
, size
* count
);
5128 trace_kvm_pio(KVM_PIO_OUT
, port
, size
, count
, vcpu
->arch
.pio_data
);
5129 return emulator_pio_in_out(vcpu
, size
, port
, (void *)val
, count
, false);
5132 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
5134 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
5137 static void emulator_invlpg(struct x86_emulate_ctxt
*ctxt
, ulong address
)
5139 kvm_mmu_invlpg(emul_to_vcpu(ctxt
), address
);
5142 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu
*vcpu
)
5144 if (!need_emulate_wbinvd(vcpu
))
5145 return X86EMUL_CONTINUE
;
5147 if (kvm_x86_ops
->has_wbinvd_exit()) {
5148 int cpu
= get_cpu();
5150 cpumask_set_cpu(cpu
, vcpu
->arch
.wbinvd_dirty_mask
);
5151 smp_call_function_many(vcpu
->arch
.wbinvd_dirty_mask
,
5152 wbinvd_ipi
, NULL
, 1);
5154 cpumask_clear(vcpu
->arch
.wbinvd_dirty_mask
);
5157 return X86EMUL_CONTINUE
;
5160 int kvm_emulate_wbinvd(struct kvm_vcpu
*vcpu
)
5162 kvm_emulate_wbinvd_noskip(vcpu
);
5163 return kvm_skip_emulated_instruction(vcpu
);
5165 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd
);
5169 static void emulator_wbinvd(struct x86_emulate_ctxt
*ctxt
)
5171 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt
));
5174 static int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
,
5175 unsigned long *dest
)
5177 return kvm_get_dr(emul_to_vcpu(ctxt
), dr
, dest
);
5180 static int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
,
5181 unsigned long value
)
5184 return __kvm_set_dr(emul_to_vcpu(ctxt
), dr
, value
);
5187 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
5189 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
5192 static unsigned long emulator_get_cr(struct x86_emulate_ctxt
*ctxt
, int cr
)
5194 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5195 unsigned long value
;
5199 value
= kvm_read_cr0(vcpu
);
5202 value
= vcpu
->arch
.cr2
;
5205 value
= kvm_read_cr3(vcpu
);
5208 value
= kvm_read_cr4(vcpu
);
5211 value
= kvm_get_cr8(vcpu
);
5214 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
5221 static int emulator_set_cr(struct x86_emulate_ctxt
*ctxt
, int cr
, ulong val
)
5223 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5228 res
= kvm_set_cr0(vcpu
, mk_cr_64(kvm_read_cr0(vcpu
), val
));
5231 vcpu
->arch
.cr2
= val
;
5234 res
= kvm_set_cr3(vcpu
, val
);
5237 res
= kvm_set_cr4(vcpu
, mk_cr_64(kvm_read_cr4(vcpu
), val
));
5240 res
= kvm_set_cr8(vcpu
, val
);
5243 kvm_err("%s: unexpected cr %u\n", __func__
, cr
);
5250 static int emulator_get_cpl(struct x86_emulate_ctxt
*ctxt
)
5252 return kvm_x86_ops
->get_cpl(emul_to_vcpu(ctxt
));
5255 static void emulator_get_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5257 kvm_x86_ops
->get_gdt(emul_to_vcpu(ctxt
), dt
);
5260 static void emulator_get_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5262 kvm_x86_ops
->get_idt(emul_to_vcpu(ctxt
), dt
);
5265 static void emulator_set_gdt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5267 kvm_x86_ops
->set_gdt(emul_to_vcpu(ctxt
), dt
);
5270 static void emulator_set_idt(struct x86_emulate_ctxt
*ctxt
, struct desc_ptr
*dt
)
5272 kvm_x86_ops
->set_idt(emul_to_vcpu(ctxt
), dt
);
5275 static unsigned long emulator_get_cached_segment_base(
5276 struct x86_emulate_ctxt
*ctxt
, int seg
)
5278 return get_segment_base(emul_to_vcpu(ctxt
), seg
);
5281 static bool emulator_get_segment(struct x86_emulate_ctxt
*ctxt
, u16
*selector
,
5282 struct desc_struct
*desc
, u32
*base3
,
5285 struct kvm_segment var
;
5287 kvm_get_segment(emul_to_vcpu(ctxt
), &var
, seg
);
5288 *selector
= var
.selector
;
5291 memset(desc
, 0, sizeof(*desc
));
5299 set_desc_limit(desc
, var
.limit
);
5300 set_desc_base(desc
, (unsigned long)var
.base
);
5301 #ifdef CONFIG_X86_64
5303 *base3
= var
.base
>> 32;
5305 desc
->type
= var
.type
;
5307 desc
->dpl
= var
.dpl
;
5308 desc
->p
= var
.present
;
5309 desc
->avl
= var
.avl
;
5317 static void emulator_set_segment(struct x86_emulate_ctxt
*ctxt
, u16 selector
,
5318 struct desc_struct
*desc
, u32 base3
,
5321 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5322 struct kvm_segment var
;
5324 var
.selector
= selector
;
5325 var
.base
= get_desc_base(desc
);
5326 #ifdef CONFIG_X86_64
5327 var
.base
|= ((u64
)base3
) << 32;
5329 var
.limit
= get_desc_limit(desc
);
5331 var
.limit
= (var
.limit
<< 12) | 0xfff;
5332 var
.type
= desc
->type
;
5333 var
.dpl
= desc
->dpl
;
5338 var
.avl
= desc
->avl
;
5339 var
.present
= desc
->p
;
5340 var
.unusable
= !var
.present
;
5343 kvm_set_segment(vcpu
, &var
, seg
);
5347 static int emulator_get_msr(struct x86_emulate_ctxt
*ctxt
,
5348 u32 msr_index
, u64
*pdata
)
5350 struct msr_data msr
;
5353 msr
.index
= msr_index
;
5354 msr
.host_initiated
= false;
5355 r
= kvm_get_msr(emul_to_vcpu(ctxt
), &msr
);
5363 static int emulator_set_msr(struct x86_emulate_ctxt
*ctxt
,
5364 u32 msr_index
, u64 data
)
5366 struct msr_data msr
;
5369 msr
.index
= msr_index
;
5370 msr
.host_initiated
= false;
5371 return kvm_set_msr(emul_to_vcpu(ctxt
), &msr
);
5374 static u64
emulator_get_smbase(struct x86_emulate_ctxt
*ctxt
)
5376 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5378 return vcpu
->arch
.smbase
;
5381 static void emulator_set_smbase(struct x86_emulate_ctxt
*ctxt
, u64 smbase
)
5383 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5385 vcpu
->arch
.smbase
= smbase
;
5388 static int emulator_check_pmc(struct x86_emulate_ctxt
*ctxt
,
5391 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt
), pmc
);
5394 static int emulator_read_pmc(struct x86_emulate_ctxt
*ctxt
,
5395 u32 pmc
, u64
*pdata
)
5397 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt
), pmc
, pdata
);
5400 static void emulator_halt(struct x86_emulate_ctxt
*ctxt
)
5402 emul_to_vcpu(ctxt
)->arch
.halt_request
= 1;
5405 static int emulator_intercept(struct x86_emulate_ctxt
*ctxt
,
5406 struct x86_instruction_info
*info
,
5407 enum x86_intercept_stage stage
)
5409 return kvm_x86_ops
->check_intercept(emul_to_vcpu(ctxt
), info
, stage
);
5412 static bool emulator_get_cpuid(struct x86_emulate_ctxt
*ctxt
,
5413 u32
*eax
, u32
*ebx
, u32
*ecx
, u32
*edx
, bool check_limit
)
5415 return kvm_cpuid(emul_to_vcpu(ctxt
), eax
, ebx
, ecx
, edx
, check_limit
);
5418 static ulong
emulator_read_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
)
5420 return kvm_register_read(emul_to_vcpu(ctxt
), reg
);
5423 static void emulator_write_gpr(struct x86_emulate_ctxt
*ctxt
, unsigned reg
, ulong val
)
5425 kvm_register_write(emul_to_vcpu(ctxt
), reg
, val
);
5428 static void emulator_set_nmi_mask(struct x86_emulate_ctxt
*ctxt
, bool masked
)
5430 kvm_x86_ops
->set_nmi_mask(emul_to_vcpu(ctxt
), masked
);
5433 static unsigned emulator_get_hflags(struct x86_emulate_ctxt
*ctxt
)
5435 return emul_to_vcpu(ctxt
)->arch
.hflags
;
5438 static void emulator_set_hflags(struct x86_emulate_ctxt
*ctxt
, unsigned emul_flags
)
5440 kvm_set_hflags(emul_to_vcpu(ctxt
), emul_flags
);
5443 static int emulator_pre_leave_smm(struct x86_emulate_ctxt
*ctxt
, u64 smbase
)
5445 return kvm_x86_ops
->pre_leave_smm(emul_to_vcpu(ctxt
), smbase
);
5448 static const struct x86_emulate_ops emulate_ops
= {
5449 .read_gpr
= emulator_read_gpr
,
5450 .write_gpr
= emulator_write_gpr
,
5451 .read_std
= emulator_read_std
,
5452 .write_std
= emulator_write_std
,
5453 .read_phys
= kvm_read_guest_phys_system
,
5454 .fetch
= kvm_fetch_guest_virt
,
5455 .read_emulated
= emulator_read_emulated
,
5456 .write_emulated
= emulator_write_emulated
,
5457 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
5458 .invlpg
= emulator_invlpg
,
5459 .pio_in_emulated
= emulator_pio_in_emulated
,
5460 .pio_out_emulated
= emulator_pio_out_emulated
,
5461 .get_segment
= emulator_get_segment
,
5462 .set_segment
= emulator_set_segment
,
5463 .get_cached_segment_base
= emulator_get_cached_segment_base
,
5464 .get_gdt
= emulator_get_gdt
,
5465 .get_idt
= emulator_get_idt
,
5466 .set_gdt
= emulator_set_gdt
,
5467 .set_idt
= emulator_set_idt
,
5468 .get_cr
= emulator_get_cr
,
5469 .set_cr
= emulator_set_cr
,
5470 .cpl
= emulator_get_cpl
,
5471 .get_dr
= emulator_get_dr
,
5472 .set_dr
= emulator_set_dr
,
5473 .get_smbase
= emulator_get_smbase
,
5474 .set_smbase
= emulator_set_smbase
,
5475 .set_msr
= emulator_set_msr
,
5476 .get_msr
= emulator_get_msr
,
5477 .check_pmc
= emulator_check_pmc
,
5478 .read_pmc
= emulator_read_pmc
,
5479 .halt
= emulator_halt
,
5480 .wbinvd
= emulator_wbinvd
,
5481 .fix_hypercall
= emulator_fix_hypercall
,
5482 .intercept
= emulator_intercept
,
5483 .get_cpuid
= emulator_get_cpuid
,
5484 .set_nmi_mask
= emulator_set_nmi_mask
,
5485 .get_hflags
= emulator_get_hflags
,
5486 .set_hflags
= emulator_set_hflags
,
5487 .pre_leave_smm
= emulator_pre_leave_smm
,
5490 static void toggle_interruptibility(struct kvm_vcpu
*vcpu
, u32 mask
)
5492 u32 int_shadow
= kvm_x86_ops
->get_interrupt_shadow(vcpu
);
5494 * an sti; sti; sequence only disable interrupts for the first
5495 * instruction. So, if the last instruction, be it emulated or
5496 * not, left the system with the INT_STI flag enabled, it
5497 * means that the last instruction is an sti. We should not
5498 * leave the flag on in this case. The same goes for mov ss
5500 if (int_shadow
& mask
)
5502 if (unlikely(int_shadow
|| mask
)) {
5503 kvm_x86_ops
->set_interrupt_shadow(vcpu
, mask
);
5505 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5509 static bool inject_emulated_exception(struct kvm_vcpu
*vcpu
)
5511 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5512 if (ctxt
->exception
.vector
== PF_VECTOR
)
5513 return kvm_propagate_fault(vcpu
, &ctxt
->exception
);
5515 if (ctxt
->exception
.error_code_valid
)
5516 kvm_queue_exception_e(vcpu
, ctxt
->exception
.vector
,
5517 ctxt
->exception
.error_code
);
5519 kvm_queue_exception(vcpu
, ctxt
->exception
.vector
);
5523 static void init_emulate_ctxt(struct kvm_vcpu
*vcpu
)
5525 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5528 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
5530 ctxt
->eflags
= kvm_get_rflags(vcpu
);
5531 ctxt
->tf
= (ctxt
->eflags
& X86_EFLAGS_TF
) != 0;
5533 ctxt
->eip
= kvm_rip_read(vcpu
);
5534 ctxt
->mode
= (!is_protmode(vcpu
)) ? X86EMUL_MODE_REAL
:
5535 (ctxt
->eflags
& X86_EFLAGS_VM
) ? X86EMUL_MODE_VM86
:
5536 (cs_l
&& is_long_mode(vcpu
)) ? X86EMUL_MODE_PROT64
:
5537 cs_db
? X86EMUL_MODE_PROT32
:
5538 X86EMUL_MODE_PROT16
;
5539 BUILD_BUG_ON(HF_GUEST_MASK
!= X86EMUL_GUEST_MASK
);
5540 BUILD_BUG_ON(HF_SMM_MASK
!= X86EMUL_SMM_MASK
);
5541 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK
!= X86EMUL_SMM_INSIDE_NMI_MASK
);
5543 init_decode_cache(ctxt
);
5544 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
5547 int kvm_inject_realmode_interrupt(struct kvm_vcpu
*vcpu
, int irq
, int inc_eip
)
5549 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5552 init_emulate_ctxt(vcpu
);
5556 ctxt
->_eip
= ctxt
->eip
+ inc_eip
;
5557 ret
= emulate_int_real(ctxt
, irq
);
5559 if (ret
!= X86EMUL_CONTINUE
)
5560 return EMULATE_FAIL
;
5562 ctxt
->eip
= ctxt
->_eip
;
5563 kvm_rip_write(vcpu
, ctxt
->eip
);
5564 kvm_set_rflags(vcpu
, ctxt
->eflags
);
5566 if (irq
== NMI_VECTOR
)
5567 vcpu
->arch
.nmi_pending
= 0;
5569 vcpu
->arch
.interrupt
.pending
= false;
5571 return EMULATE_DONE
;
5573 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt
);
5575 static int handle_emulation_failure(struct kvm_vcpu
*vcpu
)
5577 int r
= EMULATE_DONE
;
5579 ++vcpu
->stat
.insn_emulation_fail
;
5580 trace_kvm_emulate_insn_failed(vcpu
);
5581 if (!is_guest_mode(vcpu
) && kvm_x86_ops
->get_cpl(vcpu
) == 0) {
5582 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
5583 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
5584 vcpu
->run
->internal
.ndata
= 0;
5585 r
= EMULATE_USER_EXIT
;
5587 kvm_queue_exception(vcpu
, UD_VECTOR
);
5592 static bool reexecute_instruction(struct kvm_vcpu
*vcpu
, gva_t cr2
,
5593 bool write_fault_to_shadow_pgtable
,
5599 if (emulation_type
& EMULTYPE_NO_REEXECUTE
)
5602 if (!vcpu
->arch
.mmu
.direct_map
) {
5604 * Write permission should be allowed since only
5605 * write access need to be emulated.
5607 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
5610 * If the mapping is invalid in guest, let cpu retry
5611 * it to generate fault.
5613 if (gpa
== UNMAPPED_GVA
)
5618 * Do not retry the unhandleable instruction if it faults on the
5619 * readonly host memory, otherwise it will goto a infinite loop:
5620 * retry instruction -> write #PF -> emulation fail -> retry
5621 * instruction -> ...
5623 pfn
= gfn_to_pfn(vcpu
->kvm
, gpa_to_gfn(gpa
));
5626 * If the instruction failed on the error pfn, it can not be fixed,
5627 * report the error to userspace.
5629 if (is_error_noslot_pfn(pfn
))
5632 kvm_release_pfn_clean(pfn
);
5634 /* The instructions are well-emulated on direct mmu. */
5635 if (vcpu
->arch
.mmu
.direct_map
) {
5636 unsigned int indirect_shadow_pages
;
5638 spin_lock(&vcpu
->kvm
->mmu_lock
);
5639 indirect_shadow_pages
= vcpu
->kvm
->arch
.indirect_shadow_pages
;
5640 spin_unlock(&vcpu
->kvm
->mmu_lock
);
5642 if (indirect_shadow_pages
)
5643 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5649 * if emulation was due to access to shadowed page table
5650 * and it failed try to unshadow page and re-enter the
5651 * guest to let CPU execute the instruction.
5653 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5656 * If the access faults on its page table, it can not
5657 * be fixed by unprotecting shadow page and it should
5658 * be reported to userspace.
5660 return !write_fault_to_shadow_pgtable
;
5663 static bool retry_instruction(struct x86_emulate_ctxt
*ctxt
,
5664 unsigned long cr2
, int emulation_type
)
5666 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
5667 unsigned long last_retry_eip
, last_retry_addr
, gpa
= cr2
;
5669 last_retry_eip
= vcpu
->arch
.last_retry_eip
;
5670 last_retry_addr
= vcpu
->arch
.last_retry_addr
;
5673 * If the emulation is caused by #PF and it is non-page_table
5674 * writing instruction, it means the VM-EXIT is caused by shadow
5675 * page protected, we can zap the shadow page and retry this
5676 * instruction directly.
5678 * Note: if the guest uses a non-page-table modifying instruction
5679 * on the PDE that points to the instruction, then we will unmap
5680 * the instruction and go to an infinite loop. So, we cache the
5681 * last retried eip and the last fault address, if we meet the eip
5682 * and the address again, we can break out of the potential infinite
5685 vcpu
->arch
.last_retry_eip
= vcpu
->arch
.last_retry_addr
= 0;
5687 if (!(emulation_type
& EMULTYPE_RETRY
))
5690 if (x86_page_table_writing_insn(ctxt
))
5693 if (ctxt
->eip
== last_retry_eip
&& last_retry_addr
== cr2
)
5696 vcpu
->arch
.last_retry_eip
= ctxt
->eip
;
5697 vcpu
->arch
.last_retry_addr
= cr2
;
5699 if (!vcpu
->arch
.mmu
.direct_map
)
5700 gpa
= kvm_mmu_gva_to_gpa_write(vcpu
, cr2
, NULL
);
5702 kvm_mmu_unprotect_page(vcpu
->kvm
, gpa_to_gfn(gpa
));
5707 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
);
5708 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
);
5710 static void kvm_smm_changed(struct kvm_vcpu
*vcpu
)
5712 if (!(vcpu
->arch
.hflags
& HF_SMM_MASK
)) {
5713 /* This is a good place to trace that we are exiting SMM. */
5714 trace_kvm_enter_smm(vcpu
->vcpu_id
, vcpu
->arch
.smbase
, false);
5716 /* Process a latched INIT or SMI, if any. */
5717 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5720 kvm_mmu_reset_context(vcpu
);
5723 static void kvm_set_hflags(struct kvm_vcpu
*vcpu
, unsigned emul_flags
)
5725 unsigned changed
= vcpu
->arch
.hflags
^ emul_flags
;
5727 vcpu
->arch
.hflags
= emul_flags
;
5729 if (changed
& HF_SMM_MASK
)
5730 kvm_smm_changed(vcpu
);
5733 static int kvm_vcpu_check_hw_bp(unsigned long addr
, u32 type
, u32 dr7
,
5742 for (i
= 0; i
< 4; i
++, enable
>>= 2, rwlen
>>= 4)
5743 if ((enable
& 3) && (rwlen
& 15) == type
&& db
[i
] == addr
)
5748 static void kvm_vcpu_do_singlestep(struct kvm_vcpu
*vcpu
, int *r
)
5750 struct kvm_run
*kvm_run
= vcpu
->run
;
5752 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
) {
5753 kvm_run
->debug
.arch
.dr6
= DR6_BS
| DR6_FIXED_1
| DR6_RTM
;
5754 kvm_run
->debug
.arch
.pc
= vcpu
->arch
.singlestep_rip
;
5755 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
5756 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
5757 *r
= EMULATE_USER_EXIT
;
5760 * "Certain debug exceptions may clear bit 0-3. The
5761 * remaining contents of the DR6 register are never
5762 * cleared by the processor".
5764 vcpu
->arch
.dr6
&= ~15;
5765 vcpu
->arch
.dr6
|= DR6_BS
| DR6_RTM
;
5766 kvm_queue_exception(vcpu
, DB_VECTOR
);
5770 int kvm_skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
5772 unsigned long rflags
= kvm_x86_ops
->get_rflags(vcpu
);
5773 int r
= EMULATE_DONE
;
5775 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
5778 * rflags is the old, "raw" value of the flags. The new value has
5779 * not been saved yet.
5781 * This is correct even for TF set by the guest, because "the
5782 * processor will not generate this exception after the instruction
5783 * that sets the TF flag".
5785 if (unlikely(rflags
& X86_EFLAGS_TF
))
5786 kvm_vcpu_do_singlestep(vcpu
, &r
);
5787 return r
== EMULATE_DONE
;
5789 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction
);
5791 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu
*vcpu
, int *r
)
5793 if (unlikely(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) &&
5794 (vcpu
->arch
.guest_debug_dr7
& DR7_BP_EN_MASK
)) {
5795 struct kvm_run
*kvm_run
= vcpu
->run
;
5796 unsigned long eip
= kvm_get_linear_rip(vcpu
);
5797 u32 dr6
= kvm_vcpu_check_hw_bp(eip
, 0,
5798 vcpu
->arch
.guest_debug_dr7
,
5802 kvm_run
->debug
.arch
.dr6
= dr6
| DR6_FIXED_1
| DR6_RTM
;
5803 kvm_run
->debug
.arch
.pc
= eip
;
5804 kvm_run
->debug
.arch
.exception
= DB_VECTOR
;
5805 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
5806 *r
= EMULATE_USER_EXIT
;
5811 if (unlikely(vcpu
->arch
.dr7
& DR7_BP_EN_MASK
) &&
5812 !(kvm_get_rflags(vcpu
) & X86_EFLAGS_RF
)) {
5813 unsigned long eip
= kvm_get_linear_rip(vcpu
);
5814 u32 dr6
= kvm_vcpu_check_hw_bp(eip
, 0,
5819 vcpu
->arch
.dr6
&= ~15;
5820 vcpu
->arch
.dr6
|= dr6
| DR6_RTM
;
5821 kvm_queue_exception(vcpu
, DB_VECTOR
);
5830 int x86_emulate_instruction(struct kvm_vcpu
*vcpu
,
5837 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
5838 bool writeback
= true;
5839 bool write_fault_to_spt
= vcpu
->arch
.write_fault_to_shadow_pgtable
;
5841 vcpu
->arch
.l1tf_flush_l1d
= true;
5844 * Clear write_fault_to_shadow_pgtable here to ensure it is
5847 vcpu
->arch
.write_fault_to_shadow_pgtable
= false;
5848 kvm_clear_exception_queue(vcpu
);
5850 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
5851 init_emulate_ctxt(vcpu
);
5854 * We will reenter on the same instruction since
5855 * we do not set complete_userspace_io. This does not
5856 * handle watchpoints yet, those would be handled in
5859 if (!(emulation_type
& EMULTYPE_SKIP
) &&
5860 kvm_vcpu_check_breakpoint(vcpu
, &r
))
5863 ctxt
->interruptibility
= 0;
5864 ctxt
->have_exception
= false;
5865 ctxt
->exception
.vector
= -1;
5866 ctxt
->perm_ok
= false;
5868 ctxt
->ud
= emulation_type
& EMULTYPE_TRAP_UD
;
5870 r
= x86_decode_insn(ctxt
, insn
, insn_len
);
5872 trace_kvm_emulate_insn_start(vcpu
);
5873 ++vcpu
->stat
.insn_emulation
;
5874 if (r
!= EMULATION_OK
) {
5875 if (emulation_type
& EMULTYPE_TRAP_UD
)
5876 return EMULATE_FAIL
;
5877 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
5879 return EMULATE_DONE
;
5880 if (ctxt
->have_exception
&& inject_emulated_exception(vcpu
))
5881 return EMULATE_DONE
;
5882 if (emulation_type
& EMULTYPE_SKIP
)
5883 return EMULATE_FAIL
;
5884 return handle_emulation_failure(vcpu
);
5888 if (emulation_type
& EMULTYPE_SKIP
) {
5889 kvm_rip_write(vcpu
, ctxt
->_eip
);
5890 if (ctxt
->eflags
& X86_EFLAGS_RF
)
5891 kvm_set_rflags(vcpu
, ctxt
->eflags
& ~X86_EFLAGS_RF
);
5892 return EMULATE_DONE
;
5895 if (retry_instruction(ctxt
, cr2
, emulation_type
))
5896 return EMULATE_DONE
;
5898 /* this is needed for vmware backdoor interface to work since it
5899 changes registers values during IO operation */
5900 if (vcpu
->arch
.emulate_regs_need_sync_from_vcpu
) {
5901 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= false;
5902 emulator_invalidate_register_cache(ctxt
);
5906 /* Save the faulting GPA (cr2) in the address field */
5907 ctxt
->exception
.address
= cr2
;
5909 r
= x86_emulate_insn(ctxt
);
5911 if (r
== EMULATION_INTERCEPTED
)
5912 return EMULATE_DONE
;
5914 if (r
== EMULATION_FAILED
) {
5915 if (reexecute_instruction(vcpu
, cr2
, write_fault_to_spt
,
5917 return EMULATE_DONE
;
5919 return handle_emulation_failure(vcpu
);
5922 if (ctxt
->have_exception
) {
5924 if (inject_emulated_exception(vcpu
))
5926 } else if (vcpu
->arch
.pio
.count
) {
5927 if (!vcpu
->arch
.pio
.in
) {
5928 /* FIXME: return into emulator if single-stepping. */
5929 vcpu
->arch
.pio
.count
= 0;
5932 vcpu
->arch
.complete_userspace_io
= complete_emulated_pio
;
5934 r
= EMULATE_USER_EXIT
;
5935 } else if (vcpu
->mmio_needed
) {
5936 if (!vcpu
->mmio_is_write
)
5938 r
= EMULATE_USER_EXIT
;
5939 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
5940 } else if (r
== EMULATION_RESTART
)
5946 unsigned long rflags
= kvm_x86_ops
->get_rflags(vcpu
);
5947 toggle_interruptibility(vcpu
, ctxt
->interruptibility
);
5948 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
5949 kvm_rip_write(vcpu
, ctxt
->eip
);
5950 if (r
== EMULATE_DONE
&&
5951 (ctxt
->tf
|| (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)))
5952 kvm_vcpu_do_singlestep(vcpu
, &r
);
5953 if (!ctxt
->have_exception
||
5954 exception_type(ctxt
->exception
.vector
) == EXCPT_TRAP
)
5955 __kvm_set_rflags(vcpu
, ctxt
->eflags
);
5958 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5959 * do nothing, and it will be requested again as soon as
5960 * the shadow expires. But we still need to check here,
5961 * because POPF has no interrupt shadow.
5963 if (unlikely((ctxt
->eflags
& ~rflags
) & X86_EFLAGS_IF
))
5964 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5966 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= true;
5970 EXPORT_SYMBOL_GPL(x86_emulate_instruction
);
5972 int kvm_fast_pio_out(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
5974 unsigned long val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
5975 int ret
= emulator_pio_out_emulated(&vcpu
->arch
.emulate_ctxt
,
5976 size
, port
, &val
, 1);
5977 /* do not return to emulator after return from userspace */
5978 vcpu
->arch
.pio
.count
= 0;
5981 EXPORT_SYMBOL_GPL(kvm_fast_pio_out
);
5983 static int complete_fast_pio_in(struct kvm_vcpu
*vcpu
)
5987 /* We should only ever be called with arch.pio.count equal to 1 */
5988 BUG_ON(vcpu
->arch
.pio
.count
!= 1);
5990 /* For size less than 4 we merge, else we zero extend */
5991 val
= (vcpu
->arch
.pio
.size
< 4) ? kvm_register_read(vcpu
, VCPU_REGS_RAX
)
5995 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5996 * the copy and tracing
5998 emulator_pio_in_emulated(&vcpu
->arch
.emulate_ctxt
, vcpu
->arch
.pio
.size
,
5999 vcpu
->arch
.pio
.port
, &val
, 1);
6000 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
6005 int kvm_fast_pio_in(struct kvm_vcpu
*vcpu
, int size
, unsigned short port
)
6010 /* For size less than 4 we merge, else we zero extend */
6011 val
= (size
< 4) ? kvm_register_read(vcpu
, VCPU_REGS_RAX
) : 0;
6013 ret
= emulator_pio_in_emulated(&vcpu
->arch
.emulate_ctxt
, size
, port
,
6016 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
6020 vcpu
->arch
.complete_userspace_io
= complete_fast_pio_in
;
6024 EXPORT_SYMBOL_GPL(kvm_fast_pio_in
);
6026 static int kvmclock_cpu_down_prep(unsigned int cpu
)
6028 __this_cpu_write(cpu_tsc_khz
, 0);
6032 static void tsc_khz_changed(void *data
)
6034 struct cpufreq_freqs
*freq
= data
;
6035 unsigned long khz
= 0;
6039 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
6040 khz
= cpufreq_quick_get(raw_smp_processor_id());
6043 __this_cpu_write(cpu_tsc_khz
, khz
);
6046 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
6049 struct cpufreq_freqs
*freq
= data
;
6051 struct kvm_vcpu
*vcpu
;
6052 int i
, send_ipi
= 0;
6055 * We allow guests to temporarily run on slowing clocks,
6056 * provided we notify them after, or to run on accelerating
6057 * clocks, provided we notify them before. Thus time never
6060 * However, we have a problem. We can't atomically update
6061 * the frequency of a given CPU from this function; it is
6062 * merely a notifier, which can be called from any CPU.
6063 * Changing the TSC frequency at arbitrary points in time
6064 * requires a recomputation of local variables related to
6065 * the TSC for each VCPU. We must flag these local variables
6066 * to be updated and be sure the update takes place with the
6067 * new frequency before any guests proceed.
6069 * Unfortunately, the combination of hotplug CPU and frequency
6070 * change creates an intractable locking scenario; the order
6071 * of when these callouts happen is undefined with respect to
6072 * CPU hotplug, and they can race with each other. As such,
6073 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6074 * undefined; you can actually have a CPU frequency change take
6075 * place in between the computation of X and the setting of the
6076 * variable. To protect against this problem, all updates of
6077 * the per_cpu tsc_khz variable are done in an interrupt
6078 * protected IPI, and all callers wishing to update the value
6079 * must wait for a synchronous IPI to complete (which is trivial
6080 * if the caller is on the CPU already). This establishes the
6081 * necessary total order on variable updates.
6083 * Note that because a guest time update may take place
6084 * anytime after the setting of the VCPU's request bit, the
6085 * correct TSC value must be set before the request. However,
6086 * to ensure the update actually makes it to any guest which
6087 * starts running in hardware virtualization between the set
6088 * and the acquisition of the spinlock, we must also ping the
6089 * CPU after setting the request bit.
6093 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
6095 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
6098 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
6100 spin_lock(&kvm_lock
);
6101 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
6102 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
6103 if (vcpu
->cpu
!= freq
->cpu
)
6105 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
6106 if (vcpu
->cpu
!= smp_processor_id())
6110 spin_unlock(&kvm_lock
);
6112 if (freq
->old
< freq
->new && send_ipi
) {
6114 * We upscale the frequency. Must make the guest
6115 * doesn't see old kvmclock values while running with
6116 * the new frequency, otherwise we risk the guest sees
6117 * time go backwards.
6119 * In case we update the frequency for another cpu
6120 * (which might be in guest context) send an interrupt
6121 * to kick the cpu out of guest context. Next time
6122 * guest context is entered kvmclock will be updated,
6123 * so the guest will not see stale values.
6125 smp_call_function_single(freq
->cpu
, tsc_khz_changed
, freq
, 1);
6130 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
6131 .notifier_call
= kvmclock_cpufreq_notifier
6134 static int kvmclock_cpu_online(unsigned int cpu
)
6136 tsc_khz_changed(NULL
);
6140 static void kvm_timer_init(void)
6142 max_tsc_khz
= tsc_khz
;
6144 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
6145 #ifdef CONFIG_CPU_FREQ
6146 struct cpufreq_policy policy
;
6149 memset(&policy
, 0, sizeof(policy
));
6151 cpufreq_get_policy(&policy
, cpu
);
6152 if (policy
.cpuinfo
.max_freq
)
6153 max_tsc_khz
= policy
.cpuinfo
.max_freq
;
6156 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
6157 CPUFREQ_TRANSITION_NOTIFIER
);
6159 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz
);
6161 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE
, "x86/kvm/clk:online",
6162 kvmclock_cpu_online
, kvmclock_cpu_down_prep
);
6165 static DEFINE_PER_CPU(struct kvm_vcpu
*, current_vcpu
);
6167 int kvm_is_in_guest(void)
6169 return __this_cpu_read(current_vcpu
) != NULL
;
6172 static int kvm_is_user_mode(void)
6176 if (__this_cpu_read(current_vcpu
))
6177 user_mode
= kvm_x86_ops
->get_cpl(__this_cpu_read(current_vcpu
));
6179 return user_mode
!= 0;
6182 static unsigned long kvm_get_guest_ip(void)
6184 unsigned long ip
= 0;
6186 if (__this_cpu_read(current_vcpu
))
6187 ip
= kvm_rip_read(__this_cpu_read(current_vcpu
));
6192 static struct perf_guest_info_callbacks kvm_guest_cbs
= {
6193 .is_in_guest
= kvm_is_in_guest
,
6194 .is_user_mode
= kvm_is_user_mode
,
6195 .get_guest_ip
= kvm_get_guest_ip
,
6198 void kvm_before_handle_nmi(struct kvm_vcpu
*vcpu
)
6200 __this_cpu_write(current_vcpu
, vcpu
);
6202 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi
);
6204 void kvm_after_handle_nmi(struct kvm_vcpu
*vcpu
)
6206 __this_cpu_write(current_vcpu
, NULL
);
6208 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi
);
6210 static void kvm_set_mmio_spte_mask(void)
6213 int maxphyaddr
= boot_cpu_data
.x86_phys_bits
;
6216 * Set the reserved bits and the present bit of an paging-structure
6217 * entry to generate page fault with PFER.RSV = 1.
6219 /* Mask the reserved physical address bits. */
6220 mask
= rsvd_bits(maxphyaddr
, 51);
6222 /* Set the present bit. */
6225 #ifdef CONFIG_X86_64
6227 * If reserved bit is not supported, clear the present bit to disable
6230 if (maxphyaddr
== 52)
6234 kvm_mmu_set_mmio_spte_mask(mask
, mask
);
6237 #ifdef CONFIG_X86_64
6238 static void pvclock_gtod_update_fn(struct work_struct
*work
)
6242 struct kvm_vcpu
*vcpu
;
6245 spin_lock(&kvm_lock
);
6246 list_for_each_entry(kvm
, &vm_list
, vm_list
)
6247 kvm_for_each_vcpu(i
, vcpu
, kvm
)
6248 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
6249 atomic_set(&kvm_guest_has_master_clock
, 0);
6250 spin_unlock(&kvm_lock
);
6253 static DECLARE_WORK(pvclock_gtod_work
, pvclock_gtod_update_fn
);
6256 * Notification about pvclock gtod data update.
6258 static int pvclock_gtod_notify(struct notifier_block
*nb
, unsigned long unused
,
6261 struct pvclock_gtod_data
*gtod
= &pvclock_gtod_data
;
6262 struct timekeeper
*tk
= priv
;
6264 update_pvclock_gtod(tk
);
6266 /* disable master clock if host does not trust, or does not
6267 * use, TSC clocksource
6269 if (gtod
->clock
.vclock_mode
!= VCLOCK_TSC
&&
6270 atomic_read(&kvm_guest_has_master_clock
) != 0)
6271 queue_work(system_long_wq
, &pvclock_gtod_work
);
6276 static struct notifier_block pvclock_gtod_notifier
= {
6277 .notifier_call
= pvclock_gtod_notify
,
6281 int kvm_arch_init(void *opaque
)
6284 struct kvm_x86_ops
*ops
= opaque
;
6287 printk(KERN_ERR
"kvm: already loaded the other module\n");
6292 if (!ops
->cpu_has_kvm_support()) {
6293 printk(KERN_ERR
"kvm: no hardware support\n");
6297 if (ops
->disabled_by_bios()) {
6298 printk(KERN_WARNING
"kvm: disabled by bios\n");
6304 shared_msrs
= alloc_percpu(struct kvm_shared_msrs
);
6306 printk(KERN_ERR
"kvm: failed to allocate percpu kvm_shared_msrs\n");
6310 r
= kvm_mmu_module_init();
6312 goto out_free_percpu
;
6314 kvm_set_mmio_spte_mask();
6318 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
6319 PT_DIRTY_MASK
, PT64_NX_MASK
, 0,
6320 PT_PRESENT_MASK
, 0, sme_me_mask
);
6323 perf_register_guest_info_callbacks(&kvm_guest_cbs
);
6325 if (boot_cpu_has(X86_FEATURE_XSAVE
))
6326 host_xcr0
= xgetbv(XCR_XFEATURE_ENABLED_MASK
);
6329 #ifdef CONFIG_X86_64
6330 pvclock_gtod_register_notifier(&pvclock_gtod_notifier
);
6336 free_percpu(shared_msrs
);
6341 void kvm_arch_exit(void)
6344 perf_unregister_guest_info_callbacks(&kvm_guest_cbs
);
6346 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
6347 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
6348 CPUFREQ_TRANSITION_NOTIFIER
);
6349 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE
);
6350 #ifdef CONFIG_X86_64
6351 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier
);
6354 kvm_mmu_module_exit();
6355 free_percpu(shared_msrs
);
6358 int kvm_vcpu_halt(struct kvm_vcpu
*vcpu
)
6360 ++vcpu
->stat
.halt_exits
;
6361 if (lapic_in_kernel(vcpu
)) {
6362 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
6365 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
6369 EXPORT_SYMBOL_GPL(kvm_vcpu_halt
);
6371 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
6373 int ret
= kvm_skip_emulated_instruction(vcpu
);
6375 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6376 * KVM_EXIT_DEBUG here.
6378 return kvm_vcpu_halt(vcpu
) && ret
;
6380 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
6382 #ifdef CONFIG_X86_64
6383 static int kvm_pv_clock_pairing(struct kvm_vcpu
*vcpu
, gpa_t paddr
,
6384 unsigned long clock_type
)
6386 struct kvm_clock_pairing clock_pairing
;
6391 if (clock_type
!= KVM_CLOCK_PAIRING_WALLCLOCK
)
6392 return -KVM_EOPNOTSUPP
;
6394 if (kvm_get_walltime_and_clockread(&ts
, &cycle
) == false)
6395 return -KVM_EOPNOTSUPP
;
6397 clock_pairing
.sec
= ts
.tv_sec
;
6398 clock_pairing
.nsec
= ts
.tv_nsec
;
6399 clock_pairing
.tsc
= kvm_read_l1_tsc(vcpu
, cycle
);
6400 clock_pairing
.flags
= 0;
6403 if (kvm_write_guest(vcpu
->kvm
, paddr
, &clock_pairing
,
6404 sizeof(struct kvm_clock_pairing
)))
6412 * kvm_pv_kick_cpu_op: Kick a vcpu.
6414 * @apicid - apicid of vcpu to be kicked.
6416 static void kvm_pv_kick_cpu_op(struct kvm
*kvm
, unsigned long flags
, int apicid
)
6418 struct kvm_lapic_irq lapic_irq
;
6420 lapic_irq
.shorthand
= 0;
6421 lapic_irq
.dest_mode
= 0;
6422 lapic_irq
.level
= 0;
6423 lapic_irq
.dest_id
= apicid
;
6424 lapic_irq
.msi_redir_hint
= false;
6426 lapic_irq
.delivery_mode
= APIC_DM_REMRD
;
6427 kvm_irq_delivery_to_apic(kvm
, NULL
, &lapic_irq
, NULL
);
6430 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu
*vcpu
)
6432 vcpu
->arch
.apicv_active
= false;
6433 kvm_x86_ops
->refresh_apicv_exec_ctrl(vcpu
);
6436 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
6438 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
6441 if (kvm_hv_hypercall_enabled(vcpu
->kvm
)) {
6442 if (!kvm_hv_hypercall(vcpu
))
6447 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
6448 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
6449 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
6450 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
6451 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
6453 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
6455 op_64_bit
= is_64_bit_mode(vcpu
);
6464 if (kvm_x86_ops
->get_cpl(vcpu
) != 0) {
6470 case KVM_HC_VAPIC_POLL_IRQ
:
6473 case KVM_HC_KICK_CPU
:
6474 kvm_pv_kick_cpu_op(vcpu
->kvm
, a0
, a1
);
6477 #ifdef CONFIG_X86_64
6478 case KVM_HC_CLOCK_PAIRING
:
6479 ret
= kvm_pv_clock_pairing(vcpu
, a0
, a1
);
6489 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
6492 ++vcpu
->stat
.hypercalls
;
6493 return kvm_skip_emulated_instruction(vcpu
);
6495 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
6497 static int emulator_fix_hypercall(struct x86_emulate_ctxt
*ctxt
)
6499 struct kvm_vcpu
*vcpu
= emul_to_vcpu(ctxt
);
6500 char instruction
[3];
6501 unsigned long rip
= kvm_rip_read(vcpu
);
6503 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
6505 return emulator_write_emulated(ctxt
, rip
, instruction
, 3,
6509 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
)
6511 return vcpu
->run
->request_interrupt_window
&&
6512 likely(!pic_in_kernel(vcpu
->kvm
));
6515 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
)
6517 struct kvm_run
*kvm_run
= vcpu
->run
;
6519 kvm_run
->if_flag
= (kvm_get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
6520 kvm_run
->flags
= is_smm(vcpu
) ? KVM_RUN_X86_SMM
: 0;
6521 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
6522 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
6523 kvm_run
->ready_for_interrupt_injection
=
6524 pic_in_kernel(vcpu
->kvm
) ||
6525 kvm_vcpu_ready_for_interrupt_injection(vcpu
);
6528 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
6532 if (!kvm_x86_ops
->update_cr8_intercept
)
6535 if (!lapic_in_kernel(vcpu
))
6538 if (vcpu
->arch
.apicv_active
)
6541 if (!vcpu
->arch
.apic
->vapic_addr
)
6542 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
6549 tpr
= kvm_lapic_get_cr8(vcpu
);
6551 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
6554 static int inject_pending_event(struct kvm_vcpu
*vcpu
, bool req_int_win
)
6558 /* try to reinject previous events if any */
6559 if (vcpu
->arch
.exception
.injected
) {
6560 kvm_x86_ops
->queue_exception(vcpu
);
6565 * Exceptions must be injected immediately, or the exception
6566 * frame will have the address of the NMI or interrupt handler.
6568 if (!vcpu
->arch
.exception
.pending
) {
6569 if (vcpu
->arch
.nmi_injected
) {
6570 kvm_x86_ops
->set_nmi(vcpu
);
6574 if (vcpu
->arch
.interrupt
.pending
) {
6575 kvm_x86_ops
->set_irq(vcpu
);
6580 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
) {
6581 r
= kvm_x86_ops
->check_nested_events(vcpu
, req_int_win
);
6586 /* try to inject new event if pending */
6587 if (vcpu
->arch
.exception
.pending
) {
6588 trace_kvm_inj_exception(vcpu
->arch
.exception
.nr
,
6589 vcpu
->arch
.exception
.has_error_code
,
6590 vcpu
->arch
.exception
.error_code
);
6592 vcpu
->arch
.exception
.pending
= false;
6593 vcpu
->arch
.exception
.injected
= true;
6595 if (exception_type(vcpu
->arch
.exception
.nr
) == EXCPT_FAULT
)
6596 __kvm_set_rflags(vcpu
, kvm_get_rflags(vcpu
) |
6599 if (vcpu
->arch
.exception
.nr
== DB_VECTOR
&&
6600 (vcpu
->arch
.dr7
& DR7_GD
)) {
6601 vcpu
->arch
.dr7
&= ~DR7_GD
;
6602 kvm_update_dr7(vcpu
);
6605 kvm_x86_ops
->queue_exception(vcpu
);
6606 } else if (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
) && kvm_x86_ops
->smi_allowed(vcpu
)) {
6607 vcpu
->arch
.smi_pending
= false;
6609 } else if (vcpu
->arch
.nmi_pending
&& kvm_x86_ops
->nmi_allowed(vcpu
)) {
6610 --vcpu
->arch
.nmi_pending
;
6611 vcpu
->arch
.nmi_injected
= true;
6612 kvm_x86_ops
->set_nmi(vcpu
);
6613 } else if (kvm_cpu_has_injectable_intr(vcpu
)) {
6615 * Because interrupts can be injected asynchronously, we are
6616 * calling check_nested_events again here to avoid a race condition.
6617 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6618 * proposal and current concerns. Perhaps we should be setting
6619 * KVM_REQ_EVENT only on certain events and not unconditionally?
6621 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
) {
6622 r
= kvm_x86_ops
->check_nested_events(vcpu
, req_int_win
);
6626 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
6627 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
6629 kvm_x86_ops
->set_irq(vcpu
);
6636 static void process_nmi(struct kvm_vcpu
*vcpu
)
6641 * x86 is limited to one NMI running, and one NMI pending after it.
6642 * If an NMI is already in progress, limit further NMIs to just one.
6643 * Otherwise, allow two (and we'll inject the first one immediately).
6645 if (kvm_x86_ops
->get_nmi_mask(vcpu
) || vcpu
->arch
.nmi_injected
)
6648 vcpu
->arch
.nmi_pending
+= atomic_xchg(&vcpu
->arch
.nmi_queued
, 0);
6649 vcpu
->arch
.nmi_pending
= min(vcpu
->arch
.nmi_pending
, limit
);
6650 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6653 static u32
enter_smm_get_segment_flags(struct kvm_segment
*seg
)
6656 flags
|= seg
->g
<< 23;
6657 flags
|= seg
->db
<< 22;
6658 flags
|= seg
->l
<< 21;
6659 flags
|= seg
->avl
<< 20;
6660 flags
|= seg
->present
<< 15;
6661 flags
|= seg
->dpl
<< 13;
6662 flags
|= seg
->s
<< 12;
6663 flags
|= seg
->type
<< 8;
6667 static void enter_smm_save_seg_32(struct kvm_vcpu
*vcpu
, char *buf
, int n
)
6669 struct kvm_segment seg
;
6672 kvm_get_segment(vcpu
, &seg
, n
);
6673 put_smstate(u32
, buf
, 0x7fa8 + n
* 4, seg
.selector
);
6676 offset
= 0x7f84 + n
* 12;
6678 offset
= 0x7f2c + (n
- 3) * 12;
6680 put_smstate(u32
, buf
, offset
+ 8, seg
.base
);
6681 put_smstate(u32
, buf
, offset
+ 4, seg
.limit
);
6682 put_smstate(u32
, buf
, offset
, enter_smm_get_segment_flags(&seg
));
6685 #ifdef CONFIG_X86_64
6686 static void enter_smm_save_seg_64(struct kvm_vcpu
*vcpu
, char *buf
, int n
)
6688 struct kvm_segment seg
;
6692 kvm_get_segment(vcpu
, &seg
, n
);
6693 offset
= 0x7e00 + n
* 16;
6695 flags
= enter_smm_get_segment_flags(&seg
) >> 8;
6696 put_smstate(u16
, buf
, offset
, seg
.selector
);
6697 put_smstate(u16
, buf
, offset
+ 2, flags
);
6698 put_smstate(u32
, buf
, offset
+ 4, seg
.limit
);
6699 put_smstate(u64
, buf
, offset
+ 8, seg
.base
);
6703 static void enter_smm_save_state_32(struct kvm_vcpu
*vcpu
, char *buf
)
6706 struct kvm_segment seg
;
6710 put_smstate(u32
, buf
, 0x7ffc, kvm_read_cr0(vcpu
));
6711 put_smstate(u32
, buf
, 0x7ff8, kvm_read_cr3(vcpu
));
6712 put_smstate(u32
, buf
, 0x7ff4, kvm_get_rflags(vcpu
));
6713 put_smstate(u32
, buf
, 0x7ff0, kvm_rip_read(vcpu
));
6715 for (i
= 0; i
< 8; i
++)
6716 put_smstate(u32
, buf
, 0x7fd0 + i
* 4, kvm_register_read(vcpu
, i
));
6718 kvm_get_dr(vcpu
, 6, &val
);
6719 put_smstate(u32
, buf
, 0x7fcc, (u32
)val
);
6720 kvm_get_dr(vcpu
, 7, &val
);
6721 put_smstate(u32
, buf
, 0x7fc8, (u32
)val
);
6723 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_TR
);
6724 put_smstate(u32
, buf
, 0x7fc4, seg
.selector
);
6725 put_smstate(u32
, buf
, 0x7f64, seg
.base
);
6726 put_smstate(u32
, buf
, 0x7f60, seg
.limit
);
6727 put_smstate(u32
, buf
, 0x7f5c, enter_smm_get_segment_flags(&seg
));
6729 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_LDTR
);
6730 put_smstate(u32
, buf
, 0x7fc0, seg
.selector
);
6731 put_smstate(u32
, buf
, 0x7f80, seg
.base
);
6732 put_smstate(u32
, buf
, 0x7f7c, seg
.limit
);
6733 put_smstate(u32
, buf
, 0x7f78, enter_smm_get_segment_flags(&seg
));
6735 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6736 put_smstate(u32
, buf
, 0x7f74, dt
.address
);
6737 put_smstate(u32
, buf
, 0x7f70, dt
.size
);
6739 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6740 put_smstate(u32
, buf
, 0x7f58, dt
.address
);
6741 put_smstate(u32
, buf
, 0x7f54, dt
.size
);
6743 for (i
= 0; i
< 6; i
++)
6744 enter_smm_save_seg_32(vcpu
, buf
, i
);
6746 put_smstate(u32
, buf
, 0x7f14, kvm_read_cr4(vcpu
));
6749 put_smstate(u32
, buf
, 0x7efc, 0x00020000);
6750 put_smstate(u32
, buf
, 0x7ef8, vcpu
->arch
.smbase
);
6753 static void enter_smm_save_state_64(struct kvm_vcpu
*vcpu
, char *buf
)
6755 #ifdef CONFIG_X86_64
6757 struct kvm_segment seg
;
6761 for (i
= 0; i
< 16; i
++)
6762 put_smstate(u64
, buf
, 0x7ff8 - i
* 8, kvm_register_read(vcpu
, i
));
6764 put_smstate(u64
, buf
, 0x7f78, kvm_rip_read(vcpu
));
6765 put_smstate(u32
, buf
, 0x7f70, kvm_get_rflags(vcpu
));
6767 kvm_get_dr(vcpu
, 6, &val
);
6768 put_smstate(u64
, buf
, 0x7f68, val
);
6769 kvm_get_dr(vcpu
, 7, &val
);
6770 put_smstate(u64
, buf
, 0x7f60, val
);
6772 put_smstate(u64
, buf
, 0x7f58, kvm_read_cr0(vcpu
));
6773 put_smstate(u64
, buf
, 0x7f50, kvm_read_cr3(vcpu
));
6774 put_smstate(u64
, buf
, 0x7f48, kvm_read_cr4(vcpu
));
6776 put_smstate(u32
, buf
, 0x7f00, vcpu
->arch
.smbase
);
6779 put_smstate(u32
, buf
, 0x7efc, 0x00020064);
6781 put_smstate(u64
, buf
, 0x7ed0, vcpu
->arch
.efer
);
6783 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_TR
);
6784 put_smstate(u16
, buf
, 0x7e90, seg
.selector
);
6785 put_smstate(u16
, buf
, 0x7e92, enter_smm_get_segment_flags(&seg
) >> 8);
6786 put_smstate(u32
, buf
, 0x7e94, seg
.limit
);
6787 put_smstate(u64
, buf
, 0x7e98, seg
.base
);
6789 kvm_x86_ops
->get_idt(vcpu
, &dt
);
6790 put_smstate(u32
, buf
, 0x7e84, dt
.size
);
6791 put_smstate(u64
, buf
, 0x7e88, dt
.address
);
6793 kvm_get_segment(vcpu
, &seg
, VCPU_SREG_LDTR
);
6794 put_smstate(u16
, buf
, 0x7e70, seg
.selector
);
6795 put_smstate(u16
, buf
, 0x7e72, enter_smm_get_segment_flags(&seg
) >> 8);
6796 put_smstate(u32
, buf
, 0x7e74, seg
.limit
);
6797 put_smstate(u64
, buf
, 0x7e78, seg
.base
);
6799 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
6800 put_smstate(u32
, buf
, 0x7e64, dt
.size
);
6801 put_smstate(u64
, buf
, 0x7e68, dt
.address
);
6803 for (i
= 0; i
< 6; i
++)
6804 enter_smm_save_seg_64(vcpu
, buf
, i
);
6810 static void enter_smm(struct kvm_vcpu
*vcpu
)
6812 struct kvm_segment cs
, ds
;
6817 trace_kvm_enter_smm(vcpu
->vcpu_id
, vcpu
->arch
.smbase
, true);
6818 memset(buf
, 0, 512);
6819 if (guest_cpuid_has(vcpu
, X86_FEATURE_LM
))
6820 enter_smm_save_state_64(vcpu
, buf
);
6822 enter_smm_save_state_32(vcpu
, buf
);
6825 * Give pre_enter_smm() a chance to make ISA-specific changes to the
6826 * vCPU state (e.g. leave guest mode) after we've saved the state into
6827 * the SMM state-save area.
6829 kvm_x86_ops
->pre_enter_smm(vcpu
, buf
);
6831 vcpu
->arch
.hflags
|= HF_SMM_MASK
;
6832 kvm_vcpu_write_guest(vcpu
, vcpu
->arch
.smbase
+ 0xfe00, buf
, sizeof(buf
));
6834 if (kvm_x86_ops
->get_nmi_mask(vcpu
))
6835 vcpu
->arch
.hflags
|= HF_SMM_INSIDE_NMI_MASK
;
6837 kvm_x86_ops
->set_nmi_mask(vcpu
, true);
6839 kvm_set_rflags(vcpu
, X86_EFLAGS_FIXED
);
6840 kvm_rip_write(vcpu
, 0x8000);
6842 cr0
= vcpu
->arch
.cr0
& ~(X86_CR0_PE
| X86_CR0_EM
| X86_CR0_TS
| X86_CR0_PG
);
6843 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
6844 vcpu
->arch
.cr0
= cr0
;
6846 kvm_x86_ops
->set_cr4(vcpu
, 0);
6848 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6849 dt
.address
= dt
.size
= 0;
6850 kvm_x86_ops
->set_idt(vcpu
, &dt
);
6852 __kvm_set_dr(vcpu
, 7, DR7_FIXED_1
);
6854 cs
.selector
= (vcpu
->arch
.smbase
>> 4) & 0xffff;
6855 cs
.base
= vcpu
->arch
.smbase
;
6860 cs
.limit
= ds
.limit
= 0xffffffff;
6861 cs
.type
= ds
.type
= 0x3;
6862 cs
.dpl
= ds
.dpl
= 0;
6867 cs
.avl
= ds
.avl
= 0;
6868 cs
.present
= ds
.present
= 1;
6869 cs
.unusable
= ds
.unusable
= 0;
6870 cs
.padding
= ds
.padding
= 0;
6872 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
6873 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_DS
);
6874 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_ES
);
6875 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_FS
);
6876 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_GS
);
6877 kvm_set_segment(vcpu
, &ds
, VCPU_SREG_SS
);
6879 if (guest_cpuid_has(vcpu
, X86_FEATURE_LM
))
6880 kvm_x86_ops
->set_efer(vcpu
, 0);
6882 kvm_update_cpuid(vcpu
);
6883 kvm_mmu_reset_context(vcpu
);
6886 static void process_smi(struct kvm_vcpu
*vcpu
)
6888 vcpu
->arch
.smi_pending
= true;
6889 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6892 void kvm_make_scan_ioapic_request(struct kvm
*kvm
)
6894 kvm_make_all_cpus_request(kvm
, KVM_REQ_SCAN_IOAPIC
);
6897 static void vcpu_scan_ioapic(struct kvm_vcpu
*vcpu
)
6899 u64 eoi_exit_bitmap
[4];
6901 if (!kvm_apic_hw_enabled(vcpu
->arch
.apic
))
6904 bitmap_zero(vcpu
->arch
.ioapic_handled_vectors
, 256);
6906 if (irqchip_split(vcpu
->kvm
))
6907 kvm_scan_ioapic_routes(vcpu
, vcpu
->arch
.ioapic_handled_vectors
);
6909 if (kvm_x86_ops
->sync_pir_to_irr
&& vcpu
->arch
.apicv_active
)
6910 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
6911 if (ioapic_in_kernel(vcpu
->kvm
))
6912 kvm_ioapic_scan_entry(vcpu
, vcpu
->arch
.ioapic_handled_vectors
);
6914 bitmap_or((ulong
*)eoi_exit_bitmap
, vcpu
->arch
.ioapic_handled_vectors
,
6915 vcpu_to_synic(vcpu
)->vec_bitmap
, 256);
6916 kvm_x86_ops
->load_eoi_exitmap(vcpu
, eoi_exit_bitmap
);
6919 static void kvm_vcpu_flush_tlb(struct kvm_vcpu
*vcpu
)
6921 ++vcpu
->stat
.tlb_flush
;
6922 kvm_x86_ops
->tlb_flush(vcpu
);
6925 void kvm_arch_mmu_notifier_invalidate_range(struct kvm
*kvm
,
6926 unsigned long start
, unsigned long end
)
6928 unsigned long apic_address
;
6931 * The physical address of apic access page is stored in the VMCS.
6932 * Update it when it becomes invalid.
6934 apic_address
= gfn_to_hva(kvm
, APIC_DEFAULT_PHYS_BASE
>> PAGE_SHIFT
);
6935 if (start
<= apic_address
&& apic_address
< end
)
6936 kvm_make_all_cpus_request(kvm
, KVM_REQ_APIC_PAGE_RELOAD
);
6939 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu
*vcpu
)
6941 struct page
*page
= NULL
;
6943 if (!lapic_in_kernel(vcpu
))
6946 if (!kvm_x86_ops
->set_apic_access_page_addr
)
6949 page
= gfn_to_page(vcpu
->kvm
, APIC_DEFAULT_PHYS_BASE
>> PAGE_SHIFT
);
6950 if (is_error_page(page
))
6952 kvm_x86_ops
->set_apic_access_page_addr(vcpu
, page_to_phys(page
));
6955 * Do not pin apic access page in memory, the MMU notifier
6956 * will call us again if it is migrated or swapped out.
6960 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page
);
6963 * Returns 1 to let vcpu_run() continue the guest execution loop without
6964 * exiting to the userspace. Otherwise, the value will be returned to the
6967 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
)
6971 dm_request_for_irq_injection(vcpu
) &&
6972 kvm_cpu_accept_dm_intr(vcpu
);
6974 bool req_immediate_exit
= false;
6976 if (kvm_request_pending(vcpu
)) {
6977 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
))
6978 kvm_mmu_unload(vcpu
);
6979 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER
, vcpu
))
6980 __kvm_migrate_timers(vcpu
);
6981 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
))
6982 kvm_gen_update_masterclock(vcpu
->kvm
);
6983 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE
, vcpu
))
6984 kvm_gen_kvmclock_update(vcpu
);
6985 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE
, vcpu
)) {
6986 r
= kvm_guest_time_update(vcpu
);
6990 if (kvm_check_request(KVM_REQ_MMU_SYNC
, vcpu
))
6991 kvm_mmu_sync_roots(vcpu
);
6992 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
6993 kvm_vcpu_flush_tlb(vcpu
);
6994 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS
, vcpu
)) {
6995 vcpu
->run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
6999 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT
, vcpu
)) {
7000 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
7001 vcpu
->mmio_needed
= 0;
7005 if (kvm_check_request(KVM_REQ_APF_HALT
, vcpu
)) {
7006 /* Page is swapped out. Do synthetic halt */
7007 vcpu
->arch
.apf
.halted
= true;
7011 if (kvm_check_request(KVM_REQ_STEAL_UPDATE
, vcpu
))
7012 record_steal_time(vcpu
);
7013 if (kvm_check_request(KVM_REQ_SMI
, vcpu
))
7015 if (kvm_check_request(KVM_REQ_NMI
, vcpu
))
7017 if (kvm_check_request(KVM_REQ_PMU
, vcpu
))
7018 kvm_pmu_handle_event(vcpu
);
7019 if (kvm_check_request(KVM_REQ_PMI
, vcpu
))
7020 kvm_pmu_deliver_pmi(vcpu
);
7021 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT
, vcpu
)) {
7022 BUG_ON(vcpu
->arch
.pending_ioapic_eoi
> 255);
7023 if (test_bit(vcpu
->arch
.pending_ioapic_eoi
,
7024 vcpu
->arch
.ioapic_handled_vectors
)) {
7025 vcpu
->run
->exit_reason
= KVM_EXIT_IOAPIC_EOI
;
7026 vcpu
->run
->eoi
.vector
=
7027 vcpu
->arch
.pending_ioapic_eoi
;
7032 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC
, vcpu
))
7033 vcpu_scan_ioapic(vcpu
);
7034 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD
, vcpu
))
7035 kvm_vcpu_reload_apic_access_page(vcpu
);
7036 if (kvm_check_request(KVM_REQ_HV_CRASH
, vcpu
)) {
7037 vcpu
->run
->exit_reason
= KVM_EXIT_SYSTEM_EVENT
;
7038 vcpu
->run
->system_event
.type
= KVM_SYSTEM_EVENT_CRASH
;
7042 if (kvm_check_request(KVM_REQ_HV_RESET
, vcpu
)) {
7043 vcpu
->run
->exit_reason
= KVM_EXIT_SYSTEM_EVENT
;
7044 vcpu
->run
->system_event
.type
= KVM_SYSTEM_EVENT_RESET
;
7048 if (kvm_check_request(KVM_REQ_HV_EXIT
, vcpu
)) {
7049 vcpu
->run
->exit_reason
= KVM_EXIT_HYPERV
;
7050 vcpu
->run
->hyperv
= vcpu
->arch
.hyperv
.exit
;
7056 * KVM_REQ_HV_STIMER has to be processed after
7057 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7058 * depend on the guest clock being up-to-date
7060 if (kvm_check_request(KVM_REQ_HV_STIMER
, vcpu
))
7061 kvm_hv_process_stimers(vcpu
);
7064 if (kvm_check_request(KVM_REQ_EVENT
, vcpu
) || req_int_win
) {
7065 ++vcpu
->stat
.req_event
;
7066 kvm_apic_accept_events(vcpu
);
7067 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_INIT_RECEIVED
) {
7072 if (inject_pending_event(vcpu
, req_int_win
) != 0)
7073 req_immediate_exit
= true;
7075 /* Enable SMI/NMI/IRQ window open exits if needed.
7077 * SMIs have three cases:
7078 * 1) They can be nested, and then there is nothing to
7079 * do here because RSM will cause a vmexit anyway.
7080 * 2) There is an ISA-specific reason why SMI cannot be
7081 * injected, and the moment when this changes can be
7083 * 3) Or the SMI can be pending because
7084 * inject_pending_event has completed the injection
7085 * of an IRQ or NMI from the previous vmexit, and
7086 * then we request an immediate exit to inject the
7089 if (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
))
7090 if (!kvm_x86_ops
->enable_smi_window(vcpu
))
7091 req_immediate_exit
= true;
7092 if (vcpu
->arch
.nmi_pending
)
7093 kvm_x86_ops
->enable_nmi_window(vcpu
);
7094 if (kvm_cpu_has_injectable_intr(vcpu
) || req_int_win
)
7095 kvm_x86_ops
->enable_irq_window(vcpu
);
7096 WARN_ON(vcpu
->arch
.exception
.pending
);
7099 if (kvm_lapic_enabled(vcpu
)) {
7100 update_cr8_intercept(vcpu
);
7101 kvm_lapic_sync_to_vapic(vcpu
);
7105 r
= kvm_mmu_reload(vcpu
);
7107 goto cancel_injection
;
7112 kvm_x86_ops
->prepare_guest_switch(vcpu
);
7115 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7116 * IPI are then delayed after guest entry, which ensures that they
7117 * result in virtual interrupt delivery.
7119 local_irq_disable();
7120 vcpu
->mode
= IN_GUEST_MODE
;
7122 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
7125 * 1) We should set ->mode before checking ->requests. Please see
7126 * the comment in kvm_vcpu_exiting_guest_mode().
7128 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7129 * pairs with the memory barrier implicit in pi_test_and_set_on
7130 * (see vmx_deliver_posted_interrupt).
7132 * 3) This also orders the write to mode from any reads to the page
7133 * tables done while the VCPU is running. Please see the comment
7134 * in kvm_flush_remote_tlbs.
7136 smp_mb__after_srcu_read_unlock();
7139 * This handles the case where a posted interrupt was
7140 * notified with kvm_vcpu_kick.
7142 if (kvm_lapic_enabled(vcpu
)) {
7143 if (kvm_x86_ops
->sync_pir_to_irr
&& vcpu
->arch
.apicv_active
)
7144 kvm_x86_ops
->sync_pir_to_irr(vcpu
);
7147 if (vcpu
->mode
== EXITING_GUEST_MODE
|| kvm_request_pending(vcpu
)
7148 || need_resched() || signal_pending(current
)) {
7149 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
7153 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7155 goto cancel_injection
;
7158 kvm_load_guest_xcr0(vcpu
);
7160 if (req_immediate_exit
) {
7161 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7162 smp_send_reschedule(vcpu
->cpu
);
7165 trace_kvm_entry(vcpu
->vcpu_id
);
7166 wait_lapic_expire(vcpu
);
7167 guest_enter_irqoff();
7169 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
7171 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
7172 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
7173 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
7174 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
7175 set_debugreg(vcpu
->arch
.dr6
, 6);
7176 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_RELOAD
;
7179 kvm_x86_ops
->run(vcpu
);
7182 * Do this here before restoring debug registers on the host. And
7183 * since we do this before handling the vmexit, a DR access vmexit
7184 * can (a) read the correct value of the debug registers, (b) set
7185 * KVM_DEBUGREG_WONT_EXIT again.
7187 if (unlikely(vcpu
->arch
.switch_db_regs
& KVM_DEBUGREG_WONT_EXIT
)) {
7188 WARN_ON(vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
);
7189 kvm_x86_ops
->sync_dirty_debug_regs(vcpu
);
7190 kvm_update_dr0123(vcpu
);
7191 kvm_update_dr6(vcpu
);
7192 kvm_update_dr7(vcpu
);
7193 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_RELOAD
;
7197 * If the guest has used debug registers, at least dr7
7198 * will be disabled while returning to the host.
7199 * If we don't have active breakpoints in the host, we don't
7200 * care about the messed up debug address registers. But if
7201 * we have some of them active, restore the old state.
7203 if (hw_breakpoint_active())
7204 hw_breakpoint_restore();
7206 vcpu
->arch
.last_guest_tsc
= kvm_read_l1_tsc(vcpu
, rdtsc());
7208 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
7211 kvm_put_guest_xcr0(vcpu
);
7213 kvm_x86_ops
->handle_external_intr(vcpu
);
7217 guest_exit_irqoff();
7222 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7225 * Profile KVM exit RIPs:
7227 if (unlikely(prof_on
== KVM_PROFILING
)) {
7228 unsigned long rip
= kvm_rip_read(vcpu
);
7229 profile_hit(KVM_PROFILING
, (void *)rip
);
7232 if (unlikely(vcpu
->arch
.tsc_always_catchup
))
7233 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
7235 if (vcpu
->arch
.apic_attention
)
7236 kvm_lapic_sync_from_vapic(vcpu
);
7238 vcpu
->arch
.gpa_available
= false;
7239 r
= kvm_x86_ops
->handle_exit(vcpu
);
7243 kvm_x86_ops
->cancel_injection(vcpu
);
7244 if (unlikely(vcpu
->arch
.apic_attention
))
7245 kvm_lapic_sync_from_vapic(vcpu
);
7250 static inline int vcpu_block(struct kvm
*kvm
, struct kvm_vcpu
*vcpu
)
7252 if (!kvm_arch_vcpu_runnable(vcpu
) &&
7253 (!kvm_x86_ops
->pre_block
|| kvm_x86_ops
->pre_block(vcpu
) == 0)) {
7254 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
7255 kvm_vcpu_block(vcpu
);
7256 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
7258 if (kvm_x86_ops
->post_block
)
7259 kvm_x86_ops
->post_block(vcpu
);
7261 if (!kvm_check_request(KVM_REQ_UNHALT
, vcpu
))
7265 kvm_apic_accept_events(vcpu
);
7266 switch(vcpu
->arch
.mp_state
) {
7267 case KVM_MP_STATE_HALTED
:
7268 vcpu
->arch
.pv
.pv_unhalted
= false;
7269 vcpu
->arch
.mp_state
=
7270 KVM_MP_STATE_RUNNABLE
;
7271 case KVM_MP_STATE_RUNNABLE
:
7272 vcpu
->arch
.apf
.halted
= false;
7274 case KVM_MP_STATE_INIT_RECEIVED
:
7283 static inline bool kvm_vcpu_running(struct kvm_vcpu
*vcpu
)
7285 if (is_guest_mode(vcpu
) && kvm_x86_ops
->check_nested_events
)
7286 kvm_x86_ops
->check_nested_events(vcpu
, false);
7288 return (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
&&
7289 !vcpu
->arch
.apf
.halted
);
7292 static int vcpu_run(struct kvm_vcpu
*vcpu
)
7295 struct kvm
*kvm
= vcpu
->kvm
;
7297 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
7298 vcpu
->arch
.l1tf_flush_l1d
= true;
7301 if (kvm_vcpu_running(vcpu
)) {
7302 r
= vcpu_enter_guest(vcpu
);
7304 r
= vcpu_block(kvm
, vcpu
);
7310 kvm_clear_request(KVM_REQ_PENDING_TIMER
, vcpu
);
7311 if (kvm_cpu_has_pending_timer(vcpu
))
7312 kvm_inject_pending_timer_irqs(vcpu
);
7314 if (dm_request_for_irq_injection(vcpu
) &&
7315 kvm_vcpu_ready_for_interrupt_injection(vcpu
)) {
7317 vcpu
->run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
7318 ++vcpu
->stat
.request_irq_exits
;
7322 kvm_check_async_pf_completion(vcpu
);
7324 if (signal_pending(current
)) {
7326 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
7327 ++vcpu
->stat
.signal_exits
;
7330 if (need_resched()) {
7331 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
7333 vcpu
->srcu_idx
= srcu_read_lock(&kvm
->srcu
);
7337 srcu_read_unlock(&kvm
->srcu
, vcpu
->srcu_idx
);
7342 static inline int complete_emulated_io(struct kvm_vcpu
*vcpu
)
7345 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7346 r
= emulate_instruction(vcpu
, EMULTYPE_NO_DECODE
);
7347 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
7348 if (r
!= EMULATE_DONE
)
7353 static int complete_emulated_pio(struct kvm_vcpu
*vcpu
)
7355 BUG_ON(!vcpu
->arch
.pio
.count
);
7357 return complete_emulated_io(vcpu
);
7361 * Implements the following, as a state machine:
7365 * for each mmio piece in the fragment
7373 * for each mmio piece in the fragment
7378 static int complete_emulated_mmio(struct kvm_vcpu
*vcpu
)
7380 struct kvm_run
*run
= vcpu
->run
;
7381 struct kvm_mmio_fragment
*frag
;
7384 BUG_ON(!vcpu
->mmio_needed
);
7386 /* Complete previous fragment */
7387 frag
= &vcpu
->mmio_fragments
[vcpu
->mmio_cur_fragment
];
7388 len
= min(8u, frag
->len
);
7389 if (!vcpu
->mmio_is_write
)
7390 memcpy(frag
->data
, run
->mmio
.data
, len
);
7392 if (frag
->len
<= 8) {
7393 /* Switch to the next fragment. */
7395 vcpu
->mmio_cur_fragment
++;
7397 /* Go forward to the next mmio piece. */
7403 if (vcpu
->mmio_cur_fragment
>= vcpu
->mmio_nr_fragments
) {
7404 vcpu
->mmio_needed
= 0;
7406 /* FIXME: return into emulator if single-stepping. */
7407 if (vcpu
->mmio_is_write
)
7409 vcpu
->mmio_read_completed
= 1;
7410 return complete_emulated_io(vcpu
);
7413 run
->exit_reason
= KVM_EXIT_MMIO
;
7414 run
->mmio
.phys_addr
= frag
->gpa
;
7415 if (vcpu
->mmio_is_write
)
7416 memcpy(run
->mmio
.data
, frag
->data
, min(8u, frag
->len
));
7417 run
->mmio
.len
= min(8u, frag
->len
);
7418 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
7419 vcpu
->arch
.complete_userspace_io
= complete_emulated_mmio
;
7424 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
7428 kvm_sigset_activate(vcpu
);
7430 kvm_load_guest_fpu(vcpu
);
7432 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
7433 if (kvm_run
->immediate_exit
) {
7437 kvm_vcpu_block(vcpu
);
7438 kvm_apic_accept_events(vcpu
);
7439 kvm_clear_request(KVM_REQ_UNHALT
, vcpu
);
7441 if (signal_pending(current
)) {
7443 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
7444 ++vcpu
->stat
.signal_exits
;
7449 /* re-sync apic's tpr */
7450 if (!lapic_in_kernel(vcpu
)) {
7451 if (kvm_set_cr8(vcpu
, kvm_run
->cr8
) != 0) {
7457 if (unlikely(vcpu
->arch
.complete_userspace_io
)) {
7458 int (*cui
)(struct kvm_vcpu
*) = vcpu
->arch
.complete_userspace_io
;
7459 vcpu
->arch
.complete_userspace_io
= NULL
;
7464 WARN_ON(vcpu
->arch
.pio
.count
|| vcpu
->mmio_needed
);
7466 if (kvm_run
->immediate_exit
)
7472 kvm_put_guest_fpu(vcpu
);
7473 post_kvm_run_save(vcpu
);
7474 kvm_sigset_deactivate(vcpu
);
7479 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
7481 if (vcpu
->arch
.emulate_regs_need_sync_to_vcpu
) {
7483 * We are here if userspace calls get_regs() in the middle of
7484 * instruction emulation. Registers state needs to be copied
7485 * back from emulation context to vcpu. Userspace shouldn't do
7486 * that usually, but some bad designed PV devices (vmware
7487 * backdoor interface) need this to work
7489 emulator_writeback_register_cache(&vcpu
->arch
.emulate_ctxt
);
7490 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
7492 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
7493 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
7494 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
7495 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
7496 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
7497 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
7498 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
7499 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
7500 #ifdef CONFIG_X86_64
7501 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
7502 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
7503 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
7504 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
7505 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
7506 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
7507 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
7508 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
7511 regs
->rip
= kvm_rip_read(vcpu
);
7512 regs
->rflags
= kvm_get_rflags(vcpu
);
7517 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
7519 vcpu
->arch
.emulate_regs_need_sync_from_vcpu
= true;
7520 vcpu
->arch
.emulate_regs_need_sync_to_vcpu
= false;
7522 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
7523 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
7524 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
7525 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
7526 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
7527 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
7528 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
7529 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
7530 #ifdef CONFIG_X86_64
7531 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
7532 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
7533 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
7534 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
7535 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
7536 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
7537 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
7538 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
7541 kvm_rip_write(vcpu
, regs
->rip
);
7542 kvm_set_rflags(vcpu
, regs
->rflags
| X86_EFLAGS_FIXED
);
7544 vcpu
->arch
.exception
.pending
= false;
7546 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7551 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
7553 struct kvm_segment cs
;
7555 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
7559 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
7561 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
7562 struct kvm_sregs
*sregs
)
7566 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
7567 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
7568 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
7569 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
7570 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
7571 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
7573 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
7574 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
7576 kvm_x86_ops
->get_idt(vcpu
, &dt
);
7577 sregs
->idt
.limit
= dt
.size
;
7578 sregs
->idt
.base
= dt
.address
;
7579 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
7580 sregs
->gdt
.limit
= dt
.size
;
7581 sregs
->gdt
.base
= dt
.address
;
7583 sregs
->cr0
= kvm_read_cr0(vcpu
);
7584 sregs
->cr2
= vcpu
->arch
.cr2
;
7585 sregs
->cr3
= kvm_read_cr3(vcpu
);
7586 sregs
->cr4
= kvm_read_cr4(vcpu
);
7587 sregs
->cr8
= kvm_get_cr8(vcpu
);
7588 sregs
->efer
= vcpu
->arch
.efer
;
7589 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
7591 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
7593 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
7594 set_bit(vcpu
->arch
.interrupt
.nr
,
7595 (unsigned long *)sregs
->interrupt_bitmap
);
7600 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
7601 struct kvm_mp_state
*mp_state
)
7603 kvm_apic_accept_events(vcpu
);
7604 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_HALTED
&&
7605 vcpu
->arch
.pv
.pv_unhalted
)
7606 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
7608 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
7613 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
7614 struct kvm_mp_state
*mp_state
)
7616 if (!lapic_in_kernel(vcpu
) &&
7617 mp_state
->mp_state
!= KVM_MP_STATE_RUNNABLE
)
7620 /* INITs are latched while in SMM */
7621 if ((is_smm(vcpu
) || vcpu
->arch
.smi_pending
) &&
7622 (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
||
7623 mp_state
->mp_state
== KVM_MP_STATE_INIT_RECEIVED
))
7626 if (mp_state
->mp_state
== KVM_MP_STATE_SIPI_RECEIVED
) {
7627 vcpu
->arch
.mp_state
= KVM_MP_STATE_INIT_RECEIVED
;
7628 set_bit(KVM_APIC_SIPI
, &vcpu
->arch
.apic
->pending_events
);
7630 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
7631 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7635 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int idt_index
,
7636 int reason
, bool has_error_code
, u32 error_code
)
7638 struct x86_emulate_ctxt
*ctxt
= &vcpu
->arch
.emulate_ctxt
;
7641 init_emulate_ctxt(vcpu
);
7643 ret
= emulator_task_switch(ctxt
, tss_selector
, idt_index
, reason
,
7644 has_error_code
, error_code
);
7647 return EMULATE_FAIL
;
7649 kvm_rip_write(vcpu
, ctxt
->eip
);
7650 kvm_set_rflags(vcpu
, ctxt
->eflags
);
7651 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7652 return EMULATE_DONE
;
7654 EXPORT_SYMBOL_GPL(kvm_task_switch
);
7656 int kvm_valid_sregs(struct kvm_vcpu
*vcpu
, struct kvm_sregs
*sregs
)
7658 if ((sregs
->efer
& EFER_LME
) && (sregs
->cr0
& X86_CR0_PG
)) {
7660 * When EFER.LME and CR0.PG are set, the processor is in
7661 * 64-bit mode (though maybe in a 32-bit code segment).
7662 * CR4.PAE and EFER.LMA must be set.
7664 if (!(sregs
->cr4
& X86_CR4_PAE
)
7665 || !(sregs
->efer
& EFER_LMA
))
7669 * Not in 64-bit mode: EFER.LMA is clear and the code
7670 * segment cannot be 64-bit.
7672 if (sregs
->efer
& EFER_LMA
|| sregs
->cs
.l
)
7679 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
7680 struct kvm_sregs
*sregs
)
7682 struct msr_data apic_base_msr
;
7683 int mmu_reset_needed
= 0;
7684 int cpuid_update_needed
= 0;
7685 int pending_vec
, max_bits
, idx
;
7688 if (!guest_cpuid_has(vcpu
, X86_FEATURE_XSAVE
) &&
7689 (sregs
->cr4
& X86_CR4_OSXSAVE
))
7692 if (kvm_valid_sregs(vcpu
, sregs
))
7695 apic_base_msr
.data
= sregs
->apic_base
;
7696 apic_base_msr
.host_initiated
= true;
7697 if (kvm_set_apic_base(vcpu
, &apic_base_msr
))
7700 dt
.size
= sregs
->idt
.limit
;
7701 dt
.address
= sregs
->idt
.base
;
7702 kvm_x86_ops
->set_idt(vcpu
, &dt
);
7703 dt
.size
= sregs
->gdt
.limit
;
7704 dt
.address
= sregs
->gdt
.base
;
7705 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
7707 vcpu
->arch
.cr2
= sregs
->cr2
;
7708 mmu_reset_needed
|= kvm_read_cr3(vcpu
) != sregs
->cr3
;
7709 vcpu
->arch
.cr3
= sregs
->cr3
;
7710 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
7712 kvm_set_cr8(vcpu
, sregs
->cr8
);
7714 mmu_reset_needed
|= vcpu
->arch
.efer
!= sregs
->efer
;
7715 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
7717 mmu_reset_needed
|= kvm_read_cr0(vcpu
) != sregs
->cr0
;
7718 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
7719 vcpu
->arch
.cr0
= sregs
->cr0
;
7721 mmu_reset_needed
|= kvm_read_cr4(vcpu
) != sregs
->cr4
;
7722 cpuid_update_needed
|= ((kvm_read_cr4(vcpu
) ^ sregs
->cr4
) &
7723 (X86_CR4_OSXSAVE
| X86_CR4_PKE
));
7724 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
7725 if (cpuid_update_needed
)
7726 kvm_update_cpuid(vcpu
);
7728 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7729 if (!is_long_mode(vcpu
) && is_pae(vcpu
)) {
7730 load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, kvm_read_cr3(vcpu
));
7731 mmu_reset_needed
= 1;
7733 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
7735 if (mmu_reset_needed
)
7736 kvm_mmu_reset_context(vcpu
);
7738 max_bits
= KVM_NR_INTERRUPTS
;
7739 pending_vec
= find_first_bit(
7740 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
7741 if (pending_vec
< max_bits
) {
7742 kvm_queue_interrupt(vcpu
, pending_vec
, false);
7743 pr_debug("Set back pending irq %d\n", pending_vec
);
7746 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
7747 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
7748 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
7749 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
7750 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
7751 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
7753 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
7754 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
7756 update_cr8_intercept(vcpu
);
7758 /* Older userspace won't unhalt the vcpu on reset. */
7759 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
7760 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
7762 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
7764 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
7769 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
7770 struct kvm_guest_debug
*dbg
)
7772 unsigned long rflags
;
7775 if (dbg
->control
& (KVM_GUESTDBG_INJECT_DB
| KVM_GUESTDBG_INJECT_BP
)) {
7777 if (vcpu
->arch
.exception
.pending
)
7779 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
7780 kvm_queue_exception(vcpu
, DB_VECTOR
);
7782 kvm_queue_exception(vcpu
, BP_VECTOR
);
7786 * Read rflags as long as potentially injected trace flags are still
7789 rflags
= kvm_get_rflags(vcpu
);
7791 vcpu
->guest_debug
= dbg
->control
;
7792 if (!(vcpu
->guest_debug
& KVM_GUESTDBG_ENABLE
))
7793 vcpu
->guest_debug
= 0;
7795 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
7796 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
7797 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
7798 vcpu
->arch
.guest_debug_dr7
= dbg
->arch
.debugreg
[7];
7800 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
7801 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
7803 kvm_update_dr7(vcpu
);
7805 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
7806 vcpu
->arch
.singlestep_rip
= kvm_rip_read(vcpu
) +
7807 get_segment_base(vcpu
, VCPU_SREG_CS
);
7810 * Trigger an rflags update that will inject or remove the trace
7813 kvm_set_rflags(vcpu
, rflags
);
7815 kvm_x86_ops
->update_bp_intercept(vcpu
);
7825 * Translate a guest virtual address to a guest physical address.
7827 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
7828 struct kvm_translation
*tr
)
7830 unsigned long vaddr
= tr
->linear_address
;
7834 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
7835 gpa
= kvm_mmu_gva_to_gpa_system(vcpu
, vaddr
, NULL
);
7836 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
7837 tr
->physical_address
= gpa
;
7838 tr
->valid
= gpa
!= UNMAPPED_GVA
;
7845 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
7847 struct fxregs_state
*fxsave
=
7848 &vcpu
->arch
.guest_fpu
.state
.fxsave
;
7850 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
7851 fpu
->fcw
= fxsave
->cwd
;
7852 fpu
->fsw
= fxsave
->swd
;
7853 fpu
->ftwx
= fxsave
->twd
;
7854 fpu
->last_opcode
= fxsave
->fop
;
7855 fpu
->last_ip
= fxsave
->rip
;
7856 fpu
->last_dp
= fxsave
->rdp
;
7857 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
7862 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
7864 struct fxregs_state
*fxsave
=
7865 &vcpu
->arch
.guest_fpu
.state
.fxsave
;
7867 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
7868 fxsave
->cwd
= fpu
->fcw
;
7869 fxsave
->swd
= fpu
->fsw
;
7870 fxsave
->twd
= fpu
->ftwx
;
7871 fxsave
->fop
= fpu
->last_opcode
;
7872 fxsave
->rip
= fpu
->last_ip
;
7873 fxsave
->rdp
= fpu
->last_dp
;
7874 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
7879 static void fx_init(struct kvm_vcpu
*vcpu
)
7881 fpstate_init(&vcpu
->arch
.guest_fpu
.state
);
7882 if (boot_cpu_has(X86_FEATURE_XSAVES
))
7883 vcpu
->arch
.guest_fpu
.state
.xsave
.header
.xcomp_bv
=
7884 host_xcr0
| XSTATE_COMPACTION_ENABLED
;
7887 * Ensure guest xcr0 is valid for loading
7889 vcpu
->arch
.xcr0
= XFEATURE_MASK_FP
;
7891 vcpu
->arch
.cr0
|= X86_CR0_ET
;
7894 /* Swap (qemu) user FPU context for the guest FPU context. */
7895 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
7898 copy_fpregs_to_fpstate(&vcpu
->arch
.user_fpu
);
7899 /* PKRU is separately restored in kvm_x86_ops->run. */
7900 __copy_kernel_to_fpregs(&vcpu
->arch
.guest_fpu
.state
,
7901 ~XFEATURE_MASK_PKRU
);
7906 /* When vcpu_run ends, restore user space FPU context. */
7907 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
7910 copy_fpregs_to_fpstate(&vcpu
->arch
.guest_fpu
);
7911 copy_kernel_to_fpregs(&vcpu
->arch
.user_fpu
.state
);
7913 ++vcpu
->stat
.fpu_reload
;
7917 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
7919 void *wbinvd_dirty_mask
= vcpu
->arch
.wbinvd_dirty_mask
;
7921 kvmclock_reset(vcpu
);
7923 kvm_x86_ops
->vcpu_free(vcpu
);
7924 free_cpumask_var(wbinvd_dirty_mask
);
7927 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
7930 struct kvm_vcpu
*vcpu
;
7932 if (check_tsc_unstable() && atomic_read(&kvm
->online_vcpus
) != 0)
7933 printk_once(KERN_WARNING
7934 "kvm: SMP vm created on host with unstable TSC; "
7935 "guest TSC will not be reliable\n");
7937 vcpu
= kvm_x86_ops
->vcpu_create(kvm
, id
);
7942 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
7946 kvm_vcpu_mtrr_init(vcpu
);
7947 r
= vcpu_load(vcpu
);
7950 kvm_vcpu_reset(vcpu
, false);
7951 kvm_mmu_setup(vcpu
);
7956 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
7958 struct msr_data msr
;
7959 struct kvm
*kvm
= vcpu
->kvm
;
7961 kvm_hv_vcpu_postcreate(vcpu
);
7963 if (vcpu_load(vcpu
))
7966 msr
.index
= MSR_IA32_TSC
;
7967 msr
.host_initiated
= true;
7968 kvm_write_tsc(vcpu
, &msr
);
7971 if (!kvmclock_periodic_sync
)
7974 schedule_delayed_work(&kvm
->arch
.kvmclock_sync_work
,
7975 KVMCLOCK_SYNC_PERIOD
);
7978 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
7981 vcpu
->arch
.apf
.msr_val
= 0;
7983 r
= vcpu_load(vcpu
);
7985 kvm_mmu_unload(vcpu
);
7988 kvm_x86_ops
->vcpu_free(vcpu
);
7991 void kvm_vcpu_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
7993 kvm_lapic_reset(vcpu
, init_event
);
7995 vcpu
->arch
.hflags
= 0;
7997 vcpu
->arch
.smi_pending
= 0;
7998 atomic_set(&vcpu
->arch
.nmi_queued
, 0);
7999 vcpu
->arch
.nmi_pending
= 0;
8000 vcpu
->arch
.nmi_injected
= false;
8001 kvm_clear_interrupt_queue(vcpu
);
8002 kvm_clear_exception_queue(vcpu
);
8003 vcpu
->arch
.exception
.pending
= false;
8005 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
8006 kvm_update_dr0123(vcpu
);
8007 vcpu
->arch
.dr6
= DR6_INIT
;
8008 kvm_update_dr6(vcpu
);
8009 vcpu
->arch
.dr7
= DR7_FIXED_1
;
8010 kvm_update_dr7(vcpu
);
8014 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
8015 vcpu
->arch
.apf
.msr_val
= 0;
8016 vcpu
->arch
.st
.msr_val
= 0;
8018 kvmclock_reset(vcpu
);
8020 kvm_clear_async_pf_completion_queue(vcpu
);
8021 kvm_async_pf_hash_reset(vcpu
);
8022 vcpu
->arch
.apf
.halted
= false;
8024 if (kvm_mpx_supported()) {
8025 void *mpx_state_buffer
;
8028 * To avoid have the INIT path from kvm_apic_has_events() that be
8029 * called with loaded FPU and does not let userspace fix the state.
8032 kvm_put_guest_fpu(vcpu
);
8033 mpx_state_buffer
= get_xsave_addr(&vcpu
->arch
.guest_fpu
.state
.xsave
,
8034 XFEATURE_MASK_BNDREGS
);
8035 if (mpx_state_buffer
)
8036 memset(mpx_state_buffer
, 0, sizeof(struct mpx_bndreg_state
));
8037 mpx_state_buffer
= get_xsave_addr(&vcpu
->arch
.guest_fpu
.state
.xsave
,
8038 XFEATURE_MASK_BNDCSR
);
8039 if (mpx_state_buffer
)
8040 memset(mpx_state_buffer
, 0, sizeof(struct mpx_bndcsr
));
8042 kvm_load_guest_fpu(vcpu
);
8046 kvm_pmu_reset(vcpu
);
8047 vcpu
->arch
.smbase
= 0x30000;
8049 vcpu
->arch
.msr_platform_info
= MSR_PLATFORM_INFO_CPUID_FAULT
;
8050 vcpu
->arch
.msr_misc_features_enables
= 0;
8052 vcpu
->arch
.xcr0
= XFEATURE_MASK_FP
;
8055 memset(vcpu
->arch
.regs
, 0, sizeof(vcpu
->arch
.regs
));
8056 vcpu
->arch
.regs_avail
= ~0;
8057 vcpu
->arch
.regs_dirty
= ~0;
8059 vcpu
->arch
.ia32_xss
= 0;
8061 kvm_x86_ops
->vcpu_reset(vcpu
, init_event
);
8064 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu
*vcpu
, u8 vector
)
8066 struct kvm_segment cs
;
8068 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
8069 cs
.selector
= vector
<< 8;
8070 cs
.base
= vector
<< 12;
8071 kvm_set_segment(vcpu
, &cs
, VCPU_SREG_CS
);
8072 kvm_rip_write(vcpu
, 0);
8075 int kvm_arch_hardware_enable(void)
8078 struct kvm_vcpu
*vcpu
;
8083 bool stable
, backwards_tsc
= false;
8085 kvm_shared_msr_cpu_online();
8086 ret
= kvm_x86_ops
->hardware_enable();
8090 local_tsc
= rdtsc();
8091 stable
= !check_tsc_unstable();
8092 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
8093 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
8094 if (!stable
&& vcpu
->cpu
== smp_processor_id())
8095 kvm_make_request(KVM_REQ_CLOCK_UPDATE
, vcpu
);
8096 if (stable
&& vcpu
->arch
.last_host_tsc
> local_tsc
) {
8097 backwards_tsc
= true;
8098 if (vcpu
->arch
.last_host_tsc
> max_tsc
)
8099 max_tsc
= vcpu
->arch
.last_host_tsc
;
8105 * Sometimes, even reliable TSCs go backwards. This happens on
8106 * platforms that reset TSC during suspend or hibernate actions, but
8107 * maintain synchronization. We must compensate. Fortunately, we can
8108 * detect that condition here, which happens early in CPU bringup,
8109 * before any KVM threads can be running. Unfortunately, we can't
8110 * bring the TSCs fully up to date with real time, as we aren't yet far
8111 * enough into CPU bringup that we know how much real time has actually
8112 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8113 * variables that haven't been updated yet.
8115 * So we simply find the maximum observed TSC above, then record the
8116 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8117 * the adjustment will be applied. Note that we accumulate
8118 * adjustments, in case multiple suspend cycles happen before some VCPU
8119 * gets a chance to run again. In the event that no KVM threads get a
8120 * chance to run, we will miss the entire elapsed period, as we'll have
8121 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8122 * loose cycle time. This isn't too big a deal, since the loss will be
8123 * uniform across all VCPUs (not to mention the scenario is extremely
8124 * unlikely). It is possible that a second hibernate recovery happens
8125 * much faster than a first, causing the observed TSC here to be
8126 * smaller; this would require additional padding adjustment, which is
8127 * why we set last_host_tsc to the local tsc observed here.
8129 * N.B. - this code below runs only on platforms with reliable TSC,
8130 * as that is the only way backwards_tsc is set above. Also note
8131 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8132 * have the same delta_cyc adjustment applied if backwards_tsc
8133 * is detected. Note further, this adjustment is only done once,
8134 * as we reset last_host_tsc on all VCPUs to stop this from being
8135 * called multiple times (one for each physical CPU bringup).
8137 * Platforms with unreliable TSCs don't have to deal with this, they
8138 * will be compensated by the logic in vcpu_load, which sets the TSC to
8139 * catchup mode. This will catchup all VCPUs to real time, but cannot
8140 * guarantee that they stay in perfect synchronization.
8142 if (backwards_tsc
) {
8143 u64 delta_cyc
= max_tsc
- local_tsc
;
8144 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
8145 kvm
->arch
.backwards_tsc_observed
= true;
8146 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
8147 vcpu
->arch
.tsc_offset_adjustment
+= delta_cyc
;
8148 vcpu
->arch
.last_host_tsc
= local_tsc
;
8149 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE
, vcpu
);
8153 * We have to disable TSC offset matching.. if you were
8154 * booting a VM while issuing an S4 host suspend....
8155 * you may have some problem. Solving this issue is
8156 * left as an exercise to the reader.
8158 kvm
->arch
.last_tsc_nsec
= 0;
8159 kvm
->arch
.last_tsc_write
= 0;
8166 void kvm_arch_hardware_disable(void)
8168 kvm_x86_ops
->hardware_disable();
8169 drop_user_return_notifiers();
8172 int kvm_arch_hardware_setup(void)
8176 r
= kvm_x86_ops
->hardware_setup();
8180 if (kvm_has_tsc_control
) {
8182 * Make sure the user can only configure tsc_khz values that
8183 * fit into a signed integer.
8184 * A min value is not calculated needed because it will always
8185 * be 1 on all machines.
8187 u64 max
= min(0x7fffffffULL
,
8188 __scale_tsc(kvm_max_tsc_scaling_ratio
, tsc_khz
));
8189 kvm_max_guest_tsc_khz
= max
;
8191 kvm_default_tsc_scaling_ratio
= 1ULL << kvm_tsc_scaling_ratio_frac_bits
;
8194 kvm_init_msr_list();
8198 void kvm_arch_hardware_unsetup(void)
8200 kvm_x86_ops
->hardware_unsetup();
8203 void kvm_arch_check_processor_compat(void *rtn
)
8205 kvm_x86_ops
->check_processor_compatibility(rtn
);
8208 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu
*vcpu
)
8210 return vcpu
->kvm
->arch
.bsp_vcpu_id
== vcpu
->vcpu_id
;
8212 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp
);
8214 bool kvm_vcpu_is_bsp(struct kvm_vcpu
*vcpu
)
8216 return (vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_BSP
) != 0;
8219 struct static_key kvm_no_apic_vcpu __read_mostly
;
8220 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu
);
8222 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
8227 vcpu
->arch
.apicv_active
= kvm_x86_ops
->get_enable_apicv(vcpu
);
8228 vcpu
->arch
.emulate_ctxt
.ops
= &emulate_ops
;
8229 if (!irqchip_in_kernel(vcpu
->kvm
) || kvm_vcpu_is_reset_bsp(vcpu
))
8230 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
8232 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
8234 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
8239 vcpu
->arch
.pio_data
= page_address(page
);
8241 kvm_set_tsc_khz(vcpu
, max_tsc_khz
);
8243 r
= kvm_mmu_create(vcpu
);
8245 goto fail_free_pio_data
;
8247 if (irqchip_in_kernel(vcpu
->kvm
)) {
8248 r
= kvm_create_lapic(vcpu
);
8250 goto fail_mmu_destroy
;
8252 static_key_slow_inc(&kvm_no_apic_vcpu
);
8254 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
8256 if (!vcpu
->arch
.mce_banks
) {
8258 goto fail_free_lapic
;
8260 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
8262 if (!zalloc_cpumask_var(&vcpu
->arch
.wbinvd_dirty_mask
, GFP_KERNEL
)) {
8264 goto fail_free_mce_banks
;
8269 vcpu
->arch
.guest_xstate_size
= XSAVE_HDR_SIZE
+ XSAVE_HDR_OFFSET
;
8271 vcpu
->arch
.maxphyaddr
= cpuid_query_maxphyaddr(vcpu
);
8273 vcpu
->arch
.pat
= MSR_IA32_CR_PAT_DEFAULT
;
8275 kvm_async_pf_hash_reset(vcpu
);
8278 vcpu
->arch
.pending_external_vector
= -1;
8279 vcpu
->arch
.preempted_in_kernel
= false;
8281 kvm_hv_vcpu_init(vcpu
);
8285 fail_free_mce_banks
:
8286 kfree(vcpu
->arch
.mce_banks
);
8288 kvm_free_lapic(vcpu
);
8290 kvm_mmu_destroy(vcpu
);
8292 free_page((unsigned long)vcpu
->arch
.pio_data
);
8297 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
8301 kvm_hv_vcpu_uninit(vcpu
);
8302 kvm_pmu_destroy(vcpu
);
8303 kfree(vcpu
->arch
.mce_banks
);
8304 kvm_free_lapic(vcpu
);
8305 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
8306 kvm_mmu_destroy(vcpu
);
8307 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
8308 free_page((unsigned long)vcpu
->arch
.pio_data
);
8309 if (!lapic_in_kernel(vcpu
))
8310 static_key_slow_dec(&kvm_no_apic_vcpu
);
8313 void kvm_arch_sched_in(struct kvm_vcpu
*vcpu
, int cpu
)
8315 vcpu
->arch
.l1tf_flush_l1d
= true;
8316 kvm_x86_ops
->sched_in(vcpu
, cpu
);
8319 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
8324 INIT_HLIST_HEAD(&kvm
->arch
.mask_notifier_list
);
8325 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
8326 INIT_LIST_HEAD(&kvm
->arch
.zapped_obsolete_pages
);
8327 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
8328 atomic_set(&kvm
->arch
.noncoherent_dma_count
, 0);
8330 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8331 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
8332 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8333 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID
,
8334 &kvm
->arch
.irq_sources_bitmap
);
8336 raw_spin_lock_init(&kvm
->arch
.tsc_write_lock
);
8337 mutex_init(&kvm
->arch
.apic_map_lock
);
8338 mutex_init(&kvm
->arch
.hyperv
.hv_lock
);
8339 spin_lock_init(&kvm
->arch
.pvclock_gtod_sync_lock
);
8341 kvm
->arch
.kvmclock_offset
= -ktime_get_boot_ns();
8342 pvclock_update_vm_gtod_copy(kvm
);
8344 INIT_DELAYED_WORK(&kvm
->arch
.kvmclock_update_work
, kvmclock_update_fn
);
8345 INIT_DELAYED_WORK(&kvm
->arch
.kvmclock_sync_work
, kvmclock_sync_fn
);
8347 kvm_page_track_init(kvm
);
8348 kvm_mmu_init_vm(kvm
);
8350 if (kvm_x86_ops
->vm_init
)
8351 return kvm_x86_ops
->vm_init(kvm
);
8356 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
8359 r
= vcpu_load(vcpu
);
8361 kvm_mmu_unload(vcpu
);
8365 static void kvm_free_vcpus(struct kvm
*kvm
)
8368 struct kvm_vcpu
*vcpu
;
8371 * Unpin any mmu pages first.
8373 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
8374 kvm_clear_async_pf_completion_queue(vcpu
);
8375 kvm_unload_vcpu_mmu(vcpu
);
8377 kvm_for_each_vcpu(i
, vcpu
, kvm
)
8378 kvm_arch_vcpu_free(vcpu
);
8380 mutex_lock(&kvm
->lock
);
8381 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
8382 kvm
->vcpus
[i
] = NULL
;
8384 atomic_set(&kvm
->online_vcpus
, 0);
8385 mutex_unlock(&kvm
->lock
);
8388 void kvm_arch_sync_events(struct kvm
*kvm
)
8390 cancel_delayed_work_sync(&kvm
->arch
.kvmclock_sync_work
);
8391 cancel_delayed_work_sync(&kvm
->arch
.kvmclock_update_work
);
8395 int __x86_set_memory_region(struct kvm
*kvm
, int id
, gpa_t gpa
, u32 size
)
8399 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
8400 struct kvm_memory_slot
*slot
, old
;
8402 /* Called with kvm->slots_lock held. */
8403 if (WARN_ON(id
>= KVM_MEM_SLOTS_NUM
))
8406 slot
= id_to_memslot(slots
, id
);
8412 * MAP_SHARED to prevent internal slot pages from being moved
8415 hva
= vm_mmap(NULL
, 0, size
, PROT_READ
| PROT_WRITE
,
8416 MAP_SHARED
| MAP_ANONYMOUS
, 0);
8417 if (IS_ERR((void *)hva
))
8418 return PTR_ERR((void *)hva
);
8427 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
8428 struct kvm_userspace_memory_region m
;
8430 m
.slot
= id
| (i
<< 16);
8432 m
.guest_phys_addr
= gpa
;
8433 m
.userspace_addr
= hva
;
8434 m
.memory_size
= size
;
8435 r
= __kvm_set_memory_region(kvm
, &m
);
8441 vm_munmap(old
.userspace_addr
, old
.npages
* PAGE_SIZE
);
8445 EXPORT_SYMBOL_GPL(__x86_set_memory_region
);
8447 int x86_set_memory_region(struct kvm
*kvm
, int id
, gpa_t gpa
, u32 size
)
8451 mutex_lock(&kvm
->slots_lock
);
8452 r
= __x86_set_memory_region(kvm
, id
, gpa
, size
);
8453 mutex_unlock(&kvm
->slots_lock
);
8457 EXPORT_SYMBOL_GPL(x86_set_memory_region
);
8459 void kvm_arch_destroy_vm(struct kvm
*kvm
)
8461 if (current
->mm
== kvm
->mm
) {
8463 * Free memory regions allocated on behalf of userspace,
8464 * unless the the memory map has changed due to process exit
8467 x86_set_memory_region(kvm
, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
, 0, 0);
8468 x86_set_memory_region(kvm
, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT
, 0, 0);
8469 x86_set_memory_region(kvm
, TSS_PRIVATE_MEMSLOT
, 0, 0);
8471 if (kvm_x86_ops
->vm_destroy
)
8472 kvm_x86_ops
->vm_destroy(kvm
);
8473 kvm_pic_destroy(kvm
);
8474 kvm_ioapic_destroy(kvm
);
8475 kvm_free_vcpus(kvm
);
8476 kvfree(rcu_dereference_check(kvm
->arch
.apic_map
, 1));
8477 kvm_mmu_uninit_vm(kvm
);
8478 kvm_page_track_cleanup(kvm
);
8481 void kvm_arch_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
8482 struct kvm_memory_slot
*dont
)
8486 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
8487 if (!dont
|| free
->arch
.rmap
[i
] != dont
->arch
.rmap
[i
]) {
8488 kvfree(free
->arch
.rmap
[i
]);
8489 free
->arch
.rmap
[i
] = NULL
;
8494 if (!dont
|| free
->arch
.lpage_info
[i
- 1] !=
8495 dont
->arch
.lpage_info
[i
- 1]) {
8496 kvfree(free
->arch
.lpage_info
[i
- 1]);
8497 free
->arch
.lpage_info
[i
- 1] = NULL
;
8501 kvm_page_track_free_memslot(free
, dont
);
8504 int kvm_arch_create_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
,
8505 unsigned long npages
)
8509 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
8510 struct kvm_lpage_info
*linfo
;
8515 lpages
= gfn_to_index(slot
->base_gfn
+ npages
- 1,
8516 slot
->base_gfn
, level
) + 1;
8518 slot
->arch
.rmap
[i
] =
8519 kvzalloc(lpages
* sizeof(*slot
->arch
.rmap
[i
]), GFP_KERNEL
);
8520 if (!slot
->arch
.rmap
[i
])
8525 linfo
= kvzalloc(lpages
* sizeof(*linfo
), GFP_KERNEL
);
8529 slot
->arch
.lpage_info
[i
- 1] = linfo
;
8531 if (slot
->base_gfn
& (KVM_PAGES_PER_HPAGE(level
) - 1))
8532 linfo
[0].disallow_lpage
= 1;
8533 if ((slot
->base_gfn
+ npages
) & (KVM_PAGES_PER_HPAGE(level
) - 1))
8534 linfo
[lpages
- 1].disallow_lpage
= 1;
8535 ugfn
= slot
->userspace_addr
>> PAGE_SHIFT
;
8537 * If the gfn and userspace address are not aligned wrt each
8538 * other, or if explicitly asked to, disable large page
8539 * support for this slot
8541 if ((slot
->base_gfn
^ ugfn
) & (KVM_PAGES_PER_HPAGE(level
) - 1) ||
8542 !kvm_largepages_enabled()) {
8545 for (j
= 0; j
< lpages
; ++j
)
8546 linfo
[j
].disallow_lpage
= 1;
8550 if (kvm_page_track_create_memslot(slot
, npages
))
8556 for (i
= 0; i
< KVM_NR_PAGE_SIZES
; ++i
) {
8557 kvfree(slot
->arch
.rmap
[i
]);
8558 slot
->arch
.rmap
[i
] = NULL
;
8562 kvfree(slot
->arch
.lpage_info
[i
- 1]);
8563 slot
->arch
.lpage_info
[i
- 1] = NULL
;
8568 void kvm_arch_memslots_updated(struct kvm
*kvm
, struct kvm_memslots
*slots
)
8571 * memslots->generation has been incremented.
8572 * mmio generation may have reached its maximum value.
8574 kvm_mmu_invalidate_mmio_sptes(kvm
, slots
);
8577 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
8578 struct kvm_memory_slot
*memslot
,
8579 const struct kvm_userspace_memory_region
*mem
,
8580 enum kvm_mr_change change
)
8585 static void kvm_mmu_slot_apply_flags(struct kvm
*kvm
,
8586 struct kvm_memory_slot
*new)
8588 /* Still write protect RO slot */
8589 if (new->flags
& KVM_MEM_READONLY
) {
8590 kvm_mmu_slot_remove_write_access(kvm
, new);
8595 * Call kvm_x86_ops dirty logging hooks when they are valid.
8597 * kvm_x86_ops->slot_disable_log_dirty is called when:
8599 * - KVM_MR_CREATE with dirty logging is disabled
8600 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8602 * The reason is, in case of PML, we need to set D-bit for any slots
8603 * with dirty logging disabled in order to eliminate unnecessary GPA
8604 * logging in PML buffer (and potential PML buffer full VMEXT). This
8605 * guarantees leaving PML enabled during guest's lifetime won't have
8606 * any additonal overhead from PML when guest is running with dirty
8607 * logging disabled for memory slots.
8609 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8610 * to dirty logging mode.
8612 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8614 * In case of write protect:
8616 * Write protect all pages for dirty logging.
8618 * All the sptes including the large sptes which point to this
8619 * slot are set to readonly. We can not create any new large
8620 * spte on this slot until the end of the logging.
8622 * See the comments in fast_page_fault().
8624 if (new->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
8625 if (kvm_x86_ops
->slot_enable_log_dirty
)
8626 kvm_x86_ops
->slot_enable_log_dirty(kvm
, new);
8628 kvm_mmu_slot_remove_write_access(kvm
, new);
8630 if (kvm_x86_ops
->slot_disable_log_dirty
)
8631 kvm_x86_ops
->slot_disable_log_dirty(kvm
, new);
8635 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
8636 const struct kvm_userspace_memory_region
*mem
,
8637 const struct kvm_memory_slot
*old
,
8638 const struct kvm_memory_slot
*new,
8639 enum kvm_mr_change change
)
8641 int nr_mmu_pages
= 0;
8643 if (!kvm
->arch
.n_requested_mmu_pages
)
8644 nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
8647 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
8650 * Dirty logging tracks sptes in 4k granularity, meaning that large
8651 * sptes have to be split. If live migration is successful, the guest
8652 * in the source machine will be destroyed and large sptes will be
8653 * created in the destination. However, if the guest continues to run
8654 * in the source machine (for example if live migration fails), small
8655 * sptes will remain around and cause bad performance.
8657 * Scan sptes if dirty logging has been stopped, dropping those
8658 * which can be collapsed into a single large-page spte. Later
8659 * page faults will create the large-page sptes.
8661 if ((change
!= KVM_MR_DELETE
) &&
8662 (old
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) &&
8663 !(new->flags
& KVM_MEM_LOG_DIRTY_PAGES
))
8664 kvm_mmu_zap_collapsible_sptes(kvm
, new);
8667 * Set up write protection and/or dirty logging for the new slot.
8669 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8670 * been zapped so no dirty logging staff is needed for old slot. For
8671 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8672 * new and it's also covered when dealing with the new slot.
8674 * FIXME: const-ify all uses of struct kvm_memory_slot.
8676 if (change
!= KVM_MR_DELETE
)
8677 kvm_mmu_slot_apply_flags(kvm
, (struct kvm_memory_slot
*) new);
8680 void kvm_arch_flush_shadow_all(struct kvm
*kvm
)
8682 kvm_mmu_invalidate_zap_all_pages(kvm
);
8685 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
8686 struct kvm_memory_slot
*slot
)
8688 kvm_page_track_flush_slot(kvm
, slot
);
8691 static inline bool kvm_vcpu_has_events(struct kvm_vcpu
*vcpu
)
8693 if (!list_empty_careful(&vcpu
->async_pf
.done
))
8696 if (kvm_apic_has_events(vcpu
))
8699 if (vcpu
->arch
.pv
.pv_unhalted
)
8702 if (vcpu
->arch
.exception
.pending
)
8705 if (kvm_test_request(KVM_REQ_NMI
, vcpu
) ||
8706 (vcpu
->arch
.nmi_pending
&&
8707 kvm_x86_ops
->nmi_allowed(vcpu
)))
8710 if (kvm_test_request(KVM_REQ_SMI
, vcpu
) ||
8711 (vcpu
->arch
.smi_pending
&& !is_smm(vcpu
)))
8714 if (kvm_arch_interrupt_allowed(vcpu
) &&
8715 kvm_cpu_has_interrupt(vcpu
))
8718 if (kvm_hv_has_stimer_pending(vcpu
))
8724 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
8726 return kvm_vcpu_running(vcpu
) || kvm_vcpu_has_events(vcpu
);
8729 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu
*vcpu
)
8731 return vcpu
->arch
.preempted_in_kernel
;
8734 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
8736 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
8739 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
8741 return kvm_x86_ops
->interrupt_allowed(vcpu
);
8744 unsigned long kvm_get_linear_rip(struct kvm_vcpu
*vcpu
)
8746 if (is_64_bit_mode(vcpu
))
8747 return kvm_rip_read(vcpu
);
8748 return (u32
)(get_segment_base(vcpu
, VCPU_SREG_CS
) +
8749 kvm_rip_read(vcpu
));
8751 EXPORT_SYMBOL_GPL(kvm_get_linear_rip
);
8753 bool kvm_is_linear_rip(struct kvm_vcpu
*vcpu
, unsigned long linear_rip
)
8755 return kvm_get_linear_rip(vcpu
) == linear_rip
;
8757 EXPORT_SYMBOL_GPL(kvm_is_linear_rip
);
8759 unsigned long kvm_get_rflags(struct kvm_vcpu
*vcpu
)
8761 unsigned long rflags
;
8763 rflags
= kvm_x86_ops
->get_rflags(vcpu
);
8764 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
8765 rflags
&= ~X86_EFLAGS_TF
;
8768 EXPORT_SYMBOL_GPL(kvm_get_rflags
);
8770 static void __kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
8772 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
&&
8773 kvm_is_linear_rip(vcpu
, vcpu
->arch
.singlestep_rip
))
8774 rflags
|= X86_EFLAGS_TF
;
8775 kvm_x86_ops
->set_rflags(vcpu
, rflags
);
8778 void kvm_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
8780 __kvm_set_rflags(vcpu
, rflags
);
8781 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
8783 EXPORT_SYMBOL_GPL(kvm_set_rflags
);
8785 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
, struct kvm_async_pf
*work
)
8789 if ((vcpu
->arch
.mmu
.direct_map
!= work
->arch
.direct_map
) ||
8793 r
= kvm_mmu_reload(vcpu
);
8797 if (!vcpu
->arch
.mmu
.direct_map
&&
8798 work
->arch
.cr3
!= vcpu
->arch
.mmu
.get_cr3(vcpu
))
8801 vcpu
->arch
.mmu
.page_fault(vcpu
, work
->gva
, 0, true);
8804 static inline u32
kvm_async_pf_hash_fn(gfn_t gfn
)
8806 return hash_32(gfn
& 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU
));
8809 static inline u32
kvm_async_pf_next_probe(u32 key
)
8811 return (key
+ 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU
) - 1);
8814 static void kvm_add_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8816 u32 key
= kvm_async_pf_hash_fn(gfn
);
8818 while (vcpu
->arch
.apf
.gfns
[key
] != ~0)
8819 key
= kvm_async_pf_next_probe(key
);
8821 vcpu
->arch
.apf
.gfns
[key
] = gfn
;
8824 static u32
kvm_async_pf_gfn_slot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8827 u32 key
= kvm_async_pf_hash_fn(gfn
);
8829 for (i
= 0; i
< roundup_pow_of_two(ASYNC_PF_PER_VCPU
) &&
8830 (vcpu
->arch
.apf
.gfns
[key
] != gfn
&&
8831 vcpu
->arch
.apf
.gfns
[key
] != ~0); i
++)
8832 key
= kvm_async_pf_next_probe(key
);
8837 bool kvm_find_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8839 return vcpu
->arch
.apf
.gfns
[kvm_async_pf_gfn_slot(vcpu
, gfn
)] == gfn
;
8842 static void kvm_del_async_pf_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
8846 i
= j
= kvm_async_pf_gfn_slot(vcpu
, gfn
);
8848 vcpu
->arch
.apf
.gfns
[i
] = ~0;
8850 j
= kvm_async_pf_next_probe(j
);
8851 if (vcpu
->arch
.apf
.gfns
[j
] == ~0)
8853 k
= kvm_async_pf_hash_fn(vcpu
->arch
.apf
.gfns
[j
]);
8855 * k lies cyclically in ]i,j]
8857 * |....j i.k.| or |.k..j i...|
8859 } while ((i
<= j
) ? (i
< k
&& k
<= j
) : (i
< k
|| k
<= j
));
8860 vcpu
->arch
.apf
.gfns
[i
] = vcpu
->arch
.apf
.gfns
[j
];
8865 static int apf_put_user(struct kvm_vcpu
*vcpu
, u32 val
)
8868 return kvm_write_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, &val
,
8872 static int apf_get_user(struct kvm_vcpu
*vcpu
, u32
*val
)
8875 return kvm_read_guest_cached(vcpu
->kvm
, &vcpu
->arch
.apf
.data
, val
,
8879 void kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
8880 struct kvm_async_pf
*work
)
8882 struct x86_exception fault
;
8884 trace_kvm_async_pf_not_present(work
->arch
.token
, work
->gva
);
8885 kvm_add_async_pf_gfn(vcpu
, work
->arch
.gfn
);
8887 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
) ||
8888 (vcpu
->arch
.apf
.send_user_only
&&
8889 kvm_x86_ops
->get_cpl(vcpu
) == 0))
8890 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
8891 else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_NOT_PRESENT
)) {
8892 fault
.vector
= PF_VECTOR
;
8893 fault
.error_code_valid
= true;
8894 fault
.error_code
= 0;
8895 fault
.nested_page_fault
= false;
8896 fault
.address
= work
->arch
.token
;
8897 fault
.async_page_fault
= true;
8898 kvm_inject_page_fault(vcpu
, &fault
);
8902 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
8903 struct kvm_async_pf
*work
)
8905 struct x86_exception fault
;
8908 if (work
->wakeup_all
)
8909 work
->arch
.token
= ~0; /* broadcast wakeup */
8911 kvm_del_async_pf_gfn(vcpu
, work
->arch
.gfn
);
8912 trace_kvm_async_pf_ready(work
->arch
.token
, work
->gva
);
8914 if (vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
&&
8915 !apf_get_user(vcpu
, &val
)) {
8916 if (val
== KVM_PV_REASON_PAGE_NOT_PRESENT
&&
8917 vcpu
->arch
.exception
.pending
&&
8918 vcpu
->arch
.exception
.nr
== PF_VECTOR
&&
8919 !apf_put_user(vcpu
, 0)) {
8920 vcpu
->arch
.exception
.injected
= false;
8921 vcpu
->arch
.exception
.pending
= false;
8922 vcpu
->arch
.exception
.nr
= 0;
8923 vcpu
->arch
.exception
.has_error_code
= false;
8924 vcpu
->arch
.exception
.error_code
= 0;
8925 } else if (!apf_put_user(vcpu
, KVM_PV_REASON_PAGE_READY
)) {
8926 fault
.vector
= PF_VECTOR
;
8927 fault
.error_code_valid
= true;
8928 fault
.error_code
= 0;
8929 fault
.nested_page_fault
= false;
8930 fault
.address
= work
->arch
.token
;
8931 fault
.async_page_fault
= true;
8932 kvm_inject_page_fault(vcpu
, &fault
);
8935 vcpu
->arch
.apf
.halted
= false;
8936 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
8939 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu
*vcpu
)
8941 if (!(vcpu
->arch
.apf
.msr_val
& KVM_ASYNC_PF_ENABLED
))
8944 return kvm_can_do_async_pf(vcpu
);
8947 void kvm_arch_start_assignment(struct kvm
*kvm
)
8949 atomic_inc(&kvm
->arch
.assigned_device_count
);
8951 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment
);
8953 void kvm_arch_end_assignment(struct kvm
*kvm
)
8955 atomic_dec(&kvm
->arch
.assigned_device_count
);
8957 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment
);
8959 bool kvm_arch_has_assigned_device(struct kvm
*kvm
)
8961 return atomic_read(&kvm
->arch
.assigned_device_count
);
8963 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device
);
8965 void kvm_arch_register_noncoherent_dma(struct kvm
*kvm
)
8967 atomic_inc(&kvm
->arch
.noncoherent_dma_count
);
8969 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma
);
8971 void kvm_arch_unregister_noncoherent_dma(struct kvm
*kvm
)
8973 atomic_dec(&kvm
->arch
.noncoherent_dma_count
);
8975 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma
);
8977 bool kvm_arch_has_noncoherent_dma(struct kvm
*kvm
)
8979 return atomic_read(&kvm
->arch
.noncoherent_dma_count
);
8981 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma
);
8983 bool kvm_arch_has_irq_bypass(void)
8985 return kvm_x86_ops
->update_pi_irte
!= NULL
;
8988 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer
*cons
,
8989 struct irq_bypass_producer
*prod
)
8991 struct kvm_kernel_irqfd
*irqfd
=
8992 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
8994 irqfd
->producer
= prod
;
8996 return kvm_x86_ops
->update_pi_irte(irqfd
->kvm
,
8997 prod
->irq
, irqfd
->gsi
, 1);
9000 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer
*cons
,
9001 struct irq_bypass_producer
*prod
)
9004 struct kvm_kernel_irqfd
*irqfd
=
9005 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
9007 WARN_ON(irqfd
->producer
!= prod
);
9008 irqfd
->producer
= NULL
;
9011 * When producer of consumer is unregistered, we change back to
9012 * remapped mode, so we can re-use the current implementation
9013 * when the irq is masked/disabled or the consumer side (KVM
9014 * int this case doesn't want to receive the interrupts.
9016 ret
= kvm_x86_ops
->update_pi_irte(irqfd
->kvm
, prod
->irq
, irqfd
->gsi
, 0);
9018 printk(KERN_INFO
"irq bypass consumer (token %p) unregistration"
9019 " fails: %d\n", irqfd
->consumer
.token
, ret
);
9022 int kvm_arch_update_irqfd_routing(struct kvm
*kvm
, unsigned int host_irq
,
9023 uint32_t guest_irq
, bool set
)
9025 if (!kvm_x86_ops
->update_pi_irte
)
9028 return kvm_x86_ops
->update_pi_irte(kvm
, host_irq
, guest_irq
, set
);
9031 bool kvm_vector_hashing_enabled(void)
9033 return vector_hashing
;
9035 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled
);
9037 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
9038 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio
);
9039 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
9040 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
9041 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
9042 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);
9043 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun
);
9044 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit
);
9045 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject
);
9046 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit
);
9047 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga
);
9048 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit
);
9049 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts
);
9050 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset
);
9051 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window
);
9052 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full
);
9053 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update
);
9054 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access
);
9055 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi
);